The Implementation of a ReALIS-based Method of Static Intensional

Interpretation

Márton Károly

and Gábor Alberti

”Science, Please!” Project Office, University of Pécs, 4 Vasvári Pál Street, Pécs, Hungary

Department of Linguistics, University of Pécs, Pécs, Hungary

Keywords: Extensional and Intensional Predicates, Static and Dynamic Interpretation, Modeling Human Interactions.

Abstract: This paper is grounded in the joint work within our project to implement a static (and, later, dynamic)

interpretation of sentences, using the theory of eALIS. It outlines definitions and implementations of

truth-evaluating extensional and intensional predicates, with an aim to present the Prolog-based core (i.e. the

future server-side) of the program. This program has been partially demonstrated in recent publications; it is

now mature enough to have a (presumably Java/JSP-based) graphical interface deployed to it. In this paper

we still use a bottom-up approach because the theoretical complexity of eALIS does not concur with a too

early finalization of the architecture. Most notably, the world model data in eALIS, equipped with the

semantic postulates, share a great many features with the program code itself. This causes problems when

planning the user interface. As for future work, the most important goals are, on the one hand, to implement

verbal semantics (while sticking to the bottom-up approach) and, on the other hand, to finalize the

architecture, the use-cases, the components and the deployment (switching to a top-down approach).

1 INTRODUCTION

eALIS is a post-Montagovian theory which

concerns the formal interpretation of sentences

constituting coherent discourses (Asher and

Lascarides, 2003) and which uses an extension of

Kampian DRT (Kamp et al., 2011). As stated in

Alberti and Károly, 2010, eALIS provides a

theoretical background to represent the contents of

the human interpreters’ minds, including their BDI

(belief, desire and intention, see Vadász et al., 2013),

background knowledge and pragmatic relations –

beyond the actual contents and structure of the

discourse. eALIS also overcomes the problem of

representationalism (see Alberti and Károly, 2011)

by embedding the interpreters themselves into the

world model, along with their mental states.

It has become clear over the past few months that

a pure NL-based input should be replaced by an

extendable (and replaceable) language model, based

on a context-free grammar. Using this, it is possible

to build a eALIS-based model generator by which

the -labels of worldlets (definition: Alberti, 2011,

pp. 139–177, see also Alberti and Károly, 2012) are

assigned to linguistic elements.

This paper describes the problem of switching

between extensionality and intensionality (in

eALIS, W

and W[i,t], see Alberti and Károly,

2010), and it also offers a partial solution to it:

predicates can be grouped by how they are related to

the outer world W

and to the internal W[i,t]

worldlets of the interpreters. Predicate types can be

regarded as semantic postulates which are, in fact,

defined in the oracle. Unlike any represented human

interpreter, the oracle has no BDI; but it still

contains semantic postulates and logical rules in

non-BDI worldlets, in order to define intensional

predicates). Code fragments of actual and possible

implementations of certain predicate types are

shown in Sections 2 and 3 where we use a greatly

simplified approach to NP-anchoring which is,

indeed, at least in theory, refined in Section 4.

The planned architecture of the eALIS

software is summarized in the last section, while

possible applications of eALIS in NLP-based

expert systems were shown in Alberti and Károly,

2010. Most importantly, eALIS-like theories and

software based on them could be used in the

jurisdiction: by investigators, judges, prosecutors

and barristers.

393

Károly M. and Alberti G..

The Implementation of a ReALIS-based Method of Static Intensional Interpretation.

DOI: 10.5220/0004627503930398

In Proceedings of the International Conference on Knowledge Engineering and Ontology Development (KEOD-2013), pages 393-398

ISBN: 978-989-8565-81-5

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

2 LINGUISTIC AND MODEL

(LEXICAL AND SEMANTIC)

GENERATION

In its current stage of development, the linguistic

generation is CFG-based so that we can overcome

the difficulties of linguistic acceptance. The Prolog

facts and data structures look much like the ones

sketched in Alberti and Károly, 2011. Linguistic

input is provided by applying the appropriate rules

of the CFG. Possible - (worldlet) labels are

assigned to certain modal lexemes and morphemes,

or, in the case demonstrated here, phrases like D

thinks (S):

Example 1: Linguistic generation (see CFG below) and

truth evaluation of the extensional predicate to be

married. The presence of the Prolog facts below (see code

fragments) is checked in the Prolog factual database.

