TENSE AS DISCOURSE ANAPHOR 
Bonnie Lynn Webber 
Department of Computer and Information Science 
University of Pennsylvania 
Philadelphia, PA 19104-6389 
In this paper, I consider a range of English expressions and show that their context-dependency can be 
characterized in terms of two properties: 
1. They specify entities in an evolving model of the discourse that the listener is constructing; 
2. The particular entity specified depends on another entity in that part of the evolving "discourse 
model" that the listener is currently attending to. 
Such expressions have been called anaphors. I show how tensed clauses share these characteristics, 
usually just attributed to anaphoric noun phrases. This not only allows us to capture in a simple way the 
oft-stated but difficult-to-prove intuition that tense is anaphoric, but also contributes to our knowledge 
of what is needed for understanding narrative text. 
1 INTRODUCTION 
In this paper, I consider a range of English expressions 
and show that their context-dependency can be charac- 
terized in terms of two properties: 
1. They specify entities in an evolving model of the 
discourse that the listener is constructing; 
2. the particular entity specified depends on another 
entity in that part of the evolving "discourse 
model" that the listener is currently attending to. 
Two types of expressions have previously been de- 
scribed in these terms: definite pronouns and certain 
definite noun phrases (NPs). Researchers in computa- 
tional linguistics and in artificial intelligence have called 
these expressions anaphors (cf., Woods 1978, Sidner 
1983, Bobrow 1977, Hirst 1981, Webber 1983). 
Linguists, however, have used this term somewhat 
differently. Many have restricted its use to expressions 
(usually pronouns) that can be treated analogously to 
variables in a logical language (Chomsky 1980). A view 
in linguistics that comes somewhat closer to the AI 
model can be found in a paper by Sag and Hankamer 
(1984), who distinguish what they call deep (or model- 
interpretive) anaphora from what they call surface ana- 
phora (or ellipsis). Under the former, they include 
personal pronouns, sentential "it," and null-comple- 
ment anaphora, and under the latter, verb phrase (VP) 
ellipsis, sluicing, gapping, and stripping. The two types 
are distinguished by whether they make reference to the 
interpretation of an antecedent--i.e., some object in a 
model of the world constructed by the interpreter of the 
sentence of discourse (deep anaphora)---or whether 
they are interpreted with respect to a previous logical 
form (surface anaphora). While their deep anaphors 
include pronouns, Hankamer and Sag do not consider 
other expressions like NPs in discourse that might also 
be described in similar model-interpretive terms, nor do 
they describe in any detail how model interpretation 
works for the expressions they consider. 
To avoid confusion then, I will use the term discourse 
anaphors for expressions that have these two 
properties.l My main point will be that tensed clauses 
share these properties as well, and hence should also be 
considered discourse anaphors. This will capture in a 
simple way the oft-stated, but difficult-to-prove intu- 
ition that tense is anaphoric. 
To begin with, in Section 2, I characterize the depen- 
dency of an anaphoric expression X b on a discourse 
entity Ea in terms of an anaphoric function a(Xb,Ea), 
that itself depends on 1. the ontology of the specified 
entity E a and 2. discourse structure and its focusing 
effect on which Ea entities the listener is attending to. 
With respect to definite pronouns and NPs, this will 
essentially be a review of previous research. However, 
I will argue that some indefinite NPs should also be 
considered discourse anaphors in just this same way. In 
Section 3, I will move on to tensed clauses and the 
notion of tense as anaphor, a notion that goes back to at 
least Leech in his monograph Meaning and the English 
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/88/010061-73503.00 
Computational Linguistics, Volume 14, Number 2, June 1988 61 
Bonnie Lynn Webber Tense as Discourse Anaphor 
Verb (1987). I will review previous attempts to make the 
notion precise, attempts that require special-purpose 
machinery to get them to work. Then I will show, in 
contrast, that the notion can more simply be made 
precise in terms of a set of similar anaphoric functions 
that again depend on ontology and discourse structure. 
Making clear these dependencies contributes to our 
knowledge of what is needed for understanding narra- 
tive text. 
2 BACKGROUND 
2.1 DISCOURSE MODELS AND SPECIFICATION 
The notion specify that I am using in my definition of 
discourse anaphora is based on the notion of a Discourse 
Model, earlier described in Webber (1983). My basic 
premise is that in processing a narrative text, a listener 
is developing a model of at least two things: 1. the 
entities under discussion, along with their properties 
and relationships to one another, and 2. the events and 
situations under discussion, along with their relation- 
ships to one another (e.g., consequential relations, 
simple ordering relations, elaboration relations, etc.). 
The representation as a whole I call the listener's 
Discourse Model. 2 
In this section, I will focus on NPs. (In Section 3, I 
will turn attention to tensed clauses.) NPs may evoke 
entities into the listener's Discourse Model correspond- 
ing to individuals (Example 1), sets (Example 2), ab- 
stract individuals (Example 3), classes (Example 4), 
etc. 3 
1. a. Wendy gave Eliot a T-shirt for Christmas. 
b. Unfortunately, (it, the T-shirt) had the logo 
"You ate it, Ralph". 
2. a. Wendy gave each boy a T-shirt. 
b. (They, The T-shirts) each had a different logo 
on the front. 
3. a. The vice president must be over 35 years old. 
b. He or she must also be able to count. 
4. a. The dachshund down the block bit me 
yesterday. 
b. They're really vicious beasts. 
An NP which evokes a discourse entity also specifies it. 4 
One way an NP would be considered anaphoric by the 
above definition would be if it specified an entity E a in 
the model that had already been evoked by some other 
NP. (In that case, one would say that the two NPs 
co-specified the same entity.) This basic arrangement is 
illustrated in Examples 1-3 above and is shown in 
Figure la. 5 Formally, one could say that there is an 
anaphoric function a, whose value, given the anaphoric 
noun phrase NP b and the discourse entity E a, is Ea-- 
that is, a(NPb,Ea) = E a. This can also be read as NP b 
specifies E~ by virtue of Ea. Definite pronouns are most 
often anaphoric in just this way. 
The other way an NP would be considered a dis- 
course anaphor would be if it used some existing 
discourse entity E a to evoke and specify a new dis- 
course entity Eb, as in 
 o_s 0ify/ b 
evoko \ / specify ~ / specify evoke~ /evoke 
Ea Ea--E b 
Figure 1. Evoke, specify, and co-specify. 
5. a. A bus came round the corner. 
b. I signaled to the driver to stop. 
where NPb--the driver--makes use of the entity asso- 
ciated with the bus mentioned in 5a to specify a new 
entity--the driver of that bus. 
Here the anaphoric function is of the form a(NPb,Ea) 
= E b. In cooperative discourse, there have to be 
constraints on the value of a(NPb,Ea), since only NP b is 
given explicitly. In short, a cooperative speaker must be 
able to assume that the listener is able to both infer a 
possible a and single out Ea in his/her evolving Dis- 
course Model. 6 (This is illustrated in Figure lb.) I will 
consider each of these two types of constraints in turn. 
