You are on page 1of 23

Towards a Referential Analysis of Temporal Expressions Author(s): Mrvet En Source: Linguistics and Philosophy, Vol. 9, No. 4 (Nov.

, 1986), pp. 405-426 Published by: Springer Stable URL: http://www.jstor.org/stable/25001253 Accessed: 25/02/2009 01:45
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=springer. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that promotes the discovery and use of these resources. For more information about JSTOR, please contact support@jstor.org.

Springer is collaborating with JSTOR to digitize, preserve and extend access to Linguistics and Philosophy.

http://www.jstor.org

MURVET

EN(

TOWARDS

A REFERENTIAL

ANALYSIS

OF TEMPORAL

EXPRESSIONS

1. INTRODUCTION

Natural language tenses are generally treated as sentential operators in semantic theories. What motivates such treatments is the widespread assumption that the tense of a sentence affects the temporal inter pretation of all expressions in the sentence. This in turn rests on the assumption that all expressions are interpreted relative to times.1 This paper has two goals. First, it aims to show that, in a tensed clause, verbs are the only expressions whose interpretation is neccessarily affected by the tense, and that this point is obscured if one insists on semantically analyzing tenses as sentential operators. Second, it aims to show that only predicates of individuals are interpreted relative to times, suggesting that times should be viewed as individuals selected as arguments by the predicates in addition to non-temporal arguments.

2. THE

SCOPE OF TENSE

AND NP'S

Classical analyses, where tense is treated as a sentential operator, automatically make predictions about the temporal interpretation of NP's. I will first summarize these predictions, then show them to be empirically inadequate. The evidence provided will indicate that the interpretation of NP's is temporally independent of the tense that is present in the syntax. This will provide the most significant evidence against treating tense as a sentential operator. At first glance, a semantic analysis which treats tense as a sentential operator seems to account for certain temporal ambiguities in a simple and elegant way, by appealing to differences in scope relations between NP's and the tense. For example, (1) is taken to be ambiguous, depending on whether we are talking about a past president or the present president. (1) the president was a fool

The ambiguity of (1) is explained by saying that the denotation of the noun president will vary according to whether or not it is in the scope of the past tense operator. This is accomplished by providing semantic rules for tensed sentences along the lines in (2).2
Linguistics and Philosophy 9 (1986) 405-426. ? 1986 byD. Reidel Publishing Company.

406

MURVET

ENC

(2)

Where
(i)
at

qpis a sentence, P is Past and F is Future,


t iff there is a time t' such that <p is true
t' and t' < t,

Pqp is true at time

(ii)

Fcp is true at time


at t' and t< t'.

t iff there

is a time

t' such that qp is true

According to (2), the interpretation of a sentence with past tense or future tense involves evaluating the untensed sentence at a past time or a future time. Thus the tense operator in effect shifts the evaluation time. For example, Mary came is true when uttered at time t if and only if
there is a time t' prior to t such that Mary come is true at t'. And of is

course Mary come will be true at t' just in case the individual denoted by
of the set denoted at t' is a member Mary by come included in the set of individuals who came at t'. at t', i.e. ifMary

Assuming that names are rigid designators, the denotation of Mary will be the same at all times, therefore the scope of the tense operator will not affect its interpretation. However, when the head of the NP is a set denoting non-rigid noun like president, bird, brain, the scope of tense will be crucial in determining the denotation of the NP. Going back to our example in (1), suppose that tense has wide scope over the subject. Then, according to (2), the sentence will be true on that reading if and only if there is a past timewhen thepresident be a fool is true.And this tenseless sentence will be true at that past time if and only if the unique member of
the set of presidents at that time is also a member of the set of fools at

that time. Thus the tense operator shifts the evaluation time before the subject NP is evaluated, and the denotation of the head noun president is obtained by looking at the past time. This gives us the reading where we are saying of an individualwho was president at some past time that he
was a fool during that same past time.

Now suppose that the subject in (1) has wide scope over tense.3 Then the denotation of thisNP will depend on the set denoted by president at the original evaluation time, i.e. the time of utterance. (1)will be true on
this reading if and only if whoever is the president at the present was a

fool at some past time. The approach outlined here predicts that sentences like (1) will be two-ways ambiguous, since there are exactly two elements whose scope has semantic significance: the subject NP and the tense. One may consider it amajor virtue of these treatments that they strongly constrain the number of available readings. However, the evidence Iwill discuss in the next section indicates that such analyses are too restrictive, that they rule out readings which should be available.

TEMPORAL

EXPRESSIONS

407

3.

PROBLEMS

FOR

THE

CLASSICAL

ANALYSIS

The analysis of tense outlined above predicts that (3) below, like (1),will have two readings, depending on the relative scope of the subject NP and tense. (3) all richmen were obnoxious children

These readings are represented by (4) and (5).4 Vx [rich-man(x)-> P obnoxious-child(x)] (4) (5) P Vx [rich-man(x) -> obnoxious-child(x)]

In (4), the subject is outside the scope of the past tense operator. The sentence will be true on this reading if and only if there is a past time when everybody who is a rich man now was an obnoxious child. In (5), the subject is inside the scope of the past tense operator, and the
will be true on this reading if and only if there is a time in the when everybody who was a rich man at that time was an obnoxious past that the sentence will never be true on this child at that time. Note sentence

reading, because it requires that the individuals be richmen and obnox ious children at the same. According to traditional analyses, then, (3) can
never be used to make a true statement about past rich men.

