Developing a Formal Model of Argumentation-based Dialogue
Mare Koit
Institute of Computer Science, University of Tartu, J. Liivi 2, Tartu, Estonia
Keywords: Dialogue, Argument, Dialogue Model, Dialogue Corpus, Knowledge Representation.
Abstract: We are considering dialogues in natural language where the participants (A and B) are arguing for and against
of doing an action D by B. The participants can have similar or opposite communicative goals. If both A and
B have the same goal („B will do D“ or, respectively, „B will not do D“) then they are cooperatively looking
for arguments that will eliminate possible obstacles before achieving the goal. If the goals are opposite then
the participants exchange arguments and counterarguments and one of them has finally to abandon his or her
initial communicative goal. A model of dialogue has being developed which includes a model of argument.
An analysis of human-human dialogue corpus is carried out in order to give a preliminary evaluation of the
introduced model. A limited version of the model is implemented on the computer. Full implementation is
planned as a future work.
1 INTRODUCTION
Many researchers have been modelling
argumentation on the computer.
Rahwan et al (2004) discuss three approaches to
automated negotiation: game-theoretic, heuristic-
based and argumentation-based. Argumentation-
based approaches to negotiation allow agents to
‘argue’ about their beliefs and other mental attitudes
during the negotiation process.
Besnard and Hunter (2008) formalize
argumentation by using classical logic and define an
argument as a pair <Φ, α> where Φ is a set of
formulas (a subset of the knowledge base) and α is a
formula such that (1) Φ is consistent; (2) Φ entails α;
(3) Φ is a minimal subset of the knowledge base
which satisfies 2. If <Φ, α> is an argument, it is said
that it is an argument for α and it is also said that Φ is
a support for α. Here α is called the claim of the
argument.
Logical models of argument support decision
making by participants, guide negotiation and allow
to reach agreements (Amgoud and Cayrol, 2002).
Rahwan and Larson (2011) explore the
relationships between mechanism design and formal
logic, particularly in the design of logical inference
procedures when knowledge is shared among
multiple participants.
Hadjinikolis et al. (2012) provide an
argumentation-based framework for persuasion
dialogues, using a logical conception of arguments,
that an agent may undertake in a dialogue game,
based on its model of its opponents.
Overviews of the state of art in modelling
argumentation can be found e.g. in (Chesňevar et al.,
2000) and (Besnard and Hunter, 2008).
We are studying the interactions in natural
language between two participants (A and B) where A
is convincing B to do or, respectively, not to do an
action D. We have worked out a dialogue model
which includes a reasoning model as its part and
implemented it in a simple dialogue system (Koit and
Õim, 2000; 2014; Koit, 2015).
In the current paper, we will further develop the
model. The participants of dialogue exchange
arguments for and against of doing D. They can also
ask and answer questions in order to make choices
among the arguments for averting the partner’s
counterarguments.
The rest of the paper is structured as follows.
Section 2 introduces our current model of
argumentation-based dialogue. Section 3 gives the
results of analysis of human-human dialogues, in
order to justify the model. Section 4 discusses some
questions related to the concepts of argumentation,
negotiation and debate in human-human interaction
and in our computational model. Conclusions will be
made in Section 5.
258
Koit, M.
Developing a Formal Model of Argumentation-based Dialogue.
DOI: 10.5220/0005665502580263
In Proceedings of the 8th International Conference on Agents and Artificial Intelligence (ICAART 2016) - Volume 2, pages 258-263
ISBN: 978-989-758-172-4
Copyright
c
 2016 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved

2 DIALOGUE MODEL
2.1 The Structure of Dialogue
Let us consider a dialogue in natural language
between two participants (humans or artificial agents)
– A and B (Koit and Õim, 2014). Let the initiator of
dialogue be A, and let his communicative goal be “B
will do an action D” or, respectively, “B will not do
D”. B’s communicative goal can conform to A’s one
or it can be opposite. In interaction, A is influencing
B to make a decision about doing D which coincides
with his communicative goal. The following cases
can occur:
(1) A’s goal is “B will do D”, and B’s goal is “B will
do D”;
(2) A’s goal is “B will do D”, and B’s goal is “B will
not do D”;
(3) A’s goal is “B will not do D”, and B’s goal is “B
will do D”;
(4) A’s goal is “B will not do D”, and B’s goal is “B
will not do D”.