Speakers assume listeners will have no problem with 
a when a(NPb,E~) = E a. Inferring a in other cases 
follows in large part from the ontology of the entities 
specified by NPs--i.e., the ontology of our concepts of 
individuals, sets, mass terms, generics, etc. We view 
these as having parts (e.g., car: the engine, the wheels), 
having functional relations (e.g., car: the driver), having 
roles (e.g., wedding: the bride), etc. These needn't be 
necessary parts, relations, roles, etc. Our ontology 
includes possible parts, relations, etc., and these too 
make it possible for the listener to infer an a such that 
a(NPb,Ea) = E b (e.g., room: the chandelier; car: the 
chauffeur; wedding: the flower girl). Such inferences 
are discussed at length in the literature, including Clark 
and Marshall 1981, and Hobbs 1987. 7 
Before closing this section, there are two more things 
to say about NPs. First, the above definition of dis- 
course anaphor does not apply to all definite NPs: a 
definite NP can be used to refer to something unique in 
the speaker and listener's shared spatio-temporal con- 
text (e.g., the telephone--i.e., the one that they both 
hear ringing) or their shared culture (e.g., the govern- 
ment), to the unique representative of a class (e.g., the 
duck-billed platypus), to an entire class or set (e.g., the 
stars), or to a functionally defined entity (e.g., the 
62 Compulational Linguistics, Volume 14, Number 2, June 1988 
Bonnie Lynn Webber Tense as Discourse Anaphor 
largest tomato in Scotland). None of these would be 
considered discourse anaphoric by the above definition. 
Secondly, though the definition implies that one must 
consider some indefinite NPs to be discourse anaphors, 
since they are essentially parasitic on a corresponding 
anaphoric definite NP, as in the following example: 
6. a. The driver stopped the bus when a passenger 
began to sing, "Aida". 
b. The driver stopped the bus when the passen- 
gers began to sing "Aida". 
The indefinite NP a passenger in (6a) can be para- 
phrased as some one of the passengers, and thus is 
parasitic on the anaphoric definite NP the passengers 
mentioned explicitly in (6b). This does not imply that all 
indefinite NPs are discourse anaphors. In Mary met a 
boy with green hair or Fred built an oak desk, the 
indefinite NPs do not need to be interpreted with 
respect to another discourse entity and some inferrable 
relationship with that entity, in order to characterize the 
discourse entity they specify. 
In the next section, I will discuss the second kind of 
constraint on the function a(NPb,E a) necessary for 
cooperative use of an anaphor--constraints on identifi- 
able Eas. These involve notions of discourse structure 
and discourse focus. Before I close, though, I want to 
point to where I'm going vis-a-vis the anaphoric char- 
acter of tense and tensed clauses. In contrast with 
previous accounts of tense as pronoun or tense as 
loosely context-dependent, I am going to claim that, like 
an anaphoric definite NP, 
• a tensed clause C b may either specify an existing event 
or situation Eb in the listener's Discourse Model, or it 
may both evoke and specify a new entity. 
• As with anaphoric NPs, there are constraints on 
possible anaphoric functions and on the Eas that can 
participate in them at any one time. 
• These functions are sensitive to that part of a tensed 
clause C b called by Reichenbach (1947) point of 
reference (here abbreviated RT), as well as its rela- 
tionship to Reichenbach's point of the event (ET). 
• These functions can be defined in part in terms of an 
independently justifiable ontology of events proposed 
(independently) by Moens and Steedman (this vol- 
ume) and Passonneau (this volume). 
• The constraints on E a are tied in with a temporal 
analogue of discourse focus that I have called tempo- 
ral focus (TF), and through TF, with discourse struc- 
ture as well. 
2.2 DISCOURSE FOCUS AND DISCOURSE STRUCTURE 
The ideas presented in this section have been formu- 
lated and developed by Barbara Grosz and Candy 
Sidner, originally independently and later in joint re- 
search. It is not a summary of their work: 8 it is limited 
to those of their ideas that are necessary to the concept 
of anaphor that I am advancing here and the concept of 
tense as anaphor, in particular. 
Sidner's thesis (1979, 1983) presents an account of 
understanding definite pronouns and anaphoric definite 
NPs that reflects the ease with which people identify the 
intended specificand of definite pronouns (except in 
highly ambiguous cases), as well as the intended spec- 
ificand of anaphoric definite NPs. 
With respect to noun phrases (but not clauses), 
Sidner makes the same assumption about evoking, 
specifying, and co-specifying in a Discourse Model that 
I have made here. To understand anaphoric expres- 
sions, Sidner postulates three mechanisms: 
1. a current discourse focus (DF) 
2. an ordered list of potential foci (PFL) for the next 
utterance 
3. a stack for saving the current DF and resuming it 
later. 
The DF corresponds to that entity the listener is most 
attending to. Pronouns can most easily specify the 
current DF, slightly less easily a member of the PFL, 
and with slightly more difficulty, a stacked focus. 
Specifying an entity pronominally can shift the listen- 
er's attention to it, thereby promoting it to be the next 
DF. Anything else specified in the clause ends up on the 
PFL, ordered by its original syntactic position. (Sidner 
introduced a separate "agent focus" to allow two 
entities to be specified pronominally in the same clause, 
but it was not a critical feature of her approach.) As for 
anaphoric definite NPs, they can specify anything pre- 
viously introduced (whether on the PFL, a stacked 
focus, or anything else) or anything related in a mutually 
inferrable way with the current DF or a member of the 
PFL. In terms of the constraints I mentioned above, it 
is only those discourse entities that are either the DF or 
on the PFL that can serve as Ea for an anaphoric 
definite NP. 9 
In Sidner (1983) DFs always are stacked for possible 
resumption later. In Grosz and Sidner (1986) it is an 
entire focus space (FS) (Grosz 1977) that gets stacked 
(i.e., the collection of entities L is attending to by virtue 
of the current discourse segment (DS)) but only when the 
9purpose of the current DS is taken to dominate that of 
the one upcoming. Dominance relations are also speci- 
fied further according to the type of discourse. In Grosz 
and Sidner, they are defined for task-related' dialogues 
and arguments. For example, in arguments, one DS 
purpose (DSP) dominates another if the second pro- 
vides evidence for a point made in the first. When the 
dominated DSP is satisfied, its corresponding FS is 
popped. This stack mechanism models the listener's 
attentional state. The relations between DSPs constitute 
the intentional structure of the text. Getting a listener to 
resume a DS via the stack mechanism is taken to require 
less effort on a speaker's part than returning to elabo- 
rate an argument or subtask description later on. 
The significance of Sidner (1983) and Grosz and 
Sidner (1986) for the current enterprise is that: 
Computational Linguistics, Volume 14, Number 2, June 1988 63 
Bonnie Lynn Webber Tense as Discourse Anaphor 
• Sidner essentially shows how DF can move gradually 
among the discourse entities that make up a focus 
space, as the listener is processing its associated 
discourse segment; 
• Grosz and Sidner show how DF can make a radical 
jump to a different (possibly newly evoked) discourse 
entity as the listener moves to process the next 
discourse segment. ~o 
I reinterpret this in the current framework in terms of 
the anaphoric function a(NPb,Ea). Within a discourse 
segment, the entity that is the DF is the most likely E a. 
Over the discourse segment, other discourse entities in 
the segment's focus space may in turn become DF. 
With a change in discourse segment, however, the DF 
can change radically to an entity in the focus space 
associated with the new segment. 
To hint again at what is to come: in Section 3.2, I will 
propose a temporal analogue of DF, which I have called 
temporal focus (TF). In Section 3.3, I will show how 
gradual movements of the TF are tied in with the 
ontology of what a tensed clause specifies--i.e., an 
ontology of events and situations--while more radical 
movements reflect the effect of discourse structure on 
TF. 
3 TENSE AS ANAPHOR 
Tense may not seem prima facie anaphoric: an isolated 
sentence like John went to bed or I met a man who 
looked like a basset hound appears to make sense in a 
way that a stand-alone He went to bed or The man went 
to bed does not. On the other hand, if some time or 
event is established by the context (i.e., either by an 
event or situation described in the previous discourse or 
by a temporal adverbial in the current sentence--cf. 
