1. The Problem

In 1936, Benjamin Lee Whorf wrote a justly famous paper entitled "An American Indian Model of the Universe" (Carroll, 1956). In that paper, Whorf criticized the easy assumption that people in different cultures, speaking radically different languages, share common presuppositions about what the world is like. He contrasted the Hopi view of space and time with what he called elsewhere the Standard Average European view. For the Hopi, space and time are inherently relativistic; for the speaker of Western European languages, like English, the universe is basically Newtonian, time and space are absolute, "containers" of things and events.

It is not my purpose here to discuss Whorf's assessment of Hopi metaphysics. Rather I would like to consider a little more carefully than Whorf did the other side of the comparison. This essay may be thought of as an exercise in "ethnometaphysics," an attempt to dig out the hidden assumptions made by speakers of English about the way the world is. My topic will be Time.

What methods can we use in such an enterprise? One approach would [64] be to ask the native speakers what they think about various questions. Is time dense, discrete, or continuous? Is there a single time line into the future or do we have to allow for branching futures, reflecting different ways we might choose or expect the world to be tomorrow? Can events recur in all their particularity? Many philosophers, members of "Standard Average European" cultures, have sought answers to such questions, and their answers have been far from uniform. So it seems we must take a different and more difficult approach.

The tack that I will take is this: We attempt to find out about the hidden structures of meanings in a language and culture by constructing formal theories about the syntax, semantics, and pragmatics of the language. We seek indirect evidence for our hypotheses by trying to give the most accurate and general explanations for various facets of linguistic knowledge. In carrying out this task we find that we can account for a number of seemingly disparate facts -- intuitions about sentences, their well-formedness, interrelations, truth conditions -- by separating out certain assumptions that seem to be metaphysical in character, rather than semantic or syntactic. Thus, the validity of our conclusions rests ultimately on the coherence and explanatory value of our entire picture.

In the context of such an inquiry I would like to consider a number of puzzles about the English tense-aspect system and its interaction with temporal expressions. I will try to show how a number of these puzzles can be solved by positing certain metaphysical assumptions made by speakers of the language. I believe that my conclusions have a wider application than just English and hope that I will be able to show that the facts we uncover about English point toward a more basic human matrix of assumptions about the world, that the Hopi and English world views are not at bottom incompatible, but rather are different orchestrations of material that is part of our common human heritage.

The framework within which I work draws on two main sources, the work in generative grammar initiated by Chomsky and the philosophical and logical methods first applied fruitfully to the study of natural languages by Richard Montague. Even though the paper is presented in an informal way, it is to be judged ultimately on the extent to which I am able to translate it into a precise account of English syntax, semantics, and pragmatics. To the extent that my account is successful I think that it supports the view that the bewilderingly complex facts of natural languages are best understood as resulting from the interplay of a number of relatively simple and autonomous subsystems. In particular, it will support the view that it is a good move to think of semantics itself as a relatively narrow and simple system.

[65]

2. English Tense Logic

There is by now a vast literature on tense logic. Philosophers and logicians have studied the properties of a wide variety of different systems. Montague's treatment of some English tenses, set forth primarily in his paper "The proper treatment of quantification in ordinary English" (Paper 8 in Montague 1974, henceforth PTQ) falls within this tradition. Montague's semantics is model-theoretic and intensional. Truth is defined relative to a possible world and a time. We might begin by outlining the system of PTQ.

The interpretation of English given in PTQ assumes among other things two sets, I and J, the set of possible worlds and the set of times, respectively. There is a simple ordering relation, ≤ , with J as its field and J is understood to be the set of moments of time. To illustrate the method, John has loved Mary is taken to denote the Truth at a world i and time j just in case there is a time j', strictly earlier than j, such that John loves Mary is true at i, j'. If we unpack the terse definitions and terminology of PTQ that relate to time we can restate what I have just said as follows:

1. Time is absolute across possible worlds. It is always possible to relate temporally things that happen in different possible worlds. From this it follows also that things that happen in THIS world can always be temporally related in a unique way.