S’  S | D think(s) S’ | according to D S’ | …

S  D am/are/is married | D am/are/is pretty | …

D I | you | Mary | Joe | Peter | the/a boy | …

(the rules are triggered by the interface)

EGO (entity no.1.) talks to YOU (entity no.3.),

NOW (20130515 as date), HERE (Budapest) are all set by

the user.

Sentence: Peter is married.

referent(102,ref,r,’Peter’).

lambda(102,ref,1,[]).

%root worldlet (label []) of EGO

%EGO knows a certain Peter

alpha(102,ref,2,ent,out,gest).

%referent 102 is out-anchored to

%entity 2 by gesture

external(2,ext,ent,[’Peter

Smith’,i,19790201,20601010]).

%external/4: ID,two type args,

%parameter list depending on type

%(name,i=interpreter,birth,death)

lambda(302,ref,3,[]).

%ref. 302 in the root worldlet of YOU

alpha(302,ref,2,ent,out,gest).

%YOU also knows ’Peter’ whose referent

%is anchored to the same entity

referent(115,ref,p,’married’).

lambda(115,ref,1,[]).

%’married’ is an extensional predicate

%it has an external equivalent to

%which it is alpha-(out-)anchored:

alpha(115,ref,5,crel,out,gest).

external(5,ext,crel,’married’).

%extensional predicates are anchored

%to core relations (crel)

external(5,ext,inf,[[5,crel],[20101121,

20140515],[2,ent]]). %arg=[2,ent]

%parameter list of infons: core

%relation,[time interval], 1 or more

%entities as arguments]

%inf=infon (SSIS only, see Seligman–Moss, 1997):

%YES, Peter is married. Without the

%last external fact (cf. if the last

%check failed), the answer would be NO

Examples 1 and 2 also demonstrate how

anchoring is checked before truth evaluation (see

Section 4). As noted above, the values of the 

cursor function in eALIS (Alberti and Károly,

2011) i.e. speaker (I/EGO), addressee

(YOU/Hearer), time (NOW) and location (HERE) –

must be preset before the actual analysis. This is

necessary not only for the adequate handling of

pronouns and/or interpersonal knowledge, but also

to uniquely anchor the referents denoted by the topic

of the sentence (here: Peter) – since the name Peter

is not unique on its own. (The name Peter Smith is

considered unique.) This means that, in real life, the

hearer could ask the question: Which Peter? Taking

all this into account, semantic analysis has two

phases: the anchoring phase, in which pre-

suppositions are checked (if any of these checks fail,

the sentence is neither true nor false, e.g. The

reigning French king is bald), and the actual truth

evaluation. Anchoring always starts from the actual

-level of the discourse (here, it is the root worldlet).

Referents of other worldlets can also be used

(obeying the “up or left” rule of Kampian DRT), but

the truth values of the predicates must be checked

from the viewpoint of the outer world, stepping into

-worldlets if needed. In order to achieve this, it is

best to create a special interpreter – an oracle. If a

certain -worldlet does not exist, its creation is

simply done by asserting new

lambda predicates).

Example 2: Think is an intensional verb, it turns off the

extensional checking of Peter is married. In this case, only

the presence of an internal eventual referent is checked

(and not that of an infon).

Sentence: Joe thinks Peter is married.

%Joe is entity no. 4.

referent(104,ref,r,’Joe’).

lambda(104,ref,1,[]).

%root worldlet (label []) of EGO

%EGO knows a certain Joe

alpha(104,ref,2,ent,out,gest).

external(4,ext,ent,[’Joseph

Taylor’],i,19770311,20591112]).

lambda(304,ref,4,[]).

%root worldlet of YOU

alpha(304,ref,4,ent,out,gest).

%YOU also knows ’Joe’

referent(102,ref,r,’Peter’).

lambda(102,ref,1,[]).

KEOD2013-InternationalConferenceonKnowledgeEngineeringandOntologyDevelopment

394

%root worldlet (label []) of EGO

%EGO knows a certain Peter

alpha(102,ref,2,ent,out,gest).

external(2,ext,ent,[’Peter

Smith’],i,19790201,20601010]).

lambda(302,ref,3,[]).

%root worldlet of YOU

alpha(302,ref,2,ent,out,gest).

%YOU also knows ’Peter’

%Joe, too, has to know the same Peter

referent(402,ref,r,’Peter’).

lambda(402,ref,4,[]).