Passonneau, and Moens and Steedman, this volume), 
tense will invariably be interpreted with respect to it, as 
in: 
7. a. After he finished his chores, John went to bed. 
b. John partied until 3 am. He came home and 
went to bed. 
In each case, the interpretation of John's going to bed is 
linked to an explicitly mentioned time or event. This is 
what underlies all discussion of the anaphoric quality of 
tense. 
3.1 BACKGROUND 
The assumption that tense is anaphoric (i.e., that its 
interpretation is linked to some time or event derived 
from context) goes back many years, although it is not 
a universally held belief (cf. Comrie 1985). Leech seems 
to express this view in his Meaning and the English 
Verb: 
63 INDEFINITE TIME Whereas the Present Per- 
fect, in its indefinite past sense, does not name a 
specific point of time, a definite POINT OF ORI- 
ENTATION in the past is normally required for 
the appropriate use of the Simple Past Tense. The 
point of orientation may be specified in one of 
three ways: (a) by an adverbial express of time- 
when; (b) by a preceding use of a Past or Perfect 
Tense; and (c) by implicit definition; i.e., by 
assumption of a particular time from context. 
73 Tile Past Perfect Tense has the meaning of past- 
in-the-past, or more accurately, 'a time further in 
the past, seen from the viewpoint of a definite 
point of time already in the past'. That is, like the 
Simple Past Tense, the Past Perfect demands an 
already established past point of reference. 
(Leech: 47) 
Leech did not elaborate further on how reference points 
are used in the interpretation of simple past tense and 
past perfect tense, or on what has become the main 
problem in the semantics and pragmatics of tense: 
reconciling the (usual) forward movement of events in 
narratives with a belief in the anaphoric (or context- 
dependent) character of tense. 
The first explicit reference I have to tense being 
anaphoric like a definite pronoun is in an article by 
McCawley (1971:110), who said: 
However the tense morpheme does not just express 
the time relationship between the clause it is in and 
the next higher clause--it also refers to the time of 
the clause that it is in, and indeed, refers to it in a way 
that is rather like the way in which personal pronouns 
refer to what they stand for. 
McCawley also tried to fit in his view of tense as 
pronoun with the interpretation of tense in simple 
narratives. Here he proposed that the event described in 
one clause serves as the antecedent of the event de- 
scribed in the next, but that it may be related to that 
event by being either at the same time or "shortly after" 
it. He did not elaborate on when one relation would be 
assumed and when the other. 
Partee (1973) also noted the similarities between 
tense and definite pronouns. However, she subse- 
quently recognized that taking simple past tense as 
directly analogous with pronouns was incompatible 
with the usual forward movement of time in the inter- 
pretation in a sequence of sentences denoting events 
(Partee 1984). Her response was a modification of the 
claim that tense is anaphoric, saying: 
I still believe it is reasonable to characterize tense as 
anaphoric, or more broadly as context-dependent, 
but I would no longer suggest that this requires them 
to be viewed as 'referring' to times as pronouns 
'refer' to entities, or to treat times as arguments of 
predicates (256). 
The particular context-dependent process she proposes 
for interpreting tensed clauses follows that of Hinrichs 
1986, briefly described below. 
The examples presented above to illustrate the ana- 
phoric quality of tense were all simple past. However, 
as Leech notes (see above), the past perfect also makes 
demands on having some reference point already estab- 
64 Computational Linguistics, Volume 14, Number 2, June 1988 
Bonnie Lynn Webber Tense as Discourse Anaphor 
lished in the context. Thus it cannot be in terms of the 
event described in a tensed clause that tense is anapho- 
ric. Instead, several people (Steedman 1982, Hinrichs 
1986, Bauerle 1979) have argued that it is that part of 
tense called by Reichenbach (1947) the point of refer- 
ence (here abbreviated RT) that is anaphoric. This can 
be seen by considering the following example: 
8. a. John went to the hospital. 
b. He had twisted his ankle on a patch of ice. 
It is not the point of the event (here abbreviated ET) of 
John's twisting his ankle that is interpreted anaphori- 
cally with respect to his going to the hospital. Rather, it 
is the RT of the second clause: its ET is interpreted as 
prior to that because the clause is in the past perfect (see 
above). 
I will now review briefly Hinrichs's proposal as to 
how tensed clauses are interpreted in context, in order 
to contrast it with the current proposal. 
In Hinrichs 1986, Hinrichs makes the simplifying 
assumption that in a sequence of simple past sentences, 
the temporal order of events described cannot contra- 
dict the order of the sentences. This allows him to focus 
on the problem of characterizing those circumstances in 
which the event described by one sentence follows that 
described by the previous one (Example 9---Hinrichs's 
Example 15) and when it overlaps it (Example 10--- 
Hinrichs's Example 21): 
9. The elderly gentleman wrote out the check, tore 
it from the book, and handed it to Costain. 
10. Mr. Darby slapped his forehead, then collected 
himself and opened the door again. The brush 
man was smiling at him hesitantly. 
Hinrichs bases his account on the Aktionsart of a tensed 
clause (i.e., its Vendlerian classification as an accom- 
plishment, achievement, activity, or state--including 
progressives). Assuming an initial reference point in a 
discourse, the event described by a tensed clause inter- 
preted as an accomplishment or achievement will be 
included in that reference point and will also introduce 
a new reference point ordered after the old one. Events 
associated with the other Aktionsarten include the cur- 
rent reference point in the event time. This means that 
given a sequence of two clauses interpreted as accom- 
plishments or achievements, their corresponding events 
will follow one another (cf. Example 9). On the other 
hand, given a sequence with at least one tensed clause 
interpreted as an activity or state (including progres- 
sive), their corresponding events will be interpreted as 
overlapping each other (cf. Example 10). 
Hinrichs relates his reference point to that of Rei- 
chenbach. (Thus, the anaphoric character of tense is 
based on RT and not on the events directly.) However, 
Hinrichs's notion and Reichenbach's differ with respect 
to the time of the event described in the tensed clause. 
While Reichenbach talks about ET and RT being the 
same for nonprogressive past-tense clauses, in Hin- 
richs's account the reference point can fall after the 
event if a nonprogressive past is interpreted as an 
accomplishment or an achievement. This is necessary 
to achieve the forward movement of narrative that 
Hinrichs assumes is always the case (his simplifying 
assumption) but it is not the same as Reichenbach's RT. 
It also leads to problems in cases where this simplifying 
assumption is just wrong--where in a sequence of 
simple past tenses, there is what appears to be a 
"backward" movement of time, as in 
11. a. For an encore, John played the "Moonlight 
Sonata". 
b. The opening movement he took rather tenta- 
tively, but then... 
where the second clause should be understood as de- 
scribing the beginning of the playing event in more 
detail, not as describing a subsequent event. 
In the account given below, both forward and back- 
ward movement of time fall out of the anaphoric char- 
acter of tensed clauses, and the dependency of dis- 
course anaphora on discourse structure.11 
3.2 TENSE AS DISCOURSE ANAPHOR: IN WHAT SENSE 
"SPECIFY"? 
With that background, I will now show how tensed 
clauses share the two properties I set out in Section 1 
(repeated here) and hence are further examples of 
discourse anaphora: 
1. Anaphors specify entities in an evolving model of 
the discourse that the listener is constructing; 
2. the particular entity specified depends on another 
entity in that part of the evolving Discourse Model 
that the listener is currently attending to. 