2. The relation ≤ is transitive, reflexive, and antisymmetric (cf. Montague, 1974, p. 106, fn. 6). (A relation R is antisymmetric iff x R y and y R x always entail x = y.)

No other assumptions are made, but this much already commits us to certain conclusions:¹

(i) Newtonian physics is correct; Einsteinian physics is wrong. This follows because by (1) it is always possible to establish an ordering between two events, independent of observer.

(ii) Time travel is logically impossible.

(iii) Sentences about time relations in different possible courses-of-events (worlds) always have a definite truth value, example:

[66]

(1) If Mary had left on the space probe yesterday she would now be eating breakfast.

(iv) The fundamental units of the time series are moments, not intervals

(taking "moments" to mean "instants," cf. Paper 5 in Montague 1974).

Certain other questions are left open:

(v) Is the set of times infinite or finite?

(vi) Is there a first (last) moment of time?

(vii) Is time discrete, dense, continuous?

Again, answers to each of these questions imply certain conclusions about our judgments about English sentences. For example, suppose the set of times is finite. Then, certain sentences will be logically false, and their negations logically true. Let N be the cardinality of the finite set J. Then any sentence of the form "p₁ and then p₂... and then p_N+1" will be true at no possible world and its negation true at all possible worlds. If the answers to (vi) are yes, then at the last (first) moment of time any future (past) sentence will be false at all worlds and hence any implication with it as antecedent will be logically true. And so on.

Considerations such as these lead me to the view that our specifications of THE tense logic of English, if there is such a thing, should be cast in a very liberal way.

3. Pragmatics

More particularly, my framework is this: English is a pragmatic language, by which I mean that it contains expressions like I, now, here and tenses, that must receive a value from a context before the expressions in which they stand can be evaluated. I am here now spoken by me at 8:03, May 10, 1979, in such and such a room in such and such a place must be turned into a proposition by filling in all the values for the indexical expressions. Thus, I adopt the view of Stalnaker (1972) and others according to which the interpretation of expressions of a language like English is accomplished in two stages. In particular, declarative sentences receive sentence meanings which are functions from contexts to propositions, the latter in turn are functions from worlds (or histories as I prefer to call them) to truth values.

It is usual to say that among the elements of the context is a time of utterance. This is not quite accurate, as there may not be an actual utter- [67]ance (think of a street sign). So I will speak instead of a context of evaluation, thought of as a kind of performance. As we are interested here in time and tense, I will confine attention to this element or index: the time of the evaluation. So declarative sentences will be thought of in the first place as functions from times to propositions (see Kratzer, 1977; von Stechow, 1977, on the problems of a theory of contexts).

4. Events, States, Processes

There is a well-known typology that has its roots in Aristotle but has been brought into linguistic and philosophical thinking in recent times mainly by Anthony Kenny (1963) and Zeno Vendler (1957). It has to do with different kinds of sentences (or verb phrases) and the kinds of states-of-affairs that they describe. According to this view we need to contrast STATES, PROCESSES, and two kinds of EVENTS: PROTRACTED and INSTANTANEOUS (the latter correspond to Vendler's ACCOMPLISHMENTS and ACHIEVEMENTS, respectively). Some sentences illustrating the four types are these:

(2) Mary is in New York. (STATE)

(3) John ran (for an hour). (PROCESS)

(4) Mary built a cabin.(PROTRACTED EVENT)

(5) Mary found a unicorn. (INSTANTANEOUS EVENT)

(You should note that these terms are not uniformly used in the literature.) There is another dimension which crosscuts these distinctions, in English at least, having to do with the notions of agentivity, volition, and the like. The necessity for taking the distinctions into account for the semantics of English has become most evident in a series of papers by various philosophers and linguists who have dealt with the English progressive and similar constructions in other languages.²

Let us note some contrasts in the acceptability and interpretation of sentences in the progressive. A state sentence cannot occur in the progressive:

(6) *Mary is being in New York.

(I will discuss apparent exceptions to this in what follows.) The other three types can:

[68]

(7) John was running.

