Time and Tense in English 
Mary P. Harper and Eugene Charniak 
Brown University 
Department of Computer Science 
Box 1910 
Providence, RI 02912 
Abstract 
Tense, temporal adverbs, and temporal connectives 
provide information about when events described in 
English sentences occur. To extract this temporal 
information from a sentence, it must be parsed into a 
semantic representation which captures the meaning of 
tense, temporal adverbs, and temporal connectives. 
Representations were developed for the basic tenses, some 
temporal adverbs, as well as some of the temporal 
connectives. Five criteria were suggested for judging these 
representations, and based on these criteria the 
representations were judged. 
Introduction 
English sentences contain many types of temporal 
information. Tense is used to inform the reader (listener) 
of when the event associated with the main verb occurs 
with repect to the time of utterance. That is, tense informs 
the reader that an event occurs before, after, or during the 
time of utterance. Temporal adverbs (such as tomorrow or 
now) add additional information about the events in a 
sentence. Temporal connectives tell the reader about the 
temporal relationship between the events in the main clause 
and the events in the subordinate clause. While there is 
other temporal information that can be found in sentences, 
the following will concentrate on these three. 
To extract temporal information from a sentence, it must 
be parsed into a semantic representation which captures the 
meaning of tense, temporal adverbs, and temporal 
connectives. A temporal representation of tense, adverbs, 
and temporal connectives must : 
1. provide a way to reject temporally incorrect 
sentences, such as * "I will run yesterday." 
2. allow one to reason about the temporal relationship 
between events. For instance, the sentence "I had 
run when he arrived" implies that the run event 
occurs before the arrival, whereas in the sentence 
"I was running when he arrived," the arrival and 
run events overlap. 
3. allow the exact time of event to be unfixed until it is 
pinpointed based on contextual information or adverbial 
modification. 
4. allow reference to points and intervals of time (eg. 
precisely at 3 PM VS. for 5 hours). 
This work has been supported in part by the National 
Science Foundation under grants IST 8416034 and IST 
8515005, and Office of Naval Research under grant 
N00014-79-C-0529. 
5. allow parsing of temporal information in sentences 
to be simple and compositional. 
These criteria were used to judge previous temporal 
representation research (Bruce (1972), Hornstein (1977, 
1981), Yip (1985)). None fulfilled all five criteria. The 
criteria will also be used to judge the representations 
developed here. 
Tense 
The representations for tense, adverbs, and temporal 
connectives developed here is based on McDermott's 
(1982) temporal logic. McDermott's "point-based" 
temporal logic was chosen because it is not unusual to talk 
about the beginning and end points of a period of time or 
an event. In fact, the semantics of tense developed here 
relate the endpoints of events in sentences. This 
representation of tense provides a flexibility not found in 
many other representations of tense (eg. (Hornstein, 
1977,1981)). Flexibility is important since events can 
extend over tense boundaries (for instance, "In 3 minutes, 
the boy will have run for 24 hours."). Any representation 
of events in time must capture the fact that some events do 
not always wholly occur in the past, present, or future with 
respect to the time of utterance. 
The tense rules are compositional and require the 
following relations : < (before), > (after), = 
(cotemporaneous), < (before or cotemporaneous), and -> 
(after or cotemporaneous). It is assumed that events are 
"unit" events and have a beginning and an end point, 
where the beginning of an event is before or simultaneous 
to its end point. The endpoint of an event need not imply 
the achievement of the purpose with which the event was 
initiated (eg. the existence of the end point of a winning 
event need not imply that the state of having won is 
achieved). To capture the meaning of simple as well as 
more complex tenses, we introduce the following events : 
1. ~ - This is simply the speaking event 
associated with a sentence. 2. ~ - This is the event indicated by the main 
verb of the sentence. For instance, the run event in the 
following sentence is the main event : "I have been 
running to the store." 3. ~ - This is the time interval referred to 
"Bill in sentences like : had eaten at 3 PM," which 
describes an eat event in the "distant past." This 
sentence implies the existence of an event or time 
interval which occurs after the main event (eat) but 
before the utterance event. 