(4) and (5) are the only alternatives for the relative scope of the quantifier and the tense operator in (3).We can quantify either over present richmen or over past richmen. But suppose we want to quantify over both present and past rich men simultaneously. It is easy for many speakers to imagine (3) as saying something about every person who is a
rich man now or was a rich man in the past. Given standard analyses,

there is no way we can allow the universal quantifier to range over both past and present richmen. The subject is either within the scope of tense and gets interpreted at a past time, or it is outside the scope of tense and gets interpreted at the present.
Below one afford tense is another example where slot at a time. Suppose we may want to quantify in more than that there is a successful investment

club whose members are assured of getting rich. The members can then
to buy a house.

(6)
investment

every member of our investment club will buy a house


club now or will be a member or not he is a member in the future that he will buy a at the time of the purchase. Again, in standard analyses. If the subject has wide

There is a reading where (6) says of everybody who is a member of our


house, whether this reading is unavailable

408

MURVET

ENC

scope over future tense, we will be saying something of present members only. If it has narrow scope with respect to the future tense, we will be saying something of future members only, and furthermore, we will be claiming that they will still be members when they buy a house. What we need for the reading we want is to be able to quantify over present and future (and possibly past) members simultaneously, without making any claims about theirmembership when they buy a house. The two examples I have given involve subject NP's. But the problem is not restricted to subjects. (7) the Citizens' Committee will sue every politician violating conflict of interest laws

In (7), the universal quantifier in the object NP must be able to range over present and future politicians, and even of past politicians. If, for
example, Bob is not a politician now but will be one in the future and will

violate conflict of interest laws, on one reading of the sentence, we can expect him to be sued along with present crooked politicians. The difficulties we encountered are not restricted to the interpretation of NP's with universal quantifiers. (8) every senior will have met a president

Let us imagine that (8) is uttered in the following context. We are talking about the present seniors who consist of Tom, Mary and Bob. Reagan is president now and he will lose the next election to Kennedy. Bob met Johnson while Johnson was president. Tom will meet Reagan tomorrow while he is still president. Mary met Kennedy yesterday. In February 1989, itwill be true that every senior has met a president. Therefore it is true now that every senior will have met a president in February 1989.
Let us now senior look at what to have readings scope are allowed over for (8) under And the classical since we are

analyses. Since we want a different president for every senior, we want


every wide a president.

talking about present seniors, we want every senior to have wide scope over the tenses.We then have three options.5 (9) (10) (11) Vx[senior(x) -- F P 3 y[president(y) & meet(x, y)]] Vx[senior(x) - 3 y[president(y) & F P meet(x, y)]] Vx[senior(x)F 3y[president(y) & P meet(x, y)]]

In (9), the object NP is inside the scope of both tenses, and the sentence will be false on this reading because at the timeMary met Kennedy, he was not yet president. In (10), the object NP has wide scope over both tenses, and the sentence will be false on this reading because Johnson

TEMPORAL

EXPRESSIONS

409

and Kennedy are not presidents now. In (11), the object has narrow scope with respect to the future tense, but wide scope over past tense. The sentence will be false on this reading, because this reading is about
future presidents and Johnson will not be a president in the future. All

available readings will thus turn out to be false, although every senior will have met a president. In the examples discussed so far, the sentences seemed to have
readings which required NP's to be evaluated in more than one tense slot

simultaneously, e.g. in the past, present and future, something not allowed by classical analyses. The time needed for the NP's properly
included ples the time allowed that NP's may indicate by the tense operator. The next set of exam at times which do not need to be interpreted

overlap either with the present or with the time provided by the tense. Suppose that the president is giving a party for people who were held
hostage in Iran. John will attend this party. We want (12) to be true if

John meets all the hostages. (12) John will meet every hostage at the president's party

If we give every hostage wide scope over tense, we will be saying something about present hostages. Ifwe give tense wide scope over every hostage, we will be saying something about future hostages. Again, these are the only available alternatives under traditional analyses. This gives the wrong results, since we are talking about past hostages. But there is no past tense in (12), and therefore we will not get the reading we want.6 Note that (12)may be followed felicitously by "and he will ask them how
it feels not to be hostages anymore." In one breath, we may refer to these

individuals as hostages and claim that they are no longer hostages, without uttering a contradiction. (13) below is a similar example. (13)
This

every fugitive is now in jail


has present be tense and standard But analyses it also predict has only the

sentence

reading which involves individualswho are fugitives now. Therefore the


sentence that should contradictory. clearly, a non-con

tradictory reading where we are saying of individualswho were fugitives


they are now in jail. Once sentence, yet we want the subject there is no past tense in the again, to be interpreted at a past time.

A similar example was observed inCooper (1978). The following is his example (58). (14) every congressman who remembers a president will be at the party

410

MURVET

ENC

Cooper points out that this sentence should have a reading that turns out to be true in the following context. The presidency will be abolished in the future, and years after that, a party will be given for individualswho are congressmen then, and who are old enough to remember a president. We want every congressman to be in the scope of future tense (we are talking about future congressmen), and a president to be in the scope of every congressman (they may remember different presidents). But this puts a president inside the scope of future tense and gives us a reading about future presidents. We wanted, instead, a reading about past presidents, even though there is no past tense in the sentence. A further problem raised by the classical analysis is that it gives rise to scope paradoxes. Consider (15). (15) every congressman who remembers an astronaut will be at the party