(8) Mary was building a cabin.

(9) Mary was finding a unicorn.

If we follow Vlach (1981), who has given to my mind the best account of the English progressive, in the idea that the progressive construction itself is stative, then it follows that there can be no progressives of progressives:³

(10) *John was being running.

Further, if the English perfect forms state expressions, it also follows that we will not have progressives of perfects:

(11) *John is always having been hurt.

The necessity to keep processes apart from events can be seen, among other things, in the semantics of the progressive. It is necessary to give different truth conditions for sentences involving progressives of processes and events (cf. Vlach, 1981). Various writers (e.g., Dowty, 1977) have tried to give an account of the semantics of the progressive as a uniform function of the semantics of corresponding simple sentences. This works only for processes. For example, we can say that on one understanding, a sentence like John was running for an hour is true just in case the sentence John runs is true at sufficiently many subintervals (or moments) of some hour. This is clearly wrong for sentences like Mary was building a cabin for three years or Mary was finding a unicorn yesterday.

The hardest distinction to understand, so far, is that between processes and states. That there must be such a distinction is clear from the interpretations we must give to simple present sentences:

(12) Mary loves John.

(13) Mary runs.

The latter but not the former has two understandings: On the one hand it can be telling us something about Mary's habits or dispositions (call this the gnomic sense); on the other hand it can be used as a vivid description of an entire event (so-called "reportive") or as a historical present: "Last night I go downtown. I see Mary. She runs." (It is possible that these last [69] two understandings should be assimilated.) I say "understandings," rather than "meanings" or "readings," as I do not want to prejudge the question whether such sentences are genuinely ambiguous. (Of course, we can give nonce nonstative meanings to stative verbs like love; these come out when the verbs are in the progressive: Mary is loving John, I'll return to these in what follows.)

There have been a number of attempts to explicate these notions on the basis of time structures. For example, Montague (1974, Paper 5) proposed that generic instantaneous events be analyzed as properties of moments of time. I prefer to go in the other direction and analyze our notions of time on the basis of the typology given earlier, taking it as primitive. One reason for this is of purely autobiographical interest. I find it much easier to understand these notions than various abstract properties of time. But there are more interesting reasons for taking this approach, which will emerge in what follows.⁴ For my semantics, I take as basic the notion of a POSSIBLE HISTORY. This is in a sense nothing but a possible world, but I use the word history to stress the fact that the temporal relations that will come out of the analysis are world-immanent, not as in PTQ independent of worlds. Along with a collection of possible individuals (in the ordinary sense) I will assume as given a set of EVENTUALITIES, this being the generic term for the things we have just been considering: states, processes, and events (of two kinds).

Following mainly Whitehead (1920, but cf. Note 4), I will assume that there are two primitive relations that can hold between eventualities; one is a relation of strict precedence < for which I will write BEFORE, the other is a notion of covering (Whitehead's notion of "extending over"), for which I will use the term WHILE. (These words are chosen advisedly, but should be considered technical terms that will not be used quite in the way the English words that they resemble are used.) About these relations I will assume the following:

(i) BEFORE is transitive, asymmetric, and hence irreflexive.

(ii) WHILE is transitive, asymmetric, and hence irreflexive.

(iii) BEFORE and WHILE are mutually exclusive.

(Thus WHILE is to be interpreted as "properly covering": In time talk, the duration of the second eventuality is a nontangential proper part of that of [70] the first.) A HISTORY is now taken to be a set of eventualities and relations among them. Please take note that it is explicitly not assumed that every pair of eventualities in a history is related by BEFORE or WHILE. A maximal set of eventualities that ARE so related I will call a LOCAL HISTORY. (Thus, if the actual history of the universe is a local history, then it is a Newtonian history.) We can define SIMULTANEITY as another relation between eventualities as follows:

(iv) Sim(e,e') =_df for all eventualities e", (WHILE (e",e) iff WHILE(e",e')) and (WHILE(e,e") iff WHILE(e',e"))

