Proceedings of the 7th SIGdial Workshop on Discourse and Dialogue, pages 46–53,
Sydney, July 2006. c©2006 Association for Computational Linguistics
DRT Representation of Degrees of Belief
Yafa Al-Raheb
National Centre for Language Technology
Dublin City University
Ireland
yafa.alraheb@gmail.com
Abstract
This paper investigates the problems facing mod-
elling agents’ beliefs in Discourse Representation
Theory (DRT) and presents a viable solution in the
form of a dialogue-based DRT representation of
beliefs. Integrating modelling dialogue interaction
into DRT allows modelling agents’ beliefs, inten-
tions and mutual beliefs. Furthermore, it is one of
the aims of the paper to account for the important
notion of agents’ varying degrees of belief in differ-
ent contexts.1
1 Introduction
Heydrich et al. remark that ‘serious description of
natural dialogue seems to necessitate that we con-
sider the mental states of the speakers involved’
(1998).2 This is a step that is by no means easy. It
is the aim of this paper to integrate previous work
on beliefs in DRT and dialogue theory in order to
model the mental states of agents in dialogue.
The connection between beliefs, intentions and
speech or dialogue acts has been noted in the liter-
ature. Stalnaker notes, for instance, that
[i]f we understand contexts, and the
speech acts made in contexts, in terms
of the speaker’s beliefs and intentions,
we have a better chance of giving sim-
pler and more transparent explanations
of linguistic behaviour (Stalnaker 2002:
720).
The kind of agent beliefs we are concerned with
here arises in dialogue interaction. The nature of
1I gratefully acknowledge support from Science Founda-
tion Ireland grant 04/IN/I527.
2Other names for mental state used in the literature in-
clude ‘information state’, ‘conversational score’, and ‘dis-
course context’ (Larsson and Traum 2000).
interaction dictates that the strength or degree of
belief varies depending on contextual factors. This
can be seen from the following example:
(1) A: I want to make a booking for my
wife.
B: Yeah.
A: What time is the Thailand flight on
Monday?
B: It’s at 2 pm.
In example (1) B does not necessarily need to be-
lieve the presupposition (given information) that
A has a wife. For the purposes of the conversa-
tion, which is providing A with information, B can
simply ‘go along with’ the presupposition and not
have it as a member of his beliefs (i.e. his belief
set) (Stalnaker 2002). Similarly, let us consider the
following example, (2). The speaker is a customer
in a clothing shop.
(2) S1: I want to buy a dress for my wife.
H1: Is it for a formal occasion?
S2: Yes.
H2: What is her favourite colour?
S3: She doesn’t like red anymore.
H3: Does your wife like black?
S4: Yes
As the speaker, S, introduces the presupposition
that he has a wife, the hearer, H, can come to the
conclusion that S believes S has a wife. However,
when the hearer comes to refer to S’s wife, H does
not necessarily have to believe S has a wife. H
can simply go along with the information that the
speaker has a wife and use this form of acceptance
in H2 without committing to ‘strongly believing’
it. Indeed, the speaker may be buying a dress for
his mistress rather than his wife. By going along
46
with it, the hearer does not have to commit him-
self to believing that the speaker has a wife. What
is more at stake than believing that the speaker in-
deed has a wife and not a mistress is closing the
sale. Contrast examples (1) and (2) with example
(3):
(3) S1: You have to get Peter’s son a Chris-
tening present.
H1: Peter has a son?
S2: Sorry I forgot to mention that be-
fore.
H2: Ok, what sort of present should I
get him?
S3: A toy would be nice.
In this context, the hearer, H, is required to commit
more strongly to the presupposition of Peter hav-
ing a son than simply going along with it, since H
is being asked to buy a Christening present. The
fact that H2 agrees to buying a present for Peter’s
son reflects more commitment to the presuppo-
sition than B shows in example (1). Considera-
tions of this kind lead to the conclusion that dif-
ferent contexts call for varying strengths of beliefs
and belief representation. We shall not attempt to
describe all the contextual factors that can cause
strength of belief to vary. The point is, rather, that
we clearly need to model strength of belief and
no current model of DRT incorporates such a pro-
posal. This paper, thus, makes an original proposal
for including a system for graded beliefs in the be-
lief spaces (or sets) of both the speaker and the
hearer.