Suppose this sentence is about present astronauts and future congress men, i.e. there will be a party in the future which will be attended by individualswho are congressmen at that future time and who remember the individualswho are astronauts now. The classical analysis allows us to talk about present astronauts, because we are assuming some mechanism for taking NP's outside the scope of future. This example is different from (14) in that it does not require us to access a time that is neither future nor present. It does, however, lead to a scope paradox. The sentence can be true on this reading if the congressmen remember different astronauts. Therefore, we need to give every congressman wide scope over an astronaut. every congressman also needs to be inside the scope of the future tense, since we are talking about future congressmen.
This would put an astronaut also in the scope of the future tense, not

allowing a reading about present astronauts. If we gave an astronaut wide scope over the future tense, thiswould also give itwide scope over every congressman, requiring the congressmen to remember the same astronauts, again an undesirable result. There is no way of giving an astronaut wide scope over the tense and narrow scope with respect to every congressman ifwe are talking about future congressmen. Thus the desired reading is unavailable.
4. TENSE OPERATORS IN NP'S

The evidence presented in the previous section is problematic for the classical analysis only under certain assumptions about themapping from syntactic structures to semantic structures. The three crucial assumptions

TEMPORAL

EXPRESSIONS

411

in this case are the following: 1) every tense operator allowed in the semantic representations of the sentences of a natural language cor responds to a syntactic tense, which in turn corresponds to a tense morpheme (possibly null) found in the surface forms of that language,7 2) a tense operator is introduced into the semantic representation of a sentence only if the corresponding tense is present in the syntactic structure of the sentence and the surface form contains the appropriate tensemorpheme, and 3) the scope of a tense operator always includes a predicate which comes from the verb of the sentence.8 These assumptions about tense operators are implicit inmost analyses of tense in natural languages, and they seem to provide the strongest constraints on the semantics of tense in natural languages, reflecting a view which can roughly be summarized as 'whatyou see iswhat you get'. However, it is also these assumptions that render classical analyses empirically inadequate. Getting rid of the constraints they impose would allow us to provide a semantics thatwould avoid the problems discussed above.9 For example, the reading we wanted for (13) can be represented as in (16). (16) Vx[P[fugitive(x)]-> in-jail(x)]

According to our semantic rule in (2), (16) says of past fugitives that they are now in jail.We are thus able to represent the non-contradictory reading of the sentence. In this representation, the operator introduced is the past tense operator and English does indeed have a past tense morpheme. However, the past tense operator is introduced into the semantic representation in the absence of past tense morphology in the sentence. Furthermore, this past tense operator is attached to an atomic formula whose predicate comes from the head noun of the subject NP, thus leaving the predicate contributed by themain verb outside its scope. All three assumptions have to be abandoned in order to account for some of the readings discussed in the previous section. For example, for the readings of (6), (7) and (8) thatwe were interested in, it seems that no ingenuity in the order of introduction of past, present and future opera tors will get us the desired readings. We could attempt to solve our problems, however, by defining new tense operators in addition to past, present and future. Suppose we introduce a new operatorW and define it in the following way:
(17) Wqp is true at time t iff Pep or p or F<p is true at t.

This is the I-don't-care-when operator.When W applies to any sentence A, the resulting sentence is true if and only ifA is true in the past, in the

412

MURVET

ENC

present or in the future. Given this new operator, we can now represent the desired reading of (8) as in (18). (18) Vx[senior(x) --*3 y[W president(y) & F P meet(x, y)]]

One could define other tense operators thatmean true-now-or-at-some future-time, true-now-or-at-some-past-time, true-at-some-past-time that-includes-a-sunny-day, true-on-my-birthday, etc. Assuming that the class of tense operators necessary for analyzing natural languages is somehow constrained, our task would be to uncover these constraints. What kinds of tense operators should be allowed, and what particular operators are required for an adequate treatment of tense in natural languages would, then, be an empirical matter. Note that a number of such operators have already been proposed, e.g. Kamp's NOW (Kamp 1971), and operators familiar from tense logic, like G which is 'itwill
always be true that.. .', and H which is 'it has always been true that...'"

5. STRICTLY

LOCAL

OPERATORS

If we pursued this line, we would hope to end up with an empirically motivated class of tense operators necessary for the analysis of natural languages. Thus one might think that the problems with the analysis of tense reduce to the problem of defining the right operators. However, such an approach raises questions regarding the scope of these operators. Let me note to begin with that the tense operators introduced into the semantic representations to get NP interpretations right,must have very limited scopes. For example, having acknowledged the need for some thing like the past tense operator for (13) (we are saying of past fugitives
that they are now in jail), we must make sure that the scope of this

operator is exactly as in (16), and that (19) is blocked as the represen tation of themeaning of (13). (19)
conditional. past

Vx[P[fugitive(x) -- in-jail(x)]]

In (19), the scope of the past tense operator extends over the whole
was a fugitive at a (19) would be true if and only if whoever also in jail at that past time, clearly a reading never time was

available for (13). Similarly for (12) (12)


we would

John will meet every hostage at the president's party


need a past tense operator inside every hostage, since these

individuals have been hostages in the past on the intended reading. But

TEMPORAL

EXPRESSIONS

413

this operator must not have scope over the future tense operator, because all readings of the sentence require the time of meeting to be in the future, and giving the past operator wide scope over the future operator would allow the time of meeting to be in the past. This evidence shows that verbs are always interpreted according to the tense provided by the tense morpheme in the sentence, and whatever apparatus is devised to get theNP meanings right, itmust not affect verb interpretations. That is to say, the tense operators in the semantic representations of NP's must never have scope over the verb or the verb tense (the 'true' tense). Should an NP tense operator be allowed to have scope over other NP's in the sentence? In other words, are the various NP's in a sentence temporally dependent on each other? A brief look at the data shows that they are not. Consider (8), repeated here. (8)
We wanted

