SITUATIONS AND INTERVALS 1 
Rebecca J. Passonneau 
Paoli Research Center, UNISYS Defense Systems ~ 
P.O.Box 517 
Paoli, PA 19301 USA 
ABSTRACT 
The PUNDIT system processes natural 
language descriptions of situations and the inter- 
vais over which they hold using an algorithm that 
integrates ~peet and ~en~e \[ogle. It analyzes the 
tense and aspect of the real- verb to generate 
representations of three types of situations-- 
states, processes and events-- and to locate the 
situations with respect to the time at which the 
text was produced. Each situation type has a dis- 
tinct temporal structure, represented in terms of 
one or more intervals. Further, every interval has 
two features whose different values capture the 
aspectual differences between the three different 
situation types. Capturing these differences makes 
it possible to represent very precisely the times for 
which predications are asserted to hold. 
1. Introduetlon 
This paper describes a semantics of situa- 
tions and the intervals over which they hold 
that is neither situation semantics (Barwise and 
Perry, 1983) nor interval semantics (Dowty, 1979, 
1982, 1986; Taylor, 1977). It is unfortunately 
d~f~cult to avoid the overlap in terminology 
because what will be described here shares certain 
goals and assumptions with each. The concerns 
addressed here, however, arise from the computa- 
tional task of processing references to situations in 
natural language text in order to represent what 
predicates are asserted to hold over what entities 
and when. 
Situation as used here pertains to the 
linguistic means for referring to things in the 
world, i.e., to sentences or predications. More 
specifically, situation is the superordinate 
category in Mourelatos' typology of aspectual 
classes of predications, schematised in Fig. 1. 
SITUATIONS I 
I I 
STATES OCCURRENCES 
Preu.re. low. I 
I I 
PROCESSES EVENTS Alarm iJ Jounding. Engine/ailed. 
Fig. 1: Mourelatos' typology of situations 
The PUNDIT text-processing system s processes 
references to situations using an algorithm that 
integrates tense logic (Reichenbach, 1947) with 
aspect, or what Talmy (1985) calls the~pa~e~w-~ ~--~ 
d~tribution o~ ac~io, throug~ time. This paper 
describes how PUNDIT represents the temporal 
structure of three types of situations, namely 
states, processes and events, and how these 
situations are located in time. 
2. Problems in Computing Appropriate 
Representations 
The critical problems in the semantic 
analysis of references to situations and their 
associated times are: I) language encodes 
several different kinds of temporal information, 2) 
this information is distributed in many distinct 
linguistic elements, and finany, 3) the semantic 
contribution of many of these elements is 
context-dependent and cannot be computed 
without looking at co-occurring elements. 
ZThis work was supported by DARPA under contract 
N00014-85-C-0012, administered by the Office of Naval 
Research. APPROVED FOR PUBLIC RELEASE, DISTR/BU- 
TION UNLIMITED. 
ZFormerly SDC--A Burroughs Company. 
zPUNDIT is an acronym for Prolog UNderstands and In- 
teErates Text. It is a modular system, implemented in 
Quintus Prolog, with distinct syntactic, semantic and prag- 
matic components (cf. Dahl et al., 1987). 
16 
These problems have been addressed as fol- 
lows. A decision was made to focus on the kinds 
of temporal information embodied in the verb and 
its categories of tense, taxis 4 and grammatical 
aspect, s and to temporarily ignore other kinds of 
temporal information, s Computation of this infor- 
mation was then divided into two relatively 
independent tasks, with appropriate information 
passed between the modules performlng these 
tasks in order to accommodate context- 
dependencies. The first task, carried out by 
Module 1, makes use of the aspectual information 
in the verb phrase (lexical and grammatical 
aspect) to determine the situation type being 
referred to and its temporal structure. An abstract 
component of temporal structure, referred to as 
the event time (following Reichenbach, 1947), 
serves as input to Module 2, where the deictic 
information in the verb phrase (tense and taxis) is 
used to compute temporal ordering relations, i.e., 
where the situation is located with respect to the 
time of text production. Section §3 outlines the 
general goals for computing temporal structure 
and §4 describes in detail how it is computed. 
Then §5 briefly illustrates how the event time 
which Module 1 passes to Module 2 simplifies the 
interaction of tense and aspect. 
