ADAPTIVE DIALOGUE - THE BASIS FOR PERSONAL COMPUTER SYSTEM 
Victor Briabrin 
Computing Center, Academy of Sciences, Hosoow, USSR 
1. Personal Computer Systems (POS) represent nowadays a 
signifteaut trend in the professiona~, and amateur use of 
computers, in particular of mini- and microcomputers° We be- 
lieve that such systems must have the means for: 
(a) knowledge representation about the problem donm/n, the 
system itself and the partieular users; 
(b) man-n~chine interaction, taking benefits of both partners" 
abilities; 
(c) access to application-oz-lented programs and data as well 
as to other computer systems° 
Our experience with natural language question-anewer£n8 
systems has proved that the share of. QAS users preferr~n6 real 
RL is rather small. Among the reasons one can point out such 
features as inadequate heaviness and slowness of NL-proeessors, 
their unportablltty. These reasons, on one hand, and the appar- 
ent success of simple table-driven and production rule systems, 
on the ether, have led us to the development of Adaptive Dial- 
ogue System as a basis for PCS, 
Our approach is based on the integration of 3 core 
components allowing an efficient system implementation within 
the limited micz~computer resources: 
the Object Base (OB) constitutes a media for storing and 
manipulation of systems ° , problem-oz~lented and lexical 
obJects; 
- the Adaptive Dialogue Monitor (AI~) provides the guidance 
- 55 - 
for man-computer comnm£oation through die~ogue processes! 
- the P-4kach£ne (I~) accomplishes the necessary operations 
supporting system s own beha~ou~ as well as the access to 
applied programs and data. 
process in the system takes place within a cert~ 
qn~ronment which is characterized by a set of mutu~ly 
accessible objects and by initial dialogue process. A dialogue 
process or D-urocess comprises a number of interconnected 
states and supports particular man-computer activity on 
achieving some goal. Almost every stat~ of D-process acoompl- 
ieh~ a e~mple act of com-mn£cat~ with the user, an~yzJ,ug 
hie reply, performing associated actions and transiting to 
another state. Thus an analo~r could be drawn between D- 
-process and ATN, with the important difference that D-process 
supports oneoin8 d£e~ogue with the user rather than an~yeis 
of preentered st~-Lnee 
There are ) state types: 
MENU - promptin8 the user to choose among e~ernativee! 
A3EVAL - accepting the asked Ye~ue from the user! 
ZVAL - getting computed value from external'program, 
By means of MENU and ~KVAL D-process iS communicating 
with the user°s terminal, IVAL connects it with applied pro- 
8raew. The eotions with£n the states are expressed as. strings 
of ~ perfor~tJ~ predication of values, manipulation of 
different objects, call~Jae/return£ng to another D-processes, 
runnL~ extern~ programs etc. 
Adapti~rity is of prt~aa~ concern in the presented 
approach since it ~vee the user, on one hand, the ability 
to subject the system to his particular needs! on the other 
- to put himself into an active or passive position during 
some coa~mication process. Adaptivity is provided by strict- 
ly followin 8 one principlei any kind of system's acti~Lty 
pertinent to man-computer oonzm~Lcation must be controlled by 
definite objects whLch can be created and modified by the 
neme or other activities. 
- 56 - 
2. There are 3 basic klnds of objects in the OBs system 
support~ (S-objects), problem-oriented (PO-obJeote) and 
le~tcal (L-objects). Each of these kinds is further divided 
into categories or types. 
S-ob.teots hold the control over man, computer interaction 
and the ~ocess to applied progrems and dBta. They describe 
az~rth~ng residing or happezLing in the system and therefore ~re 
called "descriptors". The following descriptors are of the 
most t,nportance: 
DE - enwlronment descriptors 
DP - dialogue process descriptors . 
DS - dialogue state desoz~ptors 
DX - program descriptors 
DD - data descriptors 
DG - dialogue scenario' ~ ("game") descriptors 
DU - user descriptors 
PO-ob.tects describe concepts, properties, relations and 
values allowing to build problem-oriented models. PO-obJects 
are difterentiated by their categories (typ~s): 
C - classes of physical objects: "country", "person" 
R- se~ti@ relations: "negotiation", ',disarmament" 
V - structured value descriptors: "date", "NATO" 
T- terminal value descriptors: "ATOM", "STRING", "INTEGER" 
P - object properties substantiated by V and T - values~ 
"pop-lation ", "b~thday", "address", "weight" 
CP - object properties substantiated by the other C-obJeotss 
"capital" (of country), "relatives" (of person) 
RP - semantic relations ar~entss "participants" (of negotiat- 
ion), "place", "goal" (of any aotiwlty) 
CI, B.I, Vl, TI - individual objects and actual values - subst- 
antiations of C, R, V and T-objectso 
For external representation of objects we use a formalism of 
F-l~e developed for DILOS system /1/. S~nple exemples: 
