In this article I present an outline of a theory of phrase structure that incorporates a series of functional projections forming the basis for the interpretation of tense. Most previous theories of tense have relied on an idiosyncratic set of semantic rules to account for the distribution and interpretation of particular tense forms. The theory that I develop here seeks to derive most of the semantics of tense from independently motivated principles of syntactic theory. To this end, I introduce a minimum of new theoretical machinery, though a more complex syntactic structure is required, involving an additional functional projection dominating the verb phrase and various phonetically null elements. In this respect, the theory is in the spirit of work in the Principles and Parameters framework associated with the interpretation of argument structure, thematic roles, and pronominal reference.

Temporal interpretation is a complex affair; tense morphemes can be construed in seemingly quite different ways, depending on the interaction of several factors, including scope relations with other tenses, verbal aspectual class (stative vs. eventive), grammatical aspect (progressive vs. perfective), verbal epistemological class (intensional vs. extensional), and clause type (complement clause vs. relative clause, and finite clause vs. infinitive). Needless to say, I cannot discuss all of these issues here. I will concentrate on the phrase structure associated with the English morphemes present and past and the elements forming the perfect construction (have plus past participle). In addition, I will be concerned with the semantic interpretation of constructions containing these elements. For a fuller discussion of these and related matters, I refer the reader to Stowell (1993, 1995, to appear).

As a point of departure, I would like to consider the past perfect (pluperfect) construction illustrated in (1), mainly as a vehicle for clarifying some points of terminology.

(1) The thieves had (already) escaped when the police arrived.

There are some general points of agreement about the interpretation and morphosyntactic composition of this construction. From a morphosyntactic perspective, the past perfect is composed of the preterite (simple past) form of the aspectual auxiliary verb have (had), followed by the past participle form of another verb. From a semantic perspective, the past perfect is used to locate an event or situation in the remote past. In the familiar rendition offered by Reichenbach (1947), the time of the event or situation associated with the thematic verb (the Event Time, E) is located prior to some other time (the Reference Time, R), which is itself located prior to the time at which the full sentence is uttered (the Speech Time, S).

In Reichenbach's treatment, as in many traditional descriptions of English grammar, the past perfect construction is referred to as a tense, one of a number of tenses in English. (Other Reichenbachian tenses include the simple present, the simple past, the future perfect, and so on.) Within the research tradition of generative grammar, however, the term tense has generally been used to refer specifically to the tense inflection of the finite verb form, which in English takes one of two forms: present or past (preterite). From this perspective, the past perfect is not a true tense per se; it is a complex syntactic construction that contains Tense as one of its component parts.

Although my terminology adheres in spirit to this latter usage, I draw a further distinction between the morphological tense features (or affixes) that show up on the inflected finite verb and the abstract syntactic X-bar category Tense (T) introduced by Pollock (1989), which plays a role in theory of phrase structure.⁽¹⁾ I use the italicized lower-case terms tense, present, and past to refer to the morphological tense affixes, and the uppercase terms TENSE, PAST, and PRESENT to refer to Pollock's X-bar phrase structure heads. When referring to tenses and times in an informal way, I use regular lower-case orthography. Thus PAST is an instantiation of TENSE (T), the X-bar theoretic head of TP, while past is simply a morphological affix. This distinction (past vs. PAST, etc.) is motivated by my account of the interpretation of various constructions containing present and past, which I will touch on only very briefly here.⁽²⁾

Let us now consider the syntactic status of TENSE and its relation to the English morphological tense affixes. As the null hypothesis, I will assume that English present and past originate as affixes on V, and that they must be licensed by virtue of occurring syntactically in the T⁰ position at some point in the mapping to Logical Form (LF), presumably by means of head-to-head verb movement. Thus, either tense occurs in T⁰ at LF, or it occurs in some higher position at LF, and binds a chain-internal trace in the T⁰ position. For my purposes, it doesn't matter precisely how tense and TENSE are connected in mechanical terms; the central empirical assumption is that tense uses the T⁰ position as its "base of operations" in the syntax of LF representations. I must emphasize that I am adopting this standard assumption only as a point of departure.

