A Process Model for Recognizing 
Communicative Acts and Modeling 
Negotiation Subdialogues 
Sandra Carberry* 
University of Delaware 
Lynn Lambert t 
Christopher Newport University 
Negotiation is an important part of task-oriented expert-consultation dialogues. This paper 
presents a plan-based model for understanding cooperative negotiation subdialogues. Our sys- 
tem infers both the communicative actions that people pursue when speaking and the beliefs 
underlying these actions. Beliefs, and the strength of these beliefs, are recognized from the surface 
form of utterances,from discourse acts, and from the explicit and implicit acceptance of previous 
utterances. Our algorithm for recognizing discourse actions combines linguistic, world, and con- 
textual knowledge in a unified framework. By combining these different knowledge sources, we 
are able to recognize complex discourse acts such as expressing doubt, to identify the relationship 
of utterances to one another, and to model negotiation subdialogues. Since negotiation is an inte- 
gral part of multiagent activity, our process model addresses an important aspect of cooperative 
interaction and thus is a step toward an intelligent and robust natural language consultation 
system. 
1. Introduction 
In a typical expert-consultation dialogue, one participant (hereafter referred to as the 
executing agent or EA) has a goal that he 1 wants to achieve and is working with 
the other participant (referred to as the consulting agent or CA) to construct a plan 
for achieving this goal. Although both the plan construction process and the con- 
versation are collaborative activities, this does not mean that people always believe 
what they are told. In fact, part of the collaborative activity of conversation is negoti- 
ation of conflicting beliefs. This negotiation is particularly important in task-oriented 
expert-consultation dialogues, since the participants must resolve any conflicting be- 
liefs in order to work together effectively to devise a plan that is both well-formed and 
addresses the executing agent's needs. Thus, a robust natural language consultation 
system must be able to handle negotiation subdialogues. 
Even though there is wide agreement that negotiation is an integral part of multi- 
agent activity, previous natural language understanding systems have been unable to 
handle negotiation subdialogues such as the following: 
(1) 
(2) 
$1: Who is teaching CS360? 
$2: Dr. Smith is teaching CS360. 
* Department of Computer Science, Newark, DE 19716, USA 
Department of Physics, Computer Science, and Engineering, Newport News, VA 23606, USA 1 For exposition purposes, we will use the masculine gender when referring to EA and the feminine 
gender when referring to CA. 
(~) 1999 Association for Computational Linguistics 
Computational Linguistics Volume 25, Number 1 
(3) 
(4) 
(5) 
(6) 
$1: But isn't CS360 an undergraduate course? 
$2: Yes. CS360 is an undergraduate course. 
Dr. Smith teaches both graduate and undergraduate courses. 
$1: Who handles the CS360 lab? 
For example, existing systems do not recognize when an agent is expressing doubt 
at a previous response as in utterance (3), when an agent is attempting to resolve a 
conflict suggested by the other participant as in utterances (4)-(5), or when an agent 
is implicitly conveying acceptance of a communicated proposition as in utterance (6). 
These shortcomings prevent existing natural language systems from being able to 
handle dialogues in which one agent initially does not accept the proposition conveyed 
by the other agent and initiates a negotiation subdialogue to resolve their differences 
in belief. 
We have developed a plan-based model of dialogue that addresses these limita- 
tions. Our analysis of naturally occurring dialogue indicates that one way that people 
express doubt at a proposition Pdoubt is by contending that some other conflicting 
proposition Pi is true. Our process model includes an algorithm for recognizing such 
expressions of doubt, as well as other complex discourse acts. The algorithm uses 
a multistrength belief model and a combination of linguistic, world, and contextual 
knowledge. Our implemented system can recognize implicit as well as explicit accep- 
tance of a communicated proposition, multiple expressions of doubt at the same propo- 
sition, expressions of doubt at both immediately preceding and earlier utterances, and 
negotiation subdialogues embedded within other negotiation subdialogues. 
In the remainder of this paper, we describe our system and how this process is 
performed. Section 2 describes the kinds of expressions of doubt found in our cor- 
pus analysis, and Section 3 discusses the factors that must be taken into account in 
recognizing the kind of expression of doubt that we have been studying. Section 4 
presents our process model for recognizing complex discourse acts (such as expres- 
sions of doubt) and assimilating them into the dialogue context. First it discusses why 
it is necessary to capture varying degrees of belief, describes the multistrength model 
of belief used in our system, and discusses how our description of actions avoids 
assuming that a speaker will automatically adopt a communicated proposition. Then 
it introduces the notion of an action that requires evidence for its recognition and 
presents our recognition algorithm that uses a combination of linguistic, world, and 
contextual knowledge. Section 5 steps through an extended example that illustrates 
our system's ability to recognize complex discourse acts and model negotiation sub- 
dialogues. Section 6 discusses the evaluation of our system and our plans for future 
work, and Section 7 discusses related research. The examples in this paper are taken 
from a university advisement domain, since this is the domain in which we have 
implemented our system. 
2. Motivation from Naturally Occurring Dialogues 
To identify how speakers express doubt, we analyzed a corpus of naturally occurring 
dialogues in the domains of financial planning, university courses, real estate, pets, 
taxes, and travel. The real estate, pets, and financial planning (Harry Gross Transcripts 
1982) dialogues were transcribed from radio talk shows, the taxes and travel (SRI 
Transcripts 1992) dialogues were transcribed from tapes of simulated interactions, 
and the university courses dialogues (Columbia University Transcripts, 1985) were 
Carberry and Lambert Modeling Negotiation Subdialogues 
transcribed from student advisement sessions. In the corpus we found instances in 
which a speaker expressed doubt at a proposition by contending that some other 
conflicting proposition was true. 2 In addition, we extracted other examples of such 
expressions of doubt from the dialogues in novels. These kinds of expressions of 
doubt can be realized as surface negative questions or tag questions and are often 
accompanied by the cue word but, as in the following example taken from the Harry 
Gross financial planning dialogues (Harry Gross Transcripts 1982) in which S2's last 
utterance expresses doubt at Sl's recommendation: 
$1: 
$2: 
$1: 
$2: 
I would like to see that into an individual retirement account rollover in 
a mutual fund group. 
At my age? 
Yes. 
Uh, yeah but isn't there any risk? 
However, our corpus analysis also provided instances where surface negative and tag 
questions were used to seek verification, such as in the following excerpt from the set 
of financial planning dialogues: 
$1: 
$2: 
$1: 
And if you have more money left after you pay the taxes what difference 
does it make if you pay a few bucks more in taxes? 
I'm telling my wife but she won't listen. 
Well maybe she'll listen to me. If you get 200 bucks--isn't it better to 
have 200 bucks and uh have 200 left than to have nothing at all? 
Our recognition algorithm has only been concerned with recognizing instances in 
which a speaker expresses doubt by contending that some other proposition is true. 
However, in our corpus, speakers also expressed doubt in the following ways, and 
our future work will include extending our system to handle these: 3 
Drawing attention to an inconsistent feature or proposition: The 
speaker brings into focus a feature or proposition that is already part of 
the dialogue context but that is intended to discredit the proposition 
being doubted. These utterances were often realized as an elliptical 
fragment and included "And you're how old?', "Even though it's four more 
years?", and "At my age?" 
Drawing attention to violated expectations: The speaker mentions an 
expectation that is inconsistent with the doubted proposition. An 
example of this from our corpus is "You're kidding, what happened to the 
seventy-eight dollar fares or those sort of things ?" 
2 We found a very few instances in which the speaker asked the hearer if he was sure the conflicting 
proposition wasn't true; an example from our corpus is "Are you sure he didn't name himself as attorney for 
the estate?" taken from the Harry Gross financial planning dialogues (Harry Gross Transcripts 1982). 
Our current system does not handle such expressions of doubt. 
3 Walker (1996) analyzed the Harry Gross financial planning dialogues (Harry Gross Transcripts 1982) to 
identify features that distinguish acceptance from rejection. However, she did not consider expressions 
of doubt and some of her rejections would fall into our "express doubt" category. 
3 
Computational Linguistics Volume 25, Number 1 
• Repetition: The speaker queries the doubted proposition; this usually 
took the form of a declarative followed by a question mark, as in "You 
have 40 thou(sand) inn a ram fund?" 
• Explicit statements and questions: The speaker explicitly doubts what 
has been said or asks for justification; examples from our corpus include 
"I'm not so sure of that." and "Who ever said that?" 
• Cue words: The speaker uses discourse markers to convey his doubt. In 
addition to the cue word but that is often used to realize expressions of 
doubts in the other categories, cue words such as Really? and What? 
were used by themselves to express doubt and cue words such as even 
though were used to convey doubt in utterances such as "Even though it's 
four more years ?" 
3. Recognizing Expressions of Doubt 
When a listener in a collaborative interaction does not accept the proposition that the 
speaker is trying to convey, a negotiation subdialogue ensues in which the participants 
attempt to "square away" (Joshi 1982) their disparate beliefs. Negotiation subdialogues 
often involve complex discourse acts that implicitly refer to some proposition that is 
part of the existing dialogue context. We have found that such complex discourse 
acts require evidence for their recognition. Since the motivation for our work is the 
recognition of communicative acts that occur in negotiation subdialogues, this section 
examines in detail how plausibility and evidence affect the recognition of one kind of 
complex discourse act, an expression of doubt. 
For a listener to recognize a discourse action, it must be plausible that the speaker 
holds the requisite beliefs for performing the action. (A belief is plausible if the avail- 
able evidence does not refute it.) For example, in order for a listener to interpret an 
utterance as felicitously asking a question in order to obtain information, it must be 
plausible that the speaker does not already know the information and that the speaker 
believes that the listener may be able to provide it. Similarly, to interpret an utterance 
as expressing doubt at a proposition Pdoubt by contending Pi, it must be plausible that 
the speaker holds certain beliefs. Consider a university setting in which each course 
has only one instructor and a speaker uses the proposition Pi, that Dr. Brown teaches 
Architecture, to express doubt at the proposition Pdoubt, that Dr. Smith is teaching Ar- 
chitecture, as in utterance (9) of Figure 1. In order to intend (9) as an expression of 
doubt in a collaborative dialogue, the speaker must believe 
1. 
2. 
3. 
that the hearer has some belief in Pdoubt 
that Pi is true 
that if Pi is true, then Pdoubt is not 
The hearer must be able to plausibly ascribe each of these beliefs to the speaker in 
recognizing the expression of doubt. First, it must be plausible that the speaker believes 
that the hearer has some belief in the proposition that is being doubted, since it is 
pointless in a collaborative dialogue to express doubt at something about which there 
is no disagreement. 4 It must also be plausible that the speaker has some belief in Pi, 
4 We are using "express doubt" in the sense of challenging the truth of a proposition. 
4 
Carberry and Lambert Modeling Negotiation Subdialogues 
(7) EA: What is Dr. Smith teaching? 
(8) CA: Dr. Smith is teaching Architecture. 
(9) EA: Isn't Dr. Brown teaching Architecture? 
Figure 1 
A dialogue with an expression of doubt. 
since otherwise the hearer could not be expected to believe that the speaker was using 
a conflict between Pi and Pdoubt to question the validity of Paoubt. Similarly, it must be 
plausible that the speaker believes that if Pi is true, then Pdoubt is not; otherwise the 
hearer could not be expected to think that the speaker believes that the truth of Pi 
raises doubts about Pdoubt. 
Although being able to ascribe beliefs as plausible is necessary for recognition of 
all discourse actions, some discourse actions, such as the expressions of doubt that we 
consider in this paper, require further evidence. This evidence is provided by linguis- 
tic, world, and contextual knowledge. These knowledge sources can either provide 
evidence for a generic discourse act (such as an expression of doubt) or evidence that 
the conditions are satisfied for performing a specific discourse act (such as expressing 
doubt that Dr. Smith is teaching Architecture). In addition, contextual knowledge can 
suggest a particular interpretation when equivalent evidence exists for several specific 
discourse acts. These knowledge sources are discussed in the next sections. 
3.1 Linguistic Knowledge 
3.1.1 Evidence for a Generic Discourse Act. A number of researchers (Reichman 
1978, 1985; Grosz and Sidner 1986; Polanyi 1986; Cohen 1987; Hirschberg and Litman 
1987; Litman and Allen 1987; Schiffrin 1987; Hinkelman 1989; Litman and Hirschberg 
1990; Knott and Dale 1994; Knott and Mellish 1996; Marcu 1997) have investigated 
the use in discourse of special words and phrases such as but, anyway, and by the way. 
They found that these clue words, or discourse markers, have a number of different 
functions, including indicating the role of an utterance in the dialogue, conveying the 
relationship between utterances, suggesting shifts in focus of attention, conveying the 
structure of the discourse, etc. 
Consider again the dialogue shown in Figure 1. If EA had followed (7)-(8) with 
(9a) 
(9)a. EA: Isn't Architecture one of our required courses? 
then EA's utterance would not be interpreted as expressing doubt but would instead 
be understood as merely seeking information about the Architecture course. However, 
if this utterance is preceded by the clue word but, as in (9b) below, 
(9)b. EA: But isn't Architecture one of our required courses? 
then the utterance is expressing doubt, though we have difficulty ascertaining the 
reason for this doubt--perhaps EA believes that Dr. Smith does not teach courses that 
students are required to take! 
Thus, clue words comprise one source of evidence in the recognition of discourse 
acts. In particular, a clue word can provide evidence for a generic discourse act, such 
as Express-Doubt, but it remains for other sources to resolve what is being doubted. 
Computational Linguistics Volume 25, Number 1 
3.1.2 Evidence for a Specific Discourse Act. Expressions of doubt do not always in- 
clude clue words, as illustrated by utterance (9) in Figure 1. In the absence of a clue 
word, we need evidence that the speaker holds the three beliefs, listed earlier in Sec- 
tion 3, for performing a specific discourse act. Evidence for the second belief (that the 
speaker believes that Pi is true) is often provided by the surface form of the utterance, 
such as an utterance of the form "Isn't Pi ?"--for example, "Isn't Dr. Brown teaching Ar- 
chitecture?" in (9). This surface form indicates a strong belief in the queried proposition 
while a simple yes-no question, such as "Is Dr. Brown teaching Architecture?", does not. 
Therefore, if EA were to follow (7)-(8) with "Is Dr. Brown teaching Architecture?", EA 
would seem to have a misconception that more than one person can teach a course 
or perhaps be seeking information in order to subsequently express doubt--but the 
utterance itself is not conveying doubt that Dr. Smith is teaching Architecture. Thus 
the surface form of the utterance is one source of evidence that the speaker holds the 
requisite beliefs for performing a specific discourse act. 
3.2 World Knowledge 
World knowledge in the form of stereotypical beliefs is another source of evidence 
that the speaker holds the requisite beliefs for a particular discourse act. For example, 
world knowledge can provide evidence for the third speaker belief, that if Pi is true, 
then Pdoubt is not. Suppose that it is stereotypically believed that prestigious fellow- 
ships are awarded for sabbaticals, that faculty on sabbatical do not teach, and that 
faculty only teach in their area of expertise. Consider the dialogue shown in Figure 2. 
After (13), there are two propositions that have been conveyed by CA but not yet 
completely accepted by EA: the proposition that Dr. Smith is not on sabbatical and 
the proposition that Dr. Smith is teaching CS360, cormnunicated by utterances (13) 
and (11), respectively. A subsequent utterance might express doubt at one of these 
propositions or might forego the opportunity to doubt them, perhaps by pursuing 
some discourse act unrelated to either of the propositions. Consider the following 
three possible continuations of the dialogue: 
(14)a. 
b. 
C. 
EA: Wasn't Dr. Smith awarded a Fulbright? 
EA: Isn't Dr. Smith a theory person? 
EA: Isn't Dr. Smith an excellent teacher? 
While (14a) and (14b) seem to be expressing doubt, (14c) is simply seeking further 
information about Dr. Smith. The reason for this difference in interpretation is that 
in the case of (14a) and (14b), evidence from world knowledge suggests that EA 
believes that Pi (the proposition that EA contends is true) implies that one of the two 
open propositions is false, whereas no such evidence exists in the case of (14c). In 
the case of (14a), since it is stereotypically believed that prestigious fellowships are 
awarded for sabbaticals, EA's utterance should be interpreted as expressing doubt at 
the proposition that Dr. Smith is not on sabbatical. In the case of (14b), since Dr. Smith 
being a theory person is an alternative to Dr. Smith being a systems person and it is 
stereotypically believed that being a systems person is necessary for teaching CS360 
(a systems course), EA's utterance would instead be interpreted as expressing doubt 
at the proposition that Dr. Smith is teaching CS360. Thus, world knowledge in the 
form of stereotypical beliefs is another source of evidence that the speaker holds the 
requisite beliefs for performing a particular discourse act. 
If EA had uttered (14c), EA's utterance would be interpreted as merely seeking 
new information since there is no domain knowledge suggesting that EA believes that 
Dr. Smith being an excellent teacher contributes to determining whether Dr. Smith is 
Carberry and Lambert Modeling Negotiation Subdialogues 
(10) EA: Who is teachir~g CS360 (a systems course)? 
(11) CA: Dr. Smith is teaching CS360. 
(12) EA: Isn't Dr. Smith on sabbatical? 
(13) CA: No, Dr. Smith is not on sabbatical. 
Figure 2 
A dialogue with two open propositions. 
on sabbatical or to identifying the instructor of CS360. Note that (14c) demonstrates 
why plausibility alone is insufficient for recognition. Although there is no evidence 
that EA believes that Dr. Smith being an excellent teacher implies that Dr. Smith is 
on sabbatical or that Dr. Smith is not teaching CS360, there is also no evidence to the 
contrary, and thus it is plausible that EA believes that Dr. Smith being an excellent 
teacher indicates that he is on sabbatical or that he is not teaching CS360. This is not 
sufficient, however, to interpret (14c) as an expression of doubt. 
3.3 Contextual Knowledge 
An agent can infer from a dialogue many of the beliefs of the other participant. 
These acquired beliefs about the other participant's beliefs form one kind of contex- 
tual knowledge that can be used as evidence for the beliefs listed above. In addition, 
contextual knowledge determines the salience (or degree of prominence) of propo- 
sitions at the current point in the dialogue, and salience is a factor that constrains 
the interpretation of coherent discourse actions. Consider the first three utterances in 
the dialogue shown in Figure 2. EA's acceptance of CA's telling of the proposition 
that Dr. Smith is teaching CS360 establishes the mutual belief that CA believes that 
Dr. Smith is teaching CS360 and thus provides evidence for the first belief; 5 in addi- 
tion, the proposition that Dr. Smith is teaching CS360 becomes salient and is added to 
the dialogue context. Thus, while an utterance such as (12a) 
(12)a. EA: Doesn't Dr. Smith usually teach theory courses? 
might be used following (11) to express doubt at the statement that Dr. Smith is 
teaching CS360, it cannot be used following (11) to express doubt at the proposition 
that Dr. Smith teaches CS410 because 1) there is no reason for EA to believe that CA 
has any belief in the proposition that Dr. Smith teaches CS410, and 2) the proposition 
that Dr. Smith teaches CS410 is not salient at this point in the dialogue. 
In addition, contextual knowledge plays two other roles in the recognition of 
discourse acts. First, in the case of expressions of doubt, contextual knowledge dis- 
tinguishes propositions that have not yet been accepted by the speaker and thus are 
open for rejection. Consider again the dialogue in Figure 2. After (13), there are two 
propositions that have not yet been accepted by EA and are thus open for rejection 
by EA. If EA were to continue with (14b), repeated below, 
(14)b. EA: Isn't Dr. Smith a theory person? 
then EA would again be expressing doubt at the proposition that Dr. Smith is teach- 
ing CS360 and would have implicitly conveyed acceptance of the proposition that 
5 Note that here EA is only accepting CA's felicitous telling of the proposition, but EA is not adopting the proposition as one of his own beliefs. 
7 
Computational Linguistics Volume 25, Number 1 
Dr. Smith is not on sabbatical. Thus, as the conversation continues, only one propo- 
sition would remain open for rejection: the proposition that Dr. Smith is teaching 
CS360. This claim is supported by a combination of 1) the stack paradigm (Polanyi 
1986; Reichman 1978; Grosz and Sidner 1986; Litman and Allen 1987), which treats 
topic structure as following a stack-like discipline; 2) focusing heuristics (McKeown 
1983) that suggest that if a speaker has more to say about a topic, then he should do 
so before moving back to a topic deeper on the stack; and 3) the notion of implicit 
acceptance (discussed in Section 4.6) that argues that passing up the opportunity to 
reject an assertion in a collaborative dialogue communicates acceptance of it. 
