Tim Stowell


What is the meaning of the present and past tenses? The answer to this question depends on what objects these terms refer to. If the question is about the English tense morphemes present and past, we will get one answer; if it is about their Japanese or Russian counterparts, we will get another; and if it is about a semantic categories PRESENT and PAST attributed to the theory of Universal Grammar (UG), we will get still another. In this article, I discuss the semantics (and the syntax) of (English) morphemes, but my theoretical treatment of this exploits temporal categories that I attribute to UG, expressed overtly in some languages and covertly in others.

In the first section of the paper, I review some basic facts about the temporal interpretation of clauses containing present and past and then sketch out the essential generalizations that any theory must account for. In the second section, I will review two major competing theories about the intrinsic meaning of these morphemes, and point out certain problems with each approach. Finally, I will briefly sketch out an alternative approach to the issue, framed within a broader theory of tense that I have developed elsewhere (Stowell 1993, 1994), incorporating a new syntactic phrase structure for the representation of tense, and a theory of its logical form that is based in part on a theory of temporal polarity, modelled on recent approaches to negative and positive polarity.

1. The Past-Shifted, Simultaneous, and Double-Access Construals. The story begins with the behavior of present and past in simple sentences such as those in (1):

(1) a. Arby knew Adam.

b. Arby knows Adam.

In (1a), the choice of past serves to locate Arby's state of knowing Adam within a period of time prior to the time that the sentence is uttered--the Speech Time, or Utterance Time (UT). In (1b), on the other hand, the choice of present locates Arby's knowing-state at a time interval that includes the UT. I refer to these simple usages of past and present as indexical tense construals, since they involve a form of temporal reference that is directly related to the speech act (the utterance), just as pronouns like you and me are (cf. Partee's (1973) discussion of deictic present tense.) Plainly, any theory of tense must be able to capture this simple contrast in a straightforward way. However, the facts discussed thus far are so simple that they massively underdetermine the theory, and the literature on tense in the research traditions of formal syntax and formal semantics has produced several quite different accounts of the meaning of present and past that are equally well-equipped to account for this paradigm.

Before discussing any of these theories, it is useful to consider the interpretations of present and past in sentences that are somewhat more complex:

(2) a. Arby will say [that he knew Adam].

b. Arby will say [that he knows Adam].

(3) a. Arby said [that he knew Adam].

b. Arby said [that he knows Adam].

In (2), the use of the future modal will with the main clause verb say serves to locate Arby's reported speech event at a time subsequent to the UT; in (3), the use of past with the same verb locates the reported speech event at a time prior to the UT, analogous to its use in (1a). So far, the traditional pretheoretical notions of present, past, and future tense seem to be sufficient to explain the facts. But things get more complicated when we consider the choice of tense in the subordinate complement clauses in these examples. Consider first the use of past in (2a) and (3a). In (2a), the choice of past locates Arby's state of knowing Adam at a time prior to the future speech event; the knowing-state is not explicitly ordered with respect to the actual UT (it may precede, follow, or include the UT); hence, it is not an indexical usage. Enç (1987) introduced the term past-shifted to refer to this type of tense construal, i.e. where the tense on a subordinate clause verb is used to locate the time of the subordinate clause event or situation in relation to the time of the main clause event, rather than in relation to the UT directly.

The situation is somewhat more complicated in (3a); here, the choice of past on know in the complement clause leads to an ambiguity with respect to the temporal ordering of the knowing-state and the reported speech event in the main clause. First, the knowing-state may be located at a time prior to the (past) speech event, analogous to (2a); this is another instance of Enç's past-shifted reading. Second, the knowing-state may be located at an interval that includes the time of the (past) speech event. The latter type of construal is referred to by Enç as a simultaneous tense construal: on this reading, the time of the subordinate clause state bears the same relation to the matrix clause event-time that the matrix clause event-time in (1b) bears to the UT. (This similarity has led some observers to describe this usage of past as an instance of PRESENT tense in disguise; this is the essential insight underlying the so-called Sequence-of-tense analysis of such examples. I will return to this issue later on.)