Bearing this in mind, there is a need in DRT for
representing the differing beliefs of agents in dia-
logue and their beliefs (meta-beliefs) about other
agents’ beliefs or mental state. By focussing on
the intentions of speakers and hearers and infer-
ring agents’ intentions in making an utterance, the
approach presented in this paper aims at fulfilling
this need. It follows that, to have a ‘full’ theory of
beliefs and to have an insight into the mental states
of agents in dialogue (the speaker and the hearer),
it is necessary to have a representation of agents’
beliefs, degrees of beliefs, and the dialogue acts
expressed by their utterances (Asher 1986). This is
also in order to strengthen the link between utter-
ances and agents’ intentions in dialogue. The di-
alogue act or function performed by the utterance
tells us something about the speaker’s beliefs. Fur-
thermore, what is also needed is a representation
of beliefs that are shared between, or are common
to, the two agents.
The question is: how can DRT best model be-
liefs? The following section, 2, outlines the prob-
lems facing modelling beliefs in DRT. Section 3
presents a graded view of agents’ beliefs in dia-
logue as a solution to these problems. This is fol-
lowed by a description of the relationship between
belief and mutual belief, section 4, and then of the
relationship between belief and dialogue acts, sec-
tion 5.
2 Problems Facing Modelling Beliefs in
DRT
According to Heydrich et al. (1998), paradigms of
dynamic semantics (DRT, Situation Semantics and
Dynamic Predicate Logic) face three obstacles in
modelling dialogue. First, there is the problem of
adapting the paradigm, originally made to model
monological discourse, to the description of dia-
logue with different agents. The second problem
is the description of mental states and the beliefs
of the agents. The third problem is in explaining
how the mental states are related to overt linguistic
behaviour.
With respect to the first problem, DRT has grad-
ually attempted to address problems of belief rep-
resentation in dialogue. For example, in Prole-
gomena, Kamp introduces a simple model of ver-
bal communication (Kamp 1990: 71), which con-
sists of two agents, A and B, and their mental
states K(A) and K(B). Later work by Kamp et
al. (2005) introduces agent modelling for single-
sentence discourse, namely the hearer. The treat-
ment presented in this paper allows the represen-
tation of dialogue with different agents, thus, ad-
dressing the first problem identified by Heydrich
et al. (1998).
With regard to the second problem, however,
DRT has been primarily concerned with repre-
senting utterances containing propositional atti-
tudes such as ‘believe’, rather than the beliefs and
meta-beliefs of agents. Segmented-DRT (SDRT)
has mainly focused on belief update and revision
(Asher and Lascarides 2003). The treatment in
this paper takes previous work on beliefs in dy-
namic semantics as a starting point and extends it
to reach a richer representation of the interaction
between mental states and the linguistic content of
utterances. For example, both speaker and hearer
mental states are represented and the beliefs and
47
meta-beliefs of agents are reviewed after each ut-
terance.
As a semantic theory, DRT tells us which dis-
course referents are needed in context. However,
DRT does not deal with planning, nor with prag-
matic aspects of contexts rendered through re-
lating the current utterance to agents’ intentions.
Kamp et al.’s (2005) expansion of the original,
also known as ‘vanilla’, DRT (Poesio and Traum
1997a), deal minimally with intentions. To deal
with the third problem mentioned by Heydrich et
al., Al-Raheb (2005) has already outlined a prag-
matic extension to DRT that makes it appropriate
for linking the current utterance and agents’ inten-
tions.
