KNOWLEDGE REPRESENTATION
FOR HUMAN-MACHINE INTERACTION
Mare Koit, Tiit Roosmaa and Haldur Õim
Institute of Computer Science & Institute of Estonian and General Linguistics, University of Tartu, J. Liivi 2, Tartu, Estonia
Keywords: Dialogue Model, Reasoning Model, Conversation Agent, Knowledge Representation.
Abstract: The paper describes a computational model that we are implementing in an experimental dialogue system.
Conversation process is modelled where one participant is trying to influence his/her partner to agree to do
an action. Our goal is to model natural dialogue where computer as a dialogue participant follows norms
and rules of human-human communication. We have worked on different aspects of developing a model of
dialogue, including its computer realisation in the lines of BDI model. The main specific traits of our model
are: 1) taking into account the "naïve" common-sense reasoning as the basis of dialogue, 2) modelling
dialogues where the goal of the initiator is to impose the partner to do a certain action. In the paper we
concentrate on the use of frames as the knowledge representation formalism in the dynamic context of
dialogue. As a practical realisation of the model we have in view a computer program which we call
communication trainer.
1 INTRODUCTION
We are dealing with interactions where the goal of
one of the participants is to get the partner to carry
out a certain action. Such dialogue can be considered
as rational behaviour of conversation agents which is
based on beliefs, desires and intentions of agents, at
the same time being restricted by their resources
(Webber, 2001; Jokinen, 2009).
Because of this, we have modelled the reasoning
processes that people supposedly go through when
working out a decision whether to do an action or
not. In a model of conversation agent it is necessary
to represent its cognitive states as well as cognitive
processes. One of the most well-known models of
this type is the BDI model (Allen, 1994, Boella and
van der Torre, 2003). A framework for
argumentation-based negotiation is proposed in
(Amgoud et al., 2007). In this paper, we will develop
the model considered in (Koit and Õim, 2000, 2004).
2 MODELLING THE
COMMUNICATION PROCESS
Let us consider conversation between two agents - A
(he) and B (she). In the goal base of one participant
(let it be A) a certain goal G
A
related to B´s activities
gets activated and triggers in A a reasoning process.
In constructing his first turn A must plan the
dialogue acts and determine their verbal form as a
turn r
1
. This turn triggers a reasoning process in B
where two types of procedures should be
distinguished: the interpretation of A´s turn and the
generation of her response r
2
. B´s response triggers
in A the same kind of reasoning cycle in the course
of which he has to evaluate how the realization of
his goal G
A
has proceeded, and depending on this he
may activate a new sub-goal of G
A
, and the cycle is
repeated: A builds a new turn r
3
. Dialogue comes to
an end, when A has reached or abandoned his goal.
2.1 Model of Conversation Agent
A conversation agent is a program that consists of
six (interacting) modules (cf. Koit and Õim, 2004):
(PL, PS, DM, INT, GEN, LP),
where PL - planner, PS - problem solver, DM -
dialogue manager, INT - interpreter, GEN -
generator, LP - linguistic processor. Conversation
agent uses in its work goal base GB and knowledge
base KB. A necessary precondition of interaction is
existence of shared (mutual) knowledge of agents.
396
Koit M., Roosmaa T. and
˜
Oim H. (2009).
KNOWLEDGE REPRESENTATION FOR HUMAN-MACHINE INTERACTION.
In Proceedings of the International Conference on Knowledge Engineering and Ontology Development, pages 396-399
Copyright
c
 SciTePress

2.2 Reasoning Model
After A has expressed his intention (that B does D),
B can respond with agreement or rejection,
depending on the result of her reasoning. We want to
model a "naïve" theory of reasoning that people
themselves use when they are interacting with other
people and trying to predict and influence their
decisions.
The reasoning model consists of two parts: 1) a
model of human motivational sphere; 2) reasoning
schemes. In the motivational sphere three basic
factors that regulate reasoning of a subject
concerning D are differentiated. First, subject may
wish to do D, if pleasant aspects of D for him/her
overweight unpleasant ones; second, subject may
find reasonable to do D, if D is needed to reach
some higher goal, and useful aspects of D
overweight harmful ones; and third, subject can be
in a situation where (s)he must (is obliged) to do D -
if not doing D will lead to some kind of punishment.