Note that the temporal location of the subordinate clause state is not entirely free in (3a). In particular, it may not be (entirely) subsequent to the time of the main clause speech event. For this reason, it is usually assumed that there are two distinct tense construals for the subordinate clause past here, rather than one reading which fails to order the subordinate clause time in relation to the main clause event-time. I will refer to these two readings (the past-shifted and simultaneous) as dependent tense construals, since in each case temporal reference in the subordinate clause is dependent on the temporal reference of the main clause. Although the time of the subordinate clause knowing-state is unambiguously prior to the UT on both readings, I assume (following Enç and many others) that this is a result of the interaction of the separate tense construals of the main and subordinate clauses. In the main clause, past locates the reported speech event prior to the UT; the subordinate clause past then locates the subordinate clause knowing-state in relation to the main-clause event-time.

Now let us turn our attention to the use of present in the subordinate clauses in (2b) and (3b). In (2b), the use of present in the subordinate clause has a simultaneous construal analogous to that of past in (3a); that is, the time interval of Arby's state of knowing Adam must include matrix event-time--the time of the reported (future) speech event. The same is true of (3b), although here, the semantics is more complex: the interval of Arby's knowing-state must encompass both the matrix event-time--the time of the reported (past) speech event--and the actual UT. This construal of present is referred to in the recent literature as a double-access construal (Enç (1987), Abusch (1989), Ogihara (1989)). The double-access construal of present under past in (3b) is aptly named, since it is arguably a combination of the dependent simultaneous construal of present in (2b) (or past in (3a)) and the indexical use of present in (1b).

Actually, the use of present under the future main clause in (2b) also allows a double-access construal, since the interval of Arby's knowing-state can be understood to include both the reported (future) speech event and the actual UT; however, it is hard to show that this is actually a structurally distinct reading, since the simultaneous construal does not place an explicit boundary on the duration of the state in question. Put another way, it is conceivable that (2b) only allows a true simultaneous construal, and that this reading fails to specify the relation between the time interval of the subordinate clause state and the actual UT. I believe that it is possible to show that a true double-access reading is available here, but I will leave this question open here; see Stowell (1993) for discussion.

Thus far I have neglected to discuss the interpretation of past and present in a complement clause embedded under a main clause containing present:

(4) a. Arby believes [that he knew Adam].

b. Arby believes [that he knows Adam].

The reason for the omission is that it is difficult to categorize the readings definitively. Since the main clause introduces no new time distinct from the UT, it is hard to say whether the subordinate clause tenses are being interpreted indexically (where the knowing-time is directly related to the UT) or whether they are dependent tense interpretations (where the knowing-time is related to the main clause believing-time, which happens to contain the UT). In either case, the result is the same: Arby's knowing-state precedes the UT in (4a) and includes the UT in (4b). I will henceforth exclude present-tense main clauses from my discussion.

Summarizing thus far, we have seen four kinds of readings of past and present:

(5) a. Indexical:

Main Clause past and present (1a,b)

b. Past-shifted:

past in a complement clause under matrix will or past: (2a, 3a)

c. Simultaneous:

past in a complement under past in a main clause (3a);

present in a complement clause under will in a main clause (2b);

d. Double-Access:

present in a complement clause under past in a main clause (3b);

present in a complement clause under will in a main clause (3b).