In the cases (1) and (4) A and B have the same goal
and in interaction, they are cooperatively looking for
reasons (arguments) why to do (respectively, not to
do) D and how to overcome possible obstacles before
doing
D or, respectively, to prevent possible
undesirable results of not doing D.
In the cases (2) and (3), A and B have opposite
goals and in interaction, the initiator A is proposing
arguments which should influence B to accept A’s
goal and to abandon her own initial goal. At the same
time, B can propose counterarguments which should
force A to accept B’s goal and to abandon his own
initial goal.
A as the initiator has a partner model in his
disposal – an image about B which gives him an
opportunity to suppose that B will agree to accept his
communicative goal (to do or, respectively, not to do
the action D). In constructing his first turn, A must
plan the dialogue acts (e.g. proposal, request,
question, proposal together with an argument, etc.
depending on his image of B) and determine their
verbal form (i.e. utterances). The partner B interprets
A’s turn and before generating her response, triggers
a reasoning procedure in her mind in order to make a
decision – to do D or not. In the reasoning process, B
weighs her resources for doing D, positive and
negative aspects of doing D and its consequences and
finally, makes a decision. Then she in her turn will
plan the dialogue acts (e.g. agreement, refusal, refusal
with argument, etc.) and their verbal form in order to
inform A about her decision. If B agrees to accept A’s
goal then the dialogue finishes (A has reached his
communicative goal). If B’s response is refusal then
A must change his partner model (it did not
correspond to the reality because A supposed that B
will agree to accept A’s goal) and find out new
arguments in order to convince B to make a positive
decision.
Our reasoning model has been introduced in
(Koit and Õim, 2000; 2014). It consists of two parts:
(1) a model of human motivational sphere; (2)
reasoning procedures.
In the motivational sphere three basic factors are
differentiated that regulate reasoning of a subject
concerning an action D. First, a subject may wish to
do D if the pleasant aspects of D for him/her
overweight the unpleasant ones; secondly, a subject
may find it reasonable to do D if D is needed to reach
some higher goal, and the useful aspects of D
overweight the harmful ones; and thirdly, a subject
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 determinants,
respectively.
If the subject is reasoning about not doing D then
the basic factors which trigger the reasoning are
analogous: first, the subject does not wish to do D if
unpleasant aspects of D overweight the pleasant ones;
secondly, doing D is not needed for him/her if
harmful aspects of D overweight the useful aspects;
and thirdly, doing D is prohibited (not allowed) for
him/her and will cause some punishment. We call
these factors NO-WISH, NOT-NEEDED and NOT-
ALLOWED determinants, respectively.
Let us represent the model of motivational sphere
of a subject concerning an action D by the following
vector of ‘weights’ (with numerical values of its
components):
w
D
= (w(resources
D
),
w(pleasant
D
), w(unpleasant
D
),
w(useful
D
), w(harmful
D
), w(obligatory
D
),
w(prohibited
D
), w(punishment-do
D
),
w(punishment-not
D
)).
In the description, w(pleasant
D
), etc. mean the
weight of pleasant, etc. aspects of D; w(punishment-
do
D
) – weight of punishment for doing D if it is
prohibited, and w(punishment-not
D
) – weight of
punishment for not doing D if it is obligatory.
Further, w(resources
D
) = 1, if subject has all the
resources necessary to do D (otherwise 0);
w(obligatory
D
) = 1, if D is obligatory for the reasoning
subject (otherwise 0); w(prohibited
D
) = 1, if D is
prohibited (otherwise 0). The values of other weights
can be non-negative natural numbers. In the
following, we suppose that the action D is fixed and
do not indicate it explicitly in the vector.
Developing a Formal Model of Argumentation-based Dialogue
259

The second part of the reasoning model consists
of reasoning procedures that supposedly regulate
human action-oriented reasoning. A reasoning
procedure depends on the determinant which triggers
it (in our model, WISH, NEEDED, MUST, or
respectively, NO-WISH, NOT-NEEDED, NOT-
ALLOWED). As an example, let us present a
procedure triggered by the NOT-ALLOWED
determinant.