To do this, I need to explain the sense in which tensed 
clauses specify and the way in which that specification 
can depend on another element in the current context. 
Recall that I presume that a listener's developing 
discourse model represents both the entities being dis- 
cussed, along with their properties and relations, and 
the events and situations being discussed, along with 
their relationships with another. For the rest of this 
paper, I want to ignore the former and focus on the 
latter. This I will call event/situation structure, or E/S 
structure. It represents the listener's best effort at 
interpreting the speaker's ordering of those events and 
situations in time and space. One problem in text 
understanding, then, is that of establishing where in the 
evolving E/S structure to integrate the event or situation 
description in the next clause. 
In this framework, a tensed clause C b provides two 
pieces of semantic information: (a) a description of an 
event or situation, and (b) a particular configuration of 
ET, RT, and point of speech (abbreviated ST). (Here I 
may be departing from Reichenbach in treating ET, RT, 
and ST explicitly as elements of linguistic semantics, 
quite distinct from entities of type "event" in the 
Computational Linguistics, Volume 14, Number 2, June 1988 65 
Bonnie Lynn Webber Tense as Discourse Anaphor 
Discourse Model.) Cb then specifies an entity E b in the 
Discourse Model whose temporal relationship to other 
events in the model follows (in part) from Cb'S particu- 
lar configuration of ET, RT, and ST. Both the charac- 
teristics of E b (i.e., its ontology) and the configuration 
of ET, RT, and ST are critical to my account of tense as 
discourse anaphor. 
The event ontology I assume follows that of Moens 
and Steedman (this volume) and of Passonneau (this 
volume). Both propose that people interpret events as 
having a tripartite structure (a "nucleus" in Moens and 
Steedman's terminology) consisting of a preparatory 
phase (prep), a culmination (cuD, and a consequent 
phase (conseq)--as in Figure 2. This tripartite structure 
permits a uniform account to be given of aspectual types 
in English and of how the interpretation of temporal 
adverbials interacts with the interpretation of tense and 
aspect. For example, the coercion of clauses from one 
interpretation to another is defined in terms of which 
parts of a nucleus they select and how those parts are 
described. 12 
prep cul conseq (lllllllllllllltllllllllllllll ~ 
Figure 2. Tripartite ontology of events. 
The ET/RT/ST configuration is significant in that, like 
Steedman 1982, Dowty 1986, Hinrichs 1986, and Partee 
1984, I take RT as the basis for anaphora. To indicate 
this, I single it out as an independent argument to 
anaphoric functions, here labelled/3. In particular, the 
following schema holds of a clause C b linked anaphori- 
cally to an event Ea through its RT: 
/3(Cb, Ea, RTb) = E b 
The relationship between E b and Ea then falls out as a 
consequence of I. the particular ET/RT/ST configuration 
of Cb; and 2. the particular function/3 involved. 
One possibility is that/3 links RT b directly to Ea--- 
i.e.: 
/30(Cb, Ea, RTb) = E b 
In this case, the relationship between E b and Ea then 
depends on the configuration of RT b and ET b. If ET b = 
RTb, then (minimally) E b is taken to coincide in some 
way with E a. This is shown in Figure 3a. If ET b < RT b 
(as in the perfect tenses), E b is taken to precede Ea. This 
is shown in Figure 3d. 
Alternatively, /3 may embody part of the tripartite 
ontology of events mentioned earlier:/3prep links RT b to 
the preparatory phase of E a (as shown in Figure 3b)--- 
i.e.: 
/3prep(Cb, Ea, RTb) = E b 
ca Cb ca Cb ca Cb  ,1//xS 
E a Ea Eb 
Eb I Eb E a prep conseq \] 
prep conseq 
(a) ~0 (b) 13conseq (c) 13prep 
Figure 3. Anaphoric relations for tensed clauses where ET 
= RT. 
Ca C b X 
E b E a 
(d) 130 
Figure 3 (continued). Anaphoric relations where ET < RT. 
while/3conseq links RT b to the consequent phase of E a (as 
shown in Figure 3c)--i.e.: 
/3conseq(Cb, E a, RTb) = Eb 
(There is a third possibility--that RTb links to the 
culmination of Ea--but it is not clear to me that it could 
be distinguished from the simpler /3o function given 
above, which links RT b to Ea itself. Also, while /3prep 
and /3co,,seq relations for RTb might theoretically be 
possible for a perfect, it is not clear to me that these 
cases could be distinguished from the simpler/3o- In the 
case of perfects therefore, the relation between E b and 
Ea is correspondingly indirect.'3 
The following example illustrates the case where/3 = 
/3 o and ETb = RTb. 
12. a. John played the piano. 
b. Mary played the kazoo. 
Sentence 12a. evokes a new event entity Ea describable 
as the event of John playing the piano. Since the tense 
of (12b) is simple past, ET b = RTb. Given /3o(Cb,Ea, 
RTb) = E b, then E b is interpreted as coextensive with 
Ea. (Whether this is further interpreted as two simulta- 
neous events or a single event of their playing a duet 
depends on context and, perhaps, world knowledge as 
well.) This is illustrated in Figure 4. Example 8 (re- 
peated here) illustrates the case/30 where ET b < RT b. 
66 Computational Linguistics, Volume 14, Number 2, June 1988 
Bonnie Lynn Webber Tense as Discourse Anaphor 
E a , E b 
(C )) 
Figure 4. Co-extensive events. 
i Now 
8. a. John went to the hospital. 
b. He had twisted his ankle on a patch of ice. 
Clause 8a. evokes an entity E a describable as John's 
going to the hospital. Since 8b is past perfect, ET b < 
RT b. Thus if /3o(Cb,Ea,RTb) = Eb, the event E b de- 
scribed by 8b is taken to be prior to E a. As Moens & 
Steedman (this volume) point out, the consequences of 
an event described with a perfect tense are still assumed 
to hold. Hence the overlap shown in Figure 5: 
E b E a ) ( ) 
I =" 
conseq( E b ) Now 
Figure 5. Ordered events. 
The next example illustrates /3conseq: 
13. a. John went into the florist shop. 
b. He picked out three red roses, two white ones 
and one pale pink. 
Clause 13a evokes an entity E a describable as John's 
going into a flower shop. Since Clause 13b is simple 
past, ET b = RT b. Thus given/3conseq(Cb,Ea,RT b) = E b, 
event E b is taken as being part of the consequent phase 
of Ea. That is, John's picking out the roses is taken as 
happening after his going into the florist shop. This is 
shown in Figure 6. The next example illustrates the case 
of/3prep: 
14. a. John bought Mary some flowers. 
b. He picked out three red roses, two white ones 
and one pale pink. 
Since 14b is simple past, ET b = RT b. Thus given 
/3prep(Cb,Ea,RTb) = Eb, event Eb--the event of picking 
out some roses--is taken as being part of the prepara- 
tory phase of the event E a, which when completed, can 
E Eb \[ ( a ) ( ) 
I conseq( E a ) 
Now 
Figure 6. Consequent-phase relation. 
be described as having bought some flowers. This is 
shown in Figure 7. 
E b E a ( ) ( ) 
prep( E ) t a 
Now 
Figure 7. Preparatory-phase relation. 