Second, contextual knowledge orders propositions according to their relative salience 
in the current dialogue. This salience can be used to arbitrate among discourse acts 
for which there is equivalent evidence. Consider again the dialogue in Figure 2 and 
suppose that EA had continued with (14d). 
(14)d. EA: But isn't Dr. Smith an excellent teacher? 
Here we have a clue word suggesting an expression of doubt, but the speaker could be 
expressing doubt either that Dr. Smith is not on sabbatical or that Dr. Smith is teaching 
CS360. In both cases, we lack evidence for the third speaker belief. Contextual knowl- 
edge suggests that, all other things being equal, the proposition being doubted is the 
proposition that Dr. Smith is not on sabbatical, since it is the most salient proposition 
that is open for rejection at this point in the dialogue. Thus, contextual knowledge 
arbitrates when equivalent evidence is available for several specific discourse acts. 
3.4 Summary 
In addition to the requisite speaker beliefs being plausible and the constraints on the 
discourse act being satisfied (such as the constraint that a proposition be salient at the 
current point in the dialogue), certain discourse acts require additional evidence for 
their recognition. Two kinds of evidence that may be used in recognizing discourse ac- 
tions are 1) evidence (such as a clue word) for a generic discourse act, and 2) evidence 
that a speaker holds the requisite beliefs for performing a particular discourse act. 
Evidence for these beliefs can come from linguistic, world, or contextual knowledge. 
Although we have illustrated each of these knowledge sources by showing how they 
might provide evidence for one of the requisite beliefs for expressing doubt, it should 
be noted that each knowledge source might also be used as evidence for other be- 
liefs required for expressing doubt or for beliefs for other discourse acts. For example, 
although it does not generally arise in the kind of interactive dialogues that we are 
studying, world knowledge in the form of stereotypical beliefs might be used as evi- 
dence that a speaker believes that a hearer has some belief in the doubted proposition 
Pdoubt" 
4. The Process Model 
Grosz and Sidner (1986) claim that a robust model of understanding must use multi- 
ple knowledge sources in order to recognize the complex relationships that utterances 
have to one another. We have developed an algorithm that combines linguistic, world, 
and contextual knowledge, such as that identified in Section 3, in order to recognize 
complex discourse acts, including one kind of expression of doubt. Linguistic knowl- 
edge consists of clue words and the surface form of the utterance; world knowledge 
includes a set of stereotypical beliefs that users generally hold and recipes for perform- 
ing discourse acts; and contextual knowledge consists of a model of the user's beliefs 
8 
Carberry and Lambert Modeling Negotiation Subdialogues 
acquired from the preceding dialogue, the current structure of the discourse, the ex- 
isting focus of attention (that aspect of the task on which the participants' attention is 
currently centered), and the relative salience (degree of prominence) of propositions 
in the discourse. 
The remainder of this section presents the core ideas of our algorithm. Section 4.1 
shows why it is necessary to capture varying degrees of belief in a proposition and 
presents the multistrength belief model used in our system. Section 4.2 describes our 
representation of recipes for actions and shows how our recipe for an Inform action 
refrains from assuming that the listener will adopt the communicated proposition as 
part of his own beliefs; it also presents a recipe for expressing doubt and shows 
how constraints on the speaker's beliefs are captured in the recipe's applicability con- 
ditions. Section 4.3 gives an overview of our dialogue model. Section 4.4 describes 
how chaining is used to hypothesize a sequence of higher-level discourse acts that a 
speaker may be performing; Section 4.5 introduces the notion of a discourse action 
that requires evidence for its recognition; Section 4.6 discusses how our model recog- 
nizes implicit acceptance of a discourse act; and Section 4.7 presents our recognition 
algorithm that uses a combination of linguistic, world, and contextual knowledge in 
recognizing discourse acts. 
4.1 A MultiStrength Belief Model 
As argued in Section 3, if a speaker is expressing doubt at a proposition Pdoubt by 
contending some other proposition Pi, then the speaker must have some belief in 
Pi. Evidence for this belief is often provided by the surface form of the speaker's 
utterance, such as an utterance of the form "Isn't Pi?" for example, "Isn't Dr. Brown 
teaching Architecture?" But if we treat such an utterance as conveying certain belief 
that Pi is true, then we cannot handle situations in which an utterance such as this is 
merely requesting verification since a speaker cannot felicitously seek verification of 
a proposition that he already knows is true. Therefore, since modeling only ignorance 
and certainty is inadequate for recognizing complex discourse acts, it is necessary to 
model the strength of an agent's beliefs. 
4.1.1 Representing Varying Degrees of Belief. We use a multistrength model of belief, 
which captures not only ignorance and certainty about the truth of a proposition but 
also several degrees of belief in between. Utterances of the form "Isn't Pi ?" are treated 
as conveying a strong (but uncertain) belief in Pi. In this way, our system is able 
to handle instances in which an utterance of the form "Isn't Pi?" is used to request 
verification (since the utterance is viewed as conveying uncertainty about Pi) as well 
as instances in which it is used to express doubt (since the utterance is viewed as 
conveying some belief in Pi). 
Our multistrength belief model maintains three degrees of belief: certain belief 
(a belief strength of C); strong but uncertain belief, as in "Isn't Dr. Brown teaching 
Architecture?" (a belief strength of S); and weak belief, as in "I think that Dr. Cayne might 
be an education instructor" (a belief strength of W). Three degrees of disbelief (indicated 
by attaching a subscript of N, such as SN to represent strong disbelief and WN to 
represent weak disbelief) are also maintained, and one degree indicating no belief 
about a proposition (a belief strength of 0). We adopted three degrees of positive 
and negative belief in our model because that was the minimum number of degrees 
required for modeling the beliefs communicated in the dialogues that we examined 
and in the negotiation subdialogues that our system is intended to handle. 
Although an agent has some specific strength of belief in a proposition, the other 
agent may not always know precisely what that strength of belief is but may be able 
9 
Computational Linguistics Volume 25, Number 1 
to bound it--for example, he may be able to say that the first agent has some belief 
in a proposition. Our belief model uses belief intervals to capture this, where a belief 
interval specifies the range of strengths within which an agent's beliefs are thought to 
fall. 
Allen and Perrault (1980) noted the need to represent an agent's wanting to know 
the referent of a term in a proposition, without having to specify what that referent 
was. For example, if EA asks CA "Who is teaching CS360?', we cannot represent CA's 
belief that EA wants to know the teacher of CS360 as 
believe(CA, want(EA, believe(EA, Teaches(Dr.Smith, CS360)))) 
since this representation says that CA believes that EA wants to believe that the teacher 
is Dr. Smith (but EA, in asking the question, may not be predisposed to any such belief 
and may in fact reject "Dr. Smith" as the answer to the question). Allen and Perrault 
addressed this with knowref and knowif predicates, which represented an agent's know- 
ing the referent of a term in a proposition and knowing whether a proposition is true. 
Thus CA's belief that EA wants to know the teacher of CS360 in the above example 
might be represented as 
believe(CA, want(EA, knowref(EA, _fac, Teaches(_fac, CS360)))) 
In our multistrength belief model, knowref is treated as being certain about the referent 
of the term in the specified proposition and knowif is treated as being certain about 
whether a proposition is true or false. 
As the dialogue progresses, the belief model must capture the changing beliefs of 
the user. When a discourse act has been successful, its goals can be used to update 
the belief model. For example, if the user explicitly accepts the proposition conveyed 
by the utterance "Dr. Smith is teaching Architecture" (perhaps by saying "Yes, I'I1 accept 
that"), then the system can update its belief model to include the belief that the user 
himself believes that Dr. Smith is teaching Architecture. However, explicit acceptance 
is less common than implicit acceptance; Section 4.6 discusses implicit acceptance and 
its recognition. 
Since we do not currently have a response generation component, our system 
processes the utterances of both participants, alternating between playing the role of 
CA and the role of EA. Note that this differs from playing the role of a third-party 
observer--when the system plays the role of EA, the system has access to EA's beliefs 
(including EA's beliefs about the current dialogue model and EA's beliefs about CA), 
and when the system plays the role of CA, it has access to CA's beliefs. However, 
whenever the system assumes the role of a participant and processes a new utterance, 
it is assumed that this participant has correctly interpreted previous utterances and 
has a correct model of the preceding dialogue. 
4.1.2 Related Work on Modeling Belief. Young (1987) built a model in which the 
beliefs of the user are part of an explicit, missing, or stereotype module (he used 
the system's beliefs as a stereotype). Although this system provides needed differ- 
entiation among beliefs that the system knows the user holds, those that the system 
has attributed to the user, and those about which the system has no knowledge, this 
model still does not contain degrees of partial belief that are essential for modeling 
discourse acts such as expressing doubt. Ballim and Wilks (1991) developed a nested 
belief model that captures an agent's beliefs about other agents' beliefs. Their system 
combines belief ascription based on stereotypes with belief ascription based on per- 
turbations of the system's own beliefs, but they do not represent how strongly a belief 
is held. 
10 
Carberry and Lambert Modeling Negotiation Subdialogues 
Galliers (1991, 1992) has specified a nonnumeric theory of belief revision that 
relates strength of belief to persistence of belief. She points out that a belief model for 
communication must contain a multistrength model of beliefs that can be modified 
as the conversation proceeds. She uses endorsements (Cohen 1985) in an assumption- 
based truth maintenance system (ATMS \[DeKleer 1986\]) to specify a system that orders 
beliefs according to how strongly they are held. This ordering is used to calculate which 
beliefs should be revised when beliefs are challenged in the course of conversation. 
Walker (1991, 1992) has examined dialogues in which people repeat what they 
already know either in question or statement form (e.g., "I have four children." "OK. 
Four children."). Walker claims that this repetition by the second speaker is given so 
that the first speaker realizes that her utterance was understood and believed. That 
is, cooperative listeners often provide some evidence to speakers to indicate that the 
listener believes the speaker's claims. Like Galliers, Walker has based the strength 
of belief on the amount and kind of evidence available for that belief. Cohen and 
Levesque (1991a) also found this kind of corroboration. Our work has not investigated 
the belief reasoning process or how much evidence is needed for an agent to have a 
particular amount of confidence in a belief. Instead, we have been concerned with 
taking into account different communicated strengths of belief and the impact that the 
different belief strengths have on the recognition of discourse acts. 
Our multistrength belief model is very simple and is only intended to meet our sys- 
tem's need for representing how strongly an agent holds a particular belief. We recently 
became aware of work by Driankov on a logic in which belief/disbelief pairs capture 
how strongly a proposition is believed (Driankov 1988; Bonarini, Cappelletti, and Cor- 
rao 1990). 6 This work appears to be the only formally defined and well-developed logic 
that models strength of belief. With the exception that Driankov's logic does not in- 
chide a state of weak belief, it appears to provide the representational and reasoning 
capability needed by our system and we intend to investigate it for future use. 
4.2 Discourse Recipes 
In previous work, we noted the need to differentiate among domain, problem-solving, 
and discourse actions (Lambert and Carberry 1991; Elzer 1995). In task-oriented con- 
sultation dialogues, the participants are constructing a plan for achieving some domain 
goal, such as owning a home, and the resultant plan will consist of domain actions 
such as applying for a mortgage. In order to construct the domain plan, the partici- 
pants pursue problem-solving actions such as evaluating alternative domain actions 
or correcting an action in the partially constructed domain plan. Domain and problem- 
solving actions have been investigated by many researchers (Allen and Perrault 1980; 
Perrault and Allen 1980; Wilensky 1981; Litman and Allen 1987; van Beek and Cohen 
1986; Ramshaw 1989; Carberry 1990). 
Discourse actions are communicative actions that are executed by the dialogue 
participants in order to obtain or convey the information needed to pursue the problem- 
solving actions necessary for constructing the domain plan. Examples of very different 
discourse actions include answering a question, informing, and expressing doubt. Al- 
though our system models domain, problem-solving, and discourse actions, this paper 
is only concerned with recognizing discourse acts, particularly complex discourse acts 
such as expressing doubt. 
Our system's knowledge about how to perform actions is contained in a library 
6 We would like to thank one of the anonymous reviewers and Ingrid Zukerrnan for bringing this work to our attention. 
11 
Computational Linguistics Volume 25, Number 1 
of discourse, problem-solving, and domain recipes (Pollack 1990). Our representation 
of a recipe includes a header giving the action defined by the recipe, the recipe type, 
preconditions, applicability conditions, constraints, a body, effects, and a goal. The 
recipe type is primitive, specialization, or decomposition. If the recipe type is primitive, 
then the body of the recipe is empty and the header action corresponds with a primitive 
action in the domain. In a specialization recipe, the body gives a set of alternative ways 
of performing the header action (Pollack 1990; Kautz 1990). For example, one might 
earn credit in a course either by taking the course for credit or getting credit by exam. 
In a decomposition recipe, the body gives a set of subactions that constitute performing 
the header action. A # preceding a subaction in the body of a decomposition recipe 
indicates that the subaction can be performed any number of times (including zero). 7 
Constraints limit the allowable instantiation of variables in each component of a recipe 
(Litman and Allen 1987). For example, a variable might have a constraint requiring 
that it be instantiated with a proposition that is salient at the current point in the 
discourse. Which instantiations of variables will satisfy the constraints is part of the 
shared knowledge of the participants. 8 
Applicability conditions (Carberry 1987) are conditions that must be satisfied in 
order for a recipe to be reasonable to pursue in a given situation. The applicability 
conditions of our discourse recipes capture attitudes (beliefs and wants) that the agent 
of the action must hold in order for it to be felicitous (Searle 1970). Applicability 
conditions differ from preconditions in that one can plan to satisfy preconditions but 
it is generally anomalous to try to satisfy applicability conditions. For example, in order 
for _agent1 to inform _agent2 of _proposition, _agent1 must believe that _proposition 
is true and must not believe that _agent2 already believes _proposition. It would be 
anomalous for _agent1 to try to adopt the proposition as one of his beliefs solely for the 
sake of being able to inform someone else of it, and similarly it would be anomalous 
for _agent1 to get _agent2 to disbelieve a proposition so that _agent1 can subsequently 
inform him of it. 9 
Belief intervals are used in the applicability conditions to specify the range of 
strengths that an agent's behefs may assume. Intervals such as \[bi:bj\] specify 
a strength of belief between bi and bj inclusive. For example, Figure 3 displays the 
recipes for the Inform and Tell discourse acts. The goal of the Inform action, 
believe (_agent2, _proposition, \[C : C\] ), is that _agent2 be certain that _proposition 
is true. On the other hand, believe(_agent2, _proposition, \[CN:S\]) means that 
_agent2 is not convinced that _proposition is true (i.e., _agent2 could have any be- 
lief ranging from being certain that _proposition is false to having a strong belief 
that it is true). Thus the applicability condition believe (_agent l, believe (_agent2, 
_proposition, \[CN : S\] ), \[0 :C\] ) of the Inform act in Figure 3 means that _agent I must 
7 In this work, we are interested in understanding, not generating, responses. However, a generation 
system would pursue an action preceded by # if its applicability conditions are satisfied and the 
system does not believe that the action's goal will be satisfied if the action is omitted. The belief 
reasoning techniques described in Cawsey et al. (1992) can be used in modeling this. 8 In this research, we have assumed that the participants have equivalent knowledge of language and 
maintain equivalent discourse models, and we have not addressed the problem of recognizing 
miscommunication. 9 How applicability conditions on discourse acts are checked during planning is an interesting question 
that requires further research. Consider an agent who wants to determine whether a proposition is 
true; the agent might accomplish this by asking another agent about the proposition. An applicability 
condition on asking another agent is that the speaker wants to know the other agent's belief about the 
proposition. But when did this want come into existence? It certainly must be satisfied at the time the 
question is asked, but instead of being part of the initial state, it appears to result from the speaker's 
decision about how to obtain the desired information. 
12 
Carberry and Lambert Modeling Negotiation Subdialogues 
Discourse Recipe 
Action: Inform(_agentl, _agent2, _proposition) 
{_agent1 informs _agent2 of_proposition} 
Recipe-Type: Decomposition 
Appl Cond: believe(_agentl, _proposition, \[C:C\]) 
believe(_agentl, believe(_agent2, _proposition, \[CN:S\]), \[0:C\]) 
Body: Tell(_agentl, _agent2, _proposition) 
#Address-Behevability(_agentl, _agent2, _proposition) 
Effects: believe(_agent2, want(_agentl, believe(_agent2, _proposition, \[C:C\])), 
\[c:c\]) 
Goal: believe(_agent2, _proposition, \[C:C\]) 
Discourse Recipe 
Action: Tell(_agentl, _agent2, _proposition) 
{_agent1 tells _agent 2 of_proposition} 
Recipe-Type: Decomposition 
Appl Cond: believe(_agentl, _proposition, \[C:C\]) 
Body: Surface-Say-Prop(_agentl, _agent2, _proposition) 
#Address-Understanding(_agentl, _agent2, _proposition) 
Effects: told-about(_agentl, _agent2, _proposition) 
Goal: beheve(_agent2, believe(_agentl, _proposition, \[C:C\]), \[C:C\]) 
Figure S 
Recipes for Inform and Tell discourse acts. 
either be ignorant about _agent2's belief in _proposition or have some belief (possi- 
bly certain) that _agent2 is not already certain that _proposition is true, i.e., _agentl 
does not believe that _agent2 is already convinced that _proposition is true. In de- 
termining the belief strengths specified in the applicability conditions, we examined 
the beliefs that an agent must hold and tried to identify the minimum and maximum 
strength of belief that would make the discourse act reasonable to pursue. 
We have divided the effects of an action into two subclasses: 1) the results of 
correctly performing the action, which are labeled effects, and 2) the desired effects 
of the action (over which the agent may lack control), which are labeled goals. For 
example, in the case of domain actions, the effect of applying for graduate study is 
that one has applied, while the goal is that one be accepted for graduate study. This 
distinction between effects and goals is particularly important in the case of discourse 
actions, where the agent cannot be assured that an action will have its intended result. 
Variables in recipes are represented as lowercase strings preceded by an under- 
score, with the string reflecting the variable's type; for example, _course1 and _course2 
refer to variables of type course. 
In Allen's seminal model of plan recognition (Allen 1979), the bodies of opera- 
tors could contain either goals to be achieved or action names with parameters. In 
our system, preconditions are represented as goals to be achieved, while the bodies 
of recipes specify actions. Since the recipe for each action in our recipe library con- 
tains a single goal in its goal field, this suffices--the goal makes clear the purpose of 
the action in a plan. However, in a richer domain where an action could be used to 
achieve several different goals, 1° it would be necessary to specify the intended goal in 
the recipe body and chain from it to the desired action in order to capture the moti- 
vation for performing the action. During plan recognition, our system matches goals 
10 For example, one might read a book to gain knowledge or to entertain oneself. 
13 
Computational Linguistics Volume 25, Number 1 
against preconditions of other actions, and it matches actions against the subactions 
in the bodies of the recipes for other actions. The Effects field is not used for chaining 
during plan recognition; however, as discussed in Section 4.6.2, the effects and goals 
in discourse recipes are used for updating a model of the user's beliefs. 
4.2.1 Formulation of Discourse Recipes. The bodies of our discourse recipes are based 
on work by other researchers (Allen and Perrault 1980; Searle 1970; Cohen and Perrault 
1979), dialogues in which we have participated, the naturally occurring dialogues that 
we examined, and our hypotheses about how our system might be expanded in the 
future. For example, consider the discourse recipes for Obtain-Info-Ref, Ask-Ref, and 
Answer-Ref shown in the appendix. To obtain information about a proposition via dia- 
logue, _agent1 must ask another agent about the proposition (Ask-ReJ) and the second 
agent must provide the requested information (Answer-ReJ); this is typical of naturally 
occurring dialogue and is captured in the body of our Obtain-Info-Ref discourse act. The 
applicability conditions of the Obtain-Info-Ref act, such as the condition that _agent1 
believe that _agent2 knows the information, are based on criteria identified in Searle 
(1970), Cohen and Perrault (1979), and Allen and Perrault (1980). The body of the Ask- 
Ref action consists of making the request itself and making the request acceptable; this 
is because in our own interactions we have encountered situations in which an agent 
will make a request and then justify it to the listener. The applicability conditions of 
Ask-Ref refer to _agent2's beliefs about the proposition; for example, one applicability 
condition is that _agent1 wants to know the term that _agent2 believes will satisfy a 
proposition. We contend that this captures what a speaker wants in asking a listener 
about a proposition (i.e., the speaker wants to know the listener's beliefs about the 
proposition), and this formulation also allows the Ask-Ref to be used as a subaction 
of a Test-Knowledge act in a tutoring system. 11 Note that the fact that a speaker who 
is seeking information really wants to know correct information about a proposition 
was captured in the applicability conditions of the Obtain-Info-Ref discourse act. 