Of course, this does not tell us what the semantic content of TENSE is. Semantic analyses of Tense are diverse; it has been analyzed (a) as a sentential operator designating the time at which the truth of a sentence is to be evaluated; (b) as a referential expression denoting the Event Time or Reference Time (or some related notion); and (c) as a temporal predicate analogous to a verb. In the theory developed here, the traditional category Tense is broken down syntactically into its component semantic parts, roughly along the lines suggested by Zagona (1990). Tenses are dyadic predicates of temporal ordering, and as such do not refer; however, they take two time-denoting phrases as their arguments, and it is in these categories that the referential properties associated with tenses reside.

Zagona's treatment is concerned primarily with tenses in main clauses. Simplifying somewhat, she posits a tense predicate located in the head position of a functional category F⁰; I will translate this into the terms of Pollock's (1989) theory of inflection, by equating her F⁰ with Pollock's TENSE (T⁰). Zagona's TENSE predicate selects two arguments; its internal argument (i.e. its complement) is VP, which denotes the Event Time E (according to her). The external argument of TENSE originates in the Spec of TP position and denotes the "evaluation time", which for her is equivalent to the Speech Time S. The notions of "Event Time" and "Speech Time" are treated by Zagona as theta-roles assigned by the tense predicate, an idea that I will adopt here in a modified form.⁽³⁾

In Zagona's theory, the TENSE head has no intrinsic semantic content, apart from whatever theta-roles or features it assigns to its arguments. For her, the distinction between present and past tense is encoded as a feature assigned to the internal VP argument of T. If VP is assigned [+PAST], it is interpreted as bearing "a [+PAST] relation to the time of speaking". Although she states that "the ordering of times is semantic rather than syntactic in nature", she does not give a precise characterization of what that semantics is. In her view, the sole contribution of the syntax to tense interpretation is in the domain of Binding Theory, in terms of the coreference relations among temporal arguments that it allows or prohibits. For instance, a [+PAST] VP (the complement of a past tense) has the binding properties of Chomsky's (1981) R-expressions: it must be disjoint in reference to the external argument of T (the Speech Time, S). On the other hand, a [-PAST] VP (the complement of a present tense) can be bound by the external argument of T, though it need not be.

The theory that I wish to defend here adopts certain features of Zagona's account, while rejecting others. Like her, I assume that TENSE is a predicate head, taking two time-denoting arguments, and that the internal argument denotes the Event Time (E). I will also adopt her idea that the time-denoting arguments of TENSE are subject to Binding Theory, just like other types of referential expressions (pronouns, definite and indefinite descriptions, names, anaphors, etc.). However, my theory differs from hers in a number of other ways. First, I assume that temporal ordering relations are reflected in the syntax, in the form of the inherent predicative content of the TENSE predicate. Just as the inherent content of the preposition after specifies a temporal ordering relation holding between its two event-denoting arguments in a sentence such as The party is after the class, so the inherent predicative content of PAST establishes an analogous temporal ordering relation between its two arguments (S and E) in John hit the ball. Thus, the primary semantic function of TENSE is to temporally locate the denotation of E in relation to the denotation of its external argument. This makes it unnecessary to assume that T assigns a [+PAST] feature to its internal argument; the notion of "past" resides in T itself, in the form of PAST (roughly synonymous with after).

A second difference between my theory and Zagona's is that I posit an additional functional category ZP (Zeit-Phrase) intervening between TP and VP; it is this ZP category that serves as the time-denoting internal argument of T, denoting the Event Time E. ZP is a referential category analogous to DP (cf. Abney (1987), Szabolcsi (1987), Stowell (1989)). Loosely speaking, ZP bears a structural relation to VP that is analogous to the relation that DP bears to NP. For the sake of concreteness, I assume that ZP denotes a time, and that some element at the ZP-level (for concreteness, an operator residing in the Spec of ZP) binds a temporal variable within VP, corresponding to Kratzer's (1988) "external" event argument.

Concerning the status of the "external" (subject) argument of T, I assume, following Zagona, that it originates in the Subject position of TP, where it is assigned a temporal theta-role by T. However, I assume no necessary connection between the temporal theta-role assigned to this position and the denotation of Speech Time (S). I will refer to this external temporal theta-role assigned by T as the Reference Time argument (R). My use of the term "Reference Time" differs in an important way from the traditional sense of the term originating with Reichenbach (1947); I will compare these two usages further below. For present purposes, it is sufficient to think of the "Reference Time" argument in the following terms: like the internal ZP argument, it denotes a time, so presumably it too has the categorial status of ZP. Like the internal ZP argument, it has no fixed indexical denotation; it simply refers to a time relative to which the Event Time ZP is ordered. In a main clause, it so happens that the Reference Time ZP denotes the Speech Time S, but in subordinate clauses, it typically denotes the Event Time E of the immediately higher clause. I will discuss the mechanics of how this works later on.