To summarize, I have claimed that: 1. the notion of 
specification makes sense with respect to tensed 
clauses; 2. one can describe the anaphoric relation in 
terms of the RT of a tensed clause Cb, its ET/RT 
configuration, and an existing event or situation entity 
E~----that is, /3(Cb,Ea,RT b) = Eb; and 3. there are (at 
least) three /3 functions---one, /3o, linking RT b to E a 
itself, the other two (/3prep and/3conseq) embodying parts 
of a tripartite ontology of events. In the next section, I 
will discuss constraints on the second argument to 
/3(Cb,Ea,RTb)--that is, constraints on which entities in 
the evolving E/S structure the specification of a tensed 
clause can depend on. 
3.3 TEMPORAL FOCUS 
Recall from Section 2.2 that Sidner introduced the 
notion of a dynamically changing discourse focus (DF) to 
capture the intuition that at any point in the discourse, 
there is one discourse entity that is the prime focus of 
attention and that is the most likely (although not the 
only possible) specificand of a definite pronoun. In 
parallel, I propose a dynamically changing temporal 
focus (TF), to capture a similar intuition that at any 
point in the discourse, there is one entity in E/S struc- 
ture that is most attended to and hence most likely to 
stand in an anaphoric relation with the RT of the next 
clause. That is, fl(Cb,TF,RT b) = E b. If Cb is interpreted 
as part of the current discourse segment, after its 
interpretation there are three possibilities: 
I. With/30, the TF will stay where it is, independent 
of whether ET = RT or ET < RT. 
2. With/3conseq, RTb'S link to the consequent phase of 
the TF locates event E b there, shifting the TF 
forward (to Eb). This is the "natural" forward 
progression of narrative. 
3. With/3prep, RTb'S link to the preparatory phase of 
the TF locates E b there, shifting the TF backward 
to (Eb). This is used to elaborate an event or 
situation in more detail. 
These relationships, which I will call maintenance and 
local movement of the TF, correspond to Sidner's DF 
moving gradually among the discourse entities in a 
discourse segment. (They cover the same phenomena as 
the micromoves that Nakhimovsky describes in his 
paper (this volume).) More radical movement of TF 
correspond to changes in discourse structure. (These 
Computational Linguistics, Volume 14, Number 2, June 1988 67 
Bonnie Lynn Webber Tense as Discourse Anaphor 
cover similar phenomena to the macromoves described 
in Nakhimovsky, also this volume.) In cases involving 
movements into and out of an embedded discourse 
segment, either 1. the TF will shift to a different entity 
in E/S structure---either an existing entity or one cre- 
ated in recognition of an embedded narrative; or 2. it 
will return to the entity previously labeled TF, after 
completing an embedded narrative. Such movements 
are described in Section 3.3.2. Other movements, sig- 
naled by temporal adverbials and when clauses, are not 
discussed in this paper. 14 
3.3.1 TEMPORAL FOCUS: MAINTENANCE AND LOCAL 
MOVEMENT 
The following pair of examples illustrate maintenance 
and local movement of TF within a discourse segment 
and its link with E/S structure construction. The first I 
discussed in the previous section to illustrate /3conseq" 
The second is a variation on that example: 
13. a. John went into the florist shop. 
b. He picked out three red roses, two white ones 
and one pale pink. 
15. a. John went into the florist shop. 
b. He had promised Mary some flowers. 
c. He picked out three red roses, two white ones, 
and one pale pink. 
First consider Example 13. The first clause (13a) evokes 
an event entity E a describable as John's going into the 
florist shop. Since its tense is simple past, E a is inter- 
preted as prior to ST. Since it begins the discourse, its 
status is special vis-~t-vis both definite NPs and tensed 
clauses. That is, since no previous TF will have been 
established yet, the listener takes that entity E a to serve 
as TF. 15 This is shown in Figure 8: 
Partee, and Dowty were out to achieve. Here it falls out 
simply :from the discourse notion of a TF and from the 
particular anaphoric function/3¢o,scq.16 
Now consider Example 15 (repeated here) whose 
first clause is the same as Example 13a and hence would 
be processed in the same way. 
E E b a -( ) ( ) 
I conseq( E a ) 
Q Now 
Figure 9. E/S structure after processing Clause 13b. 
15. a. John went into the florist shop. 
b. He had promised Mary some flowers. 
c. He picked out three red roses, two white ones, 
and one pale pink. 
The tense of the next clause (15b) is past perfect. As I 
noted above, the only anaphoric function on RT15b and 
an event entity that makes sense for perfect tenses is 
/3o--that is, 
/30(Clsb, TF, RT15b) = El5 b 
Given that perfect tenses imply ET < RT, the event E b 
specified by (15b) will be interpreted as being prior to 
E a. Moreover, since (15b) is past perfect, the conse- 
quent phase of E b is assumed to still hold with respect to 
RT15 b. Hence the consequent phase of E b overlaps E a. 
Finally since TF is associated with the event entity at 
RT b, it remains at E a. E/S structure at this point 
resembles Figure 10: 
( a ) ® 
Now 
Figure 8. E/S structure after processing Clause 13a. 
If Clause 13b is interpreted as being part of the same 
discourse segment as (13a) it must be the case that 
fl(C13b,TF,RT136). Assume the listener takes /3 to be 
/3cons~,q on the basis of world knowledge--that is, 
/3conscq(Clab,TF,RT13b). Since the tense of (13b) is sim- 
ple past, its RT and ET coincide. Thus (13b) specifies a 
new entity Eb, located within the consequent phase of 
the TF--that is, Ea--and hence after it. I assume that, 
following the computation of the anaphoric function, TF 
becomes associated with the event entity located at 
RT b. In this case, it is E b, and TF thereby moves 
forward (cf. Figure 9). As noted, this is the gradual 
forward movement of simple narratives that Hinrichs, 
68 
E b E a 
( ) ( ) 
Q Now 
Figure 10. E/S structure after processing Clause 15b. 
Now Clause 15c is the same as (13b), and TF is the same 
as it was at the point of interpreting (13b). Thus not 
surprisingly, 15c produces the same change in E/S 
E b E E a c ( ) ( )( ) 
I conseq( E a ) ® v 
Now 
Figure 11. E/S structure after processing Clause 15c. 
Computational Linguistics, Volume 14, Number 2, June 1988 
Bonnie Lynn Webber Tense as Discourse Anaphor 
structure and in the TF as (13b), resulting in the diagram 
shown in Figure 11. 
3.3.2 TEMPORAL FOCUS: DISCOURSE-STRUCTURE 
MOVEMENTS 
To illustrate the effect of discourse structure on TF, 
consider the following variation on Example 15, which 
had the same structure vis-~t-vis sequence of tenses. 
16. a. John went into the florist shop. 
b. He had promised Mary some flowers. 
c. She said she wouldn't forgive him if he forgot. 
The first two clauses (a) and (b) are the same as in 
Example 15 and lead to the same configuration of event 
entities in E/S structure (as shown in Figure 10). But the 
most plausible interpretation of (16c) is where the 
"saying" event is interpreted anaphorically with respect 
to the "promising" event--that is, where (16b-c) are 
taken together as (the start of) an embedded discourse, 
describing an event prior to John's going to the florist's. 
To handle this, I assume, following Grosz and Sidner 
1986, that when the listener recognizes an embedded 
discourse segment, s/he stores the current TF for pos- 
sible resumption later. '7 However, I also assume the 
listener recognizes the embedding not when s/he first 
encounters a perfect-tensed clause Cb, since it needn't 
signal an embedded discourse, but later, when an 
immediately following simple past tense clause Cc is 
most sensibly interpreted with respect to the event 
entity E b that Cb evoked. ~8 At this point, the listener 
moves TF from its current position to E b, caching the 
previous value for possible resumption later. Following 
this gross movement, /3(Cc,TF,RT~) will be computed. 
