This action might not be possible to undo. Are you sure you want to continue?

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

Department of Logic

Logic

Michal Peliˇs

Logika ot

´

azek

Logic of Questions

PhD Thesis

Supervised by prof. RNDr. Jaroslav Peregrin, CSc.

2011

Prohlaˇsuji, ˇze jsem disertaˇcn´ı pr´aci napsal samostatnˇe s vyuˇzit´ım pouze uve-

den´ ych a ˇr´adnˇe citovan´ ych pramen˚u a literatury a ˇze pr´ace nebyla vyuˇzita v

r´amci jin´eho vysokoˇskolsk´eho studia ˇci k z´ısk´an´ı jin´eho nebo stejn´eho titulu.

V Praze, 11. ledna 2011

Michal Peliˇs

Abstract

The thesis deals with logic of questions (erotetic logic), which is one of the

branches of non-classical logic. In the introductory part we speak generally

about formalization of questions and the newest approaches to questions in

logic are summed up. We introduce a formalization based on sets of direct

answers and point out the role of inferences with questions. The rest of the

thesis consists of two parts that can be read independently.

The ﬁrst part focuses on relationships among consequence relations in

inferential erotetic logic (IEL). We keep the framework of original IEL, intro-

duced by Andrzej Wi´sniewski, together with the representation of questions

by sets of direct answers. Answers are strictly formulas of the declarative lan-

guage. The mix of interrogatives and declaratives occours just on the level of

consequences. Consequence relations with questions are deﬁned by means of

multiple-conclusion entailment among sets of declarative formulas. This way,

one can work with classes of models and to make transparent some properties

and relationships. We provide a general study of erotetic inferences based on

IEL that is open for non-classical applications.

The second part contains epistemic erotetic logic. A question is under-

stood as a set of direct answers; however, this time the set is ﬁnite. The

satisﬁability of a question in a state (possible world) of an epistemic model

is deﬁned by three conditions (a questioner does not know any direct answer,

each direct answer considers as possible and at least one of them must be the

right one). This approach is a new one and it is suitable for generalization

to every epistemic-like system by questions and common erotetic concepts

(e.g., various types of answers and erotetic inferences). The goal of this study

is a future application in communication theory of a group of agents. We

ﬁnish this part by multi-agent public announcement logic with application

of questions by answer mining in a group of agents.

Abstrakt

Pr´ace se zab´ yv´a jedn´ım z odvˇetv´ı neklasick´ ych logik – logikou ot´azek (erotet-

ickou logikou). V ´ uvodn´ı ˇc´asti se hovoˇr´ı obecnˇe o formalizovan´em pˇr´ıstupu

k ot´azk´am v logice a souˇcasnˇe je struˇcnˇe shrnuta zejm´ena nejnovˇejˇs´ı historie

tohoto odvˇetv´ı. Zde je t´eˇz zd˚uvodˇ nov´ana smysluplnost zachycen´ı ´ usudk˚u,

v nichˇz se ot´azky objevuj´ı, a je pˇredstavena formalizace ot´azky zaloˇzen´a na

explicitn´ım stanoven´ı mnoˇziny pˇr´ım´ ych odpovˇed´ı. Zbytek pr´ace je rozdˇelen

na dvˇe ˇc´asti, kter´e lze ˇc´ıst nez´avisle.

Prvn´ı ˇc´ast se zab´ yv´a d˚usledkov´ ymi relacemi v inferenˇcn´ı erotetick´e logice

(inferential erotetic logic). Plnˇe zde vyuˇz´ıv´ame r´amec p˚uvodn´ı inferenˇcn´ı

erotetick´e logiky zaveden´e Andrzejem Wi´sniewskim. Pouˇz´ıv´ame v´ yhradnˇe

formalizaci ot´azek pomoc´ı mnoˇziny pˇr´ım´ ych odpovˇed´ı, kdy pˇr´ım´e odpovˇedi

jsou formule deklarativn´ıho jazyka. Protoˇze jsou d˚usledkov´e relace s ot´azkami

deﬁnov´any pomoc´ı klasick´eho v´ıcez´avˇerov´eho s´emantick´eho d˚usledku, vyuˇz´ı-

v´ame pˇri d˚ukazech pˇr´ıstup zaloˇzen´ y na tˇr´ıd´ach model˚u. V t´eto ˇc´asti n´am

jde o obecn´ y pˇr´ıstup ovˇsem s omezen´ım, kdy je deklarativn´ı jazyk rozˇs´ıˇren

o ot´azky, ale k propojen´ı deklarativn´ıho a interogativn´ıho jazyka dojde aˇz

na ´ urovni d˚usledkov´ ych relac´ı. Prim´arn´ım z´ajmem je studovat vztahy mezi

jednotliv´ ymi erotetick´ ymi d˚usledkov´ ymi relacemi.

Druh´a ˇc´ast obsahuje epistemickou erotetickou logiku. Prvotn´ı je epis-

temick´ y r´amec tvoˇren´ y zvolenou epistemickou logikou. Ot´azka je nad´ale

formalizov´ana jako mnoˇzina pˇr´ım´ ych odpovˇed´ı, tentokr´at vˇsak uvaˇzujeme

koneˇcnou variantu a jej´ı splnˇenost v moˇzn´em svˇetˇe epistemick´eho modelu

je v´az´ana na platnost tˇr´ı epistemick´ ych podm´ınek (tazatel nezn´a ˇz´adnou z

odpovˇed´ı, kaˇzdou z pˇr´ım´ ych odpovˇed´ı vˇsak povaˇzuje za moˇznou a alespoˇ n

jedna z pˇr´ım´ ych odpovˇed´ı mus´ı b´ yt spr´avn´a). Jde o nov´ y pˇr´ıstup, kter´ y

umoˇzˇ nuje rozˇs´ıˇrit libovoln´ y epistemick´ y syst´em o ot´azky a s nimi spojenou

erotetickou terminologii (r˚uzn´e typy odpovˇed´ı, inference s ot´azkami a dalˇs´ı).

Pˇrestoˇze i zde se snaˇz´ıme o obecn´ y syst´em, naˇs´ım hlavn´ım c´ılem je komu-

nikace ve skupinˇe agent˚u pˇri hled´an´ı odpovˇed´ı na ot´azky. V z´avˇeru druh´e

ˇc´asti tak pˇredstavujeme multiagentn´ı logiku veˇrejn´eho vyhl´aˇsen´ı (public an-

nouncement logic) s ot´azkami a skupinov´ ymi znalostmi.

Preface

The work on this thesis started in 2003 when I met logic of questions—the

branch of logic, which seemed to me very promising in possible applications

and open for an extensive development. At the very beginning, I balanced

between the reasons for and against having a special logic of questions. Then

I was lucky to ﬁnd two erotetic logics that proved their vitality in the recent

years. The ﬁrst one was inferential erotetic logic (IEL) developed by Andrzej

Wi´sniewski and his collaborators and the second one was an intensional ap-

proach to questions of Jeroen Groenendijk and Martin Stokhof.

My interest in non-standard consequence relations brought me to the

study of inferential erotetic logic ﬁrst. IEL provides consequence relations

with declaratives as well as interrogatives. I decided to learn this system,

to go through the relationships of inferences with questions and, inspired by

the methodology of a question representation used in IEL, to suggest some

generalizations and new relationships. (See chapter 2.)

During the study of IEL, I worked with my colleagues on epistemic in-

terpretation of the relevant logic, see our results in [22] and in [2]. This

and the intensional approach of Groenendijk and Stokhof were inspiration

for my own extension of epistemic logic by questions. In many approaches

to the formalization of interrogative sentences, epistemic terminology is used

and questions are often seen as requirements of knowledge completion. I

wished to formulate a completely general framework that can be used in all

‘epistemic-like’ systems. Such aspirations were successful, and the ﬁndings

are presented in chapter 3. Naturally, if there is an epistemic system with

questions, it is only one step away to its dynamic application. In the thesis

I decided to use public announcement logic together with a group epistemic

modalities and chapter 4 contains results obtained in this ﬁeld.

Structure of the thesis

The thesis includes two main parts that can be read independently. The ﬁrst

one is chapter 2 and the second one consists of chapters 3 and 4. Chapter 1

i

provides common methodology for both parts. Chapter 2 can be understood

as an inspiration of some erotetic concepts used in the rest of the thesis. The

mentioned independence of chapter 2 and chapters 3 and 4 is also implied

by the fact that these parts are based on separate papers—chapter 2 was

published in [31] and a simpliﬁed version of chapters 3 and 4 appeared in [32]

for the ﬁrst time and [33] contains the last results from chapter 4.

Chapter 2

Chapter 1

`

Chapter 3 → Chapter 4

Chapter 2 can be read as a full introduction to the topic. In chapters 3 and

4 it is not the case, the basic knowledge of modal logic is required.

The last chapter 5 contains some ﬁnal remarks to the used set-of-answers

methodology, to the main results in the second part of the thesis, and to the

related approaches, and also discusses further directions.

Chapter 1: Logic and questions

The chapter brieﬂy introduces the multi-paradigmatic situation in the metho-

dology of erotetic logic and contains a short historical overview of this branch

of logic with a special emphasis on the recent development. We introduce

brieﬂy inferential erotetic logic, Groenendijk-Stokhof’s intensional approach,

and some developments of these theories. However, the core of the chapter is

devoted to a formalization of questions based on sets of answers. We justify

the usefulness of the methodology in the study of consequence relations with

questions as well as in an epistemic interpretation of questions.

Chapter 2: Consequence relations in inferential erotetic

logic

This part is aimed to study relationships among consequence relations that

were introduced by inferential erotetic logic (IEL). We keep the framework of

IEL, but the question representation uses the methodology from chapter 1.

IEL requires that declarative and interrogative formulas are not mixed on

the object-language level. Also, answers are strictly declarative sentences.

The deﬁned consequence relations with questions are naturally based on the

multiple-conclusion entailment among sets of declarative formulas. We add

ii

the semantic range of a question to the terminology of IEL and sets of declar-

atives are associated with classes of models. This ‘model-based approach’

makes proofs and properties transparent. This chapter, a technical overview

of some IEL concepts and their properties, can provide a general framework

and inspiration for the work with inferences among questions and declara-

tives.

Chapter 3: Epistemic logic with questions

The main goal is to incorporate questions in a general epistemic framework.

Questions are considered to be ﬁnite sets of direct answers and their satisﬁ-

ability in a state of an epistemic model is based on three conditions, which

express ignorance and presuppositions of a questioner. While this deﬁnition

of questions’ askability is fully general for any ‘epistemic-like’ system and

it is not necessary to keep the ﬁnite set-of-answers methodology, we work

with the normal multi-modal propositional logic K as a background for the

introduction of multi-agent epistemic logic. In this framework, a question

becomes a complex modal formula. Inspired by inferential structures in IEL,

we show that there are ‘philosophically’ similar structures based on classical

implication. The rest of the chapter is devoted to answerhood conditions and

the role of implication with respect to epistemic context and conjunctions of

yes-no questions.

Chapter 4: A step to dynamization of erotetic logic

This chapter takes the full advantage of the multi-agent setting from chap-

ter 3 and can be considered as an application of the introduced erotetic-epi-

stemic approach in a dynamic framework. We deﬁne here an epistemic logic

based on the modal system S5 extended by group modalities together with

public announcement modality. Askablity of questions as well as answerhood

conditions are studied from the viewpoint of groups of agents. Finally we

show the role of questions and group modalities in answer ‘mining’ among

agents.

Acknowledgements

It would be a long list of names to express my thanks to everybody who

helped me with the study as well as with the thesis. Let me mention only

some of them.

iii

First of all I have to thank my supervisor Jaroslav Peregrin who intro-

duced me to formal semantics and non-classical logics, read and discussed

my papers, and contemplated my work.

I consider myself lucky to have so many colleagues I can cooperate with.

Among them are Marta B´ılkov´a, Libor Bˇehounek, Tim Childers, and Greg

Restall.

My special thanks go to Ondrej Majer who has helped me with the work

in dynamization of erotetic epistemic logic. We wrote some papers from this

ﬁeld and, simultaneously, we have been working in epistemic relevant logic.

From the very beginning I was supported by papers, advice, and very

helpful comments by Andrzej Wi´sniewski.

The work on the thesis was supported by the following grants:

• Formal and historical approach to epistemology, Grant agency of the Czech Repub-

lic, no. P401/10/1504

• Logical models of reasoning with vague information, grant ICC/08/E018 of the

Grant Agency of the Czech Republic (a part of ESF Eurocores-LogICCC project

FP006)

• Dynamic Formal Systems, Grant agency of the Academy of Sciences of the Czech

Republic, no. IAA900090703

• Logical foundations for semantics and representation of knowledge, Grant agency

of the Czech Republic, no. 401/03/H047

I would like to thank all who provided me the possibility to participate on

the grants: Tim Childers, Petr Cintula, Ludmila Dost´alov´a, Chris Ferm¨ uller,

Petr H´ajek, Petr Jirk˚u, Vojtˇech Kolman, and Ondrej Majer.

Before I started to think about erotetic logic, Hana Skˇrivanov´a was the

ﬁrst one to bring me to this topic, and, a bit later, Martin V´ıta brought my

attention to epistemic logic with group modalities. During the work, Jeroen

Groenendijk sent me some of his papers in progress.

I would like to thank all my colleagues at the Faculty of Physical Educa-

tion and Sport (Charles University in Prague) that tolerated my interest in

logic although they would prefer me to be a sociologist.

Last but not least, my thanks go to my family as well as colleagues and

students at the Department of Logic at the Faculty of Arts (Charles Univer-

sity in Prague) and Institute of Philosophy (Academy of Sciences of the Czech

Republic). Let me mention, especially, Kamila Bendov´a, Jana Bureˇsov´a,

Radek Honz´ık, Marie Kol´ınsk´a, Svatopluk Nevrkla, Martina Pivoˇ nkov´a, Vla-

dim´ır Svoboda, and V´ıtˇezslav

ˇ

Svejdar.

iv

Contents

1 Logic and questions 1

1.1 Questions, answers, and inferences . . . . . . . . . . . . . . . . 1

1.1.1 Questions and answers . . . . . . . . . . . . . . . . . . 1

1.1.2 Inferences with questions . . . . . . . . . . . . . . . . . 3

1.2 Set-of-answers methodology . . . . . . . . . . . . . . . . . . . 5

1.2.1 Semantics of questions . . . . . . . . . . . . . . . . . . 5

1.2.2 Sets of answers . . . . . . . . . . . . . . . . . . . . . . 7

1.2.3 Epistemic aspects of SAM . . . . . . . . . . . . . . . . 9

1.3 Note on the recent history of erotetic logic . . . . . . . . . . . 10

1.3.1 Inferential erotetic logic . . . . . . . . . . . . . . . . . 11

1.3.2 Intensional erotetic logic . . . . . . . . . . . . . . . . . 12

2 Consequence relations in inferential erotetic logic 14

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.1.1 Adapted set-of-answers methodology in IEL . . . . . . 14

2.1.2 Consequence relations in IEL . . . . . . . . . . . . . . 15

2.1.3 Model-based approach . . . . . . . . . . . . . . . . . . 16

2.1.4 Basic properties of questions . . . . . . . . . . . . . . . 18

2.1.5 The road we are going to take . . . . . . . . . . . . . . 19

2.2 Questions and declaratives . . . . . . . . . . . . . . . . . . . . 19

2.2.1 Evocation . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.2.2 Presuppositions . . . . . . . . . . . . . . . . . . . . . . 22

2.3 Questions and questions . . . . . . . . . . . . . . . . . . . . . 27

2.3.1 Erotetic implication . . . . . . . . . . . . . . . . . . . . 27

2.3.2 Evocation and erotetic implication . . . . . . . . . . . 32

2.3.3 Comparing questions: relations of questions based on

direct answers . . . . . . . . . . . . . . . . . . . . . . . 34

2.3.4 Questions and sets of questions . . . . . . . . . . . . . 38

2.4 Final remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

v

3 Epistemic logic with questions 47

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

3.2 Single-agent propositional epistemic logic and questions . . . . 49

3.2.1 Incorporating questions . . . . . . . . . . . . . . . . . . 50

3.2.2 Some important classes of questions . . . . . . . . . . . 55

3.3 Epistemic erotetic implication . . . . . . . . . . . . . . . . . . 58

3.4 Askability and answerhood . . . . . . . . . . . . . . . . . . . . 60

3.5 Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

3.6 Implied questions . . . . . . . . . . . . . . . . . . . . . . . . . 65

4 A step to dynamization of erotetic logic 67

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

4.2 Multi-agent propositional epistemic logic with questions . . . . 68

4.2.1 Group epistemic modalities . . . . . . . . . . . . . . . 68

4.2.2 Group questions and answerhood . . . . . . . . . . . . 71

4.3 Public announcement . . . . . . . . . . . . . . . . . . . . . . . 73

4.3.1 Updates and questions . . . . . . . . . . . . . . . . . . 75

4.3.2 Public announcement and answerhood . . . . . . . . . 76

4.3.3 Answer mining in a group . . . . . . . . . . . . . . . . 78

4.4 Final remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

5 Conclusion 82

5.1 Related works and future directions . . . . . . . . . . . . . . . 84

Bibliography 86

vi

Chapter 1

Logic and questions

1.1 Questions, answers, and inferences

In this chapter we wish to show that it is reasonable to consider questions as a

part of logical study. In logic, declarative sentences usually have their formal

(logical) counterparts and play an important role in argumentation. We

often see logic to be primarily a study of inferences. Inferential structures are

studied in formal systems, which can diﬀer in the formalization of declaratives

as well as in admitting or rejecting of some principles.

We believe that the dealing with questions in the logical framework will be

justiﬁed if we show that questions can play an autonomous and important role

in inferences. Perhaps this point may be considered as the most important

to justify logic of questions.

1

This introductory chapter provides a brief overview of history as well as

methodology in recent approaches to erotetic logic. However, the main aim is

to concentrate on a methodology used in the rest of the paper—we introduce

and discuss a variant of the methodology based on sets of answers.

1.1.1 Questions and answers

Let us imagine a group of three players: Ann, Bill, and Catherine. Each of

them has one card and nobody can see the cards of the others. One of the

cards is the Joker and everybody knows this fact. Then

Who has the Joker?

is a reasonable sentence in this situation. We recognize it as an interrogative

sentence because of its word order and the question mark. The hearing

1

In this paper we use the term logic of questions in the same meaning as erotetic logic,

a discussion on both terms can be found in [16].

1

1.1. Questions, answers, and inferences

or uttering of an interrogative is followed by intonation and interrogative

pronounce.

An interrogative sentence includes a pragmatic aspect. It is a “request to

an addressee to provide the speaker with certain information”—interrogative

speech act [14, p. 1057]. Pragmatic approach emphasizes the roles of a

speaker and an addressee, which seem to be outside of the interrogative

context, but it seems very important in an analysis of questions. This can

be the reason why some logicians argue against erotetic logic.

If we want to work with interrogatives in a formal system, we have to

decide two problems:

1. What is the formal shape of questions?

2. What is the (formal) semantics of questions?

Reviewing the history of erotetic logic, there is no unique solution. There

are many approaches to the formalization of questions and every approach

varies according to what is considered as important. Logic of questions is

considered to be multiparadigmatic. This is nicely illustrated by Harrah’s

examples of ‘meta-axioms’, see [15, pp. 25–26]. He groups them into three

sets according to the acceptance by erotetic logicians.

1. The ﬁrst group includes meta-axioms accepted in almost all systems.

Harrah calls them absolute axioms and examples are:

(a) Every question has at least one partial answer.

(b) (In systems with negation) For every statement P, there exists a

question Q whose direct answers include P and the negation of P.

(c) Every question Q has a presupposition P such that: P is a state-

ment, and if Q has any true direct answer, then P is true.

2. The second group, standard axioms, is often accepted, but not in all

systems.

(a) Every question has at least one direct answer.

(b) Every direct answer is a statement.

(c) Every partial answer is implied by some direct answer.

(d) Every question is expressed by at least one interrogative.

(e) Each interrogative expresses exactly one question.

(f) Given an interrogative I there is an eﬀective method for deter-

mining the direct answers to the question expressed by I.

2

1.1. Questions, answers, and inferences

3. The last group is called excentric axioms. Thus, the following examples

of such axioms are accepted only in some interrogative systems.

(a) If two questions have the same direct answers, then the two ques-

tions are identical.

(b) Every question Q has a presupposition that is true just in case

some direct answer to Q is true.

Let us notice the terminology, the diﬀerence between interrogative (sentence)

and question was just introduced by standard axioms. The ﬁrst term mostly

refers to a type of sentence and the second one is a bit more complex. A ques-

tion is expressed by an interrogative (sentence) and can be ‘posed’, ‘asked’,

etc. Similarly, a proposition is expressed by a declarative (sentence), cf. [16].

Although we are used to use interrogative and question in the same meaning,

if necessary, the term interrogative sentence is reserved for a natural-language

sentence.

What seems to be common to all approaches is viewing questions as

something structured and connected with answers. The relationship

question — answer(s)

is a very conspicuous sign and the meaning of questions is closely connected

to answerhood conditions.

Since an answer to a question is often represented by a declarative, the

starting point of many erotetic theories is a formal system for declaratives.

“Any ﬁrst-order language can be supplemented with a question-and-answer

system” [44, p. 37]. This broadly accepted statement combines both the

formal shape and the meaning of a question. Questions’ autonomy de-

pends on the chosen solution. Wi´sniewski distinguishes two basic groups of

erotetic theories: reductionist and non-reductionist theories. Roughly speak-

ing, non-reductionism is characterized by questions that “are not reducible to

expressions of other syntactic categories” [44, p. 40]. The boundary between

both groups is vague. Perhaps only pure pragmatically oriented approaches

belong to the radical reductionism with a complete rejection of questions as

a speciﬁc entity. An example of such approach is commented in [30].

1.1.2 Inferences with questions

Although there is a discussion whether it is necessary to work with questions

as a new speciﬁc entity, almost all theorists agree that questions play a

speciﬁc role in inferences. Let us come back to our group of players. The

3

1.1. Questions, answers, and inferences

situation, where the Joker is held by a member of the group, can raise to the

question

Q: Who has the Joker?

from the declarative

Either Ann has the Joker or Bill has the Joker or Catherine has

the Joker.

What makes this raising reasonable are answerhood conditions of Q con-

nected to the declarative.

Another kind of inferential structure is based on declaratives as well as

questions among premises. For example, from

Q: Who has the Joker?

and

Γ: The only person from London has the Joker.

can be inferred the question

Q

1

: Who is from London?

The relationship of the inferred question Q

1

and the question Q is based on

their answerhood conditions again. An answer to Q can provide an answer

to Q

1

with respect to the context Γ. Moreover, in this example, Q can be

inferred from Q

1

and Γ as well. This shows that the relationship is structured

dependently on various kinds of answerhood conditions and contexts. Let us

have Q the same, but the context is

A person from London has the Joker.

If two persons are from London and we gain their names in an answer to Q

1

,

then we receive only a partial answer to Q.

2

If each player (or nobody) is

from London and an answer to Q

1

does not provide any help for the answering

of Q, it has no sense to speak of an inferential relation between Q and Q

1

with respect to this context.

The role of answerhood conditions in inferences among questions is clearly

obvious in the following example: From any (complete) answer to Q we gain

a (complete) answer to the question

2

Informally, a partial answer does not completely answer a question, but it eliminates

some of the possible answers.

4

1.2. Set-of-answers methodology

Has Ann the Joker?

as well as for the questions

Has Bill the Joker?

and

Has Catherine the Joker?

Roughly and informally speaking, answerhood conditions of the previous

three questions are ‘entailed’ in the answerhood conditions of the question

Q; they can be inferred from the answerhood conditions of Q. The question

Has Ann the Joker? is ‘entailed’ by Q.

Though we do not doubt that there are inference-like structures with

questions based on answerhood conditions, we have been still faced with the

problem how to formalize the relationship of questions and answers. We see

the convenient solution in a liberal set-of-answers methodology.

1.2 Set-of-answers methodology

In the previous section we emphasized the close connection of questions and

answers in most erotetic theories as well as in inferential structures with

questions. We believe that we can solve the problem of the formal shape

of questions together with the problem of the questions’ semantics. In this

section we introduce a formalization of questions based on a set of answers.

Our aim is to show that such approach can also reﬂect some semantic and

pragmatic requirements.

1.2.1 Semantics of questions

Some theories do not admit that questions could have an independent mean-

ing in logic. For example, questions are paraphrased by declarative sentences;

the question Who has the Joker? may be then paraphrased by

I ask you who has the Joker.

Another way is the paraphrasing by epistemic-imperative sentences:

Bring it about that I know who has the Joker!

5

1.2. Set-of-answers methodology

The propriety of both paraphrases as a complete meaning of a question is

rather problematic. Although we expect to utilize the importance of a ques-

tioner and an addressee later on, now it may be second-rate from the semantic

viewpoint.

Nuel Belnap formulated three methodological constraints on a theory

of questions, which he used for a classiﬁcation and evaluation of erotetic

theories:

3

1. Independence Interrogatives are entitled to a meaning of their own.

2. Equivalence Interrogatives and their embedded forms are to be tre-

ated on a par.

3. Answerhood The meaning of an interrogative resides in its answer-

hood conditions.

The most important is the ﬁrst requirement, which is the main sign of

non-reductionist theories. To accept ‘independence requirement’ means that

we are obliged to look for a speciﬁc semantics of questions. The ‘equivalence

requirement’ is closely related to a semantic entailment and is dependent on

the chosen semantics. The ‘answerhood’ requires that the meaning of ques-

tions is related to the meaning of answers. Moreover, we can work with the

idea that the semantics of answers forms a good background for the study of

the meaning of interrogatives. The approach, where answers are crucial for

the meaning of questions, is displayed in Hamblin’s postulates from 1958:

1. Knowing what counts as an answer is equivalent to knowing the ques-

tion.

2. An answer to a question is a statement.

3. The possible answers to a question are an exhaustive set of mutually

exclusive possibilities.

Each postulate may be argued against and the detailed discussion is available

in [14]. However, according to David Harrah, adopting the ﬁrst one is “the

giant step toward formalization often called set-of-answers methodology” [16,

section 2]. Although there are many kinds of set-of-answers methodology

(SAM, for short) in the literature, we will not make any survey here. In the

next subsection we introduce an easy idea of a question representation by a

set of direct answers.

3

Belnap, N.D., ‘Approaches to the semantics of questions in natural language. Part I’,

Pittsburgh, 1981. Cited from [13, p. 3–4].

6

1.2. Set-of-answers methodology

1.2.2 Sets of answers

Generally, without any context, the question Who has the Joker? can be

answered by expressions of the following form:

Ann.

Ann has it.

Ann has the Joker.

Ann and Bill.

.

.

.

Batman has the Joker.

.

.

.

Your friends.

People at this table.

.

.

.

Nobody.

etc.

The question seems to be answered if a (complete) list of Joker’s owners is

given. We can assume that answers are sentences; thus, the ﬁrst three items

in the list have the same meaning in the answering of this question. From

the viewpoint of propositional logic and in accordance with the ﬁrst two

Hamblin’s postulates, we can understand every question closely connected

with a set of (propositional) formulas—answers.

Of course, we can receive the following responses to the same question as

well:

Ann hasn’t the Joker.

or

I don’t know who has the Joker.

Neither of them answers completely the question Who has the Joker?. The

ﬁrst one can be considered to be a partial answer; it removes some answers

as impossible, e.g., all answers with Ann having the Joker. The second one

appears to bear another kind of information; an addressee says to a questioner

that she has the same problem and would ask the same question. We will

return to this topic shortly in Section 4.4.

If we had decided to represent every question by a complete set of its

answers, we would not have a clear and useful formalization of questions.

Let us return to our example. Considering the context, a questioner expects

one of the following responses to the question Who has the Joker?:

7

1.2. Set-of-answers methodology

α: Ann has the Joker.

β: Bill has the Joker.

γ: Catherine has the Joker.

In fact, the question might be reformulated to

Q

**: Who has the Joker: Ann, Bill, or Catherine?
**

The question Q

**is a combination of the question
**

Who has the Joker?

and the context

Either Ann has the Joker or Bill has the Joker or Catherine has

the Joker.

Of course, various responses to Q

**can be received again, but the ‘core’ an-
**

swers (the term direct answers will be used) are α, β, and γ. The sentence

δ: Neither Ann nor Bill have the Joker.

is an answer, from which γ is inferred thanks to the context; δ is a complete

answer to Q

**. Complete answers are ‘solutions’ of a question and the set of
**

direct answers is a subset of the set of complete ones.

Our SAM is inspired by the syntactic representation of questions in in-

ferential erotetic logic founded by Andrzej Wi´sniewski.

4

We want to be very

liberal and this leads us to considering questions to be sets of formulas,

which play the role of direct answers. A (general) declarative language L is

extended only by curly brackets (¦, ¦) and question mark (?).

A question is the following structure

?¦α

1

, α

2

, . . .¦

where α

1

, α

2

, . . . are formulas of the extended language.

No wonder that we want to impose some restrictions on direct answers

to keep their exclusive position. Such restrictions are mostly combination

of syntactic as well as semantic requirements. From the syntactic viewpoint

and being inspired by the previous examples, we require:

1. Formulas α

1

, α

2

, . . . are syntactically distinct.

2. A set of direct answers has at least two elements.

4

The best overview of questions’ formalization in inferential erotetic logic is in the

chapter 3 of the book [44]. See the article [46] as well.

8

1.2. Set-of-answers methodology

Both restrictions introduce questions as ‘tasks’ with at least two distinct

‘solutions’. Syntactical distinctness is a ﬁrst step to the idea that direct

answers form the ‘core’ of questions’ meaning. In semantics we will require

non-equivalence above that.

The most typical questions with only two direct answers are yes-no ques-

tions. The question

Has Ann the Joker?

has, in fact, the following two direct answers:

Yes. (Ann has the Joker.)

No. (Ann has not the Joker.)

Such question will be identiﬁed with the form ?¦α, α¦ and shortened as

?α. A yes-no question is a variant of a whether question, where an answer

is a choice from two possibilities.

5

The role of negation is considered to be

very important in SAM. Negation is always related to a background sys-

tem and receiving α can mean more than ‘it is not the case that α’—it

expresses something like ‘strict denial of α’. Compare it with an epistemic

interpretation of relevant logic in [22].

We believe that the introduced SAM is more or less successful in for-

malization of most types of natural language interrogatives with respect to

the chosen background logical system. However, it brings more—the next

subsection informally shows how to incorporate epistemic aspects.

1.2.3 Epistemic aspects of SAM

In section 1.2.1 we used the paraphrase

Bring it about that I know who has the Joker!

for the question Who has the Joker? as it is common in Hintikka’s analysis

[17]. It is natural to see epistemic aspects in the meaning of questions. At

ﬁrst sight, a question expresses an ignorance of a questioner delivered to an

addressee. Looking for an epistemic counterpart of questions in the history of

erotetic logic, the most known are epistemic-imperative approaches of

˚

Aqvist

and Hintikka. These theories are reductionist ones, questions are translated

into epistemic-imperative statements. Generally, both approaches are based

on the idea of a questioner who does not know any answer to a question and

who calls for a completion of knowledge.

6

5

For example, Is two even or odd? is such a question with a possible formalization

?¦α

1

, α

2

¦.

6

More details are in [16] and [44, chapter 2].

9

1.3. Note on the recent history of erotetic logic

Epistemic analysis of questions has two important levels. The ﬁrst one,

personal level, works with the knowledge and ignorance of a questioner. The

second one, pragmatic level, considers an exchange of information in a group

of agents, which is supposed to be used in a theory of communication.

Let us return to the question

Q

**: Who has the Joker: Ann, Bill, or Catherine?
**

with the formalization based on the discussion in the previous section: Q

=

?¦α, β, γ¦. Just introduced SAM makes it possible to specify expectations

and presuppositions of a questioner. A questioner expresses not only the

ignorance of Joker’s holder, but the presupposition that the holder must be

either Ann or Bill or Catherine. In the personal level, our SAM informs us

what answers are considered as possible and, moreover, what is the rank of

complete answers.

To employ the pragmatic level we have to indicate a questioner and an

addressee. In particular, if Catherine wants to ﬁnd out who has the Joker

in the group of her friends-players, she could ask ?

c

¦α, β¦ publicly among

them. This will be studied in chapters 3 and 4.

1.3 Note on the recent history of erotetic lo-

gic

We are not going to present a complete survey of erotetic theories. The

reader can ﬁnd a comprehensive overview of the history of erotetic logic in

[16]. Erotetic theories with the main inﬂuence in this ﬁeld of study are

described in [44, chapter 2] as well. Moreover, both papers provide a good

introduction to the terminology used in logic of questions and cover enough

the history of erotetic logic till the 1990s. The period from the 1950s till 1990s

is mapped in [15]. Mainly linguistic viewpoint with the detailed discussion

about the semantics of questions and pragmatic approaches can be found in

[14].

The logic of questions has, maybe surprisingly, a long history. F. Cohen

and R. Carnap seem to be the ﬁrst authors attempting to formalize questions

in a logical framework—their attempts date back to the 1920s [16, p. 3]. The

ﬁrst ‘boom’ of logical approach to questions took place in the 1950s (Ham-

blin, Prior, Stahl) and continued in the 1960s (

˚

Aqvist, Harrah, Kubi´ nski).

The ﬁrst comprehensive monograph on questions [1] brought into life many

important terms used in erotetic logic so far. The late 1970s gave birth to in-

ﬂuential reductionist theories: Hintikka’s epistemic-imperative approach and

Tich´ y’s approach based on his transparent intensional logic [37].

10

1.3. Note on the recent history of erotetic logic

Most inﬂuential modern logics of questions with the important role of

erotetic inferences are

• intensional approach of Jeroen Groenendijk and Martin Stokhof and

• inferential erotetic logic (IEL) of Andrzej Wi´sniewski.

Both theories appears fully developed in the 1990s and we consider them

giving birth to several approaches some of which are still inﬂuential.

1.3.1 Inferential erotetic logic

Wi´sniewski’s IEL is a complex system dealing with various interrogative in-

ferential structures. The inﬂuence of Belnap’s and Kubi´ nski’s works is appar-

ent. Primarily it is based on classical logic and a formalization of questions,

which is very similar to the introduced SAM. Conclusion relations among

questions and declaratives are deﬁned on the metalanguage level where the

role of multiple-conclusion entailment is important. The advantage of IEL

is its possible generalization for non-classical logics. Chapter 2 is devoted

to the slightly modiﬁed IEL. All important terms concerning interrogative

inferences are introduced and studied in their mutual relationships. In this

chapter the reader can ﬁnd a list of relevant publications to this topic. The

book [44] and article [46] contain a nicely written presentation of Wi´sniewski’s

approach.

The complex study of erotetic inferential structures in IEL predetermines

studies based on an old idea that a (principal) question can be answered

by asking auxiliary questions. This searching process can be seen as a tree

with a principal question (and context expressed by declaratives) in the root.

Nodes bear auxiliary questions (with context). A move from node to node is

justiﬁed by IEL inferences in the direction to leaves with answers. Erotetic

search scenarios is the name for this approach, see the paper [47]. The

similar idea is developed in Hintikka’s interrogative model of inquiry reﬂecting

the usefulness of questions in reasoning [18]. Hintikka’s approach has an

application in game-theoretic framework for belief revision theory, cf. [10].

7

The IEL methodology makes it possible to transform the derivability of a

declarative formula into a sequence of questions and produce an analytic-tab-

leaux style calculus—socratic proofs, see [49] for classical propositional logic

and [21] for some normal modal propositional logics.

7

IEL used to be presented as an alternative to Hintikka’s approach (cf. [46, 47]).

11

1.3. Note on the recent history of erotetic logic

1.3.2 Intensional erotetic logic

Groenendijk’s and Stokhof’s approach can be called intensional erotetic logic.

Intensional semantics is the background of the meaning of questions. The

meaning of a declarative sentence is given by truth conditions and forms a

subset of logical space. Logical space is understood as a set of all ‘possible

states’ (possible worlds, indexes, situations). Intension of a declarative is

then a set of states where the declarative is true (this is called proposition).

Extension of a declarative is its truth value in a given situation. Combin-

ing intensional semantics with the full acceptance of Hamblin’s postulates

we obtain the meaning of a question as a partitioning of logical space. In

accordance with the third postulate, answers to a question form exhaustive

set of mutually exclusive propositions. Partitioning of logical space is the

intension of a question. Extension of a question in a given situation is the

answer, which is true there. Questions’ representation is similar to SAM

with restrictions posed by Hamblin’s postulates. This approach introduces

an entailment between two questions as a reﬁnement of a partitioning:

A question Q entails a question Q

1

iﬀ each answer to Q implies

an answer to Q

1

.

(A question Q provides a reﬁnement of the Q

1

-partitioning.) Moreover, many

important terms are naturally deﬁned (partial answer, complete answer, in-

formative value of answers, and others). See [13] and [14].

This intensional approach inﬂuenced many works in the last ten years. It

is a good inspiration for epistemic representations of questions, see [6] and

[29]. Very recently the logic of questions receives more attention in connection

with dynamic aspects of epistemic logic and communication theory, cf. [40].

Groenendijk’s and Stokhof’s intensional interpretation inspired some ex-

tensional approaches. [27] brings a many-valued interpretation of declaratives

and interrogatives based on bilattices and in the paper [25] Gentzen style cal-

culus is presented.

8

[34] gives a syntactic characterization of answerhood for

the partition semantics of questions and then the authors implement parti-

tion semantics in question answering algorithm based on tableaux theorem

proving [35].

9

Presented logics of questions deal with questions as having ‘crisp’ answers.

However, we can imagine that there is a scale of answers, e.g., in case of yes-no

questions, the scale

yes — rather yes — rather no — no

8

An algebraic approach, where sets of answers form distributive lattices, is studied also

in [19].

9

Nice and brief comments are in [12].

12

1.3. Note on the recent history of erotetic logic

is usual in questionnaires. Such kind of scale does not require to intro-

duce four new answers, but it corresponds to comparative degrees of truth

of the original yes-/no-answer. Truth-degrees are studied in multi-valued

logics. The paper [4] presents propositional Groenendijk-Stokhof’s erotetic

logic with fuzzy intensional semantics based on fuzzy class theory. Although

fuzzy logic seems to be suitable for the study of reasoning under vagueness,

its combination with logic of questions is still rather underdeveloped.

13

Chapter 2

Consequence relations in

inferential erotetic logic

2.1 Introduction

Inferential structures that will be introduced and studied in this chapter

are based on the slightly adapted inferential erotetic logic (IEL). We utilize

the framework of this theory and show some properties, relationships, and

possible generalizations.

2.1.1 Adapted set-of-answers methodology in IEL

Inferential erotetic logic accepts only the ﬁrst two Hamblin’s postulates and

tries to keep the maximum of the (classical) declarative logic and its conse-

quence relation. On the syntactic level of a considered formalized language, a

question is assigned to a set of sentences (direct answers). Direct answers are

declarative formulas, each question has at least two direct answers, and each

ﬁnite and at least two-element set of sentences is the set of direct answers to

some question [46, p. 11].

Let us apply our SAM introduced in the previous chapter. We deﬁne

a (general) erotetic language L

Q

. A (general) declarative language L is ex-

tended by curly brackets (¦, ¦) and question mark (?).

1

In the correspondence

with Section1.2 a question Q is the following structure

?¦α

1

, α

2

, . . .¦

For the set of direct answers of a question Q we will use the symbol dQ.

1

In this chapter we use only propositional examples in the language with common

connectives (∧, ∨, →, ).

14

2.1. Introduction

Let us repeat that direct answers are syntactically distinct and [dQ[ ≥ 2.

Moreover, we require here that elements of dQ are declarative sentences.

In case of ﬁnite versions of questions ?¦α

1

, . . . , α

n

¦ we suppose that the

listed direct answers are semantically non-equivalent. The class of ﬁnite

questions corresponds to the class of questions of the ﬁrst kind in [44]. Some

of them are important in our future examples and counterexamples. Let us

mention two abbreviations and terms that are very frequent in this thesis:

• Simple yes-no questions are of the form ?α, which is an abbreviation

for ?¦α, α¦. If α is an atomic formula, then the term atomic yes-no

question is used.

• A conjunctive question ?[α, β[ requires the answer whether α (and not

β), or β (and not α), or neither α nor β, or both (α and β). It is

an abbreviation for ?¦(α ∧ β), (α ∧ β), (α ∧ β), (α ∧ β)¦. Similar

versions are ?[α, β, γ[, ?[α, β, γ, δ[, and so on.

2

In the original version of IEL, questions are not identiﬁed with sets of

direct answers: questions belong to an object-level language and are expres-

sions of a strictly deﬁned form, but the form is designed in such a way that,

on the metalanguage level (and only here), the expression which occurs af-

ter the question mark designates the set of direct answers to the question.

Questions are deﬁned in such a way that sets of direct answers to them are

explicitly speciﬁed. The general framework of IEL allows for other ways of

formalizing questions.

3

To avoid a misunderstanding, we will use the following metavariables in

this chapter:

• small Greek letters (α, β, ϕ, . . .) for declarative sentences,

• Q, Q

1

, . . . for questions,

• capital Greek letters (Γ, ∆, . . .) for sets of declaratives, and

• Φ, Φ

1

, . . . for sets of questions.

2.1.2 Consequence relations in IEL

Consequence relations are the central point of logic. Declarative logic can be

deﬁned by its consequence relation as a set of pairs 'Γ, ∆`, where Γ and ∆ are

2

Original IEL uses the symbol ?±[α, β[, etc.

3

Personal communication with Andrzej Wi´sniewski.

15

2.1. Introduction

sets of (declarative) formulas and ∆ is usually considered to be a singleton.

Inferential erotetic logic makes one step more and adds new consequence re-

lations mixing declaratives and interrogatives. The most important relations,

which we are going to introduce, are the following:

• Evocation is a binary relation 'Γ, Q` between a set of declaratives Γ

and a question.

• Erotetic implication is a ternary relation 'Q, Γ, Q

1

` between an initial

question Q and an implied question Q

1

with respect to a set of declar-

atives Γ.

• Reducibility is a ternary relation 'Q, Γ, Φ` between an initial question

Q and a set of questions Φ with respect to a set of declaratives Γ.

Motivations and natural-language examples of these consequence relations

will be introduced in the next sections. In the literature, let us recommend

texts [44, 46] for both the evocation and erotetic implication; reducibility is

studied in [20, 44, 50].

Our aim is to study erotetic consequence relations in a very general man-

ner, independently of the logic behind. The deﬁnitions of IEL consequences

are based on the semantic entailment and the model approach relative to the

chosen logical background.

2.1.3 Model-based approach

The following model-based approach was inspired by minimal erotetic se-

mantics from [46]. Let us introduce the set of all models for a declarative

language as follows:

´

L

= ¦M

**M is a (semantic) model for L¦.
**

The term model varies dependently on a background logic L. If L is classical

propositional logic (CPL, for short), then ´

CPL

is a set of all valuations.

In case of predicate logic it is a set of all structures with a realizations of

non-logical symbols. Because of the possibility of adding some other con-

straints for models we will deal with (e.g., ﬁniteness, preferred models, etc.),

let us generally use a set ´ ⊆ ´

L

. If necessary, the background logic and

restrictions posed on models will be stated explicitly.

Speaking about tautologies of a logic L we mean the set of formulas

Taut

L

= ¦ϕ

(∀M ∈ ´

L

)(M [= ϕ)¦.

16

2.1. Introduction

If a restricted set of models ´ is in use, we speak about ´-tautologies

Taut

M

L

= ¦ϕ

(∀M ∈ ´)(M [= ϕ)¦.

All semantic terms may be relativized to ´. Each declarative sentence

ϕ (in the language L) has its (restricted) set of models

´

ϕ

= ¦M ∈ ´

M [= ϕ¦

and similarly for a set of sentences Γ

´

Γ

= ¦M ∈ ´

(∀γ ∈ Γ)(M [= γ)¦.

(Semantic) entailment

Let us recall the common (semantic) entailment relation. For any set of

formulas Γ and any formula ψ:

Γ [= ψ iﬀ ´

Γ

⊆ ´

ψ

.

In case Γ = ¦ϕ¦ we write only ϕ [= ψ.

ϕ [= ψ iﬀ ´

ϕ

⊆ ´

ψ

Now, we introduce multiple-conclusion entailment (mc-entailment, for

short).

Γ [[= ∆ iﬀ ´

Γ

⊆

¸

δ∈∆

´

δ

If ´

Γ

= ´

∆

, let us write Γ ≡ ∆.

4

Mc-entailment is reﬂexive (Γ [[= Γ), but it is neither symmetric nor tran-

sitive relation:

Example 1. Let Γ ⊆ Taut

L

, ∆ be a set of sentences containing at least one

tautology and at least one contradiction, and Σ be such that

¸

σ∈Σ

´

σ

⊂ ´

L

.

Then Γ [[= ∆ and ∆ [[= Σ, but Γ [[= Σ.

Entailment is deﬁnable by mc-entailment:

Γ [= ϕ iﬀ Γ [[= ¦ϕ¦

On the other hand, mc-entailment is not deﬁnable by entailment. In this

context, the following theorem could be surprising at the ﬁrst sight.

5

4

In case of the semantic equivalence of formulas ϕ and ψ it will be only written ϕ ≡ ψ.

On the other hand, two diﬀerent sets of models do not imply the existence of two diﬀerent

sets of sentences (in L).

5

We say that mc-entailment is compact iﬀ for each Γ [[= ∆ there are ﬁnite subsets

G ⊆ Γ and D ⊆ ∆ such that G [[= D.

17

2.1. Introduction

Theorem 1. Entailment (for logic L) is compact iﬀ mc-entailment (for L)

is compact.

Proof. See [44, pp. 109–110].

2.1.4 Basic properties of questions

After we have introduced the SAM representation of questions and the mo-

del-based approach, we can mention some basic properties of questions. First,

let us introduce the term soundness, which is one of the most important terms

in IEL.

Deﬁnition 1. A question Q is sound in M iﬀ ∃α ∈ dQ such that M [= α.

A question is sound with respect to a model M whenever it has at least

one direct answer true in M. See [44, p. 113].

For all IEL consequnce relations, it is important to state the soundness

of a question with respect to a set of declaratives.

Deﬁnition 2. A question Q is sound relative to Γ iﬀ Γ [[= dQ.

The sum of all classes of models of each direct answer α, i.e.,

¸

α∈dQ

´

α

,

is called semantic range of a question Q. Considering semantic range, our

liberal approach admits some strange questions; one of them is a completely

contradictory question that has only contradictions in its set of direct an-

swers, its semantic range beeing just ∅. Another type is a question with a

tautology among its direct answers, then the semantic range expands to the

whole ´. Questions with such a range are called safe.

6

Of course, it need

not be any tautology among direct answers for to be a safe question.

Deﬁnition 3. • A question Q is safe iﬀ

¸

α∈dQ

´

α

= ´.

• A question Q is risky iﬀ

¸

α∈dQ

´

α

⊂ ´.

Questions ?α, ?[α, β[ are safe in CPL, but neither is safe in Bochvar logic.

