Constructing a Model of Dialog 
Mare KOIT 
Tartu State University 
Tartu USSR 
Abstract 
In the present paper communicative cycle 
(CC) is considered as a dialog arising 
between two partners during solving a 
problem. Communicative strategy (C5) is 
seen as a general scheme by which a dialog 
participant is guided in achieving his goal. 
CC is described in terms of the goals of the 
participants and CS. 
I, Introduction 
Dialog is a bilateral process the par ~ 
ticipants of which communicate each other 
messages concerning the surrounding world. 
Each participant interchangingly undertakes 
the role of the author or the recipient. 
For communication to be possible the commu- 
nicants must understand each other, i.e. 
%halt knowledge of language and about the 
world must have a shared part. Each partici- 
pant should also know something about the 
partner - his goals, knowledge, possibili- 
ties, etc. 
The task of a dialog model is a formal 
description of the process of communication 
or, in a narrower sense, description of the 
communicative structure of the text (speech) 
generated by the communicants in this pro- 
cess. Such description may be represented, 
for example, in the form of a generative 
grammar or graph /Robinson 1982, Metzlng 
1981/ and it can be realized on a computer. 
One of the aims of working out the dialog 
models is just to facilitate the communica- 
tion of man with the computer, bringing it 
nearer to man-to-man communication. 
Next we consider some problems arising 
in simulating dialog interaction. 
2. Communicative cycle and communicative 
strategy 
We limit ourselves only to treating 
such dialogs where the communicants (let 
332 
them be A and B) as co-operative partners 
solve some problems. 
The unit of simulation will be communi- 
cative cycle (CC). We define CC as dialog 
arising between two partners during solving 
a problem. 
CC consists of the alternate turn-ta- 
kings of the communicants though not every 
such sequence forms a CC. The turns are 
built up of communicative acts (CA). The 
minimal combination in a CC is a couple or 
triplet of turn-takings Of the participants 
(cf. /Langleben 1984/), for example A: ques- 
tion - B: answer, or A: question - B: answer 
- A: acceptance of the answer. Any CC starts 
with setting up a communicative goal by one 
of the participants (A). Let the goal of A 
be "B does a deed D'°. In natural communica- 
tlon achieving a goal requires more than one 
communicative turn-taking from both A and B. 
Typically, A has some obstacles in achieving 
D: e.g. B does not want to do D; B does not 
realize that not doing D involves negative 
consequences; B thinke that he is not able 
to do D, etc. For this reason A must, in 
addition to fixing the goal, also determine 
the resources he has at his disposal for 
overcoming these obstacles, and the way of 
using these resources - it means that A must 
outline the communicative strategy. 
Having fixed the goal G, A tries to 
direct the relevant intellectual and emotio- 
nal processes of B (his interests, assess- 
ments, opinions) during the process of com- 
munication in such a way as to eventually 
lead B to a decision to do D. In doing so 
influencing the partner is actually directed 
at certain aspects of his mind: his knowled- 
ge, assumptions, assessments, wishes. These 
relevant aspects of the mind of the partner 
may be called his psychic parametrss, in 
short P-paramstres. The decision of B to 
perform D depends upon the concrete confi- 
guration of the values of his P-parametreso 
The unfavourabls values are %he obstacles 
mentioned above. The essence of applying a 
~t~ategy lies in the fact that A organizes 
his t~iPn-takings to B in such a way that 
their r~)sult6 will be a change in the values 
of fi~)d P-parametres in a required direc- 
tion: increasing his knowledge, changing his 
assessm~3nts, increasing his conviction, Pe- 
ducing or increasing his wishes, and so on. 