%root worldlet (label []) of Joe

%EGO knows a certain Peter

alpha(402,ref,2,ent,out,gest).

external(2,ext,ent,[’Peter

Smith’],i,19790201,20601010]).

referent(998,ref,e,[[410,ref,p,

’think’],20130515,[104,ref,r,’Joe’],

[999,ref,e,null]]). (…)

referent(999,ref,e,[[415,ref,p,

’married’],20130515,[404,ref,r,’Peter’

]]). %eventual referent

lambda(999,ref,4,[[bel,med,20130515,+]

]). %lambda-equivalent of to think

%YES, Joe thinks Peter is married.

It is very important to note that in Example 1, the

eventuality of Peter is married is not necessarily

present in the mind of the speaker or the hearer, or at

least not with a positive polarity (

referent(

…,ref,e,…)

is not listed because it is not checked

at all). The description of lies, bluffs, fibbing or

killing the joke equally involve a change in the

polarities of the worldlets to negative, zero or even

‘non-zero’ (e.g. for yes/no questions, see Kilián,

2013 and Vadász et al., 2013). The hearer also has

“free choice” to believe or to not believe the

sentence: (s)he can move, or more precisely,

accommodate the eventual referent to any of his/her

internal worldlets of belief.

The speaker, too, can have the eventual referent

in his mind and he can refer to it: Peter is married

and this is what makes Mary sad. But the question of

having or not having the eventual referent in one’s

mind does not necessarily depend on the truth value

of Peter is married. In the Middle Ages, the fact that

many people bore in their minds the eventuality of

Susan is a witch often caused dire consequences…

The introduction of Joe thinks results in

beclouding the actual truth value of Peter is married.

Private intensional predicates like think, believe,

desire, want, lap up as well as most modal verbs do

not have external equivalents so no extensional

check is performed (apart from checking the

uniqueness of Peter and Joe in the given context).

Some possible types of predicates are listed in

Table 1: they are assigned to modal particles,

intensional verbs, adjectives etc. The semantic

postulates by which these types are defined in the

oracle and which roughly correspond to the “code

segment” of a truth-evaluating program (functioning

in a different way for each type, after checking the

predicates – in a similar way as in Examples 1 and

2) no longer belong the linguistic generator but to

the model generator. The CFG (the linguistic

generator) is used as input, and this invokes the

model generator to determine truth value (or to not

determine it in case of anchoring problems – e.g.

lack of uniqueness, anchoring mismatch, or falsity of

the presupposition).

In the near future, the fact database will be

extended to a real model generator in which the

former database will actually operate as a “data

segment”: internal users (administrators, linguists)

will be able to load the world model, including the

interpreters’ simulated brains with data – by using

the (web) interface.

Summarizing the above, the program is

constituted of two parts connected to each other:

Figure 1: The linguistic generator and the model generator

– corresponding to the linguistic form and the meaning of

words.

Polysemy and synonymy are depicted by the

crossing arrows in Figure 1: if we exploit the

backtracking features of Prolog, it can result in

multiple solutions and/or the same solution for two

or more different inputs.

3 POPULATING THE

LINGUISTIC DATABASE.

PREDICATE TYPES

Linguistic elements are inserted into the CFG as

possibilities. Nouns, adjectives and verbs are

inserted one by one or in groups into the grammar

by internal users, applying

assert on the Prolog

server. (The CFG should be extended to an indexed

grammar which is more suitable for morphologically

rich languages like Hungarian or Turkish.)

The real problem is to determine the types of the

certain predicates (adjectives and verbs) and to

invoke the model generator to extend the world

model. Extensional predicates have external

LINGUISTIC

GENERATOR

MODEL

GENERATOR

TheImplementationofaReALIS-basedMethodofStaticIntensionalInterpretation

395

equivalents in the (Montagovian, see Dowty et al.,

1980) outer world, which causes their semantic

postulate to establish at least one -anchoring into

the outer world (i.e. a referent is anchored to an

entity and not to another referent) and also to

include a definition of a core relation in it (see

definitions in Alberti and Károly, 2010, 2011 and

the full description in Alberti, 2011). Intensional

predicates, however, have no external equivalents.

Instead, their semantic postulates include creating

and/or checking certain - (modal) labels or more

complex intensional structures – such as checking

the mental states of more than one interpreter. One

such example is be pretty – see Table 1. The heads

of the checking predicates are mostly omitted

because they are taken out of context.

Table 1: Predicate types and their definitions / truth value

evaluations.