If β is not equivalent to α, then ?¦α, β¦ is risky in CPL. Neither ?α nor

?[α, β[ are safe in intuitionistic logic, but there are safe questions in this logic;

just each question with at least one tautology among direct answers. Simple

yes-no questions are safe in logics that accept the law of excluded middle.

It is good to emphasize that the set of direct answers of a safe question

is mc-entailed by every set of declaratives. On the other hand, knowing a

question to be sound relative to every set of declaratives implies its safeness.

Fact 1. Q is safe iﬀ (∀Γ)(Γ [[= dQ).

Specially, safe questions are sound relative to Γ = ∅.

6

This term originates from Nuel Belnap.

18

2.2. Questions and declaratives

2.1.5 The road we are going to take

After introducing evocation and the term presupposition of a question in

section 2.2, we will show the role of maximal and prospective presuppositions

in the relationship to semantic range of questions. Some classes of questions

will be based on it. One could be surprised that we are not going to discuss

answers in this section; in fact, there is not much to say about them. It

turns out that various types of answers do not play any special role in the

inferential structures.

Section 2.3 is crucial from the chosen viewpoint. We investigate erotetic

implication and reducibility there. An important part is devoted to a dis-

cussion of the role of an auxiliary set of declaratives. We will demonstrate

some variants of erotetic implication and their properties. The chosen formal

shape of questions in IEL makes it possible to compare questions in the sense

of their answerhood power. Inspired by [13] and [44, section 5.2.3] we will

examine the relationship of ‘giving an answer’ of one question to another,

which is a generalisation of Kubi´ nski’s term ‘weaker question’.

Questions will be considered as independent structures not being com-

bined by logical connectives. Reviewing the deﬁnitions of erotetic implication

and reducibility we can recognize their ‘both-sidedness’ and just reducibility

can substitute such combination of questions.

This brings us to the last note on the use of symbols [= and ¬. Because of

the clear border between declarative and interrogative parts of the language

L

Q

we will use them in many meanings. However, the meaning will be

transparent by the context the symbols are used in. Compare the deﬁnition of

evocation and various deﬁnitions of erotetic implications in the next sections.

2.2 Questions and declaratives

In this section, we introduce two terms: evocation and presupposition. The

ﬁrst one will provide a consequence relation between a set of declaratives

and a question. The second one is an important term in almost all logics of

questions and there are some classes of questions based on it in IEL.

2.2.1 Evocation

Consider the following example: after a lecture, we expect a lecturer to be

ready to answer some questions that were evoked by his or her talk. Thus,

evocation seems to be the most obvious relationship among declarative sen-

tences and questions. (Of course, next to the connection question—answer.)

19

2.2. Questions and declaratives

Almost every information can give rise to a question. What is the aim of

such a question?

First, it should complete our knowledge in some direction. Asking a

question we want to get more then by the conclusion based on a background

knowledge. A question Q should be informative relative to Γ, it means, there

is no direct answer to Q which is a conclusion of Γ.

Second, after answering an evoked question, no matter how, the answer

must be consistent with the evoking knowledge. Moreover, transmission of

truth into soundness is required: if an evoking set of declaratives has a model,

there must be at least one direct answer of the evoked question that is true

in this model. An evoked question should be sound relative to an evoking

set of declaratives (see Deﬁnition 2).

7

The deﬁnition of evocation is based on the previous two points (cf. [44,

46]). A question Q is evoked by a set of declaratives Γ if Q is sound and

informative relative to Γ.

Deﬁnition 4. A set of declarative sentences Γ evokes a question Q (let us

write Γ [= Q) iﬀ

1. Γ [[= dQ,

2. (∀α ∈ dQ)(Γ [= α).

In our model-based approach we can rewrite both conditions this way:

1. ´

Γ

⊆

¸

α∈dQ

´

α

2. (∀α ∈ dQ)(´

Γ

⊆ ´

α

)

In some special cases (e.g., dQ is ﬁnite or entailment is compact) we can

deﬁne evocation without the link to mc-entailment. The ﬁrst condition is of

the form: there are α

1

, . . . , α

n

∈ dQ such that Γ [=

n

1

α

i

.

This is the case of one of our introductory examples. Let us remind the

group of three card players. The context

Γ: Either Ann has the Joker or Bill has the Joker or Catherine

has the Joker.

evokes the question

Q: Who has the Joker: Ann, Bill, or Catherine?

7

For now, as we do not discuss epistemic issues, we shall not use the word ‘knowledge’

but the phrases ‘set of declarative(s) (sentences)’ or ‘database’ will be used instead.

20

2.2. Questions and declaratives

with direct answers

α: Ann has the Joker.

β: Bill has the Joker.

γ: Catherine has the Joker.

Let Γ consist of one formula in the form of disjunction of direct answers of

Q, thus, the ﬁrst condition is satisﬁed. The second one is satisﬁed because

no direct answer is entailed by Γ.

Evocation yields some clear and useful properties of both a set of declar-

atives and an evoked question. The following fact lists some of them.

Fact 2. If Γ [= Q, then

• Γ is not a contradictory set,

• there is no tautology in dQ, and

• Q is not a completely contradictory question.

However, by Fact 1, we obtain a less intuitive conclusion: every safe

question is evoked by any Γ that does not entail any direct answer to it. It

underlines the special position of safe questions and their semantic range.

When we restrict the deﬁnition of evocation to risky questions only, we get

the deﬁnition of generation, see [44, chapter 6].

Generation does not solve all problems with irrelevant and ineﬃcient

evoked questions either. We can accept another restriction to avoid ques-

tions that have direct answers which are incompatible with declaratives in Γ.

Borrowing an example from [7], Γ = ¦α ∨ β, γ¦ evokes also ?¦α, β, γ¦. To

eliminate this, the consistency of each direct answer with respect to Γ could

be required, i.e., we could add the third condition to Deﬁnition 4:

(∀α ∈ dQ)(´

Γ

∩ ´

α

= ∅)

Some solutions of the problem of irrelevant and ineﬃcient questions based

on a semantics in the background are discussed in the just mentioned pa-

per [7]. For our purpose, the study of consequence relations in IEL, we keep

Deﬁnition 4 unchanged.

Back to safe questions, let us mention the following fact:

Fact 3. If ∅ [= Q, then Q is safe.

As a conclusion of semantic deﬁnition of evocation we have the following

expected behavior of evocation: semantically equivalent databases evoke the

same questions.

21

2.2. Questions and declaratives

Fact 4. For every Γ, ∆ and Q, if Γ ≡ ∆, then Γ [= Q iﬀ ∆ [= Q.

If Γ evokes Q, then we have to be careful of concluding that there is a

subset ∆ ⊆ Γ such that ∆ evokes Q, see the ﬁrst item in the following fact.

Fact 5. If Γ [= Q and ∆ ⊆ Γ ⊆ Σ, then

• ∆ [= Q if ∆ [[= dQ,

• Σ [= Q if (∀α ∈ dQ)(Σ [= α).

The second item points out the non-monotonicity of evocation (in declar-

atives). Considering questions as sets of answers, evocation is non-monotonic

in interrogatives as well, see section 2.3.3.

Fact 6. If Γ [= Q and the entailment is compact, then ∆ [= Q

1

for some

ﬁnite subset dQ

1

of dQ and some ﬁnite subset ∆ of Γ.

These and some more properties of evocation (and generation) are dis-

cussed in the book [44].

2.2.2 Presuppositions

Many properties of questions are based on the concept of presupposition.

Everyone who has attended a basic course of research methods in social

sciences has heard of importance to consider presuppositions of a question in

questionaries.

If we receive the question Who has the Joker: Ann, Bill, or Catherine?,

we can recognize that it is presupposed that Ann has it or Bill has it or

Catherine has it. What is presupposed must be valid under each answer to a

question. Moreover, an answer to a question should bring at least the same

information as presupposition does. The following deﬁnition (originally given

by Nuel Belnap) is from [44]:

Deﬁnition 5. A declarative formula ϕ is a presupposition of a question Q

iﬀ (∀α ∈ dQ)(α [= ϕ).

A presupposition of a question is entailed by each direct answer to the

question. Let us write PresQ for the set of all presuppositions of Q.

At the ﬁrst sight, the set PresQ could contain a lot of sentences. Let

us have a question Q = ?¦α

1

, α

2

¦, the set of presuppositions (e.g., in CPL)

contains (α

1

∨ α

2

), (α

1

∨ α

2

∨ ϕ), (α

1

∨ α

2

∨ ϕ), etc. Looking at the very

relevant member (α

1

∨ α

2

) it is useful to introduce the concept of maximal

presupposition. Formula (α

1

∨α

2

) entails each presupposition of the question

Q.

22

2.2. Questions and declaratives

Deﬁnition 6. A declarative formula ϕ is a maximal presupposition of a

question Q iﬀ ϕ ∈ PresQ and (∀ψ ∈ PresQ)(ϕ [= ψ).

The model-theoretical view shows it in a direct way. The deﬁnition of

presupposition gives

¸

α∈dQ

´

α

⊆ ´

ϕ

, for each ϕ ∈ PresQ, which means

¸

α∈dQ

´

α

⊆

¸

ϕ∈PresQ

´

ϕ

= ´

PresQ

and the set ´

PresQ

is a model-based counterpart to the deﬁnition of maximal

presuppositions.

If the background logic has tautologies, each of them is in PresQ.

Taut

M

L

⊆ PresQ

Considering safe questions we get

Fact 7. If Q is safe, then PresQ = Taut

M

L

.

This fact says that if Q is safe, then

¸

α∈dQ

´

α

= ´

PresQ

. In classical

propositional logic the disjunction of all direct answers of a question is a

presupposition of this question and if PresQ = Taut

M

CPL

, then Q is safe.

This evokes a (meta)question whether the implication from right to left is

valid. If Q is not safe, then we know that

¸

α∈dQ

´

α

is a proper subset of

´. But what about ´

PresQ

? After introducing a class of normal questions

(see page 25) it will be valid ´

PresQ

⊂ ´ as well as the implication from

right to left (see Fact 9).

A presupposition can be seen as an information which is announced by

asking a question, without answering it. Such information is relatively small.

The semantic range of all maximal presuppositions is wider than the range

of a question. Looking at ﬁnite CPL example where the disjunction of all

direct answers forms just the semantic range of the question brings us to the

idea of prospective presupposition. It is a presupposition which a question Q

is sound relative to.

Deﬁnition 7. A declarative formula ϕ is a prospective presupposition of a

question Q iﬀ ϕ ∈ PresQ and ϕ [[= dQ. Let us write ϕ ∈ PPresQ.

All prospective presuppositions of a question are equivalent:

Lemma 1. If ϕ, ψ ∈ PPresQ, then ϕ ≡ ψ.

Proof. If M [= ϕ, then there is α ∈ dQ such that M [= α. Since ψ ∈ PresQ,

α [= ψ and it gives M [= ψ. We got ϕ [= ψ.

ψ [= ϕ is proved by the same way.

23

2.2. Questions and declaratives

A prospective presupposition forms exactly the semantic range of a ques-

tion.

¸

α∈dQ

´

α

= ´

PPresQ

If Q has a prospective presupposition, it can be understood as the ‘strongest’

one.

Two questions with the same sets of presuppositions have the same pros-

pective presuppositions.

Lemma 2. If PresQ = PresQ

1

and both PPresQ and PPresQ

1

are not empty,

then PPresQ = PPresQ

1

.

Proof. We show that if ϕ ∈ PPresQ and ψ ∈ PPresQ

1

, then ϕ ≡ ψ.

ϕ ∈ PPresQ implies ´

ϕ

=

¸

α∈dQ

´

α

and

¸

α∈dQ

´

α

⊆ ´

ψ

, because

ψ ∈ PresQ. It gives ´

ϕ

⊆ ´

ψ

and ϕ [= ψ.

The proof that ψ [= ϕ is similar.

Presuppositions of evoked questions are entailed by the evoking set of

declaratives.

Fact 8. If Γ [= Q, then Γ [= ϕ, for each ϕ ∈ PresQ.

The implication from right to left does not hold. If we only know ´

Γ

⊆

´

PresQ

, we are not sure about ´

Γ

⊆

¸

α∈dQ

´

α

as required by the ﬁrst

condition of evocation. Clearly, the informativeness must be ensured as well.

Let us note that it cannot be improved by replacing of PresQ by PPresQ.

We will return to this in the next subsection at the topic of normal ques-

tions. To sum up all general conditions of an evoked question (by Γ) and its

presuppositions let us look at this diagram:

´

Γ

⊆

¸

α∈dQ

´

α

= ´

PPresQ

⊆ ´

PresQ

Classes of questions based on presuppositions

Using the term presupposition we can deﬁne some classes of questions. Names

and deﬁnitions of the classes are from [44]. We only add the model-based ap-

proach and make transparent some results on presuppositions and evocations

(chapters 4 and 5 in [44]).

24

2.2. Questions and declaratives

Normal questions A question Q is called normal if it is sound relative to

its set of presuppositions (PresQ [[= dQ).

• Q ∈ normal iﬀ

¸

α∈dQ

´

α

= ´

PresQ

Model-based approach introduces normal questions as questions with seman-

tic range delimitated by models of maximal presuppositions. Working with

ﬁnite sets of direct answers and in logical systems with the ‘classical’ be-

havior of disjunction (each direct answer entails the disjunction of all direct

answers) we do not leave the class normal. Non-normal questions can be

found in classical predicate logic.

Two announced facts follow. They continue on the discussions at Fact 7

and Fact 8.

Fact 9. If PresQ = Taut

M

L

and Q is normal, then Q is safe.

Let us only add the clear fact, that the class of safe questions is a subset

of the class of normal questions.

safe ⊆ normal

The following fact and Fact 8 give the conditions for evocation of normal

questions.

8

Fact 10. If Γ [= ϕ, for each ϕ ∈ PresQ, and Γ [= α, for each α ∈ dQ of a

normal question Q, then Γ [= Q.

Regular questions Each question with the non-empty set of prospective

presuppositions is regular.

• Q ∈ regular iﬀ (∃ϕ ∈ PresQ)(ϕ [[= dQ)

Regularity of Q gives ´

PresQ

⊆ ´

ϕ

⊆

¸

α∈dQ

´

α

and it holds

regular ⊆ normal

If entailment is compact, both classes are equal.

Normal questions are sound relative to PresQ and regular questions are

sound relative to PPresQ. The following example shows an expected fact

that it is still not suﬃcient for evocation.

Example 2 (in CPL). Let Q = ?¦(α ∨ β), α¦. This question is normal

and regular, the formula (α ∨ β) is a prospective presupposition of Q, but

PresQ [= Q.

8

Cf. Theorem 5.23 in [44].

25

2.2. Questions and declaratives

If there is a set of declaratives Γ such that Γ [= Q, then normal (regular)

questions are sound as well as informative relative to PresQ (PPresQ). This

is summed up by

Lemma 3. Let Γ [= Q, for some set of declaratives Γ. Then

1. Q ∈ normal implies PresQ [= Q.

2. Q ∈ regular implies ϕ [= Q, for ϕ ∈ PPresQ.

Proof. For the ﬁrst item, only informativeness (relative to PresQ) must be

showed. But if it is not valid, then Fact 8 causes non-informativeness of Q

relative to Γ.

The second item is provable by the same idea.

Self-rhetorical questions Another special class of questions are self-rhe-

torical questions. They have at least one direct answer entailed by the set of

presuppositions.

• Q ∈ self-rhetorical iﬀ (∃α ∈ dQ)(PresQ [= α)

From this deﬁnition, it is clear that self-rhetorical questions are normal. How-

ever, do we ask such questions? This class includes such strange questions as

completely contradictory questions that have only contradictions in the set

of direct answers, and questions with tautologies among direct answers.

An evoked question is not of this kind.

Lemma 4. If there is Γ such that Γ [= Q, then Q is not self-rhetorical.

Proof. From Fact 8.

Proper questions Normal and not self-rhetorical questions are called pro-

per. Proper questions are evoked by their set of presuppositions.

• Q ∈ proper iﬀ PresQ [= Q

Evoked normal questions are proper (compare both Lemma 3 and Lemma 4)

and this makes the class proper prominent. The set of all presuppositions

of a question Q is believed to be a natural (declarative) context for evocation

of Q.

26

2.3. Questions and questions

2.3 Questions and questions

This section is devoted to inferential structures in which questions appear

on both sides (erotetic implication and reducibility of questions to sets of

questions) and to relations between two questions based on their sets of

direct answers. The second point focuses on ‘answerhood power’ of questions

formalized by the adapted set-of-answers methodology.

2.3.1 Erotetic implication

Now, we extend the class of inferences by ‘implication’ between two questions

with a possible assistance of some set of declaratives. Let us start with an

easy and a bit tricky example. If I ask

Q: What is Peter a graduate of: a faculty of law or a faculty of

economy?

then I can be satisﬁed by the answer

He is a lawyer.

even if I did not ask

Q

1

: What is Peter: a lawyer or an economist?

The connection between both questions could be shown by the following set

of declaratives:

Someone is a graduate of a faculty of law iﬀ he/she is a lawyer.

Someone is a graduate of a faculty of economy iﬀ he/she is an

economist.

The ﬁrst question Q can be formalized by ?¦α

1

, α

2

¦ and the latter one,

speaking of Peter’s position, can be ?¦β

1

, β

2

¦. Looking at the questions

there is no connection between them. The relationship is based on the set of

declaratives Γ = ¦(α

1

↔ β

1

), (α

2

↔ β

2

)¦. Now, we say that Q implies Q

1

on

the basis of Γ and write Γ, Q [= Q

1

.

This relation is called erotetic implication (e-implication, for short) and

the following deﬁnition is from [44]:

9

Deﬁnition 8. A question Q implies a question Q

1

on the basis of a set of

declaratives Γ iﬀ

9

We will write shortly Γ ∪ ϕ instead of Γ ∪ ¦ϕ¦.

27

2.3. Questions and questions

1. (∀α ∈ dQ)(Γ ∪ α [[= dQ

1

),

2. (∀β ∈ dQ

1

)(∃∆ ⊂ dQ)(∆ = ∅ and Γ ∪ β [[= ∆).

Returning to the introductory example, both questions are even eroteti-

cally equivalent with respect to Γ: Γ, Q [= Q

1

as well as Γ, Q

1

[= Q.

The deﬁnition requires a little comment. The ﬁrst clause should express

the soundness of an implied question relative to each extension of Γ by α ∈

dQ. This transmission of truth/soundness into soundness has the following

meaning: if there is a model of Γ and a direct answer to Q, then there must

be a direct answer to Q

1

that is valid in this model. If Q

1

is safe, then this

condition is always valid (see Fact 1).

The second clause requires direct answers to Q

1

to be cognitively useful

in restricting the set of direct answers of the implying question Q.

In comparison with evocation, the role of the set of declaratives is a bit

diﬀerent. Γ plays, especially, the auxiliary role; e-implication is monotonic

in declaratives and it gives the following [44, p. 173]:

Fact 11. If Γ, Q [= Q

1

, then ∆, Γ, Q [= Q

1

, for any set of declaratives ∆.

This could be called weakening in declaratives. From this, it is clear that

⊥, Q [= Q

1

, for each Q and Q

1

.

We wil say a word or two about auxiliary sets of declaratives in the next

subsection.

Pure erotetic implication

Pure e-implication is e-implication with the empty set of declaratives. In

our semantic approach, Γ includes only tautologies of a chosen logical sys-

tem. From Fact 11, whenever two questions are in the relation of pure

e-implication, then they are in the relation of e-implication for each set of

declaratives.

If one question purely e-implies another question, then both questions

have the same semantic range.

Lemma 5. If Q [= Q

1

, then

¸

α∈dQ

´

α

=

¸

β∈dQ

1

´

β

.

Proof. From the ﬁrst condition of Deﬁnition 8

¸

α∈dQ

´

α

⊆

¸

β∈dQ

1

´

β

and from the second one

¸

β∈dQ

1

´

β

⊆

¸

∆

¸

α∈∆

´

α

⊆

¸

α∈dQ

´

α

.

28

2.3. Questions and questions

From this we can conclude that classes of safe and risky questions are

closed under pure e-implication for both implied and implying questions.

10

Fact 12. If Q [= Q

1

, then Q is safe (risky) iﬀ Q

1

is safe (risky).

The same semantic range of questions linked together by pure e-implica-

tion does not form an equivalence relation on questions (see non-symmetry

in Example 4 and non-transitivity in Example 5). On the other hand, pure

e-implication has some important consequences for classes of presupposi-

tions.

11

Lemma 6. If Q [= Q

1

, then PresQ = PresQ

1

.

Proof. First, let us prove PresQ ⊆ PresQ

1

. Let ϕ ∈ PresQ, so

¸

α∈dQ

´

α

⊆

´

ϕ

. Simultaneously, we know that from the second condition of the deﬁni-

tion of pure e-implication there is a non-empty ∆ ⊂ dQ, for each β ∈ dQ

1

,

such that ´

β

⊆

¸

α∈∆

´

α

⊆

¸

α∈dQ

´

α

. Thus, ´

β

⊆ ´

ϕ

, for each

β ∈ dQ

1

.

Second, for proving PresQ

1

⊆ PresQ suppose ϕ ∈ PresQ

1

. The following

inclusions are valid ´

α

⊆

¸

β∈dQ

1

´

β

⊆ ´

ϕ

, for each α ∈ dQ.

The claim of Lemma 6 is not extendable to the general e-implication (cf.

Example 3).

On this lemma we can base the following statement about an inﬂuence

of pure e-implication on classes of normal and regular questions.

Theorem 2. If Q [= Q

1

, then Q is normal iﬀ Q

1

is normal.

Proof. If Q is normal, then

¸

β∈dQ

1

´

β

=

¸

α∈dQ

´

α

= ´

PresQ

= ´

PresQ

1

The ﬁrst equation is from Lemma 5, the second one is from the normality of

Q, and the third one is from Lemma 6.

It is easy to prove a similar fact for regular questions.

Theorem 3. If Q [= Q

1

, then Q is regular iﬀ Q

1

is regular.

10

Cf. Theorem 7.29 in [44, p. 184]

11

See the same result in [44, p. 184], Theorem 7.33.

29

2.3. Questions and questions

Proof. Let us suppose Q [= Q

1

and Q is regular. From Q [= Q

1

and Lemma 5

we obtain

¸

α∈dQ

´

α

=

¸

β∈dQ

1

´

β

, which means that ´

PPresQ

= ´

PPresQ

1

.

The regularity of Q implies that there is a formula ϕ ∈ PPresQ. Putting it

together, ´

ϕ

=

¸

β∈dQ

1

´

β

. Thus, ϕ ∈ PPresQ

1

.

Both theorems have similar results we have got for safe (risky) ques-

tions in Fact 12. Classes of normal and regular questions are closed to pure

e-implication. Normal (regular) questions purely imply only normal (regular)

questions and they are purely implied by the same kind of questions.

12

Concerning classes of questions in relationship with e-implication, let us

add that whenever Q [= Q

1

, then Q is completely contradicory question iﬀ

Q

1

is.

Note on auxiliary sets of declaratives in e-implication Let us re-

mind the introductory example on page 27 to emphasize the importance of

declaratives for e-implication. Similarly, the following example will point out

the role of implicitly and explicitly expressed presuppositions.

Example 3 (in CPL). If Q

1

= ?¦α, β, γ¦ and Q

2

= ?¦α, β¦ (for atomic

α, β, γ), then neither Q

1

[= Q

2

nor Q

2

[= Q

1

(see the diﬀerent semantic

ranges of both questions). On the other hand, if we would know that it must

be (α ∨ β), then (α ∨ β), Q

1

[= Q

2

.

Keeping the context of this example: the question Q

2

is normal as well

as regular, then PPresQ

2

[[= dQ

2

and, in addition, there is ∆, non-empty

proper subset of dQ

1

, such that PPresQ

2

[[= ∆. It gives PPresQ

2

, Q

1

[= Q

2

.

If the set PPresQ

2

is explicitly expressed, the implication from Q

1

to Q

2

is

justiﬁed.

But now, back to the general approach. In the following fact we display

when we can say that two questions and a set of declaratives are in the

relationship of e-implication.

Fact 13. Let us have Γ and two questions Q

1

and Q

2

. In order to conclude

Γ, Q

1

[= Q

2

it is suﬃcient to have Γ [[= dQ

2

and Γ [[= ∆, where ∆ is a

non-empty proper subset of dQ

1

.

This fact can be formulated in this form: if Q

2

is sound relative to Γ and

Γ gives a partial answer to Q

1

, then Q

1

implies Q

2

with respect to Γ. We will

add some points to this discussion in section 2.3.2 and in the last paragraph

of section 2.3.4.

12

Theorems 2 and 3 put together results in [44, pp. 185–186].

30

2.3. Questions and questions

Regular erotetic implication

A special kind of e-implication arises if there is exactly one direct answer in

each ∆ in the second clause of Deﬁnition 8. Then we say that Q regularly

implies Q

1

(on the basis of Γ). The following deﬁnition originates from [46,

p. 26]:

Deﬁnition 9. A question Q regularly implies a question Q

1

on the basis of

a set of declaratives Γ iﬀ

1. (∀α ∈ dQ)(Γ ∪ α [[= dQ

1

),

2. (∀β ∈ dQ

1

)(∃α ∈ dQ)(Γ ∪ β [= α).

Because of the special importance of this relation let us use the symbol

¬ for it (so we write Γ, Q ¬ Q

1

).

In the case of pure regular e-implication, both conditions are changed into

the form:

1. (∀α ∈ dQ)(α [[= dQ

1

),

2. (∀β ∈ dQ

1

)(∃α ∈ dQ)(β [= α).

If Q ¬ Q

1

such that we can answer Q

1

, then we have an answer to Q.

The relationship of pure regular e-implication between two questions says

that the implied question is ‘stronger’ than the implying one in the sense of

answerhood (see also section 2.3.3).

Regularity can be enforced by the minimal number of direct answers of

an implying question: if Q [= Q

1

and [dQ[ = 2, then Q ¬ Q

1

.

Basic properties of erotetic implication

In this subsection, we are going to be interested in such properties as reﬂex-

ivity, symmetry, and transitivity of e-implication.

Erotetic implication is a reﬂexive relation.

Fact 14. Γ, Q [= Q, for each Γ and Q.

Even if there are examples of the symmetric behavior of e-implication, it

is not a symmetric relation, generally.

Example 4 (in CPL). Let Q

1

= ?¦(α ∨ β), α¦ and Q

2

= ?¦α, β¦. Then

Q

1

[= Q

2

, but Q

2

[= Q

1

.

31

2.3. Questions and questions

In this example there is no non-empty proper subset of dQ

2

for the for-

mula (α ∨ β) to fulﬁl the second condition in the deﬁnition of e-implication.

Moreover, it is useful to add that Q

1

regularly implies Q

2

.

Erotetic implication is not transitive either.

Example 5 (in CPL). ?(α ∧ β) ¬ ?[α, β[ and ?[α, β[ [= ?α, but ?(α ∧ β) [=

?α.

On the other hand, if we consider regular e-implication only, the following

theorem is valid.

Theorem 4. If Q

1

¬ Q

2

and Q

2

¬ Q

3

, then Q

1

¬ Q

3

.

Proof. The ﬁrst condition of Deﬁnition 9 is proved by Lemma 5.

The second clause of this deﬁnition is based on regularity that gives (∀γ ∈

dQ

3

)(∃α ∈ dQ

1

)(´

γ

⊆ ´

α

).

We can do a cautious strengthening by the following fact:

Fact 15. If Γ, Q

1

¬ Q

2

and Q

2

¬ Q

3

, then Γ, Q

1

¬ Q

3

.

As a ﬁnal remark, let us add that presuppositions of an implied question

are entailed by each direct answer of an implying question (with respect to

an auxiliary set of declaratives).

Fact 16. Let Γ, Q [= Q

1

. Then

1. (∀α ∈ dQ)(∀ϕ ∈ PresQ

1

)(Γ ∪ α [= ϕ)

2. If e-implication is regular, then (∀β ∈ dQ

1

)(∀ϕ ∈ PresQ)(Γ ∪ β [= ϕ).

2.3.2 Evocation and erotetic implication

Both types of inferential structures can appear together and we are going to

investigate their interaction.

As shown in the next example, e-implication does not preserve evocation.

If we know Γ [= Q

1

and Q

1

[= Q

2

, it does not mean that it must be Γ [= Q

2

.

Example 6 (in CPL). • ¦(α ∨ β)¦ [= ?[α, β[ and ?[α, β[ [= ?(α ∨ β),

but ¦(α ∨ β)¦ [= ?(α ∨ β).

• ¦α¦ [= ?[α, β[ and ¦α¦, ?[α, β[ [= ?α, but there is an answer to ?α in

¦α¦.

32

2.3. Questions and questions

Of course, we do not see anything pathological in this example. Knowing

(α ∨ β), resp. α, it is superﬂuous to ask ?(α ∨ β), resp. ?α.

Generally, this brings us back to the role of an auxiliary set of declara-

tives in e-implication. Due to the admissibility of weakening in declaratives

(Fact 11) we can arrive at structures of e-implications with Γ containing (di-

rect) answers to some of the two questions. On the other hand, there are

some solutions of this problem proposed by erotetic logicians.

13

In contrast to the previous example, we can prove that evocation carries

over through a regular e-implication.

Lemma 7. If Γ [= Q

1

and Q

1

¬ Q

2

, then Γ [= Q

2

.

Proof. The ﬁrst condition requires ´

Γ

⊆

¸

β∈dQ

2

´

β

. It is valid because of

the same semantic range of both questions.

Let us suppose that there is β ∈ dQ

2

entailed by Γ. Then ´

Γ

⊆ ´

β

and, thanks to regularity of e-implication, ´

Γ

⊆ ´

α

, for some α ∈ dQ

1

.

But it is in contradiction with Γ [= Q

1

.

Lemma 7 can be formulated not only in the version of pure regular

e-implication.

Theorem 5. If Γ [= Q

1

and Γ, Q

1

¬ Q

2

, then Γ [= Q

2

.

Proof. First, we prove Γ [[= dQ

2

. Supposing it is not true, then there is a

model M

0

of Γ such that M

0

[= β, for each β ∈ dQ

2

. Because of Γ [[= dQ

1

,

there is α

0

∈ dQ

1

and M

0

[= α

0

. From Γ∪α [[= dQ

2

, for each α ∈ dQ

1

, there

must be some β

0

∈ dQ

2

such that M

0

[= β

0

and that is a contradiction.

Secondly, let us suppose that there is β

0

∈ dQ

2

and Γ [= β

0

. Regularity

and second condition of e-implication give Γ ∪ β

0

[= α and it follows Γ [= α

that is in contradiction with Γ [= Q

1

.

Since regularity was used only in the second part of the proof, we get an

expected fact that Γ [= Q

1

and Γ, Q

1

¬ Q

2

gives soundness of an implied

question Q

2

relative to Γ.

14

At the ﬁrst sight, it need not be Γ, Q

1

[= Q

2

or Γ, Q

2

[= Q

1

if we only

know that Γ [= Q

1

as well as Γ [= Q

2

.

15

Generally, neither evocation nor

13

See, for example, the deﬁnition of strong e-implication given by Wi´sniewski in [44].

Fact 13 includes the original inspiration for the deﬁnition of strong e-implication. The

deﬁnition is the same as that of e-implication, but Γ [[= ∆ is added into the second clause.

14

The second part of the proof of Theorem 5 could be slightly changed and we obtain

that strong e-implication carries over as well. (Andrzej Wi´sniewski called my attention to

this.)

15

Let us take as an example (in CPL) the case ¦ϕ¦ [= ?α and ¦ϕ¦ [= ?β. Then neither

¦ϕ¦, ?α [= ?β nor ¦ϕ¦, ?β [= ?α.

33

2.3. Questions and questions

e-implication says something new about structures of engaged questions.

Nevertheless, we can expect that some clearing up of the structure of sets

of direct answers could be helpful for the study of inferences. This will be

discussed in the next section.

2.3.3 Comparing questions: relations of questions ba-

sed on direct answers

So far we have introduced inferences that can provide certain relations be-

tween questions. Moreover, it would be useful to be able to compare ques-

tions with respect to their ‘answerhood power’. The chosen set-of-answers

methodology brings us to a natural approach. Sets of direct answers can be

purely compared or we can investigate their relationship based on entailment

relation, moreover, we can control the cardinality of sets of direct answers by

a mapping from one set to the other.

Let us start with relations among questions based on pure comparison of

sets of direct answers.

Deﬁnition 10. • Two questions are equal (Q

1

= Q

2

) iﬀ they have the

same set of direct answers (dQ

1

= dQ

2

).

16

• A question Q

1

is included in a question Q

2

(Q

1

⊂ Q

2

) iﬀ dQ

1

⊂ dQ

2

.

This approach could be extended in a semantic way. We say that (an

answer) α gives an answer to a question Q iﬀ there is β ∈ dQ such that

α [= β. Having two questions Q

1

and Q

2

we can deﬁne a relationship of

‘giving answers’:

Deﬁnition 11. A question Q

1

gives a (direct) answer to Q

2

iﬀ (∀α ∈

dQ

1

)(∃β ∈ dQ

2

)(α [= β).

In this deﬁnition the ﬁrst question is considered as to be (semantically)

‘stronger’ than the second one. For this we use the symbol ≥ and write

Q

1

≥ Q

2

.

If Q

1

= Q

2

or Q

1

⊂ Q

2

, then Q

1

≥ Q

2

and, moreover, each direct answer

to Q

1

not only gives an answer to Q

2

but also is a (direct) answer to Q

2

, i.e.,

(∀α ∈ dQ

1

)(∃β ∈ dQ

2

)(α ≡ β).

16

The original deﬁnition refers to equivalent questions instead of equal (cf. [44, p. 135]),

but we use the ﬁrst term for erotetically equivalent or semantically equivalent. In our

set-of-answers methodology (questions are deﬁned by sets of direct answers), this term is

redundant.

34

2.3. Questions and questions

The ordering based on the relation ≥ has a slightly non-intuitive conse-

quence: a completely contradictory question is the strongest one. However,

the class of evoked questions is not aﬀected by this problem.

Let us note an expected fact—stronger questions presuppose more than

weaker ones.

Fact 17. If Q

1

≥ Q

2

, then PresQ

2

⊆ PresQ

1

.

This fact is not too useful. It is better to notice the relationship among

maximal presuppositions. We have ´

PresQ

1

⊆ ´

PresQ

2

. Each maximal pre-

supposition of a stronger question entails a maximal presupposition of a

weaker one, respectively, a prospective presuposition of a stronger question

entails a prospective presupposition of a weaker question. The semantic range

of a stronger question is included in the semantic range of a weaker question.

Fact 18. If Q

1

≥ Q

2

, then

¸

α∈dQ

1

´

α

⊆

¸

β∈dQ

2

´

β

.

It follows that the set of safe questions is closed under weaker questions.

Fact 19. If Q

1

is safe and Q

1

≥ Q

2

, then Q

2

is safe.

The next example shows that safeness of weaker questions is not trans-

ferred to stronger ones.

Example 7 (in CPL). ?¦β ∧ α, β¦ ≥ ?β

Answerhood, evocation, and erotetic implication

We can show some results of evocation and e-implication based on properties

of the ≥-relation. The ﬁrst one is an obvious fact that an implied stronger

question is regularly implied.

Lemma 8. If Γ, Q

1

[= Q

2

and Q

2

≥ Q

1

, then Γ, Q

1

¬ Q

2

.

Recall what is required of the regular e-implication: (∀β ∈ dQ

2

)(∃α ∈

dQ

1

)(Γ ∪ β [= α). Then the lemma follows.

If the relation ≥ is turned, i.e., Q

1

gives an answer to Q

2

, then whenever

Q

1

implies Q

2

, Q

2

regularly implies Q

1

(both with respect to Γ). Moreover,

both questions are erotetically equivalent relative to Γ.

Theorem 6. If Γ, Q

1

[= Q

2

and Q

1

≥ Q

2

, then Γ, Q

2

¬ Q

1

.

Proof. First, we need to show that Γ∪β [[= dQ

1

, for each β ∈ dQ

2

. But it is

an easy conclusion from Γ, Q

1

[= Q

2

because there is a subset ∆ ⊆ dQ

1

for

each β ∈ dQ

2

such that Γ ∪ β [[= ∆.

The second condition of regular e-implication is obvious, it follows from

Q

1

≥ Q

2

.

35

2.3. Questions and questions

Now, as it was stated before, we are going to study the inﬂuence of ‘giving

answers’ on the relationship of evocation and e-implication. We know that,

generally, if Γ evokes Q

1

and Q

2

, it need not be that either Q

1

implies Q

2

or Q

2

implies Q

1

(with respect to Γ). If a stronger question is evoked by Γ,

then every weaker question regularly implies this stronger one with respect

to Γ.

Theorem 7. If Γ [= Q

1

and Q

1

≥ Q

2

, then Γ, Q

2

¬ Q

1

.

Proof. First, Γ ∪ β [[= dQ

1

is required for each β ∈ dQ

2

. We get Γ [[= dQ

1

from Γ [= Q

1

.

Second, from Q

1

≥ Q

2

we have (∀α ∈ dQ

1

)(∃β ∈ dQ

2

)(α [= β) and it

gives the second condition of regular e-implication (∀α ∈ dQ

1

)(∃β ∈ dQ

2

)(Γ∪

α [= β).

To digress for a moment, this repeated connection of ≥ and regular

e-implication is not an accident. The deﬁnition of regular e-implication says

that if Q

1

¬ Q

2

, then Q

2

gives an answer to Q

1

, i.e., Q

2

≥ Q

1

. However,

‘giving an answer’ does not produce e-implication, see the next example.

Example 8 (in CPL). ?¦(α ∧ ϕ), (β ∧ ψ)¦ ≥ ?¦α, β¦, but neither ?¦(α ∧

ϕ), (β ∧ ψ)¦ [= ?¦α, β¦ nor ?¦α, β¦ [= ?¦(α ∧ ϕ), (β ∧ ψ)¦.

To go back to evocation, it is clear that two equal questions are both

evoked by a set of declaratives if one of them is evoked by this set. Generally,

it is not suﬃcient to know Γ [= Q

1

and Q

1

≥ Q

2

to conclude Γ [= Q

2

.

An evoked stronger question only implies the soundness of weaker questions

relative to Γ. Let us illustrate it in the case that the ﬁrst question is included

in the second one (Q

1

⊂ Q

2

); there could be a direct answer to Q

2

which is

entailed by Γ. This reminds us of the non-monotonic behavior of evocation.

Notice that Q

2

⊂ Q

1

will not help us either. In the connection with the

relation ≥ we have to require a version of an equality.

17

Fact 20. Let Q

1

≥ Q

2

and Q

2

≥ Q

1

. Then

• Γ [= Q

1

iﬀ Γ [= Q

2

,

• Q

1

¬ Q

2

as well as Q

2

¬ Q

1

.

17

We are not going to introduce a special name for this relationship; it is included in

the erotetic equivalence.

36

2.3. Questions and questions

Controlling the cardinality of sets of direct answers

The set of direct answers of a weaker question can be much larger than that

of a stronger question. The book [44] introduces two relations originated

from Tadeusz Kubi´ nski that prevent this uncontrolled cardinality.

Deﬁnition 12. A question Q

1

is stronger then Q

2

(Q

1

_ Q

2

) iﬀ there is a

surjection j : dQ

1

→ dQ

2

such that for each α ∈ dQ

1

, α [= j(α).

The number of direct answers of the weaker question Q

2

does not exceed

the cardinality of dQ

1

, i.e., [dQ

1

[ ≥ [dQ

2

[. From the surjection, additionaly,

we know that each direct answer of a weaker question is given by some direct

answer to a stronger question. We have used the term stronger in a bit

informal way for questions that ‘give an answer’ to weaker ones. It is clear

that if Q

1

_ Q

2

, then Q

1

≥ Q

2

. But, unfortunately, we cannot provide

any special improvement of previous results for _-relation. In particular,

Examples 7 and 8 are valid for _-relation as well.

The other deﬁnition corresponds to a both-way relationship of ‘being

stronger’.

Deﬁnition 13. A question Q

1

is equipollent to a question Q

2

(Q

1

≡ Q

2

) iﬀ

there is a bijection i : dQ

1

→ dQ

2

such that for each α ∈ dQ

1

, α ≡ i(α).

In this case, both sets of direct answers have the same cardinality ([dQ

1

[ =

[dQ

2

[). Let us add expected results gained from equipollency.

Fact 21. If Q

1

≡ Q

2

, then

• both Q

1

_ Q

2

and Q

2

_ Q

1

,

• Γ [= Q

1

iﬀ Γ [= Q

2

,

• Q

1

¬ Q

2

as well as Q

2

¬ Q

1

.

Of course, two equal questions are equipollent.

Partial answerhood

We declared that the study of various types of answers (generally speaking,

answerhood) is not the central point of this chapter. However, we can utilize

the idea evoked by the second clause of Deﬁnition 8. Narrowing down the

set of direct answers of an implying question seems to be a good base for the

deﬁnition of partial answer.

37

2.3. Questions and questions

Deﬁnition 14. A declarative ϕ gives a partial answer to a question Q iﬀ

there is a non-empty proper subset ∆ ⊂ dQ such that ϕ [[= ∆.

18

This deﬁnition allows us to cover many terms from the concept of the

answerhood. Every direct answer gives a partial answer. Whenever ψ gives

a (direct) answer, then ψ gives a partial answer. As a useful conclusion we

obtain a weaker version of Theorem 7:

Fact 22. If Γ [= Q

1

and each α ∈ dQ

1

gives a partial answer to Q

2

, then

Γ, Q

2

[= Q

1

.

2.3.4 Questions and sets of questions

Working in the classsical logic, let us imagine we would like to know whether

it is the case that α or it is the case that β. The question ?¦α, β¦ is posed.

But there could be a problem when an entity, to which we are going to

address this question, is not able to accept it. This can be caused, e.g.,

by a restricted language-acceptability, i.e., a device cannot ‘understand’ this

question. However, assume that there exist two devices such that: the ﬁrst

one can be asked by the question ?α, and the other one is able to work

with the question ?β. From both machines, independently, we can get the

following pairs of answers: ¦α, β¦, ¦α, β¦, ¦α, β¦ or ¦α, β¦.

Posing the question Q = ?¦α, β¦ we expect that if an answer to Q is true,

then there must be a true answer to each question from the set ¦?α, ?β¦.

Thus, we obtain a soundness transmission from an initial question to a set

of questions.

Generally speaking, let us suppose that there are a question Q and a set

of questions Φ = ¦Q

1

, Q

2

, . . .¦. For each model of a direct answer to Q there

must be a direct answer in each Q

i

valid in this model.

(∀α ∈ dQ)(∀Q

i

∈ Φ)(α [[= dQ

i

)

Possible states (of the world) given by answers to questions in the set Φ must

be in a similar relation to the initial question. Whenever we keep a model

of the choice of direct answers from each question in Φ, then there must be

a direct answer to Q true in this model. For this, let us introduce a choice

function ξ such that ξ(Q

i

) chooses exactly one direct answer from dQ

i

. For

each set of questions Φ and a choice function ξ there is a choice set A

Φ

ξ

=

¦ξ(Q

i

)

Q

i

∈ Φ¦.

19

The soundness condition in the other direction (from a

set to initial question) will be expressed, generally, by (∀A

Φ

ξ

)(A

Φ

ξ

[[= dQ).

18

Compare this deﬁnition with Deﬁnition 4.10 in [44].

19

If the set Φ is clear from the context, we will write only A

ξ

.

38

2.3. Questions and questions

Back to our example, there are four choice sets:

A

ξ

1

= ¦α, β¦

A

ξ

2

= ¦α, β¦

A

ξ

3

= ¦α, β¦

A

ξ

4

= ¦α, β¦

But the fourth one is not in compliance with the second soundness require-

ment, it is in contradiction with our (prospective) presupposition (α ∨ β).

If we admit the additional answer (α ∧ β) and a question in the form

?¦α, β, (α ∧ β)¦, mutual soundness of this question and the set of ques-

tions ¦?α, ?β¦ will be valid. But this solution seems to be rather awkward.

A questioner posing the question ?¦α, β¦ evidently presupposes (α∨β). This

will bring us to the deﬁnition of reducibility with respect to an auxiliary set

of declaratives and the mutual soundness will be required in the following

forms:

(∀α ∈ dQ)(∀Q

i

∈ Φ)(Γ ∪ α [[= dQ

i

)

and

(∀A

Φ

ξ

)(Γ ∪ A

Φ

ξ

[[= dQ).

Our example produces more than soundness of Q relative to each A

Φ

ξ

(with respect to Γ), also eﬃcacy of each A

Φ

ξ

with respect to a question Q is

valid:

(∀A

Φ

ξ

)(∃α ∈ dQ)(Γ ∪ A

Φ

ξ

[= α).

It will be reasonable to keep this strengthening. We require to obtain at

least one answer to an initial question from a choice set. Whenever Γ and

A

Φ

ξ

describe the state of the world, there must be a direct answer to a question

Q that does the same job.

Reducibility of questions to sets of questions

We can take advantage of the previous discussion for the direct deﬁnition

of reducibility of a question to a set of questions. Now, we introduce pure

reducibility that does not use any auxiliary set of declaratives.

Deﬁnition 15. A question Q is purely reducible to a non-empty set of

questions Φ iﬀ

1. (∀α ∈ dQ)(∀Q

i

∈ Φ)(α [[= dQ

i

)

2. (∀A

Φ

ξ

)(∃α ∈ dQ)(A

Φ

ξ

[= α)

39

2.3. Questions and questions

3. (∀Q

i

∈ Φ)([dQ

i

[ ≤ [dQ[)

First two conditions express mutual soundness, the second one adds eﬃ-

ciacy, as it was discussed, and the last one requires relative simplicity. If Q is

reducible to a set Φ, we will write Q Φ. The deﬁnition of pure reducibility

was introduced by Andrzej Wi´sniewski in [43].

Example 9 (in CPL). • ?¦α, β, (α ∧ β)¦ ¦?α, ?β¦

• ?[α, β[ ¦?α, ?β¦

• ?(α ◦ β) ¦?α, ?β¦, where ◦ is any of the connectives: ∧, ∨, →

In the ﬁrst item, there is the ‘pure’ version from the introductory discus-

sion. All items display reducibilities between initial safe questions and sets

of safe questions. The following theorem shows that it is not an accident.

Theorem 8. If Q Φ, then Q is safe iﬀ each Q

i

∈ Φ is safe,

Proof. The ﬁrst condition of Deﬁnition 15 can be rewritten as

¸

α∈dQ

´

α

⊆

¸

β∈dQ

i

´

β

, for each Q

i

∈ Φ, and it gives the implication from left to right.

For the proof of the other implication, let us suppose Q Φ and that each

Q

i

∈ Φ is safe, but Q is not. It implies the existence of model M

0

∈ ´ such

that M

0

[= α, for each α ∈ dQ. The safeness of all Q

i

gives (∀Q

i

∈ Φ)(∃β ∈

dQ

i

)(M

0

[= β). Thus, there is A

Φ

ξ

made from these βs and M

0

[= A

Φ

ξ

. But

it is in contradiction with the second condition of the deﬁnition of Q Φ,

which gives the existence of some α ∈ dQ such that M

0

[= α.

From this we know that if there is a risky question among questions in Φ

and Q Φ, then Q must be risky too (cf. [44, p. 197]).

