MODELLING THE MAN MACHINE INTERACTION

Erotetic Logic and Information Retrieval Systems

Antonio Bellacicco

Department of Theory of Systems and Organizations, University of Teramo, Viale Crucioli,122, Teramo, Italy

Mario Vacca

“B. Pascal” High School, Via P.Nenni, 48, Pomezia, Italy

Keywords: Question, answer, search engine, erotetic logic

Abstract: The usual communication between man and machine

is a one way interaction. It can be upgraded

considering a two way interaction if the basic constituents of an information retrieval system are deeply

modified in their principles. In this paper we redefine, in a syntactical way, some concepts of the erotetic

logic to make them more easily computable and show how them can be used to solve some problems in the

field of information retrieval systems. The result is the possibility to build more flexible and powerful

information retrieval systems.

1 INTRODUCTION

Cognitive science, data mining, information retrieval

or artificial intelligence, among other areas, analyze

how information are requested and exchanged by

two dialogists; the view proposed in this paper

mainly refers to the last two mentioned areas.

In retrieval systems, the questioner

comm

unicates with the questionee by a query

language and the questionee is usually modelled by

an engine able to search a database or a knowledge

base (from now KB for short) for the information

requested by the questioner. Most difficulties in the

use of search engines are just related to the

narrowness of the query language used (for example,

it is tied to isolated words and not to an entire

meaningful sentences) and to the impossibility of

interacting. Furthermore, the inefficiency of the

search or the absence of any appropriate strategies

when too many answers or no one are found often

characterize search engines. These problems,

among others, lead the research to different

directions in order to make more versatile and

powerful the information retrieval or the Q/A

systems, as pointed out by many authors among

which Bellacicco (2002, 2003), Burhans (2002) or

Loia et al. (2002).

In this paper we firstly redefine some concepts

of t

he erotetic logic in a syntactic way and hence we

show how these concepts can be exploited to solve

the problems of “multiple answers” and “plausible

answers”. As a consequence, these concepts are seen

as the core for a search engine able to output either a

sentence or another question.

Because of the paper is not self-contained, we

efer to Groenendijk and Stochow (1997) for an

introduction about the theory of questions and

references or to Ginzburg (1995a, b), Krifka (2001),

Piwek (1997), Ram (1991), Wisniewki (1995) for

different approaches.

After a brief illustration of the basic functionality

foreseen

for the model, the third section is entirely

devoted to questions: firstly a sketch of the logical

representation used for them, next some

transformation rules are discussed and examples of

their use are reported. Finally, in the fourth part, the

jump of domain is presented: a way to answer to

questions using the similarity among concepts.

In the sequel, we use the terms query, question,

respon

se and answer with the following meaning:

query: a question in a natural language form;

question: pure logical form of a query;

response: answer to a query;

answer: logical form of the response to a question.

We use the term answer when confusion do not

occurs.

Furthermore, when no ambiguities occur, we

so denote by x

the n-tuple x

,…,x

. IRS stands for

172

Bellacicco A. and Vacca M. (2004).

MODELLING THE MAN MACHINE INTERACTION - Erotetic Logic and Information Retrieval Systems.

In Proceedings of the First International Conference on Informatics in Control, Automation and Robotics, pages 172-179

DOI: 10.5220/0001139501720179

 SciTePress

Information Retrieval System.

2 THE MODEL: FOUNDATIONS

AND FEATURES

The factors that usually characterize information

retrieval systems are:

1) the organization of the knowledge base;

2) the features of the query language;

3) the concept of answer used (answerhood);

4) the search algorithms.

As above mentioned, many attempts to improve

the performances through the extension of one or

more of these features had been made. Obviously,

because of the strong connections, it is not possible

to extend significantly one of the previous

mentioned features without modifying the remaining

ones. We started the analysis from the well known

consideration that the memory organization is

crucial to the performances of an IRS. It had been

indeed showed by many authors, like Schank (1982),

Cautiero et al. (1991), Kolodner (1983a, b) and

Lebowitz (1988), that an efficient storage of

information leads either to forgetting or to a

powerful ability in reconstructing episodes. In this

paper we are concerned with retrieval methods that

don’t explore the entire Knowledge Base in order to

find some information. Following Schank (1982),

KB is constituted by concepts organized by means

of two kinds of links: generalization and packaging.

Therefore the search algorithm will know only the

structure (the organization principles) of KB.