Predicate type, definitions/checks Examples

Extensional: e

married

: p

married

t r

core relation is defined in the outer world

external(RID,ext,crel,NAME),

with infons:

external(ID,ext,inf,

[[RID,crel], [TIME_BEGIN,

TIME_END]|ARGUMENTS]) etc.

married

Ukrainian (as

citizen)

to swim

anchorings:

p  c. rel.

e  infon

Private intensional: e

believe

: p

believe

t r

no core relation, one or more -relations in

the mind of a certain interpreter

referent(EBL,ref,e,[[BID,ref

,p,’believe’],TIME,R1,E2])

lambda(EBL,ref,

INTERPRETER_ID,[[bel,

STRENGTH,TIME,POLARITY]]),

only the presence of a certain eventuality

EID ( identifier of e’) is checked

referent(E2,ref,e,

[[PREDID,ref,p,PRNAME],TIME

| ARGUMENT_REFERENTS]) etc.

argument referents are inferred from the

results of anchoring checks

to believe e’

to think e’

to desire e’

to search

(for r)

e: eventual

referent

(described by

subordinate

clause or

similar)

Quantified intensional: e

pretty

: p

pretty

t r

No external equivalents. Example: Mary is

retty. After a check for the uniqueness of

Mary (e.g. ‘Mary Johnson’; let it be

IENT), anchoring to IENT is checked for

every interpreter. Then we take the set

IIDLIST of those who anchor a referent

to the same IENT entity (know her).

If, for example, more than 2/3 of

IIDLIST think (s)he is pretty (and have

retty as a predicate name, this means they

know what pretty means), the predicate is

considered to be true.

quantcheck(PRNAME,IENT):-

beautiful

pretty

ugly

(PRNAME)

neither e nor

p has an

external

equivalent

Table 1: Predicate types and their definitions / truth value

evaluations (cont).

etof(ENTID,

uantinte

(IENT,

PRNAME,ENTID),PILIST),

(those who think IENT is PRNAME)

findall(IID,(alpha(REFID,ref

,ENTID,ent,out,_),lambda(REF

ID,ref,IID,[])),IIDLIST)(…)

(those who anchor some referent to IENT

– not necessarily with the same name)

Then, the length of PILIST is divided by

the length of IIDLIST.

The details of quantcheck are omitted.

Mixed intensional: e

bald

: p

bald

t r

Peter is bald: Same as quantified

intensional but with an external

representation of the predicate as

prototype. The eventuality may or may

not have an external equivalent (non-

prototypical case). Not only ‘Peter’ but the

predicate ‘bald’, too, must be unique and

common. The eventuality of ‘Peter is bald’

may or may not be alpha-anchored to an

infon. In the second case, both p and r

have to be anchored correctly to grant the

interpreter the “right to vote”.

bald

p  c. rel.

is e anchored

to an infon?

yes 

extensional

no 

quantified

intensional

(with voting)

Intensional naming:

see anchoring of Peter and Joe in

Examples 1-2. Similar to pronouns.

alpha(ID,ref,TO_WHAT_ID,ent/

ref,arg/pred/ant/adj/out,

cat/ord/agr/hum/gest etc.)

is checked, see also Section 4.

nicknames

Peter

 ‘Peter

Smith’ etc.

Arguments written in lower-case letters like ref,

ent

etc. are type and subtype names for referents,

entities, linguistic (

category, order etc.) and extra-

linguistic (

gesture) anchoring categories.

3.1 The “Already given” Extensional

Predicates

The seemingly simplest predicate type consists of a

predicate referent which is located in the root

worldlet and is

alpha(ID,ref,CRELID,crel,

out,_)

-anchored to a strictly homogeneous core

relation in eALIS (this is called out-anchoring),

but where, unlike in many ontologies, the arity of

core relations and predicates is not limited.

All entities of the outer world are defined with

external predicates (cf. referent in the internal

world[let]s): entities, core relations, infons, and,

later, time intervals.

Core relations consist of infons of Seligman and

Moss (1997) but only simple infons are permitted.

They also contain discrete time intervals. Each

KEOD2013-InternationalConferenceonKnowledgeEngineeringandOntologyDevelopment

396

element of a core relation is described by one or

more infons, depending on time. The last, list

argument of an

external predicate describing an

infon always has two further elements in addition to

the arity of the core relation to which the infon

belongs. These predicates are “already given”

because their existence depends neither on the

language, nor on the interpreters (e.g. to be married

is strictly defined by the law).