Presumption: D is prohibited.
1) Are there enough resources for
doing D? If not then goto 8.
2) Is w(pleasant) > w(unpleasant)? If
not then 8.
3) Is w(pleasant) > w(unpleasant) +
w(punishment-do)? If not then goto 8.
4) Is w(pleasant) > w(unpleasant) +
w(punishment-do) + w(harm)? If not then
goto 8.
5) Is w(pleasant) + w(useful) >
w(unpleasant) + w(punishment-do) +
w(harm)? If not then goto 8.
7) Decide: do D. End.
8) Decide: do not do D.
The vector w
AB
(A’s beliefs concerning B’s
evaluations in relation to the action D) is used as a
partner model while the vector w
B
– the model of B
herself – represents B’s actual evaluations of D’s
aspects (which exact values A does not know).
A communicative strategy is an algorithm used
by a participant for achieving his/her goal in the
interaction. The initiator (participant A) can realize
his communicative strategy in different ways: stress
pleasant or respectively, unpleasant aspects of D (i.e.
entice the partner B), stress usefulness or,
respectively, harmfulness of D for B (i.e. persuade B),
stress punishment for not doing D if it is obligatory or
respectively, punishment for doing D if it is
prohibited (threaten B), etc. These concrete ways of
realization of a communicative strategy we call
communicative tactics. A, trying to direct B’s
reasoning to the desirable decision, proposes
arguments for doing D (respectively, not doing D)
while B, when opposing, proposes counterarguments.
When influencing B in interaction, A can bring out
different aspects of D. Implementing certain
communicative tactics in a systematic way A will
choose one aspect of D (the ‘title’ aspect of the fixed
tactics) and proposes arguments for stressing it.
In order to achieve B’s decision to do D, A can
stress the following ‘title’ aspects:
• pleasantness of D (i.e. to trigger B’s reasoning
procedure by the WISH determinant)
• usefulness of D (to trigger the reasoning
procedure by the NEEDED determinant)
• punishment for not doing
D if D is obligatory
for B (to trigger the reasoning procedure by the
MUST determinant).
Similarly, in order to achieve B’s decision not to
do D, A can stress unpleasantness, harmfulness or
punishment for doing D.
The knowledge base for the agent A includes (1)
reasoning algorithms, (2) communicative strategies
and tactics, (3) the partner model w
AB
, (4) a list of
dialogue acts which A can use (proposal, question,
assertion, etc.), (5) a list of utterances which he can
use for verbalizing the dialogue acts.
The knowledge base for B includes similar
knowledge, the only difference is that is w
B
(the
model of B herself) is used instead of the partner
model w
AB
.
When interacting about an action, A and B
exchange arguments. The general structure of A’s
argument is as follows, cf. (Amgoud and Cayrol,
2002; Besnard and Hunter, 2008; Koit, 2015):
<{R, T, w
AB
i
, proposition
A
}, claim
A
>,
where
• R
is the reasoning procedure which A is trying
to trigger in B
• T is the communicative tactics used
• w
AB
i
= (w
AB
i
(resources), w
AB
i
(pleasant), w
AB
i
(unpleasant), w
AB
i
(useful), w
AB
i
(harmful),
w
AB
i
(obligatory), w
AB
i
(prohibited),
w
AB
i
(punishment-do), w
AB
i
(punishment-not)) is the
current partner model (at turn i of the dialogue)
• proposition
A
denotes the utterance chosen by
A in order to influence one of the weights in the
partner model, after what R will supposedly give
B’s positive decision on the changed model (which
coincides with A’s communicative goal) its weight
is w(proposition
A
• claim
A
= “B will do D“ or, respectively, “B
will not do D“.
The proposition
A
chosen by A in interaction
yields a new partner model w
AB
i+1
(at turn i+1):
• if proposition
A
∈ P
increase_resources
, then
w
AB
i+1
(resources):=1
• if proposition
A
∈ P
increase_pleasantness
, then
w
AB
i+1
(pleasant):= w
AB
i
(pleasant) +
w(proposition
A
),
etc.
Here P
increase_resources
denotes the set of propositions
(utterances) that can be used for indicating that there
exist resources for doing D; P
increase_pleasantness
denotes
ICAART 2016 - 8th International Conference on Agents and Artificial Intelligence
260

the set of utterances for increasing the pleasantness of
D, etc.
