Position Paper: Event Reference 
Bonnie Lynn Webber 
Department of Computer & Information Science 
University of Pennsylvania 
Philadelphia PA 19104-6389 
1. Introduction 
Developing a computational account of event reference requires solution to a number of difficult 
problems, not least of which is characterizing the phenomenon itself. From a listener's point of view, eveni 
reference encompasses both the task of building up a structured model of the events and situations 
underlying a given text and the task of interpreting subsequent references to these events and situations 
afterwards. A computational approach to these tasks requires at least (1) a characterization of the 
information that an individual clause may convey about an event or situation; (2) a characterization of 
explicit clues a text gives as to how the pieces described in individual clauses fit together (assuming, as I 
do, that this does not rely solely on world knowledge; (3) an account of what the listener does in 
processing an explicit event reference; (4) a characterization of what events and situations are available 
for explicit reference; and (5) a procedure for choosing among possible ways of resolving an explicit event 
reference. 
What I have to say in this paper concerns only the third problem. (In \[8\], I discuss both the second and 
third problems in detail.) My position on event reference centers on something I'm calling event/situation 
structure or e/s structure. Along with building up a discourse model of the entities salient to the.given 
text, the listener is also building up a model of the events and situations they participate in -- e/s 
structure. (I actually see e/s structure as representing a bit more than just the events and situations 
underlying the text: that is, I see it as representing the speaker's view of those events and situations. 
Hence e/s structure may also contain substructure corresponding to finer-grained descriptions of other 
events and situations in the structure, as well as substructure corresponding to the speaker's vantage 
point in describing an event or situation. 
This paper is limited to a discussion of e/s structure and its role in explicit pronominal reference to 
events. My view of pronominal reference is that it involves reference to entities in a discourse model \[7\]. 
While e/s structure in the current view is separate from the listener's discourse model, there are 
important connections between them. Of particular importance is that certain entities in the discourse 
model may directly represent nodes or substructure in e/s structure. Essentially, a discourse entity is no 
more than a common locus of reference and information attachment. Semantically, it may correspond to 
any type of individual -- an object, a set of objects, a generic class, etc. 
If and only if something is associated with a particular discourse entity, I will say that it is viewed as an 
individual. So, for example, if there are separate discourse entities corresponding to John and to Mary, 
but no discourse entity corresponding to the set comprising the two of them, I will say that that set is not 
(yet) viewed as an individual. Conversely, if a discourse entity corresponds to some set (say, one evoked 
by the phrase "three kangaroos") that set is viewed as an individual but the members may not be. The 
notion of being an individual -° that is, as having a separate, unique locus of reference and information 
attachment -- will become important shortly. 
158 
2. Individuating Reference 
In my view, e/s structure is distinct from the listener's discourse model. Therefore, an event or 
situation in e/s structure is not automatically associated with a discourse entity. However, the speaker 
may do something to individuate a piece of e/s structure and tie it to a new discourse entity in the 
listener's discourse model. This new entity is intimately tied to that particular piece of e/s structure, but 
is not equivalent to it. Individuation is done by means of definite NP or pronominal reference, as in the 
following examples 
Example 1 
It's always been presumed that when the glaciers receded, the area got very 
hot. The Folsum men couldn't adapt, and they died out. That's what's 
supposed to have happened. It's the textbook dogma. But it's wrong. They were 
human and smart. They adapted their weapons and culture, and they survived. 
Example 2 
"My friend," Valdez told him, "when you accept help, you must be prepared 
to take what one is capable of giving, not what you would like to receive. I 
have never denied my human limitations; but it is ungrateful of you to refer 
to them." 
Marvin had to be content with that, since he didn't think he could find his 
way back to the city unaided. 
In each of these examples, what I will call the individuating reference - the first, explicit reference to a 
node or larger piece of els structure - is italicized. (Example 1 also contains additional anaphoric 
references, which Ill get to shortly.) 
It should be clear that individuating reference, as presented here, involves special processing. In 
response to an individuating reference interpretable as referring to an event or situation, the listener goes 
to his/her evolving els structure, circumscribes an appropriate node or substructure, possibly coerces 
it into something of a different semantic type 1, and then evokes a new discourse entity corresponding to 
this new "individuated" event or situation. Figure 1 illustrates this relationship. The dashed ovals indicate 
circumscribed pieces of e/s structure or circumscribed sets of discourse entities. (See next page.) The 
arrows indicate the links between the new discourse entities and the newly individuated structure or sets 
they correspond to. 