The rewritten ﬁrst condition of Deﬁnition 15 is of the form

¸

α∈dQ

´

α

⊆

¸

i

¸

β∈dQ

i

´

β

and it brings out the relationship of semantic ranges. The semantic range of

a reduced question is bounded by the intersection of all semantic ranges of

Q

i

s.

The relation of pure reducibility is reﬂexive (Q ¦Q¦) and we can prove

the following version of transitivity:

20

Theorem 9. If Q Φ and each Q

i

∈ Φ is reducible to some set of questions

Φ

i

, then Q

¸

i

Φ

i

.

20

Presented Theorem 9 corresponds to Corollary 7.6 in [44].

40

2.3. Questions and questions

Proof. The third condition of Deﬁnition 15 is clearly valid.

The ﬁrst one is easy to prove. From Q Φ we get

¸

α∈dQ

´

α

⊆

¸

β∈dQ

i

´

β

, for each Q

i

∈ Φ, and from the reducibility of all Q

i

in Φ to

an appropriate Φ

i

we have

¸

β∈dQ

i

´

β

⊆

¸

γ∈dQ

j

´

γ

, for each dQ

j

∈ Φ

i

. It

gives together

¸

α∈dQ

´

α

⊆

¸

γ∈dQ

j

´

γ

, for each Q

j

∈

¸

i

Φ

i

.

For the second one we require the existence of α ∈ dQ such that A

∪

i

Φ

i

ξ

[=

α, for each A

∪

i

Φ

i

ξ

. From the reducibility of all Q

i

in Φ to an appropriate Φ

i

we have that each A

Φ

i

ξ

is a subset of some A

∪

i

Φ

i

ξ

. It implies that if there is

any model M of A

∪

i

Φ

i

ξ

, it must be a model of some A

Φ

i

ξ

. From Q Φ we

know that there is α ∈ dQ for each choice set A

Φ

ξ

on Φ. This choice set is

made by elements of all dQ

i

∈ Φ which are valid in M. It means that A

Φ

ξ

is

valid in M as well as α ∈ dQ.

Now let us look at the relationship of pure reducibility and pure e-implica-

tion. The following example shows that it need not be that e-implication

causes reducubility. Both deﬁnitions have the same ﬁrst conditions, but the

second condition of reducibility can fail.

Example 10 (in CPL). ?[α, β[ [= ?(α ∧ β), but ?[α, β[ ¦?(α ∧ β)¦.

On the other hand, we can prove that regular e-implication implies re-

ducibility.

Lemma 9. Let Φ be a set of questions such that Q ¬ Q

i

, for each Q

i

∈ Φ.

If (∀Q

i

∈ Φ)([dQ

i

[ ≤ [dQ[), then Q Φ.

Proof. Let us prove the second condition of Deﬁnition 15 that requires ex-

istence of α ∈ dQ such that ´

A

Φ

ξ

⊆ ´

α

, for each A

Φ

ξ

. It is known that

´

A

Φ

ξ

⊆ ´

β

, for each β ∈ A

Φ

ξ

. If Q ¬ Q

i

, then for each β ∈ dQ

i

there is

α ∈ dQ such that ´

β

⊆ ´

α

. Thus, ´

A

Φ

ξ

⊆ ´

α

.

What about if we know Q

i

[= Q or, even, Q

i

¬ Q, for each Q

i

∈ Φ, and

(∀Q

i

∈ Φ)([dQ

i

[ ≤ [dQ[), can we conclude that Q Φ? Example 10 gives

the negative answer to this question as well as ?(α ∧ β) ¬ ?[α, β[.

Not even reducibility produces e-implication.

Example 11 (in CPL). ?(α ∧ β) ¦?α, ?β¦, but ?(α ∧ β) is not implied

neither by ?α nor by ?β and ?(α ∧ β) does not imply neither ?α nor ?β.

In the next subsection we will study some special cases of links between

reducibility and e-implication.

41

2.3. Questions and questions

So far we have worked only with the pure reducibility. It could be useful

to introduce the general term of reducibility with respect to a context given

by a set of declaratives. The deﬁnition is almost the same as Deﬁnition 15,

but the mutual soundness and eﬃcacy conditions are supplemented by an

auxiliary set of declaratives Γ (cf. [20]). We will write Γ, Q Φ.

Deﬁnition 16. A question Q is reducible to a non-empty set of questions

Φ with respect to a set of declaratives Γ iﬀ

1. (∀α ∈ dQ)(∀Q

i

∈ Φ)(Γ ∪ α [[= dQ

i

)

2. (∀A

Φ

ξ

)(∃α ∈ dQ)(Γ ∪ A

Φ

ξ

[= α)

3. (∀Q

i

∈ Φ)([dQ

i

[ ≤ [dQ[)

The introductory discussion is displayed in this example:

Example 12 (in CPL). (α ∨ β), ?¦α, β¦ ¦?α, ?β¦

As it is expected, the role of Γ is similar to the role of an auxiliary set of

declaratives in e-implication:

Fact 23. If Q Φ, then Γ, Q Φ, for each Γ.

So we can speak of weakening in declaratives and it enables us to gener-

alize Lemma 9.

Theorem 10. If Γ, Q ¬ Q

i

, for each Q

i

∈ Φ, and (∀Q

i

∈ Φ)([dQ

i

[ ≤ [dQ[),

then Γ, Q Φ.

Proof. The third and the ﬁrst conditions of Deﬁnition 16 are obvious.

The second one requires that for each A

Φ

ξ

there is α ∈ dQ such that

´

Γ∪A

Φ

ξ

⊆ ´

α

. From the construction of choice sets we know that for each

A

Φ

ξ

and Q

i

∈ Φ there is β ∈ dQ

i

(member of A

Φ

ξ

) such that ´

Γ∪A

Φ

ξ

⊆ ´

Γ∪β

.

The regular e-implication provides that there is α ∈ dQ for each β ∈ dQ

i

such that ´

Γ∪β

⊆ ´

α

.

We close this subsection by reversing the ‘direction’ of the reducibility

relation. Let us suppose that we have generated a set of questions Φ that

are evoked by a set of declaratives Γ. Can we conclude that Γ evokes such a

complex question which is reducible to the set Φ? Generally, not. But if we

know that the complex question gives an answer to some question from Φ,

the answer is positive.

Theorem 11. If Γ evokes each question from a set Φ, Q Φ, and there is

a question Q

i

∈ Φ such that Q ≥ Q

i

, then Γ [= Q.

42

2.3. Questions and questions

Proof. Soundness of Q relative to Γ requires the existence of an answer α ∈

dQ for each model M [= Γ. From the evocation of each Q

i

∈ Φ we have

(∀M [= Γ)(∀Q

i

∈ Φ)(∃β ∈ dQ

i

)(M [= β). So, each model of Γ produces some

choice set such that (∀M [= Γ)(∃A

Φ

ξ

)(M [= A

Φ

ξ

). Together with reducibility,

where it is stated that (∀A

Φ

ξ

)(∃α ∈ dQ)(A

Φ

ξ

[= α), we get Γ [[= dQ.

Informativness of Q with respect to Γ is justiﬁed by ≥-relation for some

question Q

i

∈ Φ. If Γ [= α, for some α ∈ dQ, then it gives a contradiction

with Γ [= Q

i

.

Given the conditions of Theorem 11 are met, we obtain:

• Γ, Q [= Q

i

, for each Q

i

∈ Φ, and

• Γ, Q

i

¬ Q, for Q ≥ Q

i

.

The ﬁrst item is based on Fact 13 and the second one is given by the help of

Theorem 7.

Reducibility and sets of yes-no questions

The concept of reducibility is primarily devoted to a transformation of a

question to a set of ‘less complex’ questions. The introductory discussion

and its formalization in Example 12 evoke interesting questions:

• If we have an initial question Q = ?¦α

1

, α

2

, . . .¦ with, at worse, a

countable list of direct answers, is it possible to reduce it to a set of

yes-no questions based only on direct answers of Q?

• Moreover, could we require the e-implication relationship between Q

and questions in the set Φ?

We can ﬁnd an easy solution to these problems under condition that

yes-no questions are safe and we have an appropriate set of declaratives. We

will require Q to be sound with respect to Γ.

Theorem 12. Let us suppose that yes-no questions are safe in the background

logic. If a question Q = ?¦α

1

, α

2

, . . .¦ is sound with respect to a set Γ, then

there is a set of yes-no questions Φ such that Γ, Q Φ and Γ, Q [= Q

i

, for

each Q

i

∈ Φ.

Proof. First, we deﬁne the set of yes-no questions Φ based on the initial

question Q = ?¦α

1

, α

2

, . . .¦ such that

Φ = ¦?α

1

, ?α

2

, . . .¦.

43

2.3. Questions and questions

Secondly, we prove the condition that is common for both reducibility and

e-implication. The safeness of members of Φ implies that ´

α

⊆

¸

β∈dQ

i

´

β

,

for each α ∈ dQ and Q

i

∈ Φ. This gives ´

Γ∪α

⊆

¸

β∈dQ

i

´

β

.

To prove reducibility we have to justify the second condition of Deﬁni-

tion 16. We need to ﬁnd an α ∈ dQ for each A

Φ

ξ

such that Γ∪A

Φ

ξ

[= α. Two

cases will be distinguished.

1. If there is α from both A

Φ

ξ

and dQ, then choose this direct answer.

2. If there is no direct answer α ∈ dQ in A

Φ

ξ

, then ´

Γ∪A

Φ

ξ

= ∅ and we

can take any α from dQ.

The ﬁnal step is the proof of e-implication. We have to show that for

each Q

i

and each direct answer β ∈ dQ

i

there is a non-empty subset ∆ ⊂ dQ

such that Γ ∪ β [[= ∆. For this, we use the shape of questions in Φ.

1. If β ∈ dQ, then ∆ could be ¦β¦ and Γ ∪ β [[= ¦β¦.

2. If β ∈ dQ and Γ∪β has at least one model, we recognize that ´

Γ∪β

⊆

´

Γ

. Simultaneously, β must be of the form α

j

and ∆ can be deﬁned

as dQ`¦α

j

¦. Together with soundness of initial question Q with respect

to Γ, which means ´

Γ

⊆

¸

α∈dQ

´

α

, we get ´

Γ∪β

⊆

¸

α∈∆

´

α

.

This theorem enables us to work with classes of questions which are known

to be sound relative to sets of their presuppositions. (Normal and regular

questions are the obvious example.) Whenever we know that the initial

question is evoked by a set of declaratives, we get the following conclusion.

Fact 24. Working in logics where yes-no questions are safe, if a question

Q = ?¦α

1

, α

2

, . . .¦ is evoked by a set of declaratives Γ, then there is a set of

yes-no questions Φ such that Γ, Q Φ and Γ, Q [= Q

i

, for each Q

i

∈ Φ.

This fact corresponds to main lemma in the paper [20, pp. 104–5] where

a bit diﬀerent deﬁnition of reducibility is used, but the results are the same.

If dQ is ﬁnite or the entailment is compact, the set Φ is ﬁnite set of

yes-no questions. Simultaneously, it is useful to emphasize that the proof of

Theorem 12 shows how to construct such a set.

21

In logics with risky yes-no questions, the ﬁrst condition of reducibility as

well as e-implication can fail. It need not be ´

Γ,α

⊆ (´

β

∪ ´

¬β

), for each

α ∈ dQ and each ?β ∈ Φ. More generally, we can ask for a help the auxiliary

21

The same result is provided by theorems 7.49–7.51 in [44].

44

2.4. Final remarks

set of declaratives again. Let us remind Fact 13 and put soundness of each

Q

i

∈ Φ with respect to Γ. Going through the proof of Theorem 12, the rest

is valid independently of safeness of yes-no questions. As a conclusion we get

Fact 25. If a question Q = ?¦α

1

, α

2

, . . .¦ is sound relative to a set Γ, and

there is a set of yes-no questions Φ = ¦?α

1

, ?α

2

, . . .¦ (based on Q) such that

Γ [[= dQ

i

, for each Q

i

∈ Φ, then Γ, Q Φ and Γ, Q [= Q

i

, for each Q

i

∈ Φ.

The construction of yes-no questions provided by Theorem 12 does not

prevent the high complexity of such yes-no questions. Observing the last

item of Example 9, it seems worthwhile to enquire whether it is possible to

follow this proces and to reduce a question (with respect to an auxiliary set of

declaratives) to a set of atomic yes-no questions based on subformulas of the

initial question (cf. [50]). The ﬁrst restriction is clear, yes-no questions must

be safe. But it is not all, the second clause of pure reducibility (Deﬁnition 15)

requires the truth-functional connection of subformulas. Then the answer is

positive. We can use repeatedly a cautious extension of Theorem 9:

Fact 26. If Γ, Q Φ and each Q

i

∈ Φ is reducible to some set of questions

Φ

i

, then Γ, Q

¸

i

Φ

i

.

There is a similar concept in literature, called erotetic search scenarios,

based on properties of the classical logic and erotetic implication (see [46, 47,

48] and [39]). The idea is that there is an initial question (and a context) and

we can get an answer to it by searching through answers of some operative

(auxiliary) questions. Scenarios are trees, an initial question (and a context)

is the root and branching is based on direct answers of auxiliary questions.

The relationship between interrogative nodes is given by erotetic implication.

Some scenarios work with the descending ‘complexity’ of questions from the

root to leafs. If we recall non-transitivity of e-implication (Example 5), we

can recognize the ‘truth-functional’ auxiliary role of the question ?[α, β[ as

an interlink between ?(α ∧ β) and questions ?α and ?β.

2.4 Final remarks

IEL is introduced as a very general theory of erotetic inferences dealing with

a general language L extended by questions. Following the IEL-philosophy

and in accordance with our ﬁrst chapter we decided to introduce questions

as sets of declaratives, which are in the role of direct answers. Inferences

with questions are based on multiple-conclusion entailment that makes it

possible to work with sets of declaratives in premises as well as in conclusions.

Thus, questions are kept as special objects of the language L

Q

, they are not

45

2.4. Final remarks

combined by logical constants as declaratives are. All relationships among

questions are based on inferences and a comparison of their sets of direct

answers. The main modiﬁcation of the original IEL can bee seen in the

chosen SAM and in the model-based approach; the term semantic range of

a question made the semantic work easier and transparent.

In this chapter we studied many properties and relationships, but the

central point was whether we can formulate some relationships (meta-rules)

for erotetic consequence relations in IEL. Our general view did not tend to

provide any axiomatization. The discussed properties and relationships vary

with the chosen background system, especially, they depend on semantics.

IEL opens many possibilities for working with questions in various logical

systems, see [7] as a nice example, and serves as an inspiration for the work

in erotetic logic.

46

Chapter 3

Epistemic logic with questions

3.1 Introduction

Communication is essentially connected with an exchange of information.

The basic way to complete someone’s knowledge is the posing of questions.

Although questions diﬀer from declaratives, they play a similar role in reason-

ing. In the recent history of logic of questions we can see a success in ﬁnding

the desirable position of the formal approach to questions and in a study of

inferences based on them. Let us recall Wi´sniewski’s inferential erotetic logic

and the intensional approach of Groenendijk and Stokhof. Such approaches

made it possible to incorporate questions within some formal systems and

not to lose their speciﬁc position in inferences. This brings us to an impor-

tant point, on the one hand, questions are speciﬁc entities and they bear

some special properties, on the other hand, they are used in formal systems

that are the framework of reasoning.

In Section 1.2 we introduced a formalization of questions called set-of-ans-

wers methodology. It utilizes the close connection of questions and their

answerhood conditions. Simultaneously we mentioned there that it is very

natural to use the epistemic terms in speaking of questions.

The approach we are going to present in this chapter works with epistemic

logic as a background. Roughly speaking, epistemic logic and its semantics

are used for the modeling of knowledge and epistemic possibilities of agents

and groups of them. Epistemic logic in use will be the normal multi-modal

logic. At the very beginning we do not impose any conditions on it; thus,

the working system will be the normal modal (epistemic) logic K with its

standard relational semantics (Kripke frames and models). However, the way

of incorporating questions will be of such generality that it can be applied in

all epistemic-like logics.

47

3.1. Introduction

Moreover, the process of communication is a dynamic matter and epis-

temic logic is open to enrichment by dynamic aspects, see, e.g., [42]. These

aspects will be studied later on in chapter 4 where we accept the ‘standard’

model of knowledge based on the modal system S5. Now let us brieﬂy men-

tion what a question and its epistemic meaning are.

In the introductory chapter we use the example with three friends holding

cards. Let us suppose Catherine wants to ﬁnd out where the Joker-card is.

Then she can ask

Who has the Joker: Ann, or Bill?

From our set-of-answers methodological viewpoint, the question has a two

element set of direct answers:

• Ann has the Joker.

• Bill has the Joker.

In this situation we can recognize the question as ‘reasonable’. Asking it,

Catherine expresses that she

1. does not know what is the right answer to the question,

2. considers the answers to be possible, and, moreover,

3. presupposes what is implicitly included in the answers, i.e., it must be

the case that just either Ann has the Joker or Bill has it.

An agent-questioner provides the information of her ignorance (item 1) as

well as the expected way to complete her knowledge: 2 says that there are

two possibilities and 3 that one of the possibilities is expected. A question is a

store of information of agent’s epistemic state. An asked question means that

listeners can form their partial picture of questioner’s knowledge structure,

which is an important part in communication and solving of problems in

groups.

1

As it was mentioned at the beginning, the exchange of information

is a basis of communication. The main aim of communication in a group

is to share data and to solve problems. Very typical example is a group of

scientists trying to ﬁnd an answer to their scientiﬁc problem. Only by sharing

of their knowledge and ignorance they can reach a solution.

Communication will be studied in the next chapter, here we prepare an

‘erotetic epistemic framework’. First, we introduce propositional single-agent

1

Jeroen Groenendijk says that “assertions may provide new data, questions may provide

new issues” [12].

48

3.2. Single-agent propositional epistemic logic and questions

(normal modal) epistemic logic and extend it by questions. We fully apply

our set-of-answers methodology and allow to mix declaratives and interrog-

atives. Questions will be a natural part of inference relations based on the

background logic. Then we discuss answerhood conditions in a relationship

with conditions posed on a ‘reasonable’ question. And, ﬁnally, the role of

epistemic context as well as sets of questions are studied.

3.2 Single-agent propositional epistemic logic

and questions

Our approach to epistemic logic is very liberal. The language of classical

propositional logic L

cpl

is extended by modalities [i] and 'i`. The ﬁrst one

can be interpreted as ‘agent i knows that. . . ’, ‘agent i believes that. . . ’, etc.

The other one is an ‘epistemic possibility’. Thus, we get a language L

K

cpl

with

a subset of signs for atomic formulas { = ¦p, q, . . .¦ and formulas deﬁned as

follows:

ϕ ::= p [ ψ [ ψ

1

∨ ψ

2

[ ψ

1

∧ ψ

2

[ ψ

1

→ ψ

2

[ ψ

1

↔ ψ

2

[ [i]ψ [ 'i`ψ

Modality 'i` is understood as a dual to [i]:

'i`ϕ ≡ [i]ϕ

In multi-agent variants of epistemic logic we presuppose that there is a ﬁnite

set of agents / = ¦1, . . . , m¦, where numbers 1, . . . , m are names for agents.

This section deals with a single-agent variant for the sake of simplicity in

introducing questions and their basic properties. Moreover, we do not restrict

the interpretation of [i] to ‘knowledge’ of an agent i. Now, roughly and

vaguely said, it is just ‘epistemic necessity’ of an agent without restrictions

to knowledge conditions or belief conditions.

Semantics is based on Kripke-style models. Kripke frame is a relational

structure T = 'S, R

i

` with a set of states (points, indices, possible worlds) S

and an accessibility relation R

i

⊆ S

2

. Kripke model M is a pair 'T, v` where

v is a valuation of atomic formulas. The satisfaction relation [= is deﬁned by

a standard way:

1. (M, s) [= p iﬀ (M, s) ∈ v(p)

2. (M, s) [= ϕ iﬀ (M, s) [= ϕ

3. (M, s) [= ψ

1

∨ ψ

2

iﬀ (M, s) [= ψ

1

or (M, s) [= ψ

2

49

3.2. Single-agent propositional epistemic logic and questions

4. (M, s) [= ψ

1

∧ ψ

2

iﬀ (M, s) [= ψ

1

and (M, s) [= ψ

2

5. (M, s) [= ψ

1

→ ψ

2

iﬀ (M, s) [= ψ

1

implies (M, s) [= ψ

2

6. (M, s) [= [i]ϕ iﬀ (M, s

1

) [= ϕ, for each s

1

such that sR

i

s

1

We do not put any restrictions on accessibility relation, thus, we have se-

mantics for the modal system K.

3.2.1 Incorporating questions

We extend epistemic language L

K

cpl

by brackets ¦, ¦ and the question mark

?

i

for a question of an agent i. We get the language L

KQ

cpl

. For interrogative

formulas metavariables Q

i

, Q

i

1

, etc. will be used.

Generally, a question Q

i

is any formula of the form

?

i

¦α

1

, . . . , α

n

¦,

where dQ

i

= ¦α

1

, . . . , α

n

¦ is the set of direct answers to a question Q. Direct

answers are formulas of our extended epistemic language L

KQ

cpl

and ques-

tions can be among direct answers as well. We suppose that dQ

i

is ﬁnite

with at least two syntactically distinct elements. In accordance with our

set-of-answers methodology the intended reading of a question Q

i

is:

Is it the case that α

1

or is it the case that α

2

. . . or is it the case

that α

n

?

Whenever I ask such question, I presuppose that at least one of the direct

answers is the case. Whenever I hear such question, I know that a questioner

presupposes the same, i.e., at least one of the direct answers is the case.

This brings us to an important term presupposition, which is studied in the

next subsection. On the contrary to our liberal SAM we require dQ

i

to be

ﬁnite. Working in propositional logics we want to keep direct answers as clear

epistemic possibilities, this seems to be useful in a communication processes.

Simultaneously, it makes easier the concept of ‘presupposing’.

Presuppositions

Taking an inspiration in inferential erotetic logic (see Section 2.2.2) we de-

ﬁne presuppositions of questions as formulas that are implied by each direct

answer. A presupposition is a ‘consequence’ of each direct answer, no matter

which answer is right.

2

2

Let us make a symbol convention: if it is not necessary to use the index i, we will

omit it.

50

3.2. Single-agent propositional epistemic logic and questions

Deﬁnition 17. A formula ϕ is a presupposition of a question Q iﬀ (α → ϕ)

is valid for each α ∈ dQ. We write ϕ ∈ PresQ.

The set of presuppositions of a question is full of redundant formulas. This

leads to the deﬁnition of maximal presuppositions. Maximal presuppositions

imply every presupposition.

Deﬁnition 18. A formula ϕ is a maximal presupposition iﬀ ϕ ∈ PresQ and

(ϕ → ψ) is valid for each ψ ∈ PresQ.

Example 13. A formula (α

1

∨ . . . ∨ α

n

) is a maximal presupposition of a

question ?¦α

1

, . . . , α

n

¦.

The theory of questions in IEL introduces one more term—prospective

presupposition. The truth of a prospective presupposition at a state of a

model gives the truth of some direct answer at this state. This can be a

modal reformulation of the original IEL deﬁnition.

Deﬁnition 19. A formula ϕ is a prospective presupposition of a question Q

iﬀ ϕ ∈ PresQ and, for all models M and states s, if (M, s) [= ϕ, then there

is a direct answer α ∈ dQ such that (M, s) [= α. We write ϕ ∈ PPresQ.

A formula (α

1

∨ . . . ∨ α

n

) is a prospective presupposition of a question

?¦α

1

, . . . , α

n

¦ as well.

Because of working with ﬁnite sets of direct answers in a system extend-

ing classical propositional logic, things are easier. We need not distinguish

between maximal and prospective presuppositions.

3

Theorem 13. The set of prospective presuppositions is equal to the set of

maximal presuppositions of a question Q.

Proof. First, let ϕ ∈ PPresQ but ϕ is not maximal. Since ϕ is not maximal,

there must be (M, s) and ψ ∈ PresQ such that (M, s) [= ϕ and (M, s) [= ψ.

(M, s) [= ϕ implies the existence of a direct answer α satisﬁed in (M, s).

Each presupposition is implied by every direct answer, so is ψ in (M, s) and

it gives (M, s) [= ψ, which is a contradiction.

Second, let ϕ be maximal, but not prospective. In our (ﬁnite) case we

can suppose that Q has at least one prospective presupposition. If ϕ is

not prospective, then there is a state (M, s) [= ϕ and (M, s) [= α, for each

α ∈ dQ. All presuppositions are satisﬁed in the state (M, s), so is prospective

presuppositions, but it is in contradiction with the fact that no α is valid in

(M, s).

3

In IEL, prospective presuppositions are maximal, but not vice versa, see [44, Corollary

4.10].

51

3.2. Single-agent propositional epistemic logic and questions

In the next theorem we show the same result we obtained for IEL (see

Lemma 1): All prospective presuppositions of a question are equivalent.

Theorem 14. If ϕ, ψ ∈ PPresQ, then ϕ ≡ ψ.

Proof. For proving semantic equivalence we have to prove ϕ [= ψ as well as

ψ [= ϕ.

If (M, s) [= ϕ, then there is α ∈ dQ such that (M, s) [= α. Since

ψ ∈ PresQ, then (M, s) [= (α → ψ) gives (M, s) [= ψ. We have obtained

ϕ [= ψ.

The other case is similar.

Thus, the symbol PPresQ will be used for a formula representing prospec-

tive presuppositions of a question Q modulo the semantic equivalence.

Note on presupposing and context In the example with card players

we mentioned ‘reasonable’ Catherine’s question

Who has the Joker: Ann, or Bill?

In item 3 (see p. 48) we wrote that either Ann or Bill has the Joker is

Catherine’s presupposition, i.e., it must be the case that Ann has the Joker

(and Bill not), or it must be the case that Bill has the Joker (and Ann not).

This is indicated by comma in the interrogative sentence as well. Catherine’s

presupposition is under inﬂuence of the context given by the rules of the card

‘game’:

Just one Joker is distributed among the agents Ann, Bill, and

Catherine.

Now, let us concern the following question:

What is Peter: a lawyer or en economist?

If there is no supplementary context, the question bears a presupposition

that Peter is at least one of the two possibilities (maybe, both of them).

However, the formalization of both questions would be almost the same, it

is expected the form

?¦α, β¦.

The role of context will be studied later on, viz. subsection Relativized ask-

ability in 3.2.2, especially.

52

3.2. Single-agent propositional epistemic logic and questions

Askable questions

In semantics for a majority of logical systems we speak about truth or falsity

of a formula (in a particular state of a particular model). It is clear that it

makes little sense to speak about truth/falsity of a question. We introduce

instead a concept of askability of a question. Askability is based on our idea

of a ‘reasonable’ question in a certain situation. ‘Reasonability’ corresponds

to the three conditions we informally mentioned at the introductory example.

Let us repeat and name them:

1. Non-triviality It is not reasonable to ask a question if the answer

is known.

2. Admissibility Each direct answer is considered as possible.

3. Context At least one of the direct answers must be the right one.

Whenever an agent-questioner poses a question, she does not know any (di-

rect) answer to a question, but, simultaneously, she considers all (direct)

answers possible and she is aware of what is presupposed—she knows the

prospective presupposition of a question. The formal deﬁnition follows.

Deﬁnition 20. It holds for a question Q

i

= ?

i

¦α

1

, . . . , α

n

¦ that

(M, s) [= Q

i

iﬀ

1. (M, s) [= [i]α, for each α ∈ dQ

i

2. (M, s) [= 'i`α, for each α ∈ dQ

i

3. (M, s) [= [i]PPresQ

i

Then we say that Q

i

is askable in the state (M, s) (by an agent i).

As we can see, the freedom in the syntactical form of questions was com-

pensated by restrictions in their semantics. We say that a question is (gen-

erally) askable iﬀ there is a model and a state where the question is askable

(by an agent). Askable questions include neither contradiction nor tautology

among their direct answers. The former is excluded by the second condition

and the latter by the ﬁrst one. A question Q

i

is askable relative to a model

M by an agent i (let us write M [= Q

i

) iﬀ (M, s) [= Q

i

for each s ∈ S. The

deﬁnition of [= Q

i

is straightforward, but there are no ‘tautological’ questions

in K. A question is not askable in a state without successors. In our version

53

3.2. Single-agent propositional epistemic logic and questions

at least two successors are needed. If we work in systems extending classical

logic, the ﬁrst condition is equal to (M, s) [= [i]α, i.e., (M, s) [= 'i`α, for

each α ∈ dQ

i

. We can see the questioner as admitting the possibility of α

for each direct answer α to a question Q

i

. In these systems, questions are

complex modal formulas. However, Deﬁnition 20 is meant in a full generality

without the intention of reduction of questions to the epistemic language.

The epistemic semantic viewpoint represents agent’s ‘knowledge’ in a

state s as an afterset sR

i

given by the states related to s by an accessibility

relation R

i

, i.e., sR

i

= ¦s

: sR

i

s

¦.

s

s

.

.

.

`

Let us return to the following question:

What is Peter: a lawyer or an economist?

This question can be formalized by a formula ?

i

¦α, β¦. Its askability in a state

(M, s) requires a substructure on sR

i

consisting of (at least) two accessible

states, one of them satisﬁes α (and does not satisfy β) and the other one

β (and does not satisfy α). All states in sR

i

must satisfy the prospective

presupposition (α ∨ β), because of the context condition.

s

1

α, (α ∨ β)

?

i

¦α, β¦ s

`

s

2

β, (α ∨ β)

Of course, this is a minimal requirement given by askability conditions,

the complete afterset structure can contain other states, some of them may

satisfy both α and β, but none of them satisﬁes α and β: the question

?

i

¦α, β¦ does not consider the answer neither α nor β as possible (context

condition)—such answer would be accepted, e.g., by the question

?

i

¦α, β, (α ∧ β)¦

States in the afterset sR

i

are understood as epistemic possibilities. In accor-

dance with the non-triviality condition neither α nor β can be true in all of

them. Finally, admissibility condition requires that there must be at least

one ‘α-state’ and at least one ‘β-state’ in sR

i

.

54

3.2. Single-agent propositional epistemic logic and questions

3.2.2 Some important classes of questions

In this section we introduce some classes of questions with their semantic

behavior. Yes-no questions and conjunctive questions were introduced in the

previous chapters. In the included subsection we suggest to link conditional

and hypothetical questions with the role of a context, i.e., an auxiliary set of

formulas. The names of classes originate from IEL (cf. [44]).

The very basic questions in their syntactical as well as semantical form

are yes-no questions.

Is Prague the capital of the Czech Republic?

is a question requiring one of the following answers:

• (Yes,) Prague is the capital of the Czech Republic.

• (No,) Prague is not the capital of the Czech Republic.

with the formalization ?

i

¦α, α¦, which is shortly written as ?

i

α. This ques-

tion is askable in a state s if there are (at least) two diﬀerent states available

from s, one satisﬁes α and the other one α. The afterset sR

i

is supposed

to have this form:

s

1

α

?

i

α s

`

s

2

α

Reviewing the askability conditions for ?

i

α we can see that this question

is equivalent to a formula 'i`α ∧ 'i`α. In our system, yes-no questions can

be seen as a ‘contingency modality’. The same requirements are posed by

askability conditions for ?

i

α. In our case, both ?

i

α and ?

i

α are equiv-

alent. Yes-no questions always form a partitioning on aftersets and their

presuppositions are tautologies. Questions with presuppositions, which are

all tautological, are called safe.

Deﬁnition 21. A question Q

i

is safe iﬀ PPresQ

i

is valid.

Questions that are not safe, will be called risky.

4

Another example of safe questions are conjunctive questions. The shortest

one is ?

i

¦(α∧β), (α∧β), (α∧β), (α∧β)¦ asking for a full description

based on α and β. We write it as ?

i

[α, β[ and the following ﬁgure shows the

required substructure on the afterset.

4

The original concepts of safety and riskiness of questions come from Nuel Belnap. See

chapter 2 as well.

55

3.2. Single-agent propositional epistemic logic and questions

s

1

α, β

?

i

[α, β[ s −→ s

2

α, β

↓ `

s

3

s

4

α, β

α, β

Similarly to yes-no questions they have the exhaustive set of direct an-

swers and direct answers are mutually exclusive. In section 3.4 we will deal

with some restrictions posed on direct answers, which will enable us to clarify

answerhood conditions.

The question

What is Peter: a lawyer or an economist?

with a formalization ?

i

¦α, β¦ is a risky one. However, asking this question in

a state an agent does not admit that she can see states where PPres(?

i

¦α, β¦)

is not satisﬁed. Let us deﬁne this kind of local safeness:

Deﬁnition 22. A question Q is safe in a state (M, s) (for an agent i) iﬀ

(M, s

1

) [= PPresQ

i

, for each s

1

∈ sR

i

.

Thus, an askable question in a state (for an agent) is safe in this state

(for this agent).

Relativized askability

Let us consider the following question:

Did you stop smoking?

At ﬁrst sight, it is a yes-no question, but seeing both answers it seems, there

can be something more what is presupposed:

• Yes, I did can mean I had smoked and stopped.

• No, I didn’t can mean I smoked and go on.

Both of them presuppose the smoking in the past. Such question is an

example of a conditional yes-no question with a formalization ?

i

¦(α∧β), (α∧

β)¦. Generally, conditional questions are of the form

?

i

¦α ∧ β

1

, α ∧ β

2

, . . . , α ∧ β

n

¦

56

3.2. Single-agent propositional epistemic logic and questions

The askability of a conditional question in a state s requires the validity

of α in each accessible state, i.e., an agent ‘knows’ α in s. The notion of

an askable conditional question can be generalized with respect to a set of

formulas. This leads to relativized askability.

5

Deﬁnition 23. A question Q is askable (by an agent i) in (M, s) with re-

spect to a set of formulas Γ iﬀ (M, s) [= [i]Γ and (M, s) [= Q

i

. By [i]Γ we

abbreviate the set ¦[i]γ [ γ ∈ Γ¦. Then we write (M, s) [= (Γ, Q)

i

.

A conditional question ?

i

¦α ∧ β

1

, α ∧ β

2

, . . . , α ∧ β

n

¦ is askable in s if

and only if ?

i

¦β

1

, . . . , β

n

¦ is askable there with respect to the auxiliary set

(knowledge database) ¦α¦.

The term relativized askability will be mostly used for pointing out the im-

portance of a set Γ. Every question askable at a state is askable with respect

to its set of (prospective) presuppositions. It has an expected consequence:

Fact 27. If (M, s) [= (Γ, Q)

i

, then (M, s) [= (∆, Q)

i

, for each ∆ ⊆ Γ.

However, relativized askability is not ‘monotonic’ in knowledge databases.

If (M, s) [= (Γ, Q)

i

, then it need not be (M, s) [= (∆, Q)

i

, for ∆ ⊃ Γ.

Relativized askability of this kind can be used for an explicit expressing

of the knowledge structure. Catherine’s question

Who has the Joker: Ann, or Bill?

can be formalized by

(¦α ∨ β¦, ?¦α, β¦)

c

In addition, IEL introduces one more term, hypothetical question, which is

a bit similar to the previous one. A natural language example of hypothetical

yes-no question might be

If you open the door, will you see a bedroom?

with a formalization ?

i

¦(α → β), (α → β)¦. A general hypothetical ques-

tion is then

?

i

¦α → β

1

, . . . , α → β

n

¦

Again, the askability of such questions can be understood as based on agent’s

hypothetical knowledge. Our interpretation is: if α is known, then it is to be

decided whether β

1

, or β

2

, etc. Using a generalization similar to Deﬁnition 23

we obtain

5

This term corresponds to question in an information set introduced in [13].

57

3.3. Epistemic erotetic implication

Deﬁnition 24. A question Q is askable (by an agent i) in (M, s) with respect

to a set of hypotheses Γ iﬀ (M, s) [= [i]Γ implies (M, s) [= Q

i

. By [i]Γ we

abbreviate the set ¦[i]γ [ γ ∈ Γ¦. Let us write (M, s) [= Γ

i

→ Q

Askability of a conditional question ensures the askability of a hypothet-

ical one.

Fact 28. If (M, s) [= (Γ, Q)

i

, then (M, s) [= Γ

i

→ Q.

The only diﬀerence of both deﬁnitions lies in the words and and implies.

We will return to these concepts in Section 3.5.

3.3 Epistemic erotetic implication

Erotetic inference is implicitly based on the (standard) implication. We say

that a question Q

1

implies Q

2

(in a state s, for an agent i) whenever askability

of Q

1

(in s, for i) implies askability of Q

2

(in s, for i).

(M, s) [= Q

i

1

→ Q

i

2

iﬀ (M, s) [= Q

i

1

implies (M, s) [= Q

i

2

We have mentioned that questions ?

i

α and ?

i

α have the same askability

conditions. The eqivalence of both questions is a theorem in our system

based on modal logic K. Let us omit the index i for now.

Example 14. Both (?α → ?α) and (?α → ?α) are valid.

The informal meaning of epistemic erotetic implication is very transpar-

ent. Whenever an agent asks Q, then she can ask every question implied by

Q. A question in antecedent is ‘more complex’ then the implied one. Im-

plied question’s required substructure on the afterset must be a substructure

of that required by an implying one. The question

What is Peter: a lawyer or an economist?

implies

Is Peter a lawyer?

as well as

Is Peter an economist?

This can be generalized:

Example 15. ?¦α

1

, . . . , α

n

¦ → ?α

j

is valid, for each j ∈ ¦1, . . . , n¦.

58

3.3. Epistemic erotetic implication

The following example shows a special position of conjunctive questions

in implications.

Example 16. The following implications are valid:

1. ?[α

1

, . . . , α

n

[ → ?α

j

, for each j ∈ ¦1, . . . , n¦.

2. ?[α, β[ → ?(α ◦ β), where ◦ is any truth-functional constant.

3. ?[α, β[ → ?¦α, β, (α ∧ β)¦

4. ?[α, β, γ[ → ?[α, β[

Let us notice that conjunctive questions are safe and they imply safe

questions again. We can prove that it is a rule.

Theorem 15. If Q

1

is safe and Q

1

→ Q

2

valid, then Q

2

is safe.

Proof. Let us suppose, Q

1

is safe and (Q

1

→ Q

2

) is valid, but Q

2

is not

safe. We take a model M and a state (M, s) where (M, s) [= Q

1

, then

(M, s) [= Q

2

. Let us slightly change the afterset of s. We add a single point

s

1

accessible from s where the prospective presupposition of Q

2

is invalid.

(M, s

1

) [= [i]

l

1

β

j

, for dQ

2

= ¦β

1

, . . . β

l

¦. In this model the askability of Q

1

is not violated at s

1

, (M, s

1

) [= Q

1

, but Q

2

is not askable here. This leads to

a contradiction.

We have to be careful when speaking about ‘complexity’ of questions. Is

question ?¦α, β¦ more complex than ?[α, β[? It is easy to check that neither

?[α, β[ → ?¦α, β¦ nor ?¦α, β¦ → ?[α, β[ are valid. In both cases there is

a problem with the context condition; a risky question ?¦α, β¦ requires the

validity of the disjunction (α ∨ β) on the afterset. The next two examples

emphasize the importance of context condition again.

Example 17. [= ?¦α, β¦ → ?¦α, β¦ as well as [= ?¦α, β¦ → ?¦α, β¦

Example 18. [= ?¦α, β, γ¦ → ?¦α, β¦

The question ?

i

¦α, β, γ¦ requires [i](α∨β∨γ), but ?

i

¦α, β¦ requires ‘only’

[i](α ∨ β), which can fail in the structure suﬃcient for the askability of the

ﬁrst question.

An implying question shares presuppositions with the implied one.

Fact 29. If Q

1

→ Q

2

is valid, then if ϕ ∈ PresQ

1

, then ϕ ∈ PresQ

2

.

Epistemic erotetic implication has the expected property—transitivity:

Fact 30. If (M, s) [= Q

1

→ Q

2

and (M, s) [= Q

2

→ Q

3

, then (M, s) [=

Q

1

→ Q

3

.

59

3.4. Askability and answerhood

3.4 Askability and answerhood

Epistemic erotetic implication creates a relationship among questions. If

a question is askable at a state, so is every implied one. In our system,

Q

1

→ Q

2

if and only if Q

2

→ Q

1

. If an implied question is inaskable, so

is the implying one. The askability of a question consists of the validity of

three conditions (non-triviality, admissibility, and context) and inaskability is

a result of the violation of at least one of them. Let us imagine that we know

that Q

1

→ Q

2

and we have the answer to Q

2

(non-triviality condition fails),

then we are sure that Q

1

is inaskable, but does it mean to have an answer

to Q

1

? What if Q

1

is inaskable because of failing context condition? In this

section we will just deal with such violations of askability conditions, proper-

ties of inaskable questions from various classes of questions, and answerhood

conditions—complete and partial answers will be introduced.

To break the non-triviality condition means that there is a direct answer

which is ‘known’ by an agent (in a state of a model). In fact, an agent knows

a direct answer even if she knows a formula that is equivalent to a direct

answer or she knows a formula from which some direct answer follows. Such

formula is called complete answer.

6

Let us deﬁne the concept a question is answered at a state:

7

Deﬁnition 25. A question Q

i

= ?

i

¦α

1

, . . . , α

n

¦ is answered in (M, s) (for

an agent i) iﬀ (M, s) [=

α

j

∈dQ

i

([i]α

j

). We write (M, s) [= A

i

Q.

The case of invalid admissibility condition is a bit diﬀerent. Let us imag-

ine that our agent knows α in a state s. Then, even if she does not know an

answer to a question ?[α, β[ in that state, it is not right to ask this question.

All possibilities required by the admissibility condition are not available, in

particular, accessible states with ¦α, β¦ and ¦α, β¦ are missing. Some

answers to the question ?[α, β[ give information that is superﬂuous in the

state of agent’s knowledge. Formula α is a partial answer to a question

?[α, β[. Partial answer excludes some of the (direct) answers.

Deﬁnition 26. A question Q

i

= ?

i

¦α

1

, . . . , α

n

¦ is partially answered in

(M, s) (for an agent i) iﬀ (M, s) [=

α

j

∈dQ

i

([i]α

j

). We write (M, s) [=

P

i

Q.

In fact, if the admissibility condition fails, there is a direct answer which

is not considered as possible, i.e., (∃α ∈ dQ

i

)((M, s) [= 'i`α), which is

6

In [13] we can ﬁnd two terms: to be and to give a semantic answer. This distinction

is not necessary here.

7

We do not discuss a dynamic approach just now. An answered question is a ‘poten-

tially’ answered question, in fact, an answer need not be uttered among agents.

60

3.4. Askability and answerhood

equivalent to (∃α ∈ dQ

i

)((M, s) [= [i]α) in our system. This means that

our agent can answer the question ?α in (M, s): (∃α ∈ dQ

i

)((M, s) [= A

i

?α).

This and example 15 give the proof of the following lemma.

Lemma 10. If (M, s) [= P

i

Q, then there is a formula ϕ such that (M, s) [=

A

i

?ϕ and Q

i

→ ?

i

ϕ is valid.

Let us suppose a question Q is answered. Does it mean that Q is partially

answered? Surprisingly, it is not true that if (M, s) [= A

i

Q, then (M, s) [=

P

i

Q. See the next example.

Example 19. s

1

α, β

s

`

s

2

α, β

In the structure given by Example 19 the question ?¦α, β¦ is answered in

s (the agent knows β), but it is not partially answered; the agent is not able

to get the knowledge of either α or β, both α and β are still possible.

A

i

Q implies P

i

Q for questions with pairs of mutually exclusive direct an-

swers. It means that for each direct answer there is another one such that

both of them cannot be true. Yes-no questions as well as conjunctive ques-

tions are examples from this class. Their sets of direct answers satisfy a more

strict condition, they have mutually exclusive direct answers—the ‘truth’ of

a direct answer (in a state) means that no other direct answer is satisﬁed

there. However, both conditions are of a semantic nature, it can be caused

by a model and a state. Recall Catherine’s question

Who has the Joker: Ann, or Bill?

with the context, which is important here: it does not admit the afterset

substructure in Example 19. Mutual exclusiveness is not preserved by impli-

cation. We cannot prove anything similar to Theorem 15, in particular, the

answers of the question ?

i

¦α∧β, α∧β, α∧β¦ are mutually exclusive, but

this question implies ?

i

¦α, β¦, which is not in this class, generally.

Fact 31. If a question Q

i

has the set dQ

i

with pairs of mutually exclusive

direct answers, then A

i

Q → P

i

Q is valid.

The last option for inaskability is the violation of context condition. Then

an agent does not know (believe) the prospective presupposition of a question.

For example, the question

61

3.4. Askability and answerhood

Which town is the capital of the Czech Republic: Prague, or

Brno?

can be inaskable, although an agent does not know either complete or partial

answer, but she admits that there is another town, which could be the capital

of the Czech Republic, i.e., neither Prague nor Brno might be the right

answer. Then we say that a question is weakly presupposed by an agent.

Deﬁnition 27. A question Q

i

is weakly presupposed in (M, s) (by an agent

i) iﬀ (M, s) [= [i]PPresQ

i

. We write (M, s) [= W

i

Q.

An askable question in a state (for an agent i) satisﬁes at least the ‘safe-

ness in a state’ (see Deﬁnition 22). If a question Q is safe in (M, s), then Q

cannot be weakly presupposed in (M, s).

Theorem 16. If a question with pairs of mutually exclusive direct answers

Q

i

is safe in (M, s), then the following conditions are equivalent:

1. (M, s) [= Q

i

2. (M, s) [= P

i

Q

3. There is a formula ϕ such that (M, s) [= A

i

?ϕ and Q

i

→ ?

i

ϕ is valid.

Proof. (2⇒1) is clear and (2⇒3) is from Lemma 10.

(1⇒2) If (M, s) [= Q

i

, then there are three possibilities: (M, s) [= A

i

Q

or (M, s) [= P

i

Q or (M, s) [= W

i

Q. The last one is impossible because of

the safeness of Q

i

in (M, s). If (M, s) [= A

i

Q, then (M, s) [= P

i

Q (from

Fact 31).

(3⇒1) Let us suppose that there is a formula ϕ such that the question

?

i

ϕ is answered in (M, s). From Q

i

→ ?

i

ϕ we know that if (M, s) [= ?

i

ϕ,

then (M, s) [= Q

i

.

Partial answerhood of a question Q

i

in some state is equivalent to the

existence of a yes-no question, which is answered in that state and implied

by Q

i

. From the validity of Q

i

→ ?

i

ϕ we know that inaskability of ?

i

ϕ

8

implies inaskability of Q

i

and, therefore, ϕ (as well as ϕ) implies either

some α ∈ dQ

i

or α (for α ∈ dQ

i

).

8

(M, s) [= ?

i

ϕ iﬀ (M, s) [= A

i

?ϕ.

62

3.5. Context

3.5 Context

In subsection Relativized askability (page 56) we introduced askability

with respect to sets of formulas. While both kinds are understood as variants

of conditional or hypothetical questions, in some situations it can be useful to

display and emphasize the role of a context. Especially if it has an important

position in reasoning. Let us recall the example from Section 2.3.1, where

we discussed erotetic implication in IEL. An agent asking

Q

1

: What is Peter a graduate of: a faculty of law or a faculty of

economy?

can be satisﬁed by the answer

He is a lawyer.

even if she did not ask

Q

2

: What is Peter: a lawyer or an economist?