Within ;1 fixed strategy certain P-parametres 
are chosen which can be influenced, the 
di?ectlon and ways of influence (e.go stra- 
tegies of threa%enlng, scaring, tempting, 
convinc:Lng~ etc)~ With respect to the prima- 
ry comlaunicatlve goal the changing of the 
values of P-parametres may be called "in- 
strumental goal". 
The development of the theory of com- 
municat~ive strategies thus presupposes the 
developr~ont of a theory of the P-parametres. 
I% woui~d be part of a certain qualitative 
theory of declslon-making and should, in 
addition ±o explicating the system of P- 
parametres, explain also the influence of P- 
paramet%'es upon the decisions made by a 
person. The factors that must be taken Into 
accoun± in shaping a decision ape numerous. 
But it turns out that thls diversity can be 
reduced to a limited number of p~imary paPa- 
metres. We can bring forth the following 
types of P-parametres: assessments (rational 
evaluations and those of pleasantness); 
knowled~e (also skills, experience); into- 
rests; requirements (wishes, needs, among 
them a need fop communication). 
Th~ explication of all possible P- 
parametl'es requires additional inestlgations 
(first of all psychological). In a model of 
dialog the P-parametres of a communicative 
partner mare up one part of the model of 
partner, During communication some of these 
values ~tay undergo changes. 
3. Simulating the communicative cycle 
First of all let us describe the pro- 
cess of formalizing the notion of communica- 
%ire stt,ategy (CS). 
CS may be defined as a procedure which 
deformities the choice of an author's com~ 
municatJve acts. This choice need not be 
unique. In this respect CS differs from 
strategy in the game theory Or from a plan 
of communication. 
The author A has fixed his communlca= 
tive str.ategy if he 
i) \]=as established the set of P-parametres 
of the partner B relevant 'to his communi- 
cative goal G 
it) knows in case of every relevant P-para- 
metre whether in order to achieve the 
goal G it is necessary to increase or lower 
the required value 
iii) has determined among the relevant P- 
parametres one parametre, changing the 
value of which is the Immedla%e goal G 
of implementing the strategy 
iv) knows in case of every relevant P--para- 
metre how to change its values either 
directly with the help of certain CAs, or 
indirectly, by changing the values of other 
P-parameires. 
Let us consider one example of CS. Let 
the goal of %he au±hor A of the strategy be 
G z "B performs a deed D". 
(I) The relevant P-parame%res of B: 
a wish to do D 
assessment of positive consequences of D 
assessment of negative consequences of D 
knowledge of the fulfillment of the 
preconditions of D, 
etc. 
(2) Let A know that B's wish %o de D is 
lower than required for doing it. 
(3) Let the P-parametme A wan±s to change be 
B's wlsh to do D (which needs increasing). 
This is the immediate goal G i of the stra- 
tegy. 
(4) P-parametres "assossmen±" and "knowled- 
ge" can be influenced d~rectly by such com- 
munlcative acts as "giving information", 
"explanation", "substantiation". 
The P-papametre "wish to do D" can be 
influenced only indirectly, through some 
other P-parametres, e.g. bF increasing B's 
assessment of positive consequences of D, by 
lowering B's assessment of negative conse- 
quences of D, by increasing B's knowledge 
about the fulfillment of preconditions of D. 
The goal G 1 of applying this strategy 
is to increase B's wish to do D. This may 
proceed in different ways (by setting up 
some subgoals and applying corresponding 
suBstrategles): 
A may increase B's assessment of the posi- 
tive consequence of D (by giving information 
of them, substantiating and explaining), 
i.e. to allure, cajole; 
A can lower B's assessment of negative con~ 
sequences of D (by giving information, 
555 
substantiating and explaining their bad- 
heSS), i.e. to warn, threaten, scare: 