The present paper aims to show how that link
can be strengthened through modelling agents’ in-
tentions and relating them to the dialogue acts
communicated via utterances. In relation to this
link, the significance of degrees of belief is ex-
plained in the following section.
drs1:
i you m
drs2:
s
c1: buy(you,c2)
c2: newShoes(s)
attitude(you, ‘ACCEPT’, drs3)
drs3:
attitude(i, ‘ACCEPT’, drs2)
attitude(i, ‘BEL’, drs4)
drs4:
y
b1: mary(m)
b2: party(y)
b3: has(m,y)
attitude(you, ‘BEL’, drs5)
drs5:
s
b4: mary(m)
b5: party(y)
b6: has(m,y)
b7: buy(you,b8)
b8: newShoes(s)
attitude(you, ‘INT’, drs6)
drs6:
y s
p1: mary(m)
p2:party(y)
p3: has(m,y)
a1: buy(you,a2)
a2: newShoes(s)
inform(you, i, a1)
Figure 1: Hearer Recognition of S1
3 Degrees of Belief
To our knowledge, there is no account in DRT
that accommodates strengths or degrees of belief
of agents in dialogue. This section addresses this
gap and proposes initially two strengths of belief
involved in dialogue to be expanded in future re-
search to include further degrees of belief. Modal
expressions, including words such as ‘possibly’
and ‘might’, are evidence that there exist more de-
grees of belief than the ones discussed in this pa-
per.
The beliefs of an agent are ‘her model of how
things are’ (Traum 1994: 15). The notion of belief
(or strong belief) is to be understood in relation
to the agent: it is what the agent takes to be true.
There is an important philosophical background to
the discussion of ‘belief’ and ‘knowledge’. It is
outside the scope of this paper to review all the
literature here. Quine (1960), Hintikka (1962),
Lewis (1969, 1979), and Davidson (1983) are rep-
resentative. The term ‘belief’ is understood in this
paper to refer to propositions strongly held by the
agent to be true and when making utterances relat-
ing to them, the speaker not only commits herself
to their truth but also communicates to the hearer
that she, the speaker, believes those proposition to
be true.
Another degree of belief called acceptance is
accounted for in this model. Acceptance consists
of the agent’s weakly believed propositions. The
agent may be going along with what the speaker
is saying or has acquired a new proposition based
on the speaker’s utterance which has not yet been
confirmed into a stronger belief.
To illustrate what is meant by the distinction be-
tween belief and acceptance, let us look at:
(4) S1: I need to buy new shoes for Mary’s
party.
H1: Try Next on Henry Street.
The speaker tells the hearer that she has to buy
new shoes for Mary’s party. In this example, the
hearer already (strongly) believes there is a party
and he suggests a place where the speaker can buy
them. Figure 1 demonstrates the hearer’s mental
state after hearing the speaker’s utterance, S1. The
hearer’s mental state is represented by a Discourse
Representation Structure (DRS), which contains
three sub-DRSs, one for intention (referred to by
‘attitude(you, ‘INT’, drs6)’ and the label for the
intention DRS, drs6), another for the belief DRS
containing strong beliefs (referred to by ‘attitude(i,
‘BEL’, drs4)’ and the the label for the belief DRS,
drs4), and finally the acceptance DRS contain-
ing weak beliefs (referred to by ‘attitude(i, ‘AC-
CEPT’, drs2)’ and the the label for the acceptance
48
DRS, drs2).3
If we change example (4) so that the hearer does
not actually hold the belief that there is a party, as
in:
(5) S1: I need to buy new shoes for Mary’s
party.
H1: I didn’t realize Mary is throwing a
party.
S2: Yeah she is. It’s next Tuesday.
H2: You can probably buy them at Next.
The hearer does not necessarily need to strongly
believe that Mary is throwing a party. He can ‘go
along with’ or accept it and even suggest a place
where the speaker can buy the shoes. The exis-
tence of a party does not affect the hearer person-
ally or directly, i.e. he does not need to act on
it. However, let us now consider the effect if we
change the example again so that the hearer does
not know about Mary’s party, nor that he is re-
quired to buy new shoes, as in:
(6) S1: You need to buy new shoes for
Mary’s party.
H1: I didn’t realize Mary is throwing a
party.
S2: Yeah she is. You should try Next on
Henry Street.
H2: I will.
This time, for the hearer to commit to buying
something for a party (in H2) that he did not even
know existed suggests a stronger degree of belief
than that of ‘going along with’ the speaker having
to buy it. The existence of the party affects the
hearer personally and directly. Therefore, agree-
ing to buy new shoes justifies the inference that he
believes rather than just accepts there is a party.
This is what the paper describes as belief, or a
strong degree of belief. Contrast Figure 1 with the
figure representing the speaker’s mental state after
hearing H2 in example 6, Figure 2.
