Veenen, Prakken - A Protocol for Arguing About Rejections.pdf

(1652 KB) Pobierz
LNAI 4049 - A Protocol for Arguing About Rejections in Negotiation
A Protocol for Arguing About Rejections
in Negotiation
Jelle van Veenen 1 and Henry Prakken 2 , 3
1
Faculty of Law, Tilburg University, The Netherlands
j.vanveenen@uvt.nl
2
Department of Information and Computing Sciences,
Utrecht University, The Netherlands
henry@cs.uu.nl
3
Centre for Law & ICT, Faculty of Law, University of Groningen, The Netherlands
Abstract. One form of argument-based negotiation is when agents
argue about why an offer was rejected. If an agent can state a reason
for a rejection of an offer, the negotiation process may become more e-
cient since the other agent can take this reason into account when making
new offers. Also, if a reason for rejection can be disputed, the negotiation
process may be of higher quality since flawed reasons may be revised as
a result. This paper presents a formal protocol for negotiation in which
reasons can be asked and given for rejections and in which agents can try
to persuade each other that a reason is or is not acceptable. The proto-
col is modelled as a persuasion dialogue game embedded in a negotiation
protocol. It has a social semantics since the protocol does not refer to
the internal state of negotiating agents.
1 Introduction
Recently argumentation-based approaches to negotiation have become popular
(see [1] for an overview and motivation). The idea is that if negotiating agents
exchange reasons for their proposals and rejections, the negotiation process may
become more ecient and the negotiation outcome may be of higher quality.
This paper especially focuses on reasons given for rejections of proposals. If an
agent explains why he rejects a proposal, the other agent knows which of her
future proposals will certainly be rejected so she will not waste effort at such
proposals. Thus eciency is promoted. In such exchanges, reasons are not only
exchanged, they can also become the subject of debate. Suppose a car seller offers
a Peugeot to the customer but the customer rejects the offer on the grounds that
French cars are not safe enough. The car seller might then try to persuade the
customer that he is mistaken about the safety of French cars. If she succeeds
in persuading the customer that he was wrong, she can still offer her Peugeot.
Thus the quality of the negotiation is promoted, since the buyer has revised his
preferences to bring them in agreement with reality.
This example illustrates that a negotiation dialogue (where the aim is to reach
a deal) sometimes contains an embedded persuasion dialogue (where the aim is
S. Parsons et al. (Eds.): ArgMAS 2005,
LNAI
4049, pp. 138– 153, 2006.
Springer-Verlag Berlin Heidelberg 2006
 
A Protocol for Arguing About Rejections in Negotiation
139
to resolve a conflict of opinion). The aim of this paper is to formulate a pro-
tocol for negotiation with embedded persuasion dialogues about the reasons for
rejecting a proposal. The key idea is that the propositional commitments in-
curred by the agents in the embedded persuasion dialogue constrain their be-
haviour in the surrounding negotiation dialogue. In agreement with the current
trend [2] we intend the combined protocol to have a social semantics. For this
reason we will completely abstract from the internal design of the communi-
cating agents; in particular, the protocol will only refer to the agents’ publicly
observable behaviour within a dialogue. According to [2] a social semantics is
desirable for agent interaction protocols since if a protocol refers to an agent’s
mental state, there is no gurantee that an outside observer can verify whether
the agent complies with a protocl.
The main novelty of the present research lies in the fact that current protocols
for argument-based negotiation only allow arguments supporting proposals. One
exception is [3], which also allows arguments about rejections. However, their
protocol does not have a social semantics, since whether an agent is allowed to
assert a claim or an argument partly depends on their internal mental state. The
protocol has some other limitations, which will be discussed in Section 5.
Our proposal will be stated in a dialogue game form. It will combine a negoti-
ation protocol and language of [4] with a persuasion protocol based on [5], which
adapts and extends work of [6]. In the following sections we will first introduce
these two systems and sketch the underlying argumentation logic that we will
use. Then we will present our combined protocol, investigate some of its formal
properties and illustrate it with an example.