Thus the sentence John hit the ball would have the structure in (2):

(2) TP

"R"--> ZP T'

(="S")

T ZP_i <-- "E"

PAST OP_i Z'

"after"

Z VP

ZP_i VP

[e]

DP V'

John

hit the ball

In (2), the external ZP argument of PAST is the Reference Time (here, this ZP denotes the Speech Time S); the internal ZP argument denotes the Event Time; and PAST itself establishes an ordering between these two times. The structure translates as: "The Speech Time is after a time ZP_i at which John hit the ball."

By breaking down Tense in this way into its three component parts (TENSE, the event-time ZP, and the reference-time ZP), we can integrate each part into the theory of syntax somewhat independently of the others. This integration makes it possible to account for the complexities of tense interpretation in terms of the interaction of various familiar principles from established modules of grammar. The semantics of TENSE itself is straightforward; PAST means "after"; will means (at least) "before", and PRESENT (if it exists as a true TENSE) means "simultaneous with" or "overlaps". The semantics of the event time ZP, the internal argument of TENSE, is analogous to that of a definite or indefinite DP. With a stative predicate, the event time ZP is generally understood to have definite reference; in other words, it is typically understood as referring to a time already mentioned. With an eventive predicate, the event time ZP can have either definite or indefinite reference: it can either refer to a time already mentioned, or it can introduce a new time.

The denotation of the reference time argument is fixed by control theory. Just as the overt complement of tense (the event-time ZP) is a temporal analogue of a definite or indefinite DP, so the null external argument of Tense (the reference-time ZP) is a temporal analogue of PRO. When PAST occurs in a main clause, as in (2), its external reference time argument is a PRO-ZP lacking any c-commanding ZP to serve as its controller; in this case, the PRO-ZP behaves like an indexical; it denotes the Speech Time S, and PAST orders the Event Time E in relation to S. When PAST occurs in a complement clause, as in (3),

(3) John said [that Bill hit the ball]

its reference time argument behaves like control PRO; the PRO-ZP is controlled by the closest c-commanding ZP--the temporal argument of the matrix verb (i.e. the variable bound by the head Z of the matrix event-time ZP), as in (4). This gives rise to "shifted" tense construals, where the complement clause Event Time is fixed in relation to the matrix clause Event Time; the structure translates as: "The Speech Time is after a time ZP_i at which John say that ZP_i is after a time ZP_j at which Bill hit the ball."

In both (2) and (4), I locate the traditional subject DP within a projection of VP at D-structure. I have omitted from (2) and (4) the traditional Spec position in TP to which (or through which) the subject DP raises. This is purely for the sake of simplifying the visual presentation of the tense structure; strictly speaking, these trees are incomplete. I assume that TP contains both a thematic subject position for its external ZP argument and a nonthematic Spec position for the traditional subject DP. Subject raising is clearly insensitive to the presence of the external temporal argument of TP, just as control of PRO-ZP is insensitive to intervening DP arguments in the matrix VP. Presumably ZP control relations and DP chains can interweave through the structure because minimality conditions on binding relations are relativized in the sense of Rizzi (1990).

(4) TP

ZP T'

PRO_arb

(=S) T ZP_i

PAST

OP_i Z'

Z VP

controller--> ZP_i VP

[e]

DP V'

John

V CP

say

C TP

that

ZP_i T'

controlled--> PRO

T ZP_j

PAST

OP_j Z'

Z VP

ZP_j VP

[e]

DP V'

Bill

hit the ball

An important feature of Zagona's thematic argument structure for tense, adopted here, is that it specifies a relation holding between just two times. This is a significant departure from Reichenbach's (1947) view, according to which tenses involve a set of relations holding between three time co-ordinates on a time-line diagram. Although Reichenbach's system has been largely abandoned in recent accounts of tense in the formal semantics literature, it has been widely adopted (in modified form) within the formal syntax literature, especially within the Principles and Parameters framework (see, for instance, Hornstein (1977, 1990), Bouchard (1984), Vikner (1985), and Giorgi and Pianesi (1991).)