4 Beliefs and Mutual Beliefs
The treatment of beliefs that we are developing
here requires an explicit account of how the be-
lief spaces or DRSs of two agents can interact.
3Inside the agent’s DRS, ‘i’ is used to refer to the agent
and ‘you’ is used to refer to the other agent. Assertions are
marked by ‘an’, presuppositions by ‘pn’, believed informa-
tion by ‘bn’ and accepted information by ‘cn’.
drs1:
i you m
drs2: attitude(you, ‘ACCEPT’, drs3)
drs4:
attitude(i, ‘ACCEPT’, drs2)
attitude(i, ‘BEL’, drs4)
drs4:
s n
b1: mary(m)
b2:party(y)
b3: has(m,y)
b4: buy(you,b5)
b5: newShoes(s)
b6: try(you,b7)
b7: next(n)
attitude(you, ‘BEL’, drs5)
drs5:
b8: mary(m)
b9:party(y)
b10: has(m,y)
b11: buy(you,b12)
b12: newShoes(s)
b13: try(you,b13)
b14: next(n)
attitude(you, ‘INT’, drs7)
drs7:
s n
p1: newShoes(s)
p2: next(n)
a1: buy(you,p1)
a2: try(you,p2)
inform(you,i,a1)
inform(you,i,a2)
Figure 2: Speaker Recognition of H2
‘Mutual belief’, also referred to as ‘mutual knowl-
edge’, is the term used by Traum (1994) among
others, where a group of individuals may believe
X, where X may or may not be true. Stalnaker’s
(2002) ‘common belief’ is comparable to what
others call mutual belief. For X to be a mutual
belief, it has to be accessible to a group; all be-
lieve X and all believe that all believe X, and all
believe that all believe that all believe X.
In face-to-face communication, the hearer be-
lieves that the speaker believes what she, the
speaker, is communicating. On the other hand, un-
less the hearer indicates doubt or objects to what
the speaker is saying, the speaker assumes that the
hearer believes what the speaker has said – which
is consistent with expectations under Gricean co-
operativeness assumptions (1989). The speaker
also assumes that the hearer now has the belief that
the speaker believes what she just said. This as-
sumption is what leads to ‘mutual’ beliefs (Kamp
1990: 79).
However, mutual belief can be viewed as the
process of establishing that the speaker and the
hearer hold the same belief. One way in which
this process may occur is when the speaker holds
a belief and communicates that belief to the hearer.
49
This belief may then be adopted by the hearer who
can provide feedback to the speaker that the infor-
mation communicated has now acquired the status
of belief in an ideal situation with a cooperative
hearer. When both participants reach the conclu-
sion that S bel(ieves) X, H bel X, H bel S bel X,
and S bel H bel X, then mutual belief is estab-
lished. The speaker in example (7) believes her
neighbour is a weirdo. Whether the utterance is
informative (new) or not depends on the context.
In this example, (7), the speaker may not already
have the belief that the hearer believes her neigh-
bour is a weirdo.
(7) Speaker: My neighbour is such a
weirdo.
Hearer: Yeah, he is. I saw him peeping
through your window the other day.
However, after the hearer makes his utterance, the
speaker can now strongly believe that the hearer
believes her neighbour is a weirdo, that he believes
she believes her neighbour is a weirdo, and now
she believes he believes her neighbour is a weirdo.
Figure 3 shows the level of nesting to accommo-
date the mutual belief that the speaker’s neighbour
is a weirdo. It is possible when this level of nesting
is reached to have a separate DRS or space for mu-
tual beliefs, called ‘mutual belief DRS’. In which
case, the propositions held in drs6, can now be
removed from drs6 and added to the ‘mutual be-
lief DRS’. Figure 3 represents the speaker’s men-
tal state after the hearer makes his utterance. For
the purposes of this example, the DRT represented
in Figure 3 will mainly focus on the speaker’s be-
lief DRT.
Achieving mutual belief is immensely helped
by dialogue acts. For example, when a hearer
provides strong feedback about a new proposition
(cf. drs7 in Figure 3), the speaker can come to
believe the hearer believes that proposition. Sec-
tion 5 shows the importance of considering the di-
alogue acts expressed by an assertion (new infor-
mation) and their relationship to degrees of belief
and strengthening of beliefs.