8. Goals i~or Representing Situations 
The goal in generating representations of the 
temporal structure of situations was to closely 
link the times at which situations are said to hold 
with the iexlcal decompositions of the predicates 
used in referring to them. The decompositions 
encode aspectual information about the situation 
types which is used in determlning what type of 
situation has been referred to and what its tem- 
poral structure is. Distinct components of the 
semantic decompositions correspond to different 
features of the intervals with which they are asso- 
ciated. As §4 will demonstrate, the interpretation 
of these components of temporal meaning depends 
on the interaction between iexical and grammati- 
cal aspect. 
4T4~'i# (Jakobson. 1957) refers to the semantic effect of 
the presence or absence of the perfect auxiliary. 
SAzpect is both part of the inherent meaning of a verb 
(lexical aspect) and also signalled by the presence or absence 
of the progressive suffix -lag (grammatical aspect). 
~E.g., rate (~ven by adverbs like rapidJy), "patterns of 
frequency or habituation", and so on (of. Mourelatos. 1981). 
This approach to the compositional seman- 
tics of temporal reference is similar in spirit to 
interval semantics. Interval semantics captures 
the distinct temporal properties of situations by 
specifying a truth conditional relation between 
a full sentence and a unique interval (Dowty, 
1979, 1986). This is motivated by the observa- 
tion that the aspectual type of a sentence 
depends simultaneously on the aspectual class of 
a particular lexical item, its tense, taxis and 
grammatical aspect, and the nature of its argu- 
ments (cf. Mourelatos, 1981; note that the 
latter factor is not handled here). 
The goal of PUNDIT's temporal analysis is 
not simply to sort references to situations into 
states, processes and events, but more specifically 
to represent the differences between the three 
types of situations b~ considering in detail the 
e\]~aracteris~ice o/ ~he set o/ temporal ~nter~z~# t\]~Qt 
~,sl/ hold or occur oesr (Allen, 1983, p. 132). Thus, 
instead of specifying truth conditional properties 
of sentences, the temporal semantics outlined here 
specifies what property of an interval is entailed 
by what portion of the input sentence, and then 
compositionally constructs a detailed representa- 
tion of a state, proeess or event from the inter- 
vais and their associated properties. 
8.1. Intervals and Their Features 
Each situation type has a distinct temporal 
structure comprised of the interval or intervals 
over which it holds. Two features are associated 
with each interval, klnesle and boundedness. 
Very briefly, kinesls pertains to the internal 
structure of an interval, or in informal terms, 
whether something is happening within the inter- 
val. Boundedness pertains to the way in which 
an interval is located in time with respect to other 
times, e.g., whether it is bounded by another inter- 
val. 
4. Part One of the Algorithms Computing 
Temporal Structure 
The input used to compute the temporal 
structure of a situation consists of the grammati- 
cal aspect of the verb, that is, whether it is pro- 
gressive, and the decomposition produced by 
PUNDIT's semantic interpreter (Palmer et al., 
1986). The lexical decompositions employed by 
PUNDIT (Passonneau, 1986\])) not only represent 
the predlcate/argument structure of verbs, but in 
addition, following the example of Dowty's aspect 
17 
calculus (1979), they represent a verb's inherent 
temporal properties, or lexlcal aspect. 7 In 
PUNDIT's lexical entries, there are three values of 
lexlcal aspect corresponding to the three types of 
situations. Four of the six possible combinations 
of grammatical and lexical aspect are temporally 
distinct. This section will go through the four 
cases one by one. 
4.1. States 
The following conditional statement sum- 
marises the first of four cases of temporal struc- 
ture. The antecedent specifies the necessary input 
condition, the first clause of the consequent 
specifies the situation type, the second specifies 
the k|nesls Of its associated interval and the third 
specifies its boundedness. 
IF Lexical Aspect=stative 
THEN Situation is a state 
AND its Time Argument is a period 
AND this period is unbounded 
As shown here, if the lexical aspect of a predica- 
tion is stative, its grammatical aspect is 
irrelevant. The justification for ignoring grammat- 
ical aspect in the context of lexical stativity 
appears at the end of this section. 
A state is defined as a situation which holds 
for some interval that is both statle and 
unbounded. Example 1) illustrates a typical 
reference to a state situation along with its 
semantic decomposition. Note that the lexlcal 
head of the verb phrase is the adjective low. 