every senior will have met a president


this sentence to be true in a context where we were talking

about present seniors and individuals who were presidents in the past, present or future. Therefore the relevant times are different for the two NP's every senior and a president. Similarly for (14), (14) every congressman who remembers a president will be at the party
at a future time, whereas a president needs to be evaluated at a

on the intended reading discussed above, every congressman needs to be


evaluated

past time. It seems that in natural languages, the relevant time for each NP may be different. Thus theNP's are temporally independent of each other, and we must provide an analysis that guarantees this.11
We have seen that the temporal interpretation of an NP does not affect

the interpretation of the verb, the 'true' tense, or another NP. Therefore such elements must be outside the scope of any tense operator intro duced into the semantic representation of an NP. This still leaves open the question of whether or not the 'true' tense operator, which cor responds to the tensemorphology in the sentence and which affects verb interpretations, should take scope over the tense operators introduced into the semantic representations of NP's. In the classical analysis of tense, the verb tense crucially affects the interpretation of NP's. A two-way ambiguity is predicted for sentences like (1) because a) the
scope of the tense is the minimal sentence which includes the tense, and

b) some independently motivated mechanism such as Quantifying In (or Quantifier Raising, or Cooper-storage) allows representations where the

414

MURVET

ENC

NP is not in the scope of the tense operator. I have argued that such an approach is too restrictive, and blocks readings which the English sentences do seem to have. A sentential scope analysis which is aug mented along the lines discussed above, with the introduction of new tense operators into the representation of NP's, avoids the problems discussed, and makes it possible to maintain that the verb tense does
have scope over NP's. This is because there will always be a way of

undoing the effect of the verb tense. Interestingly however, once such an approach is adopted, it is no longer necessary to give verb tense scope over NP's. In the classical analysis, different readings are the result of different scope relations
between the tense and the NP's. Now the same readings can be obtained

by simply defining the necessary operators and introducing them as part of NP meanings. That is to say, the readings can be obtained without relying on the scope of the verb tenses. Thus it becomes possible to keep the effects of the 'true' tense entirely local to the verb. For example, the classical analysis predicts two readings for (20), one about present sophomores and one about past sophomores, depending on whether or not the subject is in the scope of the past tense. (20) every sophomore cried

However, now the same readings can be obtained by giving the past tense local scope over cry, and by introducing either a present or a past tense operator into the semantic representation of the NP. There is a significant difference between the two approaches with respect to the reading of (20) involving past sophomores. The classical analysis predicts that the past sophomores cry while they are sophomores, since there is only one past tense operator which causes
both the NP and the verb to be evaluated at a past time. In contrast, the

local-scope analysis involves two past tense operators, one for the NP and another for the verb. This allows the time where the individuals are sophomores to be disjoint from the time where they cry, as long as both times are in the past. That is to say, the sentence would be true on this reading if the individuals are sophomores in the past but cry before they become sophomores, and also if the individuals cry in the past but are sophomores before then. This reading, not available under the classical analysis, does seem to be there. Suppose the discourse is about the reaction of the students enrolled in a college ten years ago to the news that they had been accepted at that college. (20) can be used to assert that the sophomores of ten years ago, i.e. past sophomores, reacted to their acceptance by crying. The relevant time of crying is also in the past,

TEMPORAL

EXPRESSIONS

415

but precedes the time when these individualswere sophomores. For this interpretation, then, we need two different past times, one for the time of being a sophomore, and the other for the time of crying. Similarly, we wanted (6) to be true (6) every member of our investment club will buy a house

if uttered about future members of the club who buy houses after they stop being members. This reading requires two distinct future times, and cannot be obtained under an analysis which obtains the reading about futuremembers only by leaving the subject NP in the scope of the future tense which affects the interpretation of the verb. If the 'true' tense is strictly local to the verb and does not take scope over NP's, then we will never get a reading of (20) which requires that the individuals be sophomores while they cry. Instead, this analysis merely allows it, by virtue of the fact that it places no constraints on the relevant times for the NP and the verb beyond requiring that they both be in the past. The analysis, in fact, claims that whenever the time of being a sophomore is identical to the time of crying, this is not due to the scope of the tense, but rather to some other restriction on the times. This phenomenon seems to be exactly parallel to the interpretation of pronouns in sentences like (21). (21) John said that he was tired
John as its antecedent, or it can receive its referent

In (21), he can have

from outside the sentence. This is usually accounted for by indexing the pronouns, and by stipulating that coindexation yields coreference. In other words, indexation places a restriction on the interpretation of the pronoun beyond the inherent meaning of the pronoun. No principle of pronoun interpretation, however, requires coindexation. It seems to me that the simultaneous reading of (20) where the time of being a
in a parallel is the same as the time of crying can be derived sophomore and does not have to be due to the scope of the tense. fashion,

I have not provided here an argument that giving the verb tense scope over NP's makes incorrect predictions. It seems quite possible to get the right readings even when the verb tense does take scope over NP's, ifwe allow ourselves the freedom to introduce other tense operators intoNP's.
However, now verb is how I hope to have shown that in such an analysis, no reading

requires the verb tense to have scope over NP's. The question facing us
can have two analyses, one where the verb tense between it cannot. If we allow the over NP's, and another where scope tense to have scope over NP's and also have the option of introduc to choose