3.2 Extensional Predicates in an

Intensional Domain

Let the interpreter be i. “Intensional domain” means

that referents are in any of the fictive worldlets of i,

so to say, not in the outer (real) world. Infons turn

into eventual referents when they are mentally

depicted in any of the fictive worldlets. In theory,

unlike infons, the structure of eventual referents (e)

can be arbitrarily complex because the arguments of

intensional and modal verbs/particles (see Table 1)

are often eventualities themselves (e’). But because

this also implies a -level switch from w to a new w’

worldlet, e’ only has to be explicated in w’: a

subordinate clause r

believes e’, more formally

believe

: p

believe

(t,r

,e’) means that, apart from the

presence of e

believe

in w, only the presence of the

eventual referent e’ (with the appropriate contents)

should be checked in i’s w’ worldlet of belief (

bel

is present in its -label with positive polarity). Being

an argument of e, the internal structure of e’ is

omitted in w (see

null in referent 999, Example 2).

For non-positive polarities see Vadász et al., 2013,

e.g. the semantic postulate of the verb lie only

contains a simple polarity check.

3.3 Predicates with a Purely

Intensional Definition. Quantified

Intensionality

Similar to modal and intensional verbs and

adjectives, pretty has no external equivalent because

the actual meaning of ‘prettiness’ greatly depends on

who is said to be pretty by whom.

Indeed, the truth evaluation of pretty may be

done by simply saying “pretty are those who are said

to be pretty by most people”. This can be used as a

semantic postulate, making a predicate type on its

own. Even this type is already implemented with

relatively small simplifications.

Let us truth-evaluate Mary is pretty. The minds

of all interpreters in the universe of eALIS are

searched for the predicate name pretty. Then, the

entity which the referent named ‘Mary’ refers to (in

the speaker’s mind or in the common knowledge of

the speaker and the hearer, see Section 4) is taken

and all referents referring to it are collected, along

with the interpreters themselves. After that,

“thinking that Mary is pretty” – a belief, see Table 1

– is checked in all interpreters with such referents.

Then, the number of interpreters who think Mary

is pretty is divided by the number of interpreters

who actually know the same Mary as the one

(unique!) present in the common knowledge of the

speaker and the hearer.

3.4 Hidden or Mixed Intensionality

The difference between bald and pretty is that the

former actually has a prototype (when somebody

has no hair) – this counts as an external equivalent.

As such, a core relation (Z) should be defined to it

which may even be empty if we assume that

“prototypical baldness” is very rare. Z is just a

common ancestor of all predicate referents of ‘bald’

in any interpreter. Even those should know the

prototypical meaning of ‘bald’ who use it for people

who actually have some hair. Only these interpreters

should have the “right to vote” in the way described

in 3.3. Truth evaluation of Peter is bald is a

combination of 3.1 and 3.3: if someone is bald

according to Z, it is true. If not, interpreters who

know “prototypical baldness” and (a certain) Peter at

the same time, “vote” for or against the truth.

There are many more predicate types which are

yet to be implemented: the most complex verbs have

five phases (preparatory phase, starting point,

cumulative phase, cumulative point, result phase)

with five different semantic postulates. For example,

both the preparatory and the result phase of the VP

to fly home lasts at least some hours. In the

preparatory phase, everything is intensional: the

eventual referent is in the worldlet int

max

,i,,+.

Moreover, i is most likely to be preparing for his/her

journey: buying the tickets, packing, and the like.

One assumes that, in the result phase, i is

probably at home. But this probability decreases

over time: the result phase is much like some kind of

“limitation period” – if we borrow this expression

from the legal terminology. But this belongs to the

dynamic interpretation of eALIS which is yet to

be implemented and researched. The preliminary

results are expected to be published very soon.

4 ANCHORING NOUN PHRASES

Let s be the speaker and h be the addressee. Before

TheImplementationofaReALIS-basedMethodofStaticIntensionalInterpretation

397

any truth evaluation, the topical parts – mostly NPs

– of the sentence (such as Peter, the boy etc.) must

be anchored to existing entities. Here, only the

definite case is described. (For the indefinite –

specific or non-specific case, relevant sets have to be

extracted from the context, which is not yet possible

at the current stage of development.)

In most cases, uniqueness is needed to properly

interpret the sentence. (Nick)names, like those in

Examples 1 and 2 are the best examples to

demonstrate this. Therefore, let us again use the

sentence Peter is married to illustrate the case.