Some of our discourse recipes, such as Obtain-Info-Ref, include subactions by both 
the initiating agent and the other participant. This captures the intention of the initi- 
ating agent to perform his required subactions as well as the intention that the other 
agent follow through on her role in the plan for this action. Thus, once the second 
agent recognizes that the initiating agent wants to obtain information, the second 
agent will also recognize that the initiating agent intends for her to play the role of 
providing that information. While an agent can construct a plan that includes acts by 
another agent, the planning agent cannot guarantee the other agent's behavior and 
thus such discourse plans can fail. (See Chu-Carroll and Carberry \[1994\] for research 
on dialogues in which agents do not always follow through on their intended role yet 
still fulfill their collaborative responsibilities.) 
A different approach would be to maintain such knowledge about adjacency pairs 
(Schegloff and Sachs 1973) and expected continuations in a transition network separate 
from the discourse recipes, as was done by Reithinger and Maier (1995). The advantage 
of this approach is that fewer discourse recipes are needed and continuations are 
generalized. Such a representation would enable us to remove the actions that address 
acceptance from our discourse recipes but still capture the expectations for them in 
the transition network. However, the disadvantage of this approach is that the higher- 
level discourse act would no longer constrain the possible continuations. For example, 
11 Searle (1970) notes that there are two kinds of questions, ones whose objective is to obtain knowledge 
and ones whose objective is to test another's knowledge. 
14 
Carberry and Lambert Modeling Negotiation Subdialogues 
an Evaluate-Answer discourse act is an expected follow-up to an Answer-Ref when they 
are part of a higher-level Test-Knowledge discourse act but not when the Answer-Ref is 
part of an Obtain-Info-Ref discourse act. Further research is needed to identify the best 
mechanism for capturing the requisite discourse knowledge. 
4.2.2 The Inform Discourse Recipe. As noted by Grosz and Sidner (1990), the assump- 
tion that one participant will slavishly respond to the wishes of the other participant 
does not reflect collaborative interaction. In Cohen and Perrault's formulation of speech 
act operators (Cohen and Perrault 1979), the effect of an Inform was that the hearer 
believed that the speaker believed the proposition. He postulated a Convince act that 
would cause the hearer to believe the proposition, but this act was left undeveloped 
and its definition did not allow for the participants to negotiate their beliefs. The effect 
of an Inform act in Allen and Perrault's system (Allen and Perrault 1980) is that the 
hearer believes the communicated proposition--this definition would seem to say that 
the hearer always accepts the information provided by the speaker. 12 Although Allen 
and Perrault's model was only concerned with recognizing the intention to perform an 
Inform act, using his formulation to model negotiation dialogues (where Inform actions 
may not automatically accomplish their purpose) is problematic. In Perrault's (1990) 
persistence model of belief, the hearer adopts a communicated proposition unless he 
has evidence to the contrary, in which case his original belief persists. Thus, models 
such as Allen and Perrault's cannot account for a hearer who does not accept a com- 
municated proposition, and Perrault's model cannot account for a hearer who changes 
his beliefs about a proposition after negotiation. 
We want to overcome these limitations and be able to handle negotiation subdia- 
logues in which participants attempt to come to some agreement about their disparate 
beliefs. Thus, the body of the recipe for our Inform act (Figure 3) contains two subac- 
tions: one in which the speaker tells the hearer a proposition and a second in which 
the participants address the believability of the communicated proposition and try to 
come to agreement. In addition, as discussed in the preceding section, the effects of our 
discourse recipes are often different from the goals. Although this does not solve the 
problem of recognizing perlocutionary effects, it does allow us to capture the notion 
that one can, for example, perform an Inform act without the hearer adopting the com- 
municated proposition. Thus, the goal of a discourse recipe is a desired perlocutionary 
effect (an effect that the speaker wishes the action to have, e.g., believe (hearer, P, 
\[C:C\] ) in the case of an Inform action), and the effects of a discourse recipe are the 
illocutionary effects (that is, the effect that the speech act has when it is performed and 
recognized by the hearer, e.g., believe (hearer, want (speaker, believe (hearer, P, 
\[C : C\] ) ), \[C : C\] ) in the case of an Inform action). 
4.2.3 An Express-Doubt Discourse Recipe. Figure 4 presents our discourse recipe for 
expressing doubt. Note that its applicability conditions capture the requisite beliefs 
listed in Section 3. The second applicability condition in Figure 4 excludes certain 
belief in _proposition2; this is because the body of the recipe is an action of conveying 
uncertain belief in _proposition2 and represents instances where an expression of doubt 
is realized as a surface negative question or a tag question. Note also that one of the 
effects of the Express-Doubt discourse act is that the listener believes that the speaker 
wants to resolve the conflict between the two propositions and that the goal of the 
12 In personal communication, Allen has said that the effect of his Inform action was intended to capture the agent's goal in performing the action. In Allen (1979) he mentions the need for a Decide-to-Believe 
act, but nothing further is done with it. 
15 
Computational Linguistics Volume 25, Number 1 
Discourse Recipe 
Action: Express-Doubt(_agentl, _agent2, _proposition1, _proposition2) {_agent1 expresses doubt to _agent2 about _proposition1 by contending that _proposition2 
is true} 
Recipe-Type: 
Appl Cond: 
Constraints: 
Body: 
Effects: 
Goal: 
Decomposition 
believe(_agentl, believe(_agent2, _propositionl, \[S:C\]), \[S:C\]) 
believe(_agentl, _proposition2, \[W:S\]) 
believe(_agentl, _proposition2 ~ ~_propositionl, \[S:C\]) 
salient( _proposition1 ) 
Convey-Uncertain-Belief(_agentl, _agent2, _proposition2) 
believe(_agent2, believe(_agentl, _proposition1, \[SN:WN\]), \[S:C\]) 
believe(_agent2, believe(_agentl,_proposition2 ~ -~_propositionl, \[S:C\]), 
\[s:c\]) 
believe(_agent2, want(_agentl, Resolve-Conflict(_agent2, _agent1, 
_proposition1, _proposition2)), \[S:C\]) 
want(_agent2, Resolve-Conflict(_agent2, _agent1, _proposition1, 
_proposition2)) 
Figure 4 
A recipe for an Express-Doubt discourse act. 
Express-Doubt discourse act is that the listener also wants to resolve the conflict. The 
mutual desire for conflict resolution resulting from a successful Express-Doubt discourse 
act leads to a negotiation subdialogue (initiated by the Express-Doubt action). 
4.3 The Dialogue Model 
We maintain a structure called a dialogue model that captures the system's beliefs 
about the existing dialogue context. The discourse level of the dialogue model contains 
a tree structure called the discourse tree. Each node of the discourse tree represents 
a discourse or communicative act that has been initiated by one of the dialogue par- 
ticipants, and the children of a node represent discourse acts that are being pursued 
in order to perform the parent action. For example, asking and answering a ques- 
tion (Ask-Ref and Answer-ReJ) are part of obtaining information (Obtain-Info-ReJ) in 
the discourse tree in Figure 5. The lowest uncompleted action in the discourse tree is 
marked as the focus of attention; 13 it represents the first expectation for subsequent 
utterances. In Figure 5, the focus of attention is the Tell action. The active path consists 
of the sequence of actions along the path from the action that is the focus of attention 
to the root node. The actions on the active path provide successive expectations about 
the role of the next utterance in the dialogue; actions closer to the current focus of 
attention are regarded as more salient than those further back on the active path. 14 
For example, in the discourse tree of Figure 5, the first expectation is that if EA does 
not understand CA's previous statement, then EA will now choose to address her 
understanding of it and thereby contribute to the Tell act that is the existing focus 
of attention. The next expectation is that if EA has any doubt about the proposition 
conveyed by CA, then EA will choose to address its believability, thereby contributing 
to the Inform discourse act that is the next action on the active path. (As shown in 
Figure 3, Address-Understanding is a subaction in the recipe for the Tell discourse act, 
and Address-Believability is a subaction in the recipe for the Inform discourse act.) 
13 By uncompleted, we mean not as yet known to be completed. 14 The active path is a sequence of actions; this work has not considered multithreaded discourse, a topic 
that Ros6 et al. (1995) have begun to investigate. 
16 
Carberry and Lambert Modeling Negotiation Subdialogues 
Obtain-lnfo-Ref(EA, CA, _fac, TeachesC fac, CS360)) 
.' - e (EA, CA, _fac, Teaches(_fac, CS360)) I I Ask R f i 
Ref-Request(EA, CA, _fac, Teaches(_fac, CS360)) I 
\[ Surface-WH-Question(EA, CA, _fac, TeachesC.fac, CS360)) J 
EA: Who is teaching CS3607 
Key: 
* Current focus of attention 
Figure 5 
Sample discourse tree. 
I 
I Answer-Ref(CA, EA, fac, Teaches(_fac, CS360)) \[ 
Ilnform(CA, EA, Teaches(Dr.Smith, CS360)) \[ 
* \[Tell(CA, EA, Teaches(Dr.Smith, CS360)) I 
I Surface-Say-Prop(CA, EA, Teaches(Dr.Smith, CS360)) I 
CA: Dr. Smith is teaching CS360. 
4.4 Hypothesizing Discourse Acts by Chaining 
Our process model starts with the semantic representation of a new utterance and uses 
plan inference rules (Allen and Perrault 1980; Carberry 1988) along with constraint 
satisfaction (Litman and Allen 1987) to hypothesize chains of actions A1,A2 ..... An 
that the speaker might be intending to perform with the utterance. In such a chain, 
action Ai contributes to the performance of its successor action Ai+l. For example, the 
semantic representation of an utterance such as "Dr. Smith is teaching Architecture" is 
Surface-Say-Prop(_agentl, _agent2, Teaches(Dr.Smith, Architecture)) 
A Surface-Say-Prop is a subaction in the recipe for a Tell discourse act, which in turn is 
a subaction in the recipe for an Inform discourse act. Thus chaining from the surface 
utterance produces a sequence of hypothesized discourse acts, each of which plays a 
role in the performance of its successor on the chain. 
We have expanded on Litman and Allen's (1987) notion of constraint satisfaction 
and Allen and Perrault's (1980) use of beliefs. As described earlier, many of the appli- 
cability conditions in our discourse recipes are beliefs that the agent of the action must 
hold in order for the action to be felicitous. Our recognition algorithm requires that 
the system be able to plausibly ascribe these beliefs in hypothesizing a new action; if 
the belief ascription is implausible or if the constraints of the discourse recipe are not 
satisfied, the inference is rejected. 
4.5 Evidence Actions 
As we claimed in Section 3, actions such as the expressions of doubt in utterances 
(14a) and (14b) (repeated in Figure 6) require evidence for their recognition. Let us 
further examine why this is the case. Figure 7 illustrates a situation in which we 
have several discourse acts (with different degrees of salience) that the agent might 
be expected to pursue. The solid boxes show some of the actions that are part of the 
17 
Computational Linguistics Volume 25, Number 1 
(10) 
(11) 
(12) 
(13) 
EA: Who is teaching CS360 (a systems course)? 
CA: Dr. Smith is teaching CS360. 
EA: Isn't Dr. Smith on sabbatical? 
CA: No, Dr. Smith is not on sabbatical. 
(14)a. 
(14)b. 
(14)c. 
EA: Wasn't Dr. Smith awarded a Fulbright? 
EA: Isn't Dr. Smith a theory person? 
EA: Isn't Dr. Smith an excellent teacher? 
Figure 6 
A sample dialogue with surface negative questions. 
l action-l(EA, CA, PROPA) I 
l action-2(EA~ CAt PROPB) I "" 
¢....._ 
m action-3(EA, CA, PROPC) I 
"~.. 
'',, 
"''.., 
e-action(EA, CA,_propl, PROPD) j 
I surface-action(EA, CA, PROPD) r 
I ................ ._J 
Figure 7 
Relating an inference path to the existing dialogue context. 
existing dialogue context 15 and the dashed boxes show a chain of actions inferred from 
the new utterance. As depicted in the figure, the process model can infer a chain of 
actions starting with some surface speech act surface-action(EA, CA, PROPD), up to 
some other action, action (EA, CA, PROPD), up to some other action, e-action (EA, CA, 
_propl, PROPD). The action e-action contains two propositions. One of these, PROPD, 
is instantiated by chaining from the earlier action, act ion (EA, CA, PROPD). However, 
the other proposition, _propl, cannot be instantiated by plan chaining; it must be 
instantiated by unification with a proposition from the existing dialogue context. For 
example, if e-action is identified as contributing to action-3 in the existing dialogue 
context, then _propl might be instantiated as PROPC. On the other hand, if e-action 
is identified as contributing to action-i, then _propl might be instantiated as PROPA. 
Chaining might suggest three possibilities (e-action could contribute to action-3, or 
to action-I, or to neither of them), and the relative salience of the propositions in the 
existing dialogue context is not sufficient to identify this relationship. 
As a concrete example, consider again the dialogue segment in Figure 6 consisting 
of utterances (10)-(13) followed by one of utterances (14a)-(14c). After utterance (13), 
15 These are actions on the active path of the dialogue model; the actions that are deepest on the active 
path are closer to the current focus of attention and are therefore regarded as more salient. 
18 
Carberry and Lambert Modeling Negotiation Subdialogues 
the actions on the active path of the dialogue model include (among others): 
action-l: Inform(CA, EA, Teaches(Smith,CS360)) 
action-2: Address-Unacceptance(EA,CA, Teaches(Smith,CS360), On-Sabbatical(Smith)) 
action-3: Inform(CA, EA, ~On-Sabbatical(Smith)) 
If EA utters (14a), (14b), or (14c), three inference paths can be constructed: one that 
links up to action-I, one that links up to action-3, and one that does not link up to 
any action on the active path. If CA utters (14a), then CA's action should be identi- 
fied as contributing to action-3 above and thus the proposition being doubted should 
be recognized as ~On-Sabbatical(Smith), even though this proposition did not appear 
explicitly in EA's utterance. However, if EA utters (14b), then EA's action should be 
identified as contributing only to action-1 and the proposition being doubted there- 
fore should be recognized as Teaches(Smith, CS360); in this case, we are rejecting the 
inference path that links up to action-3 even though action-3 is more salient at this 
point in the dialogue. On the other hand, if EA utters (14c), then EA's action should be 
recognized as not contributing to any of the actions on the active path and interpreted 
as merely seeking information about Dr. Smith. Since chaining and salience alone are 
insufficient to identify the correct interpretation, we need some additional mechanism. 
We define an evidence-action (abbreviated e-action) to be an action that intro- 
duces a new parameter that cannot be directly instantiated by chaining from the ut- 
terance. We contend that such actions require evidence for their recognition. In our 
model, the relationship between _prop1 (the proposition whose instantiation must be 
inferred from the existing dialogue context) and PROPD (a proposition instantiated by 
chaining from the current utterance) is modeled in the applicability conditions of a 
recipe for e-action. For example, Express-Doubt, whose recipe was given in Figure 4, 
is an example of an e-action. The parameter _proposition1 cannot be instantiated 
from plan chaining from the surface utterance because _proposition1 does not ap- 
pear in the body of the Express-Doubt recipe; therefore, Express-Doubt is an e-action 
because it contains a parameter (_proposition1) that must be instantiated by unifica- 
tion with a proposition extracted from the existing dialogue context. The relationship 
that _proposition1 has to the proposition contained in the utterance that is expressing 
doubt is modeled in the last applicability condition of the Express-Doubt recipe (see 
Figure 4). This applicability condition states that the agent of the Express-Doubt ac- 
tion believes that _propesition2 implies that _proposition1 does not hold. As we've 
shown earlier, plan chaining and plausibility are insufficient for recognizing an Express- 
Doubt discourse act; evidence is required. 
4.5.1 Types of Evidence. Our recognition algorithm captures the kinds of evidence 
identified in Section 3: 1) evidence provided by world knowledge, contextual knowl- 
edge, and the surface form of the utterance indicating that the applicability conditions 
for an e-action are satisfied, and 2) linguistic evidence from clue words suggesting a 
generic discourse action. 
Grosz and Sidner (1986) claim that when evidence is available from one source, 
less evidence should be required from others. Thus, if there is evidence indicating 
that the applicability conditions for a discourse act hold, then less linguistic evidence 
suggesting the discourse act should be required. This is the case for interpreting (9) 
(repeated below) as an expression of doubt. 
(7) EA: What is Dr. Smith teaching? 
(8) CA: Dr. Smith is teaching Architecture. 
19 
Computational Linguistics Volume 25, Number 1 
(9) EA: Isn't Dr° Brown teaching Architecture? 
Even though there is no linguistic clue word suggesting an Express-Doubt discourse 
act, there is enough evidence from the surface form of the utterance and from world 
and contextual knowledge to correctly interpret (9) as an expression of doubt at the 
proposition conveyed by (8). 
Let us examine this evidence in more detail. The applicability conditions of the 
Express-Doubt discourse act (see Figure 4) specify that EA must have some belief in 
each of the following: 
a° 
b. 
C. 
that CA believes that Dr. Smith is teaching Architecture; 
that Dr. Brown is teaching Architecture; and 
that Dr. Brown teaching Architecture is an indication that Dr. Smith is 
not teaching Architecture. 
Belief (c) models how the proposition in utterance (9) (that Dr. Brown is teaching Ar- 
chitecture) relates to the proposition in the existing dialogue context (that Dr. Smith 
is teaching Architecture). Therefore, evidence that EA holds belief (c) (that Dr. Brown 
teaching Architecture is an indication that Dr. Smith is not teaching Architecture) is 
particularly significant since it shows how the utterance relates to the preceding dis- 
course. 
The system has evidence "for all three applicability conditions. The system's evi- 
dence that EA holds belief (a) is provided by beliefs derived from the goal of the Tell 
discourse act. In utterance (8), CA initiates a Tell discourse act as part of an Inform 
discourse act; thus immediately after (8), both the Inform and the Tell are part of the 
existing dialogue context. If (9) is indeed an expression of doubt, then it contributes to 
the Inform act by addressing the believability of the communicated proposition. In this 
case, EA will have implicitly conveyed that EA understood CA's previous utterance, 
i.e., EA will have passed up the opportunity to contribute to the Tell discourse act that 
is a child of the Inform act, and will have thereby implicitly conveyed that the Tell act 
was successful. This notion of implicit acceptance is discussed further in Section 4.6. 
Since the goal of CA's Tell act is that EA believe that CA believes that Dr. Smith teaches 
Architecture, the hypothesis that the Tell act has completed successfully (and therefore 
that its goal has been achieved) provides evidence that (a) is a belief held by EA. 
The surface form of (9) provides evidence that EA believes (b), since it conveys 
an uncertain but still strong belief that Dr. Brown is teaching Architecture. Finally, if 
the system's model of a stereotypical user indicates that users typically believe that 
each course has only one instructor, then this world knowledge provides evidence 
that EA believes (c). Thus, the system has evidence for all three of the applicability 
conditions. In addition, contextual knowledge indicates that the single constraint on 
the Express-Doubt discourse act is satisfied--namely, that proposition Pdoubt be salient 
at this point in the dialogue. Thus, the system would recognize (9) as an expression 
of doubt. 
However, "Isn't Dr. Smith an excellent teacher?" would not be recognized as an 
expression of doubt because the system would have no evidence that EA believes that 
being an excellent teacher suggests that Dr. Smith is not teaching Architecture. Thus, 
world knowledge helps the system to correctly differentiate between utterances that 
are expressions of doubt and those that are not. 
On the other hand, if there is sufficient linguistic knowledge suggesting a par- 
ticular discourse action, then the applicability conditions should be attributed to the 
20 
Carberry and Lambert Modeling Negotiation Subdialogues 
speaker as long as they are plausible. So, if the clue word but is used, then a nonac- 
ceptance discourse action such as expressing doubt should be easier to recognize (i.e., 
should require less evidence that the applicability conditions hold) than if the clue 
word is not present. Thus, if EA said "But isn't Dr. Smith an excellent teacher?", then 
even though there is no world knowledge indicating that all of the applicability con- 
ditions hold, the linguistic clue word is sufficient evidence to interpret this utterance 
as an expression of doubt. 