oapltal (CP): IS olty; OF country 
- 57- 
politician (C): IS person; from country; post official-post 
official-poet (V): IS list; 0F ATOM; EL (president defense- 
minister...) 
NATO (V): IS list; OF country! EL (Belgium Gr-Britain...) 
Weinberger (CI): IS politician| from USA! post defense-minist- 
er 
defense-minister (TI): IS ATOMj ELOP official-post 
The problem domain model could be ~xpressed in a fore of 
conceptual semantic network /2/ with the nodes represented by 
C, R, V and T-objects and the arcs labeled by P, CP and RP- 
-objects° Representation of actual world fragments would then 
have a form of terminal semantic network with CI, RI, VI and 
TI-objects being the nodes, connected by P, CP and RP-arcs. 
L-obteots or "lexems" provide access to all other objects 
in the 0B using the words and phrases of problem-oriented 
natural language (I~OL). Each lexeme has external represent- 
ation (. the name of L-object) derived from the input word 
form by simple morphological transformation° Each L-object has 
special properties which serve for two purposes" (1) pointing 
to other objects - denotations of the given lexeme and (2) 
carrying ~ammatical marks for efficient POL analysis. 
Presently L-objects are divided into the following 10 
categories: 
LN - lexlcal names of PO and S-objects 
LS - synonyms polnting to other L-objects 
LC - components of word combinations and idioms 
LQ - question tokens - pronouns and adverbs ("who", "how") 
LJ- con~unctlons of different types ("and", "or", "but") 
LR- referenclal pronouns ("these", "his") 
LP - prepositions with indication of implied semantic oases 
for different relations 
LD - "deflnitors" establishing paradlgmatlo relations between 
objects (."is", "denote", "contatln") 
LA - articles 
X - ignored or unknown lexames 
- 58 - 
3o The system starts its operation by running initial 
D-process which, first, makes acquaintance with particule~ 
user and, second, prompts him with the main menu to choose 
anon8 several basic activities! 
(A) filling the 0B with the new objects or modifying the old 
ones ! 
(B) entering particular problem-solvlng task! 
(C) getting information on system components and processes. 
In (A) one might concentrate on creating the new dial- 
o~e schemata which is done by building new DP and DS-obJects. 
This special kind of activity results in a new communication 
lansuage (sublauguage) which might be immediately tested and 
used for practical needs. Another kind of (A) is connected 
with the creation of P0 and L-objects. The user can do it two 
different w~s8 
- by entering pattern-controlled expressions which define 
particular object categories| 
- by calling special D-pro0esses responsible for buildin 8 
appropz~lat • objects. 
Although from the user's standpoint these alternatives differ 
in the share of man-machine aoti~-lties, both of them are bas- 
ed on the same rules having the general forms 
<pattern> \[= ~condition>~ ~ actions 
~pattern~ defines the combination of object categor- 
ies accepted from the user, no matter if they come sequential- 
ly (one object per D-state) or simultaneously - as one line 
of text. Optional Zcondition~ allows checking particula~ 
properties of the accepted ob~ectso The ~ actions~ emend 
the old objects or create the new ones. A simple rule might 
look as follbwlng= • 
(X "0~ CI m~ C2)= ~t(X, "TP, "CP) 
Put (X, "OP, C1) l~t(X, "IS, C2) l~t(C1, X, C2) 
Such a rule can handle the expression: "president of country 
is a politician" which might be entered in one sentence or 
- 59 - 
spread over several ASKVALI. The result would be crention of 
new CP-ob~eot "president" and modifyi~ the C-object "country". 
In (B) the change of initiative can be softly controll- 
ed by the user in the sense that he should not apply special 
knowledge or feel any barriers when moving from one node of 
interaction to another. In any mode, however, & result of ADM 
operation is reflected in F-expression /2/ which carries the 
semantic meaning of the input sentence. 
Thus, for example, a phrase "Defense ministers of NAT0 
met in the capital of France on October 18th" could be trans- 
formed into the following F-expressions 
(Meetings participants (politician (s): 
post defense-minister 
from (country (s) = ELOF ~ATO)) 
place (capital = 0F Francs) 
time (date = M October D 18) 
Such an expression is directed to 1~ where it is interpreted 
with the result of creating/pointing at some nodes in the 
terminal semantic network thus representing the meaning of 
the entered text° 

References

/1/ Ne Asaf~eva, A° Borkovsky, Y. Briabrin, V. Ponon~rev, 
G. Senin. Knowledge Representation and Natural L~ge 
l~rooessing in DILOBo In "The Problems of Applied Systems 
Development". Ed. A. Narin'any. Novosibirsk, 1979. 

/2/ V. Briabrin: F-language - a Formalism for Knowledge 
Representation in the Intelligent Dialo~ae System. In 
"Applied Informatics", V.1, ed. Y. Savinkov, Moscow, 
1981. 