4. Proeressive Event - This is the time interval from 
which the main event extends into the past and into the 
future. The progressive event may have no correlation 
with a "real world" event, but its existence predicts 
certain phenomena in our model of temporal adverbs 
and connectives. It can be thought of as a place holder, 
or the minimal possible duration of a main event with 
progressive aspect. 
The following five rules describe the semantics of tense 
both in English and in our representation. The verbs in a 
sentence are parsed left to right (assuming an ATN, which 
is the parser in which these tense rules were implemented). 
One of the following three rules is triggered by the tense of 
the first verb in the sentence. "Event" (in the fast three 
rules) can be a main event, a perfect event, or a 
progressive event depending on the sentence. 
1. Past rule : This rule implies that there exists some 
event that must end before the beginning of the 
utterance event. 
(< (end even0 (begin utterance-even0) 
2. ~ : This rule implies that there exists some 
event that is either cotemporaneous with the utterance 
event or can begin at or after the beginning of the 
utterance event. Which is asserted seems to depend on 
the aspect of the verb associated with event. 
If the current verb is stative then 
(and (= (begin event) (begin uterance-event)) 
(= (end event) (end uterance-event))) 
If the current verb is not a stative then 
('d (begin event) (begin uterance-event)) 
3. ~ : This rule implies that there exists some 
event that must begin after the end of the utterance 
event. 
(> (begin event) (end utterance-event)) 
The following rules are required to interpret the more 
complicated perfect and progressive tenses. 
4. ~ : This rule is triggered by the word have 
followed by a past participle. The event in the rule can 
be a progressive or a main event. 
(< (end event) (begin perfect-event)) 
5. ~ : This rule is triggered by the word 
be followed by a progressive verb form. The event in 
the rule can only be a main event. 
(and (< (begin main-event) (begin progressive-event)) 
(end main-event) (end progressive-event))) 
These rules combine in a compositional way to define 
the more complicated tenses. For instance the past perfect 
progressive tense combines the past rule with the perfect 
and progressive rules. Thus the sentence "Jack had been 
running" is represented as follows : 
(and (hast utterance6 utterance-event) 
(< (end have2) (begin utterance6)) ; past rule 
(inst have2 perfect-event) 
(<= (end be3) (begin have2)) ; perfect rule 
(inst be3 progressive-event) 
(inst run64 run) 
(<= (begin run64) (begin be3)) ; progressive rule 
(>= (end run64) (end be3)) 
(hast run64 main-event) 
(name Jack 16 Jack) 
C = '(agent run64) Jackl6)) 
A "temporal" picture can be drawn for this sentence (see 
Figure 1). Note that the picture is only one possible 
depiction of the actual meaning of this representation. 
utterea'~e6 I I 
h~ve2 I I 
I be3 I 
lam64 I I 
( I I ) l~t ~w futt~ 
Figure 1. "Jack had been running." 
A parser uses the semantic rules of tense as follows. 
After checking the tense of the first verb, the parser checks 
to see if the verb is the word will. If it is, then move to the 
next verb and mark the event associated with this verb as a 
future event. Assert either the past, present or future rule 
depending on the tense associated with the "event" of the 
current verb. Now check to see ff the current verb is have 
followed by a past participle. If so, then assert the perfect 
rule relating the perfect event (the event associated with 
have) and the event associated with the verb to the right of 
have, and move to that verb. After checking for perfect 
tense, the parser looks for a form of the word be followed 
by the progressive form of a verb. This signals the 
progressive rule, which relates the progressive event with 
the main event. 
The representation adopted has some support in 
linguistic literature, and there are some similarities to the 
representations developed by Bruce (1972), Hornstein 
(1977, 1981), Reichenbach (1947), and Yip (1985), 
although there are many differences. One difference 
between this representation and previous representations of 
tense is how present tense is defined, All past theorists 
have considered present tense as indicating that the main 
event is cotemporaneous with the time of utterance. 
However, aspect of verb seems to affect the meaning of 
present tense. In present tense sentences, there exists a 
curious phenomenon which can best be understood by 
examining the following two sentences : 
1. I leave at eight o'clock tomorrow. 
2. *I have a dog tomorrow. 