Arranging our observations somewhat differently, we get the following picture:

(6) a. past in a

main clause (1a): Indexical;

complement under a matrix will (2a): Past-shifted;

complement under a matrix past (3a): Past-shifted or Simultaneous;

b. present in a

main clause (1b): Indexical;

complement under a matrix will (2b): Simultaneous or Double-access;

complement under a matrix past (3b): Double-access

At this point we can draw some preliminary generalizations. The past-shifted construal is possible only for past. This should hardly be surprising, since the past-shifted construal is, after all, a dependent-tense analogue of the indexical construal of past, with the matrix clause event-time substituting for the UT as the point of reference. Ultimately, we should expect an economical theory of the meaning of tense morphemes to collapse these two construals under a single account of what past means, since they seem close to identical, apart from the reference-time in each case. On the other hand, the simultaneous construal of past in a complement clause embedded under a main clause containing past poses a problem for such a theory being extended to all instances of past, since here its meaning seems to be more closely analogous to the interpretation of present in other contexts. As for the meaning of present, its simultaneous interpretation in a complement clause embedded under a main clause containing will (or, alternatively, under a main clause that does NOT contain past) seems directly analogous to its indexical sense, with the matrix event-time substituting for the UT as the point of reference; we should expect an economical theory of the meaning of present to collapse the two. Once again, main clauses containing past pose a potential problem, since, in these environments, a double-access interpretation for the complement clause present is required.

Thus, both morphemes past and present seem to have a more or less constant interpretation; whereas past locates an event-time (or a state-time, in the case of a stative verb) prior to a given reference-time (the UT in a main clause, and the matrix event-time (or state-time) in a complement clause. In each case, the empirical impediment to maintaining such a view is the fact that both past and present seem to have a different type of construal when they occur in a complement clause embedded within a main clause that contains past; in such contexts, past can also be construed as though it were a "present tense in disguise", while present must be construed as though it were a combination of an embedded present (with a simultaneous construal) and a main clause present (with an indexical construal). Of course, it is possible to phrase things somewhat differently, depending on one's view of what these morphemes really mean, but I will defer a substantive discussion of this point to the next section.

The interpretation of tenses in relative clauses works somewhat differently. Consider the examples in (7) and (8):

(7) a. Adam gave an ice-cream cone to [a boy [who was sitting outside]]

b. Adam gave an ice-cream cone to [a boy [who is sitting outside]]

(8) a. Adam will give an ice-cream cone to [a boy [who was sitting outside]]

b. Adam will give an ice-cream cone to [a boy [who is sitting outside]]

Consider first the interpretation of past in the relative clauses in these examples. At first glance, past in (7a) seems to work like past in (3a), since the relative clause event-time (the sitting-time) can either include or precede the matrix event-time (the giving-time), suggesting that the sentence allows either a past-shifted or simultaneous construal. However, unlike the situation in (3a), the relative clause event-time can also follow the matrix event-time, provided that it still precedes the UT. This suggests that the relative clause tense is being interpreted indexically, i.e. like a main clause occurrence of past, as Enç (1987) and others have concluded. Although the relative clause past might, hypothetically, also have a simultaneous and/or past-shifted construal, it is hard to show this, since such readings are truth-functionally superfluous, given the availability of an indexical construal. (The indexical construal allows for the relative clause event-time to occur at any time preceding the UT, including those times allowed by the simultaneous and past-shifted construals.)

The situation is only slightly clearer in (8a). Here there appears to be a strong preference for the relative clause event-time to precede the UT, which is what we would expect if only an indexical construal were available. However, the sentence also seems to allow, albeit marginally, for a construal where the relative clause event-time follows the UT but precedes the matrix event-time, which is what we would expect if a past-shifted construal were available. This reading is much easier to get if the relative clause includes a time-adverbial that supports such a construal overtly:

(9) Adam will give an ice-cream cone to [a boy [who was sitting outside

a few minutes before]]

What should we conclude from this? My own guess is that a true past-shifted construal is basically unavailable in such cases, and that the presence of an overt time-adverbial allows for a different kind of interpretation that mimics the effect of a true past-shifted construal. The marginal availability of the past-shifted construal in (8a) could then be attributed to a covert time-adverbial, or, alternatively, to an element occuring in the prior discourse that has a similar effect. I will return to this issue shortly. Summing up the discussion of past in relative clauses, it seems that only an indexical construal is possible.