I believe that individuating reference may not have been seen before as a distinct mechanism because 
the most common form of individuating reference - i.e., via a pronoun - is the same (at least in English) as 
that used for subsequent reference to entities already in the discourse model. Of course, this may not be 
truly accidental since that part of e/s structure which can be individuated is probably, in some sense, in 
"focus" in a way similar to the discourse entities which can be referenced pronominally. (Another obvious 
question which I have not attended to yet concerns the relationship of the temporal focus and other 
potential temporal foci to what part of e/s structure can be circumscribed via an individuating reference.) 
Individuation, as I conceive it, is not limited to e/s structure. It can be applied to the discourse model, 
1Coercion is commonly required in anaphoric/deictic reference to existing discourse entities. One open question concerning 
coercion is whether coercion of an existing discourse entity into one of another type results in the creation of a new discourse entity 
associated with the new type or whether the same discourse entity can support multiple descriptions of itself viewed as different 
types. 
159 
Q 
Figure 2.1: Individuating Reference 
to circumscribe several entities already in'the model and individuate a new set. I see this happening twice 
in the following example. 
Example 3 
It was time to go to the airport to fetch John. 
Marcia went into the living room, where she found Albert, while 
Simon she found in the back yard catching frogs. 
She told them to get cleaned up immediately. 
Ten minutes later, they met in the garage, got in the car and drove off. 
The first "them" individuates the new set consisting of Simon and Albert, associating a new discourse 
entity with it. The second "they" individuates the distinct set consisting of Marcia, Simon and Albert, 
associating a separate discourse entity with that set. (From example 2, where an explanation is 
individuated, it should also be clear that either the listener is building up e/s structure for rhetorical and 
expository text as well, or that individuation also operates on the distinct structure the listener is building 
up for them. If one wants to keep e/s structure as something organized primarily by temporal relations, 
built up through the interpretation of tense and relative temporal adverbials, the latter is probably 
preferable. Hence, individuation should be possible on any discourse structure.) 
The obvious question to ask is why one may want to individuate parts of e/s structure, when it is 
unnecessary for its construction. I believe the reason is that one cannot attribute features to something 
without pointing or referring to it, and one cannot point or refer to something without its being considered 
a distinct individual. In example 1, the narrative account in the first two sentences is individuated via "that" 
in the third sentence, in order to attribute to the account as a whole the status of "current guess". In 
example 2, Valdez's explanation is individuated, so as to establish it as the object of Marvin's acceptance. 
The account I'm giving here of individuating reference as being distinct from regular subsequent 
reference is born out by its different effect on discourse focus. That is, it has already been shown 
\[1,5, 2\] that regular pronominal reference to another individual can move the listener's attention to that 
individual, making it more difficult to refer successfully to the first individual with a pronoun. (The first 
160 
individual often has to be reintroduced with the aid of a definite noun phrase.) But now consider the 
following two examples, one containing an instance of individuating reference to a new set and the other, 
an instance of individuating reference to a part of e/s structure. 
Example 4 
$1. John called Fred to see if he wanted to go sailing on the Chesapeake. 
$2. Fred said fine but only if he could be the skipper. 
$3. He didn't trust John's sailing abilities. 
$4. They argued for awhile. 
$5. But eventually he agreed to take three watches and let John take one. 
*$5'. But eventually he agreed to take one watch and let Fred take the other 
three. 
Example 5 
$1. John asked Fred for a guava. 
$2. Fred refused: 
$3. He hated the smell of guavas. 
S4a. This led to an argument. 
b. That led to an argument. 
c./t led to an argument. 
$5: But eventually he agreed to give John a guava, if he'd eat it elsewhere. 
*$5'. But eventually he agreed to stop bothering Fred for pieces of fruit. 
In both examples, Fred is the focus of the third clause ($3). Sidner's focus movement algorithm predicts it 
as the most likely focus of the next clause. That is, if $5 (either example) were to follow directly on $3, 
Fred would be assumed a priorithe referent of "he". Hence the anomaly of $5', where the referent of "he" 
must be John. However, the insertion of a sentence containing an individuating reference like $4 in 
example 4 or any of S4a-c in example 5 does not disturb this: the same a priori expectations continue to 
hold. 