The connection between both questions could be established by the following

knowledge base Γ:

Someone is a graduate of a faculty of law iﬀ he/she is a lawyer.

Someone is a graduate of a faculty of economy iﬀ he/she is an

economist.

Relativized askability helps us to express that Q

1

implies Q

2

with respect to

an auxiliary set of formulas Γ, i.e., (Γ, Q

1

)

i

→ Q

i

2

. In the example, Q

1

can be

formalized by ?¦α

1

, α

2

¦, Q

2

by ?¦β

1

, β

2

¦, and Γ = ¦(α

1

↔ β

1

), (α

2

↔ β

2

)¦,

then

(¦(α

1

↔ β

1

), (α

2

↔ β

2

)¦, ?¦β

1

, β

2

¦)

i

→ ?

i

¦α

1

, α

2

¦

is valid. Moreover, the questions Q

1

and Q

2

are equivalent with respect to

Γ: (Γ, Q

1

)

i

→ Q

i

2

as well as (Γ, Q

2

)

i

→ Q

i

1

is valid.

The prime reason for introducing of the structures (Γ, Q)

i

is to keep the

importance of a context in inferences with questions. (Γ, Q)

i

can be consid-

ered as a generalization of conditional questions in our system. (Generalized)

conditional questions consist of two parts: conditional part (context) and

query part. As an easy conclusion of Fact 27 we receive that a conditional

question implies its query part: ?

i

¦α ∧ β

1

, . . . , α ∧ β

n

¦ → ?

i

¦β

1

, . . . , β

n

¦.

Fact 32. (Γ, Q)

i

→ Q

i

is valid formula.

63

3.5. Context

Hypothetical questions consist of such two parts as well, but it is not

valid that ?

i

¦α → β

1

, . . . , α → β

n

¦ → ?

i

¦β

1

, . . . , β

n

¦.

In some sense, we can see (generalized) hypothetical questions, Γ

i

→ Q,

as a counterpart of evocation in IEL. Similarly, (generalized) conditional

questions in the interplay with implication, (Γ, Q

1

)

i

→ Q

i

2

, seem to be a

counterpart of general erotetic implication in IEL. While the correspondence

could be seen in ‘philosophy’, these structures diﬀers in properties from IEL

ones.

Now, let us list some properties appearing by combination of conditional

and hypothetical questions with implication. Most of them are expected. For

example, the following fact points out the cumulativity of explicitly expressed

presuppositions.

Fact 33. If (M, s) [= (Γ, Q

1

)

i

→ Q

i

2

and (M, s) [= (∆, Q

2

)

i

→ Q

i

3

, then

(M, s) [= (Γ ∪ ∆, Q

1

)

i

→ Q

i

3

.

Relativized askability is transfered by implication; if (M, s) [= (Γ, Q

1

)

i

and (M, s) [= (Q

i

1

→ Q

i

2

), then (M, s) [= (Γ, Q

2

)

i

. And it is easy to check

the following generalization:

Lemma 11. Whenever (M, s) [= (Γ, Q

1

)

i

and (M, s) [= (Γ, Q

1

)

i

→ Q

i

2

, then

(M, s) [= (Γ, Q

2

)

i

.

The same result can be proved for generalized hypothetical questions: If

(M, s) [= Γ

i

→ Q

1

and (M, s) [= (Q

i

1

→ Q

i

2

), then (M, s) [= Γ

i

→ Q

2

.

Finally, we obtain

Theorem 17. If (M, s) [= Γ

i

→ Q

1

and (M, s) [= (Γ, Q

1

)

i

→ Q

i

2

, then

(M, s) [= Γ

i

→ Q

2

.

Proof. Let us suppose (M, s) [= Γ

i

→ Q

2

, i.e., (M, s) [= [i]Γ and (M, s) [=

Q

i

2

. From (M, s) [= [i]Γ and (M, s) [= Γ

i

→ Q

1

we get (M, s) [= Q

i

1

and

(M, s) [= (Γ, Q)

i

. Because of (Γ, Q

1

)

i

→ Q

i

2

we gain (M, s) [= Q

i

2

, which is a

contradiction.

It is necessary to point out that both variants of relativized askability

were introduced for the explicit expression of the context conditions of a

question. This approach is useful; however, (pure) questions and implication

are of prime importance in our setting.

64

3.6. Implied questions

3.6 Implied questions

If questions are in the implicational relationship, a transmission of askability

conditions from an implying question to the implied one is justiﬁed. An

implied question is understood as ‘less complex’ in its requirements posed on

the afterset substructure. Let us recall Example 15

9

where Q → ?α is valid

for each α ∈ dQ, thus

Q →

α∈dQ

(?α)

On the one hand, each question ?α is a yes-no question with many good

properties. On the other hand, they may be ‘worse’ in the description of an

agent’s knowledge/ignorance structure than the initial question Q. Might it

be ‘better’ if we consider the whole set of implied questions based on the set

of direct answers to an initial question?

If we consider the set of questions Φ = ¦?α

1

, ?α

2

, . . .¦, then inaskability

of some ?α means that the answer is either α or α. In the ﬁrst case Q is

answered as well, in the second one Q is partially answered. This can be

understood as a form of ‘suﬃciency’ condition of the set Φ: answerability

of its members implies at least partial answerability of the initial question

Q. It means, Φ does not include ‘useless’ questions. Moreover, we receive

one more property, Φ is ‘complete’ in a way: if Q is partially answered, then

there must be a question Q

j

∈ Φ that is answered. We can understand it as

a form of ‘reducibility’ of an initial question to a set of yes-no questions. In

comparison with IEL, this reducibility is purely based on implication.

Example 16 gives a similar result for conjunctive questions. The set of

yes-no questions can be formed by constituents of their direct answers:

?[α

1

, . . . , α

n

[ →

n

j=1

(?α

j

)

In this case, we arrive to really less complex questions, but having a partial

answer to ?[α

1

, . . . , α

n

[ does not give neither answer nor partial answer to

some α

j

. It could be useful in some cases. An agent can ask questions from

the set Φ = ¦?α

1

, . . . , ?α

n

¦ and complete her knowledge step by step. The

most important property is that the set Φ does not include useless ques-

tions. Generally speaking, in some communication processes it is useful to

conceal some knowledge or ignorance of a questioner—a criminal investiga-

tion is a nice example. An agent can ask questions from the set Φ without

completely revealing her knowledge structure. Asking a conjunctive question

9

Let us omit the index i in this section.

65

3.6. Implied questions

?[α

1

, . . . , α

n

[ publicly, everybody is informed that the agent-questioner does

not know anything with respect to α

1

, . . . , α

n

.

66

Chapter 4

A step to dynamization of

erotetic logic

4.1 Introduction

At the beginning of the previous chapter we introduced the language for

multi-agent propositional epistemic logic L

K

cpl

with the set of agents / =

¦1, . . . , m¦. If we add an accessibility relation for each agent to Kripke

frames, T = 'S, R

1

, . . . , R

m

`, we will obtain multi-modal system K with

box-like modalities [1], . . . , [m]. Although we often used ‘epistemic’ termi-

nology, especially in motivations, this system is not intended for knowledge

representation. In fact, there are many discussions about the best represen-

tation of knowledge as well as belief in the philosophy of logic. Nowadays,

such discussions are brought into life again in studies of substructural logics.

The term knowledge is often subjected to new interpretations based on a

background system.

1

The very aim in introducing questions in epistemic-like systems was to

provide an interpretation of questions, which agrees with the interplay of the

idea of representing the knowledge and ignorance structure of a questioner in

the process of asking. The interpretation of questions should be mostly in-

dependent of a background system. In our philosophy, ‘knowledge structure’

and its representation is considered to be primary. This chapter is devoted

to multi-agent epistemic logic with questions based on modal system S5 and

its dynamic extension—public announcement logic.

We often referred to the importance of questions in communication pro-

cesses. This is understood as an information exchange among agents in a

group. The delivering of information in a group has the beneﬁt of (public)

1

See, e.g., the deﬁnition of knowledge modality inside relevant logic in [2].

67

4.2. Multi-agent propositional epistemic logic with questions

announcements and, as a result, there is a change of epistemic states of group

members.

S5 represents standard epistemic logic, cf. [42, 9], where knowledge is

factive and fully introspective (positively as well as negatively). Of course,

this system is subjected to criticism, see [9, 8] where the ‘logical omniscience

problem’ seems to be the most criticized aspect. However, the goal of this

chapter is to show the role of questions in a formal dynamic-epistemic sys-

tem. So, we are not going to solve any problem of this formal epistemic

representation nor to follow philosophical discussions on it.

First of all we extend the erotetic epistemic framework by group questions

and group epistemic modalities (group knowledge, common knowledge, and

distributed knowledge). This makes it possible to speak about answerhood

conditions for groups of agents. Then we apply the public announcement

modality in the process of answer mining among agents.

4.2 Multi-agent propositional epistemic logic

with questions

We have just said that our epistemic framework would be based on multi-mo-

dal logic S5. Being still in language L

K

cpl

we only have to add that each acces-

sibility relation R

i

is equivalence, i.e., reﬂexive, transitive, and symmetric re-

lation. Accessibility relations seem to play a bit diﬀerent role now. In logic K

we understand the accessible states as (epistemic) ‘alternatives’ for an actual

state, an agent can see ‘possibilities’. The equivalence relation makes con-

nected states indistinguishable, an agent considers them as having the same

‘value’. Let us recall the group of three friends and just one Joker-card dis-

tributed among them. From Catherine’s viewpoint both possibilities—either

Ann has the Joker or Bill has the Joker—are indistinguishable:

s

1

R

c

s

2

Ann has the Joker ←→ Bill has the Joker

4.2.1 Group epistemic modalities

So far we have worked with personal knowledge. However, in multi-agent

systems we are obliged to introduce new modalities to reﬂect epistemic states

in groups of agents. The language L

K

cpl

will be extended by symbols E

G

, C

G

,

and D

G

, where G ⊆ / is a group of agents.

68

4.2. Multi-agent propositional epistemic logic with questions

Group knowledge E

G

ϕ means

Each agent (from G) knows ϕ.

and is fully deﬁnable by personal knowledges of G-members:

E

G

ϕ ↔

i∈G

[i]ϕ

We shall call it group knowledge. Let us stress that E

G

does not guarantee

that a member of the group G knows that she shares the same information

with some other members of the group.

Common knowledge The group modality C

G

is stronger in the following

sense: C

G

ϕ requires not only that ϕ is a group knowledge, but also that this

fact is reﬂected by everybody in G.

Each agent (from G) knows ϕ and each agent knows that each

agent knows ϕ and each agent knows that each agent knows that

each agent knows ϕ and . . .

C

G

is called common knowledge and expresses that the knowledge is max-

imally shared by everybody in G, each agent is aware of this sharing. C

G

can be seen as an inﬁnite conjunction of all ﬁnite iterations of the group

knowledge E

G

:

C

G

ϕ ↔ E

G

ϕ ∧ E

G

E

G

ϕ ∧ E

G

E

G

E

G

ϕ ∧ . . .

The system S5C is obtained by adding the operator C

G

and the semantic

clause:

• (M, s) [= C

G

ϕ iﬀ (M, s

1

) [= ϕ for each s

1

such that s

¸

i∈G

R

i

∗

s

1

¸

i∈G

R

i

∗

is a reﬂexive and transitive closure of

¸

i∈G

R

i

and it means that

s

1

is accessible from s by each R

i

(i ∈ G) in k steps, for any k ≥ 0.

As we said, E

G

is deﬁnable in the language L

K

cpl

, so adding group knowl-

edge is just a conservative extension of the background multimodal epistemic

logic S5. Both languages L

K

cpl

and L

KE

cpl

have the same expressivity. However,

this is not the case of common knowledge. Multi-modal epistemic logic with

common knowledge S5C is not compact, as it is indicated in the deﬁnition

of C

G

and there exists formula in language L

KC

cpl

, which can distinguish two

models that are indistinguishable in language L

K

cpl

, see [42, p. 227].

The relationship of introduced epistemic modalities is the following:

69

4.2. Multi-agent propositional epistemic logic with questions

Fact 34. C

G

ϕ → E

G

ϕ → [i]ϕ is valid in S5C for each i ∈ G.

Common knowledge is essential for collective behavior and coordination

of collective actions. In game theory it is often presupposed that rules of a

game are shared by players. It is important to know rules, to know that the

other players know the same rules, to know that they know that we know

it, and so on. In case of questions we considered a question to be (partially)

answered for an agent if she knows a fact based on a direct answer. However,

when we say that a question is (partially) answered for a group of agents?

An answer must not be only known by all members, but it must be generally

known that it is known. Just common knowledge is a good candidate for

group answerhood conditions.

Deﬁnition 28. • A question Q is answered in (M, s) for a group G iﬀ

there is α ∈ dQ such that (M, s) [= C

G

α.

• A question Q is partially answered in (M, s) for a group G iﬀ there is

α ∈ dQ such that (M, s) [= C

G

(α).

Distributed knowledge The last group modality is a bit of another kind.

Let us remind the group of three friends and suppose that Ann has the

Joker. Although neither Catherine nor Bill know it, the knowledge of the

Joker-owner is implicitly contained in the group. If the agents can communi-

cate, they easily reach the hidden fact that Ann has the card. The standard

meaning of D

G

ϕ is given by the semantic clause:

• (M, s) [= D

G

ϕ iﬀ (M, s

1

) [= ϕ for each s

1

such that s

¸

i∈G

R

i

s

1

ϕ is true in all states that are accessible for every member in G. D

G

is called

distributed or implicit knowledge. The term distributed knowledge coincides

with the idea of pooling agents’ knowledge together. Let us imagine that

a solution of some problem can be obtained by the collecting of particular

data from each member of a group of agents. The crucial data are distributed

among agents, but nobody can solve the problem alone because of the need

of the other data.

If an agent knows ϕ, then ϕ is distributed knowledge in every agent’s

group:

Fact 35. [i]ϕ → D

G

ϕ is valid in S5CD for each i ∈ G.

The accessibility relation based on D

G

is a subset of each R

i

. Adding D

G

to language L

K

cpl

does not increase its expressivity.

2

2

Axioms and properties can be found in [23].

70

4.2. Multi-agent propositional epistemic logic with questions

If there is distributed knowledge for a group of agents, then it is dis-

tributed knowledge for every bigger group.

Fact 36. D

G

ϕ → D

G

ϕ is valid in S5CD for G ⊆ G

.

Again, this nicely shows the idea of a hidden information; we can obtain

it by a communication of (only) some agents in the group G

. The role of

distributed knowledge in answerhood will be discussed in the next section.

4.2.2 Group questions and answerhood

We introduced questions in a form of an agent’s personal task that ﬁts in her

knowledge structure. Whenever she wants to ﬁnd an answer to a question,

she has to communicate the question and ﬁnd someone who can answer it.

Being in a group of colleagues she asks the question and, in the best case,

there is somebody who knows an answer. A worse case is that nobody can

answer the question—the question is the task for them as well. Such question

is askable by each member of a group G and we shall call it group question.

Deﬁnition 29. A question Q is an askable group question in (M, s) (for a

group of agents G) iﬀ (∀i ∈ G)((M, s) [= Q

i

). Let us write (M, s) [= Q

G

.

Whenever there is a question which is (partially) answered by at least one

agent in a group, then we can see how to reach a (partial) answer. A question

must be publicly posed and the answer is a result of a communication.

Group questions seem to be a worse problem, there is no agent with a

(partial) answer to it in a group. If an answer should be sought inside the

group, there is only one chance to ﬁnd it. Again communication is important

for to discover ‘hidden’ information, i.e., an answer is present in the group

as distributed knowledge.

Let us have a group of two agents a and b. The following example shows

their knowledge structure:

Example 20. a b

s

1

←→ s

2

←→ s

3

α α α

β β β

Agent a cannot distinguish states s

1

and s

2

and knows α, agent b cannot

distinguish s

2

and s

3

and knows β. However, neither of them is able to

(partially) answer the yes-no question ?(α → β), it is their group question

?

{a,b}

(α → β). If they communicate, they recognize the state s

2

to be common

for them. This brings us to the term implicitly (partially) answered question.

71

4.2. Multi-agent propositional epistemic logic with questions

Deﬁnition 30. • A question Q is implicitly answered in (M, s) by a

group of agents G iﬀ (∃α ∈ dQ)((M, s) [= D

G

α).

• A question Q is implicitly partially answered in (M, s) by a group of

agents G iﬀ (∃α ∈ dQ)((M, s) [= D

G

α).

Back to the example, what should agents a and b communicate to gain an

answer to ?(α → β)? We can ﬁnd an inspiration in implied yes-no questions.

The question ?(α → β) implies the disjunction of questions ?α and ?β:

?(α → β) → (?α ∨ ?β)

Moreover, the agent a can completely answer ?α and the other one can

completely answer the question ?β. So, it would be useful to communicate

questions ?α and ?β in the group.

Generally, we can prove that if there is a set of questions, their disjunction

is implied by an initial question, and each question from the set can be

(partially) answered by some agent from a group G, then the initial question

is implicitly (partially) answered.

Theorem 18. If there is a set of questions Φ such that each Q

k

∈ Φ is

(partially) answered in (M, s) by some agent i ∈ G and (M, s) [= Q →

Q

k

∈Φ

Q

k

, then Q is implicitly (partially) answered in (M, s).

Proof. If Q is not sound group question, then Φ = ¦Q¦ and there is an agent

having (partial) answer to Q. From Facts 34 and 35 we get that it is implicit

(partial) answer in a group G.

Let us suppose Q is a sound group question in (M, s), then (partial)

answers to questions from Φ are distributed among agents from G. From

(M, s) [= Q →

Q

k

∈Φ

Q

k

we get (M, s) [=

Q

k

∈Φ

Q

k

→ Q. Unsoundness

of Q in (M, s) cannot be caused by violating of context condition because

of its status of sound group question. Now, let us introduce a new agent a,

which pools knowledge of all agents in a group G together. R

a

=

¸

i∈G

R

i

, if

R

a

= ∅, then everything is distributed knowledge. Let R

a

be nonempty. All

questions Q

k

∈ Φ are unsound for a in (M, s), so is Q, and a’s knowledge of

(partial) answer is in afterset sR

a

.

The content of the theorem is based on the S5CD-valid rule

(ψ

1

∧ . . . ∧ ψ

m

) → ϕ

([l

1

]ψ

1

∧ . . . ∧ [l

m

]ψ

m

) → D

{l

1

,...,l

m

}

ϕ

(4.1)

which expresses the mentioned idea of pooling agents’ knowledge together

for getting their distributed knowledge.

72

4.3. Public announcement

A question, which is posed among agents, can be (partially) answered

only if it is at least implicitly (partially) answered by a group. The next

section shows one of the ways of communication formalization in the role of

‘answer mining’.

4.3 Public announcement

Let us return to the group of three friends—Ann, Bill, and Catherine. It is

group knowledge that each of them has one card and nobody knows the cards

of the others and that one of the cards is the Joker. Ann received the Joker,

but neither Bill nor Catherine know which of the other two friends, has it.

In particular, both of them are not able to distinguish between the states

where Ann has the Joker and where she has not. If Ann publicly announces

“I’ve got the Joker.”,

everybody in the group learns this fact. Possible worlds (states) where Ann

does not have the Joker are excluded from the (epistemic) models of both

Bill and Catherine.

Our example gives a typical situation represented in the public announce-

ment logic—after a public announcement of a statement ϕ (“I’ve got the

Joker”), some other statement ψ holds, e.g., Bill knows Ann has the Joker

and Catherine knows Ann has the Joker. In fact, the author of an announced

statement is irrelevant in our framework. The statement is understood as in-

formation coming to each member of a group in the same way. From this

viewpoint Ann’s announcement has the same eﬀect as if an external observer

announces Ann has the Joker.

Formally we introduce logic of public announcement as an extension of

the system S5, cf. [42]. We deﬁne a box-like operator [ ], such that the

intended meaning of [ϕ]ψ is:

After the public announcement of ϕ, it holds that ψ.

The semantics of the new announcement operator is given by the following

clause:

• (M, s) [= [ϕ]ψ iﬀ (M, s) [= ϕ implies (M[

ϕ

, s) [= ψ

where M[

ϕ

= ''S

, R

1

, . . . , R

m

`, v

` is deﬁned as follows:

S

= ¦s ∈ S [ s [= ϕ¦

R

i

= R

i

∩ S

2

v

(p) = v(p) ∩ S

73

4.3. Public announcement

The model M[

ϕ

is obtained from M by deleting of all states where ϕ is not

true and by the corresponding restrictions of accessibility relations and the

valuation function. Again we can introduce a dual operator ' ` deﬁned in a

standard way as 'ϕ`ψ iﬀ [ϕ]ψ. If we rewrite the corresponding semantic

clause, we obtain

• (M, s) [= 'ϕ`ψ iﬀ (M, s) [= ϕ and (M[

ϕ

, s) [= ψ

The intended meaning of the dual operator is ‘after a truthful announcement

of ϕ, it holds that ψ’. It is easy to see that the diamond-like operator is

stronger:

Lemma 12. 'ϕ`ψ → [ϕ]ψ is valid.

The language L

K[]

cpl

has the same expressive power as the language L

K

cpl

.

This is demonstrated by the following lemma, which provides a reduction of

formulas with the public announcement operator to the epistemic ones. The

corresponding equivalences give, in fact, an axiomatization of the announce-

ment operator in the public announcement epistemic logic without common

knowledge, cf. [42, p. 81].

Lemma 13. The following equivalences are valid in S5 with public announce-

ment modality (where ◦ ∈ ¦∧, ∨, →¦):

[ϕ]p ↔ (ϕ → p)

[ϕ]ψ ↔ (ϕ → [ϕ]ψ)

[ϕ](ψ ◦ χ) ↔ ([ϕ]ψ ◦ [ϕ]χ)

[ϕ][i]ψ ↔ (ϕ → [i][ϕ]ψ)

[ϕ][ψ]χ ↔ [ϕ ∧ [ϕ]ψ]χ

For common knowledge there is no such reduction (axiom), the language

L

KC[]

cpl

is more expressive than L

KC

cpl

[42, p. 232]. We only have a rule describing

the relationship between the public announcement and common knowledge,

e.g., the one introduced in [42, p. 83]:

(χ ∧ ϕ) → [ϕ]ψ ∧ E

G

χ

(χ ∧ ϕ) → [ϕ]C

G

ψ

(4.2)

From now, our formal work will proceed in the rich propositional language

L

KECDQ[]

cpl

with formulas deﬁned as follows:

ϕ ::= p [ ψ [ ψ

1

∨ ψ

2

[ ψ

1

∧ ψ

2

[ ψ

1

→ ψ

2

[ ψ

1

↔ ψ

2

[

[i]ψ [ E

G

ψ [ C

G

ψ [ D

G

ψ [

?

i

¦ψ

1

, . . . , ψ

n

¦ [ ?

G

¦ψ

1

, . . . , ψ

n

¦ [

[ψ

1

]ψ

2

74

4.3. Public announcement

We obtain public announcement logic with common knowledge and questions

PACQ.

4.3.1 Updates and questions

Let us return to our example. As we said, members of a group learn what

was announced. In particular, if Ann says

“I’ve got the Joker.”,

the announced fact becomes commonly known in the group of players ¦Ann,

Bill, Catherine¦. This seems to suggest that a publicly announced proposi-

tion becomes common knowledge . But what if Ann says:

“You don’t know it yet, but I’ve got the Joker.”?

This announcement can be formalized by

(J

a

∧ [b](J

a

) ∧ [c](J

a

)),

where J

a

means Ann has the Joker. Although the formula is true in the

moment of announcement, it is evident that its epistemic part (you don’t

know it yet) becomes invalid after it is announced. So the formula (J

a

∧

[b](J

a

) ∧ [c](J

a

)) becomes false after the announcement.

A formula, which becomes false after it is truthfully announced (as in our

example), is called an unsuccessful update; if it becomes true, we call it a

successful update.

Deﬁnition 31.

• Formula ϕ is a successful update in (M, s) iﬀ (M, s) [= 'ϕ`ϕ.

• Formula ϕ is an unsuccessful update in (M, s) iﬀ (M, s) [= 'ϕ`ϕ.

If a formula is an unsuccessful update, it cannot be commonly known

in the updated model. Using the soundness proof of the rule (4.2) we can

prove that a formula is true after a public announcement if and only if it gets

common knowledge after the announcement (see [42, p. 83 and 86]).

Lemma 14. [ϕ]ψ is valid iﬀ [ϕ]C

G

ψ is valid.

As a consequence we get

Lemma 15. [ϕ]ϕ is valid iﬀ [ϕ]C

G

ϕ is valid.

If a formula [ϕ]ϕ is valid, we call it a successful formula.

75

4.3. Public announcement

Deﬁnition 32. Formula ϕ is a successful formula iﬀ [ϕ]ϕ is valid, otherwise

it is an unsuccessful formula

From Lemma 15 we know that publicly announced successful formulas

are commonly known. Atoms, K

i

ϕ, and K

i

ϕ (for every ϕ) are examples of

successful formulas.

Successful formulas true in a state are successful updates there:

Lemma 16. If [ϕ]ϕ is valid formula and (M, s) [= ϕ, then (M, s) [= 'ϕ`ϕ.

It is easy to verify that in our S5 background system questions are suc-

cessful formulas, i.e.,

Fact 37. [Q

i

]Q

i

is valid.

In S5-models a question Q

i

, which is askable in a state s, is askable in

all states from the equivalence class sR

i

. No ‘cutting’ of states in the model

M forced by the public announcement of Q

i

results in (M, s) [= Q

i

and

(M[

Q

i , s) [= Q

i

. Thus, a publicly announced question is commonly known

(see Lemma 15). In other words there is no model and state such that

(M, s) [= [Q

i

]Q

i

.

Successful formulas have an important property: they do not bring any-

thing new if they are announced repeatedly.

Lemma 17. Let ϕ be a successful formula. [ϕ][ϕ]ψ ↔ [ϕ]ψ is valid.

Proof. [ϕ][ϕ]ψ is equivalent to [ϕ ∧ [ϕ]ϕ]ψ (Lemma 13), which is equivalent

to [ϕ]ψ, because of the validity of [ϕ]ϕ (ϕ is successful).

It is no surprise that askable questions (in a state) are successful updates;

it follows from Lemma 16 and Fact 37.

Fact 38. (M, s) [= Q

i

iﬀ (M, s) [= 'Q

i

`Q

i

.

Whenever an agent publicly asks a question, it does not cause any change

in her epistemic model, it remains askable until she gets some new informa-

tion.

4.3.2 Public announcement and answerhood

Whenever a question is (partially) answerable in a state, then there is a

formula ϕ such that after a public announcement of ϕ the question becomes

inaskable there. In our example, Ann has the Joker, but neither Bill nor

Catherine know it. If Catherine publicly asks

76

4.3. Public announcement

“Who has got the Joker?”,

Bill can infer:

“I have not the Joker and Catherine does not know who has it,

therefore Ann has it.”

Catherine’s question was informative for Bill, it caused that the question

Who has got the Joker?, which was askable for Bill, became inaskable after

Catherine had asked it, even if her question was not (partially) answered.

This leads us to the deﬁnition of informative formula.

Deﬁnition 33. A formula ϕ is informative for an agent i with respect to a

question Q in (M, s) iﬀ (M, s) [= Q

i

∧ 'ϕ`Q

i

.

Contrary to partial answerhood (see Theorem 16) there need not be any

logical connection between an informative formula and direct answers to a

question. The informativeness can be forced by the shape of a particular

model.

However, it is a clear conclusion of Deﬁnition 33 that whenever there is

an askable question in a state for an agent, then after an announcement of an

informative formula the agent obtains at least partial answer to the question.

Lemma 18. If a formula ϕ is informative in (M, s) for an agent i with

respect to a question Q, then there is α ∈ dQ such that (M[

ϕ

, s) [= [i]α or

(M[

ϕ

, s) [= [i]α.

Proof. From the informativeness of ϕ we obtain (M[

ϕ

, s) [= Q

i

, which

means (M[

ϕ

, s) [= A

i

Q or (M[

ϕ

, s) [= P

i

Q, because of the safeness of Q

in (M, s).

If an informative formula is ‘strong’ enough (i.e., it implies (partial) an-

swer to a question)

3

, then the (partial) answer is commonly known among

agents in the updated model.

Fact 39. If a formula ϕ is informative in (M, s) for an agent i ∈ G with

respect to a question Q and there is α ∈ dQ such that (ϕ → α) or (ϕ → α)

is valid, then (M[

ϕ

, s) [= C

G

α or (M[

ϕ

, s) [= C

G

α.

The core of this fact is that the formula ϕ is true in each state in an

updated model M[

ϕ

, so is α or α. The role of informativeness is minor, it

‘only’ informs us that a question Q was askable for an agent i in s and that

ϕ can be truthfully announced there.

3

We could say that an informative formula ‘gives’ a (partial) answer to a question.

77

4.3. Public announcement

We have to be careful; informativeness of a formula for an agent does

not imply its informativeness for any other one. The next example shows a

structure where ϕ is informative for the agent a with respect to the question

?α, but it is not informative for b with respect to the same question.

Example 21. a b

s

1

←→ s

2

←→ s

3

α α α

ϕ ϕ ϕ

4.3.3 Answer mining in a group

In Example 20 we displayed two agents a and b. Neither of them can answer

the question ?(α → β)—it is their group question. However, the question is

implicitly answered. We can imagine their cooperative communication. The

question ?(α → β) is askable for the agent a and she wants to answer it.

Her colleague b may help. So, a can directly and publicly ask the question

?(α → β) and reveal her ignorance. A question is a successful formula, thus,

it is commonly known. In cooperative communication agents are supposed

to announce what they know with respect to their group question, i.e., [a]α

and [b]β. They come to the (complete) answer (α → β), which is, more-

over, common knowledge. We obtained a sequence of (truthfully) publicly

announced agents’ knowledge leading to a (commonly known) answer:

'[a]α`'[b]β`(α → β),

resp.,

[[a]α][[b]β]C

{a,b}

(α → β).

So our group was successful in seeking an answer using just ‘internal’ re-

sources. They reached an answer after a series of announcements of facts

they know.

This inspires the following idea of question’s ‘solvability’. In general, we

can say that a question is (partially) solvable, if there is a ﬁnite series of

truthful announcements of agents’ knowledge after which a (partial) answer

is a result.

Deﬁnition 34. A question Q is

• solvable for a group G in a state (M, s) iﬀ there is a set of formu-

las ¦ψ

1

, . . . , ψ

k

¦ and a direct answer α ∈ dQ such that (M, s) [=

'[l

1

]ψ

1

` . . . '[l

k

]ψ

k

`α, where l

j

∈ G,

78

4.3. Public announcement

• partially solvable for a group G in a state (M, s) iﬀ there is a set of

formulas ¦ψ

1

, . . . , ψ

k

¦ and a direct answer α ∈ dQ such that (M, s) [=

'[l

1

]ψ

1

` . . . '[l

k

]ψ

k

`α, where l

j

∈ G.

Solvability follows the idea of the rule (4.1) describing the pooling agents’

scattered knowledge together. This rule inspired an alternative deﬁnition of

distributed knowledge as well, see [11]:

• (M, s) [= D

∗

G

ϕ iﬀ ¦ψ [ (∃i ∈ G)((M, s) [= [i]ψ)¦ [= ϕ

D

∗

G

is not equivalent to D

G

and the following is proved in [11]:

Theorem 19. If (M, s) [= D

∗

G

ϕ, then (M, s) [= D

G

ϕ, but not vice versa.

However, solvability is based on ﬁnite sets. Our system with common

knowledge is not compact and this brings us to a ﬁnite version of D

∗

G

:

• (M, s) [= D

+

G

ϕ iﬀ there are ψ

1

, . . . , ψ

k

such that (M, s) [= [l

1

]ψ

1

∧. . . ∧

[l

k

]ψ

k

and formula (ψ

1

∧ . . . ∧ ψ

k

) → ϕ is valid.

Whenever a (partial) answer of a question is distributed knowledge in the

new sense of D

+

G

, then the question is solvable.

Lemma 19. If (M, s) [= D

+

G

ϕ, then (M, s) [= '[l

1

]ψ

1

` . . . '[l

k

]ψ

k

`ϕ.

Proof. From (M, s) [= D

+

G

ϕ we obtain that each [l

j

]ψ

j

is successful formula

true in (M, s). Lemma 16 says that they are successful updates in (M, s)

and, simultaneously, they are commonly known. Let us write M[

...

for the

updated model after the series of public announcements '[l

1

]ψ

1

` . . . '[l

k

]ψ

k

`.

It follows that (M[

...

, s) [= ψ

j

and due to the validity of (ψ

1

∧ . . . ∧ ψ

k

) → ϕ

we have (M[

...

, s) [= ϕ.

A formula, which is distributed knowledge in a state, is a successful up-

date there.

Theorem 20. If (M, s) [= D

+

G

ϕ, then (M, s) [= 'ϕ`ϕ.

Proof. Let us suppose that (M, s) [= D

+

G

ϕ, but (M, s) [= 'ϕ`ϕ. Then either

(M, s) [= ϕ or (M[

ϕ

, s) [= ϕ.

1. (M, s) [= ϕ is not possible: s ∈

s

¸

i∈G

R

i

in S5 and (M, s) [= D

+

G

ϕ

implies (M, s) [= D

G

ϕ (Theorem 19).

2. If (M[

ϕ

, s) [= ϕ, then M[

ϕ

must be diﬀerent from the model M[

...

we

have talked about in the proof of Lemma 19. Thus, there must be a

state s

0

which makes the diﬀerence.

79

4.4. Final remarks

(a) It is impossible that s

0

∈ S

M|

...

and s

0

∈ S

M|

ϕ

: each ψ

j

is true in

(M, s

0

) and (ψ

1

∧ . . . ∧ ψ

k

) → ϕ is valid, then (M, s

0

) [= ϕ.

(b) Let s

0

∈ S

M|

...

and s

0

∈ S

M|

ϕ

. s

0

∈ S

M|

ϕ

means that (M, s

0

) [=

ϕ and from Theorem 19 we obtain s

0

∈

s

¸

i∈G

R

i

. However,

each ψ

j

is known in s by an agent from G, i.e., (M, s) [= D

G

ψ

j

(Fact 35), thus, (M, s

0

) [= ψ

j

and it is not possible that s

0

∈ S

M|

...

.

The theorem says an important thing: if an answer to a question is acces-

sible by a communication of agents in a group, then it is successful update.

A successful update becomes common knowledge in the updated model:

Fact 40. If (M, s) [= 'ϕ`ϕ, then (M[

ϕ

, s) [= C

G

ϕ.

As a result we obtain that a question, whose answer is accessible by a

communication of agents, is answered for a group of agents in the updated

model, cf. Deﬁnition 28.

4.4 Final remarks

This chapter combines questions in epistemic framework and communication

based on public announcement logic. The background system is S5, which

can be understood as ‘introspective’. Next to positive and negative intro-

spection we can recognize that an agent ‘knows’ questions askable for her:

formula

Q

i

↔ [i]Q

i

is valid. If (M, s) [= Q

i

, then the askability of Q

i

for an agent i holds in each

state in the afterset sR

i

, i.e., in the equivalence class of s.

In a multi-agent epistemic approach we can understand a question as a

‘task’ (or a ‘problem’) to be solved by a particular group of agents. Com-

munication is one of the basic tools of a group searching for a solution to a

problem (e.g., an answer to a question) and asking questions is one of the

essential parts of this communication. Our liberal SAM makes it possible to

mix knowledge and questions. The question

Who has got the Joker?

is mostly seen as a question about facts. However, we can receive the answer

I don’t know.

80

4.4. Final remarks

In our setting this answer is not a complete answer to the question Who has

got the Joker?. It solves another kind of questions. Let us return to the

example of three card players. Bill can ask Catherine

Do you know who has got the Joker, Catherine?

This question is primarily asking for Catherine’s knowledge about the card

holder. Expected direct answers are

I know who has got the Joker.

I don’t know who has got the Joker.

The ﬁrst answer indicates that Catherine can completely answer the question

Who has got the Joker?. The second one indicates that Catherine cannot

completely answer it, but such answer does not reject the possibility that

Catherine knows a partial answer. Whenever Bill wants to ﬁnd out whether

Who has got the Joker? is a task for Catherine, he should ask

Would you ask the question ‘Who has got the Joker?’, Catherine?

Bill asks Catherine whether the question Who has got the Joker? is a rea-

sonable (askable) question for her. It is a yes-no question formalized by the

formula

?

b

¦?

c

¦J

b

, J

a

¦, ?

c

¦J

b

, J

a

¦¦

The ﬁrst direct answer means that Catherine would ask Who has got the

Joker?, i.e., the question ?

c

¦J

b

, J

a

¦ is askable for her. The second one means,

this question is not askable for her, which according to Theorem 16 means

that Catherine can (at least partially) answer that question.

Multi-agent approach with group knowledge modalities makes it possible

to speak on levels of answerhood. An agent’s personal level of answerhood

conditions is based on agent’s knowledge. In case of questions posed in a

group of agents we consider commonly known (partial) answer as a right

solution of a question with respect to a group. If an answer to a question is

sought by a communication inside a group, then an answer must be known

by some member, or it must at least be included as a distributed knowledge.

81

Chapter 5

Conclusion

Although we can consider the thesis to consist of two almost independent

parts, some points are common to both of them. Working in inferential

erotetic logic as well as in epistemic erotetic logic I primarily wanted to

provide tools for the development of both branches of erotetic logic. Even if

I do not want to contribute to a philosophical debate on what a question is, it

seems to me that a word or two should be said about the chosen methodology

in both parts of the thesis.

The main inspiration came from the original IEL. It inspired the set-of-an-

swers methodology introduced in the ﬁrst chapter as well as the emphasis

posed on inferences with questions. SAM is liberal enough to be used in the

presented approaches. It is open for additional restrictions given by both

syntax and semantics. Of course, many objections can be raised against

it, especially, whenever we want to analyze all kinds of natural language

questions. Our intention was to work with propositional logic and to keep

maximum from the logic of declaratives. In the ﬁrst chapter we showed that

our SAM is convenient for the execution of erotetic inferences and that the

epistemic variant is very natural. Coming back to additional restrictions,

chapter 2 presents SAM containing direct answers as declaratives only. It is

in correspondence with the second Hamblin’s postulate (cf. subsection 1.2.1).

On the contrary, our epistemic erotetic logic (chapters 3 and 4) admits to

have not only declaratives among direct answers.

The ﬁrst part of the thesis (chapter 2) is fully developed in the IEL frame-

work. Questions and declaratives are mixed only on the level of consequence

relations and the main goal of the chapter is to study relationships among

IEL consequences. Our general approach showed that some relations must be

supported by additional relations among direct answers, cf. results obtained

for regular e-implication, or based on the relationship of ‘strongness’ between

two questions.

82

CHAPTER 5. CONCLUSION

The second part of the thesis is diﬀerent in its substance. Epistemic

framework is the prime and questions correspond to certain states of knowl-

edge, ignorance, and presuppositions of an agent. This approach is a novelty

inspired by Groenendijk-Stokhof’s intensional erotetic logic together with

our SAM. Questions become a part of epistemic language and they can be

considered as satisﬁed in an epistemic state (in a model). The semantic

work with questions has almost the same ﬂavor as it is with (epistemic)

declaratives. Only the questions’ satisﬁability (askability) in a state is more

complex, being based on three conditions (non-triviality, admissibility, and

context). Although we introduced a general approach, in the rest of the the-

sis we use ﬁnite SAM. This makes the work with the context condition and

presuppositions much easier. Also the deﬁnition of askability of a question in

a state is of such a generality that it can be used for any epistemic-like logic.

In the thesis we presented epistemic logics K and S5 extended by questions.

Questions are representable by modal formulas there. Thus, these systems

can be considered as ‘reductionist’ ones—of course, it was not intentional.

Being inspired by IEL and Groenendijk-Stokhof’s approach we wanted to

deal with inferences with questions. In this epistemic case, inferences are

based on classical implication. However, we can explicitly work with epis-

temic contexts and obtain similar structures (on the object-language level)

that are introduced in IEL. The conditions required for askability of a ques-

tion nicely correspond to natural answerhood conditions. Discovering them

we were faced with the problem of some restrictions required for the chosen

SAM.

Moreover, all works well with group modalities as well as with public an-

nouncement. If we compare the contents of the chapters, we may ﬁnd another

division of the thesis. Chapters 2 and 3 can be called ‘logic of questions’,

they study inferences with questions and the relationships of questions and

declaratives. On the other hand, chapter 4 introduces questions as a part

of communication. Dynamic approaches can mostly bear the name ‘logic of

inquiry’. Chapter 4 shows that questions behave well together with updates

and that they play the expected role in the context of distributed knowl-

edge. The framework public announcement logic is based on S5. Publicly

asked questions are successful formulas and an askable question in a state is

a successful update there. As a ﬁnal result we presented the correspondence

of a ﬁnite version of distributed knowledge with cooperative communication

aimed at ﬁnding a commonly known (partial) answer to a group question.

83

5.1. Related works and future directions

5.1 Related works and future directions

Publications related to inferential erotetic logic were mentioned in chapter 2

and subsection 1.3.1 where we list papers having something to do with an

‘inquiry’ aspect of IEL, which was not studied in chapter 2. Moreover, we

added two papers based on Hintikka’s approach.

The historical part of the ﬁrst chapter includes many publications refer-

ring to intensional erotetic logic of Groenendijk and Stokhof. Let us point

out the cited dynamic application from [40]. This paper makes the best

of dynamic logic developed in publications of Johan van Benthem and his

collaborators. The usual epistemic model is enriched by a new equivalence

relation of indistinguishability (‘abstract issue relation’). Roughly speaking,

similarly to updates of models based on public announcements, there are

updates for asking yes-no questions.

Closely related to this approach is inquisitive semantics developed by

Jeroen Groenendijk and his collaborators. The generalized version with its

associated logic can be found in [5]. The original idea behind inquisitive se-

mantics is common to dynamic approaches. A cooperative communication is

a raising and resolving issues. Propositions are seen as proposals how to up-

date the common state. “If a proposition consists of two or more possibilities,

it is inquisitive: it invites the other participants to provide information such

that at least one of the proposed updates may be established.” [5, p. 112]

As the recent publications indicate, the combination of epistemic and

dynamic aspects seems to be a good framework for erotetic logic. The goal of

our epistemic logic of questions was just to prepare such a general framework.

It opens many directions of further work, let us mention the most obvious:

• To use another epistemic logic. Relevant epistemic logic proposed by

Ondrej Majer and myself, cf. [2, 22], can be took into account. Relevant

implication provides a good background for erotetic implication. It is

necessary to develop a multi-agent version of relevant epistemic logic.

• To combine our approach with ‘logics of communication’. This should

be based on the recent boom of dynamization, cf. [41, 40, 42]. One

possibility is to apply generalized common knowledge C

G

(ϕ, ψ), which

is, in fact, equivalent to [ϕ]C

G

ψ. A correspondence with our notion

(partially) solvable question can take advantage of a generalization:

C

G

( ϕ, ψ) iﬀ [ϕ

1

] . . . [ϕ

n

]C

G

ψ.

• To develop a predicate version. This item brings us to an extensive

study of types of answers.

84

5.1. Related works and future directions

• To apply non-monotonic approaches. Inferential erotetic logic invites

to non-monotonic applications (e.g., ordering on dQ, preferred models,

default rules based on questions).

• To fuzzify presented approaches. Recall the paper [4] mentioned in

section 1.3.2 as an inspiration.

85

Bibliography

[1] N. Belnap and T. Steel. The Logic of Questions and Answers. Yale,

1976.

[2] M. B´ılkov´a, O. Majer, M. Peliˇs, and G. Restall. Relevant agents. In

L. Beklemishev, V. Goranko, and V. Shehtman, editors, Advances in

Modal Logic, pages 22–38. College Publications, 2010.

[3] M. Borkent and R. van Rozen. On logic and questions in dialogue

systems. http://www.pictureofthemoon.net/~borkent/olaqid.pdf

(downloaded by June 2007).

[4] L. Bˇehounek. Fuzziﬁcation of Groenendijk-Stokhof propositional

erotetic logic. Logique & Analyse, 185–188:167–188, 2004.

[5] I. Ciardelli and F. Roelofsen. Inquisitive Logic. In D. Grossi, L. Kurzen,

and F. Vel´azquez-Quesada, editors, Logic and Interactive Rationality,

pages 111–146. ILLC, 2010.

[6] E. Coppock. Toward a dynamic logic for questions, 2005.

http://www.stanford.edu/~coppock/dynamic-questions.pdf

(downloaded by April 2007).

[7] K. De Clercq and L. Verhoeven. Sieving out relevant and eﬃcient ques-

tions. Logique & Analyse, 185–188:189–216, 2004.

[8] H. N. Duc. Resource-Bounded Reasoning about Knowledge. PhD thesis,

Faculty of Mathematics and Informatics, University of Leipzig, 2001.

[9] R. Fagin, J. Halpern, Y. Moses, and M. Vardi. Reasoning about Knowl-

edge. MIT Press, 2003.

[10] E. J. Genot. The game of inquiry: the interrogative approach to inquiry

and belief revision theory. Synthese, 171:271–289, 2009.

86

Bibliography

[11] J. Gerbrandy. Distributed knowledge. In J. Hulstijn and A. Nijholt,

editors, Twendial’98: Formal Semantics and Pragmatics of Dialogue,

pages 111–124. Universiteit Twente, Enschede, 1998.

[12] J. Groenendijk. Questions and answers: Semantics and logic.

In R. Bernardi and M. Moortgat, editors, Proceedings of the 2nd

CologNET-ElsET Symposium. Questions and Answers: Theoretical and

Applied Perspectives, pages 16–23. Universiteit Utrecht, 2003.

[13] J. Groenendijk and M. Stokhof. Partitioning logical space. Annotated

handout, ILLC, Department of Philosophy, Universiteit van Amster-

dam, 1990. Second European Summerschool on Logic, Language and

Information; Leuven, August 1990; version 1.02.

[14] J. Groenendijk and M. Stokhof. Questions. In J. van Benthem and A. ter

Meulen, editors, Handbook of Logic and Language, pages 1055–1125.

Elsevier, Amsterdam, 1997.

[15] D. Harrah. On the history of erotetic logic. In A. Wi´sniewski and J. Zyg-

munt, editors, Erotetic Logic, Deontic Logic, and Other Logical Matters,

volume 17, pages 19–27. Wydawnictwo Uniwersytetu Wroclawskiego,

1997.

[16] D. Harrah. The logic of questions. In D. Gabbay and F. Guenthner,

editors, Handbook of Philosophical Logic, volume 8, pages 1–60. Kluwer,

2002.

[17] J. Hintikka. The Semantics of Questions and the Questions of Seman-

tics. North-Holland Publishing Company, 1976.

[18] J. Hintikka, I. Halonen, and A. Mutanen. Interrogative logic as a gen-

eral theory of reasoning. In D. Gabbay, R. Johnson, H. Ohlbach, and

J. Woods, editors, Handbook of the Logic of Argument and Inference,

pages 295–337. Elsevier, 2002.