As noted above, Reichenbach's account of the English "tenses" provides a set of ordering relations holding among three primitive time coördinates, including not only the Event Time (E) and the Speech Time (S) but also a third time--the Reference Time (R). These time coördinates can be either simultaneous (overlapping) or discrete and temporally ordered relative to each other. Thus, the past perfect is analyzed as E-R-S (E before R; R before S); the present perfect is E-RS (E before R and S; R and S simultaneous); the simple past is supposed to be ER-S (E and R simultaneous; both before S). Reichenbach held that the semantics of every tense involves a similar scheme, arranging E, R, and S in various ways. Rephrased in terms of temporal argument structure, this is a claim that all tenses are three-place (complex) predicates of temporal ordering.

Various objections to specific analytical details of Reichenbach's treatment have been raised in the literature, mostly stemming from his lack of concern with the relationship between the morphosyntactic composition of the various tense constructions and their semantic interpretation. (Reichenbach assumed the morphosyntactic makeup of the various tenses as a given, and did not seek to provide a true syntactic analysis in the modern sense.) Bouchard (1984) and Vikner (1985) point out that one cannot provide a uniform one-to-one semantic translation for individual morphosyntactic elements in English and other languages, given Reichenbach's particular analyses of various tense constructions; they suggest a number of analytical revisions to make such a translation possible.

Comrie (1985), Vikner (1985), and Hornstein (1990) argue that Reichenbach's three-place relation involving S, R, and E draws distinctions that no natural language actually makes use of. In some cases, several Reichenbachian "tenses" correspond to a single natural language tense construction, as with the simple future and future perfect. These authors conclude that Reichenbach was wrong in assuming that E and S are directly ordered relative to each other, and suggest that his three-place relation should be replaced by a pair of binary relations--between S and R on the one hand, and between R and E on the other. These proposals maintain Reichenbach's notion of Reference Time more or less in the form that he proposed; they also accept his claim that all tenses involve three time-coördinates (i.e. three temporal arguments). Translated into the framework of temporal predicate argument structure adopted here, Vikner and Hornstein view each tense as a complex predicate, composed of one predicate that orders R relative to S, and another predicate that orders E relative to R.⁽⁴⁾

Reichenbach's Reference Time (R) provides a formalization, for the purposes of tense interpretation, of the traditional notion of "point of view". This has considerable intuitive appeal, but mainly in the case of the perfect tense constructions. In the past perfect (E-R-S),the location of R between E and S captures an intuition that the event is viewed from the perspective of a past time subsequent to it. In the present perfect (E-RS),the alignment of R with S captures an intuition that the Event is being viewed from the perspective of the Speech Time. For Reichenbach, the future perfect is a conflation of three different tenses involving various relations between S and E (E-S-R, SE-R, and S-E-R), but Comrie (1985), Vikner (1985), and Hornstein (1990) all argue persuasively that there is just one future perfect tense construction, with S and E simply unordered with respect to each other.

Although there is an intuitive basis for Reichenbach's use of R in the perfect tense constructions, this is far from clear elsewhere. In the simple past (Reichenbach's ER-S), there is no intuitive sense that the event is necessarily being viewed from the perspective of the time at which it occurs, except perhaps in strings of narrative text, where other explanations for the effect are possible.⁽⁵⁾ The same is true of the future modal will, which lacks any particular sense of a distinctive temporal point of view. Although the simple present tense certainly has the Speech Time as its point of view, we don't need Reichenbach's R to account for this. The bottom line seems to be that Reichenbach's R is needed only in the perfect constructions⁽⁶⁾

.

Although Reichenbach's formalism succeeds in distinguishing the simple past from the present and past perfect, one could capture the same distinctions by eliminating R from the semantic representation of the simple past, treating it simply as E-S. Similar remarks hold for the simple present, which could be characterized as ES (i.e. as a predicate indicating simultaneity of E and S). In fact some authors, notably Enç (1990), have suggested that the present "tense" need not be thought of as a tense at all, since it induces no temporal shift of the evaluation time away from its starting point (the Speech Time, in a main clause).

This alternative treatment of the simple present and past tenses is in keeping with the dyadic structure for tenses proposed above. Although my variation on Zagona's structure makes use of a Reference Time argument, this is not Reichenbach's notion of R. In his theory, S and R have an independent status alongside E in the argument structure of every tense. In contrast, the system proposed here does not treat the Speech Time (S) as a theta-role per se; the Speech Time is simply one of the possible denotations of the Reference Time argument, as in (2). Consequently, no single tense can have both S and R as independent arguments, so there is no way to represent the semantics of the past perfect or future perfect as a single tense.