416

MURVET

ENC

ing a number of other operators inside NP's, we have in fact allowed


ourselves a way of always neutralizing the effect of the verb tense on

NP's. This, inmy opinion, greatly trivializes the scope of the verb tense.
In the absence seems verb, of any need to give the verb tense scope over NP's, it and more illuminating to treat it as an operator local to the simpler on a par with i.e. to treat the 'true' tense operator the tense

operators introduced into the semantic representations of NP's, by allow ing it to have a very limited scope. The following section further
examines operators. the limits that would have to be placed on the scope of tense

6. PREDICATES
AND TEMPORAL

OF INDIVIDUALS
ARGUMENTS

I have suggested above that the semantic representations of tensed sentences should include several tense operators, one for each NP and one for the verb, and that the scope of these operators should be appropriately restricted. Given the necessity of constraints on their scopes, we must first question the assumption that tense operators are sentential, an assumption that natural language semantics has inherited from tense logic.12 When we claim that a tense operator in the semantic representation of a natural language is sentential, we normally mean that its scope is a constituent analyzed as a sentence by the syntax of the natural language. The data discussed above suggest strongly that the tense operators introduced into the semantic representations of NP's cannot be sentential in this sense, since we saw that theymust be prevented from affecting the temporal interpretation of the verbs, and we have no independently motivated mechanism for taking verbs out of the scope of sentential elements. Furthermore, if we follow the suggestion made at the end of the last section that the scope of the 'true' tense operator should also be
local and that it should One might wish not have scope over NP's, then we have, are sentential so far, at some

no strong arguments for treating any of the tense operators as sentential.


to claim that the tense operators

level of representation where constituents no longer correspond to those provided by the syntax of the natural language. We could assume, for example, that there is an appropriate level of representation where some
part of achieve the NP is of the same semantic natural this would be to translate type as sentences. One way to into a logical langu languages

age where the noun is represented as a one-place predicate. This is indeed

TEMPORAL

EXPRESSIONS

417

what we have been doing throughout this paper by representing the readings of the English sentences by logical formulas, as in (18). However, logical formulas were used here for the sake of convenience only, as part of our descriptive apparatus, rather than as essential representations of English sentences. Suppose thatwe modified this view and took the logical representations to be essential for an adequate analysis of natural language sentences. Then a tense operator would be
sentential at that level of representation if it had scope over formulas of

the logical language, although these formulas would not correspond to the constituents analyzed as sentences by the syntax of the natural language. That is to say, tense operators would be sentential only with respect to the logical language.
The issue I would like to address tense advantages to analyzing at this point is whether there are any at any level of as a sentential operator

representation, or whether this is merely an artifact of a particular execution, with no real semantic significance. The advantage of analyz
ing tense as a sentential operator is that it allows the tense to potentially

affect the interpretation of several elements in its scope. This advantage disappears if the scope of the tense operator always contains only one element whose denotation varies according to times. I suggested above
that the effect order of the tense whether operators was limited. We were, however,

concentrating primarily on the interpretation of nouns and verbs. In


to determine or not tenses should be analyzed as sentential

operators, we need to examine their effect, if any, on the interpretation of other expressions. This, in turn, raises a more fundamental question. Exactly which expressions in natural languages need to be interpreted relative to times?
It is widely assumed that the intensions of all expressions are functions of all expressions are from world-time i.e. that the denotations pairs,

determined relative to a time (cf. Cresswell (1973), Dowty (1979), Kaplan (1977), Montague (1976)). This view allows a uniform treatment of natural language expressions, but on closer inspection, the uniformity
turns out to be only apparent. It has already been pointed out by several

philosophers that certain classes of expressions must be 'neutralized'with respect to times. For example, if names are rigid designators, then a
name time a certain individual, no matter what the evaluation always denotes from times. is. Thus the intensions of names are constant functions (1977) has shown of utterance before denotation does not

Personal

behave in a similar way. Kaplan pronouns that the referent of a pronoun is fixed by the context sense, pronouns behave like names; their

the proposition expressed by the sentence is determined. Therefore, in


some

418

MURVET

ENC

depend on the time of evaluation. Kaplan ensures this by stipulating that indexicals have constant contents (i.e. their intensions are constant functions). There are other expressions which are obviously time-in dependent. These are the logical constants. None of the logical con nectives (and, but, or, negation), or the logical determiners (every, all, the, a, etc.) are sensitive to times, and their denotation is stipulated to be the same at every time of evaluation. So far, then, names, indexicals and logical constants turn out to have constant intensions. And this has to be stipulated somewhere in the theory. I assume that this conclusion is uncontroversial. We now need to ask whether there are other classes of expressions which are time independent. Let us first examine prepositions. (22) John died inLondon

In (22), is the preposition in sensitive to time? Suppose that it is, and suppose that the prepositional phrase in London denotes the set of individualswho are inLondon at time t. Suppose further that this time is distinct from the time of dying. Then if the denotation of the verb phrase is the intersection of the individuals denoted by the verb and the individuals denoted by the PP, the sentence would be true if John died at
a certain time, and was in London at another time. This is clearly not a

reading available for (22). Suppose, instead, that the PP denotes a function which carves out a subset of the set denoted by died. Then in what sense can this function be time-sensitive? Quite simply, the value of died inLondon depends only on the time of dying, and on no other time. One might argue that the time relevant for the preposition, and hence for the PP, is somehow bound by the verb tense. However, this is not only unnecessary, but also clouds the issue of temporal dependence. We may safely conclude that the denotations of prepositions do not vary accord ing to the time of evaluation. A similar argument can be made for adjectives and adverbs. (23) John walked slowly