[19] K. H. Knuth. Toward question-asking machines: The logic of ques-

tions and the inquiry calculus, 2005. (Tenth International Workshop

on Artiﬁcial Intelligence and Statistics (AISTATS 2005), Barbados)

http://www.gatsby.ucl.ac.uk/aistats/fullpapers/246.pdf.

[20] P. Le´sniewski and A. Wi´sniewski. Reducibility of questions to sets of

questions: Some feasibility results. Logique & Analyse, 173–175:93–111,

2001.

87

Bibliography

[21] D. Leszczy´ nska. Socratic proofs for some normal modal propositional

logics. Logique & Analyse, 185–188:259–285, 2004.

[22] O. Majer and M. Peliˇs. Epistemic logic with relevant agents. In M. Peliˇs,

editor, The Logica Yearbook 2008, pages 123–135. College Publications,

2009.

[23] J. J. C. Meyer and W. van der Hoek. Epistemic Logic for AI and Com-

puter Science. Cambridge University Press, 1995.

[24] R. Nelken and N. Francez. A bilattice-based algebraic se-

mantics of questions. In Proceedings of MOL6, Sixth Meet-

ing on Mathematics of Language, Orlando, Florida, 1999.

http://www.eecs.harvard.edu/~nelken/.

[25] R. Nelken and N. Francez. A calculus for interrogatives based on their

algebraic semantics. In Proceedings of TWLT16/AMILP2000: Algebraic

methods in language processing, pages 143–160, Iowa, 2000.

[26] R. Nelken and N. Francez. Querying temporal database using

controlled natural language. In Proceedings of COLING2000, The

18th International Conference on Computational Linguistics, 2000.

http://www.eecs.harvard.edu/~nelken/.

[27] R. Nelken and N. Francez. Bilattices and the semantics of natural lan-

guage questions, 2002.

[28] R. Nelken and C. Shan. A logic of interrogation should be internalized in

a modal logic for knowledge. In K. Watanabe and R. B. Young, editors,

SALT XIV: Semantics and Linguistic Theory. Cornell University Press,

2004.

[29] R. Nelken and C. Shan. A modal interpretation of the logic of interro-

gation. Journal of Logic, Language, and Information, 15:251–271, 2006.

[30] M. Peliˇs. Questions and logical analysis of natural language: the case

of transparent intensional logic. Logique & Analyse, 185–188:217–226,

2004.

[31] M. Peliˇs. Consequence relations in inferential erotetic logic. In

M. B´ılkov´a, editor, Consequence, Inference, Structure, pages 53–88. Fac-

ulty of Arts, Charles University, 2008.

88

Bibliography

[32] M. Peliˇs and O. Majer. Logic of questions from the viewpoint of dynamic

epistemic logic. In M. Peliˇs, editor, The Logica Yearbook 2009, pages

157–172. College Publications, 2010.

[33] M. Peliˇs and O. Majer. Logic of questions and public announcements. In

Eighth International Tbilisi Symposium on Language, Logic and Com-

putation 2009. LNCS, Springer, 2011. To appear.

[34] C. Shan and B. ten Cate. The partitioning semantics

of questions, syntactically. In M. Nissim, editor, Pro-

ceedings of the Seventh ESSLLI Student Session, 2002.

http://users.soe.ucsc.edu/~btencate/papers/esslli02ss.pdf.

[35] B. ten Cate and C. Shan. Question answering: From partitions to prolog.

In U. Egly and C. G. Fermller, editors, Proceedings of TABLEAUX 2002:

Automated reasoning with analytic tableaux and related methods, pages

251–265. Springer-Verlag, 2002.

[36] B. ten Cate and C. Shan. Axiomatizing Groenendijk’s logic of interro-

gation. In M. Aloni, A. Butler, and P. Dekker, editors, Questions in

Dynamic Semantics, pages 63–82. Elsevier, 2007.

[37] P. Tich´ y. Questions, answers, and logic. American Philosophical Quar-

terly, 15:275–284, 1978.

[38] M. Urba´ nski. Some remarks concerning modal propositional logic of

questions. Logic and Logical Philosophy, 6:187–196, 1998.

[39] M. Urba´ nski. Synthetic tableaux and erotetic search scenarios: Exten-

sion and extraction. Logique & Analyse, 173–175:69–91, 2001.

[40] J. van Benthem and S¸. Minic˘a. Toward a Dynamic Logic of Questions.

In X. He, J. Horty, and E. Pacuit, editors, Logic, Rationality, and In-

teraction, pages 27–41. Springer, 2009.

[41] J. van Benthem, J. van Eijck, and B. Kooi. Logics of communication

and change. Information and Computation, 204:1620–1662, 2006.

[42] H. van Ditmarsch, W. van der Hoek, and B. Kooi. Dynamic Epistemic

Logic. Springer, 2008.

[43] A. Wi´sniewski. On the reducibility of questions. Erkenntnis, 40:265–284,

1994.

89

Bibliography

[44] A. Wi´sniewski. The Posing of Questions: Logical Foundations of

Erotetic Inferences. Kluwer, 1995.

[45] A. Wi´sniewski. Erotetic logic and explanation by abnormic hypotheses.

Synthese, 120:295–309, 1999.

[46] A. Wi´sniewski. Questions and inferences. Logique & Analyse,

173–175:5–43, 2001.

[47] A. Wi´sniewski. Erotetic search scenarios. Synthese, 134:389–427, 2003.

[48] A. Wi´sniewski. Erotetic search scenarios, problem solving, and deduc-

tion. Logique & Analyse, 185–188:139–166, 2004.

[49] A. Wi´sniewski. Socratic proofs. Journal of Philosophical Logic,

33:299–326, 2004.

[50] A. Wi´sniewski. Reducibility of safe questions to sets of atomic yes-no

questions. In J. Jadecki and J. Pa´sniczek, editors, Lvov-Warsaw School:

The New Generation, pages 215–236. Rodopi, Amsterdam/NY, 2006.

90

Prohlaˇuji, ˇe jsem disertaˇn´ pr´ci napsal samostatnˇ s vyuˇit´ pouze uves z c ı a e z ım den´ch a ˇ´dnˇ citovan´ch pramen˚ a literatury a ˇe pr´ce nebyla vyuˇita v y ra e y u z a z r´mci jin´ho vysokoˇkolsk´ho studia ˇi k z´ an´ jin´ho nebo stejn´ho titulu. a e s e c ısk´ ı e e V Praze, 11. ledna 2011

Michal Peliˇ s

Abstract The thesis deals with logic of questions (erotetic logic), which is one of the branches of non-classical logic. In the introductory part we speak generally about formalization of questions and the newest approaches to questions in logic are summed up. We introduce a formalization based on sets of direct answers and point out the role of inferences with questions. The rest of the thesis consists of two parts that can be read independently. The ﬁrst part focuses on relationships among consequence relations in inferential erotetic logic (IEL). We keep the framework of original IEL, introduced by Andrzej Wi´niewski, together with the representation of questions s by sets of direct answers. Answers are strictly formulas of the declarative language. The mix of interrogatives and declaratives occours just on the level of consequences. Consequence relations with questions are deﬁned by means of multiple-conclusion entailment among sets of declarative formulas. This way, one can work with classes of models and to make transparent some properties and relationships. We provide a general study of erotetic inferences based on IEL that is open for non-classical applications. The second part contains epistemic erotetic logic. A question is understood as a set of direct answers; however, this time the set is ﬁnite. The satisﬁability of a question in a state (possible world) of an epistemic model is deﬁned by three conditions (a questioner does not know any direct answer, each direct answer considers as possible and at least one of them must be the right one). This approach is a new one and it is suitable for generalization to every epistemic-like system by questions and common erotetic concepts (e.g., various types of answers and erotetic inferences). The goal of this study is a future application in communication theory of a group of agents. We ﬁnish this part by multi-agent public announcement logic with application of questions by answer mining in a group of agents.

Abstrakt Pr´ce se zab´v´ jedn´ z odvˇtv´ neklasick´ch logik – logikou ot´zek (eroteta y a ım e ı y a ickou logikou). r a z a explicitn´ stanoven´ mnoˇiny pˇ´ ych odpovˇd´ Zbytek pr´ce je rozdˇlen ım ı z rım´ e ı. kter´ y umoˇnuje rozˇ´rit libovoln´ epistemick´ syst´m o ot´zky a s nimi spojenou zˇ sıˇ y y e a erotetickou terminologii (r˚zn´ typy odpovˇd´ inference s ot´zkami a dalˇ´ u e e ı. Protoˇe jsou d˚sledkov´ relace s ot´zkami ıho z u e a deﬁnov´ny pomoc´ klasick´ho v´ avˇrov´ho s´mantick´ho d˚sledku. V t´to ˇ´sti n´m z y rıd´ u e ca a jde o obecn´ pˇ´ y rıstup ovˇem s omezen´ kdy je deklarativn´ jazyk rozˇ´ren s ım. y y u y Druh´ ˇ´st obsahuje epistemickou erotetickou logiku. V z´vˇru druh´ e u r a ı e ı a a e e ˇ´sti tak pˇedstavujeme multiagentn´ logiku veˇejn´ho vyhl´ˇen´ (public anca r ı r e as ı nouncement logic) s ot´zkami a skupinov´mi znalostmi. rım´ e jsou formule deklarativn´ jazyka. a sı). Pouˇ´ ame v´hradnˇ e e s zıv´ y e formalizaci ot´zek pomoc´ mnoˇiny pˇ´ ych odpovˇd´ kdy pˇ´ e odpovˇdi a ı z rım´ e ı. V uvodn´ ˇ´sti se hovoˇ´ obecnˇ o formalizovan´m pˇ´ ´ ı ca rı e e rıstupu k ot´zk´m v logice a souˇasnˇ je struˇnˇ shrnuta zejm´na nejnovˇjˇ´ historie a a c e c e e e sı tohoto odvˇtv´ Zde je t´ˇ zd˚vodˇov´na smysluplnost zachycen´ usudk˚. a y . vyuˇ´ a ı e ıcez´ e e e e u zıv´me pˇi d˚kazech pˇ´ a r u rıstup zaloˇen´ na tˇ´ ach model˚. Ot´zka je nad´le y a r y a a formalizov´na jako mnoˇina pˇ´ ych odpovˇd´ tentokr´t vˇak uvaˇujeme a z rım´ e ı. e ı. ale k propojen´ deklarativn´ a interogativn´ jazyka dojde aˇ a ı ıho ıho z na urovni d˚sledkov´ch relac´ Prim´rn´ z´jmem je studovat vztahy mezi ´ u y ı. Plnˇ zde vyuˇ´ ame r´mec p˚vodn´ inferenˇn´ e zıv´ a u ı c ı erotetick´ logiky zaveden´ Andrzejem Wi´niewskim. naˇ´ hlavn´ c´ y e sım ım ılem je komunikace ve skupinˇ agent˚ pˇi hled´n´ odpovˇd´ na ot´zky. ez u n a ı ´ u v nichˇ se ot´zky objevuj´ a je pˇedstavena formalizace ot´zky zaloˇen´ na z a ı. e ca e cıst a Prvn´ ˇ´st se zab´v´ d˚sledkov´mi relacemi v inferenˇn´ erotetick´ logice ı ca y a u y c ı e (inferential erotetic logic). kter´ lze ˇ´ nez´visle. Prvotn´ je episa ca ı temick´ r´mec tvoˇen´ zvolenou epistemickou logikou. z rım´ e ı s z z n jedna z pˇ´ ych odpovˇd´ mus´ b´t spr´vn´). Jde o nov´ pˇ´ rım´ e ı ı y a a y rıstup. Pˇestoˇe i zde se snaˇ´ r z zıme o obecn´ syst´m. a s z koneˇnou variantu a jej´ splnˇnost v moˇn´m svˇtˇ epistemick´ho modelu c ı e z e ee e je v´z´na na platnost tˇ´ epistemick´ch podm´ a a rı y ınek (tazatel nezn´ ˇ´dnou z a za odpovˇd´ kaˇdou z pˇ´ ych odpovˇd´ vˇak povaˇuje za moˇnou a alespoˇ e ı. ı sıˇ o ot´zky. a ım a jednotliv´mi erotetick´mi d˚sledkov´mi relacemi. a e na dvˇ ˇ´sti.

Preface The work on this thesis started in 2003 when I met logic of questions—the branch of logic. to suggest some generalizations and new relationships.) During the study of IEL. see our results in [22] and in [2]. Such aspirations were successful. it is only one step away to its dynamic application. In many approaches to the formalization of interrogative sentences. IEL provides consequence relations with declaratives as well as interrogatives. The ﬁrst one is chapter 2 and the second one consists of chapters 3 and 4. I worked with my colleagues on epistemic interpretation of the relevant logic. At the very beginning. I wished to formulate a completely general framework that can be used in all ‘epistemic-like’ systems. (See chapter 2. to go through the relationships of inferences with questions and. Chapter 1 i . and the ﬁndings are presented in chapter 3. My interest in non-standard consequence relations brought me to the study of inferential erotetic logic ﬁrst. if there is an epistemic system with questions. Then I was lucky to ﬁnd two erotetic logics that proved their vitality in the recent years. The ﬁrst one was inferential erotetic logic (IEL) developed by Andrzej Wi´niewski and his collaborators and the second one was an intensional aps proach to questions of Jeroen Groenendijk and Martin Stokhof. I balanced between the reasons for and against having a special logic of questions. This and the intensional approach of Groenendijk and Stokhof were inspiration for my own extension of epistemic logic by questions. I decided to learn this system. In the thesis I decided to use public announcement logic together with a group epistemic modalities and chapter 4 contains results obtained in this ﬁeld. epistemic terminology is used and questions are often seen as requirements of knowledge completion. Structure of the thesis The thesis includes two main parts that can be read independently. inspired by the methodology of a question representation used in IEL. which seemed to me very promising in possible applications and open for an extensive development. Naturally.

The deﬁned consequence relations with questions are naturally based on the multiple-conclusion entailment among sets of declarative formulas. We keep the framework of IEL. the basic knowledge of modal logic is required. The last chapter 5 contains some ﬁnal remarks to the used set-of-answers methodology. to the main results in the second part of the thesis. Groenendijk-Stokhof’s intensional approach. the core of the chapter is devoted to a formalization of questions based on sets of answers. IEL requires that declarative and interrogative formulas are not mixed on the object-language level. The mentioned independence of chapter 2 and chapters 3 and 4 is also implied by the fact that these parts are based on separate papers—chapter 2 was published in [31] and a simpliﬁed version of chapters 3 and 4 appeared in [32] for the ﬁrst time and [33] contains the last results from chapter 4.provides common methodology for both parts. We add ii . and some developments of these theories. Also. We introduce brieﬂy inferential erotetic logic. Chapter 2 Chapter 1 Chapter 3 → Chapter 4 Chapter 2 can be read as a full introduction to the topic. and to the related approaches. Chapter 1: Logic and questions The chapter brieﬂy introduces the multi-paradigmatic situation in the methodology of erotetic logic and contains a short historical overview of this branch of logic with a special emphasis on the recent development. However. but the question representation uses the methodology from chapter 1. We justify the usefulness of the methodology in the study of consequence relations with questions as well as in an epistemic interpretation of questions. and also discusses further directions. answers are strictly declarative sentences. Chapter 2: Consequence relations in inferential erotetic logic This part is aimed to study relationships among consequence relations that were introduced by inferential erotetic logic (IEL). In chapters 3 and 4 it is not the case. Chapter 2 can be understood as an inspiration of some erotetic concepts used in the rest of the thesis.

This ‘model-based approach’ makes proofs and properties transparent. This chapter. In this framework. a question becomes a complex modal formula. can provide a general framework and inspiration for the work with inferences among questions and declaratives. Questions are considered to be ﬁnite sets of direct answers and their satisﬁability in a state of an epistemic model is based on three conditions. Inspired by inferential structures in IEL. a technical overview of some IEL concepts and their properties.the semantic range of a question to the terminology of IEL and sets of declaratives are associated with classes of models. Askablity of questions as well as answerhood conditions are studied from the viewpoint of groups of agents. we show that there are ‘philosophically’ similar structures based on classical implication. Finally we show the role of questions and group modalities in answer ‘mining’ among agents. Let me mention only some of them. Chapter 4: A step to dynamization of erotetic logic This chapter takes the full advantage of the multi-agent setting from chapter 3 and can be considered as an application of the introduced erotetic-epistemic approach in a dynamic framework. iii . The rest of the chapter is devoted to answerhood conditions and the role of implication with respect to epistemic context and conjunctions of yes-no questions. We deﬁne here an epistemic logic based on the modal system S5 extended by group modalities together with public announcement modality. Acknowledgements It would be a long list of names to express my thanks to everybody who helped me with the study as well as with the thesis. we work with the normal multi-modal propositional logic K as a background for the introduction of multi-agent epistemic logic. While this deﬁnition of questions’ askability is fully general for any ‘epistemic-like’ system and it is not necessary to keep the ﬁnite set-of-answers methodology. Chapter 3: Epistemic logic with questions The main goal is to incorporate questions in a general epistemic framework. which express ignorance and presuppositions of a questioner.

and Greg a e Restall. a u e Before I started to think about erotetic logic. Vojtˇch Kolman. simultaneously. Among them are Marta B´ ılkov´. ır ıtˇ iv . especially. no. Hana Skˇivanov´ was the r a ﬁrst one to bring me to this topic. Petr Cintula. no. Svatopluk Nevrkla. Jana Bureˇov´. grant ICC/08/E018 of the Grant Agency of the Czech Republic (a part of ESF Eurocores-LogICCC project FP006) • Dynamic Formal Systems. Let me mention. Petr Jirk˚. My special thanks go to Ondrej Majer who has helped me with the work in dynamization of erotetic epistemic logic. I would like to thank all my colleagues at the Faculty of Physical Education and Sport (Charles University in Prague) that tolerated my interest in logic although they would prefer me to be a sociologist. we have been working in epistemic relevant logic. a s a Radek Honz´ Marie Kol´ a. Grant agency of the Czech Republic. 401/03/H047 I would like to thank all who provided me the possibility to participate on the grants: Tim Childers. Kamila Bendov´. and contemplated my work. and.First of all I have to thank my supervisor Jaroslav Peregrin who introduced me to formal semantics and non-classical logics. P401/10/1504 • Logical models of reasoning with vague information. During the work. and Ondrej Majer. Grant agency of the Academy of Sciences of the Czech Republic. no. read and discussed my papers. a a u Petr H´jek. and very helpful comments by Andrzej Wi´niewski. IAA900090703 • Logical foundations for semantics and representation of knowledge. my thanks go to my family as well as colleagues and students at the Department of Logic at the Faculty of Arts (Charles University in Prague) and Institute of Philosophy (Academy of Sciences of the Czech Republic). Libor Bˇhounek. and V´ ezslav Svejdar. ınsk´ n a ˇ dim´ Svoboda. a bit later. I consider myself lucky to have so many colleagues I can cooperate with. s The work on the thesis was supported by the following grants: • Formal and historical approach to epistemology. Vlaık. Martina Pivoˇkov´. Grant agency of the Czech Republic. Tim Childers. Jeroen Groenendijk sent me some of his papers in progress. advice. Ludmila Dost´lov´. We wrote some papers from this ﬁeld and. Martin V´ brought my ıta attention to epistemic logic with group modalities. Chris Ferm¨ller. From the very beginning I was supported by papers. Last but not least.

. . .3.2 Presuppositions . . . . . 2. . . . . . . . . . . .3 Note on the recent history of erotetic logic 1. 1. .4 Basic properties of questions . 1 . . . . . .2 Consequence relations in IEL . 7 .3 Questions and questions . .1 Evocation . . . . . . . . . . . . . . . 2. 1. . . . . . . . . . . . . . 2.2.Contents 1 Logic and questions 1. .3 Model-based approach . . . 1. . . . . . . . . . . . . .2 Set-of-answers methodology . . . . . . 1. . . . . . 1.1 Adapted set-of-answers methodology in IEL . . . . . . . . .1. . . . . . . . . . . . . . . . . . . .3 Comparing questions: relations of questions based on direct answers . . . . . . . . . .2 Intensional erotetic logic . . . .1 Introduction . . . . . . . .1 Inferential erotetic logic . . . . . . . . . . . . . . . . . . . . 45 v . . . . . . 11 . . . . . .2 Sets of answers . . . . . . . . .2 Evocation and erotetic implication . . .2 Questions and declaratives . . . . . . . . . . . 2. . 9 . 2. . . . . . . . . . . . 2. . . . .2. . . . . .4 Final remarks . . . . .1. .1. . . 2. . . . . . . . . . . 2 Consequence relations in inferential erotetic logic 2. . . . . . . 5 . . . . . 2. . .2. . . . . . . . .3. 1. . . . . . .2 Inferences with questions .3. . 2. . 2.1. . . . . . . . . . . . . 1. 2. . .1. and inferences . . . . .3 Epistemic aspects of SAM . . . . . . . . . .2. . . . . . . . . . . . 1 . . . . .3. . 2.1 Questions and answers . . .1 Erotetic implication . . 10 . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . . . . . 1 . . .1. . . .3. . . 1. . . . . . . . . . . . . . . . .1 Semantics of questions . . . . . . . 2. . .1 Questions. . . 38 . . 34 . . . .5 The road we are going to take . . . . .1.2. . . . . . . . . . . . . . . . . 12 14 14 14 15 16 18 19 19 19 22 27 27 32 . . . . . . .4 Questions and sets of questions . . 3 . . . . . . . . . . . . answers. . . . . . 5 . . . . . . . . . 2. .

. . . . . . . . . .3. . . . . . . . . . . . . . . . . . .1 Updates and questions . . . .2 Group questions and answerhood . . . . . . . . . . . . . . . .1 Group epistemic modalities . . . . . . . . . .2 Multi-agent propositional epistemic logic with questions 4. . . . . . . . . . . 4. . . . . .3. . . 3. . . . . . . . . . . .4 Final remarks . . . 4. . . . . . . .5 Context . . 84 Bibliography 86 vi . . . . . . . . . . . . . . . . . . .4 Askability and answerhood .2. . . . . .2 Some important classes of questions . . . 4. . . . . . . . . . . . . . 47 47 49 50 55 58 60 63 65 67 67 68 68 71 73 75 76 78 80 .3 Epistemic logic with questions 3. 4. . . . . . . . . . . .1 Incorporating questions . . . . . . . . . . . . . . . . . . 4. 4 A step to dynamization of erotetic logic 4. . . . . . .2. . . . . .3. . . .6 Implied questions . .1 Introduction . .2 Public announcement and answerhood . . . . . . . . . . . . . . . . 3. . . . . . 3. . . 4. . . . . . . .2. . . . . . . .3 Answer mining in a group . . . . . .1 Related works and future directions . . . . . . . . 3. . . . . . . . . . . . .1 Introduction . . . . . . . . . 3. . . . . . . . . . . . . . . . . . . . 3. . . . . . . . . . . . . . . . . . . . . . 4.2 Single-agent propositional epistemic logic and questions 3. 5 Conclusion 82 5. . .3 Public announcement . . . . . . . .3 Epistemic erotetic implication . . . . . . . . . . . . .2.

declarative sentences usually have their formal (logical) counterparts and play an important role in argumentation. In logic. a discussion on both terms can be found in [16].1 Questions and answers Let us imagine a group of three players: Ann. and inferences In this chapter we wish to show that it is reasonable to consider questions as a part of logical study.1. We recognize it as an interrogative sentence because of its word order and the question mark.1 Questions. the main aim is to concentrate on a methodology used in the rest of the paper—we introduce and discuss a variant of the methodology based on sets of answers. However. The hearing In this paper we use the term logic of questions in the same meaning as erotetic logic. Inferential structures are studied in formal systems. We believe that the dealing with questions in the logical framework will be justiﬁed if we show that questions can play an autonomous and important role in inferences. which can diﬀer in the formalization of declaratives as well as in admitting or rejecting of some principles. and Catherine. Perhaps this point may be considered as the most important to justify logic of questions. 1.1 This introductory chapter provides a brief overview of history as well as methodology in recent approaches to erotetic logic. One of the cards is the Joker and everybody knows this fact. We often see logic to be primarily a study of inferences. Each of them has one card and nobody can see the cards of the others. answers.Chapter 1 Logic and questions 1. 1 1 . Then Who has the Joker? is a reasonable sentence in this situation. Bill.

1.1. Questions, answers, and inferences

or uttering of an interrogative is followed by intonation and interrogative pronounce. An interrogative sentence includes a pragmatic aspect. It is a “request to an addressee to provide the speaker with certain information”—interrogative speech act [14, p. 1057]. Pragmatic approach emphasizes the roles of a speaker and an addressee, which seem to be outside of the interrogative context, but it seems very important in an analysis of questions. This can be the reason why some logicians argue against erotetic logic. If we want to work with interrogatives in a formal system, we have to decide two problems: 1. What is the formal shape of questions? 2. What is the (formal) semantics of questions? Reviewing the history of erotetic logic, there is no unique solution. There are many approaches to the formalization of questions and every approach varies according to what is considered as important. Logic of questions is considered to be multiparadigmatic. This is nicely illustrated by Harrah’s examples of ‘meta-axioms’, see [15, pp. 25–26]. He groups them into three sets according to the acceptance by erotetic logicians. 1. The ﬁrst group includes meta-axioms accepted in almost all systems. Harrah calls them absolute axioms and examples are: (a) Every question has at least one partial answer. (b) (In systems with negation) For every statement P , there exists a question Q whose direct answers include P and the negation of P . (c) Every question Q has a presupposition P such that: P is a statement, and if Q has any true direct answer, then P is true. 2. The second group, standard axioms, is often accepted, but not in all systems. (a) Every question has at least one direct answer. (b) Every direct answer is a statement. (c) Every partial answer is implied by some direct answer. (d) Every question is expressed by at least one interrogative. (e) Each interrogative expresses exactly one question. (f) Given an interrogative I there is an eﬀective method for determining the direct answers to the question expressed by I. 2

1.1. Questions, answers, and inferences

3. The last group is called excentric axioms. Thus, the following examples of such axioms are accepted only in some interrogative systems. (a) If two questions have the same direct answers, then the two questions are identical. (b) Every question Q has a presupposition that is true just in case some direct answer to Q is true. Let us notice the terminology, the diﬀerence between interrogative (sentence) and question was just introduced by standard axioms. The ﬁrst term mostly refers to a type of sentence and the second one is a bit more complex. A question is expressed by an interrogative (sentence) and can be ‘posed’, ‘asked’, etc. Similarly, a proposition is expressed by a declarative (sentence), cf. [16]. Although we are used to use interrogative and question in the same meaning, if necessary, the term interrogative sentence is reserved for a natural-language sentence. What seems to be common to all approaches is viewing questions as something structured and connected with answers. The relationship question — answer(s) is a very conspicuous sign and the meaning of questions is closely connected to answerhood conditions. Since an answer to a question is often represented by a declarative, the starting point of many erotetic theories is a formal system for declaratives. “Any ﬁrst-order language can be supplemented with a question-and-answer system” [44, p. 37]. This broadly accepted statement combines both the formal shape and the meaning of a question. Questions’ autonomy depends on the chosen solution. Wi´niewski distinguishes two basic groups of s erotetic theories: reductionist and non-reductionist theories. Roughly speaking, non-reductionism is characterized by questions that “are not reducible to expressions of other syntactic categories” [44, p. 40]. The boundary between both groups is vague. Perhaps only pure pragmatically oriented approaches belong to the radical reductionism with a complete rejection of questions as a speciﬁc entity. An example of such approach is commented in [30].

1.1.2

Inferences with questions

Although there is a discussion whether it is necessary to work with questions as a new speciﬁc entity, almost all theorists agree that questions play a speciﬁc role in inferences. Let us come back to our group of players. The

3

1.1. Questions, answers, and inferences

situation, where the Joker is held by a member of the group, can raise to the question Q: Who has the Joker? from the declarative Either Ann has the Joker or Bill has the Joker or Catherine has the Joker. What makes this raising reasonable are answerhood conditions of Q connected to the declarative. Another kind of inferential structure is based on declaratives as well as questions among premises. For example, from Q: Who has the Joker? and Γ: The only person from London has the Joker. can be inferred the question Q1 : Who is from London? The relationship of the inferred question Q1 and the question Q is based on their answerhood conditions again. An answer to Q can provide an answer to Q1 with respect to the context Γ. Moreover, in this example, Q can be inferred from Q1 and Γ as well. This shows that the relationship is structured dependently on various kinds of answerhood conditions and contexts. Let us have Q the same, but the context is A person from London has the Joker. If two persons are from London and we gain their names in an answer to Q1 , then we receive only a partial answer to Q.2 If each player (or nobody) is from London and an answer to Q1 does not provide any help for the answering of Q, it has no sense to speak of an inferential relation between Q and Q1 with respect to this context. The role of answerhood conditions in inferences among questions is clearly obvious in the following example: From any (complete) answer to Q we gain a (complete) answer to the question

2 Informally, a partial answer does not completely answer a question, but it eliminates some of the possible answers.

4

we have been still faced with the problem how to formalize the relationship of questions and answers. Though we do not doubt that there are inference-like structures with questions based on answerhood conditions. The question Has Ann the Joker? is ‘entailed’ by Q.1. they can be inferred from the answerhood conditions of Q. For example. Another way is the paraphrasing by epistemic-imperative sentences: Bring it about that I know who has the Joker! 5 . questions are paraphrased by declarative sentences. answerhood conditions of the previous three questions are ‘entailed’ in the answerhood conditions of the question Q.2. 1. We believe that we can solve the problem of the formal shape of questions together with the problem of the questions’ semantics.2. Our aim is to show that such approach can also reﬂect some semantic and pragmatic requirements.1 Semantics of questions Some theories do not admit that questions could have an independent meaning in logic. In this section we introduce a formalization of questions based on a set of answers. Set-of-answers methodology Has Ann the Joker? as well as for the questions Has Bill the Joker? and Has Catherine the Joker? Roughly and informally speaking.2 Set-of-answers methodology In the previous section we emphasized the close connection of questions and answers in most erotetic theories as well as in inferential structures with questions. 1. the question Who has the Joker? may be then paraphrased by I ask you who has the Joker. We see the convenient solution in a liberal set-of-answers methodology.

3 6 . we will not make any survey here. N. Cited from [13. Part I’. Knowing what counts as an answer is equivalent to knowing the question. Independence Interrogatives are entitled to a meaning of their own. 3. An answer to a question is a statement. To accept ‘independence requirement’ means that we are obliged to look for a speciﬁc semantics of questions. p. The possible answers to a question are an exhaustive set of mutually exclusive possibilities. ‘Approaches to the semantics of questions in natural language. Each postulate may be argued against and the detailed discussion is available in [14]. Although we expect to utilize the importance of a questioner and an addressee later on. section 2]. 2.D. In the next subsection we introduce an easy idea of a question representation by a set of direct answers.1. now it may be second-rate from the semantic viewpoint. Belnap. for short) in the literature. 3. Moreover.2. The most important is the ﬁrst requirement. according to David Harrah. adopting the ﬁrst one is “the giant step toward formalization often called set-of-answers methodology” [16. which he used for a classiﬁcation and evaluation of erotetic theories:3 1. 1981. Answerhood The meaning of an interrogative resides in its answerhood conditions. Set-of-answers methodology The propriety of both paraphrases as a complete meaning of a question is rather problematic. we can work with the idea that the semantics of answers forms a good background for the study of the meaning of interrogatives.. Nuel Belnap formulated three methodological constraints on a theory of questions. Pittsburgh. is displayed in Hamblin’s postulates from 1958: 1. which is the main sign of non-reductionist theories. However. where answers are crucial for the meaning of questions. Equivalence Interrogatives and their embedded forms are to be treated on a par. 2. The ‘answerhood’ requires that the meaning of questions is related to the meaning of answers. 3–4]. The approach. Although there are many kinds of set-of-answers methodology (SAM. The ‘equivalence requirement’ is closely related to a semantic entailment and is dependent on the chosen semantics.

.2. If we had decided to represent every question by a complete set of its answers. the ﬁrst three items in the list have the same meaning in the answering of this question. Nobody. The question seems to be answered if a (complete) list of Joker’s owners is given. without any context. . we can understand every question closely connected with a set of (propositional) formulas—answers. . People at this table. From the viewpoint of propositional logic and in accordance with the ﬁrst two Hamblin’s postulates. We can assume that answers are sentences. we can receive the following responses to the same question as well: Ann hasn’t the Joker. The second one appears to bear another kind of information. . etc. Ann and Bill. or I don’t know who has the Joker.g. We will return to this topic shortly in Section 4. Neither of them answers completely the question Who has the Joker?. Considering the context. . . Ann has it. Let us return to our example. we would not have a clear and useful formalization of questions. Ann has the Joker. . . a questioner expects one of the following responses to the question Who has the Joker? : 7 . the question Who has the Joker? can be answered by expressions of the following form: Ann. . an addressee says to a questioner that she has the same problem and would ask the same question.1.4. it removes some answers as impossible. Your friends. all answers with Ann having the Joker.2.2 Sets of answers Generally. Set-of-answers methodology 1. e. Of course.. Batman has the Joker. The ﬁrst one can be considered to be a partial answer. thus.

2. . Complete answers are ‘solutions’ of a question and the set of direct answers is a subset of the set of complete ones. from which γ is inferred thanks to the context. α2 . .4 We want to be very s liberal and this leads us to considering questions to be sets of formulas. . Our SAM is inspired by the syntactic representation of questions in inferential erotetic logic founded by Andrzej Wi´niewski. From the syntactic viewpoint and being inspired by the previous examples. The best overview of questions’ formalization in inferential erotetic logic is in the chapter 3 of the book [44]. No wonder that we want to impose some restrictions on direct answers to keep their exclusive position. the question might be reformulated to Q : Who has the Joker: Ann. See the article [46] as well.} where α1 . . γ: Catherine has the Joker. which play the role of direct answers. α2 . Such restrictions are mostly combination of syntactic as well as semantic requirements. 2. various responses to Q can be received again. α2 . A question is the following structure ?{α1 . and γ. A set of direct answers has at least two elements. are formulas of the extended language. . Bill. δ is a complete answer to Q . . }) and question mark (?). In fact. Formulas α1 . A (general) declarative language L is extended only by curly brackets ({. .1. we require: 1. is an answer. . but the ‘core’ answers (the term direct answers will be used) are α. are syntactically distinct. . 4 8 . β: Bill has the Joker. β. Of course. The sentence δ: Neither Ann nor Bill have the Joker. or Catherine? The question Q is a combination of the question Who has the Joker? and the context Either Ann has the Joker or Bill has the Joker or Catherine has the Joker. Set-of-answers methodology α: Ann has the Joker.

(Ann has the Joker.) Such question will be identiﬁed with the form ?{α.2. the following two direct answers: Yes. ¬α} and shortened as ?α.2. Set-of-answers methodology Both restrictions introduce questions as ‘tasks’ with at least two distinct ‘solutions’. Negation is always related to a background system and receiving ¬α can mean more than ‘it is not the case that α’—it expresses something like ‘strict denial of α’. it brings more—the next subsection informally shows how to incorporate epistemic aspects. However. Syntactical distinctness is a ﬁrst step to the idea that direct answers form the ‘core’ of questions’ meaning. where an answer is a choice from two possibilities. in fact. Generally. both approaches are based on the idea of a questioner who does not know any answer to a question and who calls for a completion of knowledge. At ﬁrst sight. α2 }. 6 More details are in [16] and [44. It is natural to see epistemic aspects in the meaning of questions.) No. 1. The most typical questions with only two direct answers are yes-no questions. We believe that the introduced SAM is more or less successful in formalization of most types of natural language interrogatives with respect to the chosen background logical system. These theories are reductionist ones. questions are translated into epistemic-imperative statements. Compare it with an epistemic interpretation of relevant logic in [22]. A yes-no question is a variant of a whether question.3 Epistemic aspects of SAM In section 1.5 The role of negation is considered to be very important in SAM. chapter 2]. Looking for an epistemic counterpart of questions in the history of erotetic logic.1 we used the paraphrase Bring it about that I know who has the Joker! for the question Who has the Joker? as it is common in Hintikka’s analysis [17].2. The question Has Ann the Joker? has. a question expresses an ignorance of a questioner delivered to an addressee.1. 5 9 . the most known are epistemic-imperative approaches of ˚qvist A and Hintikka. Is two even or odd? is such a question with a possible formalization ?{α1 . In semantics we will require non-equivalence above that. (Ann has not the Joker.6 For example.

personal level. The period from the 1950s till 1990s is mapped in [15]. which is supposed to be used in a theory of communication. The reader can ﬁnd a comprehensive overview of the history of erotetic logic in [16]. Kubi´ski). Harrah. β. A n The ﬁrst comprehensive monograph on questions [1] brought into life many important terms used in erotetic logic so far. pragmatic level. The logic of questions has. a long history. Note on the recent history of erotetic logic Epistemic analysis of questions has two important levels. or Catherine? with the formalization based on the discussion in the previous section: Q = ?{α. considers an exchange of information in a group of agents. This will be studied in chapters 3 and 4. The late 1970s gave birth to inﬂuential reductionist theories: Hintikka’s epistemic-imperative approach and Tich´’s approach based on his transparent intensional logic [37]. Moreover. p. but the presupposition that the holder must be either Ann or Bill or Catherine. Let us return to the question Q : Who has the Joker: Ann. Bill. Stahl) and continued in the 1960s (˚qvist. Carnap seem to be the ﬁrst authors attempting to formalize questions in a logical framework—their attempts date back to the 1920s [16. In the personal level. moreover. β} publicly among them. chapter 2] as well.3. In particular. works with the knowledge and ignorance of a questioner. Just introduced SAM makes it possible to specify expectations and presuppositions of a questioner. if Catherine wants to ﬁnd out who has the Joker in the group of her friends-players. The ﬁrst one. The second one. To employ the pragmatic level we have to indicate a questioner and an addressee. both papers provide a good introduction to the terminology used in logic of questions and cover enough the history of erotetic logic till the 1990s. she could ask ?c {α.3 Note on the recent history of erotetic logic We are not going to present a complete survey of erotetic theories. maybe surprisingly. 1. A questioner expresses not only the ignorance of Joker’s holder. The ﬁrst ‘boom’ of logical approach to questions took place in the 1950s (Hamblin. y 10 . 3]. F. our SAM informs us what answers are considered as possible and.1. γ}. what is the rank of complete answers. Prior. Mainly linguistic viewpoint with the detailed discussion about the semantics of questions and pragmatic approaches can be found in [14]. Cohen and R. Erotetic theories with the main inﬂuence in this ﬁeld of study are described in [44.

s Both theories appears fully developed in the 1990s and we consider them giving birth to several approaches some of which are still inﬂuential. see [49] for classical propositional logic and [21] for some normal modal propositional logics. 1. This searching process can be seen as a tree with a principal question (and context expressed by declaratives) in the root. [46. A move from node to node is justiﬁed by IEL inferences in the direction to leaves with answers.1. 7 IEL used to be presented as an alternative to Hintikka’s approach (cf. The book [44] and article [46] contain a nicely written presentation of Wi´niewski’s s approach. Hintikka’s approach has an application in game-theoretic framework for belief revision theory. All important terms concerning interrogative inferences are introduced and studied in their mutual relationships. Conclusion relations among questions and declaratives are deﬁned on the metalanguage level where the role of multiple-conclusion entailment is important.3. Chapter 2 is devoted to the slightly modiﬁed IEL. [10]. cf. Primarily it is based on classical logic and a formalization of questions. which is very similar to the introduced SAM. The similar idea is developed in Hintikka’s interrogative model of inquiry reﬂecting the usefulness of questions in reasoning [18]. The advantage of IEL is its possible generalization for non-classical logics.7 The IEL methodology makes it possible to transform the derivability of a declarative formula into a sequence of questions and produce an analytic-tableaux style calculus—socratic proofs. The inﬂuence of Belnap’s and Kubi´ski’s works is apparn ent. In this chapter the reader can ﬁnd a list of relevant publications to this topic. The complex study of erotetic inferential structures in IEL predetermines studies based on an old idea that a (principal) question can be answered by asking auxiliary questions. 47]). Note on the recent history of erotetic logic Most inﬂuential modern logics of questions with the important role of erotetic inferences are • intensional approach of Jeroen Groenendijk and Martin Stokhof and • inferential erotetic logic (IEL) of Andrzej Wi´niewski.3. see the paper [47]. Nodes bear auxiliary questions (with context).1 Inferential erotetic logic Wi´niewski’s IEL is a complex system dealing with various interrogative ins ferential structures. Erotetic search scenarios is the name for this approach. 11 .

It is a good inspiration for epistemic representations of questions.2 Intensional erotetic logic Groenendijk’s and Stokhof’s approach can be called intensional erotetic logic. This intensional approach inﬂuenced many works in the last ten years.) Moreover.8 [34] gives a syntactic characterization of answerhood for the partition semantics of questions and then the authors implement partition semantics in question answering algorithm based on tableaux theorem proving [35]. Extension of a declarative is its truth value in a given situation. cf. see [6] and [29]. Intensional semantics is the background of the meaning of questions. situations). (A question Q provides a reﬁnement of the Q1 -partitioning. [27] brings a many-valued interpretation of declaratives and interrogatives based on bilattices and in the paper [25] Gentzen style calculus is presented. Logical space is understood as a set of all ‘possible states’ (possible worlds.3. Combining intensional semantics with the full acceptance of Hamblin’s postulates we obtain the meaning of a question as a partitioning of logical space. where sets of answers form distributive lattices. answers to a question form exhaustive set of mutually exclusive propositions. Questions’ representation is similar to SAM with restrictions posed by Hamblin’s postulates. many important terms are naturally deﬁned (partial answer. 8 12 . which is true there. is studied also in [19]. and others). informative value of answers. This approach introduces an entailment between two questions as a reﬁnement of a partitioning: A question Q entails a question Q1 iﬀ each answer to Q implies an answer to Q1 . 9 Nice and brief comments are in [12]. See [13] and [14]. complete answer.3.. However. in case of yes-no questions. [40]. The meaning of a declarative sentence is given by truth conditions and forms a subset of logical space. Groenendijk’s and Stokhof’s intensional interpretation inspired some extensional approaches. Partitioning of logical space is the intension of a question. Intension of a declarative is then a set of states where the declarative is true (this is called proposition). indexes. e. we can imagine that there is a scale of answers. Note on the recent history of erotetic logic 1.1.g. In accordance with the third postulate.9 Presented logics of questions deal with questions as having ‘crisp’ answers. Extension of a question in a given situation is the answer. Very recently the logic of questions receives more attention in connection with dynamic aspects of epistemic logic and communication theory. the scale yes — rather yes — rather no — no An algebraic approach.

but it corresponds to comparative degrees of truth of the original yes-/no-answer. its combination with logic of questions is still rather underdeveloped. Truth-degrees are studied in multi-valued logics. Note on the recent history of erotetic logic is usual in questionnaires. Although fuzzy logic seems to be suitable for the study of reasoning under vagueness. Such kind of scale does not require to introduce four new answers. 13 .3.1. The paper [4] presents propositional Groenendijk-Stokhof’s erotetic logic with fuzzy intensional semantics based on fuzzy class theory.

and each ﬁnite and at least two-element set of sentences is the set of direct answers to some question [46.} For the set of direct answers of a question Q we will use the symbol dQ. On the syntactic level of a considered formalized language. Direct answers are declarative formulas. each question has at least two direct answers. }) and question mark (?). 2. ∨. relationships.1 In the correspondence with Section1.1. . and possible generalizations. . A (general) declarative language L is extended by curly brackets ({. 1 14 . 11]. p.1 Adapted set-of-answers methodology in IEL Inferential erotetic logic accepts only the ﬁrst two Hamblin’s postulates and tries to keep the maximum of the (classical) declarative logic and its consequence relation.1 Introduction Inferential structures that will be introduced and studied in this chapter are based on the slightly adapted inferential erotetic logic (IEL). Let us apply our SAM introduced in the previous chapter.2 a question Q is the following structure ?{α1 . We deﬁne a (general) erotetic language LQ . . In this chapter we use only propositional examples in the language with common connectives (∧. →. a question is assigned to a set of sentences (direct answers).Chapter 2 Consequence relations in inferential erotetic logic 2. α2 . ¬). We utilize the framework of this theory and show some properties.

• A conjunctive question ?|α. β. • Q. .1. . . or β (and not α). . we require here that elements of dQ are declarative sentences. for questions. ¬α}. . . . 2. . β. and so on. The class of ﬁnite questions corresponds to the class of questions of the ﬁrst kind in [44]. The general framework of IEL allows for other ways of formalizing questions. Q1 .1. γ. Φ1 . ϕ.) for declarative sentences. . γ|. • capital Greek letters (Γ. αn } we suppose that the listed direct answers are semantically non-equivalent. In case of ﬁnite versions of questions ?{α1 . the expression which occurs after the question mark designates the set of direct answers to the question. . Similar versions are ?|α. . ?|α.3 To avoid a misunderstanding. Questions are deﬁned in such a way that sets of direct answers to them are explicitly speciﬁed. (¬α ∧ ¬β)}. . or neither α nor β. then the term atomic yes-no question is used. Moreover. Some of them are important in our future examples and counterexamples. we will use the following metavariables in this chapter: • small Greek letters (α. s 15 . which is an abbreviation for ?{α. It is an abbreviation for ?{(α ∧ β).2 In the original version of IEL. (α ∧ ¬β). . Introduction Let us repeat that direct answers are syntactically distinct and |dQ| ≥ 2. for sets of questions. . If α is an atomic formula. . Let us mention two abbreviations and terms that are very frequent in this thesis: • Simple yes-no questions are of the form ?α. β| requires the answer whether α (and not β). but the form is designed in such a way that. etc. δ|. β|. on the metalanguage level (and only here). Personal communication with Andrzej Wi´niewski. questions are not identiﬁed with sets of direct answers: questions belong to an object-level language and are expressions of a strictly deﬁned form.) for sets of declaratives. (¬α ∧ β).2 Consequence relations in IEL Consequence relations are the central point of logic. Declarative logic can be deﬁned by its consequence relation as a set of pairs Γ.2. ∆ . or both (α and β). β. ∆. . where Γ and ∆ are 2 3 Original IEL uses the symbol ?±|α. and • Φ.