The concept of accommodation in conversation (Lewis 1979; Thomason 1990) (re- 
moving obstacles to the speaker's goals) suggests that a listener might recognize a 
surface negative question as an expression of doubt by accommodating a belief about 
some incompatibility between the proposition conveyed by the surface negative ques- 
tion and a proposition that might be doubted. But in the extreme case, this means 
that any surface negative question could be recognized as an expression of doubt. We 
contend that there should be evidence for such recognition. This is similar to Pollack's 
model of plan recognition (Pollack 1990) that can account for user misconceptions; in- 
stead of inferring a relationship between every query and the speaker's goal, Pollack 
requires that the system apply only well-motivated rules that hypothesize principled 
variations of the system's own beliefs and that the system treat as incoherent any 
queries that cannot be explained via these rules. (Pollack's example of incoherence is 
the query "I want to talk to Kathy, so I need to find out how to stand on my head.") In our 
model we look for evidence of incompatibility, and in our implemented system this 
evidence takes the form of stereotypical befiefs about the domain. While our imple- 
mentation does not include other means of deducing an incompatibility, they are not 
precluded by our theory but are left for future work. Moreover, it should be noted 
that if the speaker intends for the hearer to recognize the expression of doubt from the 
incompatibility between the doubted proposition and the proposition that the speaker 
is contending is true, then the speaker must believe that the belief about the incompat- 
ibility is a mutual belief. Our stereotypical befiefs fall into this category. In other cases, 
the clue word but causes the listener to accommodate a belief about incompatibility 
that he might not otherwise have done and thereby recognize the expression of doubt. 
So, in the case of complex discourse acts such as expressing doubt, the system 
should require evidence for recognizing the discourse act and should prefer to rec- 
ognize discourse acts for which there is multiple evidence: both linguistic clue words 
suggesting the generic discourse act and evidence suggesting that the applicability 
conditions for a particular discourse act are satisfied. However, the system should be 
willing to accept just one kind of evidence when that is all that is available. 
4.6 Implicit Acceptance 
In a collaborative task-oriented dialogue, the participants are working together to 
construct a plan for accomplishing a task. If the collaboration is to be successful, the 
participants must agree on the plan being constructed and the actions being taken to 
construct it. Thus, since a communicated proposition is presumed to be relevant to 
this plan construction process, the dialogue participants are obligated to communi- 
cate as soon as possible any discrepancies in belief about such propositions (Walker 
and Whittaker 1990; Chu-Carroll and Carberry 1995b) and to enter into a negotia- 
tion subdialogue in which they attempt to "square away" (Joshi 1982) their disparate 
beliefs. 
In our earlier work (Carberry 1985, 1989), we claimed that a cooperative participant 
must accept a response or pursue discourse goals directed toward being able to accept 
the response. As we noted there, this acceptance need not be explicitly communicated 
to the other participant; for example, failure to initiate a negotiation subdialogue con- 
21 
Computational Linguistics Volume 25, Number 1 
veys implicit acceptance of the proposition communicated by an Inform action. This 
notion of implicit acceptance is similar to an expanded form of Perrault's default rea- 
soning about the effects of an inform act (Perrault 1990). Our model captures this by 
recognizing implicit acceptance when an agent foregoes the opportunity to address 
acceptance of an action and moves on to pursue other discourse actions. 
4.6.1 Acceptance Actions. If a statement is intended to answer a question, the listener 
in a collaborative dialogue must note when he believes that the statement does not 
suffice as a complete answer. However, doing so implies that the listener believes the 
statement, since it is inefficient to address a proposition's completeness as an answer if 
one does not accept the proposition. Similarly, if a listener does not believe a commu- 
nicated proposition, he must convey this disagreement as soon as possible (Walker and 
Whittaker 1990). But by questioning the validity of a proposition, the listener conveys 
that he believes that he understood the utterance. As Clark and Schaefer (1989) note, 
by passing up the opportunity to ask for a repair, a listener conveys that he has un- 
derstood an utterance. Thus we hypothesize that listeners convey their acceptance (or 
lack of acceptance) in a multistage acceptance phase: 1) understanding, 2) believability, 
3) completeness. 16 
Acceptance can be communicated explicitly or implicitly. We include actions that 
address acceptance in the body of six of our discourse recipes. These recipes were 
selected because they allow us to capture acceptance of a question (the recipes for 
Ask-Ref and Ask-IJ), acceptance of the answer to a question (the recipes for Answer- 
Ref and Answer-IJ), and acceptance of a statement (the recipes for Inform and Tell). 
In this research, we have been primarily concerned with one aspect of acceptance: 
believing the proposition communicated by an Inform action. For example, the actions 
in the body of the Inform recipe (see Figure 3) are: 1) the speaker (_agent1) tells the 
listener (_agent2) the proposition that the speaker wants the listener to believe; and 
2) the speaker and listener address believability by discussing whatever is necessary in 
order for the listener and speaker to come to an agreement about this proposition. 17 
This second action, and the subactions executed as part of performing it, account for 
subdialogues that address the believability of the proposition conveyed in the Inform 
action. Other actions related to acceptance are captured in other discourse recipes. For 
example, the Tell action has a body containing a Surface-Say-Prop action and an Address- 
Understanding action; the latter enables both participants to ensure that the utterance 
has been understood. Similarly, the Answer-Ref action contains an Inform action and 
an Address-Answer-Acceptability action that ensures that the Inform action is sufficient 
to answer the question. Further research is needed to model the full range of actions 
that address acceptance and to recognize utterances resulting from them. 
The discourse tree reflects the order of acceptance actions. As discussed above, 
lack of understanding should be addressed before believability. This is reflected in the 
discourse tree that results from a statement, such as the one in Figure 5, where the Tell 
action (whose recipe contains an Address-Understanding action) is a descendant of the 
Inform action (whose recipe contains an Address-Believability action); in addition, since 
the statement in Figure 5 is intended to answer a question, the Inform act is a descen- 
16 Questions must also be accepted and assimilated into a dialogue. Our model has recently been 
expanded to address the acceptance of questions (Bartelt 1996), but we are concentrating on statements 
in this paper. 
17 Since our system does not generate responses, we do not model what the speakers need to discuss; 
however, if a speaker expresses doubt at a proposition by contending that a second proposition is true, 
then the speaker is introducing this second proposition and its relationship to the first proposition as 
items that need to be discussed. 
22 
Carberry and Lambert Modeling Negotiation Subdialogues 
(15) 
(16) 
EA: Who is teaching CS360 (a systems course)? 
CA: Dr. Smith is teaching CS360. 
(17)a. 
(17)b. 
EA: Isn't Dr. Smith a theory person? 
EA: Who handles the CS360 lab? 
t Figure 8 
Dialogues conveying different implicit acceptance. 
dant of the Answer-Ref action (whose recipe contains an Address-Answer-Acceptability 
action). Since the Tell is the current focus of attention, it must be completed before 
other actions are pursued. Thus, if the listener believes that the telling has not been 
successful (i.e., the listener does not fully understand the utterance), then the listener 
will pursue discourse acts that contribute to its Address-Understanding subaction. Once 
the Tell has been successfully completed, then attention reverts back to the Inform act. 
The Inform must be successfully completed before other higher-level acts are pursued 
further. Thus if the Inform has not been successful (i.e., the listener does not accept the 
communicated proposition), then the listener will pursue discourse acts that contribute 
to its Address-Believability subaction. 
We have concentrated primarily on recognizing the acceptance and nonaccep- 
tance of propositions communicated by Inform actions: i.e., modeling negotiation sub- 
dialogues in which participants do not automatically believe everything that they 
are told. Others, Allen and Schubert (1991), Clark and Schaefer (1989), Traum and 
Hinkelman (1992), and Traum (1994) have investigated how understanding and lack 
of understanding are communicated and can be recognized. 
4.6.2 Modeling Implicit Acceptance. Our system models implicit acceptance in col- 
laborative dialogue as passing up the opportunity to express lack of acceptance. For 
example, consider the two dialogue variations shown in Figure 8. Figure 5 depicts the 
discourse tree constructed from utterances (15) and (16) in Figure 8, with the current 
focus of attention, the Tell action, marked with an asterisk. In attempting to assimilate 
(17a) into this discourse tree, the system's first expectation is that (17a) will address 
the understanding of (16) if EA does not understand it (i.e., as part of the Tell ac- 
tion that is the current focus of attention in Figure 5). The next expectation is that 
(17a) will relate to the Inform action in Figure 5, by addressing the believability of the 
proposition conveyed by (16). The system finds that the best interpretation of (17a) 
is that of expressing doubt at the proposition that Dr. Smith is teaching CS360, thus 
confirming the secondary expectation that (17a) is addressing the believability of the 
proposition conveyed by (16). This recognition of (17a) as part of the Inform action in 
Figure 5 indicates that EA has implicitly indicated understanding, by passing up the 
opportunity to address understanding in the Tell action that appears at a lower level 
in the discourse tree and by moving instead to a relevant higher-level action; (17a) is 
thus (implicitly) conveying that the Tell action has been successful. 
Thus, when an utterance contributes to an ancestor of an action Ai and all of 
Ai's applicability conditions, except those negated by the goal, are still satisfied, then 
Ai is assumed to have completed successfully; if that were not true, the dialogue 
participants would have been required to address those actions. '8 When an action 
18 By requiring that all applicability conditions, except those negated by the goal, still be satisfied in order for the action to be viewed as successful, we eliminate situations in which the agent of the Inform act 
23 
Computational Linguistics Volume 25, Number 1 
is recognized as successful, the system updates its model of the user's beliefs with 
the effects and goals of the completed action. For example, in determining whether 
(17a) in Figure 8 is expressing doubt at (16) (thereby implicitly indicating that (16) has 
been understood and that the Tell action has therefore been successful), the system 
tentatively hypothesizes that the effects and goals of the Tell action hold, resulting 
in the tentative belief that EA believes that CA believes that Dr. Smith is teaching 
CS360. If the system determines that this Express-Doubt action is the most coherent 
interpretation of (17a), it attributes the hypothesized beliefs to EA. 
Now consider a dialogue in which utterances (15) and (16) in Figure 8 are instead 
followed by utterance (17b). In this case, the system finds that the best interpretation 
of (17b) does not contribute to any of the actions in the existing discourse tree; instead 
(17b) is identified as initiating an entirely new Obtain-Info-Ref action at the discourse 
level, resulting in a new discourse tree. Since EA has gone on to pursue some other 
discourse action unrelated to any of the acts that were part of the previous discourse 
tree, the system recognizes not only EA's understanding of (16) but also EA's implicit 
acceptance of the proposition conveyed by (16). That is, because the system interprets 
EA's utterance as foregoing the opportunity to initiate a negotiation subdialogue to 
address the acceptance of the proposition communicated by the Inform action, the 
system recognizes that the Inform action has been successful and that EA has implicitly 
conveyed acceptance of the proposition that Dr. Smith is teaching CS360. 
4.7 The Recognition Algorithm 
Our recognition algorithm, outlined in Figure 9, assimilates a new utterance into the 
existing dialogue context and identifies discourse acts that the speaker is pursuing. 
It proceeds as follows: Start with the semantic representation of the utterance and 
extract from it two kinds of linguistic information: 1) clue words that might suggest 
a generic discourse act, and 2) beliefs that are conveyed by the surface form of the 
utterance. In our implemented system, possible clue words are explicitly noted in the 
semantic representation of the utterance, and beliefs conveyed by the surface form of 
an utterance are extracted from the applicability conditions of the recipe for the surface 
speech act. For example, the surface form of an utterance such as "Isn't Dr. Smith on 
sabbatical?" conveys that the speaker has a strong but uncertain belief in the queried 
proposition; this is captured in the applicability conditions of the recipe for a Surface- 
Neg-YN-Question discourse act (see the appendix). 
Next, use plan inference rules to hypothesize sequences of actions A1,Ai2 ..... Ai~ i 
(inference paths) such that A1 is the surface action directly associated with the speak- 
er's utterance and Aidi is an action on the active path in the existing dialogue context. 
By requiring that an inference path link up with an action that is already part of the 
existing dialogue context, we are capturing the expectation that the new utterance 
will contribute to an action that has already been initiated. This corresponds to a 
focusing heuristic that captures expectations for new utterances in an ongoing dialogue 
(Carberry 1990). For any inference path, if Ai~ is not the focus of attention in the 
existing dialogue context, then Aid~ must be an ancestor of the action that is the focus of 
attention; tentatively hypothesize that each of the actions on the active path between 
the existing focus of attention (the focus of attention immediately prior to the new 
utterance) and Aidl have completed successfully and use this hypothesis in reasoning 
has become convinced by the other participant that the proposition he was trying to convey is really 
false. In such cases, the applicability condition believe(_agentl, _proposition) of the Inform will no longer be true and thus the Inform act will not be viewed as completing successfully. We have not 
addressed situations in which the participants cannot resolve their disagreements and agree to disagree. 
24 
Carberry and Lambert Modeling Negotiation Subdialogues 
A1 = surface action associated with the speaker's utterance 
LE = clue words extracted from semantic representation of utterance 
D = dialogue model 
B -- listener's beliefs 
A d = action at current focus of attention in D 
;;Construct inference paths that link up to active path of dialogue model 
S*--{Pi = A1,Ai2 ..... Aiei \] on-active-path(Ai~i, D) 
APi is an inference path constructed from A1 } 
;;Eliminate inference paths with unsatisfied constraints or implausible applicability conditions 
For each Pi C S Do 
Begin 
Bi~---B 
If A~, ¢ A i 
Then Bi ~ BiU {beliefs that A d and all actions on the active path between Ai and Aid i 
have completed successfully} 
If (3Aj)(3Ck) Aj C Pi A is-constraint(Ck, Aj) A -~Ck 
Then S ~ S- Pi 
Else If (3Aj)(3ACk) Aj E Pi A is-app-cond(ACk,Aj) A-~plausible(ACk,Bi) 
Then S ~ S - Pi 
End 
;;Determine how much evidence is available for each e-action 
So ~ O, $1 ~ O, $2 ~ 0 
For each Pi E S Do 
If (3Aj) Aj C Pi A e-action(Aj) 
Then Begin 
If ling-evid(Aj,LE) A app-cond-evid(Aj,Bi) 
Then $2 ~ S2U{Pi} 
Else If ling-evid(Aj,LE) V app-cond-evid(Aj,Bi) 
Then $1 *-- S1U{Pi} 
End 
Else So ~-- SoU{Pi} ;;So contains inference paths with no e-actions 
;;Select inference paths containing actions with the most evidence 
If $2 ~ 0 Then S ~- $2 ;;S contains inference paths with multiple evidence 
Else If $1 # 0 Then S ~ $1 ;;S contains inference paths with evidence 
Else S ~ So ;;S contains inference paths with no e-actions 
;;Select inference path containing the action closest to current focus of attention 
If S#0 
Then P ,-- Pi \] Pi E S A Pi = A1, A 6 .... , Aidi 
A-~(3Pj) Pj E S APj = A1,Aj2 ..... Aid j 
A closer-to-curr-discourse-focus(Aja j,Aiei,D) 
Else Begin 
B ~- BU {beliefs that all actions on active path have completed successfully} 
S ~ {Pi = A1,A6 ..... Ain \] Pi is an inference path constructed from A1 
A no-elts-on-active-path(Pi,D) 
A (VAj)(VCk)(Aj ff Pi A is-constraint(Ck, Aj) ~ Ck) 
A (VAj)(VACk)(Aj E Pi A is-app-cond(ACk, Aj) --~ plausible(ACk, B)) 
P ~-- Pi \[ Pi E S A -~(3Pj) Pj E S A links-closer-to-ps-dom-focus(Pj,Pi,D) 
End 
;; Assimilate utterance into dialogue model 
Add P = A1, Ap2 .... , Apk to D 
Mark Ap2 as new focus of attention in D 
Figure 9 
Pseudocode outlining our recognition algorithm. 
25 
Computational Linguistics Volume 25, Number 1 
about the actions on the inference path. 19 If the applicability conditions for any of the 
actions on an inference path are implausible or if the constraints are not satisfied, reject 
the inference path. 
For actions that are e-actions, determine how much evidence is available for the 
action. Reject any inference paths containing an e-action for which there is neither 
linguistic evidence suggesting the generic discourse act (such as the clue word but 
suggesting an Express-Doubt action) 2° nor evidence from the surface form of the ut- 
terance, world knowledge, and contextual knowledge indicating that the applicability 
conditions for the particular discourse action are satisfied. If there is an e-action for 
which both kinds of evidence exist (both linguistic evidence for the generic discourse 
act and evidence that the applicability conditions are satisfied), then consider only 
inference paths containing an e-action for which there is such multiple evidence and 
select the inference path A1,Ai2,...,Ai~ for which Ai~ i is closest to the focus of atten- 
tion in the existing dialogue context. Otherwise, if there is an inference path contain- 
ing an e-action for which one kind of evidence exists, then select the inference path 
A1,Ai2 ..... Ai~i for which Ai~i is closest to the focus of attention in the existing dialogue 
context. 
If a satisfactory inference path containing an e-action cannot be found, then con- 
sider inference paths that contain no e-actions. 21 If there is more than one such inference 
path, select the one that links up to an action that is closest to the focus of attention on 
the discourse level. If there is no inference path linking up to an action on the existing 
discourse level, then select the inference path that links up to an action that is closest 
to the focus of attention on the problem-solving and domain levels. (Our dialogue 
model actually contains three levels: domain, problem-solving, and discourse. This 
paper is primarily concerned with recognizing actions on the discourse level; we will 
briefly discuss the domain and problem-solving levels in Section 5.1.) This latter case 
corresponds to initiating a new discourse segment, and thus a new discourse tree is 
constructed at the discourse level. 
Our algorithm identifies a best interpretation of the speaker's utterance. However, 
since the algorithm uses heuristics, its interpretation can be incorrect and miscommu- 
nication can result. Our current system does not include mechanisms for detecting 
and recovering from such errors. Clark and Schaeffer (1989) discuss second, third, and 
fourth turn repairs in discourse, and McRoy and Hirst (1995) provide an excellent 
formal model of repair in dialogue. 
5. Modeling Negotiation Subdialogues 
The preceding sections have provided the key mechanisms necessary for modeling 
negotiation subdialogues. Our recipes differentiate between the effects and the goals 
of a discourse act. Thus, instead of assuming that a communicated proposition will 
automatically be accepted by the listener, the effect of our Inform action is only that 
19 The action at the focus of attention and some of its ancestor actions may have completed successfully, 
which becomes evident when the participants choose not to address them further. For example, as a 
result of providing an answer to a question, the active path may include the discourse actions Answer-Ref, Inform, 
and Tell with the Tell discourse act being the current focus of attention; if the other 
participant then performs a discourse act that is a subaction of the Answer-Ref but not of the Inform, 
then he has accepted the proposition conveyed by the Inform and it has completed successfully. 
20 Our system currently maintains a list of clue words and discourse acts that each clue word might 
suggest. 
21 Due to length restrictions, we have omitted a part of the algorithm that deals with focusing heuristics 
that are not needed for the kinds of utterances addressed in this paper; an example of utterances 
needing the full algorithm is given in Lambert and Carberry (1991). 
26 
Carberry and Lambert Modeling Negotiation Subdialogues 
the listener believes that the speaker wants the listener to believe the communicated 
proposition, while its goal is that the listener will actually adopt the proposition as 
one of his own beliefs. In addition, the body of the Inform discourse recipe contains 
not only an action capturing the telling of the proposition but also an action capturing 
the participants' addressing the believability of the communicated proposition. Our 
algorithm for recognizing discourse actions and assimilating them into the dialogue 
model can recognize when an agent is expressing doubt at a communicated proposition 
by contending that some other proposition is true. Our ability to recognize implicit 
as well as explicit acceptance of a communicated proposition enables us to identify 
when an agent has adopted a communicated proposition as part of his beliefs. 
This section describes our implementation and demonstrates our system's capabil- 
ity with two extended negotiation subdialogues that illustrate 1) the role of linguistic, 
contextual, and world knowledge in resolving expressions of doubt; 2) expressions of 
doubt at both immediately preceding and earlier utterances; 3) multiple expressions 
of doubt at the same proposition; 4) negotiation subdialogues embedded within other 
negotiation subdialogues; and 5) explicit and implicit acceptance. The recipes for the 
discourse acts used in these examples can be found in the appendix. 
5.1 Implementation 
Our system for recognizing complex discourse acts and handling negotiation subdia- 
logues has been integrated into the tripartite dialogue model presented in Lambert and 
Carberry (1991). This dialogue model contains three levels of tree structures, one for 
each kind of action discussed in Section 4.2 (domain, problem-solving, and discourse) 
with links among the actions on different levels. At the lowest level, the discourse 
actions are represented; these actions may contribute to the problem-solving actions at 
the middle level which, in turn, may contribute to the domain actions at the highest 
level. Figure 10 illustrates the tripartite dialogue model for a situation in which CA 
has previously answered a question about the cost of registering for CS180, and then 
EA asks "When does CS180 meet?" Note that the discourse level in Figure 10 only re- 
flects the current query about when CS180 meets, since previous queries have already 
achieved their discourse goals. Since this paper is concerned almost exclusively with 
the discourse level of the dialogue model, we will not discuss the overall tripartite 
model further, except to note that the construction of a new discourse tree requires that 
the system identify its relationship to existing or new actions at the problem-solving 
and domain levels (Lambert and Carberry 1991). 