The structure of B’s argument is analogous:
<{R
B
, T
B
, w
B
, proposition
B
}, claim
B
>,
where
• the reasoning algorithm R
B
gives the decision
“do not do D” or respectively, “do D” (claim
B
) on
the model w
B
• proposition
B
indicates the aspect of D which
(too small or too big) value causes this decision
• T
B
is the current communicative tactics of B.
Here B’s proposition
B
gives to A information for
choosing his next proposition (as argument) in
interaction. For example, if A is arguing for doing D
and proposition
B
∈ P
missing_resources
, then the actual
value of w
AB
i
(resources) is 0 and the next utterance
will be chosen by A from the set P
increasing_resources
(after
that, w
AB
i+1
(resources) = 1 will hold) and another
proposition will be chosen from the set of
propositions which correspond to the title aspect of
the reasoning algorithm R which A is trying to trigger
in B using the communicative tactics T.
In order to choose the next proposition
(counterargument), B triggers her current reasoning
procedure R
B
in her model w
B
, and finally, B is able
to determine the aspect of D which brought her to the
negative decision. For example, she can choose an
utterance indicating to missing resources, e.g. by
saying I don't have so much money as needed to do D
but she can also refuse by saying I do not do D. In the
last case, A cannot avert any counterargument but he
has to make a choice among the utterances for
stressing the title aspect of the implemented
communicative tactics T.
2.2 Argumentation-based Dialogue
If A and B have contradictory goals when starting
interaction then they are involved into debate (e.g. A’s
communicative goal is “B will do D”, B’s goal is “B
will not do D”). One participant will achieve his or
her communicative goal (‘wins’ debate) and another
has to abandon her or his initial goal (‘loses’ debate).
If A and B have common communicative goals
then they are cooperatively looking for arguments
that support achieving this collective goal. Still, for
example, B can indicate to obstacles which do not
allow achieve the goal. Then A has to find arguments
for showing how the obstacles can be eliminated. The
final result of discussion is whether achieving the
collective goal or its withdrawal if some of the
obstacles cannot be eliminated.
Let us suppose that both A and B have a common
set of reasoning procedures. We also suppose that
both A and B can use fixed sets of dialogue acts (e.g.
proposal, question, agreement, refusal, statements for
increasing or decreasing the values of different
components of the vector of motivational sphere
which will be used as arguments for doing or not
doing D) and corresponding utterances which are
classified semantically, e.g. P
increasing_resources
for
indicating that there exist resources for doing D,
P
increasing_pleasantness
for stressing pleasantness of D,
P
missing_resources
for indicating that some resources for
doing D are missing, P
decreasing_pleasantness
for decreasing
pleasantness of D, etc.
Starting interaction, A fixes a partner model w
AB
using his pre-knowledge about B, and determines the
communicative tactics T which he will use, i.e. he
accordingly fixes a reasoning algorithm R which he
will try to trigger in B’s mind. B has her own model
w
B
. She determines a reasoning procedure R
B
which
she will use in order to make a decision about doing
D.
The structure of argumentation-based dialogue
looks like follows (the dialogue acts in parentheses
can miss):
A: proposal (+ argument)
REPEAT
(
B: question
A: answer/giving information
)
B: agreement OR refusal (+ argument)
(
A: question
B: answer/giving information
)
A: argument
UNTIL a finishing condition is
fulfilled.
Whether A or B can indicate that a finishing
condition is fulfilled. The finishing conditions are: (1)
the communicative goal is already achieved, (2) the
participant gives up (2.1) regardless of having
utterances for expressing new arguments, or (2.2)
there are no utterances to continue the fixed
communicative tactics but no new tactics will be
chosen regardless of having some tactics not
implemented so far, or (2.3) all the tactics are already
implemented and all the utterances are used but the
communicative goal is not achieved.
Questions can be asked by participants in order to
make choices between different utterances which can
be used in argumentation.
Developing a Formal Model of Argumentation-based Dialogue
261

3 ANALYSIS OF
HUMAN-HUMAN DIALOGUES
Does the structure of actual human-human dialogues
coincide with the structure presented in Section 2.2?