1) The pressure was low. 
low (patlent(\[pressurel\]) s 
As in Dowty's aspect calculus (1979), the decompo- 
sitions of stative predicates consist of semantic 
predicates with no aspectual operators or connec- 
tives. Computing the temporal structure associ- 
ated with 1) means finding a single interval with 
the appropriate features of kinesis and bounded- 
ness to associate with the stative predicate 
low(patlent(X)). 
rThe literature on upectual classes of verbs provides a 
variety of diagnostics for determining the inherent upect of 
verbs (cf. Vendler, 1967; Dowty, 1979). 
*PUNDIT's current application is to process short 
messages texts called CASREPS (CASualty REPorts) which 
describe Navy equipment failures. The arguments in the 
decompositions, e.g., \[preuurel\], are unique identifiers of the 
entities denoted by the surface noun phrues. They are crest- 
Kinesls of states. A static interval is tem- 
porally homogeneous. With respect to the relevant 
predication, there is no change within the interval; 
consequently, any subinterval is equivalent to any 
other subinterval. Thus, a static interval is 
defined much as stative predications are defined in 
interval semantics: 
An inter~al I associated with some predication ~b 18 
static iff it follows from the truth of ~ at I that ~ is 
true at all aublnter,Jal8 of I (cf. Dowty, 1986, p. 42). 
Situations are represented as predicates identify- 
ing the situation type (e.g., state). The situation 
denoted by 1) would be represented as follows: 
state(\[lowl\], 
low (patient (\[pressureT\]), 
period (\[Iowl\]) 
The three arguments are: the unique identifier of 
the situation (e.g., \[lowl\]), the semantic decompo- 
sition, and the time argument (e.g., period 
(\[lowl\])). The same symbol (e.g.,\[lowl\]) identifies 
both the situation and its time argument because 
it is the actual time for which a situation holds 
which uniquely identifies it. 0 A period time argu- 
ment in the context of a state predicate always 
represents a statie interval. 
Boundedness of states. The intervals asso- 
ciated with states are inherently unbounded. A 
temporal bound can be provided by an appropri- 
ate temporal adverbial (e.g., The pressure wag ~or- 
real lwh~ the pump seize~), 10 but here we consider 
only the temporal semantics specified by the verb 
form itself. When an unbounded interval is 
located with respect to a particular point in time, 
it is assumed to extend indefinitely in both direc- 
tions around that time. In 1), at least part of the 
interval for which the predication 
low(pat|ent(\[pressurel\])) is asserted to hold is 
located in the past. However, this interval may or 
may not end prior to the present. The unbounded 
property of the interval can be illustrated more 
precisely by examining the relationship between 
the predication and the temporal adverbial 
ed by PUNDIT's reference resolution component (Dahl, 1986). 
SThough a situation is something quite different for Bar- 
wise and Perry (1983), they take n similar view of the role of a 
particular space-time location in tokenising a situation type 
(of. esp. pp. 51ff). 
Xlln general, temporal adverbials can modify an existing 
component of temporal structure or add components of tem- 
poral structure. 
18 
modifying it in example 2): 
2) The pressure was low at 08:00. 
This sentence asserts that the state of 
low(patient(\[pressvwel\])) holds at 08:00 and 
possibly prior and subsequent to 08:00. That is, 
the sentence would be true if the pressure were 
low for only an instant coincident with 08:00, but 
it is not asserted to hold only for that instant. 
This is captured by defining the interval as 
unbounded. 
A situation representation does not itself 
indicate the boundedness of its period time 
argument. Instead, this feature is passed as a 
parameter to the component which interprets 
tense and taxis (cf. §5). 
As will be shown in the following section, the 
progressive assigns the features active and 
unbounded to non-stative verbs. But with sta- 
tire verbs, the progressive contributes no temporal 
information. Inability to occur with the progres- 
sive has in fact been cited as a diagnostic test of 
statives, but as Dowry notes (1979), there is a 
class of progressives which denotes locative states 
(e.g., The soei~ are l~/ing under the bed). Such sta- 
tires occur in PUNDIT's current application 
domain in examples like the following sentence 
fragment: 
3) Material clogging strainer. 
A complete discussion of the interaction between 
progressive grammatical aspect and stative lexical 
aspect would have to address cases in which the 
progressive contributes non-temporal information 
(cf. Smith, 1983). However, these issues are not 
pertinent to the computation of temporal struc- 
ture. 