Our system has been implemented in Common Lisp on a Sun Sparcstation and 
tested in a university advisement domain. Figure 11 lists some of the beliefs included 
in the system's model of a stereotypical user. In our current implementation, only the 
clue word but is recognized as linguistic evidence for an Express-Doubt discourse act. 
In future work, we will expand the clue words taken into account by our system. 
5.2 An Extended Example 
Figure 12 contains an extended negotiation dialogue (portions of this dialogue have 
been given earlier). This dialogue illustrates a number of features that our system can 
handle. Utterances (18) and (19) establish the initial context in which CA has pro- 
22 ~ is usually implies rather than strict implication. The semantics of this predicate is that there may be a 
small number of cases where the antecedent is true and the consequent is not. This is similar to a 
default rule. For example, the On-Campus rule might be viewed as Vy: on-campus(y) A faculty(y) A M 
~on-sabbatical(y) ~ ~on-sabbatical(y). However, as with any default rule, there could be exceptions; 
for example, one might be on sabbatical but have returned to campus to give a colloquium. 
27 
Computational Linguistics Volume 25, Number 1 
Domain Level .............. ,,, ,. ...................................................... ................... ................ ,, .......... ........ , 
: \[ T e-Course< A, CSlS0) I 
\[ Register(EA, CS180) \] I * I Learn-Material(EA, CSl80,_fac) ~<~- ---- 
,., ....................... . .................................. | .............................................. ....... . ....................... 
I 
I 
Problem-Solving Level I 
I Build-Plan(EA, CA, Take-Course(EA,CS180)) \] 
I Build-Plan(EA, CA, Leam-Material(EA, CSl80,_fae)) \[ ............. 
Instantiate-Vars(EA, CA, Attend-Class(EA, _place, _time), Learn-Matefial(EA, CS 180, _fac)) \] 
Instantiate-Single-Var(EA, CA, _time, Attend-Class(EA, _place, _time), Learn-Material(EA, CS180, _fac)) I i 
A 
I 
I 
Discourse Level 
I i I I 
Obtain-Info-Ref(EA, CA, _time, Meets(CS 180, _time)) \] 
I I 
\[ Ask-Ref(EA, CA, _time, Meets(CS 180, _time)) \] 
i I 
* I Ref-Request(EA, CA, _time, Meets(CS180, _time)) I 
I I 
I Surface-WH-Question(EA, CA, _time, Meets(CS 180, _time)) \] 
........................ ........................................................................... , ....... , ............. , ........................................ + 
EA: When does CS180 meet? 
Key: 
- -> Enable arc 
Subaction arc 
* Current focus of attention 
Figure 10 
A simple tripartite dialogue model. 
vided information in response to a question from EA. Utterances (20), (23), and (27) 
illustrate the use of world knowledge in resolving expressions of doubt. Utterance 
(20) is an expression of doubt that initiates a negotiation subdialogue, indicating that 
EA has not accepted the proposition communicated by CA. Utterances (21) and (22) 
attempt to resolve this doubt by stating that Dr. Brown is not teaching Architecture 
and providing support for this claim. Utterance (23) expresses doubt at the supporting 
information; utterances (24) and (25) attempt to resolve this doubt; and utterance (26) 
explicitly accepts the proposition communicated by (25). Thus utterances (23)-(26) 
constitute a negotiation subdialogue embedded within the negotiation subdialogue 
of utterances (20)-(27). Although utterance (26) explicitly conveys acceptance of at 
least the most salient communicated proposition (the proposition that Dr. Brown was 
giving a University colloquium), there are still several propositions that have not yet 
been completely accepted by EA and are thus open for rejection. Utterance (27) again 
expresses doubt at the proposition that Dr. Smith is teaching Architecture, and implic- 
itly conveys acceptance of the propositions that Dr. Brown is on sabbatical and that 
28 
Carberry and Lambert Modeling Negotiation Subdialogues 
Sabbatical rule: Teachers on sabbatical usually do not teach. 
(Vx Vy course(y) A faculty(x) A on-sabbatical(x) ~ -~teaches(x, y) ) 
On Campus rule: Teachers on campus usually are not on sabbatical. 
(Vy faculty(y) A on-campus(y) ~ -~on-sabbatical(y) ) 
One Course rule: Teachers usually teach only one course a semester. 
(Vx Vy Vz # y faculty(x) A course(y) A course(z) A teaches(x, y) ~ -~teaches(x, z) ) 
One Professor rule: Each course usually has only one instructor. 
(Vx Vy Vz # y course(x) A faculty(y) A faculty(z) A teaches(y, x) ~ -~teaches(z, x) ) 
Expertise rule: Teachers usually do not teach courses outside their area of 
expertise. 
(Vx Vy Vz faculty(x) A course(y) A area(z) A specialty(x, z) A -~in-area(y, z) 
-~teaches(x, y) ) 
Figure 11 
A few stereotypical beliefs. 22 
(18) EA: 
(19) CA: 
(20) EA: 
(21) CA: 
(22) 
(23) EA: 
(24) CA: 
(25) 
(26) EA: 
(27) 
Figure 12 
Extended 
What is Dr. Smith teaching? 
Dr. Smith is teaching Architecture. 
Isn't Dr. Brown teaching Architecture? 
No, Dr. Brown is not teaching Architecture. 
Dr. Brown is on sabbatical. 
But wasn't Dr. Brown on campus yesterday? 
Yes, Dr. Brown was on campus yesterday. 
Dr. Brown was giving a University colloquium. 
OK. 
But isn't Dr. Smith a theory person? 
negotiation subdialogue. 
Dr. Brown is not teaching Architecture. 
The rest of this section works through the details of how our system processes 
these utterances, recognizes the discourse acts they are pursuing, and incrementally 
builds the discourse tree of the dialogue model. In our examples, the system will 
switch between playing the role of EA and the role of CA. However, when processing 
an utterance, the system will have access only to the beliefs of the participant whose 
identity it has assumed (namely, the listener), along with the correct dialogue model 
at the time the utterance is made. 
5.2.1 Utterance (18): Establishing the Initial Context. The system first plays the role 
of CA (the listener) and must understand EA's utterance of (18). The semantic repre- 
sentation of (18) is: 
Surface-WH-question(EA, CA, _course, Teaches(Dr. Smith, _course)) 
The Surface-WH-Question is a subaction in the body of a recipe for a Ref-Request dis- 
course act; the Ref-Request is a subaction in the recipe for an Ask-Ref discourse act; and 
29 
Computational Linguistics Volume 25, Number 1 
the Ask-Ref is a subaction in the recipe for an Obtain-Info-Ref discourse act. Therefore 
the following chain of actions is hypothesized: 
0btain-Info-Ref (EA, CA, _course, Teaches(Dr.Smith, _course)) 
T 
Ask-Ref(EA, CA, _course, Teaches (Dr. Smith, _course)) 
T 
Ref-Request(EA, CA, _course, Teaches(Dr. Smith, _course)) 
T 
Surface-WH-Question(EA, CA, _course, Teaches(Dr. Smith, _course)) 
As each of the above actions is inferred, the system checks that its constraints are 
satisfied and that its applicability conditions are plausible. Since this is the only chain 
of actions suggested by plan inferencing on the discourse level, the system recognizes 
these discourse actions; it then infers problem-solving actions from the discourse ac- 
tions and, eventually, domain actions from the problem-solving actions. As actions are 
recognized, the system updates its model of EA's beliefs, wants, and knowledge from 
the actions' applicability conditions. Figure 13 shows the initial tripartite dialogue 
model that is produced. Since this paper is primarily concerned with the recognition 
of actions on the discourse level, the remainder of our figures will only display the 
discourse level and will omit the problem-solving and domain level actions. 
5.2.2 Utterance (19): Answering the Question. The system is now playing the role of 
EA (listener) and must understand CA's utterance of (19). The semantic representation 
of (19) is: 
Surface-Say-Prop(CA, EA, Teaches(Dr. Smith, Arch)) 
Chaining suggests that the surface speech act might be part of a Tell action, which 
might be part of an Inform action since the surface speech act and the Tell act are part 
of the body of the Tell and Inform acts, respectively. The applicability conditions for all 
of these actions are plausible. 
The system tries to extend the inference chain from the Inform action. An Inform 
can be part of the recipes for several discourse actions, including Give-Background and 
Answer-Ref. However, these actions are e-actions and, with the exception of Answer- 
Ref, inference of these e-actions is rejected. For example, Give-Background is an e-action 
because it relates the proposition in the current utterance to some other proposition, 
the proposition about which background is being given. The recipe for Give-Background 
contains a constraint that there be a particular relationship between the proposition in 
the Inform action in its body and some other proposition conveyed by CA. Since CA 
has made no previous utterances, there is no other proposition conveyed by CA and 
thus this constraint cannot be satisfied. Consequently, Give-Background is rejected. 23 A 
full discussion of the Give-Background action and its recipe can be found in (Lambert 
1993). 
On the other hand, Answer-Kef(CA, EA, _term, _proposition) can be inferred 
from Inform(CA, EA, Teaches (Dr. Smith, Arch) ) and the system has evidence for its 
recognition. Answer-Re/is an e-action since the parameters _term and _proposition 
cannot be instantiated from the Inform action that precedes it on the inference chain. 
23 Although CA can provide background information prior to conveying the proposition about which the 
background is being given, the Give-Background action in these instances will be recognized in 
assimilating CA's second utterance (the utterance about which the background is being given) 
(Lambert and Carberry 1991). 
30 
Carberry and Lambert Modeling Negotiation Subdialogues 
D.omaio.Lmtet ................ , 
i 
? r qb.19m_-_Sqlv!q g. L_evel ..... 
• - ..................... i- ................................................... , 
: I Build-Plan(EA, CA, Take-Course(EA,_course)) r I 
I Instantiate-Vars(EA, CA, Learn-Material(EA, _course, Dr. Smith), Take-Course(EA, _course)) I 
* I lnstantiate-Single-Var(EA, CA,_course, Learn-Material(EA, course, Dr. Smith), Take-Coursel EA,_course)) I ii 
I 
I 
D!s_egu.rs_e_ _Level. ............. , .............................. 
I I Obtain-Info-Ref(EA, CA, _course, Teaches(Dr. Smith, _course)) \] 
$ 
I Ask-Ref(EA, CA, course, Teaches(Dr. Smith,_course)) \] 
e Ref-Request(EA, CA, _course, Teaches(Dr.Smith,_course)) 
i 
Surface-WH-Question(EA,CA,_c'ourse,Teaches(Dr.Smith,_course)) \] 
............................................................. o 
Key: EA: What is Dr. Smith teaching? 
- -~ EnableAre 
Subactlon Arc 
* Current focus of attention 
Figure 13 
Tripartite dialogue model for utterance (18). 
As discussed in Section 4.5.1, evidence for e-actions may take one of two forms: 1) 
evidence from world and contextual knowledge and the surface form of the utter- 
ance indicating that the applicability conditions for a particular e-action are satisfied, 
and 2) linguistic evidence from clue words suggesting a generic discourse action. In 
this case, there are no clue words, so any evidence must be from world and con- 
textual knowledge or the surface form of the utterance. Answer-Re/(CA, EA, _term, 
_proposition) can be a subaction in the body of 0btain-Info-Ref (EA, CA, _term, 
_proposition); unifying with the Obtain-In/o-Re/ action that is part of the existing 
discourse tree causes the parameters _proposition and _term to be instantiated as 
Teaches (Dr . Smith, _course) and _course, respectively, in both the Obtain-In/o-Re/ 
and Answer-Re/actions. 
World and contextual knowledge provide evidence that the applicability condi- 
tions of Answer-Re/are satisfied with these instantiations. The third applicability con- 
dition in the Answer-Re/recipe captures the required relationship between the new 
parameter _proposition and the parameter _propanswer that appears in the Inform 
discourse act. It indicates that CA must believe that _propanswer (where _propanswer 
is instantiated from the Inform act as Teaches (Dr. Smith, Arch)) is an instance of the 
queried proposition, _proposition, with the queried term _term instantiated. Since 
the system (playing the role of EA in this case) believes that the participants have 
31 
Computational Linguistics Volume 25, Number 1 
equivalent knowledge about language and how terms can be instantiated 24 and since 
the system believes that the two propositions unify, the system has evidence that the 
third applicability condition is satisfied. 
In addition, there is evidence that the other two applicability conditions are satis- 
fied. With the above instantiations, these applicability conditions become: 
Applicability conditions for Answer-Ref : 
believe~A,want(EA,knowref(EA,_course,believe~A,Teaches(Dr. Smith,_course), 
\[c:c\] ))), \[w:c\] ) 
believe(CA, ~knowref(EA,_course,believe(CA,Teaches(Dr. Smith,_course), 
\[C:C\])), \[W:C\]) 
If utterance 19 is in fact an Answer-Ref that contributes to the Obtain-Info-Ref that is part 
of the existing dialogue context, then CA has recognized and is responding to the Ask- 
Ref that is a child of the Obtain-Info-Ref in the existing dialogue context. Consequently, 
in considering the Answer-Ref interpretation, the system can hypothesize that CA has 
recognized the Ask-Ref and can tentatively attribute to CA the belief that the Ask-Refs 
applicability conditions were satisfied, as shown below. 
Beliefs attributed to CA (by virtue of hypothesis that CA has recognized the Ask-Ref): 
want(EA, knowref(EA, _course, believe(CA, Teaches(Dr. Smith, _course), 
\[c:c\])) 
~knowref(EA, _course, believe(CA, Teaches(Dr. Smith, _course), \[C:C\])) 
This is equivalent to tentatively hypothesizing that CA has recognized the intentions 
communicated by utterance (18) and inferred the discourse level of the dialogue model 
depicted in Figure 13. Thus the system's model of CA's beliefs (resulting from CA's 
recognition of the Ask-ReJ) provides evidence that the applicability conditions of the 
Answer-Ref discourse act are satisfied. 
Since this inference chain is the only one containing an e-action for which there 
is evidence, the system recognizes CA's utterance as providing Architecture as the 
answer to EA's question about what Dr. Smith is teaching and thereby contributing 
to the Obtain-Info-Ref action initiated by EA. The updated discourse tree is shown in 
Figure 14, with the new focus of attention marked with an asterisk. 
5.2.3 Utterance (20): Initial Expression of Doubt. The system is again playing the role 
of CA (listener) and must understand EA's utterance of (20). The semantic represen- 
tation of (20) is: 
Surface-Neg-YN-Question(EA, CA, Teaches(Dr.Brown, Arch)) 
The surface form of (20) suggests that EA thinks that Dr. Brown is teaching Archi- 
tecture, but is not certain of it. This belief is captured in the applicability condition 
24 This does not mean that the instantiation will result in a true proposition, only that it is a legal 
instantiation of the term. For example, CS180 is a legal instantiation of the _course term in the 
proposition Teaches(Jones _course) although Teaches(Jones, CS180) may be false. We have not 
addressed the problem of misconceptions about class membership. 
32 
Carberry and Lambert Modeling Negotiation Subdialogues 
Obtain-lnfo-Ref(EA, CA, _course, Teaches(Dr. Smith, course))\] 
\[ Ask-Ref(EA, CA. _course, Teaches(Dr. Smith, _course)) I I Answer-Ref(CA. EA, _course, Teaches(Dr.Smith,_course)) \[ 
t 
Ref-Request(EA, CA, _course, Teaches(Dr.Smith,_course)) I 
Surface-WH-Question(EA, CA, _course, Teaches(Dr.Smith,_course)) \[ 
(18) EA: What is Dr. Smith teaching? 
Key: 
* Current focus of attention 
Figure 14 
Discourse tree for first two utterances in Figure 12. 
Inform(CA, EA, Teaches(Dr.Smith,Arch)) I 
• \[Te,,(CA, EA, Teac os OrSmith Arch)) \] 
\[ Surface-Say-Prop(CA, EA, Teaches(Dr. Smith, Arch)) \] 
(19) CA: Dr. Smith is teaching Architecture. 
of the recipe for a Surface-Neg-YN-Question. Since we assume a noise-free medium 
and well-formed utterances, surface speech acts always execute successfully and are 
correctly recognized. Thus, the beliefs captured in the applicability conditions of the 
surface speech act are immediately entered into the system's model of EA's beliefs. 
The most salient interpretation of (20), that it is addressing the understanding of (19) 
and thus contributing to the Tell discourse act that is the current focus of attention in 
the dialogue, is rejected. 25 
The system can construct an inference path suggesting that the utterance con- 
tributes to the Inform discourse act that is the parent of the Tell act in the existing 
discourse tree. In particular, by chaining from subactions to parent actions (actions 
whose recipes contain the subaction), the system constructs an inference path contain- 
ing the following chain of actions: 
Inform(CA, EA, _propositionl) 
T 
Address-Believability(CA, EA, _propositionl) 
T Address-Unacceptance(EA, CA, _propositionl, Teaches(Dr.Brown, Arch)) 
T 
Express-Doubt(EA, CA, _propositionl, Teaches(Dr.Brown, Arch)) 
T 
Convey-Uncertain-Belief (EA, CA, Teaches(Dr.Brown, Arch)) 
T 
Surface-Neg-YN-Question(EA, CA, Teaches (Dr.Brown, Arch)) 
If the last action on this inference path is unified with the Inform act in the exisUng 
discourse tree, then _propositionl in the recipe for Address-Believability will be instan- 
tiated as Teaches (Dr. Smith, Arch), indicating that EA uttered (20) in order to express 
25 In our implemented system, it is rejected because there is no recipe for the Address-Understanding action 
that is part of the body of the recipe for the Tell discourse act, and thus it is not possible to construct an 
inference path from the utterance to the Address-Understanding act. In the future, our expanded system 
will include such recipes, and the interpretation will be rejected because of lack of evidence for the 
e-action on the inference path or because its constraints are not satisfied. 
33 
Computational Linguistics Volume 25, Number 1 
doubt at the proposition that Dr. Smith is teaching Architecture and thereby contribute 
to addressing the believability of that proposition. This interpretation would indicate 
that EA had passed up the opportunity to contribute to the Tell discourse act that is 
the focus of attention in the existing discourse tree. Thus when the system considers 
this interpretation, it hypothesizes that the Tell act has been successful and that its goal 
has been achieved, and it tentatively adds 
believe(EA, believe(CA, Teaches(Dr. Smith, Arch), \[C:C\]), \[C:C\]) 
to the belief model. 
As we have seen previously, Express-Doubt is an e-action since it is the action on 
the inference path at which the parameter _propo s it ion 1 is first introduced. Therefore, 
although the applicability conditions for each of the actions on the above inference 
path are plausible, we need evidence for the Express-Doubt act. There is no linguistic 
clue suggesting that (20) is an Express-Doubt action. The system then checks to see if 
it has evidence that the applicability conditions for the Express-Doubt action hold. The 
applicability conditions are: 
believe(EA, believe(CA, Teaches (Dr. Smith, Arch), \[S:C\]), \[S:C\]) 
believe(EA, Teaches(Dr. Brown, Arch), \[W:S\]) 
believe(EA, Teaches(Dr. Brown, Arch) --~ ~Teaches(Dr. Smith, Arch), \[S:C\]) 
The system's belief model provides evidence for the first applicability condition, that 
EA believes that CA believes that Dr. Smith teaches Architecture, since it has been 
tentatively updated to include the goal of the Tell discourse act, as noted above. The 
belief model also provides evidence for the second applicability condition, since it 
has been updated to include the beliefs captured in the applicability conditions of the 
recipe for the surface speech act. The system's model of a stereotypical user contains 
the beliefs given in Figure 11, including the belief that there is only one professor per 
course. This stereotypical belief provides evidence for the final applicability condition 
(that EA believes that Dr. Brown teaching Architecture implies that Dr. Smith is not 
teaching Architecture). Since users typically believe that only one teacher is used per 
course, perhaps EA does also. If EA believes that there is only one professor per course 
and that Dr. Brown is teaching Architecture, then EA would believe that Dr. Smith 
would not be teaching Architecture. So the system has evidence for all three of the 
applicability conditions in the Express-Doubt recipe. In addition, the constraint of the 
Express-Doubt action is satisfied since the proposition that Dr. Smith teaches Architec- 
ture is a parameter of an action on the active path and thus is salient. 