As we suppose such a memory organization, it

follows that a basic mechanism to search needs a

more dynamic and flexible algorithm than the

existing one.

Erotetic reasoning seems adequate to perform

this task.

The importance of the erotetic reasoning

had been pointed out by many authors: Schank

(1986) showed that questions play a crucial role in

understanding and that reasoning with questions

helps to find or build up answers (also creatively);

Wisniewsky (1994, 1995) introduced the semantic

notion of the evocation (the generation of questions

from declarative sentences) and of the erotetic

implication (the generation of questions from

questions and declarative sentences). Unfortunately,

both approaches are very difficult to apply: the first

one because it gives specific domain dependent rules

primarily, while the second one, yet being very

general, considers the logic of questions from the

semantics point of view, with the consequent

processing difficulties.

In this paper, mainly concerned with the

extension of the search engine, we propose some

general syntactic rules to transform questions in

order to search a base of knowledge in an

“intelligent” way. Even thought it is inspired both to

the cognitive and erotetic logic approaches, our

proposal is different from those in which the rules to

transform questions are of a syntactic nature. The

main ideas are synthesized in the following two

principles:

1) answers to a given question are defined in a

syntactic way;

2) questions are transformed syntactically in order to

find suitable answers.

We foresee to embody the following features in

the model we propose:

1)communication abilities, to rule the flow

query/answer between questioner and questionee;

2)search engine/question processor, that takes a

question to a declarative sentence or to another

question;

3)parser, that inputs a query (or a declarative

sentence) in natural language form and outputs its

logical representation; the translation process we

think of is a two stage one:

queryÆ (intermediate form)Æ question (logical

form);

4)output, that translates a logical sentence in a

natural language.

3 THE ANSWER RETRIEVAL

We follow the non reductionistic approach, in which

questions are specific expressions not reducible to

any other object. Therefore, we will consider a logic

language with both an assertoric and an erotetic

part, the latter consisting of various kinds of

questions.

3.1 Question representation and

answerhood

The representation we will use is like that proposed

by Wisniewski (1995).

Broadly speaking, a logical question consists of

five parts: the erotetic symbol ? (telling that the

formula is a question), a list of variables (what the

question asks for), a list of categorial qualifiers (that

specifies the categories of object involved in the

question), an erotetic symbol (that specifies the kind

of question) and a body (a predicate that represent

the core of the question), in agreement with the

following syntax:

?<list of the variables> <list of the categorial

qualifiers> < symbol specifying the kind> <body>

A question will be structured as follow:

MODELLING THE MAN MACHINE INTERACTION - Erotetic Logic and Information Retrieval Systems

173

..x

,…,x

)]...[P

,…,x

)]Σ{B(x

,…,x

)}

where:

-? indicates an interrogative formula;

-x

..x

are variables;

-predicates P

are categorial qualifiers;

-Σ ∈{S, W, O, U, T} or is absent

-the predicate B(x

,…,x

) is the body of the

question.

In the case that both Σ and list of variables are

absent, we have a yes-no question.

See Wisniewski (1995) for a detailed account about

the meaning of the previous constants.

For example, the query “Have Begin and Vance

wives ever met?” could be translated as:

?λt[time(t)][woman(α)][woman(β)][wife(α,Vance)]

[wife(β,Begin)]S{occurs(t,meet(α,β))}

We define the following operators on questions

that will be useful in the sequel:

a) Body(Q) that returns the body of Q;

b) Qualifiers(Q) that returns the list of categorial

qualifiers;

c) Variables(Q) that returns the set of question

variables.

As it happens with declarative formulas, also

questions can be transformed using substitutions:

Q’ = Qθ.

Definition

Let θ be a substitution and Q a question.

Qθ is the question such that:

Variables(Qθ) = (Variables(Q) - the set of variables

substituted by constants) θ

Qualifiers(Qθ) = Qualifiers(Q)θ

Body(Qθ) = Body(Q) θ

If Variables(Qθ) = ∅ then Qθ is a first kind

question (i.e. the symbol specifying the kind is not

present).