If slowly denotes the set of individualswho do something slowly at a time t and this time is distinct from the time of John's walking, then the
sentence available instead walkers would have a reading would sets of where John walked to ensure at some time and

where he did something slowly at another time.Again, this reading is not


for (23). We a function in (23), then need that the time associated

with the adverb is always bound by the verb tense. If slowly denotes
from the value to sets, carving out a subset of the set of is the set of individuals who this function

TEMPORAL

EXPRESSIONS

419

walked slowly at the time determined by the verb tense. The only time that plays a role in determining the value of theVP comes from the verb tense, and there is no evidence for a temporal argument for the function denoted by the adverb. The generalization that seems to be emerging from the above dis cussion is that the time dependence of a VP containing amodifier is due to the time dependence of the head verb, not themodifier. This does not rule out time-dependent expressions inside themodifier. For example, in the capital contains a NP which is time dependent, given that cities can have the property of being the capital at some times and not at others. However, once the value of the NP is fixed and the location is deter mined, there seems to be no further time dependence in the PP.We may then conclude that the intensions of these modifiers are constant func tions from times of evaluation. A parallel situation obtains with modifiers inNP's. (24) Mary spoke to every intelligent linguist

(24) does not seem to have a reading where the individuals thatMary spoke to are linguists at some time and intelligent at another time.We
may, then, take AP's to also denote functions which are time-in

dependent.13 Apparent exceptions to the above generalization involve relative clauses and temporal adverbs like yesterday. In the following NP (25) the president who died ten years after he left the office
a president. But this as the main verb of

the time of dying is not the same as the time of being case is parallel to PP's like in the capital. die occurs

the relative clause and its time is determined by the tense in the relative clause. Once that is fixed, however, if the relative clause is interpreted as a function carving out a subset of the relevant presidents, then this function has no further time arguments. Let us assume that temporal expressions like yesterday denote times. Then in (26) (26)
a function function We

John died yesterday


from sets to sets, that the adverb seen it would carve out a subset of the individuals a further time argument.

ifwe further assume that the adverb constructed out of yesterday denotes who died, giving us the set of individualswho died yesterday.14 But the
denotes need not have

Similarly for temporal PP's like onMonday.


have that PP's, AP's and relative clauses are time dependent

420

MURVET

ENC

only so far as they contain nouns and verbs.15Neither the heads of these modifier phrases nor the phrases themselves seem to need a further time
argument. I take it then that we can add these categories to the class of

natural language expressions whose intensions are constant functions from times. The only time-dependent expressions in natural languages
seem So to be nouns far, we have and verbs.16 assumed the time dependence of nouns and verbs

without questioning it.Given that the 'true' tense fixes a time for verbs, we may safely conclude that verbs are time dependent.17 The most telling arguments for the time dependence of nouns comes from Larson (1983) which examines temporal relative clauses inNP's. (27) the duty officer when I was on deck will soon file his report

Larson points out that in sentences like (27), the relative clause when I was on deck fixes the time for duty officer. This is possible only if nouns are interpreted relative to times. Therefore the assumption we have been making all along, that nouns are time dependent, seems justified.18 Let us now return to the assumption that all natural language expres sions have intensions which are functions from times (aswell as possible worlds). We have noted that this allows a uniform treatment of in tensions, but have also observed that we need to stipulate that all intensions except those of nouns and verbs are constant functions from times. Such a stipulation captures in perhaps a roundabout and obscure way what seems to be a true generalization about natural languages, that only nouns and verbs are sensitive to times. The generalization that all natural language expressions have intensions which are functions from
times now seems vacuous. This is a point we will return to later. For the

time being, let us assume that there is some such stipulation in our theory. We have been exploring an analysis where tense operators are intro duced at random into NP representations at an appropriate level, with new tense operators defined as necessary, and where all tense operators are sentential, although the sentences at this level do not correspond to the sentences of the natural language syntax. We noted that treating tense operators as sentential at this level would be advantageous if it could be shown that tenses affected the interpretation of several elements
in their scope in a significant way. It seems now that this is not the case. I

have argued that only nouns and verbs are sensitive to times, and also that the temporal interpretation of each noun is independent of any other noun, or the verb, or the 'true' tense. Suppose that these generalizations are captured by allowing one tense operator for each NP in addition to

TEMPORAL

EXPRESSIONS

421

the one corresponding to the syntactic tense, and by stipulating that intensions of all expressions except nouns and verbs are constant func tions. It follows that each tense operator will affect the interpretation of only one element, a noun or a verb. Therefore, analyzing tenses as sentential operators has no advantages. It does, however, have the disadvantage of requiring a level of representation where sentences do not correspond to the sentences normally generated by the syntax. I take this to indicate that the temporal analysis of natural languages should not employ sentential tense operators. We might want to treat tense operators as elements with strictly local scopes, i.e. as elements which operate only on a noun or a verb. In the classical analysis outlined at the beginning of the paper, the past tense introduces a past time which becomes the new evaluation time of the sentence. Since natural languages do not generally allow iteration of tenses, the tense in effect provides the temporal argument of the in tensions of the expressions in the sentence.19 Given that only nouns and verbs seem to have non-constant intensions, we might treat tense opera tors as elements which fix the temporal arguments of the intensions of these expressions. How do tense operators work, and at what level of representation of the object language, if any, are they present? The answers to these questions will establish the relation between tenses on the one hand, and nouns and verbs on the other. But before these issues can be adequately dealt with, we must ask a more basic question. Why is it that nouns and verbs are time-sensitive? The answer to this question seems to be straightforward: Nouns and verbs are predicates of in dividuals. Itmay then be a universal property of natural languages that all and only predicates which take individuals as arguments also take times as arguments.20 This property of natural languages is not captured under the view that all intensions are functions from times, since such a view claims that every expression has a temporal argument, although in a number of cases these temporal arguments do not affect the denotation of the expression. The evidence discussed here suggests an alternative approach where we
abandon the notion that intensions are functions from times and worlds,

and maintain perhaps that they are only functions from possible worlds. In addition, we recognize that all predicates of individuals have one more argument than previously assumed, e.g. analyze kiss as a three-place predicate, sleep and president as two-place predicates, etc., and require
to come from the temporal domain. Thus temporal one of the arguments no in a manner of individuals are selected by predicates arguments

different from the selection of non-temporal arguments.