4.2. Temporally Unbounded Processes 
The second case of temporal structure 
involves progressive uses of non-stative verbs, i.e., 
process or transition event verbs. 
IF Lexical Aspect~stative 
AND Grammatical Aspect--progressive 
THEN Situation is a process 
AND its Time Argument is a period 
AND this period is unbounded 
In this case and the two subsequent ones, both lex- 
ical and grammatical aspect are relevant input. 
Processes are situations which hold over 
active intervals of time. 11 Active intervals can be 
unbounded or zmApecifled for boundedness, 
depending on the grammatical aspect of the predi- 
cation. The two possible temporal structures asso- 
ciated with processes are discussed in this and the 
following section. 
Example 4) illustrates a typical predication 
denoting a temporally unbounded process 
along with its semantic decomposition. 
4) The alarm was sounding. DO(sound(actor(lal m4\]))) 
DO is an aspectual operator identifying a decom- 
position as a process predicate (cf. Dowty, 1979). 12 
As with statives, computing the temporal struc- 
ture for sentences llke 4) involves finding a single 
interval to associate with the semantic decomposi- 
tion. 
Kinesls of processes. The presence of a DO 
operator in a decomposition indicates that the 
interval for which it holds must be active. Active 
and static intervals contrast in that change 
occurs within an active interval with respect to 
the relevant predication. For example, for any 
interval for which DO(sound(actor(\[alarm4\]))) 
is true, the \[alarm4\] participant must undergo 
changes that qualify as sounding, and must con- 
tinue to do so throughout the interval. As Moure~ 
latos (1981) has pointed out, process predicates 
vary regarding how narrowly one can subdivide 
such intervals and still recognize the same process. 
Dowty has used this threshold of granularity as 
the defining characteristic of process sentences, 
and it is borrowed here to define active intervals: 
An interval \[ a~sociated with some predication ~ is 
aetlve iff it follows from the truth of ~b at I that ~ is 
true at all subintervals of I down to a certain limit 
in size. 
As the process representation for 4) illustrates, 
processes and states are represented similarly. 
process(\[soundS\], 
low (patlent(\[alarmT\]), 
period (\[soundS\]) 
lithe distinction between static and active intervals is 
useful for interpreting manner adverbials indicating rate of 
change. Since statics predications denote the absence of 
change over time, they cannot be modified by rate adverbiak. 
lZBecause the aspectual operator DO always hu an ac- 
tor semantic role a~ociated with it, PUNDIT's semantic 
decompositions actually omit DO and use the presence of the 
actor role to identify proce~ predicates. 
19 
The situation predicate identifies the situation 
type as a process. Note that a period time 
argument in the context of a process predicate 
indicates an act|re interval. 
The rule given above specifies that transl- 
tlon event verbs in the progressive also denote 
temporally unbounded processes (cf. 5). 
5) The engineer is installing the oU filter. 
cause(DO (|nstall(agent(\[englneer 8\]))), BECOME(ins lled(theme(\[mter4\]), 
Iocatlon(X)))) 
The cause predicate in the decomposition of 
in~tall indicates that it is a causative verb, and 
the BECOME operator that its lexical aspect is 
transition event. This aspectual class is a hetero- 
geneous one, but in general, transition event 
verbs are temporally more complex than stative or 
process verbs, and have a correspondingly more 
complex relation between their semantic decompo- 
sitions and temporal structure. Consequently, the 
discussion of the treatment of progressive transi- 
tion event verbs is postponed until after the func- 
tion of the aspectual operator BECOME has been 
explained. 
Boundedness. In 6), the interval associated 
with the alarm sounding is unbounded. It bears 
the same relationship to the at adverbial phrase 
modifying the predication as does the statlc inter- 
val in 2) above, repeated here as 7). 
6) The alarm was sounding at 08:00. 
7) The pressure was low at 08:00. 
This siml\]arity between statives and progressives 
has led Vlach (1981) to identify them with each 
other. Here, the commonality among sentences like 
1), 2), 4) and 8) is captured by associating the 
feature value unbounded both with stative lexi- 
cal aspect and with progressive grammatical 
aspect. The differences between the predications 
in 6) and 7), which show up in the types of 
modification and anaphorlc processes to which 
such predications are susceptible, are encapsulated 
in their contrasting values of klnes|s (cf. fn. 11 
above). 