On the other hand, it is relatively simple to represent the perfect tenses as complex temporal constructions, composed of two elementary tense predicates, as Vikner (1985), Zagona (1990), and many others, have recognized. This assumption allows us to derive the temporal argument structure of these constructions from their morphosyntactic structure in an obvious way. Suppose that the past participle contains a past tense predicate synonymous with PAST, so that the past, present, and future perfect tense constructions are each composed of two tenses; in this limited sense, the perfective constructions are biclausal. Suppose, further, that the perfect auxiliary have denotes a result state, with an Event Time argument like any other stative verb. Then the higher tense (PAST, PRESENT, or will) provides its usual temporal argument structure, locating the time of the result state (the time of "having") in relation to a Reference Time (i.e. the Speech Time, in a main clause). The past participle provides an additional PAST tense, which orders the thematic verb's Event Time in relation to that of the result state. The D-structure of the past perfect in (5) emerges as (6).

(5) John had telephoned

The PrtP in (6) has a status analogous to that of the embedded clause TP in (4); its external argument PRO-ZP is controlled by the temporal argument of the minimal VP dominating it. Thus (6) translates as: "The Speech Time is after a "having" time ZP_i, which is after a time ZP_j, at which John telephone". The "having time" ZP_i in this translation of (6) corresponds to the intuitive intermediate past-time perspective point from which the main event time is viewed. Reichenbach's system characterizes this intermediate time simply as R; but in (6), ZP_i is both the Event Time of the higher tense predicate PAST, and the Reference Time of the perfective tense predicate -en "PAST".

(6) TP

ZP T'

PRO_arb

T ZP_i

PAST

OP_i Z'

Z VP

controller--> ZP_i VP

[e]

V PrtP

have

ZP_i Prt'

controlled--> PRO

Prt ZP_j

-en

"PAST" OP_j Z'

Z VP

ZP_j VP

[e]

DP V'

John

telephone

So far I have not discussed the status of temporal adverbial phrases. These are usually described in the literature as "adjuncts", but from the perspective of the theory of temporal argument structure explored here, this notion requires some rethinking. Temporal adverbials are usually thought to denote the time of the event associated with the VP in which they occur. In terms of the structure for tense given above, this implies that temporal adverbials are associated syntactically either with the ZP complement of TENSE (its internal argument) or with the VP-internal variable (the position theta-marked by the verb as its temporal argument). It is tempting to think of the temporal adverbial as being the verb's temporal argument, replacing the VP-internal ZP variable in (2) and (4). But the variable is needed for the proper interpretation of the ZP structure, so it is more likely that temporal adverbials are adjunct modifiers, either of the VP-internal variable or of the full ZP dominating VP. (This is consistent with the adjunct-like behavior of temporal adverbials with respect to extraction.) For concreteness, I will assume that a temporal adverbial is a restricting modifier for the VP-internal variable (the verb's temporal argument). Whether or not this adverbial modifier is actually adjoined to the variable need not concern us here.

The discussion thus far has treated all temporal adverbials alike. Actually, it is well known that they are a diverse lot. Some denote intervals (for five minutes), and are incompatible with Vendler's (1967) achievement verbs, while others denote instantaneous points (at ten past six), and are incompatible with his accomplishment verbs. Some are clausal (when Bill left the room), while others are not (then). Some are indexical (yesterday, now), others are anaphoric (then) or relational (the day before); while others are referential (at six o'clock). I will not attempt to give a precise account of any of these distinctions here, but a few brief remarks are in order.