We carried out an analysis of dialogues taken from
the Estonian dialogue corpus (Hennoste et al., 2008):
(1) 22 everyday calls, and 4 face-to-face
conversations between acquaintances, and (2) 24 calls
of a customer who is planning a trip with a travel
agent.
Let us consider two examples. The first example
is an everyday phone call between mother and
daughter. The second example is a face-to-face
conversation in a travel agency. Transcription of
Conversation Analysis (Sidnell and Stivers, 2012) is
used in the examples.
Example 1. Here, the mother A presents several
arguments in order to increase her daughter’s wish to
bake gingersnaps (the action D). A’s last argument (I
will not be at home) turns out to be sufficient for
making a positive decision by B.
/---/
A: .hhhhh kas sulle pakuks ´pinget ´piparkookide
´küpsetamine.
would you like to bake gingersnaps
proposal
B: .hhhhhhh ma=i=´tea vist ´mitte.
I don’t know, perhaps not
refusal
A: ja=sis gla´suurimine=ja=´nii.
and then glazing and so on
proposition
A
1
(0.6)
B: ´ei, ´ei, ´ei ei=´ei.
no, no, no, no, no
refusal
(0.9)
A: me saaksime nad ´vanaema=jurde ´kaasa võtta.
we could take them with us when going to visit grandmother
proposition
A
2
(0.4)
B: ´präägu ei=´taha.
I don’t want just now
refusal
/---/
A: ma mõtlen: kui mind kodus ei=´ole.
I suppose when I will not be at home
proposition
A
3
B: aa.
ah
(0.5) .hhh et ´lähen ostan ´tainast=vä.
then I’ll go to buy paste, yes
agreement
Example 2. The travel agent A presents several
arguments attempting to indicate that the proposed
trip (which is here the action D) is interesting/ useful
for the customer. B asks questions in order to make a
decision.
/---/
A: m:eil on ´sellel aastal (.) uus ´reis välja pakkuda, see on
Sit´siilia.
we offer a new trip this year to Sicily
proposal
(.) see peaks teid kindlasti ´huvitama, see on nimelt niisuge
omapärane mt=.hh ´kant I´taalias.
you should like it, this is an original place in Italy
proposition
A
1
(0.6)
B: ee (0.6) mis:=mis:=ee (0.4) mis=a- aja- ´aegadel teil on
which time do you offer
question
/---/
A: @ te näete antiik ja ba´rokkunsti ja saate suurepärase
´võimaluse {-} ´puhata Dürreeni mere ´ran[nikul.] @
you will see ancient and baroque art and you will have an
excellent chance to take a rest at the coast of Tyrrhenian
sea
proposition
A
2
/---/
B: et=ee (.) kas see nagu ´väljasõidud ja=kõik=e (.) kas
ned=on=nagu: ´hinna ´sees kohe või net: tuleb ´eraldi
arvestada.
are the set-offs included in the price or have they to be paid
separately
question
/---/
The results of the corpus analysis show that the
introduced model can be in general lines suitable for
analysis of Estonian human-human dialogues and it
can be taken as a basis of a dialogue system.
4 DISCUSSION
We are considering dialogues where two participants
argue about doing an action D by one of them. Here
we would like to explain our understanding of the
relationships between such concepts as
argumentation, negotiation, and debate as used in the
paper.
Argumentation (as a discussion in which reasons
are advanced for and against some proposition or
proposal) constitutes a necessary part of negotiations
and debates. Both in negotiation and in debate there
are clearly fixed ‘sides’ with different goals when
considering the outcome of the communicative event.
However, negotiation covers much more divergent
possible variants than debate. “Negotiation is a form
of interaction in which a group of agents with
ICAART 2016 - 8th International Conference on Agents and Artificial Intelligence
262

conflicting interests and a desire to cooperate try to
come to a mutually acceptable agreement on the
division of scarce resources“ (Rahwan et al., 2004).
The main uniting feature of all variants of negotiation
is that the participants start the communicative event
with the ultimate aim to reach an agreement which is
seen as a compromise, that is, all sides are ready to
accept some losses. Debate is an adversarial event
from the start: the participants have conflicting goals
and the aim of each participant is to promote his or
her goal only.
