USER MODELS AND DISCOURSE MODELS 
David N. Chin 
Department of Information and Computer Services 
University of Hawaii at Manoa 
2565 The Mall 
Honolulu, HI 96822 
A user model (UM) contains information about users, 
such as users' goals, beliefs, knowledge, preferences, 
and capabilities. A discourse model (DM) contains 
information about the conversation, such as the linguis- 
tic structure, the attentional state, and the intentional 
structure \[Grosz and Sidner 1986). Given these defini- 
tions, I will argue that the UM intersects the DM. That 
is, the UM contains items that are missing from the UM; 
the DM contains items that are missing from the DM; 
and the two share some items. 
First, Schuster (1987) argues convincingly that the 
UM contains items that are missing from the DM. This 
is especially evident in cases where the speaker and 
listener have a long association, and hence the speaker 
has a large amount of prior knowledge about the listener 
which is stored in the speaker's UM of the listener. 
However, this information is not present in the DM, 
which starts off empty. 
Next, the DM contains items that are missing from 
the UM. To support this argument, I will cite an 
example given by Wahlster. Suppose the speaker men- 
tions a long list of names that is not familiar to the 
listener. Then the speaker knows that the listener 
cannot know nameX, which is a particular name in the 
middle of the list. This is represented in the speaker's 
UM of the listener. However, nameX is in the linguistic 
structure of the speaker's DM since nameX was part of 
the discourse. Thus part of the linguistic structure of the 
DM is not represented in the UM. 
Actually, Wahlster interprets his example as an ar- 
gument that the DM and the UM must be different. He 
argues that since nameX is in the DM, then the UM 
must be different, or else the speaker could not repre- 
sent the fact that the user does not know nameX. This 
argument misses a crucial point. Although it is true that 
the names are listed in the linguistic structure of the 
DM, these names are not present in the attentional 
structure of the DM. Since the attentional structure of 
the DM is that part of the DM which represents objects, 
properties, and relations with which the user is familiar, 
this is the proper part of the DM to compare with the 
UM. To show that these names are not in the attentional 
structure of the DM, consider whether it is possible to 
use a pronoun to refer to nameX. If all of the other 
names were male, and nameX were female, then the 
speaker should be able to use the pronoun "she" in 
place of nameX. However, human speakers generally 
would not use such a pronoun. So there is no disagree- 
ment between the attentional structure of the speaker's 
DM and the speaker's UM of the listener. 
Not only is there no disagreement between the atten- 
tional structure of the DM and the UM, but I would like 
to argue that the items in the attentional structure of the 
DM are part of the UM. Take the related example of 
when the speaker introduces a person's name, nameX, 
unknown to the listener. In this case, the DM model 
represents the fact that nameX is the name of a person 
and that nameX refers to some person. Later, the 
speaker can use either nameX to refer to this person, or 
a pronoun. Likewise in the UM, the speaker represents 
the fact that the listener knows nameX refers to some 
unknown person, and represents the fact that the lis- 
tener does not know the person referred to by nameX. 
In this example, the contents of the attentional structure 
of the DM is a subset of the UM. 
Another part of the DM that intersects with the UM 
is the intentional structure of the DM. The intentional 
structure is made up of the immediate goals of the user 
as expressed in the user's utterances plus some higher 
level goals. These are also present by definition in the 
UM. 
Another compelling argument for the view that the 
DM intersects the UM is the phenomenon of multi- 
speaker discourses. In multispeaker discourses, each 
speaker needs to keep not only a separate UM for each 
listener, but also a separate DM for each participant. 
For example, consider the following dialog among three 
people debugging a circuit board. 
Copyright 1988 by the Association for Computational Linguistics. Permission to copy without fee all or part of this material is granted provided 
that the copies are not made for direct commercial advantage and the CL reference and this copyright notice are included on the first page. To 
copy otherwise, or to republish, requires a fee and/or specific permission. 
0362-613X/88/0100e-e$03.00 
86 Computational Linguistics, Volume 14, Number 3, September 1988 
David N. Chin User Models and Discourse Models 
Tom: The 777 timer is really heating up. 
Dick: Let's check it. 
Tom (whose hands are full): OK. Dick, could you 
check the frequency at the output pin. Harry, 
could you check the voltage at the power pin here. 
(Tom points to the power pin.) 