If/3 is then interpreted as /3conseq or/3prep, there will be 
a second movement of TF.'9 
Coming back to Example 16, if Clause 16c is taken as 
being part of a single discourse segment with (16a-b), 
she saying something would have to be interpreted with 
respect to the current TF (Ea)---John's going to the 
florist. This is implausible under all possible interpreta- 
tions of/3.2o However, under the assumption that E b is 
part of an embedded narrative, the listener can a 
posteriori shift TF to E b and consider the anaphoric 
relation 
/3(C16c, TF, RT16c) = El6 c 
with Eb as TF. At this point, the listener can plausibly 
take/3 to be/3co,seq based on world knowledge. Since 
(16c) is simple past, ETc = RT~, the "saying" event E~ 
is viewed as part of the consequent phase (and hence 
following) the "promising" event E b. As in the first 
case, TF moves to the event located at RT~--i.e., to Ec. 
This is shown roughly in Figure 12. Notice that this 
involved two movements of TF---once in response to a 
perceived embedded segment and a second time, in 
response to interpreting/3 as/3¢o,seq- 
Now consider the following extension to (16): 
17. a. John went into the florist shop. 
b. He had promised Mary some flowers. 
c. She said she wouldn't forgive him if he forgot. 
d. So he picked out three red roses, two white 
ones, and one pale pink. 
As before, Clauses 17b-c form an embedded narrative, 
but here the main narrative of John's visit to the florist 
shop, started at (17a), is continued at (17d). To handle 
this, I again assume that TF behaves much like Sidner's 
DF in response to the listener's recognition of the end of 
an embedded narrative: that is, the cached TF is re- 
sumed and processing continues. 2~ 
F-\] 
Cache 
E b E c E a 
( )( ) ( ) 
I~nseq( E b ) ® 
Now 
Figure 12. E/S structure after processing Clause 16c. 
Under this assumption, Clauses 17a-c are interpreted 
as in the previous example (cf. Figure 12). Recognizing 
Clause 17d as resuming the embedding segment, 2z the 
previously cached TF (Ea---the going into the florist 
shop event) is resumed. Again assume that the listener 
takes the anaphoric function to be flconseq(Cd,TF,RTd) 
= E o on the basis of world knowledge. Since Clause 17d 
is simple past (ET = RT), the picking out roses event E d 
is viewed as part of the consequent phase and hence 
following the going into the florist shop event. This is 
shown roughly in Figure 13: 
Cache 
Eb Ec Ea Ed I 
)( ) ( )( ) 
Jconseq ( Jconseq( E b ) E c ) 
Q Now 
Figure 13. E/S structure after processing Clause 17d. 
Now getting the listener to interpret a text as an 
embedded narrative requires providing him/her with 
another event or situation that TF can move to. One 
way in English is via a perfect-tensed clause, which 
Computational Linguistics, Volume 14, Number 2, June 1988 69 
Bonnie Lynn Webber Tense as Discourse Anaphor 
explicitly evokes another event, temporally earlier than 
the one currently in focus. Another way is by lexical 
indications of an embedded narrative, such as verbs of 
telling and NPs that themselves denote events or situa- 
tions (e.g., ones headed by de-verbal nouns). 
This is illustrated in Example 18. Even though all its 
clauses are simple past (ET = RT), Clauses 18c-d are 
most plausibly interpreted as indirect speech describing 
an event that has occurred prior to the "telling" event. 
I assume that in response to recognizing this kind of 
embedded narrative, the listener creates a new node of 
E/S structure and shifts TF there, caching the previous 
value of TF for possible resumption later. The temporal 
location of this new node vis-a-vis the previous TF will 
depend on information in the tensed clause and on the 
listener's world knowledge. 
18. a. I was at Mary's house yesterday. 
b. We talked about her sister Jane. 
c. She spent five weeks in Alaska with two 
friends. 
d. Together, they climbed Mt. McKinley. 
e. Mary asked whether I would want to go to 
Alaska some time. 
Notice that, as with embedded narratives cued by the 
use of a perfect tense, caching the previous TF for 
resumption later enables the correct interpretation of 
Clause 18e, which is most plausibly interpreted as 
following the telling about her sister event. 
An NP denoting an event or situation (such as one 
headed by a noun like trip or by a de-verbal noun like 
installation) can also signal the upcoming possibility of 
an embedded narrative that will elaborate that event or 
situation (past, upcoming, or hypothetical) in more 
detail, as in Example 19. In this case, the original NP 
and the subsequent clause(s) will be taken as co- 
specifying the same thing. The question here is how and 
when TF moves. 
19. a. I was talking with Mary yesterday. 
b. She told me about her trip to Alaska. 
c. She spent five weeks above the Arctic Circle 
with two friends. 
d. The three of them climbed Mt. McKinley. 
After interpreting Clause 19b, the TF is at the "telling" 
event. I claim that the NP her trip to Alaska, while 
evoking a discourse entity, does not affect the TF. If 
Clause 19c is interpreted as the start of an embedded 
narrative (as it is here), TF moves to the event entity E c 
it evokes (caching the previous value Eb). At this point, 
using additional reasoning, the listener may recognize 
an anaphoric relation between Clause 19c and the 
discourse entity evoked by her trip to Alaska. Support 
for this, rather than assuming that an event-denoting NP 
sets up a potential focus, just as I claim a perfect-tensed 
clause does, comes from the reasoning required to 
understand the following parallel example, where I 
would claim TF does not move. 
20. a. I was talking with Mary yesterday. 
b. She told me about her trip to Alaska. 
c:. She had spent five weeks above the Arctic 
Circle with two friends. 
d. The three of them had climbed Mt. McKinley. 
e. She said that next year they would go for 
Aconcagua. 
The event described in Clause 20c is the same as that 
described in Clause 19c, and should be interpreted 
anaphorically with respect to the entity her trip to 
Alaska in the same way. If this is the case, however, 
then the anaphoric link does not follow from the move- 
ment of TF. 
3.3.3 TEMPORAL FOCUS: MISCELLANY 
Example 20 above illustrates one case of an anaphoric 
function on an NP and a tensed clause, specifically 
fl(Cb,Ea,RTb) where the entity Ea has been evoked by 
an NP rather than a clause. Another possibility is that 
a(NPb,Ea) = Eb, where NP b is definite by virtue of an 
entity evoked by a clause rather than an NP that is, 
Eb, is associated with either the preparatory/culmina- 
tion/consequent structure of Ea, as in 
21. a. Mary climbed Mt. McKinley. 
b. The preparations took her longer than the 
ascent. 
or its associated role structure, as in 
22. a. John bought a television. 
b. Although he had intended to buy a 13" b/w 
set, the salesman convinced him to buy a 25" 
color, back-projection job. 
where the salesman fills a particular role in the buying 
event. 
Next, notice that ambiguities arise when there is 
more than one way to plausibly segment the discourse, 
as in the following example: 
23. a. I told Frank about my meeting with Ira. 
b. We talked about ordering a Butterfly. 
Here it is plausible to take Clause 23b as the beginning 
of an embedded narrative, whereby the "talking about" 
event is interpreted against a new node of E/S structure, 
situated prior to the "telling Frank" event. (In this case, 
we is Ira and me.) It is also plausible to take (23b) as 
continuing the current narrative, whereby the "talking 
about" event is interpreted with respect to the "telling 
Frank" event. (In contrast here, we is Frank and me.) 