Aspect interacts with present tense requiring a more 
complicated present rule in a theory of tense. 
Adverbials 
The representation of several types of temporal adverbs 
will be considered, as well as how the meaning of these 
adverbs combines with the meaning of the tense. As in the 
representation of tense, we require the following relations : 
<, >, ~, >, and =. We will consider how to predict 
incorrect combinations of tense and adverbs based on the 
representations of tense and adverbs developed here. 
As suggested by Homstein (1977), we adopt the idea 
that which event is modified by an adverbial is an 
important issue (since we introduce multiple events in our 
definition of some of the basic tenses). The ambiguity 
concerning which event is modified can best be seen in the 
following example: "I had eaten at 3." This sentence has a 
utterance event, which can not be directly modified by an 
adverb. It can be modified by context, and it can be 
modified when some event which is cotemporaneous to the 
utterance event is modified. The past perfect sentence 
introduces a perfect event and a main event (eat) in addition 
to the utterance event. If we assume that the main event is 
modified, then the time of"eating" must overlap 3 o'clock. 
If it modfies the perfect event, then by the time 3 o'clock 
came around the "eating" was complete. In general, we 
adopt the idea that which event is modified is ambiguous, 
and thus a disjunction of possibilities is asserted. 
Since Hornstein (1977) and Yip (1985) examined the 
three adverbials tomorrow, yesterday, and now, we will 
concentrate on these three. Each of these adverbs shares 
the fact that they are defined with respect to the time of the 
utterance event (today is also included in this category of 
adverbs though not discussed here). The representations 
of now, tomorrow, and yesterday follow : 
Now : Now is defined to be a time interval which is 
cotemporaneous with the utterance event. Thus, the 
representation of some specific now is : 
(and (inst nowl6 time-interval) 
(= (begin howl6) (begin utterance2)) 
(= (end now 16) (end utterance2))) 
Tomorrow : Tomorrow is also defined with respect to 
the time of utterance. Notice that the duration of tomorrow 
is precisely 24 hours (as indicated in the fourth conjunct). 
(and (inst tomorrow3 day) 
(> (begin tomorow3) (end utterance2)) 
(< (begin tomorrow3) 
(+ (end utterance2)(* 24 hour))) 
(= (- (end tomorrow3) (begin tomorrow3)) 
(* 24 hour))) 
Yesterday : Yesterday is defined with respect to the time 
of utterance, and has a 24 hour duration. 
(and (inst yesterday3 day) 
(< (end yesterday3) (begin utterance2)) 
(> (End yesterday3) 
(- (begin utterance2) (* 24 hour))) 
(= (- (end yesterday3) (begin yesterday3)) 
(* 24 hour))) 
To satisfy criterion 1, this model should be able to 
predict temporal inconsistencies between temporal adverbs 
and tense. Any event in a sentence can be modified by an 
adverb if the event can potentially overlap the period of 
time associated with the adverb. Thus we introduce the 
overlap rule of adverb-tense agreement : 
Overlap Rule : An event can be modified by a temporal 
adverb iff the time period associated with an adverb can 
overlap the time period associated with the event without 
some temporal contradiction. That is, if the following 
assertion does not contradict other temporal assertions 
associated with the sentence, then the events can overlap : 
(and (< (begin event) (end adverb)) 
('d (end event) (begin adverb))) 
Because events are defined flexibly in this tense 
representation, some events can cross tense boundaries. 
For correct adverb-tense agreement, the events in the 
sentence must be "anchored" to the event associated with 
the first verb in the sentence, that is the event that 
determines the tense of the sentence (note that will has no 
event associated with it). The need for this anchoring can 
best be shown with following examples : 
*Now, he will have eaten. (excluding modal reading) 
*Yesterday, he will have eaten. (excluding modal reading) 
Tomorrow, he will have eaten. 
The tense stucture of each of these sentences (as given by 
our tense rules) introduces three events, an utterance event, 
a perfect event, and a main event. Notice that the only 
event that is necessarily in the future is the perfect event. 