Tom and Dick are experts, while Harry knows little 
about hardware. Tom knows this, so Tom has a sepa- 
rate DM for Dick and Harry. In Tom's DM for Dick, 
Dick's attentional state includes the 777 timer in his 
focus space. However, since Tom knows that Harry 
cannot identify a 777 timer, the timer is not in Tom's 
DM for Harry. So, when Tom tells Dick to check the 
frequency at the output pin, Tom knows that Dick will 
understand this referent. On the other hand, Harry 
would not know the referent of the power pin, so Tom 
points this out to Harry. 
If there were only a single DM for the entire conver- 
sation, then Tom would not be able to represent the 
different attentional states of Dick and Harry. This 
argues for the view that DMs are user dependent and 
hence are subparts of UMs. 
In some sense, the above scenario is somewhat 
aberrant in that usually speakers assume that their 
listeners share the same attentional state. So it may be 
more efficient to only represent the separate attentional 
states of different listeners as differences from the 
norm. Thus in most cases, the list of differences would 
be very small and the efficiency would be effectively the 
same as having only one DM. 
Although the DM and UM share some data, they are 
used for fairly different purposes. DMs are used in the 
generation and understanding of references such as 
noun phrases and pronouns. DMs are also used in the 
generation and understanding of connectives such as 
cue words and phrases. On the other hand, UMs mainly 
used in deciding how to respond to the user. For 
example, UMs are useful for detecting user misconcep- 
tions (McCoy 1983, 1985) and deciding which concepts 
need to be explained to the user (Chin 1986). Sometimes 
these differences lead to the confusion that the data 
stored in UMs and DMs must be different, since their 
applications are so different. 
Another difference between UMs and DMs is in how 
they are built up. Although both DMs and UMs are built 
up from propositions expressed in the conversation, the 
DM expires at the end of the discourse, while parts of 
the UM are kept for future use. Grosz and Sidner (1986) 
discuss how DMs are built up, and Chin (1986), Litman 
and Allen (1984), Carberry (1983), and Allen and Per- 
rault (1980), among others, discuss this process for 
different aspects of UMs. 
In summary, the DM and UM are not separate, but 
rather share common parts. Shared parts include the 
intentional structure of the discourse and the attentional 
structure of the discourse. In addition, the DM contains 
the linguistic structure of the discourse, which is not 
present in the UM. Likewise, the UM contains many 
items that are not present in the DM. These include 
facts about the user which were learned in previou s 
dialogs and uncertain facts that were inferred from 
stereotypes to which the user belongs. 

REFERENCES 
Allen, J.F. and Perrault C.R. 1980 Analyzing Intention in Utterances. 
In Artificial Intelligence 15:143-178. 
Carberry, S. 1983 Tracking User Goals in an Information-Seeking 
Environment. In Proceedings of the 3rd National Conference on 
Artificial Intelligence, Washington, DC: 59--63. 
Chin, D.N. 1986 User Modeling in UC, the UNIX Consultant. In 
Proceedings of the Conference on Human Factors in Computing 
Systems, Boston, MA:24-28. 
Grosz, B. and Sidner C. 1986 Attention, Intentions, and the Structure 
of Discourse. In Computational Linguistics 12: 175-204. 
Litman, D.J. and Allen J.F. 1984 A Plan Recognition Model for 
Clarification Sub-dialogues. In Proceedings of the lOth Interna- 
tional Conference on Computational Linguistics and 22nd Annual 
Meeting of the Association for Computational Linguistics, Stan- 
ford University, Stanford, CA" 302-311. 
McCoy, K.F. 1983 Correcting Misconceptions: What to Say When the 
User is Mistaken. In Proceedings of the Conference on Human 
Factors in Computing Systems, Boston, MA: 197-201. 
McCoy, K.F. 1985 Correcting Object-Related Misconceptions. Ph.D. 
thesis, MS-CIS-85-57, Department of Computer and Information 
Science, University of Pennsylvania, Philadelphia, PA. 
McKeown, K.R.; Wish, M.; and Matthews, K. 1985 Tailoring Expla- 
nations for the User. In Proceedings of the 9th International Joint 
Conference on Artificial Intelligence, Los Angeles, CA:794-798. 
Wahlster, W. and Kobsa A. 1988 User Models in Dialog Systems. In 
Kobsa A. and Wahlster W. (eds.), User Models in Dialog Systems, 
Springer-Verlag, Berlin--New York. 