A can increase B's ~nowledge about the 
fulfillment of the preconditions of D (e.g. 
by informing B about the availabilitF of 
resources required for doing D, explaining 
it), i.e. tO convince. 
The communicative strategy does not 
de±ermine uniquely the way of achieving a 
goal (the sequence of communicative acts of 
the partners) but leaves the freedom of 
choice. Constructing the next turn depends 
among oiheP things upon the changes which 
have been brought about in the partne~°s 
model by the information received during 
interpreting the previous turn-takings of 
the partner (for eMample, it may happen %hat 
some assumptions of the author about the 
partner do not hold). 
The communicative cycle proceeds as 
follows (participants A and B, A is the 
initiator of communication, G is the com~ 
municative goal of A). 
In accordance with special rules of 
interest /Saluveer, elm 1986/ A finds the P- 
parametres of B relevant to the goal G. On 
the basis of the partner model A determines 
the necessary changes in the values of these 
parametres. Next, A chooses a relevant P.- 
parametre which value needs changing. In 
±his way he chooses his communicative s±ra- 
tegy the goal G i of which is changing the 
value of the chosen P-parametr~ (in such a 
~ay as is ~equired for achieving the goal 
G). In case of every F-pa~ametre A knows 
either the list of CAs or the list of other 
P-parame%res through which i% is possible to 
increase or lower %he value of the chosen P~ 
parametre. 
In order to achieve the goal G 1 A has 
therefore either to choose a CA from a given ~ 
list or determlne a new P-parametre (which, 
in its turn, determined a new communicative 
stra±egy which is par% of 'the initial CS) 
and mare up the next turn-ta~ing. 
Partner B whose goal does not need 
coinside with the goal G determines, accor- 
ding to %he rules of interest, th~ P-pare- 
metres of partne~ A which are relevant to 
his goal. On the basis of the model of A 
(i.e. his partner model) B determines %he 
necessary changes in these pa~ametres. B 
a\]so fiMes his communicative st~ategy~ a~ 
did A. 
334 
In interpreting th@ "titian of %h~ pA~ri.~~r 
bo~h communicant~ car'~y o~t. ±.tic n~c,(-~E~a~-$~ 
changes in their paY~%~le~ ' mod~Is and d~t~id~ 
whether their initial goal or ah i~a~% aortae 
of its subgoals hays bs~n achit~ved., If 'thie 
is not the case then oiths~, a new subgoai la 
set up (or a former subgoal is ma%J.:~vtaihed) 
and a new 'turn--ta~ing i~ mad~ %o ac}~/i.ove i%,. 
or th-~ initial goal is given tlp, Colimlunit:a-. 
tion proceeds until the ini'bia¢'toP of "th~:~ 
communication cycle A has achieved hlg goal 
or given it up, 
Th6~ model outlined abov(~ und~grli(~ i.h~ 
pi 1 ot human--comput~ ~ d i al og Sys tGm TA~LIIS 
/Koit~ Saltlc~ce'r 1986/~ The vt~'glon of %h<% 
system which has been ~(~alizsd up to now (o2% 
}~yad 2 and SM computers ) th~ t~ys't¢9~u can 
perform morphological, syntactic and s£~man- 
"tic analysis and generation of Es'toni an 
'text~ In addi%io~l "to that the) v~'~,ion o:d 
'the SM col0putor can recogniz~ in a text th¢~ 
description of such criminal eve~~%i~ a~ 
theft, robbery, etc., and answer' question, s 
about that tent in the Estoniai~ languagoo 
References
Koi't M°, Saluveer Me, Generaring Natu~,~l 
Language Text in a Dialog S~'tem° -, Frooo 
COLING'86, Bonn, 1986, 576.-o580o 

Langleben M., On the Structure of Dialoguoo 
- Mic 'o and Macro Conne~ity of Texts. Ed° 
by J. Pot6f i, E. S6ze~'. Hambn~'g ; Ht-~ I ~ut 
Bus~e, 1984, 220~2B6. 

Metzing Do , Zur Entwick Iung prozedural G~ ~ 
Dialogmodelle. ~ Dialogmus%e~ ~ un~ Dialog ~" 
pr'ogesse. Hg~ D.Meizing~ Hambu~'g: Hslmut 
Buske~ 1981, 51.-72. 

l~obinsen JoJ., DIAGRAM~ A 6x~am~a~ ' for Dialo 
gu~s, ~ Cosec of %h~ AC~4, 1982, vole 25~ ~ 
I~ 27~.47 

Saluveer M.,~Jim Ho, ~ule~ and i%easoning in 
'I'e~t; Compi-ehens i on. -' NSU~ An~tze i~ 
Maschlneller Spl~ach~bo~'sstzung: Wis~sngx'~

HoJoWeber (Hggo). T~bingen: }~'~g Niemaye~%. 
1986, 139=163 