Finally, consider things from the point of view of 
generation. If some event E b is part of the preparatory 
phase of some event Ea, and a description of Ea has just 
been generated using the simple past tense, then E b 
could be described using either the simple past, as in 
Example 24 or past perfect, as in Example 25. 
70 Computational Linguistics, Volume 14, Number 2, June 1988 
Bonnie Lynn Webber Tense as Discourse Anaphor 
24. a. John went to the hospital. 
b. He took a taxi, because his car was in the shop 
25. a. John went to the hospital. 
b. He had taken a taxi, because his car was in the 
shop. 
In the case of Example 24, the listener/reader recog- 
nizes that Eb is part of the preparatory phase of E a and 
that Eb therefore precedes E a. In the case of Example 
25, the listener would first recognize that E b precedes 
Ea because of the past perfect, but then recognize Eb as 
part of the preparatory phase of E a. 
On the other hand, if E b simply precedes E a, but a 
description of E a has been generated first, then Eb must 
be described with a past perfect (Example 26): simple 
past would not be sufficient (Example 27). 
26. a. John went to the hospital. 
b. He had broken his ankle, walking on a patch 
of ice. 
27. a. John went to the hospital. 
b. *He broke his ankle, walking on a patch of 
ice. 
4 CONCLUSION 
In this paper, I have presented a uniform characteriza- 
tion of discourse anaphora in a way that includes 
definite pronouns, definite NPs, and tensed clauses. In 
doing so, I have argued that the successful use of 
discourse anaphors depends on two different things: 1. 
speakers' and listeners' (mutual) beliefs about the on- 
tology of the things and events being discussed, and 2. 
speakers' and listeners' (mutual) focus of attention. The 
former implicates semantics in the explanation of dis- 
course anaphora, the latter, discourse itself. It is impor- 
tant that we as researchers recognize these as two 
separate systems, as the properties of discourse as an 
explanatory device are very different from those of 
semantics. 
ACKNOWLEDGMENTS 
This work was partially supported by ARO grant 
DAA29-884-9-0027, NSF grant MCS-8219116-CER, and 
DARPA grant N00014-85-K-0018 to the University of 
Pennsylvania, by DARPA grant N00014-85-C-0012 to 
UNISYS Paoli Research Center, and an Alvey grant to 
the Centre for Speech Technology Research, University 
of Edinburgh. 
My thanks to Becky Passonneau, Debby Dahl, Mark 
Steedman, Ethel Schuster, Candy Sidner, Barbara 
Grosz, Ellen Bard, Anne Anderson, Tony Sanford, 
Simon Garrod, and Rich Thomason for their helpful 
comments on the many earlier versions of this paper. 
NOTES 
1. Computationally, these properties imply particular things about 
processing. For example, in interpreting an NP, one may always 
have to consider the possibility--probably in parallel with other 
possibilities--that it is a discourse anaphor. For this alternative, 
the processor will need to (a) keep track of entities whose 
specification the NP may depend on and (b) make appropriate 
inferences with respect to these entities. Other forms of context 
dependencies will have other implications for processing. In this 
paper, I will touch on processing issues for discourse anaphora, 
but not other context-dependent phenomena. 
2. In earlier work, it was only the former that I discussed under the 
concept of Discourse Model. The notion is not all that different 
from Kamp's "Discourse Representation Structures" (Kamp 
1984) or Helm's "File Cards" (Heim 1982). 
3. This does not mean that all NPs evoke entities in the listener's 
Discourse Model: for example, I would not want to say that a 
predicate nominal (as in "John is a neuro-surgeon.") evokes a 
separate entity. On the other hand, I do assume that NPs in a 
quantified context evoke discourse entities that are accessible as 
tong as that context holds. For example, "Whenever a philoso- 
pher debates a computer scientist, the philosopher turns out to 
have a naive idea of what the computer scientist is doing. Of 
course, the computer scientist always does her best to explain." 
(I thank Rich Thomason for the example.) Keith Stenning has 
also discussed examples such as this one. 
4. In this use of "specify", I am following Sidner, who used it to 
mean, essentially, refer in a model, as opposed to refer in the 
outside world (cf. Sidner 1983). 
5. Note that a discourse entity may or may not stand for some 
real-world individual or set. Whether it does or doesn't, how- 
ever, is irrelevant to these model-based notions of evoke, 
specify, and co-specify. 
6. As with any aspect of discourse that relies on the speaker making 
assumptions about the listener's knowledge, those assumptions 
may in some cases be wrong. Listeners, for their part, try to do 
their best in making sense anyhow--for example, hoping that the 
further discourse will provide enough information to work things 
out. 
7. The inferability of a in other cases follows from the ontology of 
entities specified clausally. This I will touch upon in Section 3. 
Often a speaker can assume that ct relates an individual to the 
generic class it belongs to, as in Example 4, or the intension of 
a definite description to its extension--or vice versa--as in "The 
President is elected by the Electoral College. In the 1964 
election, he got all but two of the votes." I am not claiming that 
all inferrable a's follow from ontology---only that ontology 
sanctifies a great many. 
8. Nor does it indicate wholesale acceptance of their theory of 
discc,urse as has so far emerged. I believe that they would be the 
first to admit that it's a "theory in progress". 
9. Grosz, Joshi, and Weinstein (1983) have reinterpreted part of 
Sidner's work in terms of their theory of centering. Her DF 
becomes their "backward-looking center" Cb and her PFL 
becomes their ordered list of "forward-looking centers" Cf. 
They have proposed heuristics for which entities (Cb and/or Cfs) 
will be specified pronominally (when specified at all) and for the 
consequences for such specification on subsequent C b and Cfs. 
10. This jump is not necessary: the DF can stay the same over 
discourse segments for example, discussing the same entity 
from different points of view. 
l 1. Dowty 1986 gives an account similar to Hinrichs's in its a priori 
assumption of the non-backward-movement of time in simple 
narratives and its focus on how different aksionart lead to 
interpretation of co-temporal versus subsequent events. The two 
accounts differ in how the latter is achieved. Dowty's account 
avoids some problems that Hinrichs's cannot, but still cannot 
deal with the fact that time sometimes "moves backwards" even 
in sequences of simple past tense. 
12. The ontology of an event/situation also includes its role struc- 
ture. This latter appears to play a part in possible anaphoric 
relations between an NP and an event entity, but not between a 
tensed clause and an event entity. 
13. I have not considered other aspectual types in English (such as 
progressive) vis-a-vis possible anaphoric relations. Integrating 
the current account with work on aspect and Aktionsarten, \[cf. 
Moens & Steedman, Passonneau, Hinrichs, Nakhimovsky--this 
volume\] is left to future work. 
14. I should also note that Rohrer 1985 suggests that there may exist 
a set of possible temporal referents, possibly ordered by sa- 
liency, among which the tense in a sentence may find its 
reference time, but doesn't elaborate how. That is the only thing 
I have seen that comes close to the current proposal. 
15. This is similar to the listener's response to the definite NP the 
florist shop, which in the middle of a discourse would have to be 
taken anaphorically. At the beginning of a discourse, the listener 
will just create a new discourse entity. 
16. In parallel with this, given /3prep, TF will move incrementally 
back into the preparatory phase of the event that was the 
previous TF. Given/3 o, RTb is at the TF and TF does not move. 
17. While Sidner and thereafter Grosz & Sidner assume a stack of 
focus spaces with their associated DFs, each of which is 
resumable, I do not have evidence that the listener can keep 
track of more than one other node of E/S structure in addition to 
the TF. Resuming any other node seems to require more effort-- 
such as using a when clause. Hence I assume a single item cache 
for the previous value of TF. 