5 Beliefs and Dialogue Acts
When someone makes an assertion, they commu-
nicate not only information they assume to be new
to the hearer, but also communicate to the hearer
information about their own beliefs. In order to
drs1:
i you
drs2:
attitude(you, ‘ACCEPT’, drs3)
drs3:
attitude(i, ‘ACCEPT’, drs2)
attitude(i, ‘BEL’, drs4)
drs4:
x y
b1: neighbour(x)
b2: have(i,x)
b3: weirdo(x)
b4: window(y)
b5: have(i,y)
b6: peeping-through(x,y)
b7: saw(you,b6)
attitude(you, ‘BEL’, drs5)
drs5:
b8: neighbour(x)
b9: have(i,x)
b10: weirdo(x)
b11: window(y)
b12: have(i,y)
b13: peeping-through(x,y)
b14: saw(you,b13)
attitude(i, ‘BEL’, drs6)
drs6:
b15: neighbour(x)
b16: have(i,x)
b17: weirdo(x)
b18: window(y)
b19: have(i,y)
b20: peeping-through(x,y)
b21: saw(you,b20)
attitude(you, ‘INT’, drs7)
drs7:
x y
p1: neighbour(x)
p2: weirdo(x)
p3: window(y)
p4: have(i,y)
a1: peeping-through(x,y)
a2: saw(you,a1)
strongPosFeedback(you,i,p2)
inform(you,i,a2)
Figure 3: Speaker Recognition
model beliefs in dialogue, it is necessary to un-
derstand what the representation of dialogue in-
volves. A dialogue is ‘a cooperative undertaking
of agents engaged in developing and transform-
ing their common situation’, involving verbal and
non-verbal action (Heydrich et al. 1998: 21). In
a dialogue, utterances give rise to dialogue acts
(cf. agents’ intention DRSs in Figures 1, 2 and
3), named speech acts by some, and conversation
acts by others (Traum 1994).
One of the features of dialogue acts is how they
affect the agents’ mental states. As Traum points
out, ‘... speech acts are a good link between the
mental states of agents and purposeful communi-
cation’ (Traum 1999: 30). Each agent in dialogue
needs to have a representation of their beliefs and
the other agent’s beliefs or cognitive state in or-
der for a dialogue act to be felicitous in Austin’s
and Searle’s sense (Asher 1986). That is to say,
dialogue acts depend on agents’ beliefs for inter-
pretation.
50
Each assertion made has one ‘function’ or more.
For example, the function of a statement could be
to make a claim about the world. Traum (1997) di-
vides statements into ‘assert’, ‘re-assert’, and ‘in-
form’. ‘Assert’ is trying to ‘change’ the belief of
the addressee. The result of assert is that the hearer
now assumes that the speaker is trying to get the
hearer to believe the assertion. ‘Re-assert’ can be
used when participants try to verify old informa-
tion, and not necessarily inform of something new.
‘Inform’ means that the speaker is trying to pro-
vide the hearer with information that the hearer
did not have before. However, Traum does not
go further to discuss cases where agents believe
their utterances (Traum 1994: 14). It is one of the
claims of this paper that agents in dialogue either
strongly or weakly believe their utterances in order
to be cooperative. It is possible to extend this ap-
proach in order to include cases where agents are
purposefully deceitful. However, this is left for fu-
ture research.
The adapted dialogue acts, or functions, in thiss
paper’s treatment of beliefs in DRT are mainly
‘inform’, ‘change belief’ and ‘other’. ‘Inform’
is used to communicate new information to the
hearer, whereas ‘change belief’ (or to use Poesio
and Traum’s (1997b) dialogue act term ‘assert’)
is used to change the hearer’s beliefs about some
proposition. The importance of the representation
introduced in section 3 in relation to dialogue acts
transpires in allowing us to make the distinction
between the dialogue acts ‘inform’ and ‘change
belief’ (‘assert’). To ‘inform’ the hearer of X, the
speaker needs to have the belief in her beliefs that
the hearer does not believe X, i.e. bel(S,¬ bel(H,
X)). This is a constraint to making an informative
utterance. Figure 4 shows the speaker’s beliefs be-
fore making the utterance in example (8).