Even thought Qualifiers(Q)θ is an abused

notation, we use it here for sake of clarity. It should

be also noticed that closed substitutions produce

yes/no questions. For example the query “On which

circumstance do Vance and Begin meet?”,

represented by the question

? s[circumstance(s)]S{pack(s,meet(Begin,Vance))}

can be transformed using the substitution

θ={s Å DiplomaticVisit}

in the question

Qθ=

DiplomaticVisit[circumstance(DiplomaticVisit)]

S{pack(DiplomaticVisit,meet(Begin,Vance))}

i.e. “Does a diplomatic visit in which Vance and

Begin meet exist?”

The concept of answerhood we consider is like

the one in the logic programming.

Definition

Let Q be a question and θ a substitution. θ generate

an answer to Q iff Body(Q)θ is deductible from KB

with all qualifiers in Qualifiers(Q) θ. In the latter

case Qualifiers(Q) θ ∧ Body(Q)θ is the answer to Q

generated by θ.

Notice that Qualifiers(Q) θ has to be interpreted as a

conjunction.

Because of the correspondence between

substitutions and answers, when no confusion

occurs, we can use indifferently the two terms

meaning answer.

Example:

The question

?x [Food(x)]S(occurs(eat(Mary,x), today)

could be answered by the substitution

θ = {xÅpizza }.

The corresponding answer will be:

Food(pizza)∧occurs(eat(Mary, pizza),today).

In our approach, a computational one in his

nature, soundness of a question is tied to the

existence of an answer in the KB. Therefore, a

question can be sound with respect to a KB and not

sound with respect to another.

3.2 Syntactical Processing of

Questions

It is possible to define when a sentence is more

general than another one, as in Lu J. et al (1998):

Definition

Let a’ and a’’ be two declarative sentences.

a’ ≤ a’’ if and only if a substitution θ exists such that

a’’ θ= a’

We consider a relation between questions, we

call it transformation and denote it by Q ⊢Q’ that is

like erotetic implication, but is performed

syntactically.

From now on, Ans(Q) will be the set of possible

answers to question Q.

ICINCO 2004 - INTELLIGENT CONTROL SYSTEMS AND OPTIMIZATION

174

Definition

Let Q’ and Q’’ be two questions. Q’ is less general

than Q’’ (denoted by Q ⊢Q’) iff

∀a’’∈Ans(Q’’)∃X⊂Ans(Q’) (∀a’∈X (a’ ≤ a’’))

The answers to Q’’ are called partial answer to Q’.

Obviously, it is not necessarily true that

∀a’∈Ans(Q’)∃a’’∈Ans(Q’’) (a’ ≤ a’’)

Example

The answers to the query “On which circumstance

do Vance and Begin’s wives meet?” are more

specific than the answers to the query “On which

circumstance do diplomats’ wives meet?”

Now we show how the relation ⊢ between

questions can be exploited to solve two kinds of

problems: the construction of plausible answers and

the multiple answers selection.

The first problem arises when a search in a KB

gives no answer, but it is possible to build a

plausible answer. Firstly, we observe that from the

previous definition it follows that if Q’ and Q’’ are

two questions such that Q’ ⊢Q’’ and A is an answer

to Q’’, also Q’A is a question. This kind of question

can be useful in the search process. Indeed,

answering an implicated question restricts,

sometimes drastically, the search, as the following

result shows.

Lemma

Let Q be a question and A a partial answer to Q. If

an answer to the question QA exists, then an answer

to Q exists too.

Proof: we observe that if θ is an answer to QA, by

the properties of substitutions, Aθ is an answer to Q

and the lemma holds.

This lemma enable us to build plausible answers

applying repeatedly the following steps:

- to compute an implied question;

- to retrieve the answer to the implied

question;

- to apply this answer to the original

question in order to restrict the search.

We consider for example the well known query

about the existence of a meeting between Vance and

Begin’s wives taken from (Schank, 1986) or

(Kolodner 1983a, b) in order to show how the

transformation process can be obtained in a

syntactical way. In the example (slightly simplified

to make it more readable), questions are indicated by

a Q, declarative sentences by a D and answers by an

A or PA (partial answers), capital letters followed by

a number.

Q1)Has Vance’s wife ever met Begin’s wife?

Q2)On which circumstance do Vance and Begin’s

wives meet?

Q3)On which circumstance do diplomats’ wives

meet?

D1) Every time they go with their husbands and they

meet, diplomats’ wives meet each other.

Q4)On which circumstance do diplomats usually

meet?

A4) During international meetings.

PA3) When they go with their husbands to those

meetings.

PA2) When they go with Vance and Begin to those

meetings.