The main event could overlap yesterday or now, as well 
as tomorrow. Thus it would seem that given that the main 
event can be modified by yesterday or now, the first two 
sentences should be correct, However, except for 
possible modal readings, these sentences are not 
acceptable. We account for this with the following rule : 
Anchoring rule : If the time period of the event 
associated with the first verb of a sentence can overlap the 
time period associated with an adverb, then the adverb can 
modify that event and can potentially modify the other 
events in the sentence (based on the overlap rule). The 
utterance event can not be modified using the anchoring 
rule. 
To show how these two rules (anchoring and overlap) 
are used, examine the sentence: "He is running now." 
Step I : Get the basic representations of the adverbial and 
the tense. 
(and (inst utterance6 utterance-event) 
; adverb representation 
(inst now5 time-interval) 
(= (begin now5) (begin utterance6)) 
(= (end now5) (end utterance6)) 
; tense representation 
(inst bel progressive-event) 
(= (begin bel) (begin utterance6)) 
(= (end bel) (end utterance6)) 
(inst run4 run) 
(inst run4 main-event) 
(< (begin run4) (begin bel)) 
Cd- (end run4) (end bel))) 
Step 2 : Check to see ff the anchor event can overlap the 
adverb. Assume that CHECK is a function that returns 
true if the overlap is possible. Since Bel and Now5 occur 
at the same time, the result of the test is true. 
(CHECK (and (< (begin bel) (end now5)) 
(> (end bel) (begin now5)))) 
Step 3 : If the overlap check of the anchor returns true, 
then do overlap checks on the remaining events. For those 
that return true, assert a disjunction of ways that the adverb 
can modify the events. In this case assert : 
(or (and (< (begin bel) (end nowS)) 
(end bel) (begin nowS))) 
(and (< (begin run4) (end now5)) 
(> (end run4) (begin now5)))) 
An example of a sentence in which the anchor event and 
the adverb can not overlap is *"He ran tomorrow." The 
Tense-Adverb Compatibility 
Now 
Table I. 
Yes~rday 
ok 
ok or Ping. Rule* 
ok 
error 
error 
en'Dr 
error 
en'Dr 
errDr 
P~t 
Past Progressive 
Past Perfect 
Present 
Present Progressive 
Present Perfect 
Future 
Futu~ Progressive 
Future Perfect 
Tomorrow 
error error 
Ping. Rule only Prvg. Rule only 
error error 
ok ok or error 
ok Ping. Rule only 
ok error 
error ok 
error ok or Prvg. Rule 
error ok 
* Reference ~o "Ping. Rule" refers ~ a modification of the Pest Progressive 
Rule sugges~d by Hon~in (1977), wkich is ~nored in this paper. 
run event can not overlap tomorrow (because the run event 
ends in the past and tomorrow begins in the future), and 
the sentence is therefore reported as erroneous. See Table 
1 for the adverb-tense predictions of our model. Modal 
readings are ignored in this paper. 
There are other adverbials which are interpreted relative 
to the time of utterance (for instance, this week, next 
week, and last year). It is not difficult to imagine how to 
represent these adverbials. There are also some adverbials 
which need not be defined relative to the time of utterance. 
These include all of the clock calendar adverbials, such as 
Sunday and midnight. For example the representation of a 
specific Sunday is : 
(and (hast sunday3 day) 
(= (- (end sunday3) (begin sunday3)) 
(* 24 hour))) 
Sunday3 can not be placed in the past, present, or future. 
However, when Sunday is used in a sentence, we can 
determine whether we mean a past, present, or future 
Sunday. Durational adverbials can also be easily 
represented (somewhat like the definition of Sunday). 
There are other adverbials which like clock calendar 
adverbials are not interpretted with respect to the time of 
speech. One such temporal adverb is just. This adverb is 
distinguished from the word just, meaning only. To see 
how it is used, examine the following sentences : 
1. I just ate lunch. 
2. I was just eating lunch. 
3. I had just eaten lunch. 
4. * I just eat. 
5. I am just eating lunch. 
6. I have just eaten lunch. 
7. * I will just eat lunch. 