2 The Building Blocks
In this section we present the negotiation and persuasion system that we aim
to combine. Both systems are formulated as a dialogue game. Dialogue games
formulate principles for coherent dialogue, and coherence depends on the goal of
a dialogue. The goal of negotiation dialogues is to reach agreement on the division
of scarce resources [1] and the goal of persuasion dialogues is to resolve a conflict
of opinion [7]. Formal dialogue games have a topic language L t with a logic
L
2.1 A Language and Protocol for Multi-attribute Negotiation
The negotiation system we will use is that of Wooldridge and Parsons [4]. The
negotiation topic language L t of this system assumes that in a negotiation
agents try to reach agreement over the values of a finite set V =
{
,
and a communication language L c with a protocol P . The protocol specifies the
allowed moves at each point in a dialogue. A dialogue system also has effect
rules , which specify the effects of utterances on the participants’ commitments,
and outcome rules , defining the outcome of a dialogue.
v 1 , ..., v m }
of negotiation issues . Each issue v can be assigned at most one value from a
range C ( v )ofvalues.An outcome of a negotiation is an assignment of values to
140
J. van Veenen and H. Prakken
C ( v )
or c =?, (where ? technically is a free variable, capturing that the issue has not
been assigned a value) and R denotes one of the relations = ,<,>,
V and c
.
The negotiation communication language L c can be used to talk about
proposals. The left column of table 1 shows the speech acts that agents can
perform and the right column their possible replies. The formulas ϕ and ϕ
are elements of L t . Request(ϕ) is a request for an offer. Here ϕ typically is
wholly or partially uninstantiated (i.e., it may contains occurrences of ?): the
speech act request ( price =?
or
warranty = 12) can be read as “What is the
price if I want a 12 months warranty?”. The speech act offer ( ϕ ) makes a fully
instantiated proposal ϕ ,andwith accept ( ϕ ) an agent accepts an offer ϕ made
by another agent. With reject ( ϕ ) such an offer is rejected. With withdraw an
agent withdraws from the negotiation.
We next outline the negotiation protocol of [4] for this language, with no-
tation slightly adapted to our purposes. A negotiation takes place between two
agents, one of whom starts with either an offer or a request . The agents then
take turns after each utterance, selecting their replies from Table 1. As the table
indicates, a negotiation terminates when an agent accepts an offer or withdraws
from the negotiation. Finally, moves may not be repeated by the same player.
Table 1. Speech acts and replies in L c
Acts Replies:
request(ϕ)offer(ϕ )
offer(ϕ) ϕ ) or accept(ϕ) or
reject(ϕ) or withdraw
reject(ϕ)off ϕ ) or withdraw
accept(ϕ) end of negotiation
withdraw end of negotiation
( ϕ = ϕ )
To ensure that the offers exchanged during a negotiation and its outcome are
related to an initial request, we add the following rule to the protocol of [4]:
If request ( ϕ ) is the initial request of a dialogue then for any move offer ( ψ )
in the dialogue:
ψ is logically consistent with ϕ ;and
ψ contains at least the same issues as ϕ .
Since issues have at most one value, this rule implies that an instantiated part of
a request cannot be changed by an offer (but the offer may contain more issues
than the request). Therefore:
Proposition 1. If a negotiation that starts with a request terminates with accep-
tance of an offer, that offer is consistent with and fully instantiates the request.
a subset of V . A proposal is expressed in a subset of the language of first-order
logic as a conjunction of expressions of the form vRc ,where v
590220748.001.png
A Protocol for Arguing About Rejections in Negotiation
141
We illustrate the system with an example in which two agents, Paul ( P )and
Olga ( O ), negotiate over the sale of a car. The dialogue starts when Paul requests
to buy a car, and shows that he is interested in the brand and the price.
P 1 : request ( brand =?
price = 10000)
price = 10000)
(Olga has offered a Peugeot for 10000, but Paul has rejected the offer. Olga
makes him another offer.)
O 4 : offer ( brand = renault
price = 8000
stereo = yes )
P 5 : reject ( brand = renault
price = 8000
stereo = yes )
O 6 : offer ( brand = audi
price = 10000)
price = 10000)
(Olga offers a Renault with stereo for 8000. Paul again rejects after which Olga
offers a non-French car for 10000. Paul accepts and the dialogue terminates.
Move O 4 illustrates that an offer may introduce additional issues, for instance,
to make an offer more attractive or to make a trade-off possible.)
2.2 The Underlying Argumentation Logic
. In doing so we will abstract
from details of the language and inference rules wherever possible, to allow for
different instantiations of the logic and language. Thus we in fact specify a set
of constraints on L t and
L
assumed by our persuasion dialogue system.