Now, although I cannot say exactly what the various kinds of eventualities are, we can say something about some of their properties (the following owes a great deal to unpublished work and lectures by Lauri Carlson, cf. also Mourelatos, 1978). First of all, there is a natural affinity between the kinds of things that we are thinking about and theories of MEREOLOGY, that is the logic of part-whole relations. Such a theory takes as its basis a single binary relation, that of being a part, and a single operation, that of forming a new individual out of several individuals. Using these ideas it is possible to state intuitively reasonable characteristics of the various kinds of eventualities. Consider first two typical events: a finding of a unicorn and a building of a cabin. Whatever else is true of such events, no proper part of one can be an event of the same kind. Call this property ANTISUB-DIVISIBILITY. This property is clearly not shared by processes. Note that it is not correct to say that a process can always be subdivided into parts that are also processes of the same kind. The point is that sometimes processes can be so subdivided but events never can. Further, if you have two distinct events of the same kind, their sum is never an event of the same kind; but if you sum two or more processes of the same kind you will or may have a process of the same kind: call the latter ADDITIVITY. Thus we can say that events are antisubdivisible and nonadditive; processes lack these properties (Note: Again it is not the case that processes are necessarily subdivisible and additive). We are reminded of the similar distinctions having to do with the nominal system of English and the difference between bare plurals (dogs) and mass terms (gold) on the one hand and count nouns (dog) on the other.

The foregoing considerations provide a strong argument against all attempts to reconstruct events and the like on the basis of times. For example, if an event is a property of a certain kind of interval, we simply can not say that no proper part of the interval is also an event of the same kind. Further, certain entailments that have been noted in the literature can be captured in a very direct way. Consider, for example, the "entailment" (cf. Kenny, 1963):

(14) If Mary is finding a unicorn, Mary has not found a unicorn.

[71] Although this entailment is clearly false, there is a valid intuition behind it. The intuition is something like this: If Mary is finding a unicorn, then THAT finding of a unicorn by Mary is not over. This statement uses the notion of an individuated event directly. Another sort of "entailment" has to do with processes:

(15) If John is running, then John has run.

Although again I do not think that this is literally a semantically valid entailment (consider the very beginning of John's running), what I have said about subdivisibility makes it a natural kind of pragmatic "entailment."

Within the preceding framework of contrasts, what can we say about states? The ontological status of states is considerably more obscure than that of events and processes. Moreover, the relation between states and temporal notions often seems mysterious. In the case of events and processes we were led naturally to think about their temporal parts. Events may have beginnings and ends and possibly "middles." We think easily about a process going on for a time and think about smaller chunks of the same process. States have an atemporal and abstract quality. Think about the questions: Where did a certain event take place? Where is a certain process going on? Now think about the questions: Where is the state of John's being in New York located? Where does Mary know algebra? It is hard to make sense of these last sentences. The atemporality of states (at least some states) can be brought out in the following Gedankenexperiment: Imagine a possible history (world) with only one time (which in a sense amounts to having no time); it is possible to think of various states that might obtain in such a world, but impossible to imagine events and processes that occur or go on in such a world.

Bennett and Partee (1978) have suggested using (among others) the "subinterval" property to distinguish event predicates from states and processes. If a sentence has a subinterval verb phrase as its main predicate and is true at some interval I, then it will be "true at every subinterval including every moment of time in I." This is an important idea (and clearly related to the notion of antisubdivisibility). However, taken literally it is applicable only to states, not processes. If we take a process verb like run and consider an instant included in some interval of running, then the plain predicate run will NOT hold for the argument. But a stative predicate, including in particular the progressive predicate be running (Vlach, 1981), does have this property. Rather tentatively, we can sum up these differences as follows. We will require that if e is an event, then there must be at least one other event [72] connected to e by the relation BEFORE, and if e is a process then there must be at least two events overlapped by e that stand in the BEFORE relation. States will have no such requirement. Thus possible histories that contain events or processes are guaranteed to have a certain temporal "thickness." Our world does have such thickness, hence the abstract character of states.