4.8. Temporally Unspecified Processes 
The third case of temporal structure 
accounts for the differences between sentences like 
4), having a process verb in the progressive, and 
8), where the process verb is non-progressive. 
8) The alarm sounded. 
The differences, which will be explained below, are 
captured in the following rule indicating that the 
actlve interval for which the predication is said to 
hold is unspecified for boundedness. 
IF Lexical Aspect=process 
AND Grammatical Aspect=non-progressive 
THEN Situation is a process 
AND its Time Argument is a period 
AND this period is unmpeeifled 
Again, the parameter indicating that the interval 
associated with 8) is unspecified gets passed to 
Module 2 which interprets tense and taxis. In 
every other respect, the analysis of the temporal 
structure associated with 8) resembles that for 4). 
A comparison of progressive and non- 
progressive process verbs in the context of an at 
adverbial phrase illustrates the relative indeter- 
rninacy of the non-progressive use. In the context 
of the progressive process verb in 8), the clock 
time is interpreted as falling within the active 
interval of sounding but in 9), where the verb is 
not progressive, 08:00 can be interpreted as falling 
at the inception of the process or as roughly Iocat- 
ing the entire process. 
9) The alarm sounded at 08:00. 
Non-progresslve process verbs exhibit a wide vari- 
ation with respect to what part of the temporal 
structure is located by tense (Passonneau, 1986a). 
The influencing factors seem to be pragmatic in 
nature, rather than semantic. The solution taken 
here is to characterize the event tlme of such 
predications as having an unnpecifled relation to 
the active interval associated with the denoted 
process. 
4.4. Transition Events 
As mentioned in the previous section, the 
temporal structure of transition events is more 
complex than that of states or processes. 
Correspondingly, the rule which applies to this 
case has more output conditions. 
IF Lexical Aspect=transition event 
A.ND Grammatical Aspect=non-progressive 
THEN Situatlon=event 
AND Time Argument---moment 
AND this moment culminates an interval 
associated with a process 
AND this moment introduces an interval 
associated with a state or process 
20 
A transltlon event is a complex situation consist- 
ing of a process which cu\]mlnates in a transition 
to a new state or proeess. Its temporal structure 
is thus an aetlve interval followed by--and 
bounded by--a new aet|ve or stat|e interval. 
The new state or process comes into being as a 
result of the initial process, is 
As in Dowty (1986), both Vendler's achieve- 
ments and his accomplishments collapse into one 
class, vie., tr~n_Aitlon events. That is, achieve- 
ments 
are those ~'nesi8 predicates which are not only typi- 
eally of shorter duration than accomplishments, but 
aJso thoee which toe do not normally understand o.8 
entailing a sequence o/ sub-events, given o r e, eaJ 
every-dal/ criterla for identifying the even~ 
named by the predicate (Dowty, 1988, p. 43). 
Causative verbs, in which the action of one parti- 
cipant results in a change in another participant, 
are typical accomplishment verbs. 
10) The pump sheared the drive shaft. 
eause(D O(shear (agent (|pumpS|))), 
BEC OME(sheared(patlent(\[shai~e\])))) 
Sentence 10) asserts that a process in which the 
pump participated (shear/ng) caused a change in 
the drive shaft (being sheared). Note that the 
decomposition explicitly indicates a causal relation 
between two conjoined predicates, one represent- 
ing an activity perfomed by an agent, and the 
other representing the resulting situation. 
BECOME serves as the aspectual operator for 
marking transltlon event decompositions. The 
argument to BECOME constitutes the semantic 
decomposition of the new state or process arising 
at the culmination of the event. 
Non-causative verbs can also denote transi- 
tion events. With inchoatives, the same entity 
participates in both the initial process and the 
resulting situation denoted by the predication. 
11) The engine failed. 
D O (fall (agent(\[englnel\]))), 
BE COME (failed(patlent(\[englnel\]))) 
In 11), an engine is said to participate in some 
process ~ailing) which culminates in a new state 
(e.g., being inoperative). The semantic 
ISA state may be a necessary precondition for a certain 
change to occur, but since states are defined by the absence of 
change, or negative klnesla, they are inherently incapable of 
generating new situations. 
decompositions used in PUNDIT do not explicitly 
represent the initial processes involved in transi- 
tion events because they are completely predict- 
able from the presence of the BECOME operator. 