The distinctive properties exhibited by Vendler's aspectual verb classes, with respect to the kinds of adverbials that they are compatible with, clearly reveals that (at least from a semantic perspective) the notion of the "event time" has a fine substructure that distinguishes among states, activities, achievements, and accomplishments. Campbell (1989) captures these distinctions in terms of thematic role assignment, by allowing the verb's argument structure to include not only the temporal interval or point corresponding to the duration of the event, but also the event's onset time and completion time. One might also conceive of Vendler's distinctions in terms of traditional S-selection, in the sense of Pesetsky (1982); just as a verb can select an animate DP argument, so it can select an interval-denoting ZP argument. The structure that I have proposed here does not provide any syntactic means for expressing Vendler's aspectual class distinctions in structural terms, and perhaps it should be enriched in order to do so. However, the differences among Vendler's event-classes (achievements, activities, and accomplishments) seem to have little effect on tense construal, apart from the well-known effects on temporal adverbial selection. On the other hand, the distinction between stative and eventive verbs is quite clearly relevant for tense construal, as virtually all studies of the syntax and semantics of tense recognize. For instance, only stative verbs can be used with the simple present in English, unless a generic/habitual reading is intended. Perhaps this indicates a fundamental difference between stative and eventive verbs with respect to the syntactic status of their temporal arguments, but I will not pursue this here; see Stowell (1995) for discussion.

Clausal temporal adverbials also raise interesting issues, but these mainly concern the internal structure of these clauses (e.g. concerning the application of Wh-movement of when or its phonetically null twin in after the police arrived). One curious property of these clauses, which is surely relevant for the theory of tense, is that they must bear the same tense:

(7) a. John arrived when/after/before the police were (*are) here.

b. John will arrive when/after/before the police are (*were) here.

This fact probably tells us something significant about the structure of these clauses and about the logic of English tense morphology, but I will not speculate about these matters here. Regardless of their internal structure, these clauses behave externally like other adverbials; they denote times and function as adjunct modifiers of the event time variable in a similar way.

I would now like to return to the question of how English tense relates to the syntactic category TENSE. So far, I have built a theory of tense interpretation around the idea that the head of TP, TENSE, is a temporal ordering predicate, and I have given no reason to doubt that tense is simply the phonological form of TENSE; in other words, past is the spellout of PAST. Every occurrence of past has been accompanied by the semantics of PAST, so it is reasonable to equate them. However, the interpretation of the subordinate clause past in (8) poses a problem:

(8) John said that Bill was sick

The subordinate past can be interpreted in two ways. First, it can refer to a time T' that is prior to John's saying time; this corresponds to the past shifted reading in (4). This past shifted reading seems to be possible with stative verbs only if T' is already under discussion,⁽⁷⁾ suggesting that the VP-internal variable here may be bound by a null Topic ZP denoting T'.

In any event, a more serious problem concerns another reading of (8), where the Event Time of the subordinate clause (i.e. the time of Bill's state of sickness) is interpreted as being simultaneous with the main clause event time (the time at which John spoke). This illustrates the phenomenon of "sequence of tense", where a past tense in indirect discourse corresponds to a present tense in direct quotation. The theory of tense construal advocated above incorrectly excludes this reading, if past means PAST, since the matrix verb's temporal argument obligatorily controls the Reference Time argument of the subordinate clause PAST, and PAST orders this time after (rather than simultaneous with) the subordinate clause event time. Therefore some aspect of the analysis must be revised. It would be inappropriate to abandon the control analysis of the PRO-ZP Reference Time argument, since this yields the correct result for eventive verbs. Rather, it seems that we must allow for the possibility that past (sometimes, at least) does not really mean PAST.

The decomposed temporal argument structure developed here suggests a possible revision to accommodate this "simultaneous" reading. Suppose that English past is really an instantiation of the head Z of the time-denoting category ZP, rather than an instantiation of T, the head of the predicative category TP. (In a sense, this represents a partial return to Enç's (1987) view that tenses are referential expressions.) Thus Z is overt (alternating between past and present) while T itself is non-overt (null), at least when it is headed by PAST or PRESENT. (Obviously the future modal will is overt, and this must be treated as a true tense predicate, since it never yields a simultaneous construal.) This being the case, we are free to assume that the subordinate clause in (8) contains no occurrence of PAST, so that the subordinate clause Event Time need not be ordered prior to any other time. Instead, we are free to assume that the subordinate clause contains either a null PRESENT, or perhaps no TENSE at all, in the spirit of Enç (1990). Suppose, for concreteness, that the null T in the subordinate clause is free to assume the value of either PAST or PRESENT. In the former case, the structure will yield the past shifted reading; in the latter case, it will yield the simultaneous construal.