Now consider the interpretation of present in (7b) and (8b). In both cases, an indexical construal is possible; i.e. the relative clause event-time can include just the UT (and not the matrix event-time). However, in (8b), a simultaneous interpretation is also possible; the relative clause event-time can include just the matrix event-time (and not the UT). This suggests that relative clauses do allow for one kind of dependent tense construal (specifically, a simultaneous construal), contrary to our tentative conclusion in the previous paragraph. (This in turn suggests that a simultaneous construal of past in (7a) may be available too.) Finally, both (7b) and (8b) seem to allow for a double-access construal, where the relative-clause event-time includes both the UT and the matrix clause event-time. However, the double-access construal is truth-functionally superfluous, given the availability of the indexical construal, since the latter construal is compatible with the possibility of the boy's sitting outside for an extended period that encompasses both times. For this reason, I will remain agnostic on this point; given the indeterminacy of the evidence, theoretical considerations should decide the question.

I summarize the relative clause facts in (10):

(10) a. Indexical construals of past and present are always available in a relative clause;

b. Simultaneous construals of present are available in a relative clause if the matrix clause contains will (or, alternatively,if the matrix clause does not contain past);

c. Simultaneous of past are possibly available in a relative clause if the matrix clause contains past;

d. True past-shifted construals of past in a relative clause are probably unavailable, given (8a) vs. (9);

e. We cannot determine whether double-access construals of present in a relative clause are available.

In some respects, the picture here is simpler than it is with complement clauses. For one thing, relative clause tenses can always be interpreted as though they occurred in a main clause. Furthermore, the issue of the double-access construal does not arise (at least, not necessarily). The situation with simultaneous construals seems to be roughly the same as with complement clauses, although the facts are not entirely clear in the case of past in examples like (7a). On the other hand, the marginal status of the past-shifted construal for past in a relative clause in examples like (8a) poses a new puzzle, given that such a construal is necessary in the case of complement clauses.

Before examining the theoretical accounts of these generalizations, I would like to introduce one further factor that is relevant to our discussion, namely the contrast between eventive and stative verbs with respect to the various tense construals under discussion. Thus far, I have judiciously chosen examples to conceal the effects of this distinction, but it is widely recognized that the stative/eventive contrast is crucial. As a point of departure, consider past and present with eventive predicates in simple clauses analogous to (1):

(11) a. Arby ate an ice-cream cone.

b. Arby eats an ice-cream cone.

The use of past with an eventive predicate in (11a) is straighforward; it can be used to report a single (completed) event located prior to the UT. But the use of present with an eventive predicate is more complicated. It cannot be used to report a single event, located at the UT, regardless of whether the event in question is initiated at the UT, completed at the UT, or ongoing at the UT. To convey the latter sense, the progressive aspect is needed in such cases (with is eating substituting for eats). Put another way, the progressive aspect of an eventive verb behaves like a stative verb in being possible with the indexical use of present, but this is not possible for the simple form of an eventive verb. (It has often been observed that English differs from many other languages in this respect; for instance, the Italian or German counterpart to an example like (11b) can be used to report an ongoing eating-event.)

Although examples like (11b) cannot be used to report a single event located at the UT, they can be used with other senses. Most prominent of these is the generic or habitual interpretation; thus (11b) could be an answer to a question such as What does Arby do every evening? (I use the term generic because of examples like Dogs chase cats, which works the same way.) Actually, the generic/habitual construal is also possible with past; thus, (11a) would be an appropriate answer to a question like What did Arby (use to) do every evening? and Dinosaurs chased mammals would be an appropriate generic statement about a species that no longer exists. Thus, the generic/habitual construal of an eventive verb behaves just like a stative predicate; in its indexical use in a main clause, it locates a habit or generic property prior to the UT with past, and at the UT with present.

In addition to the generic/habitual construal, an eventive verb with present has at least two other special senses. First, an example like (11b) can be used as a caption under an illustration or photograph, or as a headline or title above a text, to label an event that is not located in time with respect to the UT (or any other time). In this type of context, present almost certainly conveys the absence of a tense. Second, present can also be used to report a (scheduled) event that has not yet occured at the UT, although there are a number of special inferences associated with this usage that I will not go into here. Apart from the generic/habitual construal, I will have nothing further to say about these special uses of present.