But both conjuncts are shown here to illustrate 
that computing the temporal structure of a transi- 
tion event situation requires finding two intervals, 
one associated with the initial process predicate 
(e.g., DO(fa|l(agent(\[eng|nel\])))) and the other 
with the predicate for the resulting situation (e.g., 
falled(pat|ent(\[eng|nel\]))). 
As indicated in the rule for this case, the 
temporal structure also includes a moment of 
transition between the two intervals, i.e., its tran- 
sltion bound. Since a trans|t|on event is one 
which results in a new situation, there is in theory 
a point in time before which the new situation 
does not exist and subsequent to which it does. A 
transition bound is a theoretical construct not 
intended to correspond to an empirlcal\]y deter- 
mined time. In fact, it should be thought of as the 
same kind of boundary between intervals implied 
by Alien's meets relation (Allen, 1983; 1984, esp. 
p. 128). However, it is a convenient abstraction for 
representing how transition events are perceived 
and talked about. 
We can now return to the question of the 
interpretation of progressive transition event 
verbs. In the context of a decomposition with a 
BECOME operator, the progressive is con- 
strained to apply to the predicate corresponding 
to the initial process, i.e., the predicate denoting 
the portion of a transition event prior to the 
moment of transition. Computing the temporal 
structure for the progressive of install in 12), for 
example, involves generating a single active, 
unbounded interval for which the predication 
D O (agent (|engineer 8\])) holds: 
12) The engineer is installing the oll filter. 
eause(D O (agent(|engineer 8\]), 
BECOME(irmtalled(theme(\[fllter4\]), 
Ioeatlon(X))))) 
In this context, the remainder of the semantic 
decomposition denotes what the person report.ing 
on the event assumes to be the eventual culmina- 
tion of the process referred to as inetalilng. 
K|nesls. Examples 13) and 14) illustrate two 
types of transition events, one resulting in a new 
state, and one resulting in a new process. As illus- 
21 
trated, ~4 transition events are represented as com- 
plex situations in which an event with a moment 
time argument results in a new state or process: 
13) The lube oil pump has seized. 
event(\[seisel\], 
BECOME(selsed(patlent(\[pnmpl\]))), moment(\[selsel\]) 
stat (\[selse \], 
seised(patlent (\[plm~pl\])), 
period(\[seise2 D 
starts(moment (\[selsel\]),perlod(\[seise2\]) 
14) The engine started. 
event(\[sta~tl\]r 
BEC OME (operating(actor (\[engineX\]))), 
moment(\[staxtl\]) 
process(\[startS\], 
opel"atlng(actor (\[eng|nel\])), 
period(\[start2\]) 
starts(moment(\[startl\]),perlod(\[start2\]) 
The starts relation indicates that a transition 
bound (e.g., moment(\[selsel\])) is the onset of the 
interval (e.g., perlod(\[selse2D) associated with 
the situation resulting from a tran~itlon event. 
Boundedness. An important role played by 
the transition bound is that it serves as the tem- 
poral component of transition events for locat- 
ing them with respect to other times. For example, 
the sentence in 15) asserts that the moment of 
transition to the new situation coincides with the 
clock time of 8:00. 
15) The engine failed at 8:00. 
The status of the engine prior to 8:00 is asserted 
to be different from its status afterwards. 
5. Part Two of the Algorithms 
Temporal Ordering Relations 
PUNDIT employs a Reichenbachian analysis 
of tense which temporally locates situations in 
terms of three abstract times: the time of the 
situation (event alms), the time of speech/text 
production (speech time), and the time with 
xtAt present, PUNDIT explicitly represents only two 
components of transition event predications: the moment at- 
sociated with an event of becoming, and a period associated 
with a resulting situation. This representation has been found 
to be adequate for the current application. The omission of 
the first interval is purely a matter of practical convenience, 
but could easily be reprelented should the need arise. 
respect to which relational adverbials are inter- 
preted (reference time). Reichenbach (1947) did 
not distinguish between the temporal structure of 
a situation and its event tlme. In PUNDIT, 
event time is a carefully defined abstract com- 
ponent of temporal structure in terms of which 
ordering relations are specified. It is determined 
on the basis of boundedness, and is always 
represented as a dimensionless moment. 