The model presented in Section 2 covers a certain
limited kind of negotiations about doing an action. If
A and B are pursuing the same communicative goal
then they start discussion in order to explain that there
are no obstacles before doing the action D or,
respectively, no undesirable consequences follow
after D will not be done. They do not necessarily
achieve their joint communicative goal. The model
does not consider the situations where the initial goal
will be modified. If the goals are opposite then A and
B are involved into debate where one participant wins
and another loses.
The structure of argument used in the model is
adapted to the limited kind of negotiations considered
here. When arguing, a participant presents only one
part of argument – proposition(s); the remaining parts
are implicit (cf. the examples in Section 3).
5 CONCLUSION AND FUTURE
WORK
We introduced a model of argumentation-based
dialogue which includes exchange of arguments. A
model of argument is presented which consists of a
partner model for A (or, respectively, a model of
herself for B), a reasoning procedure which A tries to
trigger in B (or what B is implementing herself),
communicative tactics and (a set of) proposition(s)
(utterances) which all together would bring A and/or
B to a desirable conclusion. The conclusion (a
decision about doing D by B) is interpreted as a claim
in the structure of argument.
We evaluated our model on actual human-human
dialogues taken from a dialogue corpus. The corpus
study gives an opportunity to believe that the
introduced model can be used for the analysis of
human-human dialogues and modelling them in a
dialogue system.
We have implemented on the computer a simple
argumentation-based dialogue (debate) where A’s
communicative goal is “B will do D” and B’s goal is,
on the contrary, “do not D” (Koit, 2015). Our future
work includes implementation of the whole model.
ACKNOWLEDGEMENTS
This work was supported by the Estonian Research
Council (project IUT-2056).
REFERENCES
L. Amgoud, and C. Cayrol. 2002. A Reasoning Model
Based on the Production of Acceptable Arguments. In
Ann. Math. Artif. Intell. 34(1-3): 197–215.
P. Besnard, and A. Hunter. 2008. Elements of
Argumentation, MIT Press, Cambridge, MA.
C. Chesñevar, A. Maguitman, and R. Loui. 2000. Logical
Models of Argument. In ACM Computing Surveys,
32(4), 337–383.
C. Hadjinikolis, S. Modgil, E. Black, P. McBurney, and M.
Luck. 2012. Investigating Strategic Considerations in
Persuasion Dialogue Games. In STAIRS, 137–148.
T. Hennoste, O. Gerassimenko, R. Kasterpalu, M. Koit, A.
Rääbis, and K. Strandson. 2008. From Human
Communication to Intelligent User Interfaces: Corpora
of Spoken Estonian. In Proc. of the 6th International
Language Resources and Evaluation (LREC'08).
Marrakech, Morocco: European Language Resources
Association (ELRA), 2025–2032.
M. Koit. 2015. Communicative Strategy in a Formal Model
of Dispute. In Proc. of the International Conference on
Agents and Artificial Intelligence: 7th International
Conference on Agents and Artificial Intelligence
(ICAART), Lisbon, Portugal, SCITEPRESS, 489–496.
M. Koit, and H. Õim. 2014. A Computational Model of
Argumentation in Agreement Negotiation Processes. In
Argument & Computation, 5 (2-3), 209–236, Taylor &
Francis Online. DOI: 10.1080/19462166.2014.915233
M. Koit, and H. Õim. 2000. Developing a Model of Natural
Dialogue. In From spoken dialogue to full natural
interactive dialogue-theory, Empirical analysis and
evaluation. LREC2000 Workshop proceedings, 18–21.
I. Rahwan, and K. Larson. 2011. Logical Mechanism
Design. In The Knowledge Engineering Review, 26(1),
61–69.
I. Rahwan, S. D. Ramchurn, N. R. Jennings, P. Mcburney,
S. Parsons, and L. Sonenberg. 2004. Argumentation-
Based Negotiation. In The Knowledge Engineering
Review, Vol. 18:4, 343–375. Cambridge University
Press. DOI: 10.1017/S0269888904000098
J. Sidnell, and T. Stivers (eds.). 2012. Handbook of
Conversation Analysis, Boston: Wiley-Blackwell.
Developing a Formal Model of Argumentation-based Dialogue
263