Γ.). let us recommend texts [44.3 Model-based approach The following model-based approach was inspired by minimal erotetic semantics from [46]. Because of the possibility of adding some other constraints for models we will deal with (e.1. Speaking about tautologies of a logic L we mean the set of formulas TautL = {ϕ (∀M ∈ ML )(M |= ϕ)}. the background logic and restrictions posed on models will be stated explicitly. 44. 2. Motivations and natural-language examples of these consequence relations will be introduced in the next sections. Q1 between an initial question Q and an implied question Q1 with respect to a set of declaratives Γ. ﬁniteness. If necessary. 50]. The most important relations. Our aim is to study erotetic consequence relations in a very general manner. are the following: • Evocation is a binary relation Γ. In the literature. In case of predicate logic it is a set of all structures with a realizations of non-logical symbols. Q between a set of declaratives Γ and a question. preferred models. • Reducibility is a ternary relation Q. • Erotetic implication is a ternary relation Q.. The term model varies dependently on a background logic L. independently of the logic behind. which we are going to introduce. reducibility is studied in [20. then MCP L is a set of all valuations.1. The deﬁnitions of IEL consequences are based on the semantic entailment and the model approach relative to the chosen logical background. If L is classical propositional logic (CPL.g. Γ. etc. Inferential erotetic logic makes one step more and adds new consequence relations mixing declaratives and interrogatives. let us generally use a set M ⊆ ML .2. Φ between an initial question Q and a set of questions Φ with respect to a set of declaratives Γ. 16 . 46] for both the evocation and erotetic implication. for short). Let us introduce the set of all models for a declarative language as follows: ML = {M M is a (semantic) model for L}. Introduction sets of (declarative) formulas and ∆ is usually considered to be a singleton.

the following theorem could be surprising at the ﬁrst sight. for short). ∆ be a set of sentences containing at least one tautology and at least one contradiction. (Semantic) entailment Let us recall the common (semantic) entailment relation. but it is neither symmetric nor transitive relation: Example 1. let us write Γ ≡ ∆. 5 We say that mc-entailment is compact iﬀ for each Γ ||= ∆ there are ﬁnite subsets G ⊆ Γ and D ⊆ ∆ such that G ||= D. two diﬀerent sets of models do not imply the existence of two diﬀerent sets of sentences (in L). L All semantic terms may be relativized to M. we introduce multiple-conclusion entailment (mc-entailment. Entailment is deﬁnable by mc-entailment: Γ |= ϕ iﬀ Γ ||= {ϕ} On the other hand. Let Γ ⊆ TautL . On the other hand. In this context. mc-entailment is not deﬁnable by entailment. In case Γ = {ϕ} we write only ϕ |= ψ.5 In case of the semantic equivalence of formulas ϕ and ψ it will be only written ϕ ≡ ψ. 4 17 . Introduction If a restricted set of models M is in use.1.4 Mc-entailment is reﬂexive (Γ ||= Γ). ϕ |= ψ iﬀ Mϕ ⊆ Mψ Now. Γ ||= ∆ iﬀ MΓ ⊆ Mδ δ∈∆ If MΓ = M∆ . Then Γ ||= ∆ and ∆ ||= Σ. but Γ ||= Σ. Each declarative sentence ϕ (in the language L) has its (restricted) set of models Mϕ = {M ∈ M M |= ϕ} and similarly for a set of sentences Γ MΓ = {M ∈ M (∀γ ∈ Γ)(M |= γ)}.2. we speak about M-tautologies TautM = {ϕ (∀M ∈ M)(M |= ϕ)}. For any set of formulas Γ and any formula ψ: Γ |= ψ iﬀ MΓ ⊆ Mψ . and Σ be such that σ∈Σ Mσ ⊂ ML .

2.1. Introduction

Theorem 1. Entailment (for logic L) is compact iﬀ mc-entailment (for L) is compact. Proof. See [44, pp. 109–110].

2.1.4

Basic properties of questions

After we have introduced the SAM representation of questions and the model-based approach, we can mention some basic properties of questions. First, let us introduce the term soundness, which is one of the most important terms in IEL. Deﬁnition 1. A question Q is sound in M iﬀ ∃α ∈ dQ such that M |= α. A question is sound with respect to a model M whenever it has at least one direct answer true in M. See [44, p. 113]. For all IEL consequnce relations, it is important to state the soundness of a question with respect to a set of declaratives. Deﬁnition 2. A question Q is sound relative to Γ iﬀ Γ ||= dQ. The sum of all classes of models of each direct answer α, i.e., α∈dQ Mα , is called semantic range of a question Q. Considering semantic range, our liberal approach admits some strange questions; one of them is a completely contradictory question that has only contradictions in its set of direct answers, its semantic range beeing just ∅. Another type is a question with a tautology among its direct answers, then the semantic range expands to the whole M. Questions with such a range are called safe.6 Of course, it need not be any tautology among direct answers for to be a safe question. Deﬁnition 3. • A question Q is safe iﬀ

α∈dQ α∈dQ

Mα = M.

• A question Q is risky iﬀ

Mα ⊂ M.

Questions ?α, ?|α, β| are safe in CPL, but neither is safe in Bochvar logic. If β is not equivalent to ¬α, then ?{α, β} is risky in CPL. Neither ?α nor ?|α, β| are safe in intuitionistic logic, but there are safe questions in this logic; just each question with at least one tautology among direct answers. Simple yes-no questions are safe in logics that accept the law of excluded middle. It is good to emphasize that the set of direct answers of a safe question is mc-entailed by every set of declaratives. On the other hand, knowing a question to be sound relative to every set of declaratives implies its safeness. Fact 1. Q is safe iﬀ (∀Γ)(Γ ||= dQ). Specially, safe questions are sound relative to Γ = ∅.

6

This term originates from Nuel Belnap.

18

2.2. Questions and declaratives

2.1.5

The road we are going to take

After introducing evocation and the term presupposition of a question in section 2.2, we will show the role of maximal and prospective presuppositions in the relationship to semantic range of questions. Some classes of questions will be based on it. One could be surprised that we are not going to discuss answers in this section; in fact, there is not much to say about them. It turns out that various types of answers do not play any special role in the inferential structures. Section 2.3 is crucial from the chosen viewpoint. We investigate erotetic implication and reducibility there. An important part is devoted to a discussion of the role of an auxiliary set of declaratives. We will demonstrate some variants of erotetic implication and their properties. The chosen formal shape of questions in IEL makes it possible to compare questions in the sense of their answerhood power. Inspired by [13] and [44, section 5.2.3] we will examine the relationship of ‘giving an answer’ of one question to another, which is a generalisation of Kubi´ski’s term ‘weaker question’. n Questions will be considered as independent structures not being combined by logical connectives. Reviewing the deﬁnitions of erotetic implication and reducibility we can recognize their ‘both-sidedness’ and just reducibility can substitute such combination of questions. This brings us to the last note on the use of symbols |= and . Because of the clear border between declarative and interrogative parts of the language LQ we will use them in many meanings. However, the meaning will be transparent by the context the symbols are used in. Compare the deﬁnition of evocation and various deﬁnitions of erotetic implications in the next sections.

2.2

Questions and declaratives

In this section, we introduce two terms: evocation and presupposition. The ﬁrst one will provide a consequence relation between a set of declaratives and a question. The second one is an important term in almost all logics of questions and there are some classes of questions based on it in IEL.

2.2.1

Evocation

Consider the following example: after a lecture, we expect a lecturer to be ready to answer some questions that were evoked by his or her talk. Thus, evocation seems to be the most obvious relationship among declarative sentences and questions. (Of course, next to the connection question—answer.)

19

2.2. Questions and declaratives

Almost every information can give rise to a question. What is the aim of such a question? First, it should complete our knowledge in some direction. Asking a question we want to get more then by the conclusion based on a background knowledge. A question Q should be informative relative to Γ, it means, there is no direct answer to Q which is a conclusion of Γ. Second, after answering an evoked question, no matter how, the answer must be consistent with the evoking knowledge. Moreover, transmission of truth into soundness is required: if an evoking set of declaratives has a model, there must be at least one direct answer of the evoked question that is true in this model. An evoked question should be sound relative to an evoking set of declaratives (see Deﬁnition 2).7 The deﬁnition of evocation is based on the previous two points (cf. [44, 46]). A question Q is evoked by a set of declaratives Γ if Q is sound and informative relative to Γ. Deﬁnition 4. A set of declarative sentences Γ evokes a question Q (let us write Γ |= Q) iﬀ 1. Γ ||= dQ, 2. (∀α ∈ dQ)(Γ |= α). In our model-based approach we can rewrite both conditions this way: 1. MΓ ⊆

α∈dQ

Mα

2. (∀α ∈ dQ)(MΓ ⊆ Mα ) In some special cases (e.g., dQ is ﬁnite or entailment is compact) we can deﬁne evocation without the link to mc-entailment. The ﬁrst condition is of the form: there are α1 , . . . , αn ∈ dQ such that Γ |= n αi . 1 This is the case of one of our introductory examples. Let us remind the group of three card players. The context Γ: Either Ann has the Joker or Bill has the Joker or Catherine has the Joker. evokes the question Q: Who has the Joker: Ann, Bill, or Catherine?

For now, as we do not discuss epistemic issues, we shall not use the word ‘knowledge’ but the phrases ‘set of declarative(s) (sentences)’ or ‘database’ will be used instead.

7

20

21 . The following fact lists some of them. chapter 6]. we obtain a less intuitive conclusion: every safe question is evoked by any Γ that does not entail any direct answer to it. we could add the third condition to Deﬁnition 4: (∀α ∈ dQ)(MΓ ∩ Mα = ∅) Some solutions of the problem of irrelevant and ineﬃcient questions based on a semantics in the background are discussed in the just mentioned paper [7].. Γ = {α ∨ β. • there is no tautology in dQ. by Fact 1. It underlines the special position of safe questions and their semantic range. thus. β. Generation does not solve all problems with irrelevant and ineﬃcient evoked questions either. If Γ |= Q. For our purpose. Let Γ consist of one formula in the form of disjunction of direct answers of Q.2. γ: Catherine has the Joker. let us mention the following fact: Fact 3.2. the consistency of each direct answer with respect to Γ could be required. If ∅ |= Q. β: Bill has the Joker. the ﬁrst condition is satisﬁed. Back to safe questions. Borrowing an example from [7]. and • Q is not a completely contradictory question. We can accept another restriction to avoid questions that have direct answers which are incompatible with declaratives in Γ. γ} evokes also ?{α.e. then Q is safe. then • Γ is not a contradictory set. i. we keep Deﬁnition 4 unchanged. However. Fact 2. we get the deﬁnition of generation. As a conclusion of semantic deﬁnition of evocation we have the following expected behavior of evocation: semantically equivalent databases evoke the same questions. ¬γ}. The second one is satisﬁed because no direct answer is entailed by Γ. see [44. the study of consequence relations in IEL. Questions and declaratives with direct answers α: Ann has the Joker. Evocation yields some clear and useful properties of both a set of declaratives and an evoked question. When we restrict the deﬁnition of evocation to risky questions only. To eliminate this.

A declarative formula ϕ is a presupposition of a question Q iﬀ (∀α ∈ dQ)(α |= ϕ). (α1 ∨ α2 ∨ ϕ). see section 2. if Γ ≡ ∆. If Γ |= Q and the entailment is compact.2. At the ﬁrst sight. These and some more properties of evocation (and generation) are discussed in the book [44]. What is presupposed must be valid under each answer to a question. an answer to a question should bring at least the same information as presupposition does. then we have to be careful of concluding that there is a subset ∆ ⊆ Γ such that ∆ evokes Q. Formula (α1 ∨ α2 ) entails each presupposition of the question Q. the set PresQ could contain a lot of sentences. For every Γ. 22 .. α2 }. The second item points out the non-monotonicity of evocation (in declaratives). The following deﬁnition (originally given by Nuel Belnap) is from [44]: Deﬁnition 5. Bill. A presupposition of a question is entailed by each direct answer to the question. see the ﬁrst item in the following fact. (α1 ∨ α2 ∨ ¬ϕ). Questions and declaratives Fact 4. then • ∆ |= Q if ∆ ||= dQ. If Γ |= Q and ∆ ⊆ Γ ⊆ Σ. Let us have a question Q = ?{α1 . in CPL) contains (α1 ∨ α2 ). then ∆ |= Q1 for some ﬁnite subset dQ1 of dQ and some ﬁnite subset ∆ of Γ.3. Fact 6. we can recognize that it is presupposed that Ann has it or Bill has it or Catherine has it. then Γ |= Q iﬀ ∆ |= Q. If we receive the question Who has the Joker: Ann. the set of presuppositions (e. ∆ and Q. or Catherine?.g. Everyone who has attended a basic course of research methods in social sciences has heard of importance to consider presuppositions of a question in questionaries. Let us write PresQ for the set of all presuppositions of Q. Moreover. Fact 5. • Σ |= Q if (∀α ∈ dQ)(Σ |= α). If Γ evokes Q. etc.2. Considering questions as sets of answers.2.3. 2.2 Presuppositions Many properties of questions are based on the concept of presupposition. evocation is non-monotonic in interrogatives as well. Looking at the very relevant member (α1 ∨ α2 ) it is useful to introduce the concept of maximal presupposition.

Such information is relatively small. each of them is in PresQ. then ϕ ≡ ψ. If M |= ϕ. then there is α ∈ dQ such that M |= α. L This fact says that if Q is safe. then PresQ = TautM . We got ϕ |= ψ. then we know that α∈dQ Mα is a proper subset of M. without answering it. But what about MPresQ ? After introducing a class of normal questions (see page 25) it will be valid MPresQ ⊂ M as well as the implication from right to left (see Fact 9). All prospective presuppositions of a question are equivalent: Lemma 1. If the background logic has tautologies. ψ |= ϕ is proved by the same way. then α∈dQ Mα = MPresQ . A declarative formula ϕ is a prospective presupposition of a question Q iﬀ ϕ ∈ PresQ and ϕ ||= dQ. The deﬁnition of presupposition gives α∈dQ Mα ⊆ Mϕ . If Q is not safe. It is a presupposition which a question Q is sound relative to. A presupposition can be seen as an information which is announced by asking a question. Questions and declaratives Deﬁnition 6. If ϕ. The semantic range of all maximal presuppositions is wider than the range of a question. Let us write ϕ ∈ PPresQ.2.2. which means Mα ⊆ α∈dQ ϕ∈PresQ Mϕ = MPresQ and the set MPresQ is a model-based counterpart to the deﬁnition of maximal presuppositions. The model-theoretical view shows it in a direct way. TautM ⊆ PresQ L Considering safe questions we get Fact 7. Proof. Looking at ﬁnite CPL example where the disjunction of all direct answers forms just the semantic range of the question brings us to the idea of prospective presupposition. α |= ψ and it gives M |= ψ. If Q is safe. A declarative formula ϕ is a maximal presupposition of a question Q iﬀ ϕ ∈ PresQ and (∀ψ ∈ PresQ)(ϕ |= ψ). then Q is safe. ψ ∈ PPresQ. 23 . for each ϕ ∈ PresQ. In classical propositional logic the disjunction of all direct answers of a question is a presupposition of this question and if PresQ = TautM . CPL This evokes a (meta)question whether the implication from right to left is valid. Deﬁnition 7. Since ψ ∈ PresQ.

We will return to this in the next subsection at the topic of normal questions. because ψ ∈ PresQ. To sum up all general conditions of an evoked question (by Γ) and its presuppositions let us look at this diagram: MΓ ⊆ α∈dQ Mα = MPPresQ ⊆ MPresQ Classes of questions based on presuppositions Using the term presupposition we can deﬁne some classes of questions. Mα = MPPresQ α∈dQ If Q has a prospective presupposition. Clearly. 24 . Fact 8. Questions and declaratives A prospective presupposition forms exactly the semantic range of a question. We show that if ϕ ∈ PPresQ and ψ ∈ PPresQ1 . Lemma 2. the informativeness must be ensured as well. then Γ |= ϕ. Two questions with the same sets of presuppositions have the same prospective presuppositions. If PresQ = PresQ1 and both PPresQ and PPresQ1 are not empty. The implication from right to left does not hold. If Γ |= Q. Let us note that it cannot be improved by replacing of PresQ by PPresQ. Names and deﬁnitions of the classes are from [44].2. we are not sure about MΓ ⊆ α∈dQ Mα as required by the ﬁrst condition of evocation. then PPresQ = PPresQ1 . If we only know MΓ ⊆ MPresQ .2. The proof that ψ |= ϕ is similar. Presuppositions of evoked questions are entailed by the evoking set of declaratives. We only add the model-based approach and make transparent some results on presuppositions and evocations (chapters 4 and 5 in [44]). then ϕ ≡ ψ. it can be understood as the ‘strongest’ one. It gives Mϕ ⊆ Mψ and ϕ |= ψ. Proof. for each ϕ ∈ PresQ. ϕ ∈ PPresQ implies Mϕ = α∈dQ Mα and α∈dQ Mα ⊆ Mψ .

that the class of safe questions is a subset of the class of normal questions.23 in [44]. Non-normal questions can be found in classical predicate logic. for each ϕ ∈ PresQ. 8 Cf. L Let us only add the clear fact. Normal questions are sound relative to PresQ and regular questions are sound relative to PPresQ. Questions and declaratives Normal questions A question Q is called normal if it is sound relative to its set of presuppositions (PresQ ||= dQ).2. 25 . If Γ |= ϕ. They continue on the discussions at Fact 7 and Fact 8. safe ⊆ normal The following fact and Fact 8 give the conditions for evocation of normal questions.8 Fact 10.2. • Q ∈ normal iﬀ α∈dQ Mα = MPresQ Model-based approach introduces normal questions as questions with semantic range delimitated by models of maximal presuppositions. Theorem 5. Working with ﬁnite sets of direct answers and in logical systems with the ‘classical’ behavior of disjunction (each direct answer entails the disjunction of all direct answers) we do not leave the class normal. • Q ∈ regular iﬀ (∃ϕ ∈ PresQ)(ϕ ||= dQ) Regularity of Q gives MPresQ ⊆ Mϕ ⊆ α∈dQ Mα and it holds regular ⊆ normal If entailment is compact. α}. Fact 9. then Γ |= Q. Let Q = ?{(α ∨ β). The following example shows an expected fact that it is still not suﬃcient for evocation. Regular questions Each question with the non-empty set of prospective presuppositions is regular. and Γ |= α. for each α ∈ dQ of a normal question Q. If PresQ = TautM and Q is normal. Example 2 (in CPL). Two announced facts follow. but PresQ |= Q. both classes are equal. then Q is safe. the formula (α ∨ β) is a prospective presupposition of Q. This question is normal and regular.

This is summed up by Lemma 3. Then 1. do we ask such questions? This class includes such strange questions as completely contradictory questions that have only contradictions in the set of direct answers. Questions and declaratives If there is a set of declaratives Γ such that Γ |= Q. Proper questions Normal and not self-rhetorical questions are called proper. The second item is provable by the same idea. 2. Lemma 4. for some set of declaratives Γ. Q ∈ normal implies PresQ |= Q. and questions with tautologies among direct answers. However. Self-rhetorical questions Another special class of questions are self-rhetorical questions. • Q ∈ self-rhetorical iﬀ (∃α ∈ dQ)(PresQ |= α) From this deﬁnition. 26 . Q ∈ regular implies ϕ |= Q. If there is Γ such that Γ |= Q. Proper questions are evoked by their set of presuppositions. then normal (regular) questions are sound as well as informative relative to PresQ (PPresQ). The set of all presuppositions of a question Q is believed to be a natural (declarative) context for evocation of Q. it is clear that self-rhetorical questions are normal. then Fact 8 causes non-informativeness of Q relative to Γ.2. From Fact 8. An evoked question is not of this kind. Proof. Let Γ |= Q. • Q ∈ proper iﬀ PresQ |= Q Evoked normal questions are proper (compare both Lemma 3 and Lemma 4) and this makes the class proper prominent. Proof. for ϕ ∈ PPresQ. then Q is not self-rhetorical.2. For the ﬁrst item. They have at least one direct answer entailed by the set of presuppositions. only informativeness (relative to PresQ) must be showed. But if it is not valid.

we say that Q implies Q1 on the basis of Γ and write Γ. If I ask Q: What is Peter a graduate of: a faculty of law or a faculty of economy? then I can be satisﬁed by the answer He is a lawyer. we extend the class of inferences by ‘implication’ between two questions with a possible assistance of some set of declaratives.2. speaking of Peter’s position. can be ?{β1 . 2.3. for short) and the following deﬁnition is from [44]:9 Deﬁnition 8. 27 . Looking at the questions there is no connection between them. β2 }. The ﬁrst question Q can be formalized by ?{α1 . This relation is called erotetic implication (e-implication. even if I did not ask Q1 : What is Peter: a lawyer or an economist? The connection between both questions could be shown by the following set of declaratives: Someone is a graduate of a faculty of law iﬀ he/she is a lawyer. Now. α2 } and the latter one. Questions and questions 2.1 Erotetic implication Now. Let us start with an easy and a bit tricky example. Someone is a graduate of a faculty of economy iﬀ he/she is an economist. The second point focuses on ‘answerhood power’ of questions formalized by the adapted set-of-answers methodology.3 Questions and questions This section is devoted to inferential structures in which questions appear on both sides (erotetic implication and reducibility of questions to sets of questions) and to relations between two questions based on their sets of direct answers. A question Q implies a question Q1 on the basis of a set of declaratives Γ iﬀ 9 We will write shortly Γ ∪ ϕ instead of Γ ∪ {ϕ}. The relationship is based on the set of declaratives Γ = {(α1 ↔ β1 ).3. Q |= Q1 . (α2 ↔ β2 )}.

This transmission of truth/soundness into soundness has the following meaning: if there is a model of Γ and a direct answer to Q. Γ. for each Q and Q1 . In our semantic approach. Γ includes only tautologies of a chosen logical system. for any set of declaratives ∆. If Q1 is safe. 28 . If Γ. then there must be a direct answer to Q1 that is valid in this model. 173]: Fact 11. the role of the set of declaratives is a bit diﬀerent. the auxiliary role.3. then ∆. The ﬁrst clause should express the soundness of an implied question relative to each extension of Γ by α ∈ dQ. then they are in the relation of e-implication for each set of declaratives. If one question purely e-implies another question.2. Q |= Q1 as well as Γ. (∀α ∈ dQ)(Γ ∪ α ||= dQ1 ). then this condition is always valid (see Fact 1). From this. In comparison with evocation. From Fact 11. This could be called weakening in declaratives. p. especially. If Q |= Q1 . both questions are even erotetically equivalent with respect to Γ: Γ. e-implication is monotonic in declaratives and it gives the following [44. (∀β ∈ dQ1 )(∃∆ ⊂ dQ)(∆ = ∅ and Γ ∪ β ||= ∆). then α∈dQ Mα = β∈dQ1 Mβ . We wil say a word or two about auxiliary sets of declaratives in the next subsection. then both questions have the same semantic range. Questions and questions 1. whenever two questions are in the relation of pure e-implication. Q |= Q1 . Pure erotetic implication Pure e-implication is e-implication with the empty set of declaratives. 2. Q1 |= Q. it is clear that ⊥. The deﬁnition requires a little comment. Q |= Q1 . From the ﬁrst condition of Deﬁnition 8 Mα ⊆ α∈dQ β∈dQ1 Mβ and from the second one Mβ ⊆ β∈dQ1 ∆ α∈∆ Mα ⊆ α∈dQ Mα . Γ plays. The second clause requires direct answers to Q1 to be cognitively useful in restricting the set of direct answers of the implying question Q. Proof. Lemma 5. Q |= Q1 . Returning to the introductory example.

Theorem 2. 184] See the same result in [44.3. Simultaneously. then Q is safe (risky) iﬀ Q1 is safe (risky). The same semantic range of questions linked together by pure e-implication does not form an equivalence relation on questions (see non-symmetry in Example 4 and non-transitivity in Example 5). If Q |= Q1 . If Q |= Q1 . such that Mβ ⊆ α∈∆ Mα ⊆ α∈dQ Mα . and the third one is from Lemma 6. Mβ ⊆ Mϕ . 10 11 Cf. for each β ∈ dQ1 . the second one is from the normality of Q. then Q is regular iﬀ Q1 is regular. p. On the other hand. Theorem 3.29 in [44. Theorem 7. then Mβ = β∈dQ1 α∈dQ Mα = MPresQ = MPresQ1 The ﬁrst equation is from Lemma 5. On this lemma we can base the following statement about an inﬂuence of pure e-implication on classes of normal and regular questions. The claim of Lemma 6 is not extendable to the general e-implication (cf. First.33. Example 3). It is easy to prove a similar fact for regular questions.11 Lemma 6. If Q |= Q1 .10 Fact 12. Thus. If Q is normal. for each β ∈ dQ1 . for proving PresQ1 ⊆ PresQ suppose ϕ ∈ PresQ1 . If Q |= Q1 . so α∈dQ Mα ⊆ Mϕ . Let ϕ ∈ PresQ. 184]. Theorem 7. 29 . Proof. The following inclusions are valid Mα ⊆ β∈dQ1 Mβ ⊆ Mϕ . then PresQ = PresQ1 . Questions and questions From this we can conclude that classes of safe and risky questions are closed under pure e-implication for both implied and implying questions.2. Proof. p. Second. let us prove PresQ ⊆ PresQ1 . we know that from the second condition of the deﬁnition of pure e-implication there is a non-empty ∆ ⊂ dQ. for each α ∈ dQ. pure e-implication has some important consequences for classes of presuppositions. then Q is normal iﬀ Q1 is normal.

the implication from Q1 to Q2 is justiﬁed. pp. β. 185–186]. Let us have Γ and two questions Q1 and Q2 . Mϕ = β∈dQ1 Mβ .2 and in the last paragraph of section 2. Normal (regular) questions purely imply only normal (regular) questions and they are purely implied by the same kind of questions. If the set PPresQ2 is explicitly expressed. the following example will point out the role of implicitly and explicitly expressed presuppositions. Note on auxiliary sets of declaratives in e-implication Let us remind the introductory example on page 27 to emphasize the importance of declaratives for e-implication. In the following fact we display when we can say that two questions and a set of declaratives are in the relationship of e-implication. β} (for atomic α. We will add some points to this discussion in section 2. Similarly. Thus. then PPresQ2 ||= dQ2 and. γ} and Q2 = ?{α. Questions and questions Proof.3. non-empty proper subset of dQ1 .3. From Q |= Q1 and Lemma 5 we obtain α∈dQ Mα = β∈dQ1 Mβ . On the other hand. Putting it together. which means that MPPresQ = MPPresQ1 . Q1 |= Q2 it is suﬃcient to have Γ ||= dQ2 and Γ ||= ∆.12 Concerning classes of questions in relationship with e-implication.2. Classes of normal and regular questions are closed to pure e-implication. γ). β. then (α ∨ β). in addition. But now. then Q1 implies Q2 with respect to Γ. Example 3 (in CPL). Keeping the context of this example: the question Q2 is normal as well as regular. such that PPresQ2 ||= ∆. If Q1 = ?{α. Q1 |= Q2 . Both theorems have similar results we have got for safe (risky) questions in Fact 12. Q1 |= Q2 . then Q is completely contradicory question iﬀ Q1 is. The regularity of Q implies that there is a formula ϕ ∈ PPresQ. let us add that whenever Q |= Q1 . In order to conclude Γ. ϕ ∈ PPresQ1 . there is ∆. then neither Q1 |= Q2 nor Q2 |= Q1 (see the diﬀerent semantic ranges of both questions). 30 . Let us suppose Q |= Q1 and Q is regular. 12 Theorems 2 and 3 put together results in [44.4. It gives PPresQ2 . if we would know that it must be (α ∨ β). where ∆ is a non-empty proper subset of dQ1 . This fact can be formulated in this form: if Q2 is sound relative to Γ and Γ gives a partial answer to Q1 . back to the general approach.3. Fact 13.

Then we say that Q regularly implies Q1 (on the basis of Γ). 31 . (∀β ∈ dQ1 )(∃α ∈ dQ)(β |= α). If Q Q1 such that we can answer Q1 . Q |= Q. (∀β ∈ dQ1 )(∃α ∈ dQ)(Γ ∪ β |= α). (∀α ∈ dQ)(α ||= dQ1 ). 26]: Deﬁnition 9. but Q2 |= Q1 .3. 2. then we have an answer to Q. Then Q1 |= Q2 . generally. Regularity can be enforced by the minimal number of direct answers of an implying question: if Q |= Q1 and |dQ| = 2. both conditions are changed into the form: 1. it is not a symmetric relation. Questions and questions Regular erotetic implication A special kind of e-implication arises if there is exactly one direct answer in each ∆ in the second clause of Deﬁnition 8. β}.3. The following deﬁnition originates from [46. then Q Q1 . Because of the special importance of this relation let us use the symbol for it (so we write Γ.3). for each Γ and Q. The relationship of pure regular e-implication between two questions says that the implied question is ‘stronger’ than the implying one in the sense of answerhood (see also section 2. (∀α ∈ dQ)(Γ ∪ α ||= dQ1 ). Erotetic implication is a reﬂexive relation. Even if there are examples of the symmetric behavior of e-implication. we are going to be interested in such properties as reﬂexivity. symmetry. α} and Q2 = ?{α. In the case of pure regular e-implication. p. Γ. Example 4 (in CPL). A question Q regularly implies a question Q1 on the basis of a set of declaratives Γ iﬀ 1. Let Q1 = ?{(α ∨ β). Basic properties of erotetic implication In this subsection. Fact 14.2. Q Q1 ). and transitivity of e-implication. 2.

As shown in the next example. The ﬁrst condition of Deﬁnition 9 is proved by Lemma 5. ?(α ∧ β) ?α. then Γ. let us add that presuppositions of an implied question are entailed by each direct answer of an implying question (with respect to an auxiliary set of declaratives). Fact 16. Let Γ. then (∀β ∈ dQ1 )(∀ϕ ∈ PresQ)(Γ ∪ β |= ϕ). As a ﬁnal remark. We can do a cautious strengthening by the following fact: Fact 15. Questions and questions In this example there is no non-empty proper subset of dQ2 for the formula (α ∨ β) to fulﬁl the second condition in the deﬁnition of e-implication.3. β| |= ?α. Q1 Q2 and Q2 Q3 . then Q1 Q3 . β| and {α}. Q |= Q1 . Then 1. Proof. Example 6 (in CPL). If Γ. • {(α ∨ β)} |= ?|α. β| and ?|α. it does not mean that it must be Γ |= Q2 .3. e-implication does not preserve evocation.2 Evocation and erotetic implication Both types of inferential structures can appear together and we are going to investigate their interaction. ?|α. but ?(α ∧ β) |= On the other hand. • {α} |= ?|α. β| |= ?(α ∨ β). 32 . β| |= ?α. Example 5 (in CPL). it is useful to add that Q1 regularly implies Q2 . 2. (∀α ∈ dQ)(∀ϕ ∈ PresQ1 )(Γ ∪ α |= ϕ) 2. Q1 Q3 . but there is an answer to ?α in {α}. the following theorem is valid. If e-implication is regular. The second clause of this deﬁnition is based on regularity that gives (∀γ ∈ dQ3 )(∃α ∈ dQ1 )(Mγ ⊆ Mα ). Erotetic implication is not transitive either. β| and ?|α. Moreover. If we know Γ |= Q1 and Q1 |= Q2 .2. Theorem 4. but {(α ∨ β)} |= ?(α ∨ β). If Q1 Q2 and Q2 Q3 . ?|α. if we consider regular e-implication only.

Proof. Then MΓ ⊆ Mβ and. we prove Γ ||= dQ2 .3. 13 33 . 14 The second part of the proof of Theorem 5 could be slightly changed and we obtain that strong e-implication carries over as well. then Γ |= Q2 . α.2. resp. Theorem 5. Supposing it is not true. there are some solutions of this problem proposed by erotetic logicians. Q2 |= Q1 if we only know that Γ |= Q1 as well as Γ |= Q2 . Secondly. for example. Since regularity was used only in the second part of the proof. Q1 Q2 . Regularity and second condition of e-implication give Γ ∪ β0 |= α and it follows Γ |= α that is in contradiction with Γ |= Q1 . the deﬁnition of strong e-implication given by Wi´niewski in [44]. Proof. Generally. MΓ ⊆ Mα . let us suppose that there is β0 ∈ dQ2 and Γ |= β0 . The ﬁrst condition requires MΓ ⊆ β∈dQ2 Mβ . Lemma 7 can be formulated not only in the version of pure regular e-implication. First. On the other hand. The deﬁnition is the same as that of e-implication. Questions and questions Of course. for some α ∈ dQ1 .15 Generally. we can prove that evocation carries over through a regular e-implication. but Γ ||= ∆ is added into the second clause. Due to the admissibility of weakening in declaratives (Fact 11) we can arrive at structures of e-implications with Γ containing (direct) answers to some of the two questions. thanks to regularity of e-implication.13 In contrast to the previous example. it need not be Γ. Lemma 7. s Fact 13 includes the original inspiration for the deﬁnition of strong e-implication. If Γ |= Q1 and Q1 Q2 . neither evocation nor See. If Γ |= Q1 and Γ. this brings us back to the role of an auxiliary set of declaratives in e-implication. resp. for each α ∈ dQ1 . It is valid because of the same semantic range of both questions. Then neither {ϕ}. it is superﬂuous to ask ?(α ∨ β). ?β |= ?α. But it is in contradiction with Γ |= Q1 . From Γ ∪ α ||= dQ2 .) 15 Let us take as an example (in CPL) the case {ϕ} |= ?α and {ϕ} |= ?β. Q1 |= Q2 or Γ. there is α0 ∈ dQ1 and M0 |= α0 . Knowing (α ∨ β). we do not see anything pathological in this example. ?α |= ?β nor {ϕ}.14 At the ﬁrst sight. Let us suppose that there is β ∈ dQ2 entailed by Γ. then there is a model M0 of Γ such that M0 |= β. Q1 Q2 gives soundness of an implied question Q2 relative to Γ. ?α. for each β ∈ dQ2 . we get an expected fact that Γ |= Q1 and Γ. Because of Γ ||= dQ1 . there must be some β0 ∈ dQ2 such that M0 |= β0 and that is a contradiction. then Γ |= Q2 . (Andrzej Wi´niewski called my attention to s this.

In our set-of-answers methodology (questions are deﬁned by sets of direct answers).3. we can control the cardinality of sets of direct answers by a mapping from one set to the other. This will be discussed in the next section. We say that (an answer) α gives an answer to a question Q iﬀ there is β ∈ dQ such that α |= β. Having two questions Q1 and Q2 we can deﬁne a relationship of ‘giving answers’: Deﬁnition 11.e. If Q1 = Q2 or Q1 ⊂ Q2 .16 • A question Q1 is included in a question Q2 (Q1 ⊂ Q2 ) iﬀ dQ1 ⊂ dQ2 . i.. The chosen set-of-answers methodology brings us to a natural approach. Let us start with relations among questions based on pure comparison of sets of direct answers. p. then Q1 ≥ Q2 and. Deﬁnition 10. Nevertheless. but we use the ﬁrst term for erotetically equivalent or semantically equivalent. we can expect that some clearing up of the structure of sets of direct answers could be helpful for the study of inferences. [44. Moreover. For this we use the symbol ≥ and write Q1 ≥ Q2 . moreover. moreover.3 Comparing questions: relations of questions based on direct answers So far we have introduced inferences that can provide certain relations between questions. • Two questions are equal (Q1 = Q2 ) iﬀ they have the same set of direct answers (dQ1 = dQ2 ). 2. Questions and questions e-implication says something new about structures of engaged questions. 16 34 . In this deﬁnition the ﬁrst question is considered as to be (semantically) ‘stronger’ than the second one. it would be useful to be able to compare questions with respect to their ‘answerhood power’. this term is redundant. (∀α ∈ dQ1 )(∃β ∈ dQ2 )(α ≡ β).2. The original deﬁnition refers to equivalent questions instead of equal (cf. Sets of direct answers can be purely compared or we can investigate their relationship based on entailment relation.3. 135]). each direct answer to Q1 not only gives an answer to Q2 but also is a (direct) answer to Q2 . A question Q1 gives a (direct) answer to Q2 iﬀ (∀α ∈ dQ1 )(∃β ∈ dQ2 )(α |= β). This approach could be extended in a semantic way.

Example 7 (in CPL). It follows that the set of safe questions is closed under weaker questions. First. for each β ∈ dQ2 . We have MPresQ1 ⊆ MPresQ2 . This fact is not too useful. 35 . Q1 |= Q2 and Q1 ≥ Q2 . However. Moreover. If Γ. Fact 17. ¬β} ≥ ?β Answerhood. Proof. we need to show that Γ ∪ β ||= dQ1 . then Q2 is safe. Fact 19. respectively. Q1 |= Q2 and Q2 ≥ Q1 . Recall what is required of the regular e-implication: (∀β ∈ dQ2 )(∃α ∈ dQ1 )(Γ ∪ β |= α). The next example shows that safeness of weaker questions is not transferred to stronger ones. Let us note an expected fact—stronger questions presuppose more than weaker ones.. If Q1 ≥ Q2 . and erotetic implication We can show some results of evocation and e-implication based on properties of the ≥-relation. Lemma 8. Theorem 6. The second condition of regular e-implication is obvious. evocation. then α∈dQ1 Mα ⊆ β∈dQ2 Mβ . But it is an easy conclusion from Γ. Fact 18. it follows from Q1 ≥ Q2 . It is better to notice the relationship among maximal presuppositions. i. Q1 |= Q2 because there is a subset ∆ ⊆ dQ1 for each β ∈ dQ2 such that Γ ∪ β ||= ∆. If Γ. both questions are erotetically equivalent relative to Γ. Each maximal presupposition of a stronger question entails a maximal presupposition of a weaker one. the class of evoked questions is not aﬀected by this problem. If Q1 is safe and Q1 ≥ Q2 . ?{β ∧ α. then Γ. Q2 regularly implies Q1 (both with respect to Γ). a prospective presuposition of a stronger question entails a prospective presupposition of a weaker question.3. Q2 Q1 . then Γ. Q1 Q2 . then whenever Q1 implies Q2 . Questions and questions The ordering based on the relation ≥ has a slightly non-intuitive consequence: a completely contradictory question is the strongest one. If the relation ≥ is turned. then PresQ2 ⊆ PresQ1 . The ﬁrst one is an obvious fact that an implied stronger question is regularly implied. Then the lemma follows.2.e. The semantic range of a stronger question is included in the semantic range of a weaker question. If Q1 ≥ Q2 . Q1 gives an answer to Q2 .

this repeated connection of ≥ and regular e-implication is not an accident. i. Notice that Q2 ⊂ Q1 will not help us either. Q2 Q1 . but neither ?{(α ∧ ϕ). To go back to evocation. it is included in the erotetic equivalence. Second. (β ∧ ψ)}.. then Γ. We are not going to introduce a special name for this relationship. then Q2 gives an answer to Q1 . as it was stated before.2. then every weaker question regularly implies this stronger one with respect to Γ. An evoked stronger question only implies the soundness of weaker questions relative to Γ. Proof. Q2 ≥ Q1 . it need not be that either Q1 implies Q2 or Q2 implies Q1 (with respect to Γ). First. (β ∧ ψ)} |= ?{α. If Γ |= Q1 and Q1 ≥ Q2 . This reminds us of the non-monotonic behavior of evocation. it is clear that two equal questions are both evoked by a set of declaratives if one of them is evoked by this set. • Q1 17 Q2 as well as Q2 Q1 . 36 . Questions and questions Now.17 Fact 20. The deﬁnition of regular e-implication says that if Q1 Q2 . there could be a direct answer to Q2 which is entailed by Γ. Γ ∪ β ||= dQ1 is required for each β ∈ dQ2 . Theorem 7. Let us illustrate it in the case that the ﬁrst question is included in the second one (Q1 ⊂ Q2 ). β} |= ?{(α ∧ ϕ). We get Γ ||= dQ1 from Γ |= Q1 . (β ∧ ψ)} ≥ ?{α. ‘giving an answer’ does not produce e-implication. However. To digress for a moment. We know that. see the next example. Let Q1 ≥ Q2 and Q2 ≥ Q1 . Example 8 (in CPL). β}.3. Generally. we are going to study the inﬂuence of ‘giving answers’ on the relationship of evocation and e-implication. ?{(α ∧ ϕ). it is not suﬃcient to know Γ |= Q1 and Q1 ≥ Q2 to conclude Γ |= Q2 . In the connection with the relation ≥ we have to require a version of an equality. generally.e. if Γ evokes Q1 and Q2 . from Q1 ≥ Q2 we have (∀α ∈ dQ1 )(∃β ∈ dQ2 )(α |= β) and it gives the second condition of regular e-implication (∀α ∈ dQ1 )(∃β ∈ dQ2 )(Γ∪ α |= β). Then • Γ |= Q1 iﬀ Γ |= Q2 . β} nor ?{α. If a stronger question is evoked by Γ.

Let us add expected results gained from equipollency. both sets of direct answers have the same cardinality (|dQ1 | = |dQ2 |). Partial answerhood We declared that the study of various types of answers (generally speaking. additionaly.e. α ≡ i(α). Fact 21. The other deﬁnition corresponds to a both-way relationship of ‘being stronger’. we can utilize the idea evoked by the second clause of Deﬁnition 8. We have used the term stronger in a bit informal way for questions that ‘give an answer’ to weaker ones.2. However. In particular. two equal questions are equipollent. The book [44] introduces two relations originated from Tadeusz Kubi´ski that prevent this uncontrolled cardinality. i. Examples 7 and 8 are valid for -relation as well. In this case. we cannot provide any special improvement of previous results for -relation. Narrowing down the set of direct answers of an implying question seems to be a good base for the deﬁnition of partial answer. then Q1 ≥ Q2 . It is clear that if Q1 Q2 . 37 . |dQ1 | ≥ |dQ2 |. n Deﬁnition 12. we know that each direct answer of a weaker question is given by some direct answer to a stronger question. The number of direct answers of the weaker question Q2 does not exceed the cardinality of dQ1 . • Q1 Q2 as well as Q2 Q1 . But. From the surjection. then • both Q1 Q2 and Q2 Q1 . Of course. unfortunately. Deﬁnition 13. A question Q1 is stronger then Q2 (Q1 Q2 ) iﬀ there is a surjection j : dQ1 → dQ2 such that for each α ∈ dQ1 . answerhood) is not the central point of this chapter.. • Γ |= Q1 iﬀ Γ |= Q2 . α |= j(α). A question Q1 is equipollent to a question Q2 (Q1 ≡ Q2 ) iﬀ there is a bijection i : dQ1 → dQ2 such that for each α ∈ dQ1 . If Q1 ≡ Q2 .3. Questions and questions Controlling the cardinality of sets of direct answers The set of direct answers of a weaker question can be much larger than that of a stronger question.

then ψ gives a partial answer. If Γ |= Q1 and each α ∈ dQ1 gives a partial answer to Q2 . For this. β} is posed. ¬β} or {¬α.19 The soundness condition in the other direction (from a set to initial question) will be expressed. let us introduce a choice function ξ such that ξ(Qi ) chooses exactly one direct answer from dQi . . β} we expect that if an answer to Q is true. This can be caused. Thus. β}. Whenever ψ gives a (direct) answer. Whenever we keep a model of the choice of direct answers from each question in Φ. However. we can get the following pairs of answers: {α. Generally speaking. Posing the question Q = ?{α. {¬α. From both machines. independently. i.2. by (∀AΦ )(AΦ ||= dQ). For each model of a direct answer to Q there must be a direct answer in each Qi valid in this model. 38 . then there must be a direct answer to Q true in this model. (∀α ∈ dQ)(∀Qi ∈ Φ)(α ||= dQi ) Possible states (of the world) given by answers to questions in the set Φ must be in a similar relation to the initial question. For each set of questions Φ and a choice function ξ there is a choice set AΦ = ξ {ξ(Qi ) Qi ∈ Φ}. we will write only Aξ . . ¬β}. assume that there exist two devices such that: the ﬁrst one can be asked by the question ?α.. . Q2 . β}. e. The question ?{α. ξ ξ 18 19 Compare this deﬁnition with Deﬁnition 4. let us suppose that there are a question Q and a set of questions Φ = {Q1 . Questions and questions Deﬁnition 14. is not able to accept it.3.4 Questions and sets of questions Working in the classsical logic. Q2 |= Q1 . generally. a device cannot ‘understand’ this question. then Γ.18 This deﬁnition allows us to cover many terms from the concept of the answerhood. to which we are going to address this question. let us imagine we would like to know whether it is the case that α or it is the case that β.}.e. by a restricted language-acceptability. Every direct answer gives a partial answer. But there could be a problem when an entity. ?β}. A declarative ϕ gives a partial answer to a question Q iﬀ there is a non-empty proper subset ∆ ⊂ dQ such that ϕ ||= ∆.g.. As a useful conclusion we obtain a weaker version of Theorem 7: Fact 22. {α. If the set Φ is clear from the context.10 in [44]. 2. and the other one is able to work with the question ?β.3. we obtain a soundness transmission from an initial question to a set of questions. then there must be a true answer to each question from the set {?α.