18. One could say, in parallel with Sidner 1983, that this E b was a 
potential focus (PF). However, I do not postulate a Potential 
Focus List as in Sidner, because I do not think there is ever more 
than one PF that one can shift to without using a temporal 
adverbial. 
19. This is clearly an inertial strategy', it assumes that the listener 
will prefer to interpret clauses as being in the same discourse 
segment, unless forced to do otherwise, say by a continuation 
such as (16c) or a lexical item that signals a possible embedded 
narrative. This I will discuss later on. In spoken text, the speaker 
might give an intonation to (16b) that conveys that it and the 
following clauses should be understood as an embedded dis- 
course segment. This would be what breaks the inertia and 
causes the listener to shift the TF. 
20. ~It is rarely the case that one cannot come up with a story linking 
two events and/or situations. Thus it would be impossible to 
reject a hypothesis on grounds of inconsistency. All one can say 
is that one of such stories might~be more plausible than the 
72 Computational Linguistics, Volume 14, Number 2, June 1988 
Bonnie Lynn Webber Tense as Discourse Anaphor 
others by requiring, in some sense not explored here, fewer 
inferences. Crain and Steedman 1985 make a similar argument 
about prepositional phrase (PP) attachment. For example, it is 
not impossible for a cat to own a telescope---e.g., by inheritance 
from its former owner. Thus a cat with a telescope is not an 
inconsistent description. However, it must compete with other 
plausible interpretations like seeing with a telescope in I saw a 
cat with a telescope. Thus I assume that interpretations are 
computed in parallel, with the most plausible prediction being 
the one that ends up updating both E/S structure and the TF. 
21. Recall my prediction that if any but the most recently cached TF 
is to be resumed, a cooperative speaker will explicitly indicate 
this with a temporal adverbial or a when clause. 
22. So is one cue. In spoken discourse, intonation would be another. 
Computational Linguistics, Volume 14, Number 2, June 1988 73 

REFERENCES 
Bauerle, R. 1979 Temporal Deixis, temporale Frage. Gunter Narr 
Verlag, Tubigen. 
Bobrow, D.; and the PARC Understander Group. 1986 GUS, A 
Frame-Driven Dialog System. Artificial Intelligence 8: 155-173. 
Chomsky, N. 1980 Rules and Representations. Columbia University 
Press, New York, NY. 
Clark, H.; and Marshall, C. 1981 Definite Reference and Mutual 
Knowledge. In Joshi, A., Webber, B., & Sag, I. (Eds.). Elements 
of Discourse Understanding. Cambridge University Press, Cam- 
bridge, England. 
Comrie, B. 1985 Tense. Cambridge University Press, Cambridge. 
Crain, S.; and Steedman, M. 1985 On not being Led up the Garden 
Path: the Use of Context by the Psychological Syntax Processor. 
In Natural Language Parsing. Dowty, D., Karttunen, L., and 
Zwicky, A., (Eds.), Cambridge University Press, Cambridge; 
320-358. 
Dowty, D. 1986 The Effects of Aspectual Class on the Temporal 
Structure of Discourse: Semantics or Pragmatics. Linguistics and 
Philosophy 9(1): 37-62. 
Grosz, B. 1977 The Representation and Use of Focus in Dialogue 
Understanding. Ph.D. diss. University of California, Berkeley, 
CA. 
Grosz, B.; and Sidner, C. 1986 Attention, Intention, and the Structure 
of Discourse. Computational Linguistics 12(3): 175-604. 
Grosz, B.; Joshi, A.; and Weinstein, S. 1983 Providing a Unified 
Account of Definite Noun Phrases in Discourse. Proceedings of 
the 22nd Annual Meeting of the Association for Computational 
Linguistics. MIT, Cambridge, MA; 44-50. 
Heim, I. 1982 The Semantics of Definite and Indefinite Noun Phrases. 
Ph.D. diss. Department of Linguistics, University of Massachu- 
setts, Amherst, MA. 
Hinrichs, E. 1986 Temporal Anaphora in Discourses of English. 
Linguistics and Philosophy 9(1): 63-82. 
Hirst, G. 1981 Anaphora in Natural Language Understanding: A 
Survey. Springer-Verlag, Berlin, W. Germany. 
Hobbs, J. 1987 Implicature and Definite Reference. CSLI-87-99. 
Center for the Study of Language and Information, Stanford 
University, Stanford, CA. 
Kamp, H. 1984 A Theory of Truth and Semantic Representation. In 
Truth, Interpretation, and Information. Groenendijk, J., Janssen, 
T.M.V., and Stokhof, M. (Eds.). Foris, Dordrecht, Holland. 
Leech, G. 1987 Meaning and the English Verb (2nd edition). Long- 
man, London, England. 
McCawley, J. 1971 Tense and Time Reference in English. In Studies 
in Linguistic Semantics. Fillmore, C. and Langendoen, D.T. 
(Eds.). Holt, Rinehart and Winston, New York, NY; 97-114. 
Moens, M.; and Steedman, M. 1987 Temporal Ontology in Natural 
Language. Proceedings of the 25th Annual Meeting of the Asso- 
ciation for Computational Linguistics. Stanford University, Stan- 
ford, CA; 1-7. 
Partee, B. 1973 Some Structural Analogies between Tenses and 
Pronouns in English. Journal of Philosophy 70: 601-609. 
Partee, B. 1984 Nominal and Temporal Anaphora. Linguistics and 
Philosophy 7(3): 287-324. 
Passonneau, R. 1987 Situations and Intervals. Proceedings of the 25th 
Annual Meeting of the Association for Computational Linguistics. 
Stanford University, Stanford, CA; 16-24. 
Reichenbach, H. 1966 The Elements of Symbolic Logic. The Free 
Press, New York, NY. 
Rohrer, C. 1985 Indirect Discourse and "Consecutio Temporum". In 
Temporal Structure in Sentence and Discourse. LoCascio, V., and 
Vet, C. (Eds.). Foris Publications, Dordrecht, Holland; 79--98. 
Sag, I.; and Hankamer, J. 1984 Toward a Theory of Anaphoric 
Processing. Linguistics and Philosophy 7(3): 325-345. 
Sidner, C. 1979 Towards a Computational Theory of Definite Ana- 
phora Comprehension in English Discourse. Ph.D. diss. Dept. of 
Computational L~ngulsfics, Volume 14, Number 2, June 1988 71 
Bonnie Lynn Webber Tense as Discourse Anaphor 
Electrical Engineering and Computer Science, MIT, Cambridge, 
MA. 
Sidner, C. 1983 Focusing in the Comprehension of Definite Anaphora. 
In Computational Models of Discourse. Brady, M., and Berwick, 
R. (Eds.). MIT Press, Cambridge, MA; 267-330. 
Steedman, M. 1982 Reference to Past Time. In Speech, Place, and 
Action. Jarvella, R. and Klein, W. (Eds.). Wiley, New York, NY; 
125-157. 
Webber, B. 1983 So can we Talk about Now? In Computational 
Models of Discourse. Brady, M., and Berwick, R. (Eds.). MIT 
Press, Cambridge, MA; 331-371. 
Webber, B. 1987 The Interpretation of Tense in Discourse. Proceed- 
ings of the 25th Annual Meeting of the Association for Computa- 
tional Linguistics. Stanford University, Stanford, CA. 
Woods, W.A. 1978 Semantics and Quantification in Natural-Lan- 
guage Question Answering. In Advances in Computing 17: 2-64. 
M. Yovits (Ed.). Academic Press, New York, NY. 