In any case, there are certain facts about states that should be explained in any adequate theory. Here are some:

(i) The naturalness of simple present tense sentences. We have already noted this, and its concomitant: the oddity or downright anomalousness of progressives for states.

(ii) The semantics of when-clauses Vlach (1981) notes that for a sentence combining a when-clause rooted in a state sentence with an event sentence to be true the time of the event has to be in the interval defined by the state:

(16) When John was in New York, Mary left.

In time talk, there must be at least an instant of John's being in New York coincident with Mary's leaving. We can sharpen this intuition by a sentence like (17):

(17) ?When I was in bed, I got up.

The question mark serves to indicate that on the ordinary understanding of getting up (i.e., getting out of bed) this sentence is anomalous.

5. Time Adverbials

The distinctions just reviewed play a role in the interpretation of time adverbials and combinations of tense and aspect. I will first consider adverbials, restricting the discussion in this section to sentences in the simple past tense.

Following Bennett and Partee (1978), we can distinguish three main classes of adverbials:

1. FRAME ADVERBIALS:
        Point: at 3 o'clock
        Interval: on the 23rd of June, today

2. DURATION: (for) three hours, all day

3. FREQUENCY: twice, frequently

[73] Frame adverbials occur only with event or state descriptions, thus forcing an event or state interpretation when combined with a process phrase.

(18) Mary found a unicorn at dawn.

(19) Last year Bill built a cabin.

(20) In 1943, Sam was in New York.

(21) John ran yesterday.

(22) John ran last year.

Sentence (21) requires us to think of a particular event of which running is a part, say, running for a specific length of time, starting to run, or the like. Sentence (22) also has available a stative, generic reading.

Some peculiarities arise when we make various specific choices.

(23) At 4 o'clock, Bill built a cabin.

(24) At 3 o'clock, Mary knew physics.

Sentence (23) is peculiar only because we know that building a cabin, in the real world, is a protracted event. If Bill were a magician, then it would be perfectly acceptable. Sentence (24) is, I think, an unlikely but not anomalous sentence. For many states, it is odd to pin them to particular times and we tend to reinterpret the stative predicates as event predicates (come to know). On that understanding (24) would be unlikely, again for real world reasons (compare At 3 o'clock Mary knew the answer).

Point-time adverbials name events: in conventional time schemes, periodic events like dawn, the summer solstice, positions of hands on a clock, vibrations of the cesium atom. Sentences involving them are true just in case an event described in the sentence is simultaneous with the named event or a state described holds "at" the event in question. What happens when we have a process predicate like run is that the truth conditions force us to take a "piece" of running which is simultaneous to some named event or we take the process in a generic sense. We can offer the same explanation for interval adverbials if we provide for them a semantics that incorporates the idea of a reference event ("time") within the named interval.

We find just the opposite pattern for events and processes in much discussed examples involving durational expressions (not necessarily time expressions):

(25) John ran for an hour.

(26) ?For 3 hours, Mary found a unicorn.

[74] (27) ?John built a cabin for 3 days.

(28) For 3 days, John was in New York.

(29) ?For 3 years, Mary knew physics.

State descriptions occur with durationals (with the same proviso, it is odd to think about -- not just say something about -- someone knowing something for a definite period of time). Event descriptions with durationals are anomalous, not just odd, process descriptions are fine. How are we to account for this difference?

The combination of a specific durational adverbial with a process predicate (or sentence) acts in every way like an event predicate (or sentence):

[75]

(30) It took John an hour to run for an hour (naturally).

(31) ?John ran for an hour for an hour.

Once again the analogy with the count-mass distinction comes to mind. Durational expressions stand to verbal expressions as amount expressions stand to nominal expressions. Just as we do not use expressions like 3 pounds of with singular count nouns like a horse, we do not use the expressions that chunk up our experience with (singular) expressions that provide that experience already chunked up. Another way to see the same thing is this: Suppose durationals have something like the following truth condition: for I, p (p a sentence, I an interval) is true just in case it is true at all (or sufficiently many scattered) subintervals of I. Then (26) (cf. Vlach, ms.) would be literally true if Mary finds a unicorn sufficiently often over the 3 hours in question. But as speakers of English we know that the right way to say THAT is Mary found unicorns for 3 hours.