3. But this solution seems to be rather awkward. Whenever Γ and AΦ describe the state of the world. mutual soundness of this question and the set of questions {?α. We require to obtain at least one answer to an initial question from a choice set. If we admit the additional answer (¬α ∧ ¬β) and a question in the form ?{α. β} Aξ3 = {α. (∀AΦ )(∃α ∈ dQ)(AΦ |= α) ξ ξ 39 . Deﬁnition 15. ¬β} Aξ4 = {¬α. it is in contradiction with our (prospective) presupposition (α ∨ β). we introduce pure reducibility that does not use any auxiliary set of declaratives. (∀α ∈ dQ)(∀Qi ∈ Φ)(α ||= dQi ) 2. Now. β. Questions and questions Back to our example. (¬α ∧ ¬β)}. ξ ξ It will be reasonable to keep this strengthening. A question Q is purely reducible to a non-empty set of questions Φ iﬀ 1. there must be a direct answer to a question ξ Q that does the same job. A questioner posing the question ?{α. there are four choice sets: Aξ1 = {α. ξ ξ Our example produces more than soundness of Q relative to each AΦ ξ (with respect to Γ). ¬β} But the fourth one is not in compliance with the second soundness requirement. also eﬃcacy of each AΦ with respect to a question Q is ξ valid: (∀AΦ )(∃α ∈ dQ)(Γ ∪ AΦ |= α). Reducibility of questions to sets of questions We can take advantage of the previous discussion for the direct deﬁnition of reducibility of a question to a set of questions.2. This will bring us to the deﬁnition of reducibility with respect to an auxiliary set of declaratives and the mutual soundness will be required in the following forms: (∀α ∈ dQ)(∀Qi ∈ Φ)(Γ ∪ α ||= dQi ) and (∀AΦ )(Γ ∪ AΦ ||= dQ). β} Aξ2 = {¬α. ?β} will be valid. β} evidently presupposes (α ∨β).

there is AΦ made from these βs and M0 |= AΦ . let us suppose Q Φ and that each Qi ∈ Φ is safe. Proof. for each α ∈ dQ. ?β}. The deﬁnition of pure reducibility was introduced by Andrzej Wi´niewski in [43]. then Q must be risky too (cf. ?β} {?α. The relation of pure reducibility is reﬂexive (Q {Q}) and we can prove 20 the following version of transitivity: Theorem 9. (¬α ∧ ¬β)} {?α. [44. there is the ‘pure’ version from the introductory discussion. If Q is reducible to a set Φ. The ﬁrst condition of Deﬁnition 15 can be rewritten as α∈dQ Mα ⊆ β β∈dQi M . All items display reducibilities between initial safe questions and sets of safe questions. then Q Φi . The rewritten ﬁrst condition of Deﬁnition 15 is of the form Mα ⊆ α∈dQ i β∈dQi Mβ and it brings out the relationship of semantic ranges. It implies the existence of model M0 ∈ M such that M0 |= α. for each Qi ∈ Φ. β. 197]). the second one adds eﬃciacy. → • ?{α. i 20 Presented Theorem 9 corresponds to Corollary 7. as it was discussed. For the proof of the other implication. If Q Φ and each Qi ∈ Φ is reducible to some set of questions Φi . The safeness of all Qi gives (∀Qi ∈ Φ)(∃β ∈ dQi )(M0 |= β). we will write Q Φ. ?β} In the ﬁrst item. If Q Φ. then Q is safe iﬀ each Qi ∈ Φ is safe. p. s Example 9 (in CPL). Theorem 8. From this we know that if there is a risky question among questions in Φ and Q Φ. but Q is not.3. which gives the existence of some α ∈ dQ such that M0 |= α. where ◦ is any of the connectives: ∧. (∀Qi ∈ Φ)(|dQi | ≤ |dQ|) First two conditions express mutual soundness. β| • ?(α ◦ β) {?α. Questions and questions 3. and it gives the implication from left to right. But ξ ξ it is in contradiction with the second condition of the deﬁnition of Q Φ. and the last one requires relative simplicity. Thus.6 in [44].2. ∨. • ?|α. 40 . The following theorem shows that it is not an accident. The semantic range of a reduced question is bounded by the intersection of all semantic ranges of Qi s.

for each A∪i Φi . ?(α ∧ β) {?α.2. β|. but ?(α ∧ β) is not implied neither by ?α nor by ?β and ?(α ∧ β) does not imply neither ?α nor ?β. Proof. On the other hand. From Q Φ we ξ Φ know that there is α ∈ dQ for each choice set Aξ on Φ. M ξ ⊆ Mα . Qi . for each AΦ . for each Qi ∈ Φ. but ?|α. It is known that ξ AΦ ξ ⊆ Mβ . From the reducibility of all Qi in Φ to an appropriate Φi ξ we have that each AΦi is a subset of some A∪i Φi .3. Qi Q. Not even reducibility produces e-implication. even. β| |= ?(α ∧ β). for each Qj ∈ i Φi . we can prove that regular e-implication implies reducibility. It gives together α∈dQ Mα ⊆ γ∈dQj Mγ . Thus. For the second one we require the existence of α ∈ dQ such that A∪i Φi |= ξ α. for each Qi ∈ Φ. If Q M Qi . From Q Φ we get α∈dQ Mα ⊆ β β∈dQi M . The following example shows that it need not be that e-implication causes reducubility. for each Qi ∈ Φ. What about if we know Qi |= Q or. Example 10 (in CPL). and from the reducibility of all Qi in Φ to an appropriate Φi we have β∈dQi Mβ ⊆ γ∈dQj Mγ . then for each β ∈ dQi there is ξ AΦ β α α ∈ dQ such that M ⊆ M . Now let us look at the relationship of pure reducibility and pure e-implication. ?β}. Example 11 (in CPL). This choice set is made by elements of all dQi ∈ Φ which are valid in M. Questions and questions Proof. The third condition of Deﬁnition 15 is clearly valid. Both deﬁnitions have the same ﬁrst conditions. ?|α. It means that AΦ is ξ valid in M as well as α ∈ dQ. The ﬁrst one is easy to prove. for each dQj ∈ Φi . In the next subsection we will study some special cases of links between reducibility and e-implication. 41 . Lemma 9. β| {?(α ∧ β)}. It implies that if there is ξ ξ ∪i Φi any model M of Aξ . and (∀Qi ∈ Φ)(|dQi | ≤ |dQ|). can we conclude that Q Φ? Example 10 gives the negative answer to this question as well as ?(α ∧ β) ?|α. it must be a model of some AΦi . Let Φ be a set of questions such that Q If (∀Qi ∈ Φ)(|dQi | ≤ |dQ|). Let us prove the second condition of Deﬁnition 15 that requires exΦ istence of α ∈ dQ such that MAξ ⊆ Mα . for each β ∈ AΦ . but the second condition of reducibility can fail. then Q Φ.

ξ ξ The regular e-implication provides that there is α ∈ dQ for each β ∈ dQi such that MΓ∪β ⊆ Mα . Can we conclude that Γ evokes such a complex question which is reducible to the set Φ? Generally. the answer is positive. We will write Γ. But if we know that the complex question gives an answer to some question from Φ. Proof. [20]). Theorem 10. Q Φ. (α ∨ β). not. for each Qi ∈ Φ. From the construction of choice sets we know that for each M Φ AΦ and Qi ∈ Φ there is β ∈ dQi (member of AΦ ) such that MΓ∪Aξ ⊆ MΓ∪β . Q then Γ. We close this subsection by reversing the ‘direction’ of the reducibility relation. Q Φ. ?β} As it is expected. The second one requires that for each AΦ there is α ∈ dQ such that ξ Γ∪AΦ ξ ⊆ Mα .3. Theorem 11. ?{α. Qi . (∀α ∈ dQ)(∀Qi ∈ Φ)(Γ ∪ α ||= dQi ) 2. If Γ evokes each question from a set Φ. Q Φ. for each Γ. β} {?α. Let us suppose that we have generated a set of questions Φ that are evoked by a set of declaratives Γ. Q a question Qi ∈ Φ such that Q ≥ Qi . The third and the ﬁrst conditions of Deﬁnition 16 are obvious. but the mutual soundness and eﬃcacy conditions are supplemented by an auxiliary set of declaratives Γ (cf. then Γ |= Q.2. A question Q is reducible to a non-empty set of questions Φ with respect to a set of declaratives Γ iﬀ 1. It could be useful to introduce the general term of reducibility with respect to a context given by a set of declaratives. then Γ. 42 Φ. If Q Φ. Questions and questions So far we have worked only with the pure reducibility. and (∀Qi ∈ Φ)(|dQi | ≤ |dQ|). So we can speak of weakening in declaratives and it enables us to generalize Lemma 9. and there is . (∀Qi ∈ Φ)(|dQi | ≤ |dQ|) The introductory discussion is displayed in this example: Example 12 (in CPL). If Γ. The deﬁnition is almost the same as Deﬁnition 15. Deﬁnition 16. (∀AΦ )(∃α ∈ dQ)(Γ ∪ AΦ |= α) ξ ξ 3. the role of Γ is similar to the role of an auxiliary set of declaratives in e-implication: Fact 23.

at worse.3. . The introductory discussion and its formalization in Example 12 evoke interesting questions: • If we have an initial question Q = ?{α1 . We will require Q to be sound with respect to Γ. So. 43 . Reducibility and sets of yes-no questions The concept of reducibility is primarily devoted to a transformation of a question to a set of ‘less complex’ questions. for some α ∈ dQ. is it possible to reduce it to a set of yes-no questions based only on direct answers of Q? • Moreover. a countable list of direct answers. . we obtain: • Γ. Theorem 12. α2 . α2 . for each Qi ∈ Φ. we deﬁne the set of yes-no questions Φ based on the initial question Q = ?{α1 .} such that Φ = {?α1 . . Given the conditions of Theorem 11 are met. we get Γ ||= dQ.2. Together with reducibility. . . then there is a set of yes-no questions Φ such that Γ. could we require the e-implication relationship between Q and questions in the set Φ? We can ﬁnd an easy solution to these problems under condition that yes-no questions are safe and we have an appropriate set of declaratives. If a question Q = ?{α1 . for Q ≥ Qi . Q |= Qi .} is sound with respect to a set Γ. . Q |= Qi . Qi Q. The ﬁrst item is based on Fact 13 and the second one is given by the help of Theorem 7. Let us suppose that yes-no questions are safe in the background logic.} with. ?α2 . First. . From the evocation of each Qi ∈ Φ we have (∀M |= Γ)(∀Qi ∈ Φ)(∃β ∈ dQi )(M |= β). ξ ξ where it is stated that (∀AΦ )(∃α ∈ dQ)(AΦ |= α). α2 . Q Φ and Γ. Soundness of Q relative to Γ requires the existence of an answer α ∈ dQ for each model M |= Γ. . ξ ξ Informativness of Q with respect to Γ is justiﬁed by ≥-relation for some question Qi ∈ Φ. .}. . . and • Γ. each model of Γ produces some choice set such that (∀M |= Γ)(∃AΦ )(M |= AΦ ). . Questions and questions Proof. for each Qi ∈ Φ. then it gives a contradiction with Γ |= Qi . Proof. If Γ |= α.

49–7.3. . . α2 . Simultaneously.21 In logics with risky yes-no questions.) Whenever we know that the initial question is evoked by a set of declaratives. we use the shape of questions in Φ. we can ask for a help the auxiliary 21 The same result is provided by theorems 7. but the results are the same. 1. 104–5] where a bit diﬀerent deﬁnition of reducibility is used. the ﬁrst condition of reducibility as well as e-implication can fail. Two ξ ξ cases will be distinguished. for each Qi ∈ Φ. This gives MΓ∪α ⊆ β∈dQi Mβ . If there is α from both AΦ and dQ. ξ 2. If β ∈ dQ. If β ∈ dQ and Γ ∪ β has at least one model. Φ This theorem enables us to work with classes of questions which are known to be sound relative to sets of their presuppositions. The safeness of members of Φ implies that Mα ⊆ β∈dQi Mβ . Fact 24. we recognize that MΓ∪β ⊆ MΓ . For this. then choose this direct answer. 2. then there is a set of yes-no questions Φ such that Γ. If dQ is ﬁnite or the entailment is compact. . Questions and questions Secondly.} is evoked by a set of declaratives Γ. the set Φ is ﬁnite set of yes-no questions. we prove the condition that is common for both reducibility and e-implication. for each α ∈ dQ and each ?β ∈ Φ.51 in [44]. We need to ﬁnd an α ∈ dQ for each AΦ such that Γ ∪ AΦ |= α. 44 . (Normal and regular questions are the obvious example. To prove reducibility we have to justify the second condition of Deﬁnition 16. we get MΓ∪β ⊆ α∈∆ Mα .α ⊆ (Mβ ∪ M¬β ). β must be of the form ¬αj and ∆ can be deﬁned as dQ\{αj }. Working in logics where yes-no questions are safe. we get the following conclusion. it is useful to emphasize that the proof of Theorem 12 shows how to construct such a set. If there is no direct answer α ∈ dQ in AΦ . for each α ∈ dQ and Qi ∈ Φ. if a question Q = ?{α1 . More generally. This fact corresponds to main lemma in the paper [20. We have to show that for each Qi and each direct answer β ∈ dQi there is a non-empty subset ∆ ⊂ dQ such that Γ ∪ β ||= ∆. It need not be MΓ. Simultaneously. which means MΓ ⊆ α∈dQ Mα . Together with soundness of initial question Q with respect to Γ. then MΓ∪Aξ = ∅ and we ξ can take any α from dQ. Q Φ and Γ. 1. pp. then ∆ could be {β} and Γ ∪ β ||= {β}.2. The ﬁnal step is the proof of e-implication. Q |= Qi .

If a question Q = ?{α1 . yes-no questions must be safe. which are in the role of direct answers. then Γ. and there is a set of yes-no questions Φ = {?α1 . Observing the last item of Example 9. α2 . Final remarks set of declaratives again. Going through the proof of Theorem 12. The ﬁrst restriction is clear. Then the answer is positive. questions are kept as special objects of the language LQ . . Let us remind Fact 13 and put soundness of each Qi ∈ Φ with respect to Γ. Q |= Qi . Thus. based on properties of the classical logic and erotetic implication (see [46. [50]). There is a similar concept in literature. β| as an interlink between ?(α ∧ β) and questions ?α and ?β. Q Φ and each Qi ∈ Φ is reducible to some set of questions i Φi . If Γ. Some scenarios work with the descending ‘complexity’ of questions from the root to leafs. they are not 45 .4 Final remarks IEL is introduced as a very general theory of erotetic inferences dealing with a general language L extended by questions. Following the IEL-philosophy and in accordance with our ﬁrst chapter we decided to introduce questions as sets of declaratives. called erotetic search scenarios. If we recall non-transitivity of e-implication (Example 5). for each Qi ∈ Φ. 48] and [39]). . 2. we can recognize the ‘truth-functional’ auxiliary role of the question ?|α. . . an initial question (and a context) is the root and branching is based on direct answers of auxiliary questions. the second clause of pure reducibility (Deﬁnition 15) requires the truth-functional connection of subformulas. The relationship between interrogative nodes is given by erotetic implication. Scenarios are trees. We can use repeatedly a cautious extension of Theorem 9: Fact 26. As a conclusion we get Fact 25.4. ?α2 . it seems worthwhile to enquire whether it is possible to follow this proces and to reduce a question (with respect to an auxiliary set of declaratives) to a set of atomic yes-no questions based on subformulas of the initial question (cf. The construction of yes-no questions provided by Theorem 12 does not prevent the high complexity of such yes-no questions. the rest is valid independently of safeness of yes-no questions. . Inferences with questions are based on multiple-conclusion entailment that makes it possible to work with sets of declaratives in premises as well as in conclusions. for each Qi ∈ Φ. then Γ. Q Φ and Γ. But it is not all. Q Φi . .2.} (based on Q) such that Γ ||= dQi .} is sound relative to a set Γ. The idea is that there is an initial question (and a context) and we can get an answer to it by searching through answers of some operative (auxiliary) questions. 47.

4. All relationships among questions are based on inferences and a comparison of their sets of direct answers.2. Our general view did not tend to provide any axiomatization. Final remarks combined by logical constants as declaratives are. The main modiﬁcation of the original IEL can bee seen in the chosen SAM and in the model-based approach. the term semantic range of a question made the semantic work easier and transparent. The discussed properties and relationships vary with the chosen background system. see [7] as a nice example. but the central point was whether we can formulate some relationships (meta-rules) for erotetic consequence relations in IEL. IEL opens many possibilities for working with questions in various logical systems. 46 . In this chapter we studied many properties and relationships. and serves as an inspiration for the work in erotetic logic. they depend on semantics. especially.

the working system will be the normal modal (epistemic) logic K with its standard relational semantics (Kripke frames and models). This brings us to an important point. The basic way to complete someone’s knowledge is the posing of questions. However. Epistemic logic in use will be the normal multi-modal logic. The approach we are going to present in this chapter works with epistemic logic as a background. 47 . on the one hand. Such approaches made it possible to incorporate questions within some formal systems and not to lose their speciﬁc position in inferences. Roughly speaking. At the very beginning we do not impose any conditions on it. Let us recall Wi´niewski’s inferential erotetic logic s and the intensional approach of Groenendijk and Stokhof. Although questions diﬀer from declaratives.2 we introduced a formalization of questions called set-of-answers methodology. thus. Simultaneously we mentioned there that it is very natural to use the epistemic terms in speaking of questions. on the other hand. they are used in formal systems that are the framework of reasoning. the way of incorporating questions will be of such generality that it can be applied in all epistemic-like logics.1 Introduction Communication is essentially connected with an exchange of information.Chapter 3 Epistemic logic with questions 3. they play a similar role in reasoning. epistemic logic and its semantics are used for the modeling of knowledge and epistemic possibilities of agents and groups of them. It utilizes the close connection of questions and their answerhood conditions. In the recent history of logic of questions we can see a success in ﬁnding the desirable position of the formal approach to questions and in a study of inferences based on them. In Section 1. questions are speciﬁc entities and they bear some special properties.

considers the answers to be possible. 1 48 . does not know what is the right answer to the question..g. 3. i. the question has a two element set of direct answers: • Ann has the Joker.1. the process of communication is a dynamic matter and epistemic logic is open to enrichment by dynamic aspects. or Bill? From our set-of-answers methodological viewpoint. Then she can ask Who has the Joker: Ann. questions may provide new issues” [12]. In the introductory chapter we use the example with three friends holding cards. The main aim of communication in a group is to share data and to solve problems. e. These aspects will be studied later on in chapter 4 where we accept the ‘standard’ model of knowledge based on the modal system S5.3. Asking it. An asked question means that listeners can form their partial picture of questioner’s knowledge structure. it must be the case that just either Ann has the Joker or Bill has it. presupposes what is implicitly included in the answers.1 As it was mentioned at the beginning.e. the exchange of information is a basis of communication. Introduction Moreover. First. An agent-questioner provides the information of her ignorance (item 1) as well as the expected way to complete her knowledge: 2 says that there are two possibilities and 3 that one of the possibilities is expected. A question is a store of information of agent’s epistemic state. which is an important part in communication and solving of problems in groups. Very typical example is a group of scientists trying to ﬁnd an answer to their scientiﬁc problem. see. Communication will be studied in the next chapter. Now let us brieﬂy mention what a question and its epistemic meaning are. • Bill has the Joker. Catherine expresses that she 1. moreover. Let us suppose Catherine wants to ﬁnd out where the Joker-card is. here we prepare an ‘erotetic epistemic framework’. and.. [42]. 2. In this situation we can recognize the question as ‘reasonable’. Only by sharing of their knowledge and ignorance they can reach a solution. we introduce propositional single-agent Jeroen Groenendijk says that “assertions may provide new data.

(M. This section deals with a single-agent variant for the sake of simplicity in introducing questions and their basic properties. etc. possible worlds) S and an accessibility relation Ri ⊆ S 2 . Now. . The language of classical propositional logic Lcpl is extended by modalities [i] and i . . s) |= p iﬀ (M. Thus. ’. . s) |= ¬ϕ iﬀ (M. m}. . the role of epistemic context as well as sets of questions are studied. . . v where v is a valuation of atomic formulas. .3. Moreover. Ri with a set of states (points. . s) |= ψ1 or (M. . s) |= ϕ 3. s) |= ψ1 ∨ ψ2 iﬀ (M. s) ∈ v(p) 2. And. we get a language LK with cpl a subset of signs for atomic formulas P = {p. . ‘agent i believes that. Semantics is based on Kripke-style models. Kripke model M is a pair F. q. (M. roughly and vaguely said. . Questions will be a natural part of inference relations based on the background logic. We fully apply our set-of-answers methodology and allow to mix declaratives and interrogatives. ’. The satisfaction relation |= is deﬁned by a standard way: 1. Then we discuss answerhood conditions in a relationship with conditions posed on a ‘reasonable’ question. The ﬁrst one can be interpreted as ‘agent i knows that. Single-agent propositional epistemic logic and questions (normal modal) epistemic logic and extend it by questions. . Kripke frame is a relational structure F = S. it is just ‘epistemic necessity’ of an agent without restrictions to knowledge conditions or belief conditions. The other one is an ‘epistemic possibility’. (M. s) |= ψ2 49 . ﬁnally. 3.} and formulas deﬁned as follows: ϕ ::= p | ¬ψ | ψ1 ∨ ψ2 | ψ1 ∧ ψ2 | ψ1 → ψ2 | ψ1 ↔ ψ2 | [i]ψ | i ψ Modality i is understood as a dual to [i]: i ϕ ≡ ¬[i]¬ϕ In multi-agent variants of epistemic logic we presuppose that there is a ﬁnite set of agents A = {1. indices. m are names for agents.2 Single-agent propositional epistemic logic and questions Our approach to epistemic logic is very liberal.2. . where numbers 1. we do not restrict the interpretation of [i] to ‘knowledge’ of an agent i. . .

.3. In accordance with our set-of-answers methodology the intended reading of a question Qi is: Is it the case that α1 or is it the case that α2 . . I know that a questioner presupposes the same.. (M. For interrogative cpl i i formulas metavariables Q . this seems to be useful in a communication processes. s) |= ψ1 and (M. s) |= ψ2 5. no matter which answer is right. s) |= ψ1 implies (M. a question Qi is any formula of the form ?i {α1 . On the contrary to our liberal SAM we require dQi to be ﬁnite. . This brings us to an important term presupposition. } and the question mark cpl ?i for a question of an agent i. thus.2. we will omit it. or is it the case that αn ? Whenever I ask such question. . 3. I presuppose that at least one of the direct answers is the case. We suppose that dQi is ﬁnite with at least two syntactically distinct elements. which is studied in the next subsection. it makes easier the concept of ‘presupposing’. Direct answers are formulas of our extended epistemic language LKQ and quescpl tions can be among direct answers as well. Working in propositional logics we want to keep direct answers as clear epistemic possibilities. etc. Whenever I hear such question. . A presupposition is a ‘consequence’ of each direct answer. (M.1 Incorporating questions We extend epistemic language LK by brackets {.2) we deﬁne presuppositions of questions as formulas that are implied by each direct answer. . 2 50 . Generally. s) |= ψ1 ∧ ψ2 iﬀ (M. . We get the language LKQ . s) |= ψ1 → ψ2 iﬀ (M.2. Q1 . will be used. . we have semantics for the modal system K.2. Single-agent propositional epistemic logic and questions 4. αn }. s1 ) |= ϕ. αn } is the set of direct answers to a question Q. (M. Presuppositions Taking an inspiration in inferential erotetic logic (see Section 2. where dQi = {α1 . for each s1 such that sRi s1 We do not put any restrictions on accessibility relation. . Simultaneously. i. s) |= [i]ϕ iﬀ (M. s) |= ψ2 6. at least one of the direct answers is the case.2 Let us make a symbol convention: if it is not necessary to use the index i. .e.

∨ αn ) is a prospective presupposition of a question ?{α1 . Example 13. . The set of presuppositions of a question is full of redundant formulas. s). so is prospective presuppositions. for all models M and states s. Corollary 4. The theory of questions in IEL introduces one more term—prospective presupposition. which is a contradiction. Deﬁnition 18. This can be a modal reformulation of the original IEL deﬁnition. In our (ﬁnite) case we can suppose that Q has at least one prospective presupposition. . . let ϕ be maximal. but it is in contradiction with the fact that no α is valid in (M. . Maximal presuppositions imply every presupposition. Proof. s) |= ϕ implies the existence of a direct answer α satisﬁed in (M. prospective presuppositions are maximal. then there is a state (M. A formula ϕ is a prospective presupposition of a question Q iﬀ ϕ ∈ PresQ and. . The truth of a prospective presupposition at a state of a model gives the truth of some direct answer at this state. Second. . s) |= ϕ. for each α ∈ dQ. The set of prospective presuppositions is equal to the set of maximal presuppositions of a question Q. . s) |= α. . All presuppositions are satisﬁed in the state (M. s) |= ϕ and (M. .3. Deﬁnition 19. but not vice versa. Since ϕ is not maximal. s) |= ψ. If ϕ is not prospective. s) |= α. let ϕ ∈ PPresQ but ϕ is not maximal. s) |= ψ. 3 51 . s) |= ϕ and (M. things are easier. but not prospective. We write ϕ ∈ PPresQ.3 Theorem 13. s). . ∨ αn ) is a maximal presupposition of a question ?{α1 . there must be (M. A formula ϕ is a presupposition of a question Q iﬀ (α → ϕ) is valid for each α ∈ dQ. (M. Because of working with ﬁnite sets of direct answers in a system extending classical propositional logic. This leads to the deﬁnition of maximal presuppositions.2. First. . see [44. . s) and it gives (M. In IEL. A formula (α1 ∨ . s) and ψ ∈ PresQ such that (M. if (M. We write ϕ ∈ PresQ. s). A formula (α1 ∨ . then there is a direct answer α ∈ dQ such that (M. Each presupposition is implied by every direct answer. αn }. We need not distinguish between maximal and prospective presuppositions. A formula ϕ is a maximal presupposition iﬀ ϕ ∈ PresQ and (ϕ → ψ) is valid for each ψ ∈ PresQ. Single-agent propositional epistemic logic and questions Deﬁnition 17. αn } as well. so is ψ in (M.10].

2.e. or it must be the case that Bill has the Joker (and Ann not). Bill. then ϕ ≡ ψ. 48) we wrote that either Ann or Bill has the Joker is Catherine’s presupposition. and Catherine. The other case is similar. Catherine’s presupposition is under inﬂuence of the context given by the rules of the card ‘game’: Just one Joker is distributed among the agents Ann. it must be the case that Ann has the Joker (and Bill not). Note on presupposing and context In the example with card players we mentioned ‘reasonable’ Catherine’s question Who has the Joker: Ann. Thus. viz.3. especially. or Bill? In item 3 (see p. s) |= (α → ψ) gives (M.2. We have obtained ϕ |= ψ. Now. ψ ∈ PPresQ. If (M. it is expected the form ?{α. s) |= ψ. Since ψ ∈ PresQ. The role of context will be studied later on. then (M. If ϕ. For proving semantic equivalence we have to prove ϕ |= ψ as well as ψ |= ϕ. This is indicated by comma in the interrogative sentence as well. Proof. the question bears a presupposition that Peter is at least one of the two possibilities (maybe. Single-agent propositional epistemic logic and questions In the next theorem we show the same result we obtained for IEL (see Lemma 1): All prospective presuppositions of a question are equivalent. Theorem 14. s) |= ϕ. the formalization of both questions would be almost the same. the symbol PPresQ will be used for a formula representing prospective presuppositions of a question Q modulo the semantic equivalence. 52 . let us concern the following question: What is Peter: a lawyer or en economist? If there is no supplementary context. both of them). subsection Relativized askability in 3.. β}. i. However.2. s) |= α. then there is α ∈ dQ such that (M.

for each α ∈ dQi 2. but. the freedom in the syntactical form of questions was compensated by restrictions in their semantics. Non-triviality is known. ‘Reasonability’ corresponds to the three conditions we informally mentioned at the introductory example. Let us repeat and name them: 1. she does not know any (direct) answer to a question. s) |= [i]PPresQi Then we say that Qi is askable in the state (M. (M. . for each α ∈ dQi 3. s) (by an agent i). It holds for a question Qi = ?i {α1 . Whenever an agent-questioner poses a question.2. Admissibility 3. she considers all (direct) answers possible and she is aware of what is presupposed—she knows the prospective presupposition of a question. The formal deﬁnition follows. Askable questions include neither contradiction nor tautology among their direct answers. . In our version 53 . Deﬁnition 20. It is clear that it makes little sense to speak about truth/falsity of a question. . Context It is not reasonable to ask a question if the answer Each direct answer is considered as possible. simultaneously. As we can see. s) |= Qi iﬀ 1. A question Qi is askable relative to a model M by an agent i (let us write M |= Qi ) iﬀ (M. s) |= [i]α. s) |= Qi for each s ∈ S. At least one of the direct answers must be the right one. but there are no ‘tautological’ questions in K. We say that a question is (generally) askable iﬀ there is a model and a state where the question is askable (by an agent). We introduce instead a concept of askability of a question. A question is not askable in a state without successors. The former is excluded by the second condition and the latter by the ﬁrst one. . αn } that (M. 2. Askability is based on our idea of a ‘reasonable’ question in a certain situation. Single-agent propositional epistemic logic and questions Askable questions In semantics for a majority of logical systems we speak about truth or falsity of a formula (in a particular state of a particular model). s) |= i α. (M. (M.3. The deﬁnition of |= Qi is straightforward.

s) |= i ¬α. questions are complex modal formulas. (¬α ∧ ¬β)} States in the afterset sRi are understood as epistemic possibilities. Single-agent propositional epistemic logic and questions at least two successors are needed. for each α ∈ dQi . admissibility condition requires that there must be at least one ‘α-state’ and at least one ‘β-state’ in sRi . because of the context condition. by the question ?i {α.e. . Deﬁnition 20 is meant in a full generality without the intention of reduction of questions to the epistemic language. i. the complete afterset structure can contain other states. β}. β} does not consider the answer neither α nor β as possible (context condition)—such answer would be accepted. However.3. All states in sRi must satisfy the prospective presupposition (α ∨ β). sRi = {s : sRi s }. some of them may satisfy both α and β.. β} s s2 β. this is a minimal requirement given by askability conditions. (α ∨ β) Of course. We can see the questioner as admitting the possibility of ¬α for each direct answer α to a question Qi . one of them satisﬁes α (and does not satisfy β) and the other one β (and does not satisfy α). but none of them satisﬁes ¬α and ¬β: the question ?i {α. β. The epistemic semantic viewpoint represents agent’s ‘knowledge’ in a state s as an afterset sRi given by the states related to s by an accessibility relation Ri . In these systems. 54 . s s . Its askability in a state (M.g. s) requires a substructure on sRi consisting of (at least) two accessible states.2. (M. ··· ··· ··· Let us return to the following question: What is Peter: a lawyer or an economist? This question can be formalized by a formula ?i {α. the ﬁrst condition is equal to (M.. i. s) |= ¬[i]α. s1 α. e. (α ∨ β) ?i {α. If we work in systems extending classical logic. Finally.e. In accordance with the non-triviality condition neither α nor β can be true in all of them.. .

(¬α ∧ β).2.2 Some important classes of questions In this section we introduce some classes of questions with their semantic behavior. In the included subsection we suggest to link conditional and hypothetical questions with the role of a context. [44]). The same requirements are posed by askability conditions for ?i ¬α. an auxiliary set of formulas. yes-no questions can be seen as a ‘contingency modality’. (α ∧ ¬β). β| and the following ﬁgure shows the required substructure on the afterset. ¬α}. In our case. The shortest one is ?i {(α ∧ β). Yes-no questions and conjunctive questions were introduced in the previous chapters. Single-agent propositional epistemic logic and questions 3. 4 55 . which is shortly written as ?i α. In our system. Questions that are not safe.2. A question Qi is safe iﬀ PPresQi is valid.e. The original concepts of safety and riskiness of questions come from Nuel Belnap.4 Another example of safe questions are conjunctive questions. The names of classes originate from IEL (cf. will be called risky. Questions with presuppositions.3. See chapter 2 as well. Yes-no questions always form a partitioning on aftersets and their presuppositions are tautologies. This question is askable in a state s if there are (at least) two diﬀerent states available from s. both ?i α and ?i ¬α are equivalent. Is Prague the capital of the Czech Republic? is a question requiring one of the following answers: • (Yes. are called safe. one satisﬁes α and the other one ¬α. The afterset sRi is supposed to have this form: s1 α ?i α s s2 ¬α Reviewing the askability conditions for ?i α we can see that this question is equivalent to a formula i α ∧ i ¬α..) Prague is the capital of the Czech Republic. We write it as ?i |α. Deﬁnition 21.) Prague is not the capital of the Czech Republic. (¬α ∧ ¬β)} asking for a full description based on α and β. which are all tautological. The very basic questions in their syntactical as well as semantical form are yes-no questions. i. with the formalization ?i {α. • (No.

. I smoked and go on. but seeing both answers it seems. β ?i |α. I did • No. In section 3. α ∧ βn } 56 . ¬β s2 α. However. Let us deﬁne this kind of local safeness: Deﬁnition 22. β} is a risky one. there can be something more what is presupposed: • Yes. β| s −→ ↓ s3 ¬α. Relativized askability Let us consider the following question: Did you stop smoking? At ﬁrst sight. A question Q is safe in a state (M. β Similarly to yes-no questions they have the exhaustive set of direct answers and direct answers are mutually exclusive. The question What is Peter: a lawyer or an economist? with a formalization ?i {α.4 we will deal with some restrictions posed on direct answers. s1 ) |= PPresQi . . I didn’t can mean can mean I had smoked and stopped. s) (for an agent i) iﬀ (M. ¬β s4 ¬α. . . it is a yes-no question. for each s1 ∈ sRi . Both of them presuppose the smoking in the past. conditional questions are of the form ?i {α ∧ β1 . Thus. Single-agent propositional epistemic logic and questions s1 α. an askable question in a state (for an agent) is safe in this state (for this agent). (α∧ ¬β)}.2. α ∧ β2 . β}) is not satisﬁed. Such question is an example of a conditional yes-no question with a formalization ?i {(α∧β). Generally. asking this question in a state an agent does not admit that she can see states where PPres(?i {α.3. which will enable us to clarify answerhood conditions.

α ∧ βn } is askable in s if and only if ?i {β1 . s) |= Qi . will you see a bedroom? with a formalization ?i {(α → β). etc. which is a bit similar to the previous one. α → βn } Again. s) |= (∆. Every question askable at a state is askable with respect to its set of (prospective) presuppositions. Then we write (M. . Q)i . . β})c In addition. A conditional question ?i {α ∧ β1 . . Q)i . 57 . This leads to relativized askability. for each ∆ ⊆ Γ. Q)i . for ∆ ⊃ Γ. . . ?{α. s) |= (Γ. s) |= (Γ. α ∧ β2 . If (M. A general hypothetical question is then ?i {α → β1 . . or Bill? can be formalized by ({¬α ∨ ¬β}. A question Q is askable (by an agent i) in (M. . the askability of such questions can be understood as based on agent’s hypothetical knowledge. s) |= [i]Γ and (M. relativized askability is not ‘monotonic’ in knowledge databases. . Our interpretation is: if α is known.3.. then it is to be decided whether β1 . then (M.2. hypothetical question. Relativized askability of this kind can be used for an explicit expressing of the knowledge structure. It has an expected consequence: Fact 27. IEL introduces one more term. . an agent ‘knows’ α in s. or β2 . .e. Using a generalization similar to Deﬁnition 23 we obtain 5 This term corresponds to question in an information set introduced in [13]. If (M. . Q)i . The notion of an askable conditional question can be generalized with respect to a set of formulas. Q)i . Single-agent propositional epistemic logic and questions The askability of a conditional question in a state s requires the validity of α in each accessible state. s) |= (∆. (α → ¬β)}. However. The term relativized askability will be mostly used for pointing out the importance of a set Γ. i. βn } is askable there with respect to the auxiliary set (knowledge database) {α}. Catherine’s question Who has the Joker: Ann. A natural language example of hypothetical yes-no question might be If you open the door. s) with respect to a set of formulas Γ iﬀ (M. By [i]Γ we abbreviate the set {[i]γ | γ ∈ Γ}. s) |= (Γ. then it need not be (M. .5 Deﬁnition 23.

?{α1 . for an agent i) whenever askability of Q1 (in s. for each j ∈ {1. s) |= (Γ.3. . If (M. then (M.5. Q)i . We will return to these concepts in Section 3. s) |= Qi 1 2 1 2 We have mentioned that questions ?i α and ?i ¬α have the same askability conditions. The informal meaning of epistemic erotetic implication is very transparent. . Both (?α → ?¬α) and (?¬α → ?α) are valid. s) |= [i]Γ implies (M.3 Epistemic erotetic implication Erotetic inference is implicitly based on the (standard) implication. Implied question’s required substructure on the afterset must be a substructure of that required by an implying one. (M. i 3. then she can ask every question implied by Q. . Example 14. n}. We say that a question Q1 implies Q2 (in a state s. Let us omit the index i for now. . s) |= Γ → Q. for i) implies askability of Q2 (in s. . The only diﬀerence of both deﬁnitions lies in the words and and implies. The question What is Peter: a lawyer or an economist? implies Is Peter a lawyer? as well as Is Peter an economist? This can be generalized: Example 15. . The eqivalence of both questions is a theorem in our system based on modal logic K. By [i]Γ we i abbreviate the set {[i]γ | γ ∈ Γ}. s) |= Γ → Q Askability of a conditional question ensures the askability of a hypothetical one. A question Q is askable (by an agent i) in (M. αn } → ?αj is valid. Let us write (M. . Whenever an agent asks Q. s) |= Qi . s) |= Qi → Qi iﬀ (M. Epistemic erotetic implication Deﬁnition 24.3. A question in antecedent is ‘more complex’ then the implied one. . Fact 28. s) with respect to a set of hypotheses Γ iﬀ (M. s) |= Qi implies (M. 58 . for i).

γ} requires [i](α∨β ∨γ). Q1 is safe and (Q1 → Q2 ) is valid. then if ϕ ∈ PresQ1 . We add a single point s1 accessible from s where the prospective presupposition of Q2 is invalid. We take a model M and a state (M. which can fail in the structure suﬃcient for the askability of the ﬁrst question. If Q1 → Q2 is valid. In both cases there is a problem with the context condition. Proof. . Example 17. ?|α. for dQ2 = {β1 . β| Let us notice that conjunctive questions are safe and they imply safe questions again. then (M. but ?i {α.3. s) where (M. (M. ¬β} as well as |= ?{¬α. Let us suppose. . If (M. (¬α ∧ ¬β)} 4. . β} more complex than ?|α. s) |= Q1 → Q2 and (M. 3. β|? It is easy to check that neither ?|α. β. In this model the askability of Q1 1 is not violated at s1 . β| are valid. |= ?{α. ?|α. where ◦ is any truth-functional constant. This leads to a contradiction. β} requires ‘only’ [i](α ∨ β). We can prove that it is a rule. but Q2 is not askable here. (M. An implying question shares presuppositions with the implied one. but Q2 is not safe. Epistemic erotetic implication The following example shows a special position of conjunctive questions in implications. Epistemic erotetic implication has the expected property—transitivity: Fact 30. ¬β} → ?{α. . s1 ) |= [i] l βj . s) |= Q2 → Q3 . If Q1 is safe and Q1 → Q2 valid. β} → ?{¬α. The next two examples emphasize the importance of context condition again. We have to be careful when speaking about ‘complexity’ of questions. β} requires the validity of the disjunction (α ∨ β) on the afterset. . Fact 29. Example 16. then Q2 is safe. β. β| → ?{α. β} The question ?i {α. . The following implications are valid: 1. . s) |= Q1 → Q3 . β} → ?|α. ?|α. s) |= Q2 . αn | → ?αj . . ?|α1 . then (M. s1 ) |= Q1 . γ| → ?|α. a risky question ?{α. . 2. β. . β| → ?{α. β} Example 18. n}. β. βl }. then ϕ ∈ PresQ2 . β| → ?(α ◦ β). |= ?{α. Let us slightly change the afterset of s. Is question ?{α. s) |= Q1 .3. γ} → ?{α. Theorem 15. β} nor ?{α. for each j ∈ {1. 59 . .

admissibility. .4 Askability and answerhood Epistemic erotetic implication creates a relationship among questions. . All possibilities required by the admissibility condition are not available.. If a question is askable at a state. . in fact. We write (M. s) |= αj ∈dQi ([i]αj ). The case of invalid admissibility condition is a bit diﬀerent. To break the non-triviality condition means that there is a direct answer which is ‘known’ by an agent (in a state of a model). which is In [13] we can ﬁnd two terms: to be and to give a semantic answer. s) (for an agent i) iﬀ (M. . there is a direct answer which is not considered as possible. The askability of a question consists of the validity of three conditions (non-triviality. s) |= αj ∈dQi ([i]¬αj ). . αn } is partially answered in (M. αn } is answered in (M. Askability and answerhood 3. Some answers to the question ?|α. 6 60 . β} are missing.3. An answered question is a ‘potentially’ answered question. so is every implied one. We write (M. . so is the implying one. properties of inaskable questions from various classes of questions. an agent knows a direct answer even if she knows a formula that is equivalent to a direct answer or she knows a formula from which some direct answer follows. β| give information that is superﬂuous in the state of agent’s knowledge. In our system. Deﬁnition 26. β| in that state. in particular. A question Qi = ?i {α1 . if the admissibility condition fails. and answerhood conditions—complete and partial answers will be introduced. 7 We do not discuss a dynamic approach just now. Let us imagine that we know that Q1 → Q2 and we have the answer to Q2 (non-triviality condition fails). Q1 → Q2 if and only if ¬Q2 → ¬Q1 . In fact. . β|. accessible states with {¬α. If an implied question is inaskable. ¬β} and {¬α. s) |= i α). even if she does not know an answer to a question ?|α. s) |= Pi Q. Then.e. . s) (for an agent i) iﬀ (M. (∃α ∈ dQi )((M. In fact. Such formula is called complete answer.6 Let us deﬁne the concept a question is answered at a state:7 Deﬁnition 25.4. Partial answer excludes some of the (direct) answers. and context) and inaskability is a result of the violation of at least one of them. A question Qi = ?i {α1 . but does it mean to have an answer to Q1 ? What if Q1 is inaskable because of failing context condition? In this section we will just deal with such violations of askability conditions. Let us imagine that our agent knows α in a state s. it is not right to ask this question. This distinction is not necessary here. i. then we are sure that Q1 is inaskable. Formula α is a partial answer to a question ?|α. s) |= Ai Q. an answer need not be uttered among agents.

For example. β} is answered in s (the agent knows β). then Ai Q → Pi Q is valid. Lemma 10. Yes-no questions as well as conjunctive questions are examples from this class. Their sets of direct answers satisfy a more strict condition. The last option for inaskability is the violation of context condition. generally. Does it mean that Q is partially answered? Surprisingly. Mutual exclusiveness is not preserved by implication. the answers of the question ?i {α ∧ β. which is important here: it does not admit the afterset substructure in Example 19. s): (∃α ∈ dQi )((M. s) |= [i]¬α) in our system. Let us suppose a question Q is answered. it is not true that if (M. Example 19. This and example 15 give the proof of the following lemma. However. β}. the question 61 s1 ¬α. but it is not partially answered. it can be caused by a model and a state. s s2 α. It means that for each direct answer there is another one such that both of them cannot be true. both conditions are of a semantic nature. but this question implies ?i {α.3. Askability and answerhood equivalent to (∃α ∈ dQi )((M. If (M. s) |= Pi Q. s) |= Ai ?α). If a question Qi has the set dQi with pairs of mutually exclusive direct answers. or Bill? with the context. β In the structure given by Example 19 the question ?{α. the agent is not able to get the knowledge of either ¬α or ¬β. Then an agent does not know (believe) the prospective presupposition of a question.4. in particular. α ∧ ¬β. See the next example. both α and β are still possible. ¬α ∧ β} are mutually exclusive. s) |= Ai ?ϕ and Qi → ?i ϕ is valid. which is not in this class. We cannot prove anything similar to Theorem 15. Ai Q implies Pi Q for questions with pairs of mutually exclusive direct answers. β . Recall Catherine’s question Who has the Joker: Ann. This means that our agent can answer the question ?α in (M. s) |= Pi Q. then there is a formula ϕ such that (M. Fact 31. s) |= Ai Q. they have mutually exclusive direct answers—the ‘truth’ of a direct answer (in a state) means that no other direct answer is satisﬁed there. then (M.

although an agent does not know either complete or partial answer. therefore. s).4. s) |= Ai ?ϕ and Qi → ?i ϕ is valid. s) |= Wi Q. which could be the capital of the Czech Republic. s) |= Pi Q (from Fact 31). We write (M. then (M. If a question Q is safe in (M. s) |= [i]PPresQi . s) |= Ai Q. s) |= ?i ϕ. Partial answerhood of a question Qi in some state is equivalent to the existence of a yes-no question. 8 (M. A question Qi is weakly presupposed in (M. There is a formula ϕ such that (M. (2⇒1) is clear and (2⇒3) is from Lemma 10. If a question with pairs of mutually exclusive direct answers Qi is safe in (M. ϕ (as well as ¬ϕ) implies either some α ∈ dQi or ¬α (for α ∈ dQi ). Theorem 16. s). Askability and answerhood Which town is the capital of the Czech Republic: Prague. (M. Deﬁnition 27. (3⇒1) Let us suppose that there is a formula ϕ such that the question ?i ϕ is answered in (M. s) |= ¬Qi . s) |= Ai Q or (M. s) |= Pi Q or (M. then Q cannot be weakly presupposed in (M. neither Prague nor Brno might be the right answer. or Brno? can be inaskable. From the validity of Qi → ?i ϕ we know that inaskability of ?i ϕ8 implies inaskability of Qi and. s) (by an agent i) iﬀ (M. s) |= Qi . but she admits that there is another town. (M. The last one is impossible because of the safeness of Qi in (M.. From Qi → ?i ϕ we know that if (M. Then we say that a question is weakly presupposed by an agent. 62 . which is answered in that state and implied by Qi . An askable question in a state (for an agent i) satisﬁes at least the ‘safeness in a state’ (see Deﬁnition 22). If (M. s) |= Wi Q. Proof. then the following conditions are equivalent: 1. then (M.3. s). i.e. (1⇒2) If (M. then there are three possibilities: (M. s) |= ¬Qi 2. s) |= Ai ?ϕ. s). s) |= Pi Q 3. s) |= ¬?i ϕ iﬀ (M. s).

Let us recall the example from Section 2. where we discussed erotetic implication in IEL. (Generalized) conditional questions consist of two parts: conditional part (context) and query part. Q1 can be 2 formalized by ?{α1 . Q)i → Qi is valid formula. Q1 )i → Qi . . . in some situations it can be useful to display and emphasize the role of a context.5 Context In subsection Relativized askability (page 56) we introduced askability with respect to sets of formulas. An agent asking Q1 : What is Peter a graduate of: a faculty of law or a faculty of economy? can be satisﬁed by the answer He is a lawyer. Especially if it has an important position in reasoning.. Q)i is to keep the importance of a context in inferences with questions. α2 }. Relativized askability helps us to express that Q1 implies Q2 with respect to an auxiliary set of formulas Γ. α2 } is valid. (α2 ↔ β2 )}. Fact 32. βn }. Context 3. . (α2 ↔ β2 )}. . While both kinds are understood as variants of conditional or hypothetical questions. then ({(α1 ↔ β1 ).e. . i. (Γ. Moreover.3. Someone is a graduate of a faculty of economy iﬀ he/she is an economist. As an easy conclusion of Fact 27 we receive that a conditional question implies its query part: ?i {α ∧ β1 . . Q)i can be considered as a generalization of conditional questions in our system.3. . even if she did not ask Q2 : What is Peter: a lawyer or an economist? The connection between both questions could be established by the following knowledge base Γ: Someone is a graduate of a faculty of law iﬀ he/she is a lawyer. and Γ = {(α1 ↔ β1 ).1. the questions Q1 and Q2 are equivalent with respect to Γ: (Γ. (Γ. (Γ. β2 })i → ?i {α1 . Q1 )i → Qi as well as (Γ. In the example. α ∧ βn } → ?i {β1 . β2 }. Q2 )i → Qi is valid.5. ?{β1 . 2 1 The prime reason for introducing of the structures (Γ. . 63 . Q2 by ?{β1 .

s) |= (Γ. Q2 )i → Qi . Q)i . we can see (generalized) hypothetical questions. s) |= Γ → Q2 . . as a counterpart of evocation in IEL. (generalized) conditional questions in the interplay with implication. . s) |= (Γ. (Γ. Q1 )i → Qi . . the following fact points out the cumulativity of explicitly expressed presuppositions. s) |= Γ → Q2 . s) |= (Γ. s) |= [i]Γ and (M. i In some sense. which is a 2 2 contradiction. s) |= (Γ. .. It is necessary to point out that both variants of relativized askability were introduced for the explicit expression of the context conditions of a question. Q1 )i → Qi . we obtain Theorem 17. s) |= Qi .3.e. Q1 )i → Qi . s) |= (Qi → Qi ). Q1 )i → Qi . If (M. This approach is useful. βn }. then 2 3 (M. let us list some properties appearing by combination of conditional and hypothetical questions with implication.5. 3 Relativized askability is transfered by implication. but it is not valid that ?i {α → β1 . Fact 33. Whenever (M. Proof. Now. Q1 )i → Qi and (M. Let us suppose (M. . Q1 )i and (M. Context Hypothetical questions consist of such two parts as well. if (M. α → βn } → ?i {β1 . Q1 )i → Qi we gain (M. For example. s) |= Γ → Q2 . seem to be a 2 counterpart of general erotetic implication in IEL. s) |= (Γ ∪ ∆. Q2 )i . s) |= Γ → Q1 and (M. 1 2 Finally. s) |= (Qi → Qi ). While the correspondence could be seen in ‘philosophy’. these structures diﬀers in properties from IEL ones. Q1 )i and (M. s) |= [i]Γ and (M. i. And it is easy to check 1 2 the following generalization: Lemma 11. s) |= (∆. then (M. then 2 (M. i i 64 . s) |= Γ → Q1 and (M. s) |= (Γ. s) |= (Γ. Similarly. then (M. Most of them are expected. . From (M. however. then 2 i (M. s) |= Γ → Q1 we get (M. s) |= i Qi . (pure) questions and implication are of prime importance in our setting. s) |= (Γ. (M. s) |= (Γ. s) |= Qi and 2 1 (M. If (M. Q2 )i . The same result can be proved for generalized hypothetical questions: If i i (M. Because of (Γ. . Γ → Q. .