Since there is evidence from world and contextual knowledge and the surface 
form of the utterance that the applicability conditions hold for interpreting (20) as an 
expression of doubt and since there is no evidence for any other e-action, the system 
infers that this is the correct interpretation and stops. Thus, (20) is interpreted as an 
Express-Doubt action, as shown in Figure 15. 
5.2.4 Utterances (21)-(22): Attempted Resolution of Conflict. The system is now play- 
ing the role of EA (listener) and must assimilate CA's utterances (21)-(22). The semantic 
representation of (21) is: 
Surface-Say-Prop(CA, EA, ~Teaches(Dr.Brown, Architecture)) 
Plan chaining indicates that the Surface-Say-Prop may be part of Tell(CA, EA, 
~Teaches(Dr.Brown, Architecture)), which might be part of Inform(CA, 
EA, -~Teaches(Dr. Brown, Architecture)), which might be part of Resolve- 
34 
Carberry and Lambert Modeling Negotiation Subdialogues 
/ 
a 
o 
o m o 
m o 
4 
• © 
u 
m 
m 
o= o < .~ 
o i ul ~4 < 
. .o 
m 
Figure 15 
Discourse tree for 
m 
° 
Ua ~ - 
o ~ 121 
m 
m m ~u 
< < 
a, g 
<i ~ 
utterances (18)-(20). 
m o 
~ u 
m < g ~ 
"~ .~2 
o 
o Z 
~ 6 o ~ 
m 
0 ",~, 
Conflict(CA, EA, _propositionl, _proposition2), which might in turn be part 
of Address-Unacceptance(EA, CA, _propositionl, _proposition2). If this is the 
Address-Unacceptance action that is part of the existing discourse tree in Figure 15, then 
the Express-Doubt and Convey-Uncertain-Belief actions in Figure 15 have completed suc- 
cessfully. Thus, in considering this interpretation, the system hypothesizes that these 
35 
Computational Linguistics Volume 25, Number 1 
actions have been successful and tentatively updates its belief model to reflect the ef- 
fects and goals of these actions. In particular, the following two beliefs (among others) 
are tentatively added to the system's model of CA's beliefs: 
believe (CA, believe (EA, Teaches (Dr. Brown, Arch)-~Teaches (Dr. Smith, Arch), 
IS:C\]), \[s:c\]) 
believe(CA, believe(EA,Teaches(Dr.Brown,Arch), \[W:S\]), \[S:C\]) 
Resolve-Conflict (see the appendix) is an e-action since it introduces two new propo- 
sitions (the propositions about which there is conflict) that cannot be instantiated by 
chaining from the Inform action in its body, and the system must be able to deter- 
mine what conflict the utterance is trying to resolve. If the Address-Unacceptance ac- 
tion is unified with the Address-Unacceptance that is part of the existing discourse 
tree, then the conflicting propositions, _propositionl and _proposition2, are in- 
stantiated as Teaches (Dr. Smith, Arch) and Teaches (Dr.Brown, Arch), respectively, 
in both Address-Unacceptance and Resolve-Conflict. The system has evidence for the 
Resolve-Conflict action with these instantiations. The constraints that _propositionl 
and _proposition2 be salient and that _proposition2 and _proposition3 be 
the opposite of one another are satisfied. First, Teaches (Dr . Smith, Arch) and 
Teaches(Dr.Brown, Arch) are the propositions instantiating _propositionl and 
_proposition2, and they are salient since they are part of an action on the active path 
of the existing discourse tree. Second, the proposition instantiating _proposition2 is 
the opposite of the proposition conveyed by CA's current utterance. Evidence for the 
first two applicability conditions, 1) that CA believes that EA beheves that Dr. Brown's 
teaching Architecture implies that Dr. Smith is not teaching Architecture and 2) that 
CA believes that EA has an uncertain belief in the proposition that Dr. Brown teaches 
Architecture, is provided by the system's tentatively updated model of CA's beliefs. 
Evidence for the final applicability condition, that CA believes that Dr. Smith is teach- 
ing Architecture, is also provided by the system's model of CA's beliefs. When CA's 
Inform action in (19) was recognized, the system updated its model of CA's beliefs to 
include the beliefs contained in the applicability conditions for the Inform act; thus the 
belief model indicates that CA believes that Dr. Smith is teaching Architecture. Since 
the system has evidence for the e-action on the inference path (and since there are no 
other inference paths containing e-actions), the system recognizes this chain of actions 
and interprets (21) as informing EA that Dr. Brown is not teaching Architecture as 
part of attempting to resolve the conflict suggested by EA. Thus the Resolve-Conflict 
action is recognized as contributing to the Address-Unacceptance action that was begun 
in (20). 
The semantic representation of (22) is: 
Surface-Say-Prop(CA, EA, on-sabbatical(Dr.Brown)) 
The Surface-Say-Prop is part of Tell(CA, EA, on-sabbatical(Dr.Brown)), which is 
part of Inform(CA, EA, on-sabbatical(Dr.Brown)). Further chaining suggests that 
the Inform action could be part of several other actions. We will discuss two of these 
possibilities, Address-Acceptance and Explain-Claim. In the Address-Acceptance case, CA 
might be uttering (22) to support the statement that she made in (21); in the Explain- 
Claim case, CA might be uttering (22) to explain why the supposedly conflicting propo- 
sitions are not really in conflict. 
Let us examine the Address-Acceptance case first. Address-Acceptance(CA, EA, 
_propositionl) is part of a recipe for Address-Believability(CA, EA, 
_propositionl), which in turn is part of a recipe for Inform(CA, EA, _propositionl). 
36 
Carberry and Lambert Modeling Negotiation Subdialogues 
If this is CA's immediately preceding Inform act, then unifying with this Inform act will 
cause _propositionl to be instantiated as ~Teaches (Dr.Brown, Arch) in the Inform, 
Address-Believability, and Address-Acceptance actions. Address-Acceptance is an e-action, 
since it is the action on the inference path at which a new proposition is first in- 
troduced. If _proposition3 in the recipe for Address-Acceptance (see the appendix) is 
instantiated with Teaches (Dr. Brown, Arch), then the constraints are obviously sat- 
isfied. (Note that _proposition2 in the recipe for Address-Acceptance is instantiated 
with on-sabbatical (Dr. Brown) as a result of chaining from the surface speech act to 
the Inform act in the body of the Address-Acceptance action.) The system has evidence 
that the applicability conditions are satisfied with these instantiafions. Evidence for 
the first applicability condition is provided by the system's model of a stereotypical 
user, which indicates that it is generally believed that professors on sabbatical do not 
teach. Evidence for the second applicability condition is provided by the system's 
model of CA's beliefs, which was updated to contain the effect of utterance (20)'s 
Convey-Uncertain-Belief action--namely, that CA believes that EA has some belief that 
Dr. Brown is teaching Architecture. Thus there is evidence for recognizing CA's ut- 
terance as addressing acceptance of the proposition that Dr. Brown is not teaching 
Architecture by offering support for it. 
Now let us examine the Explain-Claim case. Explain-Claim(CA, EA, 
_propositionl, _proposition2) is part of Resolve-Conflict(CA, EA, 
_propositionl, _proposition2) and unifying with the Resolve-Conflict that is already 
part of the existing discourse tree causes _proposition1 and _proposition2 in the 
Resolve-Conflict and Explain-Claim actions to be instantiated with Teaches (Dr. Smith, 
Arch) and Teaches (Dr . Brown , Arch), respectively. In the recipe for Explain-Claim, 
_proposition3 has been instantiated with on-sabbatical (Dr. Brown) by chaining from 
the surface speech act to the Explain-Claim action. Explain-Claim is an e-action. How- 
ever, the system lacks evidence for its second applicability condition, that CA believes 
that Dr. Brown being on sabbatical implies that Dr. Brown teaching Architecture and 
Dr. Smith teaching Architecture are not in conflict with one another. Thus this poten- 
tial interpretation is rejected. Since the inference path containing the Address-Acceptance 
discourse act is the only one whose e-action has evidence supporting its recognition, 
the system recognizes (22) as addressing the acceptance of the proposition conveyed 
by (21)--namely, that Dr. Brown is not teaching Architecture. Thus, CA's response in 
(21) and (22) indicates that CA is trying to resolve EA's and CA's conflicting beliefs. 
The structure of the discourse tree after these utterances is shown in Figure 16, above 
the numbers (18)-(22). 26 
5.2.5 Utterances (23)-(26): Embedded Negotiation Subdialogue. The system is now 
playing the role of CA (listener) and must assimilate EA's utterance of (23). The se- 
mantic representation of (23) is: 
Surface-Neg-YN-Question(EA, CA, on-campus(Dr.Brown, Yesterday)) 
Clueword(But) 
The Surface-Neg-YN-Question in utterance (23) is one way to Convey-Uncertain-Belief. 
The linguistic clue but suggests that EA is executing a nonacceptance discourse action; 
this nonacceptance action might be addressing (22), (21), or (19), since the propositions 
conveyed by these utterances have not yet been accepted by EA and are thus open 
for rejection. Let us consider the proposition conveyed by (22), since it is the most 
26 For space reasons, only the action names are shown. 
37 
Computational Linguistics Volume 25, Number 1 
z ~ 
~1 !:v 
~J ~ .... 
m 
m ~.j 
Figure 16 
Discourse tree for dialogue in Figure 12. 
J 
t.. 
f 
J 
~c~ I 
uj ~--Q 
m i ::i = ix,.---. 
V 
J 
o IX / 
-- I~ "J 
¢--1 ¢-,1 
38 
Carberry and Lambert Modeling Negotiation Subdialogues 
salient open proposition at this point in the dialogue and thus the most expected 
candidate. Plan chaining suggests that the Convey-Uncertain-Belief could be part of an 
Express-Doubt action, which in turn could be part of an Address-Unacceptance action, 
which could be part of an Address-Believability action, which could be part of the Inform 
action in (22). As in utterance (20), there is evidence that the applicability conditions for 
the e-action (the Express-Doubt action) hold: for example, world knowledge indicates 
that a typical user believes that professors who are on campus are not on sabbatical, 
providing evidence for the third applicability condition. Thus, there is both linguistic 
evidence for a generic nonacceptance discourse act and evidence from world and 
contextual knowledge and the surface form of the utterance that the applicability 
conditions and constraints are satisfied for the specific action of expressing doubt at 
the proposition that Dr. Brown is on sabbatical. Since no other e-action has both kinds 
of evidence, (23) is interpreted as expressing doubt at the proposition conveyed by 
(22). 
The system now reverts to playing the role of EA (listener) and must assimilate 
the next two utterances in which CA resolves the doubt that EA has expressed in 
(23), by agreeing that Dr. Brown was on campus yesterday but explaining the purpose 
of his visit (one that is an exception to the rule that people on sabbatical are not on 
campus). Plan inferencing for utterance (24) is identical to that of utterance (21) and 
will not be described further. 
From the Surface-Say-Prop in (25), plan inference rules suggest that the Surface- 
Say-Prop is part of a Tell action that is part of an Inform action. As was the case for 
utterance (22), the Inform action can be part of several different higher-level actions, 
including Address-Acceptance and Explain-Claim. Since Address-Acceptance is a subac- 
tion in a recipe for Address-Believability, and Address-Believability is a subaction in a 
recipe for Inform, CA might be trying to offer support for the Inform act of (24), 
Inform(CA, EA, on-campus(Dr.Brown, Yesterday)). However, this time the appli- 
cability conditions for the Address-Acceptance action are implausible. In particular, as 
a result of the effect of the Convey-Uncertain-Belief action in (23), the system's model 
of CA's beliefs indicates that CA believes that EA has some belief in the proposi- 
tion that Dr. Brown was on campus yesterday. The second applicability condition 
of the recipe for addressing the acceptance of the proposition conveyed by (24), 
believe(CA, believe(EA, -~on-campus(Dr.Brown, Yesterday), \[W:S\]), \[W:C\]), 
conflicts with this belief--i.e., Address-Acceptance is reasonable to pursue only when 
an agent has some reason to believe that the listener disbelieves the proposition in 
question. Since the second applicability condition is implausible, the inference path 
containing the Address-Acceptance action is rejected. 
However, the system does have evidence for interpreting (25) as an Explain-Claim. 
Explain-Claim(CA, EA, _propositionl, _proposition2) is part of the recipe for 
Resolve-Conflict(CA, EA, _propositionl, _proposition2). If this is the Resolve- 
Conyqict action that is closest to the focus of attention in the existing discourse tree, 
then unification will cause _propositionl and _proposition2 in Resolve-Conflict and 
Explain-Claim to be instantiated respectively with on-sabbatical(Dr.Brown) and 
on-campus(Dr. Brown, Yesterday). In the recipe for Explain-Claim, _proposition3 
was instantiated with Give (Dr. Brown, University-Colloquiura) during chaining from 
the surface speech act. The system has evidence for the e-action Explain-Claim because 
it has evidence that its applicability conditions hold--namely, that CA believes that EA 
believes that Dr. Brown's being on campus implies that he is not on sabbatical from the 
effect of the Express-Doubt action; that CA believes that Dr. Brown's giving a University 
colloquium implies that being on campus is not in conflict with being on sabbatical, 
from the model of stereotypical beliefs; and that CA believes that EA believes that 
39 
Computational Linguistics Volume 25, Number 1 
Dr. Brown was on campus yesterday, from the effect of the Convey-Uncertain-Belief 
discourse act accomplished by (23). Since this is the only inference path containing 
an e-action for which the system has evidence, utterance (25) is interpreted as con- 
tributing to resolving the conflict suggested in (23) by explaining the claim that the 
propositions do not really conflict in this instance. 
The system now reverts to playing the role of CA (listener) and must assimilate 
EA's utterances. In (26), EA indicates explicit acceptance of the most salient Inform 
action, so the system is able to determine that EA has accepted CA's response in (25). 
Other inform actions remain open for rejection and must still be implicitly or explicitly 
accepted. In this dialogue, the Inform actions in (22) and (21) are implicitly accepted in 
utterance (27). Althougl~ utterance (27) might cause one to hypothesize that (26) was 
indicating explicit acceptance of all of the propositions conveyed by utterances (21)- 
(25), it is not possible to decide with certainty from a simple "ok" exactly how many 
Inform actions EA is accepting. Thus our system assumes that the speaker accepts as 
little as possible, which is the most salient Inform action. 
Utterances (23)-(26) illustrate our model's handling of negotiation subdialogues 
embedded within other negotiation subdialogues. The subtree contained within the 
dashed lines in Figure 16 shows the structure of this embedded negotiation subdia- 
logue. 
5.2.6 Utterance (27): Multiple Expressions of Doubt and Implicit Acceptance. The 
system is still playing the role of CA (listener). The semantic representation of EA's 
next utterance is 
Surface-Neg-YN-Question(EA, CA, Specialty(Dr. Smith, Theory)) 
Clueword(But) 
The linguistic clue but in (27) again suggests nonacceptance. Since (25) has been ex- 
plicitly accepted, the propositions open for rejection are those conveyed in (22), (21), 
and (19). Once again, chaining from the surface speech act can produce a chain of 
actions containing an Express-Doubt action and terminating with one of the Inform ac- 
tions that is on the active path of the existing discourse tree. If the Inform action is 
Inform(Ca, EA, Teaches (Dr. Smith, Arch)), then the Express-Doubt action will be in- 
stantiated as Expres s-Doubt (EA, CA, Teaches (Dr. Smith, Arch), Specialty (Dr. Smith, 
Theory) ). The system has evidence that this action's applicability conditions are sat- 
isfied. The evidence for the first two applicability conditions is similar to the evidence 
for interpreting utterance (20) as expressing doubt. World knowledge provides evi- 
dence for the third applicability condition. The system's model of stereotypical user 
beliefs indicates that it is typically believed that faculty only teach courses in their 
field. Other system knowledge states that Architecture and Theory are different fields. 
So in this case, the system's world knowledge provides evidence that Dr. Smith's being 
a theory person is an indication to the user that Dr. Smith does not teach Architec- 
ture. Thus the system has two kinds of evidence for interpreting (27) as expressing 
doubt at the proposition conveyed by (19): linguistic evidence for a generic Express- 
Doubt discourse act and evidence that the applicability conditions are satisfied for 
the particular discourse act of expressing doubt at the proposition that Dr. Smith is 
teaching Architecture. Since the system does not have multiple evidence for any of the 
other interpretations, the system recognizes (27) as again expressing doubt about the 
proposition conveyed by (19). Thus, the system is able to recognize and assimilate a 
second expression of doubt at the proposition conveyed in (19), even after intervening 
dialogue. The discourse tree for the entire dialogue is given in Figure 16. 
40 
Carberry and Lambert Modeling Negotiation Subdialogues 
(28) 
(29) 
(30) 
(31) 
(32) 
(33) EA: 
(34) CA: 
(35) 
(36) EA: 
Figure 17 
EA: When does CS510 meet? 
CA: CS510 meets on Monday night at 7PM. 
EA: But isn't Dr. Jones teaching CS510? 
CA: No, Dr. Jones is not teaching CS510. 
Dr. Hart is teaching CS510. 
But isn't CS510 a graduate course? 
Yes, CS510 is a graduate course. 
Dr. Hart teaches both graduate and undergraduate courses. 
What courses are prerequisites for CS510? 
A second negotiation subdialogue. 
Since EA's utterance reverts back to addressing the acceptance of the proposition 
conveyed by (19), EA has foregone the opportunity to challenge the claims made in 
utterances (22) and (21). Since the befief model indicates that the applicability condi- 
tions of the Inform actions are still satisfied (except those negated by achievement of 
the goal), the system infers that EA has implicitly accepted the statements in (22) and 
(21), that Dr. Brown is on sabbatical and that Dr. Brown is not teaching Architecture, 
and the system updates its model of EA's beliefs. 
5.3 A Second Example 
Figure 17 contains a second negotiation dialogue. Due to space limitations, we will 
only discuss two interesting features of the dialogue and its processing by our system. 
Utterance (30) illustrates the use of a linguistic clue word to suggest an expression of 
doubt. In interpreting utterance (30), the system constructs an inference path contain- 
ing the action: 
Express-Doubt(EA, CA, Meets(CS510, MonYPM), Teaches(Dr. Jones, CS510)) 
Although the system does not have evidence that all of the applicability conditions 
for this Express-Doubt action are satisfied, the linguistic clue but does provide evidence 
for the generic Express-Doubt act. Since this is the only inference path containing an 
e-action for which there is evidence, the system recognizes (30) as expressing doubt 
at the proposition that CS510 meets on Monday at 7PM by contending that Dr. Jones 
is teaching CS510. In this case, the system lacked evidence for the third applicability 
condition in the recipe for Express-Doubt. But having recognized that EA is expressing 
doubt, it attributes to EA the beliefs captured in the applicability conditions. In partic- 
ular, the system attributes to EA the belief that Dr. Jones teaching CS510 implies that 
CS510 would not meet on Monday at 7PM, though it has no idea why EA believes 
that this implication holds--perhaps EA befieves that Dr. Jones has to be home to take 
care of his children at night. 
When utterance (33) occurs, there are three propositions that have not yet been 
accepted by EA, and the system considers the possibility that EA is performing one 
of three express doubt actions, namely 
Express-Doubt (EA, 
Express-Doubt (EA, 
Expre s s-Doubt (EA, 
CA, Meets(CS510, MonTPM), Graduate-Course(CS510)) 
CA, ~Teaches(Dr. Jones, CS510), Graduate-Course(CS510)) 
CA, Teaehes(Dr.Hart, CS510), Graduate-Course(CS510)) 
41 
Computational Linguistics Volume 25, Number 1 
In all three cases, the system lacks evidence that the third applicability condition in 
the Express-Doubt recipe is satisfied. However, the linguistic clue word but provides 
evidence for a generic Express-Doubt action. Since the system has equivalent evidence 
for all three of the Express-Doubt acts, contextual knowledge is used to choose among 
them. Since the proposition Teaches (Dr. Hart, CS510) is closest to the existing focus 
of attention in the discourse tree, it is the most salient of the three propositions that 
are open for rejection. Utterance (33) is therefore interpreted as expressing doubt at the 
proposition that Dr. Hart is teaching CS510 by contending that it is a graduate-level 
course. Thus contextual knowledge arbitrates when equivalent evidence is available 
for several specific discourse acts. 