Turning our attention to complement clauses and relative clauses, we observe that the simultaneous construal of past and present behaves like the indexical construal of present in a main clause:

(12) a. Adam said [that Arby ate an ice-cream cone]

b. Adam will say [that Arby eats an ice-cream cone]

c. Adam said [that Arby eats an ice-cream cone]

(13) a. Adam gave a watch to [a boy [who ate an ice-cream cone]]

b. Adam will give a watch to [a boy [who eats an ice-cream cone]]

c. Adam gave a watch to [a boy [who eats an ice-cream cone]]

In (12a), only a past-shifted interpretation of the relative clause past is possible; the simultaneous interpretation available to the stative predicate in (3a) disappears. (A simultaneous reading is possible only if a quantifier over times such as every day is added, and the relative clause is interpreted as a generic/habitual state holding of Arby at the time of the reported saying-event.) In (12b), the simultaneous interpretation available in (2b) likewise disappears, except on a similar generic/habitual construal. Unsurprisingly, the double-access construal available in (2b) also disappears in both (12b) and (12c), with the same qualification; in fact, there is no non-habitual interpretation available for (12b) and (12c).

The relative clause tenses in (13) behave similarly; in each case, the simultaneous interpetation disappears. In (13a), only the indexical past construal from (7a) survives; interestingly, this cannot naturally be understood so that the two event times coincide; one event must follow the other, though the ordering appears to be arbitrary, abstracting away from whatever pragmatic inferences one might draw from a presumed causal relation holding between the events. In (13b) and (13c), the indexical, simultaneous, and double-access interpretations all vanish. (Interestingly, a new kind of interpretation seems to arise in (13b), which superficially resembles a past-shifted (!) construal; (13b) can mean something like "Adam will give a watch to a boy if he (the boy) has eaten an ice-cream cone". It can be shown, of course, that this reading is not a true past-shifted construal; rather, it is a special kind of modal construal licenced by a covert conditional. I will not examine the details of this here, however.)

What should we infer from this discussion? I believe that three essential points emerge. First, the fact that the simultaneous construals of present and past resemble the construal of the indexical present with respect to the stative/eventive distinction supports the view that these construals are closely related, and perhaps identical at a certain level of abstraction. Second, the fact that the double-access construals behave similarly supports the view that they are likewise closely related to the simple indexical usages. Finally, the fact that a simultaneous construal of past is blocked in (13a) suggests that a true simultaneous interpretation of past is actually possible in relative clauses if they contain stative verbs, as in (7a), parallel to the true simultaneous construal of present in the relative clause in (7b). This possibility was glossed over earlier, owing to lack of clear evidence (cf. (10b) and the preceding discussion). If the simultaneous construal in (7a) were simply a subcase of a more vague indexical construal of past, then one would expect such an ersatz simultaneous interpretation to be fortuitously available in (13a) too. That it is not suggests that some structural factor excludes the simultaneous reading in the latter case, while allowing it in the former; certainly there is no semantic problem with having two events coincide, as (14) shows:

(14) Adam gave a watch to Betty while Arby ate an ice-cream cone.

I conclude that relative clauses allow both indexical and simultaneous construals of both past and present, though my prior conclusion that they probably lack past-shifted construals stands.

2. Theories of the Present Tense. We are now in a position to consider the question of what present and past mean. The previous literature has provided two leading ideas about this. One possibility, explored in the work of Partee (1973), Enç (1987), Ogihara (1989) Abusch (1993), among others, is to consider tenses to be referential expressions. Just as nominal expressions (common noun phrases, names, and pronouns) refer to individuals (or events, in some cases), tenses refer to times. In particular, Partee has suggested that tenses are temporal analogues of pronouns, an idea that I will discuss in more detail below. Within such a framework, the contrast between English past and present can be thought of as analogous to the distinction between English he and she. Just as particular pronouns can refer to restricted classes of individuals (male humans vs. female humans), so particular tenses can refer to restricted classes of times. Thus past refers to a time prior to the UT, while present refers to a time that includes the UT (or perhaps it simply refers to the UT itself.) In this respect, past and present resemble time-adverbials such as today and yesterday. I will refer to this approach as the referential theory of present and past.