. Moreover. Let us recall Example 159 where Q → ?α is valid for each α ∈ dQ. each question ?α is a yes-no question with many good properties.3. An agent can ask questions from the set Φ = {?α1 . they may be ‘worse’ in the description of an agent’s knowledge/ignorance structure than the initial question Q. then inaskability of some ?α means that the answer is either α or ¬α. . . a transmission of askability conditions from an implying question to the implied one is justiﬁed. in some communication processes it is useful to conceal some knowledge or ignorance of a questioner—a criminal investigation is a nice example. . Asking a conjunctive question 9 Let us omit the index i in this section. . This can be understood as a form of ‘suﬃciency’ condition of the set Φ: answerability of its members implies at least partial answerability of the initial question Q. this reducibility is purely based on implication. . we arrive to really less complex questions. An agent can ask questions from the set Φ without completely revealing her knowledge structure. Generally speaking. On the other hand. αn | → j=1 (?αj ) In this case. then there must be a question Qj ∈ Φ that is answered. Φ does not include ‘useless’ questions. . . It means. The set of yes-no questions can be formed by constituents of their direct answers: n ?|α1 . Might it be ‘better’ if we consider the whole set of implied questions based on the set of direct answers to an initial question? If we consider the set of questions Φ = {?α1 . in the second one Q is partially answered. .6 Implied questions If questions are in the implicational relationship. ?α2 . 65 . An implied question is understood as ‘less complex’ in its requirements posed on the afterset substructure. We can understand it as a form of ‘reducibility’ of an initial question to a set of yes-no questions.6. . In the ﬁrst case Q is answered as well. . . Example 16 gives a similar result for conjunctive questions. In comparison with IEL. ?αn } and complete her knowledge step by step. Implied questions 3. but having a partial answer to ?|α1 . The most important property is that the set Φ does not include useless questions. αn | does not give neither answer nor partial answer to some αj .}. . . It could be useful in some cases. . we receive one more property. thus Q→ (?α) α∈dQ On the one hand. Φ is ‘complete’ in a way: if Q is partially answered.

. . Implied questions ?|α1 . everybody is informed that the agent-questioner does not know anything with respect to α1 .3. . . . . αn . . αn | publicly. 66 . .6.

. . especially in motivations. . there are many discussions about the best representation of knowledge as well as belief in the philosophy of logic. . We often referred to the importance of questions in communication processes. Although we often used ‘epistemic’ terminology. In fact.. . ‘knowledge structure’ and its representation is considered to be primary. . The delivering of information in a group has the beneﬁt of (public) 1 See. .Chapter 4 A step to dynamization of erotetic logic 4. . which agrees with the interplay of the idea of representing the knowledge and ignorance structure of a questioner in the process of asking. In our philosophy. The interpretation of questions should be mostly independent of a background system. .g. R1 . 67 . this system is not intended for knowledge representation. such discussions are brought into life again in studies of substructural logics.1 The very aim in introducing questions in epistemic-like systems was to provide an interpretation of questions. Rm . F = S. This chapter is devoted to multi-agent epistemic logic with questions based on modal system S5 and its dynamic extension—public announcement logic. m}. This is understood as an information exchange among agents in a group. . . The term knowledge is often subjected to new interpretations based on a background system. e. .1 Introduction At the beginning of the previous chapter we introduced the language for multi-agent propositional epistemic logic LK with the set of agents A = cpl {1. Nowadays. If we add an accessibility relation for each agent to Kripke frames. the deﬁnition of knowledge modality inside relevant logic in [2]. [m]. we will obtain multi-modal system K with box-like modalities [1].

e. in multi-agent systems we are obliged to introduce new modalities to reﬂect epistemic states in groups of agents. 68 .1 Group epistemic modalities So far we have worked with personal knowledge. we are not going to solve any problem of this formal epistemic representation nor to follow philosophical discussions on it. 8] where the ‘logical omniscience problem’ seems to be the most criticized aspect.. see [9. Multi-agent propositional epistemic logic with questions announcements and. there is a change of epistemic states of group members. transitive. reﬂexive. where knowledge is factive and fully introspective (positively as well as negatively). 9]. S5 represents standard epistemic logic. as a result.2.4. The language LK will be extended by symbols EG . i. CG . this system is subjected to criticism. an agent considers them as having the same ‘value’. Then we apply the public announcement modality in the process of answer mining among agents. This makes it possible to speak about answerhood conditions for groups of agents. Being still in language LK we only have to add that each accescpl sibility relation Ri is equivalence. So. 4. and symmetric relation. [42. However. Let us recall the group of three friends and just one Joker-card distributed among them. where G ⊆ A is a group of agents. cpl and DG . the goal of this chapter is to show the role of questions in a formal dynamic-epistemic system. Of course.2 Multi-agent propositional epistemic logic with questions We have just said that our epistemic framework would be based on multi-modal logic S5. In logic K we understand the accessible states as (epistemic) ‘alternatives’ for an actual state. The equivalence relation makes connected states indistinguishable. However.2. First of all we extend the erotetic epistemic framework by group questions and group epistemic modalities (group knowledge. From Catherine’s viewpoint both possibilities—either Ann has the Joker or Bill has the Joker —are indistinguishable: s1 Ann has the Joker Rc ←→ s2 Bill has the Joker 4. Accessibility relations seem to play a bit diﬀerent role now. an agent can see ‘possibilities’. cf. and distributed knowledge). common knowledge.

. Let us stress that EG does not guarantee that a member of the group G knows that she shares the same information with some other members of the group. cpl The relationship of introduced epistemic modalities is the following: 69 . which can distinguish two cpl models that are indistinguishable in language LK .4. . EG is deﬁnable in the language LK .2. . each agent is aware of this sharing. Each agent (from G) knows ϕ and each agent knows that each agent knows ϕ and each agent knows that each agent knows that each agent knows ϕ and . but also that this fact is reﬂected by everybody in G. Multi-agent propositional epistemic logic with questions Group knowledge EG ϕ means Each agent (from G) knows ϕ. see [42. 227]. for any k ≥ 0. CG is called common knowledge and expresses that the knowledge is maximally shared by everybody in G. As we said. s) |= CG ϕ iﬀ (M. However. and is fully deﬁnable by personal knowledges of G-members: EG ϕ ↔ [i]ϕ i∈G We shall call it group knowledge. Common knowledge The group modality CG is stronger in the following sense: CG ϕ requires not only that ϕ is a group knowledge. p. Multi-modal epistemic logic with common knowledge S5C is not compact. CG can be seen as an inﬁnite conjunction of all ﬁnite iterations of the group knowledge EG : CG ϕ ↔ EG ϕ ∧ EG EG ϕ ∧ EG EG EG ϕ ∧ . so adding group knowlcpl edge is just a conservative extension of the background multimodal epistemic logic S5. . as it is indicated in the deﬁnition of CG and there exists formula in language LKC . s1 ) |= ϕ for each s1 such that s ∗ i∈G i∈G Ri ∗ s1 Ri is a reﬂexive and transitive closure of i∈G Ri and it means that s1 is accessible from s by each Ri (i ∈ G) in k steps. cpl cpl this is not the case of common knowledge. The system S5C is obtained by adding the operator CG and the semantic clause: • (M. Both languages LK and LKE have the same expressivity.

[i]ϕ → DG ϕ is valid in S5CD for each i ∈ G. Let us imagine that a solution of some problem can be obtained by the collecting of particular data from each member of a group of agents. In game theory it is often presupposed that rules of a game are shared by players. Let us remind the group of three friends and suppose that Ann has the Joker. the knowledge of the Joker-owner is implicitly contained in the group. CG ϕ → EG ϕ → [i]ϕ is valid in S5C for each i ∈ G. The accessibility relation based on DG is a subset of each Ri . The standard meaning of DG ϕ is given by the semantic clause: • (M. and so on. they easily reach the hidden fact that Ann has the card. The crucial data are distributed among agents. Although neither Catherine nor Bill know it. DG is called distributed or implicit knowledge.2 cpl 2 Axioms and properties can be found in [23]. Distributed knowledge The last group modality is a bit of another kind. Just common knowledge is a good candidate for group answerhood conditions. It is important to know rules. If an agent knows ϕ. to know that they know that we know it. s) for a group G iﬀ there is α ∈ dQ such that (M. Common knowledge is essential for collective behavior and coordination of collective actions. The term distributed knowledge coincides with the idea of pooling agents’ knowledge together. when we say that a question is (partially) answered for a group of agents? An answer must not be only known by all members. s) |= CG α. s) |= DG ϕ iﬀ (M. Adding DG to language LK does not increase its expressivity. but nobody can solve the problem alone because of the need of the other data. then ϕ is distributed knowledge in every agent’s group: Fact 35. to know that the other players know the same rules. Multi-agent propositional epistemic logic with questions Fact 34. 70 . • A question Q is partially answered in (M. s) |= CG (¬α). s) for a group G iﬀ there is α ∈ dQ such that (M. s1 ) |= ϕ for each s1 such that s i∈G Ri s1 ϕ is true in all states that are accessible for every member in G. In case of questions we considered a question to be (partially) answered for an agent if she knows a fact based on a direct answer.4. • A question Q is answered in (M. but it must be generally known that it is known.2. However. If the agents can communicate. Deﬁnition 28.

we can obtain it by a communication of (only) some agents in the group G . 71 .2. they recognize the state s2 to be common for them. then we can see how to reach a (partial) answer. Fact 36. an answer is present in the group as distributed knowledge. The following example shows their knowledge structure: Example 20. Group questions seem to be a worse problem. i. s1 α β a ←→ s2 α ¬β b ←→ s3 ¬α ¬β Agent a cannot distinguish states s1 and s2 and knows α.2 Group questions and answerhood We introduced questions in a form of an agent’s personal task that ﬁts in her knowledge structure. then it is distributed knowledge for every bigger group. Whenever there is a question which is (partially) answered by at least one agent in a group. DG ϕ → DG ϕ is valid in S5CD for G ⊆ G .2. Again.e. A worse case is that nobody can answer the question—the question is the task for them as well. Let us have a group of two agents a and b. neither of them is able to (partially) answer the yes-no question ?(α → β). The role of distributed knowledge in answerhood will be discussed in the next section. Being in a group of colleagues she asks the question and. However. Such question is askable by each member of a group G and we shall call it group question. Again communication is important for to discover ‘hidden’ information. s) |= QG .b} (α → β). there is only one chance to ﬁnd it. there is no agent with a (partial) answer to it in a group. Whenever she wants to ﬁnd an answer to a question. she has to communicate the question and ﬁnd someone who can answer it. This brings us to the term implicitly (partially) answered question. this nicely shows the idea of a hidden information. s) |= Qi ). A question must be publicly posed and the answer is a result of a communication. Multi-agent propositional epistemic logic with questions If there is distributed knowledge for a group of agents. there is somebody who knows an answer. agent b cannot distinguish s2 and s3 and knows ¬β. s) (for a group of agents G) iﬀ (∀i ∈ G)((M. If an answer should be sought inside the group. Deﬁnition 29.4. If they communicate.. Let us write (M. in the best case. A question Q is an askable group question in (M. 4. it is their group question ?{a.

. which pools knowledge of all agents in a group G together. 72 . Let Ra be nonempty. if Ra = ∅. ∧ [lm ]ψm ) → D{l1 . we can prove that if there is a set of questions.1) which expresses the mentioned idea of pooling agents’ knowledge together for getting their distributed knowledge. ∧ ψm ) → ϕ ([l1 ]ψ1 ∧ . then (partial) answers to questions from Φ are distributed among agents from G. s). Generally. From Facts 34 and 35 we get that it is implicit (partial) answer in a group G. • A question Q is implicitly answered in (M. and a’s knowledge of (partial) answer is in afterset sRa .lm } ϕ (4. So.. Theorem 18. Let us suppose Q is a sound group question in (M.. Now. s) by a group of agents G iﬀ (∃α ∈ dQ)((M. Back to the example. If Q is not sound group question. Unsoundness of Q in (M. what should agents a and b communicate to gain an answer to ?(α → β)? We can ﬁnd an inspiration in implied yes-no questions.. s) |= Q → Qk ∈Φ Qk . the agent a can completely answer ?α and the other one can completely answer the question ?β. The question ?(α → β) implies the disjunction of questions ?α and ?β: ?(α → β) → (?α ∨ ?β) Moreover.2. • A question Q is implicitly partially answered in (M. then Φ = {Q} and there is an agent having (partial) answer to Q. s). it would be useful to communicate questions ?α and ?β in the group. The content of the theorem is based on the S5CD-valid rule (ψ1 ∧ . s). s) by some agent i ∈ G and (M. then everything is distributed knowledge. Multi-agent propositional epistemic logic with questions Deﬁnition 30. and each question from the set can be (partially) answered by some agent from a group G. . All questions Qk ∈ Φ are unsound for a in (M. so is Q. s) cannot be caused by violating of context condition because of its status of sound group question. s) |= DG ¬α). . . s) |= Qk ∈Φ ¬Qk → ¬Q. From (M. then the initial question is implicitly (partially) answered. s) by a group of agents G iﬀ (∃α ∈ dQ)((M. s) |= DG α). then Q is implicitly (partially) answered in (M. Proof.4. s) |= Q → Qk ∈Φ Qk we get (M. If there is a set of questions Φ such that each Qk ∈ Φ is (partially) answered in (M.. their disjunction is implied by an initial question. let us introduce a new agent a. Ra = i∈G Ri .

Public announcement A question. . the author of an announced statement is irrelevant in our framework. . but neither Bill nor Catherine know which of the other two friends. Bill knows Ann has the Joker and Catherine knows Ann has the Joker. s) |= ϕ implies (M|ϕ . Formally we introduce logic of public announcement as an extension of the system S5. some other statement ψ holds. If Ann publicly announces “I’ve got the Joker. and Catherine. Our example gives a typical situation represented in the public announcement logic—after a public announcement of a statement ϕ (“I’ve got the Joker”). The semantics of the new announcement operator is given by the following clause: • (M. Bill.3. such that the intended meaning of [ϕ]ψ is: After the public announcement of ϕ. . 4. From this viewpoint Ann’s announcement has the same eﬀect as if an external observer announces Ann has the Joker. Rm . v is deﬁned as follows: S = {s ∈ S | s |= ϕ} Ri = Ri ∩ S 2 v (p) = v(p) ∩ S 73 . cf. In particular. everybody in the group learns this fact.g. It is group knowledge that each of them has one card and nobody knows the cards of the others and that one of the cards is the Joker. . [42]. In fact. s) |= ψ where M|ϕ = S .4. The statement is understood as information coming to each member of a group in the same way. Possible worlds (states) where Ann does not have the Joker are excluded from the (epistemic) models of both Bill and Catherine. Ann received the Joker. it holds that ψ. which is posed among agents. s) |= [ϕ]ψ iﬀ (M.. e.3 Public announcement Let us return to the group of three friends—Ann. both of them are not able to distinguish between the states where Ann has the Joker and where she has not. can be (partially) answered only if it is at least implicitly (partially) answered by a group. We deﬁne a box-like operator [ ]. has it. R1 .”. The next section shows one of the ways of communication formalization in the role of ‘answer mining’.

s) |= ψ The intended meaning of the dual operator is ‘after a truthful announcement of ϕ.. .3. 83]: (χ ∧ ϕ) → [ϕ]ψ ∧ EG χ (χ ∧ ϕ) → [ϕ]CG ψ (4. . we obtain • (M. 232]. It is easy to see that the diamond-like operator is stronger: Lemma 12. →}): [ϕ]p [ϕ]¬ψ [ϕ](ψ ◦ χ) [ϕ][i]ψ [ϕ][ψ]χ KC[] Lcpl K[] ↔ ↔ ↔ ↔ ↔ (ϕ → p) (ϕ → ¬[ϕ]ψ) ([ϕ]ψ ◦ [ϕ]χ) (ϕ → [i][ϕ]ψ) [ϕ ∧ [ϕ]ψ]χ For common knowledge there is no such reduction (axiom). in fact. p. The language Lcpl has the same expressive power as the language LK . . [42. . The corresponding equivalences give. . Public announcement The model M|ϕ is obtained from M by deleting of all states where ϕ is not true and by the corresponding restrictions of accessibility relations and the valuation function. . Lemma 13. . . our formal work will proceed in the rich propositional language KECDQ[] Lcpl with formulas deﬁned as follows: ϕ ::= p | ¬ψ | ψ1 ∨ ψ2 | ψ1 ∧ ψ2 | ψ1 → ψ2 | ψ1 ↔ ψ2 | [i]ψ | EG ψ | CG ψ | DG ψ | ?i {ψ1 . p. ϕ ψ → [ϕ]ψ is valid. e. cf. an axiomatization of the announcement operator in the public announcement epistemic logic without common knowledge. ∨. the one introduced in [42. cpl This is demonstrated by the following lemma. which provides a reduction of formulas with the public announcement operator to the epistemic ones. 81]. Again we can introduce a dual operator deﬁned in a standard way as ϕ ψ iﬀ ¬[ϕ]¬ψ. it holds that ψ’. If we rewrite the corresponding semantic clause. We only have a rule describing cpl the relationship between the public announcement and common knowledge. ψn } | ?G {ψ1 . ψn } | [ψ1 ]ψ2 74 . The following equivalences are valid in S5 with public announcement modality (where ◦ ∈ {∧. p. s) |= ϕ ψ iﬀ (M.g.4. the language is more expressive than LKC [42.2) From now. s) |= ϕ and (M|ϕ .

p. members of a group learn what was announced. [ϕ]ϕ is valid iﬀ [ϕ]CG ϕ is valid. s) |= ϕ ¬ϕ. it cannot be commonly known in the updated model. In particular. if Ann says “I’ve got the Joker.3.2) we can prove that a formula is true after a public announcement if and only if it gets common knowledge after the announcement (see [42. 4. • Formula ϕ is a successful update in (M. If a formula [ϕ]ϕ is valid.1 Updates and questions Let us return to our example. we call it a successful formula. As a consequence we get Lemma 15. Using the soundness proof of the rule (4. • Formula ϕ is an unsuccessful update in (M. Public announcement We obtain public announcement logic with common knowledge and questions PACQ. If a formula is an unsuccessful update. A formula. Catherine}.4. is called an unsuccessful update. it is evident that its epistemic part (you don’t know it yet) becomes invalid after it is announced. which becomes false after it is truthfully announced (as in our example). s) iﬀ (M.3. we call it a successful update.”.”? This announcement can be formalized by (Ja ∧ ¬[b](Ja ) ∧ ¬[c](Ja )). [ϕ]ψ is valid iﬀ [ϕ]CG ψ is valid. s) iﬀ (M. if it becomes true. This seems to suggest that a publicly announced proposition becomes common knowledge . Deﬁnition 31. Although the formula is true in the moment of announcement. 83 and 86]). So the formula (Ja ∧ ¬[b](Ja ) ∧ ¬[c](Ja )) becomes false after the announcement. but I’ve got the Joker. As we said. s) |= ϕ ϕ. Lemma 14. 75 . Bill. the announced fact becomes commonly known in the group of players {Ann. where Ja means Ann has the Joker. But what if Ann says: “You don’t know it yet.

it remains askable until she gets some new information. which is askable in a state s. Proof. then (M. Ki ϕ.3. is askable in all states from the equivalence class sRi . but neither Bill nor Catherine know it. Thus. Fact 38. then there is a formula ϕ such that after a public announcement of ϕ the question becomes inaskable there. Let ϕ be a successful formula. Formula ϕ is a successful formula iﬀ [ϕ]ϕ is valid. because of the validity of [ϕ]ϕ (ϕ is successful). Successful formulas true in a state are successful updates there: Lemma 16. In other words there is no model and state such that (M. Fact 37..e. and ¬Ki ϕ (for every ϕ) are examples of successful formulas. It is easy to verify that in our S5 background system questions are successful formulas. s) |= Qi and (M|Qi . Public announcement Deﬁnition 32. it follows from Lemma 16 and Fact 37. [ϕ][ϕ]ψ ↔ [ϕ]ψ is valid. s) |= Qi . s) |= Qi Qi . i. which is equivalent to [ϕ]ψ. If [ϕ]ϕ is valid formula and (M. Successful formulas have an important property: they do not bring anything new if they are announced repeatedly. [Qi ]Qi is valid. In S5-models a question Qi . s) |= ϕ. Whenever an agent publicly asks a question. No ‘cutting’ of states in the model M forced by the public announcement of Qi results in (M.4. In our example. Ann has the Joker. It is no surprise that askable questions (in a state) are successful updates.3. Atoms. s) |= Qi iﬀ (M. otherwise it is an unsuccessful formula From Lemma 15 we know that publicly announced successful formulas are commonly known. it does not cause any change in her epistemic model. a publicly announced question is commonly known (see Lemma 15). s) |= ϕ ϕ. (M.2 Public announcement and answerhood Whenever a question is (partially) answerable in a state. If Catherine publicly asks 76 . 4. Lemma 17. s) |= [Qi ]Qi . [ϕ][ϕ]ψ is equivalent to [ϕ ∧ [ϕ]ϕ]ψ (Lemma 13).

s) for an agent i with respect to a question Q. s) iﬀ (M. Fact 39. However. it ‘only’ informs us that a question Q was askable for an agent i in s and that ϕ can be truthfully announced there.e. then the (partial) answer is commonly known among agents in the updated model. The core of this fact is that the formula ϕ is true in each state in an updated model M|ϕ . Bill can infer: “I have not the Joker and Catherine does not know who has it. s) |= Ai Q or (M|ϕ . 3 We could say that an informative formula ‘gives’ a (partial) answer to a question. The role of informativeness is minor. s) |= ¬Qi .. which means (M|ϕ . therefore Ann has it. Contrary to partial answerhood (see Theorem 16) there need not be any logical connection between an informative formula and direct answers to a question. it caused that the question Who has got the Joker?. which was askable for Bill. s) for an agent i ∈ G with respect to a question Q and there is α ∈ dQ such that (ϕ → α) or (ϕ → ¬α) is valid. it is a clear conclusion of Deﬁnition 33 that whenever there is an askable question in a state for an agent. became inaskable after Catherine had asked it. From the informativeness of ϕ we obtain (M|ϕ . then (M|ϕ . If a formula ϕ is informative in (M. so is α or ¬α. s) |= CG α or (M|ϕ . even if her question was not (partially) answered.” Catherine’s question was informative for Bill. This leads us to the deﬁnition of informative formula. Public announcement “Who has got the Joker?”. s). s) |= [i]¬α. The informativeness can be forced by the shape of a particular model. s) |= Pi Q. s) |= Qi ∧ ϕ ¬Qi . 77 . because of the safeness of Q in (M. Lemma 18. then there is α ∈ dQ such that (M|ϕ .3.4. If an informative formula is ‘strong’ enough (i. it implies (partial) answer to a question)3 . Proof. Deﬁnition 33. A formula ϕ is informative for an agent i with respect to a question Q in (M. then after an announcement of an informative formula the agent obtains at least partial answer to the question. s) |= [i]α or (M|ϕ . s) |= CG ¬α. If a formula ϕ is informative in (M.

Neither of them can answer the question ?(α → β)—it is their group question. Her colleague b may help. This inspires the following idea of question’s ‘solvability’. a can directly and publicly ask the question ?(α → β) and reveal her ignorance. However. if there is a ﬁnite series of truthful announcements of agents’ knowledge after which a (partial) answer is a result.b} ¬(α → β). A question Q is • solvable for a group G in a state (M. They reached an answer after a series of announcements of facts they know. s1 α ¬ϕ a ←→ b ←→ s2 ¬α ϕ s3 α ϕ 4. 78 . So. We obtained a sequence of (truthfully) publicly announced agents’ knowledge leading to a (commonly known) answer: [a]α [b]¬β ¬(α → β). . the question is implicitly answered.. but it is not informative for b with respect to the same question. Public announcement We have to be careful. [lk ]ψk α.. thus.e. . . moreover. In cooperative communication agents are supposed to announce what they know with respect to their group question.3 Answer mining in a group In Example 20 we displayed two agents a and b. . s) iﬀ there is a set of formulas {ψ1 . .3. They come to the (complete) answer ¬(α → β). In general. it is commonly known. A question is a successful formula.3. s) |= [l1 ]ψ1 . We can imagine their cooperative communication. informativeness of a formula for an agent does not imply its informativeness for any other one. Example 21. So our group was successful in seeking an answer using just ‘internal’ resources. [[a]α][[b]¬β]C{a. we can say that a question is (partially) solvable. The next example shows a structure where ϕ is informative for the agent a with respect to the question ?α. Deﬁnition 34. where lj ∈ G. [a]α and [b]¬β. ψk } and a direct answer α ∈ dQ such that (M.4. which is. resp. The question ?(α → β) is askable for the agent a and she wants to answer it. i. . common knowledge.

solvability is based on ﬁnite sets. s) |= DG ϕ.. 79 . s) |= DG ϕ (Theorem 19). + Theorem 20. then (M. . . If (M. s) |= ϕ ϕ. which is distributed knowledge in a state. It follows that (M|. If (M|ϕ . Our system with common ∗ knowledge is not compact and this brings us to a ﬁnite version of DG : + • (M. Then either (M. s) |= ψj and due to the validity of (ψ1 ∧ . Solvability follows the idea of the rule (4. s). . s) |= ϕ. ∧ ψk ) → ϕ we have (M|. . [lk ]ψk ¬α. s) |= ϕ or (M|ϕ .1) describing the pooling agents’ scattered knowledge together. + Proof. . simultaneously.. 2. we have talked about in the proof of Lemma 19. Let us suppose that (M. s) |= [l1 ]ψ1 ∧ .4. there must be a state s0 which makes the diﬀerence. then the question is solvable. ψk } and a direct answer α ∈ dQ such that (M. then M|ϕ must be diﬀerent from the model M|. However. s) |= [l1 ]ψ1 . is a successful update there. . ∧ ψk ) → ϕ is valid. From (M. . s) |= DG ϕ. . . s) |= ϕ is not possible: s ∈ s i∈G Ri in S5 and (M.. s) |= ϕ. (M. . [lk ]ψk ϕ. Whenever a (partial) answer of a question is distributed knowledge in the + new sense of DG . + Proof. A formula. If (M. s) |= [l1 ]ψ1 . .. ∧ [lk ]ψk and formula (ψ1 ∧ . Let us write M|. . . s) and. s) iﬀ there is a set of formulas {ψ1 . see [11]: ∗ • (M. . s) |= DG ϕ. for the updated model after the series of public announcements [l1 ]ψ1 . . . s) |= DG ϕ iﬀ {ψ | (∃i ∈ G)((M. [lk ]ψk .. s) |= DG ϕ. s) |= ϕ ϕ.3. This rule inspired an alternative deﬁnition of distributed knowledge as well. . Thus. . where lj ∈ G. + Lemma 19. Lemma 16 says that they are successful updates in (M. . . then (M. . . s) |= ϕ. + 1. s) |= DG ϕ implies (M. s) |= [i]ψ)} |= ϕ ∗ DG is not equivalent to DG and the following is proved in [11]: ∗ Theorem 19. but (M. ψk such that (M.. s) |= DG ϕ iﬀ there are ψ1 .. Public announcement • partially solvable for a group G in a state (M. but not vice versa. s) |= DG ϕ. If (M.. then (M. they are commonly known. s) |= DG ϕ we obtain that each [lj ]ψj is successful formula true in (M.

thus. Final remarks (a) It is impossible that s0 ∈ S M|. . Deﬁnition 28. s) |= CG ϕ. A successful update becomes common knowledge in the updated model: Fact 40. 80 .e.4 Final remarks This chapter combines questions in epistemic framework and communication based on public announcement logic. s) |= DG ψj (Fact 35). Next to positive and negative introspection we can recognize that an agent ‘knows’ questions askable for her: formula Qi ↔ [i]Qi is valid..e. whose answer is accessible by a communication of agents. is answered for a group of agents in the updated model. If (M. s0 ) |= ϕ and from Theorem 19 we obtain s0 ∈ s i∈G Ri . s0 ) and (ψ1 ∧ . The question Who has got the Joker? is mostly seen as a question about facts. then (M|ϕ . As a result we obtain that a question. an answer to a question) and asking questions is one of the essential parts of this communication.4. then (M. and s0 ∈ S M|ϕ : each ψj is true in (M. (b) Let s0 ∈ S M|. each ψj is known in s by an agent from G. However. i.. i. in the equivalence class of s..4. However. we can receive the answer I don’t know. which can be understood as ‘introspective’. . s) |= Qi . s0 ∈ S M|ϕ means that (M. then it is successful update.g. The background system is S5. If (M.. The theorem says an important thing: if an answer to a question is accessible by a communication of agents in a group. and s0 ∈ S M|ϕ ... . then the askability of Qi for an agent i holds in each state in the afterset sRi . (M. s0 ) |= ψj and it is not possible that s0 ∈ S M|. s) |= ϕ ϕ.. cf. (M. ∧ ψk ) → ϕ is valid. Communication is one of the basic tools of a group searching for a solution to a problem (e. 4. In a multi-agent epistemic approach we can understand a question as a ‘task’ (or a ‘problem’) to be solved by a particular group of agents. s0 ) |= ϕ.. Our liberal SAM makes it possible to mix knowledge and questions..

Let us return to the example of three card players. Ja }} The ﬁrst direct answer means that Catherine would ask Who has got the Joker?.4. If an answer to a question is sought by a communication inside a group. Expected direct answers are I know who has got the Joker.4. It solves another kind of questions. The ﬁrst answer indicates that Catherine can completely answer the question Who has got the Joker?. Ja }. It is a yes-no question formalized by the formula ?b {?c {Jb . Catherine? This question is primarily asking for Catherine’s knowledge about the card holder. Bill can ask Catherine Do you know who has got the Joker. The second one indicates that Catherine cannot completely answer it. ¬?c {Jb . the question ?c {Jb . Final remarks In our setting this answer is not a complete answer to the question Who has got the Joker?. I don’t know who has got the Joker. or it must at least be included as a distributed knowledge. he should ask Would you ask the question ‘Who has got the Joker?’. The second one means. but such answer does not reject the possibility that Catherine knows a partial answer. this question is not askable for her.. 81 .e. Whenever Bill wants to ﬁnd out whether Who has got the Joker? is a task for Catherine. In case of questions posed in a group of agents we consider commonly known (partial) answer as a right solution of a question with respect to a group. Catherine? Bill asks Catherine whether the question Who has got the Joker? is a reasonable (askable) question for her. i. An agent’s personal level of answerhood conditions is based on agent’s knowledge. Ja } is askable for her. which according to Theorem 16 means that Catherine can (at least partially) answer that question. Multi-agent approach with group knowledge modalities makes it possible to speak on levels of answerhood. then an answer must be known by some member.

results obtained for regular e-implication.2. On the contrary. It is open for additional restrictions given by both syntax and semantics. 82 . cf. many objections can be raised against it. especially. Our general approach showed that some relations must be supported by additional relations among direct answers. some points are common to both of them. The main inspiration came from the original IEL. It is in correspondence with the second Hamblin’s postulate (cf. chapter 2 presents SAM containing direct answers as declaratives only. Coming back to additional restrictions. Even if I do not want to contribute to a philosophical debate on what a question is. Of course. our epistemic erotetic logic (chapters 3 and 4) admits to have not only declaratives among direct answers. The ﬁrst part of the thesis (chapter 2) is fully developed in the IEL framework. In the ﬁrst chapter we showed that our SAM is convenient for the execution of erotetic inferences and that the epistemic variant is very natural. it seems to me that a word or two should be said about the chosen methodology in both parts of the thesis. Our intention was to work with propositional logic and to keep maximum from the logic of declaratives. subsection 1. or based on the relationship of ‘strongness’ between two questions. SAM is liberal enough to be used in the presented approaches. It inspired the set-of-answers methodology introduced in the ﬁrst chapter as well as the emphasis posed on inferences with questions. whenever we want to analyze all kinds of natural language questions.1). Working in inferential erotetic logic as well as in epistemic erotetic logic I primarily wanted to provide tools for the development of both branches of erotetic logic.Chapter 5 Conclusion Although we can consider the thesis to consist of two almost independent parts. Questions and declaratives are mixed only on the level of consequence relations and the main goal of the chapter is to study relationships among IEL consequences.

in the rest of the thesis we use ﬁnite SAM. On the other hand. it was not intentional. In this epistemic case.CHAPTER 5. Moreover. inferences are based on classical implication. Thus. Discovering them we were faced with the problem of some restrictions required for the chosen SAM. Epistemic framework is the prime and questions correspond to certain states of knowledge. As a ﬁnal result we presented the correspondence of a ﬁnite version of distributed knowledge with cooperative communication aimed at ﬁnding a commonly known (partial) answer to a group question. Questions become a part of epistemic language and they can be considered as satisﬁed in an epistemic state (in a model). Although we introduced a general approach. In the thesis we presented epistemic logics K and S5 extended by questions. these systems can be considered as ‘reductionist’ ones—of course. Chapters 2 and 3 can be called ‘logic of questions’. The semantic work with questions has almost the same ﬂavor as it is with (epistemic) declaratives. all works well with group modalities as well as with public announcement. ignorance. This makes the work with the context condition and presuppositions much easier. Being inspired by IEL and Groenendijk-Stokhof’s approach we wanted to deal with inferences with questions. However. and context). being based on three conditions (non-triviality. Also the deﬁnition of askability of a question in a state is of such a generality that it can be used for any epistemic-like logic. Dynamic approaches can mostly bear the name ‘logic of inquiry’. chapter 4 introduces questions as a part of communication. we may ﬁnd another division of the thesis. we can explicitly work with epistemic contexts and obtain similar structures (on the object-language level) that are introduced in IEL. admissibility. If we compare the contents of the chapters. The conditions required for askability of a question nicely correspond to natural answerhood conditions. Only the questions’ satisﬁability (askability) in a state is more complex. The framework public announcement logic is based on S5. CONCLUSION The second part of the thesis is diﬀerent in its substance. Questions are representable by modal formulas there. Chapter 4 shows that questions behave well together with updates and that they play the expected role in the context of distributed knowledge. 83 . This approach is a novelty inspired by Groenendijk-Stokhof’s intensional erotetic logic together with our SAM. they study inferences with questions and the relationships of questions and declaratives. and presuppositions of an agent. Publicly asked questions are successful formulas and an askable question in a state is a successful update there.

5. p. let us mention the most obvious: • To use another epistemic logic. The original idea behind inquisitive semantics is common to dynamic approaches. This paper makes the best of dynamic logic developed in publications of Johan van Benthem and his collaborators. . Relevant epistemic logic proposed by Ondrej Majer and myself. similarly to updates of models based on public announcements. which is. 40. 42]. “If a proposition consists of two or more possibilities. which was not studied in chapter 2. ψ). Relevant implication provides a good background for erotetic implication.3. A cooperative communication is a raising and resolving issues. can be took into account. A correspondence with our notion (partially) solvable question can take advantage of a generalization: CG (ϕ.1 where we list papers having something to do with an ‘inquiry’ aspect of IEL. The historical part of the ﬁrst chapter includes many publications referring to intensional erotetic logic of Groenendijk and Stokhof. we added two papers based on Hintikka’s approach.” [5. The usual epistemic model is enriched by a new equivalence relation of indistinguishability (‘abstract issue relation’). Propositions are seen as proposals how to update the common state. Closely related to this approach is inquisitive semantics developed by Jeroen Groenendijk and his collaborators. • To develop a predicate version. it is inquisitive: it invites the other participants to provide information such that at least one of the proposed updates may be established. the combination of epistemic and dynamic aspects seems to be a good framework for erotetic logic. This should be based on the recent boom of dynamization. 84 . Let us point out the cited dynamic application from [40]. It is necessary to develop a multi-agent version of relevant epistemic logic. Related works and future directions 5. The generalized version with its associated logic can be found in [5]. It opens many directions of further work.1. One possibility is to apply generalized common knowledge CG (ϕ. cf. [41. The goal of our epistemic logic of questions was just to prepare such a general framework. [ϕn ]CG ψ. [2.1 Related works and future directions Publications related to inferential erotetic logic were mentioned in chapter 2 and subsection 1. equivalent to [ϕ]CG ψ. Roughly speaking. 112] As the recent publications indicate. . ψ) iﬀ [ϕ1 ] . This item brings us to an extensive study of types of answers. • To combine our approach with ‘logics of communication’. cf. Moreover. in fact. there are updates for asking yes-no questions. 22].

• To fuzzify presented approaches. 85 .5. default rules based on questions). Related works and future directions • To apply non-monotonic approaches. Recall the paper [4] mentioned in section 1.g.2 as an inspiration.1.. preferred models. ordering on dQ. Inferential erotetic logic invites to non-monotonic applications (e.3.

2010. 2010. Bˇhounek. Advances in Modal Logic. and V. 2005. [5] I. V. and G. [7] K. [6] E. 2009. Coppock. In D.edu/~coppock/dynamic-questions.Bibliography [1] N. Verhoeven. 2001. and F. J. Synthese. 2004. e Fuzziﬁcation of Groenendijk-Stokhof propositional erotetic logic. In a s L. http://www. Resource-Bounded Reasoning about Knowledge. [10] E. Roelofsen. Toward a dynamic logic for questions. Vardi. De Clercq and L. PhD thesis. 1976. J. 185–188:189–216. editors. College Publications. [9] R. Vel´zquez-Quesada. Logic and Interactive Rationality.pictureofthemoon. Reasoning about Knowledge. 2004. Sieving out relevant and eﬃcient questions. Grossi. a pages 111–146. Yale. Ciardelli and F. University of Leipzig. Fagin. M. L.pdf (downloaded by April 2007). Borkent and R. 171:271–289. Faculty of Mathematics and Informatics. Duc. N. editors. MIT Press. Peliˇ. Relevant agents. B´ ılkov´. The Logic of Questions and Answers. On logic and questions in dialogue systems. 86 . [8] H. 185–188:167–188. [3] M. van Rozen. Majer. O. Goranko. Moses. 2003. Belnap and T.pdf (downloaded by June 2007). Restall. Halpern. Inquisitive Logic. Kurzen. ILLC. Y. pages 22–38. [4] L. http://www. Shehtman.net/~borkent/olaqid. The game of inquiry: the interrogative approach to inquiry and belief revision theory. Logique & Analyse. Steel. and M. Beklemishev. Logique & Analyse. Genot.stanford. [2] M.

2001. ter Meulen. In A.gatsby. Logique & Analyse. editors. H. Zygs munt. The logic of questions. [14] J. volume 8. 87 . In J. (Tenth International Workshop on Artiﬁcial Intelligence and Statistics (AISTATS 2005). volume 17. 2005. Amsterdam.pdf. Distributed knowledge. 2003. Gabbay. Mutanen. Reducibility of questions to sets of s s questions: Some feasibility results. version 1. Twendial’98: Formal Semantics and Pragmatics of Dialogue. Hintikka. pages 19–27. editors. and A. pages 1–60. ILLC. In D. Stokhof. pages 295–337. [17] J. [15] D. Erotetic Logic. pages 16–23. On the history of erotetic logic. R. Wydawnictwo Uniwersytetu Wroclawskiego. Hintikka. Universiteit Utrecht. Knuth. In R. pages 111–124. [20] P. Leuven. Barbados) http://www. Toward question-asking machines: The logic of questions and the inquiry calculus. Proceedings of the 2nd CologNET-ElsET Symposium. Harrah. Annotated handout. Handbook of Logic and Language. Groenendijk. Harrah. Deontic Logic. Questions and Answers: Theoretical and Applied Perspectives. H. Wi´niewski. and J. and Other Logical Matters. The Semantics of Questions and the Questions of Semantics. editors. Handbook of the Logic of Argument and Inference. North-Holland Publishing Company. [13] J. Le´niewski and A. Elsevier. [19] K. 173–175:93–111. Kluwer. Enschede.Bibliography [11] J. Department of Philosophy. Interrogative logic as a general theory of reasoning. [12] J. pages 1055–1125. 2002. editors.uk/aistats/fullpapers/246. Questions and answers: Semantics and logic. 1990. Questions.ucl. editors. Ohlbach. Stokhof. Guenthner. Partitioning logical space. In D. 1997. In J. Woods. Language and Information. Johnson. [16] D. Halonen. Elsevier. Universiteit Twente. Bernardi and M. Groenendijk and M. Gerbrandy. Hulstijn and A. 2002.ac. Second European Summerschool on Logic. Groenendijk and M. 1997. Handbook of Philosophical Logic. 1976. Gabbay and F. Universiteit van Amsterdam.02. Nijholt. editors. Moortgat. 1998. van Benthem and A. I. Wi´niewski and J. August 1990. [18] J.

2000. 2004.harvard. Epistemic Logic for AI and Computer Science. C.Bibliography [21] D. 2004. Nelken and N. s s editor. http://www.harvard. In s M. Iowa. [30] M. SALT XIV: Semantics and Linguistic Theory. editors. Nelken and N. Majer and M. Francez. College Publications. The 18th International Conference on Computational Linguistics. Sixth Meeting on Mathematics of Language. Nelken and N. In Proceedings of COLING2000. [28] R. Logique & Analyse. Epistemic logic with relevant agents. http://www. [29] R. [26] R. pages 123–135. Language. Shan. A logic of interrogation should be internalized in a modal logic for knowledge. [23] J. editor. Socratic proofs for some normal modal propositional n logics. Journal of Logic. pages 53–88. 1999. In M. The Logica Yearbook 2008. Charles University. A modal interpretation of the logic of interrogation. Orlando. Francez. Leszczy´ska. [24] R. Logique & Analyse. In K. A bilattice-based algebraic semantics of questions. Nelken and C. Meyer and W. Young. 2006. Watanabe and R. Peliˇ. 2009. Francez. Shan. [27] R.eecs. 88 . Structure.eecs. 15:251–271. Consequence. 2000. Francez. pages 143–160. van der Hoek. A calculus for interrogatives based on their algebraic semantics. Inference. and Information. 2002. 185–188:217–226. J. Cornell University Press. Peliˇ. [22] O. Querying temporal database using controlled natural language. Bilattices and the semantics of natural language questions. In Proceedings of MOL6. Cambridge University Press. Consequence relations in inferential erotetic logic.edu/~nelken/. [25] R. Peliˇ. B. Florida. Nelken and N. 1995. 2004.edu/~nelken/. In Proceedings of TWLT16/AMILP2000: Algebraic methods in language processing. 2008. Peliˇ. 185–188:259–285. Nelken and C. [31] M. Faca ulty of Arts. B´ ılkov´. Questions and logical analysis of natural language: the case s of transparent intensional logic.

American Philosophical Quary terly. G. van der Hoek. Logic. pages 27–41. Dekker. A. Peliˇ and O. Fermller. 2008. Erkenntnis. [39] M. Minic˘. J. W. [34] C. Logics of communication and change. editors. editors. [40] J. The partitioning semantics of questions. Proceedings of the Seventh ESSLLI Student Session. Springer-Verlag. Questions in Dynamic Semantics. He. College Publications. and P. In M. ten Cate and C. [41] J. On the reducibility of questions. 204:1620–1662. Egly and C. 2002. s 1994. In U. J. pages 251–265. [43] A. Pacuit. and Interaction.pdf. van Benthem and S. and B. Urba´ski. Logic of questions and public announcements. Synthetic tableaux and erotetic search scenarios: Extenn sion and extraction. Question answering: From partitions to prolog. van Eijck. 1978. 89 .soe. van Benthem. ten Cate and C. In M. 2007. Tich´. Butler. [38] M. LNCS. 173–175:69–91. Logic of questions from the viewpoint of dynamic s epistemic logic. ten Cate. 2010. ¸ a In X. Logic and Logical Philosophy. Majer. Logique & Analyse. Peliˇ. Horty. 2011. pages 63–82. 40:265–284. Questions. and B. editor. [36] B. 2006. [33] M. [37] P. Rationality. Shan and B. Kooi. Majer.edu/~btencate/papers/esslli02ss. Peliˇ and O. Springer. Shan. editor. editors.ucsc. syntactically. [42] H. and E. Proceedings of TABLEAUX 2002: Automated reasoning with analytic tableaux and related methods. Aloni. Some remarks concerning modal propositional logic of n questions. In s Eighth International Tbilisi Symposium on Language. van Ditmarsch. pages s 157–172. Urba´ski. Springer. 2001. Wi´niewski. answers. and logic. The Logica Yearbook 2009. To appear. Dynamic Epistemic Logic. Elsevier. 2002. Axiomatizing Groenendijk’s logic of interrogation. Shan. Kooi. [35] B. 15:275–284. Springer. Information and Computation. Logic and Computation 2009. Nissim. 2009. Toward a Dynamic Logic of Questions.Bibliography [32] M. http://users. In M. 1998. 6:187–196.

[46] A. Logique & Analyse. Pa´niczek. s Synthese. [50] A. Rodopi. 90 . [49] A. In J. Erotetic search scenarios. 2003. The Posing of Questions: Logical Foundations of s Erotetic Inferences. Erotetic logic and explanation by abnormic hypotheses. editors. s 33:299–326. Journal of Philosophical Logic. [45] A. pages 215–236. Socratic proofs. 1995. Synthese. s Questions and inferences. Kluwer. 2006. problem solving. Amsterdam/NY. Wi´niewski. s [48] A. Reducibility of safe questions to sets of atomic yes-no s questions. Wi´niewski. 2004. Logique & Analyse. Wi´niewski. 173–175:5–43. Wi´niewski. 134:389–427. Wi´niewski. Jadecki and J. 2004. Wi´niewski. 2001.Bibliography [44] A. Wi´niewski. 185–188:139–166. [47] A. Erotetic search scenarios. 1999. Lvov-Warsaw School: s The New Generation. and deducs tion. 120:295–309.

- Lecture Three
- Problem of Induction
- Pritchard Structure of Skeptical Arguments PDF
- Diversity Notes
- PPrawitzReply_1.4bis (1)
- Dialnet-ADilemmaForSkeptics-4232468
- Persistence, Parts and Presentism
- Legrand 2002
- 08 Heller +
- 307 Fictional Ism
- Car Nap Empiricism Semantics Ontology
- Kathavatthu and Vijnanakaya
- Explicaciones
- hhh
- On Certainty - Ludwig Wittgenstein - Full Text - Athenaeum Library of Philosophy
- The Nature of Logic
- Ludwig Wittgenstein on Certainty 1972
- Ontological Raws
- Bill Brewer
- Sundholm, Inference Versus Consequence
- Briner, Denyer y Rousseau (2009) Evidence Based Management Concept Cleanup Time
- Schwarz Metaph Final
- Argumentation without arguments
- Colin McGinn Problems in Philosophy the Limits of Inquiry 1993
- Chibber
- Basic Research Elements.pdf
- Mansfeld, Jaap. Ambiguity in Empedocles B17, 3-5 - A Suggestion (1972)
- Stroud - 'Unpurged Pyrrhonism'

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd