ASPECT, ASPECTUAL CLASS, AND THE TEMPORAL STRUCTURE OF 
NARRATIVE 
Alexander Nakhimovsky 
Department of Computer Science 
Colgate University 
Hamilton, NY 13346 USA 
This paper consists of two parts. The first part discusses commonsense knowledge about events as 
manifested in language. Three kinds of knowledge are identified: compositional, durational, and 
aspectual. Compositional knowledge concerns internal structuring of events into preparatory, initial, 
main (the body), final, and resulting stages. Durational knowledge concerns durational relations 
between events and stages of the same event. Durational knowledge can be expressed as qualitative 
dependencies among the parameters of the event and as its time scale. The notion of time scale is 
introduced and related to shared cyclical events (time units). 
In discussing aspectuai knowledge three notions are distinguished: aspect as a grammatical category 
of the verb, implemented by affixes, auxiliaries, and such; aspectual class, which is a characteristic of 
a lexical meaning; and the aspectual perspective of the sentence determined by the position of the 
Reference Time (RT) with respect to the event described by a finite clause. I argue that an aspectual 
classification of situations evolving in time should be based on such considerations as the kinds of 
resources they consume and the goals they achieve. A detailed classification of instantaneous and 
noninstantaneous events is developed. 
The second part of the paper discusses how this knowledge is employed in understanding extended 
narratives. Temporal discontinuities, in conjunction with other kinds of discontinuities identified in the 
paper, signal boundaries between discourse segments; within each segment, all three varieties of 
temporal knowledge help establish the temporal relations among narrated events. 
1 INTRODUCTION 
This paper consists of two parts. 1 The first part (sec- 
tions 2-5) discusses aspect and other varieties of com- 
monsense knowledge about time as manifested in lan- 
guage. The second part (sections 6-8) discusses how 
this knowledge is employed in understanding extended 
narratives. This preliminary section spells out the pa- 
per's principles and aspirations. 
A major influence on this paper has been the study of 
commonsense knowledge and reasoning in the paradigm 
established in Hayes 1985 (1979), Hobbs and Moore 
1985, and Hobbs et al., 1985, 1986. The paper proceeds 
from the premise that commonsense studies and natural 
language semantics are closely interdependent: natural- 
language semantics is essentially incomplete without a 
theory of the commonsense world, for which natural 
language provides key empirical data. These data are of 
two kinds, lexical and grammatical. So far, mostly 
lexical data have been used in work on knowledge 
representation: the contexts of a given word, the struc- 
ture of semantic fields, and metaphorical uses (Hobbs 
1984:284, 1985a:3). This paper suggests that in working 
on core knowledge (time, space, causality), it is imper- 
ative to recognize grammatical categories as powerful, 
and language specific, tools of organizing conceptual 
space. 
Among grammatical categories particularly impor- 
tant are those that allow, and indeed force, the language 
user to choose between alternative ways of viewing the 
same object or situation. I will call such categories 
subjective. A paradigmatic example of a subjective 
grammatical category is aspect. (In fact, the term aspect 
came into being as an attempt to translate the Russian 
term vid ("view"); see Lyons 1977:705.) Subjective 
grammatical categories are important for two reasons. 
First, they suggest restrictions on the formalism: since 
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/88/010029-43503.00 
Computational Linguistics, Volume 14, Number 2, June 1988 29 
Alexander Nakhimovsky Aslaect, Aspectual Class, and the Temporal Structure of Narrative 
language users can easily switch between alternative 
views of the same entity, the formalism should provide 
for alternative representations related by a simple and 
general operation. (Bunt 1985:39-40 makes a similar 
point with respect to the opposition mass-count.) Sec- 
ond, subjective grammatical categories show that the 
corresponding conceptual distinctions reflect our alter- 
native ways of viewing things rather than intrinsic 
properties of things. There are intrinsic properties that 
interact with subjective distinctions, and in a language 
that does not have a well-articulated subjective gram- 
matical opposition, the intrinsic and the subjective 
become easily confused. An example of such a confu- 
sion is, I believe, the well-known taxonomy of events, 
processes, and states (Mourelatos 1981), which is dis- 
cussed in Section 4.3. 
It follows that in designing the primitives of our 
semantic representations it is essential to look at data 
across languages, paying particular attention to subjec- 
tive grammatical categories. (They are easy to identify 
because they are notoriously difficult to learn how to 
use correctly, possibly because of having to internalize 
a new global operation on the mental model.) Decades 
of linguistic research can thus be brought to bear on the 
enterprise of formalizing commonsense knowledge. 
Conversely, this enterprise makes it possible to pose 
with greater precision the old problem of how language, 
and in particular its grammatical categories, influences 
our thought and vision of the world. 
Formalizations of the commonsense may or may not 
turn out to be language and culture specific, but they 
will have to have features particular to our planet and 
the species. Such accidental parameters of human ex- 
istence as the duration of a year, the length of an arm, 
and the temperature of the human body are universal 
constants that play an important role in our reasoning: 
they determine how we factor a continuous parameter 
like temperature or time into a quantity space (in the 
sense of Forbus 1985:80-1). Therefore, and contrary to 
a widespread view, any rigorous semantics of natural 
language has to concern itself with definitions of exten- 
sional concepts. Such definitions obviously belong in 
the lexicon. In this paper, lexicon is understood to be a 
repository of both grammatical and commonsense 
knowledge, indexed by lexical items. This view of the 
lexicon, advocated also in Hobbs et al., 1985, has 
important antecedents in Soviet and European linguis- 
tics (Apresjan 1974, Wierzbicka 1985). 
A second major influence on this paper is the notion 
of computational modeling of natural-language seman- 
tics (Hendrix and Moore 1979), i.e., the idea that 
semantical objects associated with natural-language ex- 
pressions are (typed) data structures, together with 
inference procedures that operate on them. These data 
structures form an internal representation language that 
may or may not in turn be endowed with a semantics of 
its own. In what follows I assume that semantical 
objects are frames and collections of frames, and the 
basic :ypes include frame and slot label (cf. Creary and 
Pollard 1985, and Hirst 1987, ch.3). There are sortal 
predical~.es, or simply sorts, that are defined on frames 
and organize them into an ISA hierarchy. 
The', first cut in the hierarchy is between flames 
representing objects and flames representing situations 
persis :ing or evolving in time. (Loosely following Hayes 
1985, I call the latter histories.) Orthogonal to this 
distinction is the one between generic flames or types, 
and their instances or tokens: tokens, but not types, 
have a slot for a spatio-temporal location. I will thus 
speak of history types (h-types) and history tokens (h- 
tokens); when the type-token distinction is irrelevant I 
will simply say "history." H-types can be of various 
degrec.s of generality (cf. the representations of read, 
read a book, read a book for an hour; all of these, 
however, are types of which the representation of Jon 
read a book for an hour is a token). Associated with 
each h-type is a set of slot labels and selectional 
restrictions on them, usually a conjunction of sortal 
predicates (Hobbs et al., 1986). 
Whether types or tokens, the semantical objects of 
natural-language semantics are only partially specified, 
and ":nore partial" objects are embeddable, with some 
provisions for nonmonotonicity, into less partial ones. I 
assume that representations of lexical items in the 
lexicon form a discrete level of partiality. The subject of 
the first part of the paper can then be stated as follows: 
What is the knowledge about events that is part of 
lexical meanings? How can it be coded in the h-types of 
the lexicon, and how is it used as lexical meanings are 
combined and embedded into more precisely specified 
strucrares? In the end, what it all boils down to is an 
extended argument for some slots in the lexical repre- 
sentations of histories and a few sortal predicates on 
h-types representing aspectual classes. 
Finally, with respect to narrative understanding, a 
major itffluence has been the idea that such understand- 
ing means constructing, or "building a good structure" 
(Bruce 1981:283), so that the semantic value of a 
lingui~;tic unit (phrase, sentence) is its contribution to 
the emerging representation of discourse. This idea, in 
one o1' those parallel developments that do not seem to 
be purely coincidental, emerged at about the same time 
in the Reader-Response school of literary criticism 
(e.g., Tompkins 1980), in model-theoretic semantics 
(Kamp 1981, Heim 1982) and in AI approaches to 
discourse; in combination with the idea of computa- 
tional modeling it first clearly appears in Webber 1979 
(see also Webber, this volume). 
2 VARIETIES OF TEMPORAL KNOWLEDGE AND 
COMPOSITION OF H-TYPES 
This and the next two sections summarize and elaborate 
on the proposals of Nakhimovsky 1986, 1987a. Com- 
monscnse knowledge about time falls into three catego- 
ries: compositional, durational, and aspectual. Compo- 
30 Computational Linguistics, Volume 14, Number 2, June 1988 
Alexander Nakhimovsky Aspect, Aspeetual Class, and the Temporal Structure of Narrative 
sitionally, h-type representations in the lexicon consist 
of the following components: 
a. Preparatory stage. 
b. Initial stage. 
c. Body. 
d. Final stage. 
e. Resulting stage. 
In Nakhimovsky 1986 only components b--e were rec- 
ognized, but Moens and Steedman 1987 (also this vol- 
ume) convincingly shows that a needs to be added, at 
least when the h-token is a planned human action. 
Components b--e are probably uncontroversial, except 
in certain cases, discussed below, where they are col- 
lapsed together. For example, consider the verb cross, 
as in cross the road. It describes movement from one 
side of a ribbonal surface to another, following a trajec- 
tory that is usually at approximately the right angle to 
the edge of the surface. The stages are: be located on 
one side, cross the edge, move across, cross the other 
edge, be located on the other side. So, as Steedman and 
Moens observe, the clause When the children crossed 
the road.., can be followed by any of the following: 
Example 1 
a. They waited for the teacher to give a signal. 
b. They stepped onto its concrete surface as if it 
were about to open and swallow them. 
c. They were nearly hit by a car. 
d. They reached the other side stricken with fear. 
e. They found themselves surrounded by strangers. 
Moens and Steedman argue, correctly I think, that the 
different temporal relations between the when clause 
and the continuations do not result from the ambiguity, 
or polysemy, of the temporal connective: the when 
clause simply makes both the entire h-token and all its 
component stages available for focusing. What Example 
1 does show is the weak aspectual typing of English 
verb forms: the same simple past form crossed makes 
all the stages of the h-type accessible. The most that can 
be said is that before some, but not all, of the continu- 
ations in Example 1, the simple past can be replaced by 
the progressive. In languages with strong aspectual 
typing an imperfective verb (form) would have to be 
used in the when clause before (lb) or (lc). 
This example points up what I believe to be the main 
function of aspect as a grammatical category: making 
either an h-type in its entirety, or (some subsequence of) 
its internal stages, available for focusing. In order to 
make the notion of focusing more precise I will use the 
notion of Reference Time (RT), adopted from Reichen- 
bach 1947 but reinterpreted pragmatically: RT is to be 
thought of as a point or a very short interval that 
indicates where the current focus of attention is located. 
It may be obvious that after (la) RT is located in the 
middle of the preparatory stage, after (lb) in the begin- 
ning of the body right after the initial stage, after (lc) in 
the middle of the body, after (ld) in the resulting stage 
right after the final stage, and after (le) in the resulting 
stage. If, by the end of the sentence, the RT is posi- 
tioned inside the body of the h-token described by the 
sentence I will say that the sentence has the imperfective 
aspectuai perspective; if the RT is positioned after the 
final stage or, more rarely, before the initial stage, I will 
say that the sentence has the perfective aspectual per- 
spective. In Example 1, the initial when clause with a 
verb in the simple past leaves the aspectual sentence 
perspective underdetermined, which is what I meant by 
weak aspectual typing of English. 
Note that the RT never ends up in the middle of the 
initial or final stages, because they are too short, much 
shorter than the body; they belong to a different time 
scale. This does not mean that they are necessarily 
instantaneous in an absolute sense: they are simply 
shorter than the body by at least a constant factor, and 
in a different sentence they can still be presented 
imperfectively. Consider Example 2: 
Example 2 
a. We spent the whole of last week working on 
Bob's project. 
bl. In the beginning... 
b2. As we began working on it... 
b3. When we were just beginning our work... 
The verb begin followed by an infinitive or gerund is a 
standard way of referring to the initial stage of a history. 
(Sometimes a lexical item is available as well or instead: 
"to begin to be asleep" is idiomatically expressed as fall 
asleep.) The second sentence of Example 2 thus puts 
the RT in the middle of the initial stage of the h-token 
described by the first sentence. Note that if in (2a) we 
replace the word week with month, year, decade, or 
night, the duration of the beginning varies with the 
duration of the whole while remaining much shorter 
than the body. 
The fact that we can so "zoom in" on different stages 
of a history has led many to conclude that there is no 
absolute sense in which events can be described as 
instantaneous. As Allen and Kautz 1985:253 puts it, 
"we can always 'increase the magnification' and find 
more structure" (see also Dowty 1986, Kamp 1979). I 
believe that instantaneousness is an absolute quality 
determined by our biology: instantaneous, events are 
those that are not perceived by humans as possessing 
internal structure. Languages select such events for 
special treatment by disallowing the "imperfective de- 
scription" of them: one cannot use the imperfective 
aspect to place the RT in the middle of an instantaneous 
event, so that The light was flashing does not place the 
RT inside an individual flash. Nor can the begin plus 
gerund construction refer to the initial stage of an 
instantaneous event: compare He began crossing the 
street with He began blinking. The initial stage, the 
body and the final stage of an individual blink or flash or 
Computational Linguistics, Volume 14, Number 2, June 1988 31 
Alexander Nakhimovsky Asl,ect, Aspectual Class, and the Temporal Structure of Narrative 
any other instantaneous event are thus inaccessible for 
future focusing, and it makes sense to represent them as 
all collapsed together. 2 
Note that the class of instantaneous events can be 
defined from three different perspectives corresponding 
to the three varieties of temporal knowledge postulated 
in the beginning of this section. Durationally, instanta- 
neous events belong to a time scale that is below a 
certain threshold of human perception of time (cf. 
Fraisse 1984:29-30). Compositionally, they have impov- 
erished internal structure. Aspectually, this shows up in 
imperfective forms of verbs describing instantaneous 
events, which are interpreted differently than such 
forms of verbs describing noninstantaneous histories. 
3 KNOWLEDGE ABOUT DURATIONS 
Durations can be entered in the lexicon in the following 
three ways that are not mutually exclusive: 
a. For some h-types (e.g., lecture, shower, lunch) we 
know the duration of their h-tokens to be stable, 
albeit approximate. Such durations can be entered 
in the lexicon directly as a fuzzy number, e.g., 
lecture \[1 2 hour\] (cf. Allen and Kautz 1985). This 
is a simple and least interesting case. 
b. For most aspectual classes of histories, duration is 
constrained by qualitative functional dependen- 
cies (Forbus 1985) among the participants of the 
situation; so, the time it takes to read a text 
depends on its length and genre, and the profi- 
ciency of the reader. This case is discussed in 
Section 4, after the notion of aspectual class is 
introduced. 
c. For a majority of h-types, constraints on their 
durations can be expressed using the notion of 
time scale, introduced below. 
In order to measure durations one needs cyclical 
events. These may be celestial, biological, or man- 
made. (See Campbell 1986 for a lively discussion.) 
Time-measurement units provided by cyclical events 
form several interlocking, multitiered hierarchies. It is 
commonly assumed in AI literature that the choice of a 
measurement unit is arbitrary and independent of the 
interval to be measured, because we can always switch 
from one unit to another by using axioms of the sort 
hour = 3600 seconds. Mathematically, of course, this is 
true, but I don't think language quite works this way: 
when we say It took Bob an hour to repair the faucet, 
we don't mean that it took him 3600 seconds. Given an 
interval, we choose a unit to measure it with so that the 
duration is expressed by a "reasonable" number, that 
is, not too big. Small fractions are also avoided: we say 
half a month or a quarter of an hour, but not one-fifth, 
-sixth, and so on. (Note that a quarter is the smallest 
fraction for which there is a special word; the names for 
the rest of them are formed by a regular morphological 
process.) Under these (psychologically) very under- 
standable restrictions it is rarely the case that an inter- 
val fit ~ \]precisely into the measurement unit we use. This 
results in a certain degree of fuzziness: when we say 
that something happened a year ago, we typically don't 
claim to be precise to the day. The same imprecision is 
observed when we use man-made time units such as 
hours minutes, and seconds. The sentences Joe slept 
for th,ee hours, or Joe slept from 2 to 5 do not claim 
precision to the minute; in fact, the second sentence can 
be continued with He woke up at a quarter after 5 
feeling greatly refreshed, and the narrative does not 
become self-contradictory. 
These observations suggest that a duration expressed 
by a number in hours becomes expressed by a fuzzy 
number (an interval) in minutes: \[3 3 hour\] becomes 
something like \[160 200 minute\]. (In a richer theory, the 
degree of fuzziness would depend on what we know 
about the source of information.) Since the boundaries 
of a fuzzy interval are themselves fuzzy, it might seem 
that we ought to throw in some seconds at the ends of 
the \[2,50 300\] minutes interval. However, the duration 
would then be expressed in thousands of seconds, and 
commonsense reasoning is too qualitative to deal with 
such big numbers. The length of the interval suggests a 
measurement unit, which, in turn, establishes a certain 
scale, and durations that are much smaller than this 
scale are disregarded. 
It tiros makes sense to talk of units of measurement 
that a:'e "natural" for a given interval: measured in a 
natural unit, the length of the interval will not be a very 
small :fraction (greater than some constant R) or a very 
big nttmber (less than some constant N). I wouldn't 
want to commit myself to specific values for R and N, 
but psychologists seem to have a fairly clear idea of 
what they are (see, e.g., Fraisse 1984). The time scale of 
an interval can then be characterized by the sequence of 
its nat aral units, a subsequence of all the duration units. 
An interval's duration can be represented by fuzzy 
numbers, e.g., \[1 1 year\] "a year, about a year"; \[1.5 1.5 
week\]. "a week and a half"; \[2 3 day\] "two to three 
days" and so forth. There are rules that transform fuzzy 
durations in one unit into fuzzy durations in a smaller 
unit, but only as long as both units are natural to the 
interval being measured: \[1.5 1.5 week\] becomes \[9 12 
day\], but not \[200 260 hour\]. 
To .zapture the intuition that hours are simply irrele- 
vant when we talk about weeks, or days when we talk 
about years, I introduce the relation much-greater-than, 
which is readily defined in terms of natural units: i is 
much greater thanj if and only if (iff) the largest natural 
unit ofj is smaller than the smallest natural unit for i. 
There are thus two ordering relations defined on dura- 
tions :neasured in the same scale: greater-than and 
much-greater-than; together, they give rise to an equi 
(indistlnguishability) relation (cf. Hobbs 1985, 1986b) 
defined as follows: two intervals' durations are equi iff 
the smaller of the two durations is much greater than the 
difference between them. Note that in this scheme of 
32 Computational Linguistics, Volume 14, Number 2, June 1988 
Alexander Nakhimovsky Aspect, Aspectual Class, and the Temporal Structure of Narrative 
things the equi relation is derived, rather than primitive 
as in Hobbs's work, and it is not incompatible with 
greater-than. I find it both convenient and intuitively 
correct that we can treat two intervals as equivalent 
without losing the knowledge, if we have it, that one is 
greater than the other. 
Note that the notion of natural units presented here is 
related to McDermott's (1982) lifetimes: saying that the 
natural unit for property p is year means that the 
property lasts (or persists) for years. It is also related to 
Hobbs's (1985a) grain. What I propose is that lifetimes 
and grains are monotonically related: the greater the 
lifetime, the greater the grain. Thus the same parameter, 
natural unit, can capture both. 
Duration units cut up a continuous parameter (time) 
into a finite set of quantities. The set of duration units, 
with the greater-than and much-greater-than relations 
on it, thus forms a quantity space for time, in the sense 
of Forbus 1985:96-7. We can communicate because our 
reasoning systems employ the same temporal quantity 
space, whose elements are arbitrary in precisely the 
same sense in which linguistic signs are arbitrary: each 
one of them could have been different, yet none of them 
can be changed by the individual speaker/reasoner, who 
acquires them in the process of cognitive and linguistic 
development. The same progression of ideas applies, 
with some modifications, to other one-dimensional sca- 
lars: distance, temperature, price, height/tallness, and 
speed. 
Given the role of time-measurement units in the 
proposed framework, the task of axiomatizing the clock 
and the calendar, mentioned in Hobbs 1985, becomes 
highly important. It involves identifying the time-mea- 
surement units, defining them in terms of cyclical 
events, and establishing the much-greater-than relation 
on them. There is some discussion of the task in 
Nakhimovsky 1986; here I can give just one illustration 
of the relevant empirical arguments. Apart from mea- 
surement units with single word names, there are mul- 
tiples of other units, described by phrases, that have 
themselves become measurement units. Five minutes 
provides an example: we talk about 10 or 15 or 20 
minutes much more commonly than 11 or 14 or 17 
minutes, and when asked to estimate a duration that is 
less than an hour we typically use a multiple of five 
minutes. (Note that this is not true about hours or days.) 
If we don't, an unusual precision is implied; compare: 
Example 3 
a. I'll be back in 15 minutes. 
b. I'll be back in 14 minutes. 
There's much more of a broken promise if I'm one 
minute late after (3b) than (3a), which indicates a finer 
grain (smaller scale) in (3b). 
Clearly, five minutes have become a unit because a 
minute is a unit, and we have five fingers on each hand. 
This is yet another example of the anthropocentrism 
pervading language and commonsense reasoning, the 
same antropo-centrism that was mentioned earlier in 
connection with instantaneous events. 
4 ASPECT 
4.1 GRAMMATICAL ASPECT 
In what follows it is essential to keep the following three 
concepts apart: aspect as a grammatical category of the 
verb, implemented by affixes, auxiliaries, and such; 
aspectual class, which is a characteristic of an h-type or 
lexical meaning; and the aspectual perspective of the 
sentence determined by the position of the RT with 
respect to the h-token described by a finite clause. Thus 
grammatical aspect is associated with linguistic expres- 
sions, aspectual class with h-types, and aspectual per- 
spective with h-tokens. It is the aspectual perspective of 
the sentence that is ultimately important for understand- 
ing; both grammatical aspect and aspectual class some- 
times uniquely determine, sometimes just strongly con- 
strain, the aspectual perspective. In English, for 
instance, the progressive aspect guarantees that the 
sentence perspective is imperfective, except for sen- 
tences describing instantaneous events, which, in any 
language, cannot be presented imperfectively. (As the 
discussion in the end of Section 2 put it, He was blinking 
or He started blinking does not put the RT in the middle 
of an individual blink.) This does not mean that verbs 
like blink are grammatically perfective. 
There are languages, most notably Slavic, where the 
difference in sentence perspective is hard-wired into 
verb morphology: simplifying very slightly, every Rus- 
sian verb is either perfective or imperfective (with the 
diagnostics being that only imperfective infinitives can 
follow verbs like begin or continue), and the morpho- 
logical feature of the verb determines the aspectual 
perspective of the sentence. In English, by contrast, 
simple past is fairly noncommittal as to the aspectual 
perspective of the sentence, as we saw in the discussion 
of Example 1. Another way to describe this contrast is 
to say that Russian has a perfective-imperfective 
aspectual system, while English has an imperfective- 
unmarked aspectual system; most aspectual systems 
seem to belong to one of these two kinds, with about 
three-to-one advantage for the perfective-imperfective 
systems in Dahl's (1985) analysis of 64 languages. 
A fairly common misconception is to assume that 
since English has an imperfective, it ought to have a 
perfective also. Obviously, simple past, without any 
further qualifications, cannot be considered perfective, 
in view of such obvious examples as Bob owned a house 
on Cape Cod. However, sentences like Vanessa read a 
magazine article about Mongolia do suggest the perfec- 
tive perspective: by the end of the sentence the article 
seems to have been read (cf. Hinrichs 1986:68; Dowty 
1986:46-8 for similar examples and discussion). I argue 
that it would be incorrect to call the verb or the verb 
phrase in such sentences perfective, because the per- 
fective sentence perspective emerges not so much from 
Computational Linguistics, Volume 14, Number 2, June 1988 33 
Alexander Nakhimovsky Asp ~ct, Aspectual Class, and the Temporal Structure of Narrative 
the linguistic properties of the verb phrase as from 
commonsense knowledge about the situation it de- 
scribes: a short process that progresses towards a 
well-defined terminal point beyond which it cannot 
continue. To see that we are not dealing with a gram- 
matical category but rather commonsense inference, 
observe that an appropriate context can change the 
sentence perspective, as in Example 4, where Va- 
nessa's reading is interrupted by a doorbell: 
Example 4 
After supper, Vanessa and Didi sat down in the 
living room. Vanessa read a magazine article 
about Mongolia, Didi watched her favorite car- 
toon on television. Suddenly the doorbell rang. 
I conclude, in accord, I believe, with Passonneau (1987 
also this volume), that English has no morphological 
perfective; the perfective effect arises out of the aspect- 
ual properties of the history being described, i.e., its 
aspectual class. 
4.2 ASPECTUAL CLASSES OF HISTORIES 
An aspectual classification divides lexical meanings 
according to the internal temporal structuring of the 
corresponding h-tokens. I believe they should be 
thought of as intrinsic properties of lexical meanings, 
independent of context and, in particular, aspectual 
perspective. Even out of context we know a good deal 
about how instances of, e.g., reading or sleeping or 
walking to the store are individuated and how they 
unfold in time. This knowledge interacts with, but is not 
determined by, whether the h-token is presented imper- 
fectively, as in John was walking to the store. 
The major division among noninstantaneous histo- 
ries, recognized at least since Aristotle, is between 
process (energeia) and state (stasis). In recent times, 
Vendler (1967) proposed a highly influential classifica- 
tion that is still commonly accepted, although the prin- 
ciples of classification have changed. Vendler believed, 
erroneously, that he was classifying English verbs, 
rather than their denotations, and he used such lan- 
guage-specific criteria as whether or not a verb has a 
progressive form (Vendler's statives, such as know, 
don't). In the Taylor-Dowty model-theoretical version 
(see, e.g., Dowty 1986), the classification is based on 
the relationship between the truth value of a sentence at 
an interval and at its subintervals. So, for instance, a 
sentence S is stative (denotes a state) iff it follows from 
the truth of S at an interval I that S is true at all 
subintervals of I (the so-called subinterval property, cf. 
Dowty 1986:42). 
I submit that these criteria cannot possibly capture 
the real distinctions operative in the workings of human 
language: these have to relate to something perceived 
and experienced, rather than truth values (which is not 
to deny that real distinctions may result in fairly con- 
sistent truth-functional properties). It is not accidental 
that Dowty's own example of a state (sleep) contradicts 
his definition: we can truthfully say that Bob slept from 
10 to ~ even if he got up once to go to the bathroom. My 
propo:;al is that we take the physical vocabulary of 
proce.,ses and states seriously, and classify histories 
accorcLing to their internal dynamics, the stability of 
their parameters, and the resources they consume. (Part 
of the internal dynamics, in the presence of a conscious 
agent, is the degree of required attention and volitional 
control.) We can then note the distinction between 
states that do not require any resources to sustain 
themselves (knowing English, owning a house) and 
those :hat do (sleep requires a certain amount of sleep- 
iness, which gradually wears out). The subinterval 
propel~ty holds only for zero-resource states and is, in 
fact, a simple consequence of their other properties: a 
state t aat requires no resources, including attention and 
conscious control, obtains continuously. 
An important group of resources are those that are 
consumed by the unconscious and periodic metabolic 
proce~,ses sustaining a living organism. I call them 
"genetic resources" because they are not specific to 
any particular history, but rather to all individuals of a 
given sort. Resource-consuming states all seem to re- 
quire only generic internal resources, while within 
processes, there are those that require only generic 
resources (walking) and those that require process- 
specific resources as well: reading, for example, re- 
quires not only being awake and not too hungry, but 
also a text to read. This is the first of the two important 
divisicns among processes. The other one separates 
telic processes from atelic ones. 
Telic processes can be defined as processes that have 
a buill-in terminal point that is reached in the normal 
course of events and beyond which the process cannot 
continue. Two most common, and overlapping, varie- 
ties of relic processes are: human activities directed to a 
specific goal, and processes that consume a specific 
amount of a process-specific resource. So, reading a 
magazine article is a telic process not only because 
reading is a goal-directed activity, but also because an 
article contains a limited amount of the process-specific 
resource (reading matter) that gets steadily consumed as 
the process unfolds. Temporal information about a telic 
process thus includes a typical rate of progress towards 
its goal or a typical rate of consumption of its resource. 
A comparison of the preparatory and the resulting 
stages of a telic process shows the possible changes that 
such processes bring about: a telic process can create an 
object, destroy an object, modify an object, or move a 
specified amount of material (possibly the mover him- 
self) to a specified destination. A subclass of destruction 
processes are ingestions, which convert an external 
resource into an internal one. Moving is understood to 
include all three of Schank's PTRANS, ATRANS, and 
MTRANS classes (Schank 1975). Moving also includes 
gradual (but not instantaneous) changes of state. 
It has been observed (in Verkyul 1972 and others 
after him) that the aspectual class of a telic predicate is 
34 Computational Linguistics, Volume 14, Number 2, June 1988 
Alexander Nakhimovsky Aspect, Aspectual Class, and the Temporal Structure of Narrative 
affected, in regular ways, by the properties of its 
noun-phrase (NP) complements, especially theme. Such 
distinctions as plural-singular, group-individual, and 
mass-count are particularly important, as well as the 
degree of granularity-homogeneity and internal struc- 
ture. To take a well-known example, Bill crossed the 
border describes an instantaneous event; The regiment 
crossed the border describes a noninstantaneous h- 
token with a beginning, body, and end; and Drug 
smugglers crossed the border easily describes an activ- 
ity that continues as long as there are drug smugglers. 
The aspectual properties of telic h-tokens thus arise 
compositionally in context; 3 however, it still makes 
sense to talk of the aspectual clas s of a telic h-type as it 
manifests itself in h-tokens with atomic individuals as 
themes. 
4.3 EVENTS, PROCESSES, AND STATES 
Now I briefly comment on how the conceptual frame- 
work developed here relates to a classification of histo- 
ries into events, processes, and states (Mourelatos 
1981). For a linguist, the distinction event-process is one 
of aspectual perspective: "The term 'process' means a 
dynamic situation viewed imperfectively, and the term 
'event' means a dynamic situation viewed perfectively" 
(Comrie 1976:51). The distinction process-state is one of 
aspectual class. Thus the sentence John was walking to 
the store describes a process, not a state, because 
walking in general, and walking to a specified destina- 
tion in particular, is intrinsically a process; it describes 
a process and not an event because the h-token is 
viewed imperfectively. I think the lexicon can be made 
more consistent and informative if the notion of aspect- 
ual class (lexical aspect, Aktionsart) is kept separate 
from the aspectual sentence perspective. 
An important piece of evidence for keeping the 
distinctions event-process and process-state separate is 
provided by events "made out of" states such as the 
event described by Bobby took a nap. (Cf. also did some 
reading, went for a walk.) Collocations of this nature 
have never, to my knowledge, been discussed in con- 
nection with the English aspectual system, precisely 
because, in English, it is easy to associate telicity (an 
intrinsic lexical property) and perfectivity (the value of 
a subjective grammatical category). In languages with a 
well-articulated perfective aspect a stretch of most 
atelic activity can be viewed perfectively, i.e., as an 
event. For instance, He walked (perfective), clearly 
atelic, would mean something like "He walked for a 
while/he took a walk"--clearly an event. Atelic histo- 
ries, of course, lend themselves more readily to the 
imperfective view; Nakhimovsky 1983:80-2 documents 
some subtle ways in which "aspectual" languages dis- 
tinguish between atelic events, which are events simply 
by virtue of allocating them a limited amount of time, 
and telic events, which are intrinsically bounded in time 
by the very manner in which they unfold. Telic events 
tend to be identified as a natural class because they 
support perhaps the most common temporal inference 
we make: given a telic history viewed perfectively, we 
infer that the end point is reached. (This is made more 
precise in Section 4.5.) For all that, telicity and perfec- 
tivity remain distinct, if not entirely orthogonal, 
categories .4 
4.4 A CLASSIFICATION OF INSTANTANEOUS EVENTS 
Unlike noninstantaneous histories, instantaneous 
events lack internal structure and have to be classified 
by comparing the world before and after them. An 
instantaneous event can terminate either a process or a 
state, and it can initiate either a process or a state; if it 
is sandwiched between two processes or two states, the 
two can be the same or different. The resulting classi- 
fication is shown in Table 2, next page. It can be seen 
that in English, most verbs describing instantaneous 
events fall into the first three groups, where the instan- 
taneous event meets a state. Sometimes the name of the 
process can be used to describe its beginning, as in She 
ran, i.e., "set out running." 
Among the three groups of instantaneous events, 
happenings always have a progressive form, and they 
always describe a sequence of instantaneous h-tokens. 
Transitions may or may not have a progressive form; if 
they do, progressive is again associated with plurality, 
e.g., He's been noticing that... (the implication being 
that the noticing has taken place on many occasions. 
Observe the inference that between the occasions there 
is a return to the initial state of the transition: in order to 
again notice, i.e., become aware of something, one has 
to first cease to be aware of it.). Finally, culminations 
usually have progressive forms that allow two interpre- 
tations. One is the familiar idea of plurality: The trav- 
elers, one by one, were reaching the summit. The other 
arises from the nature of culminations, which crown 
processes with sharply defined end points: the top of a 
climb, the finish of a race. The progressive of a culmi- 
nation verb describes an h-token whose lifetime is an 
open interval properly contained in the time of the 
process that leads to the instantaneous event. I con- 
clude that the proposed classification, although based 
on extra-linguistic considerations, captures salient dis- 
tinctions in the meaning of linguistic forms. 
All the classifications developed in this section are 
brought together in tables 1-3, next page. 
4.5 ASPECTUAL CLASS, SENTENCE PERSPECTIVE, AND 
INFERENCE 
Recall that the aspectual sentence perspective is deter- 
mined by the position of the RT, either inside the time 
of an h-token or after that. In interaction with the 
knowledge about the internal composition of h-tokens, 
the position of the RT creates expectations, which can 
be stated using the notions of time scale and expended 
resources. Consider first an example: the h-token of 
Sharon reading a novel can terminate in one of the 
following four ways: 
Computational Linguistics, Volume 14, Number 2, June 1988 35 
Alexander Nakhimovsky Aspect, Aspectual Class, and the Temporal Structure of Narrative 
ASPECTUAL CLASSES 
POSSESSING INTERNAL STRUCTURE 
NOT POSSESSING INTERNAL State Process 
STRUCTURE (instantaneous) \[ 
Atelic Telic 
\[ SAME, VIEWED FROM INSIDE 
Table 1. Top-level classification. 
Configuration Label Examples 
state-E-same state happening flash, cough, jump 
state-E-different state transition forget, recognize, notice, 
process-E-state culmination win the race, reach the top 
process-E-same process disturbance examples 
state-E-process activation more difficult 
process-E-different process switch to find 
Table 2. Instantaneous events. 
STATES PROCESSES II 
atelie: atelic: I telic 
generic resource specific resource I 
zero-resource \] generic resource 
I 
relations: 
own sleep 
possess stand 
resemble sit 
perceptions: lie 
see hold 
hear lean against 
feel 
mental states: 
know 
remember 
trust 
walk 
run 
think(of) 
work 
read read a book 
write write a letter 
build build a chair 
dig dig a hole 
shave 
Table 3. Noninstantaneous h-types. 
natural end: Within a certain time scale, Sharon 
finished the novel; 
exhaustion: Within a certain time scale, determined 
by a different set of resources, Sharon got tired of 
reading; 
volition: Sharon decided to put the novel down; 
interruption: Something happened that made reading 
impossible: Sharon's friend took the novel away 
from her, the lights went out, her attention was 
distracted by loud music. 
Generalizing from this example, we can say that when 
the RT finds itself inside an h-token, some or all of the 
following expectations are set up, depending on the 
internal dynamics of the h-token and the presence or 
absence of volitional control: 
• If the h-token has an Agent participant, it can be 
interrupted by the Agent's decision to do so. (The 
decision may turn out to be impossible to implement 
but the h-token is then transformed into a different 
one, in which the former Agent perhaps becomes an 
Object or Patient: think of a driver trying to stop the 
car and going into a skid.) 
• If the h-token consumes generic resources (i.e., if it is 
a process or a resource consuming state) then, within 
a certain time scale, it comes to an end when the 
generic" resource is exhausted. 
• If the h-token is a telic process, then, within a certain 
time scale, the process will come to its natural end. 
• Finally, and most difficult to detect, the h-token may 
come to an end if some of its preconditions no longer 
hold, causing an interruption of the h-token. 
The causality involved in interruption may be indirect, 
and a complete specification of everything that can go 
wrong is clearly impossible, but no such completeness 
is observed in human reasoning, either. What we can 
and ought to do is list in the lexicon the main precon- 
ditions for processes and states expressed by lexical 
items. Assuming that the lexicon is organized as an 
inheritance hierarchy, a great deal of preconditions will 
be shared by entire classes of histories. Some examples 
are given in Nakhimovsky 1987a. 
The most noticeable expectations are those arising 
when a telic process is viewed imperfectively, and the 
logical problems of reasoning with expectations were 
first observed on such processes. This is Dowty's (1977) 
"imperfective paradox", i.e., the problem of giving "an 
account of how \[John was drawing a circle\] entails that 
John was engaged in bringing-a-circle-into-existence 
activity but does not entail that he brought a circle into 
existence" (p.46). From the perspective of the interven- 
ing years it can be seen that Dowty here grapples with 
the default, nonmonotonic nature of the inferences 
involved. The progressive sentence puts the RT in the 
middle of the time of John's activity; if another sentence 
puts the RT after that interval, we infer, unless or until 
told otherwise, that the activity came to its natural end. 
More precisely, the inference works as follows: 
Telic-lmperfective inference rule: If one sentence in a 
narrative tells us that a telic history hi has begun or 
is in progress, and a later sentence moves the RT into 
or beyond a later history h2, and the time distance 
between hl and h2 is of the same or a greater scale 
than the time scale ofhl, then assume, unless or until 
told otherwise, that the end-point of hl is reached. 
The rule can be illustrated by the following narrative: In 
the afternoon, when Johnny was writing a postcard to 
his friend, Mommy went to the market. In the evening 
they played together. Knowing how long it takes, we 
infer that Johnny finished his postcard and Mommy her 
shopping. However, the inferences are withdrawn if 
contradicted: the narrative could continue with until 
Mommy discovered that the postcard never got written. 
To see the importance of time scale, replace "writing a 
postcard" with "reading War and Peace." 
4.6 DURATIONAL CONSTRAINTS 
The examples of the preceding section illustrate two 
facts about temporal reasoning that I believe are crucial 
for any implementation of narrative understanding: (a) 
knowledge about internal constituency of events and 
temporal relations between them is tightly integrated 
36 Computational Linguistics, Volume 14, Number 2, June 1988 
Alexander Nakhimovsky Aspect, Aspectual Class, and the Temporal Structure of Narrative 
with durational knowledge; (b) durational knowledge 
includes a great many constraints originating in celes- 
tial, biological, and civilization-specific cycles. The first 
of these facts is recognized and implemented in Allen 
and Kautz 1985. The implementation suggestions made 
here all derive from that paper and Forbus 1985. Allen 
and Kautz also note that "in more complicated situa- 
tions, certain units (such as the standard time units-- 
minutes, hours, weeks, etc.) can be used as reference 
durations to keep the network of manageable size" (p. 
260). How this ties in with (b) above can be seen by 
recasting Allen and Kautz's example of a unified (tem- 
poral relations + durations) implementation in the 
present framework. The example in its entirety is as 
follows (pp. 260-1): 
Example 5 
a. Moe and Larry began reading Principia Mathe- 
matica. 
b. Moe read for over an hour. 
c. Larry stopped reading and fell asleep after 10 
minutes. 
Allen and Kautz interpret (5a) to mean that Moe and 
Larry started reading simultaneously (and may or may 
not have finished together). But surely, language is not 
that precise: (5a) can be said about two students who 
both started reading the book in September for a course 
they were taking in the fall semester. The most one can 
say about (5a) is that the time distance between \[began- 
read Moe\] and \[began-read Larry\] is much-shorter-than 
the expected duration of reading. This is an instance of 
a general rule, noted in the end of Section 2, that the 
duration of an h-token is much-greater-than the dura- 
tions of its beginning and end. 
Sentence (5b) supplies both a duration and a time 
scale. Allen and Kautz's representation again under- 
constrains: they interpret the sentence to mean that the 
duration of Moe's reading is (1 ~). An obvious con- 
straint is that Moe read for less than a day, but more 
precisely "over an hour" implicates "less than two 
hours"; in general, "over/after n units of time" impli- 
cates "less than n + 1 units." (This is an instance of 
Horn's (1984) scalar implicature.) Allen and Kautz, in 
fact, make an implicit use of this implicature when they 
interpret (5c) to mean that Larry read 10 to 15 minutes 
(p. 261). The interpretation is correct because 5 minutes 
is a duration-measurement unit in our language, as 
mentioned in Section 3 above. 
It appears that the varieties of temporal knowledge 
proposed in this paper result in a richer set of con- 
straints that give a more adequate picture of (5) and can 
handle more complex narratives. They also indicate a 
fairly clear connection between the task of natural 
language understanding and the enterprise of naive or 
qualitative physics (Hayes 1985, 1985a; Forbus 1985). 
The entire aspectual classification of histories is based 
on taking seriously the physical vocabulary of processes 
and states, and the reasoning of the preceding section is 
an example of Forbus's (1985) limit analysis. In general, 
nonmonotonic constraint propagation over interacting 
processes evolving in time is the essence of naive 
physics. These similarities suggest that qualitative proc- 
ess descriptions of Forbus 1985 can serve as a model for 
lexical entries describing h-types. (Some examples can 
be found in Nakhimovsky 1987a.) 
5 NARRATIVES AND THEIR REPRESENTATIONS 
Discussions of tense and aspect in discourse invariably 
seem to converge on narrative discourse, rather than 
conversation or exposition. The reasons, I believe, have 
to do with several profound, if somewhat speculative, 
distinctions: between semantic and episodic memory; 
between on-line perception of time effected by biologi- 
cal clocks and detached time cognition rooted in long- 
term memory; and between subjective and objective 
discourse. Briefly and, I repeat, speculatively, exposi- 
tion contributes to semantic memory while narrative 
builds an episodic representation; conversation is a 
breed apart because the time of a conversation coin- 
cides with the time of its content, with simultaneity in 
time typically accompanied by a tight integration be- 
tween the linguistic and nonlinguistic behavior: the 
verbalization of how-to-get-there directions is the action 
of giving directions; task-oriented conversations be- 
tween an expert and apprentice are an integral part of 
performing the task at hand, and the unfolding text of an 
argumentative dialogue is precisely the activity of argu- 
ing. Conversation can thus properly be called performa- 
tire discourse. By contrast, the content of a narrative is 
decoupled from the linear progression of its text and 
unfolds in its own, separate timeline. 5 
Whatever the reasons, it is certainly the case that (a) 
tense, aspect, and mood systems of the verb are fre- 
quently more finely nuanced in the past than in the 
present or future, and never the other way around; (b) 
the tense of narrative discourse is past, in its many 
varieties; (c) the past tense in narratives is not inter- 
preted with respect to the situation of discourse. In 
place of the situation of discourse, a narrative is proc- 
essed with respect to a constantly maintained deictic 
center, which is "the locus in conceptual space-time of 
the objects and events depicted or described by the 
sentences currently being perceived. At any point in the 
narrative, the cognitive agent's attention is focused on 
particular characters (and other objects) standing in 
particular temporal and spatial relations to each other. 
Moreover, the agent 'looks' at the narrative from the 
perspective of a particular character, spatial location 
and temporal location. Thus the deictic center consists 
of a WHERE-point, a WHEN-point and a WHO point." 
(Bruder et al., 1986:1). In this paper, the WHEN-point 
of the deictic center is referred to as the temporal focus, 
or TF (cf. Webber 1987a,b; Nakhimovsky 1987b). This 
term is adopted because it brings out the similarity 
Computational Linguistics, Volume 14, Number 2, June 1988 37 
Alexander Nakhimovsky Aspect, Aspectual Class, and the Temporal Structure of Narrative 
between definite and temporal anaphora (cf. Webber, 
this volume, for the genesis of this idea.). 
Not one but two structures are thus needed to 
represent a narrative: the event-situation structure (ESS, 
cf. Webber 1987b), representing the narrative's unfold- 
ing contents, and the linear text structure (LTS), whose 
components are linked by rhetorical relations such as 
elaboration, resumption, or flashback (see, e.g., Hobbs 
1982). Next to ESS and LTS, there must also be a 
short-term memory structure called current focus space 
(CFS) and a buffer in which the representation of the 
current sentence is assembled. Among the components 
of the CFS are the data structures needed for compre- 
hension of definite anaphora (such as the Focus Stack 
and Alternate Focus List in Sidner 1983), as well as the 
components of the current deictic center, which are 
really pointers to the appropriate nodes of the ESS. 
After the first sentence of a narrative establishes the 
initial set-up, each subsequent sentence is processed in 
the context consisting of the ESS and the CFS. Depend- 
ing on the sentence's meaning, one of two things 
happen: either the representation of the sentence is 
incorporated in the CFS, with the focusing mechanisms 
appropriately modified; or, in the case of a focus shift, 
the contents of the CFS are incorporated into the ESS 
and LTS, and the CFS is completely reset. In terms of 
text structure, the current sentence either continues the 
same, or starts a new, discourse segment (DS; cf. Grosz 
and Sidner 1986). 
The nature of the processing at the DS juncture is 
thus quite different from the "routine" tasks to be 
performed as long as the text remains in the same DS: 
the start of new DS prompts, and is prompted by, a shift 
of attention. The circularity here is deliberate. The start 
of a new DS brings about several changes, some of them 
more immediately noticeable than others; the more 
obvious ones serve to signal that a new DS is, indeed, 
started. Some of the most important such signals come 
from changes in time scale, aspectual class, and other 
temporal characteristics of narrated h-tokens. These 
temporal discontinuities are usually accompanied by 
discontinuities of other kinds that form recurrent pat- 
terns; I suspect, and this is an empirical hypothesis, that 
rhetorical relations within a narrative text are macrola- 
bels that stand for certain oft-repeated clusters of dis- 
continuities. In any event, the task of segmenting nar- 
ratives into units that form the basic elements of 
discourse representations and constrain the application 
of focusing algorithms is obviously very important. The 
task is frequently mentioned in work on discourse (see 
Grosz and Sidner 1986:177-8 and references there), but 
the available heuristics mostly rely on clue words and 
phrases. I propose heuristics based on four kinds of 
discontinuities: discontinuities of topic, discontinuities 
of space and time, discontinuities of figure and ground, 
and discontinuities of the narrative perspective (not to 
be confused with aspectual sentence perspective). I 
explain these after a preliminary and partial illustration. 
Consider' Example 6. 
Example 6 
a. Hartley and Phoebe had been sent by their 
mother to fix the tail vane of the windmill. 
b. In the great expanse of the prairie where they 
lived, the high tower of the windmill was the 
only real landmark. (Worline 1956:1) 
Rhetorically speaking, the (b) sentence above interrupts 
a sequence of events (DS1) to start DS2, a description. 
In order to recognize this rhetorical relation between 
the two DSs it is necessary to recognize that: 
a. there is a shift of topic; 
b. there is a shift in perceptual modality to visual 
perception; 
c. there is a shift in time scale from the events of 
the current day to years or decades, associated 
with the lifetime of a windmill and the 'where 
they lived' clause; 
d. there is a shift in spatial scale from a household 
to the entire prairie; 
e. there is a shift from a foregrounded sequence of 
events to the 'mopping-up' operation of filling 
the background. The shift is signaled by changes 
in aspectual class. 
The following empirical investigation thus suggests 
itself: classify the discontinuities and clusters of discon- 
tinuities that typically accompany DS breaks in narra- 
tives; identify the linguistic and extralinguistic knowl- 
edge involved; develop heuristics for using this 
knowledge; and test the heuristics in a computer pro- 
gram. 6 This paper presents a classification that arises 
from an analysis of what constitutes a narrative. 
The first observation to make is that a narrative must 
have a \]plot, i.e., present a sequence of events that forms 
an instance of a recognizable pattern. (The patterns are 
part universal, part culture specific; the work in Lehnert 
1982 can be seen as a search for the principles on which 
such patterns are built.) Using the Gestalt terminology 
(brought into linguistics by Talmy 1983) I can say that a 
narrative's plot must present a recognizable temporal/ 
causal figure shown against some ground that minimally 
consists of spatial/visual settings (descriptions of char- 
acters are also frequent). The distinction is not always 
clear-cut because elements of the figure can be hidden 
among the details of the ground, but the temporal nature 
of the plot does stand in clear contrast to the spatial 
nature of the ground. 
Secondly, a narrative must have characters with 
which we empathize. These characters don't have to be 
human: one can easily imagine a story about an adven- 
turesome plant seed that falls off its parent, gets swal- 
lowed and excreted by a horse, and nearly drowns in a 
tropical rain before being miraculously saved by the sun 
and producing a flower. Even so, the narrative is likely 
to alternate between the objective narrator's point of 
38 Computational Linguistics, Volume 14, Number 2, June 1988 
Alexander Nakhimovsky Aspect, Aspectual Class, and the Temporal Structure of Narrative 
view and that of one of the characters. ("The belly of 
the horse was dark and noisy inside.") Some disconti- 
nuities of narratives very simply reflect the intrinsic, 
spatial, or temporal discontinuities of the plot, when, 
e.g., the story is composed of a sequence of events 
taking place during an afternoon, followed by two 
years' hiatus, followed by another action-packed after- 
noon. (A similar example of a spatial discontinuity can 
be easily imagined.) One measure of the "simplicity" of 
a narrative is how faithfully the order and structuring of 
its text reflects the order and structuring of its compo- 
nent events. Even in the simplest narrative, however, 
there are bound to be discontinuities resulting from the 
tension between the linear nature of the text and the 
multidimensional structure that it is meant to evoke. 
These are discontinuities of figure and ground, when the 
narrative shifts between the main story line and the 
surrounding circumstance, and discontinuities of per- 
spective, when the narrative crosses into a different 
"empathy space" or creates a new one. 
Given this classification of discontinuities one can 
proceed to catalogue the clues that signal them. This is 
a subject for a large empirical study, of which the next 
section is but a preliminary sketch. It is important to 
keep in mind that it is clusters of discontinuities that 
signal the beginning of a new DS. 
6 DISCONTINUITIES IN NARRATIVES 
6.1 DISCONTINUITIES OF TOPIC 
Discontinuities of topic fall in two groups. In the first, 
there is no anaphoric relation or immediate inference 
path from the new topic to a node in the CFS. What "an 
immediate inference path" is depends, of course, on the 
system's knowledge base and inferential capabilities, 
but this is a separate big issue that is not dealt with in 
this paper. The second kind of discontinuity, noted in 
Reichman 1985, arises when an anaphoric relation ex- 
ists and allows a pronominal anaphor, but the topic is 
instead reintroduced by a full noun phrase. I call this 
phenomenon topic reintroduction and illustrate it in 6.5. 
6.2 TEMPORAL DISCONTINUITIES 
The most important temporal discontinuities are: 
(a) A shift from perfective to imperfective sentence 
perspective accompanied by a shift to a much 
greater time scale. The corresponding rhetorical 
move is frequently characterized as "introducing 
background or descriptive material". The move 
is frequently accompanied by a topic reintroduc- 
tion. 
(b) The reverse shift from descriptive material to the 
main line of the narrative. This move is signaled 
by the temporal focus (TF), and the entire deictic 
center, returning to an established node in the 
ESS, with an appropriate contraction of the time 
scale. 
(c) A backwards move of the TF to an earlier point in 
time, with or without a change in time scale. 
Rhetorically, this is known as flashback. This 
move is frequently signaled by a verb in past 
perfect or by the used to + infinitive construc- 
tion, although a shift may occur without such a 
verb form, and the presence of such a verb form 
does not necessarily signal a shift: the reference 
time of the sentence may remain the same as, 
rather than precede, the current TF. (Section 7 
illustrates; see also Almeida 1987 and Webber 
1987b, this volume.) 
6.3 SPATIAL DISCONTINUITIES 
The only purely spatial discontinuities I have looked at 
are discontinuities of scale. Just as h-types have time 
scales associated with them, characterized in terms of 
"received" cyclical events such as day or year, it seems 
equally necessary to establish a gradation of spatial 
scales, based on similar considerations from human 
biology and habitat. (A hint at the same idea can be 
found in Nirenburg et al., 1985:230.) Some important 
spatial scales are: tiny, hand, one's body (may need to 
be further specialized); within arm's reach; room area, 
such as desk or bed; room; floor; house; household; 
village or neighborhood; larger area within a between- 
meals round trip; within a day's round trip; staying 
overnight. (The larger the scale, the more domain- and 
culture-specific variation there is.) 
6.4 DISCONTINUITIES OF PERSPECTIVE 
Two kinds of phenomena are clearly relevant: attitude 
reports, and indications of the spatial position from 
which things are perceived (the WHO-point in Bruder et 
al., 1986). The subject needs further study. 
6.5 AN EXAMPLE 
The remainder of this section shows how the above 
heuristics work themselves out on the initial fragment of 
Joyce's "Clay" (Joyce 1969:99-106). The fragment falls 
into three DSs: Sentence 1, sentences 2-6, and sen- 
tences 6-12; Sentence 13 starts a fourth DS by returning 
to the first one. Embedded in DS3 is DS3.1 consisting of 
sentences 9-10. 
1. The matron had given her leave to go out as soon 
as the women's tea was over and Maria looked 
forward to her evening out. 2. The kitchen was spick 
and span: the cook said you could see yourself in the 
big copper boilers. 3. The fire was nice and bright and 
on one of the side-tables were four very big barm- 
bracks. 4. These barmbracks seemed uncut; but if 
you went closer you would see that they had been cut 
into long thick even slices and were ready to be 
handed round at tea. 5. Maria had cut them herself. 
6. Maria was a very, very small person indeed but 
she had a very long nose and a very long chin. 7. She 
talked a little through her nose, always soothingly: 
Yes, my dear, and No, my dear. 8. She was always 
sent for when the women quarrelled over their tubs 
Computational Linguistics, Volume 14, Number 2, June 1988 39 
Alexander Nakhimovsky Aspect, Aspectual Class, and the Temporal Structure of Narrative 
and always succeeded in making peace. 9. One day 
the matron had said to her:--Maria, you are a veri- 
table peace-maker! 10. And the sub-matron and two 
of the Board ladies had heard the compliment. 1 I. 
And Ginger Mooney was always saying what she 
wouldn't do to the dummy who had charge of the 
irons if it wasn't for Maria. 12. Everyone was so fond 
of Maria. 
13. The women would have their tea at six o'clock 
and she would be able to get away before seven. 
The transition from Sentence 1 to 2 is very similar to the 
transition from (a) to (b) in Example 6: it is a shift from 
figure to ground marked primarily by aspectual changes 
and a shift from temporal to spatial/visual material. The 
change in time and space scales is not as dramatic here, 
but, on the other hand, there is a discontinuity of 
perspective as the story shifts from Maria's attitude 
report ("Maria looked forward") to the objective per- 
spective of the implied narrator ("The kitchen was 
spick and span... "). 
The transition from 5 to 6 (paragraph break) is 
characterized by a shift in time scale and a topic 
reintroduction. The material is a (back)ground character 
description: "Maria is such that .... " The beginning 
of DS3.1 is signaled by a well-known "clue phrase" 
One day and by the past perfect tense. Note that the 
material is still ground ("Maria is such that .... ") but 
the precise relationship between DS3 and DS3.1 is 
ambiguous: they can be siblings. Only when DS3 re- 
turns in Sentence 11 (the signals are the return to the 
simple past and the sentence-initial and) is it established 
that DS3.1 is, indeed, embedded in DS3. 
The transition from 12 to 13 (the next paragraph 
break) is a return to the event sequence of Sentence 1. 
The position of the TF, and the entire deictic center is 
indicated by the future-in-the-past tense. Note that it is 
essential to retrieve the entire deictic center, and not 
just the TF, because the WHO-point is also restored: 
the she in 13 does not evoke Maria of Sentence 12 or 
any other sentence in DS2 and 3 presented from the 
perspective of the implied narrator. Rather, this pro- 
noun is a quasi-indexical (Rapaport 1986) that replaces 
the first-person singular I of a portion of the narrative 
from Maria's perspective, signaled by Maria looked 
forward in Sentence 1. 
7 NARRATIVE MOVES 
It is a common intuition that, as a narrative progresses, 
TF changes its position in time. I will refer to the 
movement of TF from one sentence of the narrative to 
the next as the narrative move. Narrative moves can be 
classified along several dimensions. First off, the TF 
can remain in the same timeline or branch off to an 
alternative one. (Such branchings-off are usually sig- 
naled by a nonindicative mood of the verb; this paper 
has nothing more to say about them.) Within the same 
timeline, the TF can move forward, sideways (into an 
overlapping h-token), or backwards, to an earlier h- 
token. In each of these three cases, the move may or 
may not be accompanied by a change in time scale. 
Finally, the TF can remain within the current focus 
space or shift to a different one. (In terms of linguistic 
structure, the current sentence may continue the same, 
or start a new, DS.) I will coin the terms mieromove and 
macromove for the last distinction. 
The preceding section argued that, if considered in 
the context of the entire deictic center movements, 
macromoves can be fully catalogued and associated 
with a limited set of rhetorical labels. It appears even 
more likely that micromoves also form a small set of 
recurrent patterns, because they are additionally con- 
strained by the requirement that they cannot introduce 
a drastic discontinuity. I present a sample of micro- 
moves below, contrasting them with macromoves; the 
exposition is organized by the temporal classification of 
narrative moves (backward, sideways, or forward). 
Much of the material of this section overlaps with 
Webber (this volume); some of the examples are from 
Dowty 1986. 
7.1 BACKWARD MOVES 
Backward moves are always macromoves, i.e., they 
always introduce a DS break. This is not as drastic a 
claim as it might seem, because simply narrating an 
h-token that precedes the current TF does not neces- 
sarily introduce a backward move. In particular, one 
sentence in the past perfect does not produce a back- 
ward move. (This point is made, with respect to the 
French plus-que-parfait, in Kamp and Rohrer 1983:256 
fn.) If the out-of-sequence h-token is presented perfec- 
tively, i.e., the RT is after it, the TF does not move 
backward. Compare the following two examples: 
Example 7 
The telephone rang: it Was Mme Dupont; her 
husband had eaten too many oysters for lunch. 
The doctor recommended a change in lifestyle. 
Example 8 
. . . Tessie sat beside \[Phoebe\] with Araminta. 
Araminta had begun as an ear of corn. On a piece 
of cloth Mother had painted a face, using bluing 
for the eyes... She had stitched on smaller 
pieces of cloth, black for the nose, pink for the 
checks and red for the mouth. She had tied the 
cloth over one end of the ear of corn, then made a 
- bonnet to go over that... All in all Araminta was 
one of the most elegant members of the Dawson 
family. (Worline 1956:28). 
In (7) the past perfect is like the present perfect in a 
dialogue: it indicates precedence and relevance but does 
not shift the TF. In (8) the past perfect is like a past 
tense in a dialogue: it creates a new TF position to 
which subsequent sentences refer, until simple past 
40 Computational Linguistics, Volume 14, Number 2, June 1988 
Alexander Nakhimovsky Aspect, Aspectual Class, and the Temporal Structure of Narrative 
terminates the embedded DS and returns to the one 
interrupted by the backwards move. 
A sentence in the simple past following a sentence in 
the past perfect does not always terminate an embedded 
DS: nontemporal anaphora may integrate the two sen- 
tences (cf. Webber, Example 20 in this issue, and 
discussion). Such a past perfect/simple past configura- 
tion is a must if one wants to go into a long, multi-FS 
digression, because there seems to be a strong limit on 
how long an embedded DS can continue in the past 
perfect, as in (8). Consider: 
Example 9 
John was the captain of Penn's squash team. He 
had previously been captain of the Haverford 
team, but the coach there started chasing after 
John's girlfriend and John got very upset. One day 
he went to talk to the coach, but... 
Example 9 can now continue with an embedded narra- 
tive of arbitrary length. However, the interrupted DS 
has to be eventually resumed, otherwise the narrative is 
perceived as sloppy (or, if Proust does it, self-con- 
scious, cf. Genette 1980:66-67). 
7.2 SIDEWAYS AND FORWARD MOVES 
The majority of micromoves belong in these two cate- 
gories, and in this brief subsection I cannot possibly do 
justice to them all. What I try to do is establish a 
framework, based on the conceptual apparatus of this 
paper, for a more detailed investigation. 
I assume, for simplicity, that both the most recently 
processed sentence (SO) and the current sentence (S1) 
consist of just one clause each, narrating h-tokens H0 
and HI, respectively; the verbs in the sentences, V0 and 
V1, can be either past or progressive, and their aspect- 
ual classes can be any of those introduced in Section 3. 
Consider a few combinations. 
H0: telic, simple past-Hl: telic, simple past 
Example 10 
a. John entered the president's office. The presi- 
dent got up. 
b. John blew his nose. The president mopped his 
brow. 
c. John hid in the president's office. Bob locked 
himself up in the closet. 
This combination is most frequently a sequence, as in 
(a), but not necessarily so. If H0 and H1 are not related 
by an obvious chain of causality the sequencing is not 
obvious, and if they both can be construed as responses 
to the same event, they are interpreted as overlapping, 
cf. (b) and (c). 
H0: telic, simple past-Hl: atelic or state, simple past 
H0: telic, simple past-Hl: non-instantaneous, 
progressive 
Example 11 
a. John entered the president's office. The presi- 
dent was asleep. 
b. John entered the president's office. The presi- 
dent was writing a letter. 
In these examples, there is no compositional relation 
between H0 and HI, and H1 is interpreted as beginning 
before H1 and continuing after it. However, consider 
the following texts, in which H1 is part of the body, or 
the resulting stage, of H0: 
Example 12 
John knelt at the edge of the stream and washed 
his hands and face. He washed slowly, feeling the 
welcome sensation of the icy water on his parched 
skin. 
Example 13 
a. Bob turned off the light. It was dark around him. 
b. Bob heard a loud click. The door behind him 
was locked. 
c. Bob closed his eyes. He was asleep. 
In (13), the narrative moves forward, into a narrated 
h-token. The conditions for the move are that there is a 
direct inferential link from H0 to HI, and/or some sort 
of incompatibility between them. So, before Bob's eyes 
were closed they had to be open, which, for most of us, 
is incompatible with being asleep; on the other hand, 
one frequently closes one's eyes with the intention of 
going to sleep. 7 
H0: state, atelic or progressive-HI: state, atelic or 
progressive 
Example 14 
a. John was fixing an electrical appliance down- 
stairs. Suddenly, the whole house was dark. 
b. The president was conducting the interview with 
Mary in a cold, official manner. Suddenly, to her 
utter amazement, he was offering her an impor- 
tant job. 
The contrast between (11) and (13) illustrates the fol- 
lowing narrative principle: it is easy to move sideways 
into a narrated h-token, but it is difficult to move 
forward into one. (In other words, an imperfective 
sentence will not be interpreted as advancing the TF 
unless such an interpretation is forced by some form of 
incompatibility and causality.) It is especially difficult to 
move forward from inside one interval into another one, 
as in Example 14. In fact, such a move cannot be 
effected without the help of an adverbial that acknowl- 
edges the gap (try removing suddenly from the exam- 
ples). 
This concludes my discussion of narrative moves. 
The sampling given here is certainly not complete, but 
an extensive empirical investigation can, I believe, 
produce a full catalogue, organized by the configuration 
of tenses, aspectual classes, and time scales of the 
Computational Linguistics, Volume 14, Number 2, June 1988 41 
Alexander Nakhimovsky Aspect, Aspectual Class, and the Temporal Structure of Narrative 
h-tokens involved. As the examples of this section 
show, no such configuration determines the narrative 
move uniquely, but they strongly constrain the choices. 
8 CONCLUSIONS 
This paper suggests several empirical investigations 
having to do with the lexicon and discourse. In conclu- 
sion, I would like to recapitulate them and outline some 
possible ways to proceed. 
In order to enter durational information into the 
lexicon, the time-measurement units need to be identi- 
fied, from moment, minute, and five minutes to year and 
decade. Then, for many, perhaps most, h-types, their 
time scales can be characterized by a sequence of such 
units, e.g., visit-1 \[minutes, hours\] "visit somebody" (as 
opposed to visit-2 \[days\] "visit with somebody"). Two 
h-types stand in the much-greater-than relation to each 
other if their time scales do not overlap. 
Aspectually, each h-type should be assigned to a 
class according to the classification developed in Sec- 
tion 4. Regular aspectual polysemies should be noted 
and their extent investigated. It would be helpful to 
know, for example, how many, and what kind of, 
process verbs can describe the instantaneous beginning 
of the process as in She ran (i.e., began to run). For telic 
h-types, as argued in Section 4, there is a limited 
number of ways in which the arguments of the h-type 
can be affected by it (create, destroy, move, modify, 
expend: see Section 4.2). These ways can be catalogued 
and used as labels on the appropriate slots of lexical 
representations. Qualitative dependencies between ar- 
guments can be described using Forbus's (1985) ap- 
proach. (Nakhimovsky 1987a has examples.) A lexicon 
so outfitted can be used in a constraint propagation 
program, similar to the Allen and Kautz 1985 experi- 
ment, in a manner described in Section 4.6. 
With respect to narratives, two investigations are in 
order, one concerning micromoves within a DS, the 
other concerning macromoves and DS boundaries. The 
aim of the former investigation would be to identify the 
most frequent temporal/causal patterns of micromoves 
and how they are signaled by tense and aspect. The 
aims of the latter investigation would be, first, to 
identify the types of discontinuities that cause the shift 
to a new DS; second, identify typical combinations of 
such discontinuities and how they correspond to rhetor- 
ical notions and the narrator's intentions; and finally, 
develop heuristics for the proper segmentation of nar- 
ratives. At this stage, work outlined here is likely to 
connect with some of the pursuits of structuralist and 
poststructuralist poetics. 
NOTES 
1. I am grateful to Rebecca Passonneau for detailed comments on an 
earlier version, and to Bonnie Webber for many discussions and 
painstaking editorial work. Conversations with Wayles Browne, 
Donka Farkas, Tom Myers and Mark Steedman have been helpful 
in clarifying my views (which is not to claim that clarity has been 
achieved). Patricia Ryan of Colgate has made it possible for me 
and the deadline to meet. Work on the paper was mostly done 
during a study leave granted by Colgate. 
2. It was pointed out to me (cf. also Comrie 1976) that modern 
technology, and in particular slow motion movies, can spread out 
instantaneous events so that one can imagine somebody saying 
Look, he's blinking! or He's beginning to blink! in reference to a 
protracted individual blink on the screen. I don't think this 
abolishes the durational basis for the notion of instantaneous 
events: they're still treated differently than noninstantaneous 
ones, and the above sentences refer to an event on the screen, not 
a real-life blink. We wouldn't want to say that just because we can 
produce a hugh image of a mosquito we no longer think of it as 
something small, or that the phrase No larger than a mosquito can 
no longer be used. 
3. See Bach 1986, ter Meulen 1987. Dowty 1987 reports an idea of 
Hinrichs' that "'with respect to their Theme arguments, the 
meanings of telic predicates are homomorphisms from the algebra 
of entities (NP denotations) \[as suggested in, e.g., Link 1983-- 
AN\] into the algebra of events (as conceived by Bach)." 
4. Dahl (1981) confuses the issue by defining perfectivity only for 
telic histories (p. 82), in terms of reaching or not reaching their 
terminal point. Not surprisingly, he finds it difficult to keep the 
two categories apart. 
5. Clearly, both narrative and exposition can be embedded in a 
conversation: the modality (spoken or written) does not deter- 
mine the genre uniquely. I am talking about clear-cut cases. 
6. One such investigation is already under way as part of a larger 
interdisciplinary study of the Deictic Center in Narratives (Bruder 
et al., 1986). Some of the foregoing material has been influenced 
by that project and overlaps with Nakhimovsky and Rapaport (in 
preparation). 
7. Discussing examples like these, Cooper 1986:34 says that this 
kind of move is impossible with progressives. I don't think this is 
the case; once the dynamics of the move is understood, an 
appropriate example is easy to construct, e.g.: A heavy volume 
slipped off the top shelf and landed on Bob's right toe. He was 
hopping on the unaffected foot, holding the hurt one in both 
hands. 

REFERENCES 
Allen, James 1984 Towards a General Theory of Action and Time. 
Artificial Intelligence 23(2): 123-54. 
Alien, James and Kautz, H.A. 1985 A Model of Naive Temporal 
Reasoning. In Hobbs and Moore 1985. 
Apresjan, Juri 1974 Lexical Semantics. (In Russian) Nauka, Moscow. 
Almeida, Michael J. 1987 Reasoning About the Temporal Structure of 
Narratives. Ph.D. diss. and TR 87-10. Computer Science Depart- 
ment, SUNY at Buffalo, Buffalo, NY. 
Bach, Emmon 1986 The Algebra of Events. Linguistics and Philoso- 
phy 9: 5-16. 
Bobrow, Daniel, (ed.) 1985 Qualitative Reasoning about Physical 
Systems. M1T Press, Cambridge, MA. 
Bruce, Bertram 1981 A social interaction model of reading. Discourse 
Processes 4:273-311. 
Bruder, G.A. et al. 1986 Deictic Centers in Narrative: An Interdisci- 
plinary Cognitive-science Project. TR 86-20. Computer Science 
Department, SUNY at Buffalo, Buffalo, NY. 
Bunt, H. 1985 The Formal Representation of (quasi) Continuous 
Concepts. In Hobbs and Moore 1985. 
Campbell, Jeremy 1986 Winston Churchill's Afternoon Nap. Simon 
and Schuster, New York, NY. 
Comrie, B. 1976 Aspect. Cambridge University Press, Cambridge, 
England. 
Cooper, Robin 1986 Tense and Discourse Location in Situation 
Semantics. In Dowty 1986a: 17-36. 
Creary, L.; and Pollard, Carl 1985 A Computational Semantics for 
Natural Language. Proceedings of the 23rd Annual Conference of 
the Association for Computational Linguistics, Chicago, IL; 
172-9. 
Dahl, O. 1985 Tense and Aspect Systems. Basil Blackwell, Oxford, 
England. 
Dowty, David 1986 The Effects of Aspectual Class on the Temporal 
Structure of Discourse. In Dowty 1986a: 37-61. 
Dowty, David, (ed.) 1986a Tense and Aspect in Discourse. Linguistics 
and Philosophy 9(1) (special issue). 
Dowty, David 1987 Aspect and NP semantics. Logic and Linguistics 
Conference. Stanford, CA. 
Dowty, David 1977 Toward a Semantic Analysis of Verb Aspect and 
the English "imperfective" Progressive. Linguistics and Philoso- 
phy 1: 45-77. 
Forbus, K.D. 1985 Qualitative Process Theory. In Bobrow 1985. 
Fraisse, Paul 1984 Psychology of Time Perception. Annual Review of 
Psychology 1-34. 
Genette, G. 1980 Narrative Discourse. Cornell University Press, 
Ithaca, NY. 
Grosz, Barbara; and Sidner, Candace 1986 Attention, Intentions, and 
the Structure of Discourse. Computational Linguistics 12(3): 175- 
204. 
Hayes, P. 1985 The Second Naive Physics Manifesto. In Hobbs and 
Moore 1985. 
Hayes, P. 1985a Naive Physics I; Ontology for Liquids. In Hobbs and 
Moore 1985. 
Heim, Irene 1982 The Semantics of Definite and Indefinite Noun 
Phrases. PhD diss. University of Massachusetts, Amherst, MA. 
Hinrichs, Erhard 1986 Temporal Anaphora in Discourses of English. 
In Dowty 1986a: 63-82. 
Hirst, G. 1987 Semantic Interpretation and the Resolution of Ambi- 
guity. Cambridge University Press, Cambridge, England. 
Hobbs, J.R. 1982 Towards an Understanding of Coherence in Dis- 
course. In Lehnert and Ringle (eds.). Natural Language Process- 
ing. Erlbaum Associates, Hillsdale, NJ. 
Hobbs, J.R. 1985a Granularity IJCAI 85: 432-5. 
Hobbs, J.R. and R. C. Moore (eds.) 1985 Formal Theories of the 
Commonsense World. Ablex Publishing Corporation, Norwood, 
NJ 
Hobbs, J.R. et al. 1985 Commonsense Summer: Final Report (no. 
CSL1-85-35) CSLI, Stanford, CA. 
Hobbs, J.R. et al. 1986 Commonsense Metaphysics and Lexical 
Semantics. Proceedings of the 25th Annual Conference of the 
Association for Computational Linguistics, Stanford, CA; 231--40. 
Horn, Larry 1984 Toward a New Taxonomy for Pragmatic Inference. 
In D. Schiffrin (ed.) Meaning, Form and Use in Context: Linguis- 
tic Application. Georgetown University Press, Washington, DC. 
Joyce, James 1969 \[1914\] "Clay." In Dubliners. Modern Library, 
New York, NY; 99-106. 
Kamp, Hans 1981 A Theory of Truth and Semantic Representation. In 
J.A.G. Groenendijk, et al. (eds.), Formal Methods in the Study of 
Language. Amsterdam. 
Kamp, Hans and Christian Roerer 1983 Tense in Texts. In R. Bauerle, 
C. Schwarze, Arnim von Stechow (eds.). Meaning, Use and 
Interpretation in Language. Walter de Gruyter, Berlin, Germany. 
Lehnert, Wendy 1982 Plot Units: A Narrative Summarization Strat- 
egy. In W. Lehnert and M. Ringle (eds.). Strategies for Natural 
Language Processing. Erlbaum, Hillsdale, NJ. 
Link, G. 1983 The Logical Analysis of Plurals and Mass Terms: A 
Lattice-theoretical Approach. In R. Bauerle et al. (ed.), Meaning, 
Use and the Interpretation of Language. Walter de Gruyter, 
Berlin and New York. 
McDermott, Drew 1982 A Temporal Logic for Reasoning about 
Processes and Plans. Cognitive Science 6: 101-155. 
Moens, M.; and Steedman, M. 1987 Temporal Ontology in Natural 
Language. In Proceedings of the 26th Annual Conference of the 
Association for Computational Linguistics, Buffalo, NY; 1-7. 
Mourelatos, A.P.D. 1981 Events, Processes, and States. In Tedeschi 
and Zaenen. 
Nakhimovsky, A. 1983 Meaning-Text Linguistics and the Problem of 
Voice. Linguistic Research Inc., Edmonton, Alberta. 
Nakhimovsky, A. 1986 The Lexicon and Commonsense Reasoning: A 
Case Study. Proceedings of the Conference on Advances in 
Lexicology. University of Waterloo, Waterloo, Ontario. 
Nakhimovsky, A. 1987a The Lexicon, Grammatical Categories, and 
Temporal Reasoning. In J. Hallam and C. Mellish (eds.), Advances 
in Artificial Intelligence, John Wiley, Chichester, England. 
Nakhimovsky, A. 1987b Temporal Reasoning in Natural-Language 
Understanding: The Temporal Structure of the Narrative. EACL- 
87. Copenhagen. 
Nakhimovsky, A. and Rapaport, W. In preparation. Discontinuities in 
narratives. 
Nirenburg, Sergei; Raskin, Victor; and Tucker, Allen 1985 Interlingua 
Design for TRANSLATOR. Proceedings of the Colgate Confer- 
ence on Machine Translation. Hamilton, NY. 
Passonneau, Rebecca. 1987 Situations and Intervals. In Proceedings 
of the 26th Annual Conference of the Association for Computa- 
tional Linguistics, Buffalo, NY; 8--15. 
Rapaport, William J. 1986 Logical Foundations for Belief Represen- 
tation. Cognitive Science 4: 371-422. 
Reichman, R. 1985 Getting Computers to Talk Like You and Me. MIT 
Press, Cambridge, MA. 
Schank, R. (ed.) 1975 Conceptual Information Processing. North- 
Holland, Amsterdam. 
Sidner, Candace 1983 Focusing in the Comprehension of Definite 
Anaphora. In M. Brady and R. Berwick (eds.), Computational 
Models of Discourse. MIT Press, Cambridge, MA. 
Talmy, Leonard 1983 How Language Structures Space. Berkeley 
Cognitive Science, Report no. 4, Berkeley, CA. 
Tedeschi, P.; and Zaenen, Anne (eds.) 1981 Tense and Aspect 
(Syntax and Semantics 14), Academic Press, New York, NY. 
ter Meulen, Alice 1987 Structured Domains for Events. Workshop on 
Generics and Mass Terms. Stanford, CA. 
Tompkins, Jane 1981 Reader Response Criticism. Johns Hopkins 
University Press, Baltimore, MD. 
Vendler, Z. 1967 Verbs and Times. In Linguistics and Philosophy. 
Cornell University Press, Ithaca, NY. 
Verkuyl, Henk 1972 On the Compositional Nature of Aspect. Reidel, 
Dordrecht. 
Webber, Bonnie 1979 A Formal Approach to Discourse Anaphora. 
Garland, New York, NY. 
Webber, Bonnie 1987a Two Steps Closer to Event Reference. TR 
MS-CIS-86-74, Department of Computer and Information Science, 
University of Pennsylvania, Philadelphia, PA. 
Webber, Bonnie 1987b The Interpretation of Tense in Discourse. In 
Proceedings of the 26th Annual Conference of the Association for 
Computational Linguistics, Buffalo, NY; 147-154. 
Wierzbicka, Anna 1985 Lexicography and Conceptual Analysis'. Ka- 
roma, Ann Arbor, MI. 
Worline, Bonnie Bess 1956 The Children Who Stayed Alone. Scho- 
lastic Book Services, New York, NY. 