8. I will be just eating lunch. 
9. I will have just eaten lunch. 
Notice that just can not be used in simple present or simple 
future tense. This adverb requires the existence of some 
event in the sentence that begins immediately after the start 
of the event modified by just. Sentences 5 and 8 require 
progressive events to represent their tense structure. This 
tense representation allows our model to predict the 
correctness of these two sentences. The definition of just 
follows : 
Just : Just relates two events, where Evl can be the main 
event, the progressive-event, or the perfect-event, and Ev2 
can be the utterance-event, the perfect-event, or the 
progressive-event. Evl and Ev2 must not be separated by 
another event introduced by the sentence. 0 is some small 
value which is determined by context. 
(< (begin Evl) (begin Ev2)) 
if (< fEnd Evl) (begin Ev2)) then 
assert (< (- (begin Ev2) (End Evl)) 0) 
else assert (< (- (begin Ev2) (begin Evl)) ~) 
There are many other temporal adverbials that need to 
be represented, among them recently, afterwards, earlier, 
lately, already, and soon. Most of these relate two events, 
in much the same way as temporal connectives which will 
be our next topic. 
Temporal Connectives 
A few issues must be examined before we present our 
representation of temporal connectives. First it should be 
pointed out that temporal connectives are subordinators. 
Most subordinators do not restrict the tense of the 
subordinate clause given the tense of the main clause. The 
tense of the main clause does restrict the tense of the 
subordinate clause when the subordinator is a temporal 
connective. The following results are predicted by 
Hornstein (1977) : 
John left when Harry 
1. *arrives. 4. arrived. 7. *will come. 
2. *is arriving. 5. was arriving. 8. *will be coming. 
3. *has arrived. 6. had come. 9. *will have arived. 
By studying the above example, one might suggest that the 
tense of the main clause and the tense of the subordinate 
clause must have the same tense (disregarding progressive 
and perfect aspects). This seems to be true for all past and 
present tenses. There are some restrictions of this 
statement, however, since the will/shall construction of 
future tense is not allowed in temporal subordinate clauses. 
As pointed out by l.~ech (1971) : 
"In dependent clauses inlxoduced by conditional and temporal 
conjunctions if, unless, when, as soon as, as, etc., the future is 
denoted by the ordinary Present Tense instead of the construction 
with will~shall : 
I'll tell you if it hurts. 
When the spring comes, the swallows will return. 
Jeeves will announce \[he guests as they arrive." (p.59) 
If the will~shall construction is used in a subordinate 
clause introduced by a temporal connective, then the 
reading of the sentence is not a future but a modal reading. 
This fact was not noticed by Hornstein (1977, 1981) or 
Yip (1985). Hornstein allows both present tense and 
will~shall future tense to occur in temporal subordinate 
clauses. Yip only allows the will~shall future tense to 
occur in the subordinate clause 1. 
Rather than include the syntactic needs of temporal 
connectives in our semantic representation, it seems wiser 
to include the requirement at a syntactic level. That is the 
tense of the f'trst verb of the main clause restricts the tense 
of the first verb in the temporal subordinate clause. If the 
tense of the first verb in the main clause of the sentence is 
past or present, then the tense of the first verb in the 
subordinate clause must have like tense. If the tense of the 
first verb in the main clause is future tense, then the tense 
of the fhst verb in the subordinate clause must be present 
tense (though it will be semantically interpretted as future 
tense). 
Now, we must consider how to extract the temporal 
meaning of sentences of the form sentence-temporal 
connective-sentence. Each clause will be given a temporal 
representation as indicated in the tense representation, 
section of this paper. Both clauses will have the same time 
of utterance, since an utterance event is created only for a 
sentence. The only subtlety is the requirement that present 
tense in a subordinate clause be interpretted using future 
semantics when the main clause has future tense. After 
each clause is represented, the semantics for the temporal 
connective must be invoked. Each temporal connective 
requires its own definition, as pointed out by 
Hornstein(1977). These definitions will determine the 
temporal relationship between the events in the main clause 
and the events in the subordinate clause. We will present 
the definitions for five temporal connectives : when, while, 
until, before, and after. Because these definitions can use 
the representation of tense associated with each clause in a 
sentence to interrelate the events between clauses, the 
strength of the tense representation is increased. 