PA1) Yes, when they go with Vance and Begin to

those meetings.

Q5) Do meetings in which Vance and Begin meet

and their wives go with them exist?

A5) [NO]

Q6) Which conditions have to hold for Vance and

Begin to meet and their wives to go with them?

Q7) Which conditions have to hold for two

diplomats to meet?

Q8) Which conditions have to hold for two people to

meet?

A8) People need to be in the same place in order to

meet.

PA7) Diplomats need to be in the same place in

order to meet.

PA6) Vance and Begin need to be in the same place

in order to meet.

Q9) Are Vance and Begin in the same place?

A10) [NO. Vance is in U.S.A. and Begin in Israel.]

Q10) Which conditions have to hold for Vance and

Begin to be in the same place?

Q11) Which conditions have to hold for two

diplomats to be in the same place?

A11) A diplomat usually goes to another country to

meet another diplomat.

PA10) Vance needs to go to Begin’s country to meet

him.

Q12) Has Vance ever gone to Israel?

A12) [List of visit]

(Notice that some further steps would have been

necessary to obtain this answer).

Q13) In which of those visits did their wives go with

them?

A13) [Particular visits]

It is possible to describe the previous sequences of

questions and answers by the following schema, in

which Æ relates a question to an answer

Q1⊢Q2⊢Q3D1⊢Q4ÆA4ÆPA3ÆPA2ÆPA1⊢Q5

Æ[NO]

MODELLING THE MAN MACHINE INTERACTION - Erotetic Logic and Information Retrieval Systems

175

Q5⊢Q6⊢Q7⊢Q8ÆA8ÆPA7ÆPA6⊢Q9Æ[NO]

Q5 suspended

Q9⊢Q10⊢Q11⊢A11ÆPA10⊢Q12Æ[List of

visits]

Q5 reconsidered

Q5 [List of visits] ⊢Q13Æ [Particular visits]

This example shows how it is possible to build in

a syntactical way a sort of “deductive chain”

involving questions whose aim is to search in a KB

an answer to a given question. These chains are

similar to those called erotetic derivations in

(Wisniewski 2003).

The multiple answers selection takes place when

a lot of answers are available in the KB and the need

of selection arises.

Consider the following query as an example in

(Ginzburg 1995a, b):

“Who works in the Philosophy Department?”

Two kind of answers are possible:

1) a general one;

2) a more specific one.

For example

1)A group of neo-positivist philosophers and some

erotetic logicians.

2) John X , Mark A, etc.

The first answer is related to a question like

? P [P(x) ]S{work(x, Philosophy Department )}

In this case the questioner is not interested in the

objects, but in the properties they have. The second

kind of answer is related to a question such

? x S(work(x, Philosophy Department))

There can be people interested in the first kind of

answer and not interested in the list of names.

Ginzburg (1995a, b) showed that the questioner is

not always looking for a specific answer. Indeed, he

stressed that, to solve this case, it is necessary to

know the goal of the questioner. We believe that the

capability of replying to a question by another

question could be useful.

We propose the following solution:

-every query must be translated in a more general

question as possible ;

-the output is the answer to this question and some

queries asking the questioner for further

specifications.

An example of reply to the previous question could

be: a group of neo-positivist philosophers and some

erotetic logicians and the query “Would you like to

know the kind of people or just the people’s names

working in the Department?”

Finally we indicate some way to obtain a more

general (or a more specific) question from a given

question:

) “introduction of variable”

Let Q be a question and c a constant such that P(a)

for some predicate P. A more general question Q’ is

obtained in the following way:

Variables(Q’) = Variables(Q) U {x};

(x is a new variable not belonging to Variables(Q));

Qualifiers(Q’) = (Qualifiers(Q)θ;

Body(Q’) = Body(Q) θ.

(θ substitute c with x).

Obviously, the effect of this transformation is a

widening of the search field.

) “search for conditions”

Variables(Q’) = [C];

(a new variable is introduced)

Qualifiers(Q’) =Qualifiers(Q)U (condition(C));

Body(Q’) = (C ⇒Body(q)).

4 THE JUMP OF DOMAIN

In this section we will deal with a kind of inference

which is very common in the ordinary reasoning: it

is something beyond the modus ponens and the

syllogism, which considers at least a chain of two

implications like (x→y, y→z ) ¶ x → z. Another

scheme of reasoning is the so called “trial based

reasoning”. The common support is the concepts-

predicate table, which crosses asserts and predicates.