An obvious question arises immediately, however. If past does not mean PAST, and does not even require the presence of PAST in the same clause, then why is it that whenever past occurs on a verb in a simple sentence, the Event Time is always understood as being ordered prior to the Speech Time? The idea that present and past are both instantiations of Z (rather than of T) obviously does not explain the difference between them. My answer to this question, first proposed in Stowell (1992), is developed further in Stowell (1993, to appear); I will present a synopsis of the solution here.

Recall that I have suggested that the event time ZP is normally interpreted in a way similar to a definite or indefinite noun phrase, with Z having a status analogous to that of a determiner. Suppose that we now draw a more articulated analogy between Z and D, by comparing present and past with the quantificational determiners some and any. It is well known that some and any can be analyzed as existential quantifiers, distinguished from each other chiefly in terms of their scopal properties; any must appear at LF under the scope of Negation or another downward-entailing polarity-inducing element, while some must appear outside the scope domain of any such element (cf. Ladusaw (1979).) Suppose that past is also a kind of polarity item, with an LF scope position fixed under PAST; in other words, past is a PAST Polarity Item (PPI). When past occurs as the head of the event time ZP in a simple clause, that clause must contain PAST in T, or else the PPI past would not be licensed. On the other hand, when past occurs in a subordinate clause, as in (8), its polarity requirements can be satisfied by virtue of its appearing at LF under the c-command domain of the matrix PAST, whose presence is forced independently by the polarity requirements of the matrix PPI past. Hence the subordinate clause can freely contain either (null) PAST or (null) PRESENT in its T position, providing the desired ambiguity between the past shifted and simultaneous construals in (8). On the other hand, present is an "anti-PP"; just as some may not occur under the scope of Negation, so present may not occur under the scope of PAST; for this reason, a clause containing present tense morphology on the finite verb must be understood to contain (null) PRESENT rather than (null) PAST in T.

This is, of course, only a sketch of an analysis; for a more thorough justification of the claim that English tense morphology encodes polarity relations, the reader is referred to Stowell (1993, to appear), where more complex structures (involving "independent" and "double-access" tense construals) are analyzed in these terms. However, I hope to have conveyed the basic idea of how this theory can be applied to account for the familiar phenomenon of "sequence of tense", without appealing either to a "sequence of tense" rule or to a non-unified treatment of the semantics of past. It is perhaps worth observing, in this context, that the analysis of past as a PPI can be applied to imperfect tenses in many European languages, more or less directly. It can also be applied to the use of past in conditional clauses in English, which exhibits a similar polarity behavior, unrelated to any past tense meaning. On the other hand, this analysis cannot be applied either to the English perfect (which never yields a simultaneous construal) or to the perfective tenses of other languages; these must be regarded as containing a true PAST temporal ordering predicate, always resulting in a "past shifted" construal.

I have proposed a phrase structure for tense, based on the idea that TENSE is a dyadic temporal ordering predicate, taking two time-denoting phrases as its arguments, in the spirit of Zagona (1990). I have also proposed a new functional category ZP, a referential (time-denoting) category modeled on DP. In English, T is null, but Z can be instantiated overtly by either present or past, at least in a finite clause; the choice between these elements is determined by the scope of their ZP in relation to PAST at LF. The internal argument of T has the referential properties of a definite or indefinite referring expression. The external argument has the referential properties of PRO; when it occurs in a subordinate clause, its denotation is fixed by control theory. Finally, I have argued that it is possible to account for the semantics of tense strictly within the bounds of a dyadic argument structure, without appealing to a Reichenbachian "Reference Time" R distinct from the Speech Time S. The desired effects of Reichenbach's R are captured here by assuming that the perfective tense constructions contain two dyadic temporal ordering predicates, together with a verb denoting a result state.

¹ This distinction is implicit in Pollock (1989) and most other generative accounts. Nevertheless these studies generally assume that a given occurrence of morphological tense affix originates in (or must be licensed by) the X-bar head T. The theory of English tense morphology developed in Stowell (1992, 1993, to appear) explicitly rejects this assumption, in a way that will be made clear further below.

² For fuller discussion, see Stowell (1993).

³ In addition, she suggests that this predicate sometimes assigns a theta role "reference time", more or less in the spirit of Reichenbach's (1947) theory (see below.)

⁴ Actually Vikner proposes an additional complexity that I will ignore here.

⁵ For instance, suppose that narrative requires a temporal ordering of events in a chronological sequence. One way to encode this formally is to suppose that the Event Time of a given sentence S will serve as the antecedent for a null Topic time-denoting phrase binding an Event Time variable in the VP of the clause immediately following S.