Much other work on argument-based dialogue, e.g. [3, 8], regards arguments
as classical proofs from consistent sets of propositional formulas and allows clas-
sical inconsistency of the premises of two arguments as the only source of attack
on arguments. We argue that the present application requires a richer language
and notions of argument and attack. The topic language L t must include a
suitable subset of first-order predicate logic, to express arguments about val-
ues for negotiation issues. Since in persuasion dialogues arguments are often
attacked by counterarguments, the logic
L
must be a logic for defeasible argu-
mentation, or ‘argumentation system’ for short (cf. [9]). We want our system to
be an instance of the well-understood abstract framework of [10], in particular
of his grounded semantics (also used by e.g. [11, 12]), since this semantics can
be easily incorporated into a persuasion dialogue game. Since the arguments
exchanged in persuasion dialogues are often constructed stepwise during a di-
alogue in reply to challenges of the premises, the argumentation system must
allow for a tree structure of arguments, where inference rules are chained into
trees. As for notation, prem ( A )and conc ( A ) denote the premises and conclu-
sion of argument A , i.e. the leaves and root of the tree structure. Furthermore,
since arguments exchanged in persuasion dialogues are often based on defeasi-
ble argumentation schemes (such as schemes for practical reasoning or default
reasoning), the rules for constructing arguments must include defeasible as well
as deductive inference rules. Each defeasible inference rule comes with one or
more undercutters , which specify the circumstances under which the inference
L
price =?)
O 2 : offer ( brand = peugeot
P 3 : reject ( brand = peugeot
P 7 : accept ( brand = audi
We next present the logical elements assumed by our persuasion protocol, i.e.,
the persuasion topic language L t and its logic
142
J. van Veenen and H. Prakken
rule cannot be applied. Accordingly, a defeasible argument can be defeated in
two ways. It can be rebut with an argument for the opposite conclusion, while it
can be undercut with an argument why an inference rule does not apply in the
given circumstances. To be successful, an attack should be of a certain strength.
In the present paper, we will not discuss issues of strength and therefore implic-
itly assume a given measure of relative strength between arguments. Also, since
our persuasion dialogue system is intended for any underlying logic satisfying
the above constraints, we will not further specify the defeasible inference rules
here but rather introduce them semiformally when discussing our examples. For
technical details the reader is referred to e.g. [13] and [9].
Given a set of arguments and a binary defeat relation defined over it, an argu-
mentation system classifies the arguments into justified, defensible and overruled
arguments. Our persuasion system presupposes a game-theoretic formulation of
Dung’s grounded semantics [14, 11]. The proponent and opponent of a certain
argument play a game where proponent starts with an argument he wants to
defend and then both players take turns, defeating the preceding argument with
a counterargument. A player wins if the other player has run out of moves. Now
an argument is justified if proponent has a winning strategy in a game starting
with the argument; and a proposition is justified if it is the conclusion of a jus-
tified argument. This game can be optimised in several ways (see e.g. [11]) but
in order to focus on the essence we leave them undiscussed here.
2.3 A Dialogue Game for Persuasion
We now present a dialogue game for persuasion. As noted above, this game is an
instance of the framework of [5], which adapts and further develops the system
of [6]. We are particularly interested in using this framework’s idea of reply
structure on the communication language and its notions of dialogical status
of relevance. A crucial feature of our game is that its protocol is flexible in
that it allows for alternative replies to moves and for postponement of replies,
sometimes even indefinitely. This is important since when an agent sees that a
line of attack or defence fails, s/he should be allowed to play other available lines
of attack or defence. However, in order to still ensure a strong focus of dialogues
this flexibility is constrained by the notion of relevance, to be defined below.
The dialogue game will be presented here in detail since its format plays a
crucial role in Section 3 in the combination of the persuasion and negotiation
dialogue game. Dialogues are between a proponent P and opponent O of a single
dialogue topic t
L t . The game is based on the following ideas. Each dialogue
move except the initial one replies to one earlier move in the dialogue of the other
party (its target ). Thus a dialogue can be regarded in two ways: as a sequence
(reflecting the order in which the moves are made) and as a tree (reflecting the
reply relations between the moves). Each replying move is either an attacker
or a surrender . For instance, a claim ( p ) move can be attacked with a why ( p )
move and surrendered with a concede ( p )move.Anda why ( p )movecanbe
attacked with an argue ( A )movewhere A is an argument with conclusion p ,
and surrendered with a retract ( p ) move. When s is a surrendering and s is
Zgłoś jeśli naruszono regulamin