We can drag predicates by chaining concepts,

provided there is a medium concept which joins two

other concepts.

A simple dissimilarity measure between two

concepts may be the inverse of the owned

predicates, so that if two concepts do not share any

predicate it means that between them there is an

hole. The material implication between two asserts

can be interpreted as the transfer of a predicate

ownership from the explanans to the explanandum.

Sharing a common predicate means that both the

asserts can be chained by an if…then.

It is also easy to see that the implication is

merged into an hyperbolic metric structure so that

d(x,y) - d(x,z) > d(z,y). This means that the join of

two concepts which share only one predicate admits

the jump to a third concept. The triangular relation

does not hold. As a consequence, it is built in a

priority between two predicates. The identification

of a cause after a trial implies that the simultaneous

occurrence of a subset of predicates is definitively

higher than elsewhere in the table of observed cases.

From the metric point of view, the product of an

inverse function is normalized by the number of

occurrences of each predicate among the set of

considered concepts. The chain of the product of the

inverse of the number of occurrences of n predicates

and m concepts means that we can go along a

backward path which can simulate the so called

ICINCO 2004 - INTELLIGENT CONTROL SYSTEMS AND OPTIMIZATION

176

explanation of an evidence. In the and relation no

priority is supposed. The hole cuts down any direct

relation, represented by an edge between two

predicates. We can overcome the stop by a sequence

of transfers, so that we can built up an assert which

joins two distant predicates and therefore two distant

concepts.

For example, every man can be affected by a disease

and is mortal. A fail in a machine is like a disease.

We can make a bridge so that we can built up a

chain: every machine is mortal.

We can built examples whose chain can be pretty

long just using and or if…then:

i. young(P

)people(

) are itching(P

) to fight(P

);

ii. young people like(P

)driving(P

)cars(P

We can join itching to fight with like driving cars.

We can write :

iii. people itching to fight like driving cars.

This seems to be a common way of thinking.

We can also consider the reverse path, so that we

can assert:

People who like driving cars are itching to fight.

The reverse reasoning is allowed here as far as we

do not have forbidden directions in any edge of the

chain. Through young(P

)people(Π) we drag

itching(P

) to fight(P

) and join itching (P

) to

fight(P

) to

like (P0) to driving cars (P5).

The join brings to a new concept: the young drivers

are aggressive people.

We propose here the algorithm BTKSA based

on the concept-predicate table.

The algorithm BTKSA is based on the Ariadne’s

thread logic in the search of the exit in the labyrinth.

Input: a question in the form

..x

(x)]…[P

(x)]Σ{A(x)}

/*It is supposed that the leading predicate exists*/

Step 1. a partial DB is rescued by the leading

predicate Α in the query, which becomes the root

of a tree T;

Step 2. all the subtrees whose roots are on the tree

generated by A are identified in the partial DB;

/*the edges connecting vertices already connected

in the graph are deleted in order to avoid cycles.*/

Step 3. the asserts joining n-tuple of the predicates

are identified for n = 2,3,…,k, where k is max

length of the branches of the subtrees;

Step 4. all the asserts of the local DB are reduced to

branches of at least a subtree of the main tree;

Step 5. a scanning process of the whole tree is

performed to identify the branch whose vertices

are the same predicates distributed in the same

order apart for the quantifier and variables which

are substituted by a predicate and a free variable;

Step 6. a jump ¶ is a path in the tree connecting

different predicates so that the backward course

follows;

/* we recall that a material implication enjoys the

transitive property so that:

p(A)→q(B) , q(B) →w(C),¶ p(A)→w(C).

A backward path recognizes p(A) as the sources

of the jump. In the tree a path joining separately

the tree predicates attributes the extension of the

asserts containing C assigning them A. */

/* the equivalent statistical reasoning is the so

called Bayes theorem, which supports a backward

reasoning.*/

/* the negation of the last consequence implies the

negation of the whole implication. The sequence

of backward reasoning implies the contraction of

the chain of implications and the negation of the

implication of the negated consequence is also

true.*/

Step 7.we built up an assert which contains a truth

not belonging to the DB as far as it is an

induction.

Step 8. stop.