Finally, frequency adverbials like often co-occur with all sorts of descriptions but again with a difference.

(32) Mary often found a unicorn.

(33) People frequently went to the country.

(34) Sometimes Bill ran.

(35) John was in New York twice.

In event sentences like (32) and (33) the adverbials are acting like direct quantifiers over events. With processes and states we have to say a little more. For (34) to be true it must have been true on several occasions that Bill started to run, ran, and stopped. Similarly for (35). Thus by their nature frequency adverbials like these require that the processes or states involved constitute parts of events.

Sentences like (33) are particularly interesting for my basic thesis that time is a derivative notion. Sentence (33) does not mean that there were frequent "times" at which someone went to the country. Similarly, John was hit twice requires that there be two hittings but they may have occurred at the very same instant (gratia Terry Parsons). What is needed is some notion like "occasion" or "case" (Lewis, 1975). Such examples provide direct evidence against any theory that combines PTQ's treatment of time and Montague's analysis of events as properties of moments of time (1974, Paper 5).

The possibilities for combining various sorts of time adverbials in a single sentence are literally endless, and I cannot begin to do justice to the topic in this chapter. But it is worth looking at a few examples to show how metaphysical assumptions can be used to explain certain judgments. Consider this sentence:

(36) ?At 3 o'clock, John left at 4 o'clock.

Such examples are sometimes taken as showing that the grammar of English should contain a restriction (syntactic or semantic) that limits the number of point time adverbials in a simple sentence to one. But this is unnecessary. It follows from the stipulation that BEFORE is irreflexive and the semantics of the particular time expressions that (36) could be true in no possible "standard" history. Similar considerations apply to mismatches between temporal expressions and tenses:

(37) John will leave yesterday.

(38) Mary left tomorrow.

6. Tense and Aspect

In this section I wish to take up several problems in the interpretation of the English tense-aspect system. Once again, I can only skim the surface and give a few examples that show the interaction of the typology of eventualities with this system.

The first set of puzzles has to do with the interpretation of the simple present tense. Indeed, it was in part the behavior of simple present tense sentences and contrasts between simple present and progressive that led Vendler and others to set up the typology we have been examining. If be + progressive is itself a state-describing expression (Vlach, 1981), then we can say that most "ordinary" present tense sentences are state descriptions:

[76]

(39) Mary loves John.

(40) Three plus two equals five.

(41) John is in New York.

(42) John is building a cabin.

(43) Mary is running.

These are to be contrasted with simple present tense sentences involving single events and processes:

(44) Mary finds a unicorn.

(45) Harry builds a cabin.

(46) Sam runs.

These sentences require "special" interpretations: a so-called reportive reading; a futurate present (The Mets play at 3); or in the case of (46) a generic or gnomic reading (itself stative). And we find that a process verb like run, interpreted in the reportive way, requires us to think of some singular event involving running.

This configuration of facts reminds us of what we have seen about past-tense sentences with frame adverbials. This is not surprising. I have suggested that such sentences can be true in just two circumstances: an event described is simultaneous with the reference event (time) explicitly or implicitly present in the frame adverbial or a state described covers the reference event. Following our earlier discussion we can take present tense sentences to be functions from an event ("time") of evaluation (from context) to propositions. Thus the truth conditions are to be considered "at now." Moreover, if we assume that the event of evaluation (a performance) is thought of as instantaneous we can provide an explanation for a number of facts.

Consider first the problem of the reportive interpretation. I suggest a basically Gricean explanation. Suppose we take the account that was given literally: The event described must be simultaneous with the performance event. There are two circumstances when we can really be sure of this. One of these is when such sentences are used to PERFORM events, as with verbs like christen, promise, pronounce. Such verbs are used in the simple present with first person subjects in performative sentences. As we might expect, with other subjects they act like any other kind of event sentences: (He christens the child). The second is when we are, so to speak, creating a fictional history or recreating actual history. Here, by fiat, the events described are thought of as simultaneous with the evalua-[77] tion; hence the use of such sentences in poetry, narrative, historical present, and so on. The reportive use falls into the same slot.