Another possibility, adopted in much of the early work on tense in generative grammar (e.g. McCawley (1973)), and developed recently by Zagona (1990), is to consider tenses as relational or predicative expressions, expressing a relation of temporal ordering between the UT and (the time of) the event or state expressed by the verb phrase. This amounts to treating tenses as dyadic predicates of temporal ordering; one argument of the tense predicate denotes the UT, while the other denotes (the time of) the event or state denoted by the verb phrase. On this view, the distinction between past and present involves a distinction between two different predicates of temporal ordering; past orders the UT after the time of the event or state denoted by the VP, while present is a predicate of simultaneity or temporal containment, ordering the UT within the time of a state denoted by the VP. I will refer to this approach as the predicative theory of present and past.

Prima facie, both the referential and predicative theories have some intuitive appeal. Let me begin by sketching out a version of the predicative approach, since it comes fairly close to the view that I wish to promote. Suppose that tenses are true

two-place predicates indicating the temporal ordering relation between a pair of time-denoting arguments. Suppose, further, that tenses, like other predicates, appear syntactically as heads of phrasal projections bearing their category-labels; thus a tense is the head of a TP, following Pollock (1989). If tenses are like other predicates, then they assign thematic roles to their two arguments as follows: the "external" (subject) argument is theta-marked in the Spec of TP position (adopting the theory of Stowell (1983)), while the internal argument of the tense is its complement (ZP in (15) below). The external argument of the tense is a phonetically null element (PRO-ZP in (15)) that is understood to denote the UT (in a main clause, at least), while the internal argument corresponds to the time of the event or state denoted by the verb phrase, as Zagona (1990) proposes; this arrangement makes sense if TP dominates VP, as is currently widely assumed. The complete structure for the temporal argument structure described above is given in (15):

(15) TP




Tense ZPi






[e]i VP*

The ZP ("Zeit-Phrase) complement of Tense in (15) is a time-denoting category, with a phrase structure similar to that of other referential expressions, such as DP, as described in Stowell (1993). Its head Z takes VP as its complement; either Z itself or an empty operator in the Spec of ZP saturates the VP by binding a null temporal ZP variable in a (VP-internal) temporal argument position theta-marked by the verb.

This ZP yields an interpretation along the lines of "the/a time t such that VP* holds of t", analogous to the translation of the man as "the individual x such that man holds of x". The (phonetically null) category occupying the Spec of TP in (15) is a ZP-analogue of PRO; we stipulate that this PRO-ZP denotes the UT when it has no possible c-commanding antecedent, as when it occurs in the Spec position of a main clause TP.

Assuming that past and present are tenses occupying the head position of TP in (15), we can now provide an account of the indexical uses of these tenses in (1a) and (1b) by analyzing past as a temporal ordering predicate meaning something like after, and present as a temporal ordering predicate meaning something like within. Assuming that the traditional subject originates within the verb phrase (Koopman & Sportiche (1991)), (1a) comes out as "the UT (is) after the time (at which) Arby know Adam", while (1b) corresponds to "the UT (is) within the time (at which) Arby know Adam".

So far I have said nothing about the exclusion of indexical present with an eventive predicate in the verb phrase. Suppose that the tense predicate present (meaning something like "within") imposes a selectional reqirement on its ZP complement, requiring that ZP denote a particular type of time interval that is compatible only with states. The essential insight here is that only stative intervals are capable of containing another time in the sense required by this predicate. While such a stipulation is obviously far from being a true explanation of the stative restriction, it will do for our purposes here. (Presumably past (meaning "after") imposes no such restriction, and allows for either state-times or event-times as denotations of its internal argument.)