Let us take four sets of Peters: P

to P

. P

contains the entities known by s, the elements of P

are the ones known by h, P

and P

are assumed sets:

s believes that h knows the elements of P

and h

believes that s knows the elements of P

To be pragmatically correct, |P

P

| and |P

P

should be 1 and the two entities must be the same. If

this is not true, uniqueness is not guaranteed from

either the speaker’s or the hearer’s side.

Of course, in this case, uniqueness can be

inferred from a wider context: Peter has died. Both s

and h may know many Peters but it is only one

“common Peter” who actually died: both s and h

might have known which Peter that was. Although

Prolog is capable of performing even this task, it has

not been implemented yet (we are assuming strict

uniqueness), and also, it would slow down the

program considerably.

5 PLANNED ARCHITECTURE

Prolog has two interfaces to Java: PrologBeans and

Jasper. Since the prototype is mature enough for the

Prolog core and the (future) interface to be separated

and since this will render it very important to

implement a multi-user interface (for internal and

external users), we are considering building a web

application from eALIS, skipping the phase of a

stand-alone graphical application. Moreover,

because Jasper is only suitable to create stand-alone

applications, PrologBeans will be used as an

intermediate layer between the Prolog server and

Java. Communication between PrologBeans and JSP

is also quite well documented, so it seems possible

to build two web-based interfaces for eALIS: one

for internal users (linguists and administrators) and

one for external ones. Only internal users would

have the right to add new linguistic elements and

new semantic postulates.

Even later, the Prolog core might be extended

with an SQL background to handle large databases.

Although we have experimented with this, the actual

implementation will greatly depend on the memory

limits of SICStus Prolog and the actual memory

consumption of the program.

ACKNOWLEDGEMENTS

We are grateful to SROP-4.2.2.C-11/1/KONV-2012-

0005 (Well-being in the Information Society) for

their financial contribution to eALIS.

REFERENCES

Alberti, G., 2011.



eALIS. Interpretálók a világban,

világok az interpretálóban [Interpreters in the world,

worlds in the interpreter], Academic Press. Budapest.

Alberti, G., Károly, M., 2010. The eALIS Model of

Human Interpreters and Its Application in

Computational Linguistics. In Proceedings of the 5th

ICSOFT, Athens, Greece Vol. 2, SCITEPRESS. 468–

474.

Alberti, G., Károly, M., 2011. The Implemented Human

Interpreter as a Database. In Proceedings of the 5th

IC3K-KEOD, Paris, France, SCITEPRESS. 379–385.

Alberti, G., Károly, M., 2012. Multiple Level of Referents

in Information State. In Computational Linguistics and

Intelligent Text Processing – 13th International

Conference, CICLing 2012, New Delhi, India,

Springer. Berlin.

Alberti, G., Kleiber J., 2010. The Grammar of eALIS

and the Implementation of its Dynamic Interpretation.

In Informatica Vol. 34. No. 2. 103–110.

Alberti, G., Kleiber, J., 2011. Where are Possible Worlds?

(Arguments for eALIS) In 4

Syntax, Phonolohy

and Language Analysis Conference. Budapest.

Asher, N., Lascarides, A., 2003. Logics of Conversation,

Cambridge University Press. Cambridge.

Dowty, D. R., Wall, R. E., Peters, S., 1980. Introduction to

Montague Semantics, D. Reidel Publishing Company.

Dordrecht.

Kamp, H., van Genabith, J., Reyle, U., 2011. Discourse

Representation Theory. In Handbook of Philosophical

Logic, Vol. 15, Springer. Berlin. 125–394.

Kilián, I., 2013. A Metamodel-Driven Architecture for

Generating, Populating and Manipulating “Possible

Worlds” to Answer Questions. In Proceedings of the

ICSOFT, Reykjavík, Iceland, SCITEPRESS. To

appear.

Seligman, J., Moss, L. S., 1997. Situation Theory. In

Handbook of Logic and Language, Elsevier and MIT

Press. Amsterdam and Cambridge. 239–309.

Vadász, N., Alberti, G., Kleiber, J. 2013. The Matrix of

Beliefs, Desires and Intention – Sentence by Sentence.

In International Journal of Computational Linguistics

and Applications Vol. 4 No. 1. 95–110.

KEOD2013-InternationalConferenceonKnowledgeEngineeringandOntologyDevelopment

398