⁶ Bertinetto (1982) and Kamp and Reyle (1993) have also proposed to refine Reichenbach's notion of Reference Time, distinguishing between the notion relevant to the semantics of the perfect construction and a distinct notion relevant to temporal ordering in texts.

⁷ I am grateful to R. Boogaart for drawing this fact to my attention. See also fn. 5.

Abney, Steven: 1987, The English Noun Phrase in Its Sentential Aspect, unpublished

Ph.D. dissertation, Massachusetts Institute of Technology, Cambridge, Massachusetts.

Bertinetto, Pier Marco: 1982, 'Intrinsic and extrinsic temporal reference. On restricting

the notion of 'reference time'', Journal of Italian Linguistics, 71-108.

Bouchard, Denis: 1984, 'Having a Tense Time in Grammar', Cahiers Linguistiques

d'Ottawa 12, 89-113.

Campbell, Richard: 1989, The Grammatical Structure of Verbal Predicates, unpublished

Ph.D. dissertation, University of California, Los Angeles.

Chomsky, Noam: 1981, Lectures on Government and Binding, Foris, Dordrecht. Comrie, Bernard: 1985, Tense, Cambridge University Press, Cambridge.

Enç, Mürvet: 1987, 'Anchoring Conditions for Tense', Linguistic Inquiry 18, 633-657.

Enç, Mürvet: 1990, 'On the Absence of the Present Tense Morpheme in English',

unpublished manuscript, University of Wisconsin, Madison.

Giorgi, Alesandra, and Fabio Pianesi: 1991, 'Toward a Syntax of Temporal

Representations', Probus 2, 187-213.

Hornstein, Norbert: 1977, 'Towards a Theory of Tense', Linguistic Inquiry 8, 521-57.

Hornstein, Norbert: 1990, As Time Goes By, MIT Press, Cambridge, Massachusetts.

Kamp, Hans, and Uwe Reyle: 1993, From Discourse to Logic, Kluwer, Dordrecht.

Kratzer, Angelika: 1988, 'Stage Level and Individual Level Predicates', in M. Krifka,

ed. Genericity in Natural Language. Universität Tübingen.

Ladusaw, William: 1979, Polarity Sensitivity as Inherent Scope Relations, unpublished

Ph.D. dissertation, University of Texas, Austin.

Pesetsky, David: 1982, Paths and Categories, unpublished Ph.D. dissertation,

Massachusetts Institute of Technology, Cambridge, Mass.

Pollock, Jean-Yves: 1989, 'Verb Movement, Universal Grammar, and the Structure of

IP', Linguistic Inquiry 20, 365-424.

Reichenbach, Hans: 1947, Elements of Symbolic Logic, MacMillan, New York.

Rizzi, Luigi: 1990, Relativized Minimality, MIT Press, Cambridge, Massachusetts.

Stowell, Tim: 1989, 'Subjects, Specifiers, and X-bar Theory', in M. Baltin and A.

Kroch, eds., Alternative Conceptions of Phrase Structure, University of Chicago

Press, Chicago.

Stowell, Tim: 1992, 'Past Polarity', talk given at the 1992 NELS meeting, University

of Ottawa.

Stowell, Tim: 1993, 'Syntax of Tense', unpublished manuscript, University of

California, Los Angeles.

Stowell, Tim: 1995, 'Times, States, and Events', unpublished manuscript, University of

California, Los Angeles.

Stowell, Tim: to appear, 'What is the Meaning of the Present and Past Tenses?', in

P.-M. Bertinetto and M. Sqartini (eds.), Proceedings of the Cortona Tense-Aspect

Meeting, Scuola Normale Superiore, Pisa.

Szabolcsi, Anna: 1987, 'Functional Categories in the Noun Phrase', in I. Kenesei, ed.,

Approaches to Hungarian 2, University of Szeged.

Vendler, Zeno: 1967, Linguistics in Philosophy, Cornell University Press, Ithaca, New

York.

Vikner, Sten: 1985, 'Reichenbach Revisited: One, Two, or Three Temporal Relations?',

Acta Linguistica Hafniensia 19, 81-98.

Zagona, Karen: 1990, 'Times as Temporal Argument Structure', unpublished manuscript,

University of Washington, Seattle.

1.

2.

3.

4.

5.

6.

7.