Let us turn next to the question of the non-indexical uses of present and past. The predicative theory extends fairly naturally to the simultaneous construal of present in a complement clause; we need only assume that the PRO-ZP occupying the Spec of TP position falls under control theory when it occurs withing the c-command domain of possible controller. In general, this controller will be the VP-internal variable bound by the null operator in the Spec position of the matrix clause's event-time ZP (see (15)). Thus the simultaneous construal of (2b), repeated here, will work as follows:

(2) b. Arby will say [that he knows Adam].

In the main clause, the modal will (meaning "before") will place the UT prior to the matrix event-time ZP; the operator in the head of this ZP (coindexed with it) will bind the VP-internal temporal argument variable of the matrix verb; this in turn will control the PRO-ZP in the Spec position of the complement clause TP; finally, the complement clause present will order this time within the interval of the state-time of the complement clause VP. Thus, the predicative theory provides a unified account of the indexical and simultaneous construals of present.

The predicative theory of indexical past likewise extends naturally to account for the past-shifted construal in (2a) and (3a). Once again, the only difference between the indexical and dependent construals is the interpretation of the PRO-ZP in the pec of TP; in a main clause such as (1a) and in the matrix clause in (3a-b), this PRO-ZP has no commanding potential antecedent, and must therefore denote the UT by convention; but in a complement clause such as that in (3a), it is controlled by the event-time variable in the matrix VP and yields a past-shifted construal. The meaning of past itself is unchanged; only the effects of control theory distinguish the past-shifted construal of past from its indexical counterpart.

Problems arise, however, in accounting for the simultaneous construal of past in the complement clause in (3a) and the double-access construal of present in (3b). The former is problematic because of the interpretation that the predicative theory attributes to past; if control applies to the complement clause's PRO-ZP (as we expect), then only the past-shifted reading should arise; if, instead, we allow this PRO-ZP to denote the UT, then an indexical past construal is derived, which is incorrect, since it wrongly allows for the complement clause event time to follow the matrix event time. In effect, we want to interpret past here as though it meant present, but this is impossible without abandoning a unified account of what past means. The double-access construal of present in (3b) is problematic for a similar reason: if control applies to the PRO-ZP in the complement, as expected, we should derive a simultaneous construal. Even if control unexpectedly fails to apply, we should simply derive an indexical construal, with PRO-ZP denoting the UT. But what we actually get is a combination of these two.

Of course, we could "solve" these problems by claiming that past and present are both homophone pairs, with past being interpreted as present when it occurs under the c-command domain of a higher (true) past, and present having a complex double-access construal in the same environment. But this approach is surely dissatisfying. (It is perhaps worth pointing out that neither problem arises in languages like Russian and Japanese, whose "past" and "present" tense morphemes behave just as the predicative theory would predict: their "past" always yields a past-shifted construal in a complement clause, and their "present" can always yield a simple simultaneous construal, regardless of whether the main clause is past, present, or future.)

The facts of relative clause interpretation motivate an additional conclusion about the nature of LF representations. Recall that an important feature of our analysis is that reference to the UT in our theory is limited to structures where the external argument of a tense (the PRO-ZP in the Spec of TP) occurs in a structure where it has no c-commanding antecedent. In other words, neither the tense itself nor the PRO-ZP is a true indexical expression like now or pronouns such as you and me. This assumption is necessary in order to block the indexical construal of tenses in complement clauses. Therefore, in order to allow for an indexical construal of tenses in relative clauses, we must assume that their PRO-ZP external arguments appear at LF in a structural position that is not c-commanded by a matrix ZP. We can derive this result if we adopt the additional assumption that relative clauses (or the NPs or DPs containing them) undergo movement at LF out of the VP in which they originate. (The factor motivating this movement might be Case-theoretic in nature, e.g. if DPs must move to the Spec of a VP-external AgrP category.) The PRO-ZPs that they contain will then fail to have any c-commanding ZP in the matrix clause (apart from the PRO-ZP in the Spec of the matrix TP, which denotes the UT). Hence these PRO-ZPs will denote the UT too, correctly deriving the indexical construals that are characteristic of relative clause tenses, as we have seen.