Let consider the following formula:

(((B∧C)⇒A)and(C ∧ E)))¶ E (1)

The ¶ means a jump as far as we need to connect a

formula for jumping to another assert. The jump

means only a non evident join which can be

interpreted as an implication if we give a direction

to the path. Actually there are at most n! paths which

are reduced by the incidence matrix. An example of

compressed incidence matrix of a concept-predicate

table is given in the sequel. Each row can be an

assert. The direction can be given by further

restrictions, considering at first the concept with

min sharing, then deleting it and recycling. There is

no guarantee of a total ordering if there are no

specific requests given by specific implications

between couples of concepts. The predicates of C

are transferred to E. E joins two predicates which

are not joined at the beginning of the reasoning. The

transfer is assured by the jump of a hole. The

Ariadne strategy for moving out of a labyrinth is

followed here as far as we avoid to cross again the

same square in the table. Moreover, we can use as a

strategy both a forward movement and a backward

one: from the target of the jump or from the first

concept. The choice depends from the target domain

as before mentioned.

The connection to E is not obvious if we do not join

and P

as well as P

and P

The strategy on the table is to move both

horizontally and vertically. As an example we

suppose the following incidence structure

connecting concepts to predicates:

(A) contains P

, P

(D,C,B), share P

, P

MODELLING THE MAN MACHINE INTERACTION - Erotetic Logic and Information Retrieval Systems

177

(A,C) share P

(D,E) share P

, P

The jump allows the following scheme:

from the chain (E

→

C) ∧ C ¶ E.

If C is the evidence E therefore drags P

and P

.We

see that P

, P

are 4 predicates and A,B,C,D

and E are 5 asserts bearing 5 concepts, so that the

referred subsets overlap each other and a chain pulls

two predicates to E. We see that the path starting

from C stops at E if two predicates partially overlap.

The jump connects the asserts sharing at least

one of the two predicates. In other terms, the logical

connection joins the asserts, while the jump joins the

predicates and indirectly the asserts. The jump

generates

something like an implication: P

is a

common property for A and B, as well as P

and P

are common properties for B and C. Finally, P

is a

common property for C and D. E actually extends to

C the properties of D.

The backward reasoning can be ascertained

from the concept-predicate table if we set up the

network of the predicates between all the couples of

concepts. As a consequence, we set up a matrix of

accessibility between concepts. The network of

relations can be considered for the backward

reasoning. The Ariadne algorithm starts from the

end of the labyrinth; that is the evidence to get the

entrance, which is the remote cause of a chain of

causes and concurrent causes. The problem is to

avoid to meet the same cause many times. In order

to get the root of the tree, starting from the top, it is

necessary to cut an edge whose accessibility

measure is identified by the measure at the right of

the inequality. If r and s are number of predicates

shared by two overlapping concepts, x and y,

respectively, we have the following relation between

their segregation evaluated in terms of the inverse of

the number of owned predicates:

−

>−

(2)

−

(3)

where r < s.

The formula looks like a probabilistic formula:

P(x) – P (x ∩ (¬ y) ) > P(y) (4)

where x and y are the concepts whose extentions

are r and s, respectively and P is the inverse of the

extentions r and s.

There is some difference in the underlying logic

background. In a concept-predicate table we have:

x ∧ ¬ (x ∧¬ y) → y (5)

in other terms:

( x ∨ y) → y.

In probability we usually have:

P(x) + P(y) - P(x ∩ y ) > 0 (6)

In probability x and y are events while here may be

the extension of concepts as well as of conceps in

terms of number of predicates.

The sound difference here is that formula (2) can be

rewritten as

−

(7)

The geometry is quite different just as the

underlying logic. The consequence is that the

chaining supposes y ⊂ x, so that the concept x

must include y. In other terms, the predicates of y

are the predicates of x but it is not true the

viceversa. The backward chaining is therefore

allowed provided that the previous condition is

satisfied. Cutting a circle means cutting the

superposition between two concepts. The path from

the evidence to the cause is therefore ruled by the

previous inequalities.

5 CONCLUSIONS AND FURTHER

WORKS

In this paper we first redefine some concepts of

erotetic logic like question implication and then we

show how they can be used to search information in

a KB. The result is the base for a more powerful

search engines. Future works will involve an

implementation of the system.

REFERENCES

Bellacicco A., 2002. A New Tool for Textual Database

Analysis and Management, in Brebbia C.A., Pascolo

P. (eds.) Management Information Systems 2002: GIS

and Remote Sensing. Wit Press.