6. Evaluation and Future Work 
We undertook an evaluation of our prototype system both to assess whether it derived 
appropriate interpretations of utterances and to identify areas for further research. We 
obtained eight human volunteers, six of whom are not engaged in NLP research and 
two of whom are involved in unrelated NLP projects. The subjects were given a set 
of world knowledge stereotypically believed in the domain, such as that faculty on 
sabbatical do not normally teach. The subjects were presented with a set of dialogues 
and asked to analyze several utterances from each dialogue. The selected utterances 
did not include simple questions initiating the dialogue or straightforward answers 
to questions, since it seemed likely that the subjects would agree with the system's 
interpretation and thus the results would be biased in favor of the system. The selected 
utterances did include surface negative questions (both with and without a clue word 
but), statements interpreted by our system as support for a previous assertion or as 
explanations about why a proposition was not in conflict with a previous claim, and 
examples of implicit acceptance. 
For each utterance selected for analysis, the subjects were given a suggested inter- 
pretation, and asked whether the suggested interpretation was reasonable and whether 
they could identify a better interpretation. 27 For 15 of 20 utterances, the subjects unani- 
mously believed that the system's interpretation was best. It should be noted there was 
unanimous agreement that utterance (42) below should be interpreted as an expression 
of doubt but that utterance (39) should not. 
Dialogue A 
(37) 
(38) 
(39) 
EA: Who is teaching architecture? 
CA: Dr. Smith is teaching architecture. 
EA: Isn't Dr. Smith an excellent teacher? 
Dialogue B 
(40) EA: Who is teaching architecture? 
(41) CA: Dr. Smith is teaching architecture. 
(42) EA: But isn't Dr. Smith an excellent teacher? 
There were two categories of utterances where the subjects disagreed. In the case 
of surface negative questions that did not express doubt, such as utterance (39) above, 
27 The subjects were not told that the suggested interpretation was the one produced by our system but 
only that we were trying to determine how utterances in a discourse should be interpreted. 
42 
Carberry and Lambert Modeling Negotiation Subdialogues 
the suggested interpretation given to the subjects was that the speaker was seeking 
information about whether the queried proposition was true. When the subjects did not 
interpret the utterance as an expression of doubt (see below), five of them contended 
that a better interpretation would be that EA was seeking verification of the queried 
proposition. Since our system already recognizes from the surface negative question 
that the speaker has a strong (but uncertain) belief in the queried proposition, it is 
easy to extend our system so that it can explicitly identify a Seek-Veri~cation discourse 
act. 
The other category for which there was disagreement was surface negative ques- 
tions where a clue word was not present and the stereotypical domain knowledge did 
not provide a conflict. In two of five instances, some subjects used their own experi- 
ence to identify a mutual belief that might suggest a conflict, such as the belief that 
sometimes certain faculty are not allowed to teach graduate level courses. While this 
knowledge cannot be captured as a default rule, it does represent a kind of shared 
experiential knowledge that would provide weak evidence for a potential conflict. 
However, it should be noted that our subjects were split on how these problematic 
cases should be interpreted, agreeing with the system's interpretation slightly more 
than half the time. There was also another such surface negative question where one 
subject viewed the system's interpretation as reasonable but argued that an expression 
of doubt would be a better interpretation. In order to derive this interpretation, the 
subject posited an attribute for the speaker that was neither evident from the dialogue 
nor stereotypically true. (The other subjects agreed that the system's interpretation was 
best.) These examples bear on the issue of accommodation mentioned in Section 4.5.1, 
since one could argue that the subjects who interpreted the utterances as expressions 
of doubt were trying to accommodate an incompatibility. This is particularly true in 
the last instance where the subject found it necessary to resort to nonshared knowledge 
in making the interpretation. However, it is unclear whether a speaker would expect a 
listener to recognize such utterances as expressions of doubt without additional clues. 
As noted below, our future research will consider other forms of evidence (gestural 
and intonational) in order to resolve such ambiguous utterances. 
After they had finished analyzing the dialogues, we asked the subjects to construct 
three dialogues containing an expression of doubt and to explain why the expression 
of doubt should be interpreted as such. While these dialogues provided no contradic- 
tions to our approach, they did provide a couple of interesting examples, such as the 
following dialogue, that suggest areas for future work. 
(43) EA: We have basil, parsley, and oregano, but we need marjoram. 
(44) CA: Isn't marjoram the same as oregano? 
Clearly (44) is expressing doubt at the claim conveyed by (43), but it relies on shared 
world knowledge that if a list contains X items, the X items are presumed to be 
different. Our system does not currently include such knowledge. 
Our subjects commented that intonation and facial gesture might alter their in- 
terpretation of the utterances in the dialogues; we are beginning research that will 
take these kinds of evidence into account (Carberry, Chu-Carroll, and Lambert 1996). 
In addition, we will be expanding the kinds of world knowledge incorporated into 
our system, and will be considering both the strength of different pieces of evidence 
and how several pieces of weak evidence affect interpretation. We would also like to 
extend our use of linguistic clues to include a wide variety of clue words and phrases 
and to recognize the functions that these words can play. In addition, we are devel- 
oping a plan-based response generation component (Chu-Carroll and Carberry 1994). 
43 
Computational Linguistics Volume 25, Number 1 
Initial work on this component includes a subsystem that can identify what evidence 
to present to a user when conflicts arise (Chu-Carroll and Carberry 1995b, 1998) and 
what information to request when the system cannot rationally decide whether to 
accept a proposition conveyed by the user (Chu-Carroll and Carberry 1995a, 1998). 
We will also be investigating the scale-up of our system as we extend its cover- 
age. Part of the motivation for the content of the current discourse recipes was their 
future extension to other domains, such as tutoring. For example, as discussed in Sec- 
tion 4.2.1, the formulation of our Ask-Ref recipe allows it to be used as a subaction of a 
future Test-Knowledge discourse act since the recipe does not presume that the speaker 
is ignorant about the correct value of the requested term. This should aid in extending 
the kinds of discourse acts that can be handled. Although transporting our system 
to another domain will require encoding new domain knowledge and new domain 
recipes, the recipes for discourse and problem-solving acts are domain-independent 
and thus will remain unchanged. Moreover, the knowledge captured in our recipes is 
communicative knowledge shared by dialogue participants; we believe that such com- 
municative knowledge (such as how to express doubt) is finite although the possible 
intentions (such as the intention of expressing doubt at Dr. Smith teaching CS360) are 
infinite. 
7. Other Related Work 
7.1 Grosz and Sidner's Theory of Discourse Processing 
Grosz and Sidner (1986) postulated a theory of discourse structure that included lin- 
guistic, intentional, and attentional components, and they argued that the dominance 
and satisfaction-precedes relationships between discourse segments must be identi- 
fied in order to determine discourse structure. They also noted three kinds of infor- 
mation that contribute to determining the purposes of discourse segments and their 
relationship to one another: linguistic markers, utterance-level intentions, and general 
knowledge about actions and objects. Subsequently Lochbaum (1994) developed an 
algorithm based on Grosz and Sidner's SharedPlan model (Grosz and Sidner 1990) 
that recognizes discourse segment purposes and discourse structure. 
We contend that, in order to understand utterances and respond appropriately, it 
is necessary not only to determine the structure of the discourse but also to identify 
the communicative acts that an agent intends to perform with an utterance. 2s For 
example, if a listener does not recognize when an utterance such as "Wasn't Dr. Smith 
on campus yesterday?" is expressing doubt, then the listener's response might fail to 
address the reasons for this doubt. Our research provides a computational algorithm 
that uses multiple knowledge sources to recognize complex discourse acts, including 
expressions of doubt, and to identify their relationship to one another. This algorithm 
and our strategy for recognizing implicit acceptance enable us to model negotiation 
subdialogues, something that previous systems have been unable to handle. 
7.2 Argument Understanding Systems 
Several researchers have built argument understanding systems, but none has ad- 
dressed participants coming to an agreement or mutual belief about a particular situa- 
tion, either because the researchers investigated monologues only (Cohen 1987; Cohen 
and Young 1991), or because they assumed that dialogue participants do not change 
28 In a dialogue, Grosz and Sidner's discourse segment purpose is intended to capture the purpose of a 
segment consisting of a series of utterances by both participants, not the communicative intentions 
underlying each participant's discourse actions. 
44 
Carberry and Lambert Modeling Negotiation Subdialogues 
their minds (Flowers, McGuire, and Birnbaum 1982; Quilici 1991). Cohen (1987) de- 
veloped an argument understanding system that used clue words and an evidence 
oracle to build a discourse structure for arguments based on which utterances served 
as support for other utterances. Cohen's model, however, handles only monologues, so 
responses to arguments are not modeled in her system. Birnbaum, Flowers, Dyer, and 
McGuire (Flowers and Dyer 1984; McGuire, Birnbaum, and Flowers 1981; Birnbaum, 
Flowers, and McGuire 1980) developed a system that finds flaws in arguments and 
determines how to respond. Quilici (1991) created a system in which agents respond 
to each other's arguments based on a justification pattern that will support the agent's 
position. Both Quilici and Birnbaum et al., however, assume that all participants in 
an argument will retain their opinion throughout the course of the argument, and 
concentrate mainly on how to find flaws in arguments and construct responses based 
on those findings; they do not address actually winning arguments. Reichman (1981) 
modeled informal debates by using her idea of context spaces and expectations to de- 
termine who should respond and what possible topics might be addressed. However, 
she does not provide a detailed computational mechanism for recognizing the role of 
each utterance in a debate. 
7.3 Models of Collaborative Behavior 
Several models of discourse have recently been built which view conversation as a 
kind of collaborative behavior in which speakers try to make themselves understood 
and listeners work with speakers to help speakers attain this goal. 
Clark and Schaefer (1989) contend that utterances must be "grounded," or un- 
derstood, by both parties, but they do not address conflicts in belief, only lack of 
understanding. Walker (1992) has found many occasions of redundancy in collabora- 
tive dialogues, and explains these by claiming that people repeat themselves in order 
to ensure that each utterance has been understood. 29 Clark and Wilkes-Gibbs (1990) 
propose a collaborative model of dialogue in which referring is viewed as a collabo- 
rative process and each conversation unit is viewed as a contribution, which consists 
of 1) an utterance that performs a referring action, and 2) the utterances required 
to understand the referent described in the utterance. Heeman (1991) implemented 
this model in a plan-based collaborative model of dialogue that is able to plan and 
recognize referring expressions and their corrections. 
Other collaborative models assume that two participants are working together to 
achieve a common goal (Cohen and Levesque 1990a, 1991a, 1991b; Lochbaum, Grosz, 
and Sidner 1990; Lochbaum 1991; Grosz and Sidner 1990; Searle 1990). Searle (1990) 
proposes a model in which the two agents working together have a joint intention, a 
"we intention," instead of individual intentions. Cohen and Levesque (1990a, 1990b, 
1990c, 1991a, 1991b) have developed a formal theory in which agents are jointly com- 
mitted to accomplishing a goal, so both parties have individual intentions to accom- 
plish the goal as part of their joint commitment. Grosz, Lochbaum, and Sidner (Grosz 
and Sidner 1990; Lochbaum, Grosz, and Sidner 1990; Lochbaum 1991) have specified a 
system in which two agents are working to accomplish some common goal by build- 
ing a "shared plan" in which each agent holds certain beliefs and intentions. These 
beliefs and intentions indicate that the agents intend to perform some joint action, and 
that they believe they can perform this action. All of these models indicate the need 
for modeling collaborative dialogue, but none suggests a system that can handle the 
29 Another reason for repetition, she claims, is for centering (Grosz, Joshi, and Weinstein 1995), but she 
concentrates on repetitions that give evidence of understanding. 
45 
Computational Linguistics Volume 25, Number 1 
kind of negotiation subdialogues that people often engage in when trying to negotiate 
their conflicts in belief, even when they are both working towards the same goal. 
8. Conclusion 
We have presented a plan-based model for handling cooperative negotiation subdia- 
logues. Our system infers both the communicative actions that people pursue when 
speaking and the beliefs underlying these actions. Beliefs, and the strength of these 
beliefs, are recognized from the surface form of utterances and from the explicit and 
implicit acceptance of previous utterances. Our algorithm for recognizing discourse 
actions combines linguistic, contextual, and world knowledge in a unified framework. 
By combining these different knowledge sources, we are able to recognize complex 
discourse acts such as expressing doubt, to identify the relationship of utterances to 
one another, and to model negotiation subdialogues. Since negotiation is an integral 
part of multiagent activity, our process model addresses an important aspect of coop- 
erative interaction and thus is a step toward an intelligent and robust natural language 
consultation system. 
Acknowledgments 
This work was supported by the National Science Foundation under Grant No. IRI- 
9122026. The Government has certain rights in this material. We would like to thank 
Rachel Sacher for her help in our corpus analysis and the anonymous reviewers for 
their helpful comments on the manuscript. 
Appendix 
Discourse Recipe 
Action: Address-Acceptance(_agentl, _agent2, _proposition1) 
{_agent1 tries to make _proposition1 believable to _agent2} 
Recipe-Type: Decomposition 
Appl Cond: believe(_agentl, _proposition2 --~ ~_proposition3, \[S:C\]) 
believe(_agentl, believe(_agent2, _proposition3, \[W:S\]), \[W:C\]) 
Constraints: opposite(_propositionl, _proposition3) 
Body: Inform(_agentl, _agent2, _proposition2) 
Effects: believe(_agent2, believe(_agentl, _proposition2 ~ _proposition1, 
\[s:c\]), \[s:c\]) 
Goal: enhance-believability(_agentl, _agent2, _proposition1) 
Discourse Recipe 
Action: Address-Believability(_agentl, _agent2, _proposition1) 
{_agent1 and _agent2 address the believability of_proposition1} 
Recipe-Type: Decomposition 
Appl Cond: believe(_agentl, _proposition1, \[C:C\]) 
believe(_agentl, believe(_agent2, _proposition1, \[CN:S\]), \[0:C\]) 
Body: #Address-Acceptance(_agentl, _agent2, _proposition1) 
#Address-Unacceptance(_agent2, _agent1, _proposition1, _proposition2) 
#Convey-Acceptance-Explicitly(_agent2, _agent1, _proposition1) 
Effects: believability-addressed(_agentl, _agent2, _proposition1) 
Goal: same-mutual-beliefs(_agentl, _agent2, _proposition1) 
46 
Carberry and Lambert Modeling Negotiation Subdialogues 
Discourse Recipe 
Action: Address-Unacceptance(_agentl, _agent2, _proposition1, _proposition2) 
{By noting a conflicting _proposition2, _agent1 initiates negotiation of his unacceptance 
of_proposition1} 
Recipe-Type: 
Appl Cond: 
Body: 
Effects: 
Goal: 
Decomposition 
believe(_agentl, in-conflict(_propositionl, _proposition2), \[W:C\]) 
Express-Doubt(_agentl, _agent2, _proposition1, _proposition2) 
Resolve-Conflict(_agent2, _agent1, _proposition1, _proposition2) 
unacceptance-addressed(_agentl, _agent2, _proposition1) 
conflict-resolved(_propositionl, _proposition2) 
Discourse Recipe 
Action: Answer-Ref(_agentl, _agent2, _term, _proposition) 
{_agent1 answers _agent 2's question about the referent of_term in _proposition} 
Recipe-Type: Decomposition 
Appl Cond: believe(_agentl, want(_agent2, knowref(_agent2, _term, 
believe(_agentl, _proposition, \[C:C\]))), \[W:C\]) 
believe(_agentl, ~knowref(_agent2, _term, believe(_agentl, 
_proposition, \[C:C\])), \[W:C\]) 
believe(_agentl, instantiates(_propanswer, _term, _proposition), \[C:C\]) 
Constraints: salient(_proposition) 
Preconditions: question-accepted(_agentl, _agent2, _proposition) 
believe(_agentl, knowref(_agentl, _term, _proposition), \[C:C\]) 
Body: Inform(_agentl, _agent2, _propanswer) 
#Address-Answer-Acceptability(_agentl, _agent2, _propanswer) 
Effects: believe(_agent2, answered-question(_agentl, _proposition), \[W:C\]) 
Goal: knowref(_agent2, _term, believe(_agentl, _proposition, \[C:C\])) 
Discourse Recipe 
Action: Ask-Ref(_agentl, _agent2, _term, _proposition) 
{_agent1 tries to get _agent2 to tell him the referent of the _term in _proposition} 
Recipe-Type: Decomposition 
Appl Cond: want(_agentl, knowref(_agentl, _term, 
believe(_agent2, _proposition, \[C:C\]))) 
-~knowref(_agentl, _term, believe(_agent2, _proposition, \[C:C\])) 
Constraints: term-in(_term, _proposition) 
Body: Ref-Request(_agentl, _agent2, _term, _proposition) 
#Make-Question-Acceptable(_agentl, _agent2, _proposition) 
Effects: believe(_agent2, want(_agentl, 
Answer-Ref(_agent2, _agent1, _term, _proposition)), \[C:C\]) 
Goal: want(_agent2, Answer-Ref(_agent2, _agent1, _term, _proposition)) 
Discourse Recipe 
Action: Convey-Uncertain-Belief(_agentl, _agent2, _proposition) 
{_agent1 conveys an uncertain belief in _proposition} 
Recipe-Type: Specialization 
Body: Surface-Neg-YN-Question(_agentl, _agent2, _proposition) 
Surface-Tag-Question(_agentl, _agent2 _proposition) 
Effects: believe(_agent2, believe(_agentl, _proposition, \[W:S\]), \[S:C\]) 
Goal: believe(_agent2, believe(_agentl, _proposition, \[W:S\]), \[S:C\]) 
47 
Computational Linguistics Volume 25, Number 1 
Discourse Recipe 
Action: Explain-Claim(_agentl, _agent2, _proposition1, _proposition2) 
{_agentl explains why _proposition1 and _proposition2 are not in conflict} 
Recipe-Type: Decomposition 
App! Cond: believe(_agentl, believe(_agent2, _proposition2 ~ ~_propositionl, 
\[s:c\]), \[W:Cl) 
believe(_agentl, _proposition3 ~ -~in-conflict(_propositionl, 
_proposition2), \[S:C\]) 
believe(_agentl, believe(_agent2, _proposition2, \[W:C\]), \[S:C\]) 
Constraints: salient(_propositionl) 
salient (_proposition2) 
Body: Inform(_agentl, _agent2, _proposition3) 
Effects: claim-explained(_agentl, _agent2, _proposition1) 
Goal: believe(_agent2, _proposition2 ~ ~_propositionl, \[CN:CN\] ) 
Discourse Recipe 
Action: Express-Doubt(_agentl, _agent2, _proposition1, _proposition2) 
{_agent1 expresses doubt to _agent2 about _proposition1 by contending that _proposition2 
is true} 
Recipe-Type: 
Appl Cond: 
Constraints: 
Body: 
Effects: 
Goal: 
Decomposition 
believe(_agentl, believe(_agent2, _proposition1, \[S:C\]), IS:C\]) 
believe(_agentl, _proposition2, \[W:S\]) 
believe(_agentl, _proposition2 ~ ~_propositionl, \[S:C\]) 
salient( _proposition1 ) 
Convey-Uncertain-Belief(_agentl, _agent2, _proposition2) 
believe(_agent2, believe(_agentl, _proposition1, \[SN:WN\]), \[S:C\]) 
believe(_agent2, believe(_agentl, _proposition2 ~ ~_propositionl, 
\[S:Cl), \[s:c\]) 
believe(_agent2, want(_agentl, Resolve-Conflict(_agent2, _agent1, 
_proposition1, _proposition2)), \[S:C\]) 
want(_agent2, Resolve-Conflict(_agent2, _agent1, _proposition1, 
_proposition2)) 
Discourse Recipe 
Action: Inform(_agentl, _agent2, _proposition) 
{_agent1 informs _agent2 of_proposition} 
Recipe-Type: Decomposition 
Appl Cond: believe(_agentl, _proposition, \[C:C\]) 
believe(regent1, believe(_agent2, _proposition, \[CN:S\]), \[0:C\]) 
Body: Tell(_agentl, _agent2, _proposition) 
#Address-Believability(_agentl, _agent2, _proposition) 
Effects: believe(_agent2, want(_agentl, believe(_agent2, _proposition, 
\[c:c\])), \[c:c\]) 
Goal: believe(_agent2, _proposition, \[C:C\]) 
Discourse Recipe 
Action: Obtain-Info-Ref(_agentl, _agent2, _term, _proposition) 
{_agent1 learns from _agent2 the referent of_term in _proposition)} 
Recipe-type: Decomposition 
Appl Cond: believe(_agentl, knowref(_agent2, _term, _proposition), \[W:C\]) 
-~knowref(_agentl, _term, _proposition) 
48 
Carberry and Lambert Modeling Negotiation Subdialogues 
Constraints: 
Body: 
Effects: 
Goal: 
want(_agentl, knowref(_agentl, _term, _proposition)) 
term-in(_term, _proposition) 
Ask-Ref(_agentl, _agent2, _term, _proposition) 
Answer-Ref(_agent2, _agent1, _term, _proposition) 
information-sought(_agentl, _agent2, _proposition) 
knowref(_agentl, _term, _proposition) 
Discourse Recipe 
Action: Ref-Request(_agentl, _agent2, _term, _proposition) 
{_agent1 requests the referent of_term in _proposition} 
Recipe-Type: 
Constraint: 
Body: 
Effects: 
Goal: 
Specialization 
term-in(_term, proposition) 
Surface-WH-Question(_agentl, _agent2, _term, _proposition) 
believe(_agent2, requested(_agentl, _term, _proposition), \[C:C\]) 
believe(_agent2, want(_agentl, know-ref(_agentl, _term, 
believe(_agent2, _proposition, \[C:C\]))), \[C:C\]) 
Discourse Recipe 
Action: Resolve-Conflict(_agentl, _agent2, _proposition1, _proposition2) 
{_agent1 resolves the conflict of_proposition1 and _proposition2} 
Recipe-Type: Decomposition 
Appl Cond: believe(_agentl, believe(_agent2, _proposition2 ~ -~_propositionl, 
\[S:Cl), \[w:c\]) 
believe(_agentl, believe(_agent2, _proposition2, \[W:S\]), \[W:C\]) 
believe(_agentl, _proposition1, \[C:C\]) 
Constraints: equalorneg(_proposition2, _proposition3) 
salient(_propositionl) 
salient(_proposition2) 
Body: Inform(_agentl, _agent2, _proposition3) 
#Explain-Claim(_agentl, _agent2, _proposition1, _proposition2) 
Effects: conflict-addressed(_agentl, _agent2, _proposition1, _proposition2) 
Goal: believe(_agent2, in-conflict(_propositionl, _proposition2), \[CN:WN\]) 
Discourse Recipe 
Action: Surface-Neg-YN-Question(_agentl, _agent2, _proposition) 
{_agent1 makes a surface negative request about _proposition} 
Recipe-Type: Primitive 
Appl Cond: believe(_agentl, _proposition, \[S:S\]) 
Effects: asked-about(_agentl, _agent2, _proposition) 
Goal: asked-about(_agentl, _agent2, _proposition) 
Discourse Recipe 
Action: Surface-Say-Prop(_agentl, _agent2, _proposition) 
{_agent1 makes a surface utterance of_proposition to _agent2} 
Recipe-Type: Primitive 
Appl Cond: believe(_agentl, _proposition, \[C:C\]) 
Effects: said(_agentl, _agent2, _proposition) 
Goal: said(_agentl, _agent2, _proposition) 
Discourse Recipe 
Action: Surface-WH-Question(_agentl, _agent2, _term, _proposition) 
49 
Computational Linguistics Volume 25, Number 1 
{_agent1 makes a surface request for the _term in _proposition} 
Recipe-Type: Primitive 
Constraints: term-in(_term, _proposition) 
Effects: asked-for(_agentl, _agent2, _term, _proposition) 
Goal: asked-for(_agentl, _agent2, _term, _proposition) 
Discourse Recipe 
Action: Tell(_agentl, _agent2, _proposition) 
{_agent l tells _agent2 of_proposition} 
Recipe-Type: 
Appl Cond: 
Body: 
Effects: 
Goal: 
Decomposition 
believe(_agentl, _proposition, \[C:C\]) 
Surface-Say-Prop(_agentl, _agent2, _proposition) 
#Address-Understanding(_agentl, _agent2, _proposition) 
told-about(_agentl, _agent2, _proposition) 
believe(_agent2, believe(_agentl, _proposition, \[C:C\]), \[C:C\]) 
References 
Allen, James. 1979. A Plan-Based Approach to 
Speech Act Recognition. Ph.D. thesis, 
University of Toronto, Toronto, Ontario, 
Canada. 