Recall, however, that relative clauses also allow for exactly one type of dependent construal--the simultaneous construal, as shown by (7b) for present and by (7a) vs. (13a) for past. We might account for the simultaneous construal with present by assuming that the LF movement of the relative clause (or the DP containing it) is optional, rather than obligatory, so that the PRO-ZP external argument of the TP headed by present would be controlled in (7b) if the relative clause remains in situ. This analysis would extend to the simultaneous contrual of (13a) if we could find a way for past to behave as though it had the semantics of present; i.e. this reduces to the problem posed by simultaneous interpretations of past in a complement clause. But if LF movement is optional, we would be unable to explain why a true past-shifted construal of past in a relative clause is apparently blocked in (8a). Admittedly, the facts are not entirely clear-cut in (8a), as I observed with respect to (9), and perhaps we ought to conclude that such an interpretation is available after all, and seek an alternative account of the marginality of the construal in (8a). I will leave this question unresolved here.

Let us now turn to the referential account of the semantics of past and present, bearing in mind the problems that complement clause construals pose for the predicative theory. The version of the referential approach that I wish to consider first takes the position that present is itself an indexical expression denoting the UT, while past is an indexical denoting any time that precedes the UT. On this view, the tenses are not predicative categories at the clausal level, but rather referential arguments of the verb; i.e. they are to overt time-adverbials as pronouns are to DP arguments. Plainly such an approach can handle the indexical construals of main clause tenses and relative clause tenses straightforwardly; just as an indexical pronoun like you or me can refer directly, regardless of where it occurs in the syntactic structure, so can a tense. The problem with such an account, however, is that it fails to account for all of the dependent construals of tenses. In particular, it fails to allow for the simultaneous construal of both present and past in relative clauses and complements, and likewise fails to account for the past-shifted construal of past in complements. Finally, it can account for the double-access interpretation of present only by adopting special assumptions about the nature of indexical time reference in intensional contexts.

(Ogihara (1989) recognizes this, and is forced to posit that present is


Bibliographical references

Abusch, D. (1988), "Sequence of Tense, Intensionality, and Scope", WCCFL 7: 1-12.

Abusch, D. (1993), "Sequence of Tense Revisited: Two Semantic Accounts of Tense in Intensional Contexts", unpublished manuscript, University of Stuttgart.

Enç, M. (1987) "Anchoring Conditions for Tense", Linguistic Inquiry 18: 633-657.

Fillmore, C. & D.T. Langendoen, eds. (1971), Studies in Linguistic Semantics, New york, Holt Rinehart, & Winston.

Koopman, H. & D. Sportiche (1991), Lingua 85: 211-258.

McCawley, J. (1971), "Tense and Time Reference in English" in Fillmore & Langendoen (1971: 96- 113).

Ogihara, T. (1989), Temporal Reference in English and Japanese, doctoral dissertation, University of Texas, Austin.

Partee, B. (1973), "Some Structural Analogies Between Tenses and Pronouns in English", The Journal of Philosophy 70: 601-609.

Pollock, J.-Y. (1989) "Verb Movement, Universal Grammar, and the Structure of IP", Linguistic Inquiry 20.3, 365-424.

Stowell (1983), "Subjects Across Categories", The Lingustistic Review 2: 285-312.

Stowell, T. (1993) "Syntax of Tense", unpublished manuscript, University of California, Los Angeles.

Stowell, T. (to appear), "The Phrase Structure of Tense", in J. Rooryck & L. Zaring, eds., Phrase Structure and the Lexicon, Dordrecht, Kluwer.

Zagona, K. (1990) "Times as Temporal Argument Structure", unpublished manuscript, University of Washington.