Bellacicco A., 2003. Two automata linguistic

communication system (TALCS) in 7

world

ICINCO 2004 - INTELLIGENT CONTROL SYSTEMS AND OPTIMIZATION

178

multiconference on Systemics, Cybernetics and

Informatics. Orlando, Florida, USA.

Bell J.,1995. Changing attitudes. In Wooldridge M. and

Jennings N.R. (eds.) Intelligent Agents, Post-

Proceedings of the ECAI-94 Workshop on Agent

Theories, Architectures, and Languages.

Ben-Ari M., 1993. Mathematical Logic for Computer

Science. Prentice Hall

Burhans D.T. Expanding the Notion of Answer.

Proceedings of the Grace Hopper Conference,

Vancouver

Cautiero G., Sessa M.I., Sessa M., and Vacca M, 1991.

Conceptual Processing of Texts in Paleography.

Information Processing and Management. 27(2/3),

219-227.

Frauenfelder U.H, 1996. Computational models of

spoken word recognition in Dijkstra T. and de

SmedtK. (eds.) Computational Psycholinguistics: AI

and connectionist models of human language

processing. Taylor & Francis, Great Britain.

Ginzburg J., 1995a. Resolving questions I. Linguistics and

Philosophy, 18(5), 459-527.

Ginzburg J., 1995b. Resolving questions II. Linguistics

and Philosophy, 18(6), 567-609.

Groenendijk J., Stochow M., 1997. Questions. in Van

Benthem J., Ter Meulen A. (Eds.) Handbook of logic

and language. Elsevier, Amsterdam.

Kolodner, J. L., 1983a. Maintaining organization in a

dynamic long-term memory. Cognitive Science, 7,

243-280.

Kolodner J.L., 1983b. Reconstructive Memory: A

Computer Model. Cognitive Science. 7, 280-328.

Krifka M., 2001. For a Structured Meaning Account of

Questions and Answers. In Fery C., Sternfeld W.

(Eds.) Audiatur Vox Sapientia. A Festschrift for Arnim

Von Stechow, Akademie Verlag, Berlin

Lebowitz M., 1988. The Use of Memory in Text

Processing. Commun. ACM 31(12), 1483-1502

Loia V., Luongo P., Senatore S, Sessa M.I., 2002. Info-

Miner: Bridging Agent Technology with Approximate

Information Retrieval. In Nikhil R. Pal, Michio

Sugeno (Eds.) Advances in Soft Computing -

International Conference on Fuzzy Systems. Lecture

Notes in Computer Science 2275, Springer, Berlin.

Lu J., Harao M., Hagiya M.,

1998. Higher Order

Generalization. in Dix J., Fariñas del Cerro L.,

Furbach U. (Eds.) Logics in Artificial Intelligence,

European Workshop. Lecture Notes in Computer

Science, Springer, Berlin.

Ostrovskaya Y.A, 2003. Recognizing text structures:

computational applications of formulaic theory of

language. in 7th world multiconference on Systemics,

Cybernetics and Informatics. Orlando, Florida , USA.

Piwek P., 1997. The construction of Answers. In Benz A.,

Jaeger G. (Eds.) Proceedings of MunDial: the

Munchen Workshop on the Formal Semantics and

Pragmatics of Dialogue, CIS- Bericht 97-106

Department of Computational Linguistics, University

of Munich.

Ram A., 1991. A theory of questions and question asking.

The Journal of the Learning Sciences, 1 (2/3), 273-

318.

Schank R.C., 1982. Dynamic Memory: A Theory of

Learning in Computers and People. Cambridge

University Press.

Schank R.C.,1986. Explanation Patterns: Understanding

Mechanically and Creatively. Lawrence Erlbaum

Associates, Hillsdale, NJ.

Wisniewski A.,1994. Erotetic Implications. Journal of

Philosophical Logic, 23, 173-195.

Wisniewski A., 1995. The posing of questions. Logical

Foundations of Erotetic Inferences. Kluwer Academic

Publishers, Dordrecht, The Netherlands.

Wisniewski A., 1996. The logic of questions as a theory

of erotetic arguments. Synthese,109,1-25.

Wisniewski A., 2003. Erotetic search scenarios. Synthese,

134, 389-42.

MODELLING THE MAN MACHINE INTERACTION - Erotetic Logic and Information Retrieval Systems

179