When : align the anchor events to determine the 
relationship between events of the clauses. If the main 
events of both clauses are the anchor events, then the 
events may occur at exactly the same time, though not 
necessarily. 
(and (= (begin anchor-event(main-clause)) 
(begin anchor-event(subordinate-clause))) 
(= (end anchor-event(main-clause)) 
(end anchor-event(subordinate-clause)))) 
While : align the anchor and main events of the clauses. 
Check to see if the alignment of both is possible. If check 
.-etums false then reject the sentence. 
1. Yip(1985) and Hornstein(1977) try to deal with this 
temporal connective phenomenon and adverb-tense 
agreement with a unified theory. Hornstein's theory 
accepts sentences of the form *"I have eaten tomorrow" so 
that the sentence "I will leave when he has eaten" is 
acceptable. Yip modifies Hornstein's theory to get rid of 
the yesterday-present perfect error, but the modification 
does not allow a future tense main clause to have a present 
tense subordinate clause. 
(and (= (begin anchor-event(main-clause)) 
(begin anchor-event(subordinate-clause))) 
(= (end anchor-event(main-clause)) 
(end anchor-event(subordinate-clause))) 
(= (begin main-event(main-clause)) 
(begin main-event(subordinate-clause))) 
(= (end main-event(main-clause)) 
(end main-event(subordinate-clause)))) 
Until : requires in most cases that the main event of the 
main clause end when the the main event of the subordinate 
clause begins. If the tense representation of the subordinate 
clause has a perfect event and no progressive event, then 
the main event of the main clause must end when the main 
event of the subordinate clause ends. 
If subordinate clause has a perfect but no progressive event 
(= (end main-event(main-clause)) 
(end main-event(subordinate-clause))) 
Else 
(= (end main-event(main-clause)) 
(begin main-event(subordinate-clause))) 
Before : requires that the anchor event of the main clause 
end before the beginning of the main event of the 
subordinate clause. 
(< (end anchor-event(main-clause)) 
(begin main-event(subordinate-clause))) 
After : requires in most cases that the main event of the 
main clause begin after the end of the anchor event of the 
subordinate clause. If the main clause has a progressive 
event, then the anchor event of the main clause begins after 
the end of the anchor event of the subordinate clause and 
the main event of the subordinate clause ends before the 
end of the main event of the main clause. 
If main clause has a progressive event then 
(and (< (end anchor-event(subordinate-clause)) 
(begin anchor-event(main-clause))) 
(< (end main-event (subordinate-clause)) 
(end main-event (main-clause)))) 
Else 
(< (end anchor-event(subordinate-clause)) 
(begin main-event(main-clause))) 
Notice that before and after are not always inverses of one 
another. Consider the following two sentences : 
1. I ate before he was running 
2. He was running after I ate. 
If before and after were inverses, then sentence 1 and 2 
would have equivalent meanings which they do not. The 
definitions of before and after capture this assymetry. 
Two Examples are presented to acquaint the reader 
with the representation of sentences joined by temporal 
connectives. The fLrst is : "Mary ate when Joe was eating." 
I. Represent the clauses. 
(and (inst utterance3 utterance-even0 
; "Mary ate" 
(inst eat22 ca0 
(inst eat22 main-event) 
(< (end eat22) (begin utterance3)) 
(name Mary22 Mary) 
(:= '(agent eat22) Mary22) 
;"Joe was eating" 
(< (end beA) (begin utterance3)) 
(inst be4 progressive-event) 
Onst eat23 eat) 
(hast cat23 main-event) 
(< (begin eat23) (begin be.A)) 
(end cat23) (end beA)) 
(name Joel2 Joe) 
C = '(agent eat23) Joel2)) 
IL Do semantics for when. Note that the anchor event for the main clause is 
eat22, and the anchor event for the subordinate 
clause is beA. (and 
(= (begin cat22) (begin be4)) 
(= (end eat22) (end be4))) 
This sentence can depicted as follows (see Figure 2) : 
ea122 I 
be4 
I 4 
ea123 I I 
 u tenmce3 • ! 
past nov future 
Figure 2. "Mary ate vhen Joe vas eating." 