We call these factors WISH-, NEEDED- and
MUST-factors, respectively.
It is supposed here that the dimensions
pleasant/unpleasant, useful/harmful have numerical
values and that in the process of reasoning
(weighing the pro- and counter-arguments) these
values can be summed up. For examplee, for the
characterisation of pleasant and unpleasant aspects
of some action there are specific words which can be
expressed quantitatively: enticing, delightful,
enjoyable, attractive, acceptable, unattractive,
displeasing, repulsive etc.
We have represented the model of motivational
sphere of a subject by the following vector of
weights:
w = (w(resources), w(pleasant), w(unpleasant),
w(useful), w(harmful), w(obligatory), w(prohibited),
w(punishment-for-doing-a-prohibited-action),
w(punishment-for-not-doing-an-obligatory-action)).
In the description, w(pleasant), etc. means weight of
pleasant, etc. aspects of D.
The second part of the reasoning model consists of
reasoning schemes that supposedly regulate human
action-oriented reasoning. The reasoning proceeds
depending on the determinant which triggers it
(WISH, NEEDED or MUST). As an example, let us
present a reasoning procedure.
// Reasoning triggered by NEEDED-
determinant
Presumption: w(useful) > w(harmful) //
1. Are there enough resources for
doing D?
2. If not then do not do D.
3. Is w(pleasant) > w(unpleasant)?
4. If not then go to 10.
5. Is D prohibited?
6. If not then do D.
7. Is w(pleasant) + w(useful) >
w(unpleasant) + w(harmful) +
w(punishment-for-doing-a-prohibited-
action)?
8. If yes then do D.
9. Otherwise do not do D.
10. Is D obligatory?
11. If not then do not do D.
12. Is w(pleasant) + w(useful) +
w(punishment-for-not-doing-an-
obligatory-action) > w(unpleasant) +
w(harmful)?
13. If yes then do D.
14. Otherwise do not do D.
3 KNOWLEDGE
REPRESENTATION
3.1 World Knowledge
We are using frames for representing world
knowledge in our system. Let us consider the
following situation: A makes B a proposal to do an
action D. For example, Mary proposes John to make
a potato salad for the party.
There is the frame ACTION in our system:
ACTION
RESOURCES
ACTOR
ACT: a sequence of elementary acts
SETTING: ACTOR has RESOURCES
GOAL
CONSEQUENCE
The frame ACTION has sub-frames, e.g.:
PREPARING-POTATO-SALAD
SUP: ACTION
RESOURCES:
Components: boiled potato,
boiled egg, pickled cucumber, hacked
onion, sour cream, salt, bowl
Skills: take, chop up, mix,
decorate, add
Time: 30 minutes
ACT: take Components; chop up
potato, egg, cucumber; mix in bowl;
decorate with onion; add salt
GOAL, CONSEQUENCE: potato salad
3.2 Communication Knowledge
We are using two kinds of knowledge about
communication: 1) descriptions of dialogue acts
(proposal, question, argument, etc.), and 2)
communication algorithms - communicative
strategies and tactics.
3.2.1 Dialogue Acts
The dynamic parts of dialogue acts work for a
coherent dialogue - there are limited sets of dialogue
acts that can be come after the current act.
3.2.2 Communicative Strategies and Tactics
A communicative strategy is an algorithm used by a
participant for achieving his/her goal in interaction.
KNOWLEDGE REPRESENTATION FOR HUMAN-MACHINE INTERACTION
397

Communication takes place in a communicative
space which is determined by a number of
coordinates that characterize the relationships of
participants. Communication can be collaborative or
confrontational, personal or impersonal; it can be
characterized by the social distance between
participants; by the modality (friendly, ironic,
hostile, etc.) and by intensity (peaceful, vehement,
etc.).
The choice of communicative tactics depends on
the point of the communicative space in which the
participants place themselves. The values of the
coordinates are again given in the form of numerical
values.
The participant A can realize his communicative
strategy in different ways: stress pleasant aspects of
D (i.e. entice B), stress usefulness of D for B (i.e.
persuade B), stress punishment for not doing D if it
is obligatory (threaten B). We call communicative
tactics these concrete ways of realization of a
communicative strategy. The participant A, trying to
direct B´s reasoning to the positive decision (to do
D), proposes various arguments for doing D while
B, when opposing, proposes counter-arguments.