(8) The X-Files DVD is on sale on Amazon.
The speaker believes the hearer does not al-
ready believe that the X-Files DVD is on sale on
Amazon, drs3. This is demonstrated by the miss-
ing propositions representing ‘on sale on Amazon’
‘onSale(x, b4)’ and ‘at(a)’ from drs3 in Figure 4.
On the other hand, to make a ‘change belief’ or
an ‘assert’, the speaker would have reason to be-
lieve that the hearer believes something different
or the opposite of what the speaker believes, bel(S,
bel(H, ¬ X)). The DRT treatment of beliefs pro-
posed in this paper allows us to reflect this in
drs1:
i you x a
attitude(i, ‘BEL’, drs2)
drs2:
b1: xFilesDVD(x)
b2: amazon(a)
b3: onSale(x, b4)
b4: at(a)
attitude(you, ‘BEL’, drs3)
drs3:
b5: xFilesDVD(x)
Figure 4: Inform: Speaker’s utterance
drs1:
i you x a
attitude(i, ‘BEL’, drs2)
drs2:
b1: xFilesDVD(x)
b2: amazon(a)
b3: onSale(x, b4)
b4: at(a)
attitude(you, ‘BEL’, drs3)
drs3: b5: xFilesDVD(x)
b6: not(onSale(x))
Figure 5: Change belief
Figure 5, drs3, in which the speaker believes the
hearer believes the X-Files DVD is not on sale,
‘not(onSale(x))’.
The category ‘Other’ embraces any dialogue act
other than ‘inform’ and ‘change belief’, whose
recognition involves the same process explained
for others, e.g. ‘suggest’, ‘clarify’, and ‘explain’.4
The dialogue acts ‘accept’ and ‘reject’ come un-
der the umbrella of feedback as they can be in re-
sponse to, for instance, a ‘suggest’ dialogue act.
The dialogue act ‘clarify’ is used when a hearer
is having difficulty recognizing the speaker’s ut-
terance.5 On the other hand, ‘explain’ is when
the speaker responds to the hearer’s clarification
request and provides a clarifying utterance. The
hearer can accept, believe, or reject that explana-
tion. The dialogue act ‘suggest’ also instigates one
of three reactions: the hearer can accept, believe or
reject that suggestion and may provide feedback to
indicate which is his reaction. It is of more inter-
est to this paper to examine the effects of dialogue
acts on the hearer’s beliefs, and what dialogue acts
suggest about the speaker’s beliefs.
4It is possible for this category to be expanded to in-
clude more dialogue acts such as ‘question’, ‘answer’, ‘self-
correct’ and ‘offer’.
5Clarification is a form of feedback. ‘I didn’t hear what
you said’ is both ‘feedback’ act and an ‘inform’ (Schegloff et
al. 1977).
51

	



	








  






Figure 6: Feedback
5.1 Feedback and Agents’ Beliefs
Traum (1994) suggests that when an assertion is
made, the hearer has an obligation to produce
an ‘understanding act’. In general, acknowledge-
ment is expected in Traum’s treatment of speech
acts. This means that when a hearer responds with
‘okay’, the hearer can be taken to be providing
an acknowledgement and an acceptance. How-
ever, the hearer does not always provide feedback.
Grounding often happens as a result of implicit
rather than overt feedback and acknowledgement
(Bunt 1995).6 In fact, the treatment outlined in
this paper maintains that the lack of feedback is to
be considered a form of ‘weak positive feedback’,
an extension to Dynamic Interpretation Theory’s
(DIT) positive feedback (Bunt 1995). The hearer
does not object to the speaker’s utterance by not
providing feedback, since if the hearer did object,
he would explicitly do so.
When the speaker makes an assertion, the
hearer may indicate that the message has been re-
ceived (weak positive feedback), example (9.b).