Support for the idea that the time of evaluation is thought of as an instant comes from the consideration of sentences like these:

(47) I utter a sentence.

(48) I tell the truth.

If the performance of a sentence were identified with an utterance (something that happens in the physical world and takes time) then the temporal condition for such sentences would always be satisfied, and they should be perfectly ordinary. In fact, they have the same range of interpretations as any other event sentences. I believe the key to understanding such facts and others that have received attention in the literature (the habits of verbs like recognize) lies in the assumption that it is above all, or exclusively, mental acts that are felt to be truly instantaneous.

The second puzzle I would like to consider is illustrated in sentences like these:

(49) ?I am knowing the answer.

(50) John is believing that the earth is flat.

Every English grammar tells us that such sentences are peculiar if not ungrammatical or "unsemantic." But in every list of such "stative" verbs I have ever seen I have been able to find natural examples of progressives with stative verbs (with the sole exception of be when it combines with a prepositional phrase or nominal):

(51) I'm really loving the play.

(52) I'm understanding you but I'm not believing you.

We have already considered one class of such examples, where a normally stative verb is given a nonce meaning of a nonstative sort (Mary is loving John). But examples like (51) and (52) do not seem to fall into this usage.

I mentioned earlier the idea that the progressive is simply not defined for states (this is Vlach's position, 1981). If we accept this idea then we are forced to conclude that examples like (51) and (52) involve different predicates from the simple counterparts, that is, we would have to assume different lexical items: love₁, love₂, etc. But I believe the phenomenon is more systematic than that. To focus our ideas, let us consider a more perspicuous example:

(53) I live in Massachusetts.

[78] (54) I am living in California.

The conditions under which I would use these two sentences are fairly clear. Sentence (54) suggests or means that I am living in California temporarily. Sentence (53) has no such implications. For (54), there is the feeling that we are thinking of some bounded interval (this year, these days). The same obtains for progressives of gnomic sentences:

(55) John drives a truck.

(56) John is driving a truck (these days).

If we take these facts seriously we will have to abandon the idea that the progressive is not defined for states and add yet another case to the statement of the truth conditions for the progressive.

Let us call the kind of state described by a progressive of a stative verb TEMPORARY and the kind of state described by simple statives -- including in particular the gnomic situation as in (55) -- NONTEMPORARY. This distinction can also be seen in adjectives: on the one hand tall, intelligent, sane; on the other drunk, present, sick. Greg Carlson (1977) has explicated this difference as follows. Suppose we distinguish two kinds of entities. In our universe: ordinary individuals and their temporally limited manifestations (or "realizations"). Then we can say that adjectives are divided into those that denote properties of individuals and those that denote properties of their manifestations over some limited length of time. This distinction can be used to account for the kind of difference seen in examples just given. For Carlson, a progressive verb phrase denotes the property of being an individual such that there is a manifestation (or realization) of that individual of which the basic predicate holds. In mereological terms we can identify the manifestation of an individual with some temporally limited proper part of the individual. Thus, the progressive of a stative verb like love (e.g., (51)) has to do with a temporary state more or less directly connected with the meaning of the more basic nonprogressive predicate.

A consequence of the decision to let the progressive operator or construction occur with statives is that we must seek some new explanation for the lack of a progressive of a perfect or progressive:

(57) ?John is always having been hurt.

(58) ?Mary is being running in the park.