This makes sense if one treats individuating reference as introducing a new entity, even though its 
surface appearance makes it look like an instance of subsequent reference. In this way, it will not 
immediately affect the focus tracking process until the speaker makes another pronominal reference to it, 
which is now a real instance of subsequent reference that may indicate a real shift of focus off the 
previously focussed individual and onto the entity corresponding to the event, situation or set. 
Note that Sidner's focus movement algorithm \[5\] makes the same predictions as here about the effect 
of initial event/situation reference (although her account is limited to the event/situation set up by a single 
clause). What is different is that Sidner's predictions do not follow from a particular view of event/situation 
reference: rather, they follow from a particular way of using her regular focus mechanisms. In particular, 
using 'it' or 'this' to refer to an event/situation resets the current discourse focus (DF) of the text, pushing 
the previous one onto a separate focus stack whose entries are still available for pronominal reference. If 
the current DF is rejected, and the potential DFs (if any) are rejected as well, the previous DF would be 
popped off the stack and reset as the current DF. On the other hand, event/situation reference via 'that' 
has no effect on the current DF, which remains the preferred referent of an anaphoric expression in the 
next clause. Sidner's algorithm also makes the same predictions for elaborated sets, which she calls 
focus sets. Reference to a focus set resets the DF, and stacking the previous DF. Again that previous DF 
is still available for pronominal reference, whereupon it would be popped off the stack and reset as the 
current DF. 
161 
The difference between the account given in this paper and Sidner's account is that here the behavior 
is specifically attributable to a different mechanism -- individuating reference. The different pronouns (it, 
this, that) may perhaps differ in what they can individuate, but the individuation operation is still the same 
and has the same effect on the discourse: it introduces a new entity into the discourse model which, like 
any other entity, is a potential focus for the next utterance. But until it is actually taken to be the focus of 
the next utterance, it has no effect on the current focus. 
3. Conclusion 
I have shown here how a notion of e/s structure, separate from the listener's discourse model can 
form the basis of an account of explicit reference to events via a process of individuating reference. In 
\[8\], I show how e/s structure can also be used as a basis for the anaphoric interpretation of clause-level 
tense and relative temporal adverbials (which in turn extends els structure through the attachment of the 
clause's interpretation. That e/s structure can be seen to participate in these two quite different linguistic 
phenomena gives credence to its value in our pursuit of a complete theory of discourse understanding. 
Next steps include (1) integrating this with current work on the representation of events and situations 
\[6, 3\], (2) developing an account of what individuating reference has access to in e/s structure, and (3) 
developing a procedure for identifying and choosing among possible ways of resolving an individuating 
reference \[4\]. 
Acknowledgments 
The author would like to extend her grateful thanks to Debby Dahl, Martha Palmer and Becky 
Schiffman at SDC/A Burroughs Company for their enthusiastic support=and their trenchant criticism of 
earlier muddy ideas. 
162 
\[1\] 
\[2\] 
\[3\] 
\[4\] 
\[5\] 
\[6\] 
\[7\] 
\[8\] 
References 
Grosz, B. 
Focusing and Description in Natural Language Dialogues. 
In A. Joshi, B. Webber & I.A. Sag (editor), Elements of Discourse Understanding, pages 84-105. 
Cambridge Univ. Press, Cambridge, England, 1981. 
Grosz, B., Joshi, A. & Weinstein, S. 
Towards a Computational Theory of Discourse Interpretation. 
Paper in progress. 
Moens, M. 
Referring to Times. 
PhD thesis, University of Edinburgh, 1986 (forthcoming). 
Schuster, E. 
Reference to Events and Actions. 
Technical Report CIS-MS-86-??, Dept. of Comp. & Info Science, Univ of Pennsylvania, May, 
1986. 
Dcotoral thesis proposal. 
Sidner, C. 
Focusing in the Comprehension of Definite Anaphora. 
In M. Brady & R. Berwick (editor), Computational Models of Discourse, pages 267-330. MIT 
Press, Cambridge MA, 1982. 
Steedman, M.J. 
Reference to Past Time. 
In R. Jarvella & W. Klein (editor), Speech, Place and Action, pages 125-157. Wiley, New York, 
1982. 
Webber, B. 
So What Can We Talk about Now? 
In M. Brady & R. Berwick (editor), Computational Models of Discourse, pages 331-371. MIT 
Press, Cambridge MA, 1982. 
Webber, B. 
Two Steps Closer to Event Reference. 
Technical Report CIS-86-75, Dept. of Comp. & Info Science, Univ. of Pennsylvania, October, 
1986. 
163 