Weak positive feedback may indicate understand-
ing, continued attention, or acknowledgement,
such as ‘uh huh’, and ‘yeah’ (Clark and Schaefer
1989). Another case of weak positive feedback is
provided by example (9.a) where the hearer does
not say anything. It is assumed that the hearer did
not have any problems and has received the asser-
tion, A. In the case of weak feedback, it can be
argued that this represents the ‘acceptance’ of A.7
Another response for the hearer is ‘strong posi-
6Grounding is a term adapted by Traum (1994) from
Clark and Schaefer’s (1989) work on establishing common
ground.
7This does not cancel cases where for social reasons, such
as politeness, the hearer does not necessarily agree with the
speaker, but does not wish to indicate it. The speaker can
wrongly or rightly come to the conclusion that the hearer ac-
cepts the assertion.
tive feedback’ (another extension to DIT’s positive
feedback), where the hearer not only indicates re-
ception of A, but also that she agrees that A (cf.
drs7 Figure 3). This is where confirming adoption
of new beliefs takes place, example (9.c). Reject-
ing A is another way of giving feedback, negative
feedback, as in example (9.d).
(9) Speaker: Mary loves John.
a. Hearer:
b. Hearer: aha.
c. Hearer: I couldn’t agree more!
d. Hearer: No, Mary is besotted with
Tom!
There are also degrees of belief that can be ex-
pressed according to the speech act used, firm ver-
sus ‘tentative’. Poesio and Traum pay less atten-
tion to ‘the attitudes expressed by the acts’ (Poesio
and Traum 1998: 221). Unlike Traum’s model, the
effects of the dialogue acts’ employed in agents’
DRSs on agents’ beliefs are considered in this pa-
per. Figure 6 demonstrates the link between feed-
back dialogue acts and agents’ beliefs.
6 Conclusion
As this paper has demonstrated, beliefs vary in
strength according to context. Beliefs also change
with the coming of new information. The DRT
treatment discussed here allows for the represen-
tation of strong beliefs and weaker beliefs as well
as changes to beliefs. Agents in a dialogue may
form stronger beliefs as the dialogue progresses,
requiring moving the content of their weaker be-
liefs to the stronger belief space.
In sum, there is no account in standard DRT that
accommodates degrees of belief of agents in dia-
logue. This paper has addressed this omission and
suggested two degrees of belief involved in dia-
logue, namely ‘belief’ and ‘acceptance’. It is sug-
52
gested that this is the initial step in representing
agents’ mental states in dialogue-oriented DRT.
However, this paper does not deal with words
which introduce more degrees of belief than the
two addressed in the model. It would be interest-
ing to see more degrees of belief represented in a
DRT dialogue model of agents in future research.
It is possible that such modal expressions can be
arranged on a scale corresponding to degrees of
belief (cf. Werth 1999). Moreover, this paper has
accounted for agent’s mutual beliefs and linked
agents’ beliefs and intentions to the dialogue acts
of their utterances, in order to address the prob-
lematic nature of accounting for belief in DRT.

References
Al-Raheb, Y. 2005. Speaker/Hearer Representation in a Dis-
course Representation Theory Model of Presupposition: A
Computational-Linguistic Approach. Phd. University of
East Anglia.
Asher, N. 1986. ‘Belief in Discourse Representation The-
ory’. Journal of Philosophical Logic 15, pp. 127–189.
Asher, N. and Lascarides, A. 2003. Logics of Conversation.
Cambridge: Cambridge University Press.
Bunt, H. 1995. ‘Dynamic Interpretation and Dialogue The-
ory’. In: M. Taylor, F. Neel, and D. Bouwhuis (Eds.). The
Structure of Multimodal Dialogue, Volume 2. pp. 139–166.
Amsterdam: John Benjamins 2000.
Clark, H. and Schaefer, E. 1989. ‘Contributing to Discourse’.
Cognitive Science 13, pp. 259–294.
Davidson, D. 1983. ‘A Coherence Theory of Truth and
Knowledge’. In: D. Henrich (Ed.). Kant oder Hegel. pp.
433–438. Stuttgart: Klett-Cotta Buchhandlung.
Gazdar, G. 1979. ‘A Solution to the Projection Problem’.
In: C. Oh and D. Dineen (Eds.). Syntax and Semantics II:
Presupposition. New York: Academic Press.
Grice, P. 1989. Studies in the Way of Words. Cambridge,
MA: Harvard University Press.