5.1. Event Time 
The three values of boundedness outlined 
above correspond to three possible relations of 
event time to a time argument. Examples 16) 
through 18) illustrate these relations. If an inter- 
val is unbounded, its event time is represented 
as an arbitrary moment iz~lmled within the 
period time argument: 
16) The pressure is low. 
Boundedness: unbounded 
Event time: ~/1 such that 
|ncludes(period(~owl\])jnoment(~VIl\])) 
For an interval unspecified for boundedness the 
event time /ms a non-committal relation to the 
interval, i.e., it may be an endpoint of or included 
within the period time argument: 
17) The alarm sounded. 
Boundedness: unspecified 
Event time: M l such that 
has(period(\[soundl\]),moment (~vfs\])) 
The moment time argument of a transition event 
is id~e=~/~/to its event time. Identity, or the lack 
of referential distinctness, is handled through Pro- 
log unification. 
18) The engine failed. 
Boundedness: transition hound 
Event time: M l unifies with 
moment(\[fail1\]) 
Defining these three different relations of event 
time to temporal structure simplifies the computa- 
tion of the ordering relations given by the perfect 
and non-perfect tenses. 
5.2. Temporal Ordering RelatlonB 
The event time computed in Module 1 and 
the verb's tense and taxis comprise the input used 
in computing temporal ordering relations. Due to 
the pragmatic complexity of the perfect tenses and 
22 
to space \]~m~tatlons, neither referenee time nor 
taxis is discussed here (but cf. Passonneau, 1986a). 
The rules for the past and present tenses are quite 
simple. They locate the event time as coincident 
with or prior to the time of text production (i.e., 
the Report Time): 
IF Tense=present 
AND Taxis/non-perfect 
THEN coincide(Event Time, Report Time) 
IF Tense=past 
AND Taxis/non-perfect 
THEN precedes(Event Time, Report Time) 
These two rules in combination with the different 
relations of event time to the temporal structures 
of situations makes it possible to capture impor- 
tant facts about the interaction of tense and 
aspect. For example, present tense denotes an 
actual time only when applied to unbounded 
intervals. Thus a reference to an actual situation 
is computed for sentences like 19) hut not 20). 
19) The engine is failing. 
20) The engine fails. 
In 20), the present tense pertains not to a specific 
event of engine failure, but rather to the ten- 
dency for this type of situation to recur. 
A predication denoting a past unbounded 
situation can be followed by a predication assert- 
ing the continuation or cessation of the same 
situation: 
21) The pump was operating at 08.~0 and is still 
operating. 
A single interval would be generated for the two 
clauses in 21). However, a similar assertion follow- 
ing a predication with a transition event verb in 
the simple past is contradictory if still is inter- 
preted as indicating persistence of the same 
event. Is 
22) ?The pump sheared the drive shaft and is 
still shearing it. 
The event time for the first conjunct in 22) is a 
moment necessarily culminating in a new situa- 
tion (i.e., a state of being sheared). Since the 
transition bound is dimensionless, the adverb still 
cannot refer to its persistence. A predication 
evoking an unspecified interval in a similar 
ISAnother reading of 22) refers to a uniqe event followed 
by iterations of the same type of event. 
context can be interpreted analogously to either 
21) or 22): 
23) The pump operated at 08.~0 and is still 
operating. 
The non-commlttal relation of event time to tem- 
poral structure for unspecified intervals makes 
both interpretations of 23) possible, and selecting 
among them is undoubtedly a pragmatic task 
rather than a semantic one. As we will see next, 
the utility of distinguishing between unbounded 
and lm~peeifled process predications is especially 
apparent in the context of temporal adverbials. 
6. Coneluslom Adverbial Modification 
The representations described above were 
inspired by remarks found in the literature on 
tense and aspect to the effect that the time sche- 
mata (Vendler, 1967) associated with different 
situations are crucial to the way we perceive and 
talk about them. One of the crucial types of evi- 
dence used in deriving PUNDIT's temporal seman- 
tics as the interpretation of temporal adverbials in 
different contexts (Passonneau, 1988a). Conse- 
quently, one of the advantages to the representa- 
tions is that they make it possible to tailor the 
interpretation of a temporal adverb to the tem- 
poral structure of the modified situation. 