For the first case, the explanation is straightforward. Suppose John has been hurt is true in a history. Then forever after in that history it will be true. Hence, there is no temporary state corresponding to the stative

[79] predicate have been hurt . More generally, "eternal" sentences (Two plus two equals four) do not have progressive forms. There is also a straightforward explanation for (58) on Carlson's account of the progressive (or any parallel explanation using a distinction between individuals and their manifestations). The semantics assigns to Sentence (58) the following interpretation: Sentence (58) is true just in case (at history h at the contextually determined time of evaluation) Mary has the property of being an individual such that there is a Mary-manifestation that has the property of being an individual such that some manifestation of that individual runs. But (in Carlson's terms) this meaning is sortally incongruous: Manifestations cannot stand in the realization relation to manifestations.

7. Conclusions

I hope to have shown how certain metaphysical assumptions are essential to an understanding of English tenses and aspects. These assumptions have to do with the way reality -- or our experience of it -- is structured. I believe that our more sophisticated and abstract ideas about time have their roots in the fundamental types of eventualities and their relationships (as perceived by us). Whatever truth there may be in Whorf's account of Hopi metaphysics, I believe that he was simply wrong about the "Standard Average European" metaphysical assumptions about time. The minimal assumptions we must make about temporal relations in order to explain our intuitions about sentences lend themselves just as well to the construction of an Einsteinian as a Newtonian world view. Moreover, I believe that the notions we have been examining are universal: I have yet to encounter a language in which there is no reflection of the contrasts between states, events, and processes. This is not to say that the use they are put to or the reflexes that we find are identical across languages.

Certain aspects of human experience are common to all people. It is this matrix of common experience that is the stuff of which grammars are made: causation, human responsibility and intentionality, temporal and spatial relations, important classifications of the things in the world (animateness, sex), number, social hierarchy, family relations. Such notions probably enter in one way or another into every human language either as covert or overt categories (to use Whorf's terms). I believe that many universal aspects of language will be understood in the end as resulting from an interaction between the innate language-creating gift of the human animal and this common matrix.

I have said very little about the technical problems of incorporating such metaphysical distinctions into a linguistic description. I do not in- [80] tend to address such questions here. But I think we can draw one conclusion (and I believe I am in agreement with Chomsky here, though probably not on the question of what to do about it): A formal semantics for a natural language, if it is to be truth-conditional, cannot ignore metaphysical questions. If it is correct that the truth conditions related to tense, aspect, and temporal adverbials must make reference to the event structures we have posited, and if these distinctions are basically metaphysical, this conclusion is inescapable. Moreover, since what is required to work this out is reference to particular words, we cannot draw a sharp line between truth-conditional "structural" semantics and "lexical" semantics (to borrow Partee's terminology). To do justice to such aspects of language in an explicit theory is a challenging task. But even partial solutions cannot fail to tell us something important about language, people, and their place in the world.

Acknowledgements

The research reported on here was initially undertaken at the Center for Advanced Studies in the Behavioral Sciences, Stanford, under support from the Center and the National Endowment for the Humanities. I wish to express gratitude for this support and for the help of the Center staff, particularly Mary Tye. For criticism and discussion I would like to thank the following friends and colleagues (none of whom, of course, should be held responsible for my mistakes or misunderstandings): Greg Carlson, Lauri Carlson, David Dowty, Terry Parsons, Mark Stein, and especially Barbara H. Partee. I am also indebted to the late Michael Bennett for many good discussions.

Footnotes:

¹Strictly speaking, this is not quite right. In any model it will be the case that two events are temporally related in a unique way. It is unclear whether Montague intends his interpretation of English to be making any claims about the real world. One could also interpret his semantics for English as saying that Einsteinian physics is inexpressible in (ordinary) English.

²Dowty (1972, 1977), Vlach (1981) are two examples.

³This idea was also put forward by various linguists. I have been unable to find a reference, but I recall discussion of this point in the late sixties among especially generative semanticists.

⁴The idea of constructing "time" on the basis of primitive relations among events may be found in Russell 1929 and Whitehead 1920. More recently, it has been taken up by Kamp (1980) and Van Benthem (1980). Kamp shows the usefulness or necessity of this approach for two applications: a theory of discourse and a treatment of temporal relations for the interpretation of sentences involving vague predicates. [EB 2007: see added reference to Norbert Wiener (1914)]