422

MURVET

ENC

We now return to the status of tense operators. I noted the need for one operator for each noun and verb, and also argued that these operators should not be sentential. I also pointed out that they could be analyzed as operators on predicates. But the discussion above indicates that labelling such elements 'operators' is not particularly appropriate here. The facts seem to be simple: a) Nouns and verbs must be provided with temporal arguments, b) there is a necessary link between the 'true' tense and the temporal argument of the verb, c) there is no necessary link between the 'true' tense and the temporal arguments of nouns. Let us assume a referential theory of tense, i.e. a theory where tense simply denotes a time.21 We may then claim that the temporal argument of the verb is the time provided by the tense in the sentence.22 We explored the idea of abstract tense operators for NP's. However, the facts can be accounted forwithout introducing novel tense operators into our semantic system, because a temporal argument for nouns can be supplied without recourse to tense operators. There are two plausible ways of achieving this. One is to assume that each noun is assigned a time by the context of use. This is the analysis proposed in Eng (1980), and highlights the fact that generally, the temporal interpretation of a noun depends on the pragmatic situation.23 It assumes that the times assigned to nouns have been introduced into the domain of discourse previously, plausibly by tenses and temporal adverbs. In such an analysis, the temporal arguments of nouns are not represented by any constituents in the syntax of the language. Rather, they are supplied in the course of the interpretation. An alternative is to assume that the temporal arguments of nouns are represented in the syntax as empty categories, and that they are pronominal. This is the view adopted in Enc (1985). The interpretation of the empty temporal arguments is then exactly parallel to the interpretation of indexicals like he, and once again the value of these arguments is supplied by the context. These two ap proaches are equivalent from a semantic point of view, and the choice will depend ultimately on syntactic matters not addressed here. For our
is important is that each noun is assigned a time by the purposes, what for the context. To see how an analysis along these lines can account

data discussed above, let us turn to (12), repeated here. (12)


We were before

John will meet every hostage at the president's party


interested the time of utterance in the reading about individuals who were hostages of the sentence. This reading is no longer a individuals were in fact hostages. Note that this is

problem, since the time assigned by the context to hostage can be the
past time when these

TEMPORAL

EXPRESSIONS

423

possible only if such a time was previously introduced into the domain of discourse, i.e. if it is pragmatically recoverable. There does not seem to be any need for any constraints on the temporal arguments of nouns beyond the pragmatic ones. Therefore we assume that the context can assign to a noun a specific past time, or a time that includes all of the past moments and the present moment, or a time that includes past, present and futuremoments, etc.24 As in the case of personal pronouns, this approach allows the temporal arguments of nouns to have sentence-internal antecedents. Since a noun can be assigned any time from the context, it can also be assigned the time introduced by the tense or a temporal adverb in the sentence.25 For example, we discussed the two readings of (20) which were about past sophomores, where the readings differed according towhether or not the past time of being a sophomore was the same as the past time of crying. (20) every sophomore cried

We will assume that the past tense yields a past timewhich is the time of crying. If this time is assigned as the value of the temporal argument of sophomore, thenwe have the reading which is true if individualswho are sophomores at a past time cry at that same past time. This reading, which would have been obtained in the classical analysis by giving the past tense wide scope over the subject, is now available simply because the tense is a possible antecedent for the temporal argument of the noun. However, it is possible for the context to assign some other past interval to the noun in the subject, provided it is pragmatically available. This would give us a reading not available under the classical analysis, a reading where we are saying of past sophomores that they cried in the past, but allowing the two past times to be distinct.
7. CONCLUSION

I have argued against analyzing natural language tenses as sentential operators on the grounds that only verbs are necessarily interpreted relative to the time provided by the tense. I have further argued that the time-sensitive expressions of natural languages consist of nouns and verbs, which strongly suggests that times should be viewed as individuals which serve as arguments of predicates of individuals. The natural
language data seems to favor a referential analysis of temporal expres

sions, where temporal expressions are fundamentally similar to non temporalNP's, rather than the classical analyses which employ sentential
operators.

424

MURVET

ENC

ACKNOWLEDGEMENT This paper develops ideas first put forward inEng (1980) and Enc (1981). I am grateful to Robin Cooper and Larry Horn for very helpful discussions and numerous suggestions, and for their encouragement and support. I have also benefited from comments and criticisms by Hagit Borer, Michael Byrd, David Dowty, Berent Eng, Elisabet Engdahl, Irene Heim, Frank Heny, Hans Kamp, Lauri Karttunen, Richard Larson, David Pesetsky, Stanley Peters, Geoffrey Pullum, Ivan Sag, Susan Stucky, Johan van Benthem, Frank Vlach, Tom Wasow and an anonymous referee of this journal.