There exist three tactics for A in our model
connected with three reasoning procedures (WISH,
NEEDED, MUST). By tactics of enticing the
reasoning procedure WISH, by tactics of persuading
the procedure NEEDED and by tactics of
threatening the procedure MUST will be tried to
trigger in the partner.
The participant A when implementing a
communicative strategy uses a partner model - a
vector w
AB
- which includes his imagination about
weights of the aspects of the action D for B. The
more A knows about B the more similar is the vector
w
AB
with the vector w
B
of the motivational sphere of
the partner B. We can suppose that A has sets of
statements for influencing the weights of different
aspects of D for the partner B: {st
A
i
-asp
j
, i=1,
…,n
A
asp j
; j=1, …, n} where asp
j
is the j-th aspect of
D and n is the number of different aspects. All the
statements have their weights as well.
For illustration, let us present a schematic
description of the tactic of persuasion, based on the
reasoning procedure NEEDED. B may verbalize her
rejection to do D bringing out a certain statement
about a certain aspect of D (e.g. if B says I do not
have enough time then she indicates that resources
are missing for doing D). We can suppose that B has
a set of statements {st
B
i
-asp
j
, i=1,…,n
B
asp j
; j=1,…,
n} for indicating the aspect which weight caused her
rejection where asp
j
is the j-th aspect of D and n is
the number of aspects.
// Persuasion: A persuades B to do D //
WHILE B is rejecting AND A is not giving up
DO
CASE B´s answer OF
st
B
-resources //no resources//:
IF there are statements st
A
-resources THEN
present a statement st
A
-resources
i
in order to
point at the possibility to gain the resources,
at the same time showing that the cost of
gaining these resources is lower than the
weight of the usefulness of D. //The expected
result:
w
B
(resources):=w
B
(recourses)+w(st
A
-
resources
i
)//
ELSE exit //there are no more
statements, give up//
st
B
-harmful //much harm//:
IF there are statements st
A
-harmful THEN
present a statement st
A
-harmful
i
to decrease
the value of harmfulness in comparison with the
weight of usefulness
//The expected result:
w
B
(harmful):=w
B
(harmful)-w(st
A
-harmful
i
)//
st
B
-unpleasant //much unpleasant//:
IF there are statements st
A
-unpleasant THEN
present a statement st
A
-unpleasant
i
in order to
downgrade the unpleasant aspects of D as
compared to the useful aspects of D
//The expected result:
w
B
(unpleasant):=w
B
(unpleasant)-w(st
A
-
unpleasant
i
)//
st
B
-punishment-for-doing-a-prohibited-
action //D is prohibited and the punishment is
great//:
IF there are statements st
A
-punishment-for-
doing-a-prohibited-action THEN present a
statement st
A
-punishment-for-doing-a-
prohibited-action
i
in order to downgrade the
weight of punishment as compared to the
usefulness of D
//The expected result:
w
B
(punishment-for-doing-a-prohibited-
action):=w
B
(punishment-for-doing-a-prohibited-
action) - w(st
A
- punishment-for-doing-a-
prohibited-action
i
)//
st
B
-pleasant //little pleasant//:
IF there are statements st
A
-pleasant THEN
present a statement st
A
-pleasant
i
in order to
stress pleasantness
ELSE IF there are statements st
A
-unpleasant
THEN present a statement st
A
-unpleasant
i
in
order to downgrade unpleasantness
st
B
-obligatory //not obligatory; in such
a case, B´s reasoning finished on the step 11,
see above//
IF there are statements st
A
-pleasant THEN
present a statement st
A
-pleasant
i
in order to
stress the pleasant aspects of D
ELSE IF there are statements st
A
-unpleasant
THEN present a statement st
A
-unpleasant
i
in
order to downgrade the unpleasant aspects of D
END CASE
IF there are statements st
A
-useful THEN
present a statement st
A
-useful
i
in order to
stress usefulness
//The expected result:
w
B
(useful):=w
B
(useful)+w(st
A
-useful
i
)//
ELSE exit //give up//.