This implies that eat23 can begin before and end after 
eat22, though they could be exactly coincident. This 
seems to be the desired interpretation of this sentence. 
This is not the meaning that Hornstein's model would give 
this sentence. Yip(1985) introduces progressive aspect 
rules to Hornstein's tense rules to get exactly this result. 
The second example consists of an analysis of the 
sentence : "Mary ate when he had eaten." 
I. Represent the clauses. (and (hast utterance3 utterance-event) 
; '~Mary ate" representation 
(inst cat22 cat) 
(inst cat22 main-event) 
(< (end cat22) (begin utterance3)) 
(name Mary22 Mary) 
C = '(agent cat22) Mary22) 
; "He had eaten" representation 
(< (end have3) (begin utterance3)) 
(inst have3 perfect-event) 
(hast cat23 eat) 
(hast eat23 main-event) 
(~ (end cat23) (begin have3)) 
(inst Jackl 2 Jack) 
C = '(agent cat23) Jackl2)) 
II. Do semantics for when. 
Note that the anchor event for the main clause is 
cat22, and the anchor event for the subordinate 
clause is have3. (and (= (begin cat22) (begin have3)) 
(= (end eat22) (end have3))) 
This sentence can be depicted as shown in Figure 3. 
Thus, it can be seen that eat23 must end by the beginning 
of eat22, This seems to be the correct interpretation of this 
sentence, and was exactly the interpretation that 
Hornstein's when rule makes. These two examples show 
that the when rule predicts very different relationships 
between events depending on the tenses in the clauses. 
u~emnee3 
, ea'~2 
I ea~3 I 
past now future 
Figure 3. "Mary ate vhen Jack had eaten " 
Conclusion 
This paper describes a preliminary study of the temporal 
phenomena found in English sentences. Many issues have 
been ignored for simplicity. For instance, the issue of 
habitual readings of verbs was not examined. The 
meanings of verbs with temporal aspects (such as plan ) 
were also not considered. In addition, we did not consider 
how to relate (in time) events from different sentences. 
The only events from different sentences that can be related 
are the utterance events. If two sentences occur in 
sequence, one can conclude only that the utterance event of 
the In'st ends before the utterance event of the second. 
The model developed here can, however, temporally 
order events within a sentence. Five criteria were 
suggested at the beginning of the paper for the 
representation of temporal information found in an English 
sentence. These criteria guided the development of our 
model. All criteria were met, except the compositional 
parse criterion in a few cases. There seem to be 
unavoidable special cases which can not be captured in 
compositional tense, adverb, and temporal connective 
rules. For instance, the meanings of some adverbs require 
tense information to determine their correct representations 
(e.g. just). 
References 
Allen, James. Maintaining Knowledge About Temporal 
Intervals. CACM, 1983, 26, 832-843. 
Bruce, Bertram C. A Model for Temporal References and 
Its Application in a Question Answering Program. 
Artificial Intelligence, 1972, 3, 1-25. 
Charuiak, E., Gavin, M., and Hendler, J. The Frail/Nasl 
Reference Manual. Brown University Technical Report 
CS-83-06, 1983. 
Charniak, E. and McDermott, D. Introduction to 
Artificial Intelligence. Reading, MA : Addison-Wesley 
Publishing Company, 1985. 
Hornstein, Norbert. Towards a Theory of Tense. 
Linguistic Inquiry, 1977, 8, 521-557. 
Hornstein, Norbert. The Study of Meaning in Natural 
Language, In N. Hornstein & D. Lightfoot (Eds.), 
Explanation in Linguistics. New York : Longman, 
1981. 
Leech, Geoffrey N. Meaning and the English Verb. 
London : Longman, 1971. 
McDermott, Drew. A Temporal Logic For Reasoning 
About Processes And Plans. Cognitive Science, 1982, 
6, 101-155. 
Reichenbach, Hans. Elements of Symbolic Logic. 
New York : MacMillan, 1947. 
Yip, Kenneth M. Tense, Aspect and the Cognitive 
Representation of Time. IJCAI Proceedings, 1985, 
806-814. 