Allen, James and C. Raymond Perrault. 
1980. Analyzing intention in utterances. 
Artificial Intelligence, 15:143-178. 
Allen, James and Lenhart Schubert. 1991. 
The trains project. Technical Report 91-1, 
Department of Computer Science, 
University of Rochester, Rochester, NY. 
Ballim, Afzal and Yorick Wilks. 1991. 
Beliefs, stereotypes, and dynamic agent 
modeling. User Modeling and User-Adapted 
Interaction, 1(1):33-66. 
Bartelt, Margaret. 1996. A computer 
program that recognizes rejected 
questions computationally. In Proceedings 
of NCUR-IO, pages 989-993. 
Birnbaum, Lawrence, Margot Flowers, and 
Rod McGuire. 1980. Towards an AI model 
of argumentation. In Proceedings of the First 
National Conference on Artificial Intelligence, 
pages 306-309. 
Bonarini, Andrea, Ernesto Cappelletti, and 
Antonio Corrao. 1990. Network-based 
management of subjective judgements: A 
proposal accepting cyclic dependencies. 
Technical Report 90-067, Dipartimento di 
Elettronica, Politecnico di Milano, Milano, 
Italy. 
Carberry, Sandra. 1985. A pragmatics based 
approach to understanding intersentential 
ellipsis. In Proceedings of the 23rd Annual 
Meeting, pages 188-197. Association for 
Computational Linguistics. 
Carberry, Sandra. 1987. Pragmatic modeling: 
Toward a robust natural language 
interface. Computational Intelligence, 
3:117-136. 
Carberry, Sandra. 1988. Modeling the user's 
plans and goals. Computational Linguistics, 
14(3):23-37. 
Carberry, Sandra. 1989. A Pragmatics-Based 
Approach to Ellipsis Resolution. 
Computational Linguistics, 15(2):75-96. 
Carberry, Sandra. 1990. Plan Recognition in 
Natural Language Dialogue. ACL-MIT Press 
Series on Natural Language Processing. 
MIT Press, Cambridge, MA. 
Carberry, Sandra, Jennifer Chu-Carroll, and 
Lynn Lambert. 1996. Modeling intention: 
Issues for spoken language dialogue 
systems. In Proceedings of the International 
Symposium on Spoken Dialogue, pages 
13-24. 
Cawsey, Alison, Julia Galiiers, Steven Reece, 
and Karen Sparck Jones. 1992. 
Automating the librarian: A fundamental 
approach using belief revision. Technical 
Report 243, University of Cambridge 
Computer Laboratory, Cambridge, 
England. 
Chu-Carroll, Jennifer and Sandra Carberry. 
1994. A plan-based model for response 
generation in collaborative task-oriented 
dialogues. In Proceedings of the Twelfth 
National Conference on Artificial Intelligence, 
pages 799-805. 
Chu-Carroll, Jennifer and Sandra Carberry. 
1995a. Generating information-sharing 
subdialogues in expert-user consultation. 
In Proceedings of the 14th International Joint 
Conference on Artificial Intelligence, pages 
1,243-1,250. 
Chu-Carroll, Jennifer and Sandra Carberry. 
1995b. Response generation in 
collaborative negotiation. In Proceedings of 
the 33rd Annual Meeting, pages 136-143. 
50 
Carberry and Lambert Modeling Negotiation Subdialogues 
Association for Computational 
Linguistics. 
Chu-Carroll, Jennifer and Sandra Carberry. 
1998. Collaborative response generation 
in planning dialogues. Computational 
Linguistics, 24(3):355-400. 
Clark, Herbert and Edward Schaefer. 1989. 
Contributing to discourse. Cognitive 
Science, pages 259-294. 
Clark, Herbert and Deanna Wilkes-Gibbs. 
1990. Referring as a collaborative process. 
In Philip Cohen, Jerry Morgan, and 
Martha Pollack, editors, Intentions in 
Communication. MIT Press, Cambridge, 
MA, pages 463-493. 
Cohen, Paul R. 1985. Heuristic Reasoning 
about Uncertainty: An Artificial Intelligence 
Approach. Pitman Publishing Company. 
Cohen, Philip and Hector Levesque. 1990a. 
Intention is choice with commitment. 
Artificial Intelligence, 42:213-261. 
Cohen, Philip and Hector Levesque. 1990b. 
Persistence, intention, and commitment. 
In Philip Cohen, Jerry Morgan, and 
Martha Pollack, editors, Intentions in 
Communication. MIT Press, Cambridge, 
MA, pages 33-70. 
Cohen, Philip and Hector Levesque. 1990c. 
Rational interaction as the basis for 
communication. In Philip Cohen, Jerry 
Morgan, and Martha Pollack, editors, 
Intentions in Communication. MIT Press, 
Cambridge, MA, pages 221-256. 
Cohen, Philip and Hector Levesque. 1991a. 
Confirmations and joint action. In 
Proceedings of the International Joint 
Conference on Artificial Intelligence, pages 
951-957. 
Cohen, Philip and Hector Levesque. 1991b. 
Teamwork. Technical Report 504, SRI 
International, Menlo Park, California. 
Cohen, Philip and C. Raymond Perrault. 
1979. Elements of a plan-based theory of 
speech acts. Cognitive Science, 3:177-212. 
Cohen, Robin. 1987. Analyzing the structure 
of argumentative discourse. Computational 
Linguistics, 13(1-2):11-24. 
Cohen, Robin and Mark Anthony Young. 
1991. Determining intended evidence 
relations in natural language arguments. 
Computational Intelligence, 7:110-118. 
Columbia University Transcripts. 1985. 
Transcripts derived from audiotape 
conversations made at Columbia 
University, New York, NY. Provided by 
Kathleen McKeown. 
DeKleer, Johan. 1986. An assumption-based 
TMS. Arti~cial Intelligence, 28:269-301. 
Driankov, Dimiter. 1988. Towards a 
Many-Valued Logic of Quanti~ed Belief. 
Ph.D. thesis, Linkoping University, 
Department of Computer and 
Information Science, Linkoping, Sweden. 
Elzer, Stephanie. 1995. The role of user 
preferences and problem-solving 
knowledge in plan recognition for expert 
consultation systems. In Proceedings of the 
IJCAI Workshop on the Next Generation of 
Plan Recognition Systems, pages 37-41. 
Flowers, Margot and Michael E. Dyer. 1984. 
Really arguing with your computer. In 
Proceedings of the National Computer 
Conference, pages 653-659. 
Flowers, Margot, Rod McGuire, and 
Lawrence Birnbaum. 1982. Adversary 
arguments and the logic of personal 
attack. In W. Lehnert and M. Ringle, 
editors, Strategies for Natural Language 
Processing. Lawrence Erlbaum Associates, 
Hillsdale, NJ, pages 275-294. 
Galliers, Julia Rose. 1991. Belief revision and 
a theory of con~nunication. Technical 
Report 193, University of Cambridge, 
Cambridge, England. 
Galliers, Julia Rose. 1992. Autonomous 
belief revision and communication. In 
P. Gardenfors, editor, Belief Revision, 
Cambridge tracts in theoretical computer 
science. Cambridge University Press, 
Cambridge, England. 
Grosz, Barbara, Aravind K. Joshi, and Scott 
Weinstein. 1995. Centering: A framework 
for modeling the local coherence of 
discourse. Computational Linguistics, 
21(2):203-225. 
Grosz, Barbara and Candace Sidner. 1986. 
Attention, intentions, and the structure of 
discourse. Computational Linguistics, 
12(3):175-204. 
Grosz, Barbara and Candace Sidner. 1990. 
Plans for discourse. In Philip Cohen, Jerry 
Morgan, and Martha Pollack, editors, 
Intentions in Communication. MIT Press, 
Cambridge, MA, pages 417-444. 
Harry Gross Transcripts. 1982. Transcripts 
derived from tapes of the radio talk show 
Harry Gross: Speaking of your money. 
Provided by the Dept. of Computer 
Science at the University of Pennsylvania. 
Heeman, Peter. 1991. A computational 
model of collaboration on referring 
expressions. Master's thesis, University of 
Toronto, September. Also Technical 
Report CSRI-251. 
Hinkelman, Elizabeth. 1989. Two constraints 
on speech act ambiguity. In Proceedings of 
the 27th Annual Meeting, pages 212-219. 
Association for Computational 
Linguistics. 
Hirschberg, Julia and Diane Litman. 1987. 
Now let's talk about now. In Proceedings 
of the 25th Annual Meeting, pages 163-171. 
51 
Computational Linguistics Volume 25, Number 1 
Association for Computational 
Linguistics. 
Joshi, Aravind K. 1982. Mutual beliefs in 
question-answer systems. In N. Smith, 
editor, Mutual Beliefs. Academic Press, NY, 
pages 181-197. 
Kautz, Henry. 1990. A circumscriptive 
theory of plan recognition. In Philip 
Cohen, Jerry Morgan, and Martha Pollack, 
editors, Intentions in Communication. MIT 
Press, Cambridge, MA, pages 105-133. 
Knott, Alistair and Robert Dale. 1994. Using 
linguistic phenomena to motivate a set of 
coherence relations. Discourse Processes, 
18(1):35-62. 
Knott, Alistair and Chris Mellish. 1996. A 
feature-based account of the relations 
signalled by sentence and clause 
connectives. Language and Speech, 
39(2-3):143-183. 
Lambert, Lynn. 1993. Recognizing Complex 
Discourse Acts: A Tripartite Plan-Based Model 
of Dialogue. Ph.D. thesis, University of 
Delaware, June. 
Lambert, Lynn and Sandra Carberry. 1991. 
A tripartite plan-based model of dialogue. 
In Proceedings of the 29th Annual Meeting, 
pages 47-54. Association for 
Computational Linguistics. 
Lambert, Lynn and Sandra Carberry. 1992. 
Modeling negotiation subdialogues. In 
Proceedings of the 30th Annual Meeting, 
pages 193-200. Association for 
Computational Linguistics. 
Lewis, D. 1979. Scorekeeping in a language 
game. Journal of Philosophical Logic, 
8:339-359. 
Litman, Diane and James Allen. 1987. A 
plan recognition model for subdialogues 
in conversation. Cognitive Science, 
11:163-200. 
Litman, Diane and Julia Hirschberg. 1990. 
Disambiguating cue phrases in text and 
speech. In Proceedings of the 13th 
International Conference on Computational 
Linguistics, pages 251-256. 
Lochbaum, Karen. 1991. An algorithm for 
plan recognition in collaborative 
discourse. In Proceedings of the 29th Annual 
Meeting, pages 33-38. Association for 
Computational Linguistics. 
Lochbaum, Karen. 1994. Using Collaborative 
Plans to Model the Intentional Structure of 
Discourse. Ph.D. thesis, Harvard 
University, Cambridge, MA. Technical 
Report: TR-25-94. 
Lochbaum, Karen, Barbara Grosz, and 
Candace Sidner. 1990. Models of plans to 
support communication: An initial report. 
In Proceedings of the Eighth National 
Conference on Artificial Intelligence, pages 
485-490. 
Marcu, Daniel. 1997. The rhetorical parsing 
of natural language text. In Proceedings of 
the 35th Annual Meeting, pages 96-103. 
Association for Computational 
Linguistics. 
McGuire, Rod, Lawrence Birnbaum, and 
Margot Flowers. 1981. Opportunistic 
processing in arguments. In Proceedings of 
the 1981 International Joint Conference on 
Artificial Intelligence, pages 58-60. 
McKeown, Kathleen R. 1983. Focus 
constraints on language generation. In 
Proceedings of the Third National Conference 
on Artificial Intelligence, pages 582-587. 
McRoy, Susan and Graeme Hirst. 1995. The 
repair of speech act misunderstandings 
by abductive inference. Computational 
Linguistics, 21(4):435-478. 
Perrault, Raymond. 1990. An application of 
default logic to speech act theory. In 
Philip Cohen, Jerry Morgan, and Martha 
Pollack, editors, Intentions in 
Communication. MIT Press, Cambridge, 
MA, pages 161-185. 
Perrault, Raymond and James Allen. 1980. 
A plan-based analysis of indirect speech 
acts. American Journal of Computational 
Linguistics, 6(3-4):167-182. 
Polanyi, Livia. 1986. The linguistic discourse 
model: Towards a formal theory of 
discourse structure. Technical Report 
6409, Bolt Beranek and Newman 
Laboratories Inc., Cambridge, MA. 
Pollack, Martha. 1990. Plans as complex 
mental attitudes. In Philip Cohen, Jerry 
Morgan, and Martha Pollack, editors, 
Intentions in Communication. MIT Press, 
Cambridge, MA, pages 77-104. 
Quilici, Alexander. 1991. The Correction 
Machine: A Computer Model of Recognizing 
and Producing Belief Justifications in 
Argumentative Dialogs. Ph.D. thesis, 
Department of Computer Science, 
University of California at Los Angeles, 
Los Angeles, CA. 
Ramshaw, Lance A. 1989. A Metaplan 
model for problem-solving discourse. In 
Proceedings of the Fourth Conference of the 
European Chapter of the Association for 
Computational Linguistics, pages 35-42. 
Reichman, Rachel. 1978. Conversational 
coherency. Cognitive Science, 2:283-327. 
Reichman, Rachel. 1981. Modeling informal 
debates. In Proceedings of the 1981 
International Joint Conference on Artificial 
Intelligence, pages 19-24. 
Reichman, Rachel. 1985. Getting Computers to 
Talk Like You and Me. MIT Press, 
Cambridge, MA. 
52 
Carberry and Lambert Modeling Negotiation Subdialogues 
Reithinger, Norbert and Elisabeth Maier. 
1995. Utilizing statistical dialogue act 
processing in verbmobil. In Proceedings of 
the 33rd Annual Meeting, pages 116--121. 
Association for Computational 
Linguistics. 
RosG Carolyn Penstein, Barbara Di Eugenio, 
Lori Levin, and Carol Van Ess-Dykema. 
1995. Discourse processing of dialogues 
with multiple threads. In Proceedings of the 
33rd Annual Meeting, pages 31-38. 
Association for Computational 
Linguistics. 
Schegloff, Emanuel and Harvey Sachs. 1973. 
Opening up closings. Semiotica, 8:289-327. 
Schiffrin, Deborah. 1987. Discourse Markers. 
Cambridge University Press, Cambridge, 
England. 
Searle, John. 1970. Speech Acts: An Essay in 
the Philosophy of Language. Cambridge 
University Press, London, England. 
Searle, John. 1990. Collective Intentions and 
Actions. In Philip Cohen, Jerry Morgan, 
and Martha Pollack, editors, Intentions in 
Communication. MIT Press, Cambridge, 
MA, pages 401-416. 
SRI Transcripts. 1992. Transcripts derived 
from audiotape conversations made at 
SRI International, Menlo Park, CA. 
Prepared by Jacqueline Kowtko under the 
direction of Patti Price. 
Thomason, Richmond. 1990. 
Accommodation, meaning, and 
implicature: Interdisciplinary foundations 
for pragmatics. In Philip Cohen, Jerry 
Morgan, and Martha Pollack, editors, 
Intentions in Communication. MIT Press, 
Cambridge, MA, pages 325-363. 
Traum, David. 1994. A Computational Theory 
of Grounding in Natural Language 
Conversation. Ph.D. thesis, University of 
Rochester. 
Traum, David and Elizabeth Hinkelman. 
1992. Conversation acts in task-oriented 
spoken dialogue. Computational 
Intelligence, 8(3):575-599. 
van Beek, Peter and Robin Cohen. 1986. 
Towards user specific explanations from 
expert systems. In Proceedings of the Sixth 
Canadian Conference on Arti~'cial Intelligence, 
pages 194-198. 
Walker, Marilyn. 1991. Redundancy in 
collaborative dialogue. In Working Notes 
for the AAAI Fall Symposium: Discourse 
Structure in Natural Language Understanding 
and Generation, pages 124-129. 
Walker, Marilyn. 1992. Redundancy in 
collaborative dialogue. In Proceedings of the 
Fifteenth International Conference on 
Computational Linguistics, pages 345-351. 
Walker, Marilyn. 1996. Inferring acceptance 
and rejection in dialog by default rules of 
inference. Language and Speech, 
39(2-3):265-304. 
Walker, Marilyn and Steve Whittaker. 1990. 
Mixed initiative in dialogue: An 
investigation into discourse segmentation. 
In Proceedings of the 28th Annual Meeting, 
pages 70-78. Association for 
Computational Linguistics. 
Wilensky, Robert. 1981. Meta-Planning: 
Representing and using knowledge about 
planning in problem solving and natural 
language understanding. Cognitive Science, 
5:197-233. 
Young, Mark Anthony. 1987. The design 
and implementation of an evidence oracle 
for the understanding of arguments. 
Technical report, University of Waterloo, 
Waterloo, Ontario, Canada. 
53 