NOTES Cf. Montague (1976), where all expressions have intensions which are functions from times and possible worlds. 2 Cf. Montague (1976). For the purposes of this paper, we will go along with the idea that natural languages have a future tense, a view rejected inEnc (1985). 3We assume here that some mechanism exists for giving NP's wide scope, the exact nature of thismechanism being irrelevant. 4 For convenience, logical formulas will be used to represent the readings of English sentences. 5 I am treating the Present Perfect here as a simple past tense operator. The differences are irrelevant to the point under discussion and are therefore ignored. 6 I am assuming that the introduction of tense operators is triggered by tensemorphology. I return to this question later. 7 Thus the tense operators of English would be limited to the past, the present and the future, if will is indeed a future tense morpheme. 8 This is because tense is treated as a sentential operator, and while there are general mechanisms for giving NP's wide scope over operators, there are none for verbs. 9 The problems surface in analyses which are based on these assumptions and which treat tenses as sentential operators, whether the semantics is interval-based or moment-based. 10These operators, however, were not proposed to account for the relationship between NP's and tense. 1 Ens (1981) provides arguments for the independence of NP's also with respect to locations. 12 It is irrelevant here whether the tense operators occur at some level of representation of the object language, or whether they are introduced as part of the interpretive rules. 13 It seems the dead president can pick out an individual who is dead now and was a president in the past. The analysis proposed later for nouns would be extended to adjectives if they turn out to pattern with nouns. 14These comments are relevant only to analyses which treat such modifiers as denoting functions from sets to sets, an approach rejected inEnc (1985). 15 Participles seem to have independent time reference. In a dying president, the time of
dying and the time of being a president need not be the same. I take this to be due to the

verbal nature of the participles. 16But see Note 13. Also, an anonymous referee of this journal has brought tomy attention the temporal properties of non-restricting modifiers like alleged or fake. In these cases, the temporal restriction does not seem to come from the head noun alone. A fake gun is not an
object which is a gun at a certain time and which is also fake at that time. Rather, we need

an object which, at a certain time, has the property of being a fake gun, i.e. semantically, there is only one relevant property here and this property is obtained from the non

TEMPORAL

EXPRESSIONS

425

restrictive adjective plus the noun. Correspondingly, there is only one temporal depen dence, the only time needed being the time this object has the property of being a fake gun. On this point, correct resultsmay be obtained by treating fake gun as a compound noun. In any event, it is not necessary for the non-restrictive adjective and the noun to have separate temporal arguments. 17This view differs from Enc (1981) where verbs and nouns were interpreted directly, without reference to times, and tenses contributed pragmatic well-formedness conditions. See Larson (1983) for arguments against that position. 18This view again differs from Ens (1981), and is the one adopted in Ens (1980) which assigns to every NP independent contextual indices. 19This paper ignores the complications arising from aspect and embedded clauses. See Dowty (1979), Enc (1985). 20One might want to treat restrictive adjectives as individual predicates, inwhich case theywould be expected to pattern with nouns and verbs in this respect. See Note 13. 21A referential view of tense is proposed in Partee (1973). 22 This is an oversimplification. Enc (1985) argues for a more complex relation between the temporal argument of the verb and the tense. 23 The analysis inLarson (1983), where nouns are assigned independent locations, is also in the same spirit. 24Nor are the times required to be intervals. It seems possible, for example, for the discourse to be about a timewhich consists of all the election years, excluding the years in between. 25This would require the antecedent in the sentence to be processed before the noun. I assume this depends on the syntactic configuration. See Heim (1982) for the possible antecedents of definites in general.

REFERENCES Cooper, R.: 1978, 'Variable Binding and Relative Clauses', in F. Guenthner and S. Schmidt (eds.), Formal Semantics and Pragmatics for Natural Languages, Reidel, Dor drecht, 131-169. Cresswell, M.: 1973, Logics and Languages, Methuen and Co. Ltd., London. Cresswell, M.: 1977, 'Interval Semantics and Logical Words', in C. Rohrer (ed.), On the Logical Analysis of Tense and Aspect, TBL Verlag Gunter Narr, Tubingen. Dowty, D.: 1979, Word Meaning and Montague Grammar, Reidel, Dordrecht. Enc, M.: 1980, 'QuantifyingOut', unpublished manuscript of talk delivered at the Annual LSA Meeting. Enc, M.: 1981, Tense Without Scope: An Analysis of Nouns as Indexicals, University of Wisconsin, Madison dissertation. Distributed by IULC. Enc, M.: 1985, 'Temporal Interpretation', unpublished USC manuscript. Heim, I.: 1982, The Semantics of Definite and Indefinite Noun Phrases, University of Massachusetts, Amherst dissertation. Kamp, J.: 1971, 'FormalProperties of "Now"', Theoria 37, 227-273. Kaplan, D.: 1977, 'Demonstratives', unpublished manuscript. Ladusaw, W.: 1977, 'Some Problems with Tense in PTQ', in S. Schmerling and C. Smith (eds.), Texas Linguistic Forum 6, The University of Texas, Austin, 89-102. Larson, R.: 1983, Restrictive Modification: Relative Clauses and Adverbs, University of Wisconsin, Madison dissertation. Larson, R.: 1985, 'Bare-NP Adverbs', Linguistic Inquiry 16, 595-621. Montague, R.: 1976, Formal Philosophy, edited by R. Thomason, Yale University Press, New Haven.

426

MURVET

ENC

Partee, B.: 1973, 'Some Structural Analogies Between Tenses and Pronouns in English', The Journal of Philosophy 70, 601-610. Partee, B.: 1984, 'Nominal and Temporal Anaphora', Linguistics and Philosophy 7, 243-286.

Department of Linguistics, University of Southern California, Los Angeles, CA 90089-1693, U.S.A.

You might also like