Heydrich, W., Kuhnlein, P., and Rieser, H. 1998. ‘A DRT-
Style Modelling of Agents’ Mental States in Construction
Dialogue’. In: Proceedings of Workshop on Language
Technology 13 (Twendial ’98), TWLT. Faculty of Infor-
matics, the University of Twente: The Netherlands.
Hintikka, J. 1962. Knowledge and Belief: An Introduction to
the Logic of the Two Notions. Mimeo: Indiana University
Linguisitics Club.
Horton, D. and Hirst, G. 1988. ‘Presuppositions as Beliefs’.
In: Coling-88: Proceedings of the 12th International Con-
ference on Computational Linguistics. pp. 255–260. Bu-
dapest: Hungary.
Kamp, H. 1990. ‘Prolegomena to a Structural Account of Be-
lief and Other Attitudes’. In: C. Anderson and J. Owens
(Eds.). Propositional Attitudes: The Role of Content in
Logic, Language, and Mind. Stanford, CA: CSLI Publi-
cations.
Kamp, H., van Genabith, J., and Reyle, U. 2005.
The Handbook of Logic. Unpublished Manuscript.
http://www.ims.uni-stuttgart.de/˜hans/.
Larsson, S. and Traum, D. 2000. ‘Information State and Dia-
logue Management in the TRINDI Dialogue Move Engine
Toolkit’. In: Natural Language Engineering. Special Is-
sue on Spoken Language Dialogue System Engineering.
pp. 323–340.
Lewis, D. 1969. Convention: A Philosophical Study. Har-
vard University Press.
Lewis, D. 1979. ‘Attitudes de dicto and de re’. Philosophical
Review 88, pp. 513–543.
Poesio, M. and Traum, D. 1997a. ‘Conversational Actions
and Discourse Situations’. Computational Intelligence 13,
pp. 309–347.
Poesio, M. and Traum, D. 1997b. ‘Representing Conversa-
tion Acts in a Unified Semantic/Pragmatic Framework’.
In: Working Notes of AAAI Fall Symposium on Com-
municative Action in Humans and Machines. pp. 67–74.
Cambridge, MA: MIT Press.
Poesio, M. and Traum, D. 1998. ‘Towards an Axiomatization
of Dialogue Acts’. In: J. Hulstijn and A. Nijholt (Eds.).
Formal Semantics and Pragmatics of Dialogue, Proceed-
ings of Twendial’ 98. pp. 207–221. Universiteit Twente:
Enschede.
Quine, W. 1960. Word and Object. Cambridge MA: MIT
Press.
Schegloff, E., Jefferson, G., and Sacks, H. 1977. ‘The Prefer-
ence for Self-Correction in the Organization of Repair in
Conversation’. Language 53, pp. 361–382.
Stalnaker, R. 1974. ‘Pragmatic Presupposition’. In: M. Mu-
nitz and P. Unger (Eds.). Semantic and Philosophy. pp.
197–214. New York: New York University Press.
Stalnaker, R. 1988. ‘Belief Attribution and Context’. In: R.
Grimm and D. Merrill (Eds.). Contents of Thought. Pro-
ceedings of the 1985 Oberlin Colloquium in Philosophy.
pp. 140–156. Tucson State: The University of Arizona
Press.
Stalnaker, R. 1999. Context and Content. Oxford: Oxford
University Press.
Stalnaker, R. 2002. ‘Common ground’. Linguistics and Phi-
losophy 25(5-6), pp. 701–721.
Traum, D. 1994. A Computational Theory of Grounding in
Natural Language Conversation. Phd and tr 545. Com-
puter Science Department, Univeristy of Rochester.
Traum, D. 1997. ‘Report on Multiparty Dialogue Sub-group
on Forward-looking Communicative Function’. In: Stan-
dards for Dialogue Coding in Natural Language Process-
ing, Dagstuhl-Seminar Report no. 167.
Traum, D. 1999. ‘Computational Models of Grounding in
Collaborative Systems’. In: Working notes of AAAI Fall
Symposium on Psychological Models of Communication.
pp. 124-131. North Falmouth, Massachusetts.
Werth, P. 1999. Text Worlds: Representing Conceptual Space
in Discourse. New York: Longman.