KEOD 2009 - International Conference on Knowledge Engineering and Ontology Development
398

4 DISCUSSION
When A tries to influence B in order to bring her to
a decision, A uses several statements to increase the
weights of the positive aspects and to decrease the
weights of the negative aspects of the action D under
consideration.
If B indicates a certain aspect which does not
allow her to do D then A simply can choose a
statement for attacking this aspect. If B does not
indicate a certain reason of rejection then A only can
stress the usefulness.
Let us consider an interaction where A is the
computer, and B - the user. When starting a dialogue
the computer chooses a point in the communicative
space and a communicative tactic and generates such
a partner model w
AB
(a set of weights) that a
reasoning procedure will give a positive decision.
Let us consider a brief example where the action D
is "to prepare a potato salad". A chooses a co-
operative and personal character of communication,
a short distance between participants and the neutral
intensity (that means that A and B are friends), and
generates such a partner model that the reasoning
procedure NEEDED will give a positive decision. A
will implement the tactic of persuasion. The
computer composes exemplars of the frames
PREPARING-POTATO-SALAD and PROPOSAL.
A (computer): Please prepare a potato salad.
B (user): I do not have enough time.
The computer must correct the value of w(resources)
in the partner model and chooses a dialogue act
ARGUMENT.
A: I will help you.
B: It is very hot in the kitchen.
The user pointed out the harmfulness of the action.
Thus the weight of w(harmful) will be corrected in
the user model.
A: My kitchen has a good ventilation.
etc.
An experimental dialogue system is implemented
which in interaction with a user can play the role of
both A or B. At the moment the computer operates
with semantic representations of linguistic
input/output only, the surface linguistic part of
interaction is provided in the form of a list of ready-
made and classified utterances which are used both
by the computer and user.
5 CONCLUSIONS
The main specific traits of our model are: 1) taking
into account the "naïve" common-sense reasoning as
the basis of dialogue, 2) modelling dialogues where
the initiator´s goal is to impose the partner to do a
certain action. We are continuing our work in the
following directions: 1) refining the reasoning
model, 2) developing linguistic knowledge, 3)
analysis of human-human dialogues in the Estonian
dialogue corpus in order to verify the model.
ACKNOWLEDGEMENTS
This work was supported in part by the Estonian
Science Foundation (grant No 7503) and the
Estonian Ministry of Research and Education (the
projects No SF0180078s08 and EKKTT09-57).
REFERENCES
Allen, J., 1994. Natural Language Understanding. Second
Edition. The Benjamins/Cummings Publ. Co.
Redwood City etc., 678 p.
Amgoud, L., Dimopoulus, Y., Moraitis, P., 2007. A
Unified and General Framework for Argumentation-
based Negotiation. In Proc. of AAMAS´07 - the 6th
international joint conference on Autonomous agents
and multiagent systems, http://portal.acm.org/-
citation.cfm?doid=329125.1329317
Boella, G., van der Torre, L., 2003. BDI and BOID
Argumentation. In Proc. of CMNA-03
http://www.computing.dundee.ac.uk/staff/creed/resear
ch/previous/cmna/finals/boella-final.pdf
Jokinen, K., 2009. Constructive Dialogue Modelling.
Speech Interaction and Rational Agents. Wiley, 160
p.
Koit, M., Õim, H., 2004. Argumentation in the Agreement
Negotiation Process: A Model that Involves Natural
Reasoning. In CMNA-04 - Proc. of the Workshop W12
on Computational Models of Natural Argument. 16
th
European Conference on Artificial Intelligence, 53-56.
Valencia, Spain.
Koit, M., Õim, H., 2000. Reasoning in interaction: a
model of dialogue. In TALN 2000 - 7th Conference on
Automatic Natural Language Processing, 217-224.
Ed. E. Wehrli. Lausanne, Switzerland.
Webber, B., 2001. Computational Perspectives on
Discourse and Dialogue. In The Handbook of
Discourse Analysis, 798-816. Eds. D. Schiffrin, D.
Tannen, H. Hamilton. Blackwell Publishers Ltd.
KNOWLEDGE REPRESENTATION FOR HUMAN-MACHINE INTERACTION
399