For example, specifying a different relation 
for the event time of an active interval, depend- 
ing on its boundedness, yields different temporal 
relations between the situations described in sen- 
tences llke 24-26), as shown informally in the 
examples. 
24) The pump failed when the engine was rotat- 
ing. 
transition o/failure during period of rotation 
25) The pump failed when the engine rotated. 
tran~itlon o/failure during 
OR at one endpoi,~t of period o/rotation 
26) The engine rotated when the pump failed. 
Same =a ~S) 
Sentences like 25) and 26) are often interpreted 
with the process (e.g., rotation) beginning at or 
after the transition event moment (e.g., failure). 
PUNDIT's representations of the temporal seman- 
tics of predications are explicit enough yet 
sufficiently non-commlttal to provide suitable 
input to a pragmatic reasoner that could decide 
these cases. 
23 
Acknowledgements 
I would like to thank Martha Palmer, Lynette 
Hirschman, Bonnie Webber and Dehbie Dahl for 
their comments, encouragement and patience. 
REFERENCES 
Allen, James F. 1984. Towards a general theory 
of action and time. AI 23: 123-154. 
Allen, James. F. 1983. 1V~aintaining knowledge 
about temporal intervals. ACI~ 25.11:832-843. 
Barwise, Jon and John Perry. 1983. Situations and 
Att|tudes. Cambridge, Massachusetts: The MIT 
Press. 
Dahl, Deborahl. 1986. Focusing and reference reso- 
lution in PUNDIT. Presented at AA.~. Philadel- 
phia, PA~ 
Dahl, Deborah; Dowding, John; Hirschman, 
Lyuette; Lang, Francois; Linebarger, Marcia; 
Palmer, Martha; Passonneau, Rebecca; Riley, 
Leslie. 1987. Integrating Syntax, Semantics, 
and Discourse: DARPA Natural Language 
Understanding Program. Final Report May, 
1985--May, 1987. 
Dowty, David R. 1986. The effects of aspectual 
class on the temporal structure of discourse: 
semantics or pragmatics? Linguistics and Philo- 
sophy 9: 37-61. 
Dowty, David R. 1979. Word Meaning and Mon- 
tague Grammar. Dordrecht: D. Reidel. 
Jakobson, Roman. 1971 \[1957\]. Shifters, verbal 
categories and the Russian verb. In his Selected 
Writings, "v'ol. 2, pp. 130-147. The Hague: Mou- 
ton. 
Mourelatos, Alexander P. D. 1981. Events, 
processes, and states. In Tedeschi and Zaenen, 
pp. 191-212. 
Palmer, 1Vlartha; Dahl, Deborah A.; Schiffman, 
Rebecca J. ~)assonneau\]; H~rschman, Lynette; 
Linebarger, Marcia; Dowding, John. 1986. 
Recovering Implicit Information. 24th Annual 
Meeting of the ACL. Columbia University, New 
York. 
Passonneau, Rebecca. 1986a. A Computational 
Model of the Semantics of Tense and Aspect. 
Logic-Based Systems Technical Memo No. 43. 
Paoli Research Center. SDC. December, 1986. 
Passonneau, Rebecca. 1986b. Designing Lexical 
Entries for a Limited Domain. Logk-Based Sys- 
tems Technical Memo No. 42. Paoli Research 
Center. SDC. November, 1986. 
Reichenbach, Hans. 1947. Elements of Symbolic 
Logic. New York: The Free Press. 
Talmy, Leonard. 1985. Lexicalisation patterns. In 
Language Typology and Syntactic Description, 
vol. 3: Grammatkal Categories and the Lexicon, 
pp. 57-151. Edited by Timothy Shopen. Cam- 
bridge: Cambridge University Press. 
Taylor, Barry. 1977. Tense and continuity. 
Linguistics and Philosophy 1. 
Tedeschi, P. J. and A. Zaenen, eds. 19810 Syntax 
and Semantics, vol 14: Tense and Aspect. New 
York: Academic Press. 
Vendler, Zeno. 1967. Verbs and times. Linguistics 
in Philosophy. New York: CorneU University 
Press. 
"v'lach, Frank. 1981. The Semantics of the pro- 
gressive. In Tedesch\] and Zaenen, pp. 271-292. 
24 
