Journal of Semantics 20: 3571 ccirclecopyrtOxford University Press 2003 [602105]

Journal of Semantics 20: 35–71 c/circlecopyrtOxford University Press 2003
Tense Meanings and Temporal Interpretation
SILVIA P. GENNARI
University of Maryland
Abstract
For any theory of tense meanings, subordinate sentences are particularly problematic
because embedded tenses do not seem to receive the same interpretations as their non-embedded counterparts. Previous approaches to this problem have often proposedsome syntactic mechanism or sequence of tense rule that allows the embedded tensemorphemes to receive interpretations that differ from those typically assumed fornon-embedded tenses. This paper explores an alternative view in which tenses areassumed to be uniformly defined for both independent and embedded occurrences. Itargues that the problematic subordinate interpretations can be explained if appropriatedefinitions of tense meanings are provided and independent factors influencing thetemporal interpretation are taken into account. Specifically, it is suggested that themeaning of the tense morphemes alone do not completely determine the temporalinterpretation of a sentence. In a systematic and predictable way, aktionsart propertiesfurther specify the exact duration and location of the interval in which the sentenceis true. Thus, the interaction of tense meanings and general facts of the grammar suchas aktionsart properties, rather than sequence of tense specific mechanisms, conspireto explain temporal interpretation in both embedded and non-embedded sentences.
1INTRODUCTION
In English, subordinate sentences are particularly problematic because
embedded tenses do not seem to receive the same interpretationsas their non-embedded counterparts. This contrasts with languageslike Japanese or Russian, in which uniform interpretations apply toboth embedded and non-embedded tenses. Previous approaches tothis problem have proposed language specific syntactic mechanismsor sequence of tense rule such as tense deletion or tense agreementrules. Such rules allow embedded tense morphemes in English toreceive interpretations that differ from those typically assumed for non-embedded tenses.
In this paper, I explore an alternative view in which English tenses
are uniformly defined for both non-embedded and embedded occur-rences and consequently, no language specific syntactic mechanism isrequired. I argue that the problematic subordinate interpretations canbe explained if appropriate definitions of tense meanings are provided

36Tense Meanings and Temporal Interpretation
and independent factors influencing the temporal interpretation are
taken into account. I suggest that truth-conditional tense meaningssometimes underdetermine the actual temporal interpretation in aparticular utterance situation. The exact duration and location ofthe interval denoted by the sentence may depend on other factsof the grammar such as aktionsart and pragmatic implicatures. Inparticular, I claim that aktionsart properties systematically influencetemporal interpretation. For example, stative sentences are typicallyassumed to persist in time in a way that differs from eventivesentences. This allows stative sentences to obtain beyond the intervalspecified by the tense morphemes and therefore, to overlap with thetruth intervals of sentences previously mentioned in discourse. Thisproperty, generally available in discourse contexts, is also operative insubordinate constructions, thus accounting for non-canonical temporalinterpretations identified in the literature.
The result of this proposal is that lexical tense meanings and
lexical/sentential aktionsart, rather than language specific mechanisms,interact to explain temporal interpretation in both embedded andnon-embedded sentences. If correct, this approach would simplify thesyntax-semantics interface and would further illuminate the interactionof tense meanings with aktionsart and pragmatic considerations.
1.1 The problem
Abasic task of any theory of tense meanings is to predict the temporal
interpretations of tenses embedded under the scope of other tenses.The difficulty arises when determining the meaning of a given tense inauniform manner for both independent and embedded occurrences: It
seems that embedded tenses and their independent counterparts do notreceive the same interpretation. Consider simple past tense:
(1) The senator heard about the president’s secret meetings.(2) The secretary believed that the senator was happy.
Assume for the sake of argument that each past tense in (2) denotes any
time prior to the speaker’s speech time (ST), as suggested by (1). Then,(2) should be true in a situation in which the interval of being happyis prior to the ST but later than the attitude time, i.e. the secretary’sbelieving (a forward shifted reading ). However, this interpretation is not
available in (2). The interval in which the embedded sentence is truecan either overlap with or precede the attitude time. Therefore, if pasttense is evaluated solely relative to the ST, incorrect predictions are

Silvia Gennari 37
made. I will call a view of a tense meaning in which the temporal
interpretations is exclusively relative to the ST the indexical view .
Suppose now that past tense does not always denote an interval prior
to the ST, but rather, an interval prior to whichever time happens tooccur in its immediate context. That is, suppose that in non-embeddedsentences, past tense is evaluated relative to the ST, but in embeddedsentences is evaluated relative to the time of the reported attitude. Thisview predicts that the embedded sentence in (2) is true at an intervalprior to the attitude time (a backward shifted reading )and thus fails to
predict the overlapping reading of (2). I will call such a view of tensemeaning the relative view .
Both the present and the future tense also show non-uniform
interpretations in embedded and independent occurrences. Whileindependent sentences are interpreted relative to the ST (suggestingan indexical meaning), embedded sentences might also be interpretedrelative to the attitude time:
(3) The president believed that his party is furious.
(4) The press will think that the president is out of town.(5) In two days, an official will announce that the president will
apologize (
∗tomorrow).
(6) A journalist said that the president will resign (∗yesterday).
(3) is true if the interval at which the complement is true overlaps
with both the attitude time and the ST. This is the so-called double
access reading, (Abusch 1991, 1997), because the truth-interval of the
embedded sentence has ‘access’ to both the attitude and the ST. Incontrast, (4) could be true in situations in which the embedded intervalof truth either overlaps with both the future attitude time and theST (another double access reading) or simply overlaps with the futureattitude time. As the reader can confirm, neither the indexical viewnor the relative view explains the double access readings of the presenttense.
Likewise, neither view accounts for the temporal interpretations of
future tense. The indexical view (future of the ST) would allow theembedded complement of (5) to be true at an interval later than theST but prior to the attitude time. However, this reading is not possibleas the starred adverbial modification in (5) shows. In turn, the relativeview (future of the attitude time) would incorrectly predict a readingof (6) in which the embedded complement is true at an interval laterthan the attitude time but earlier than the ST (i.e. an interpretationequivalent to the use of would ).

38Tense Meanings and Temporal Interpretation
1.2 Previous solutions
Given the failure of a uniform approach to tense meanings, the
solution proposed by most theorists is to claim that embedded tensesmake a semantic contribution different from that of their independentmorphological counterparts (non-uniform theories). Some surfaceembedded tense morphemes are assumed to be semantically vacuousand are underlyingly analysed as overlapping the attitude time. Theway in which the embedded tense morphology is semantically nullifiedvaries from theory to theory. Ladusaw (1977), for example, proposesatransformation that changes an underlying tense morpheme (e.g.
present) into a past morpheme if embedded under past morphology,while the semantic interpretation of the underlying tense (e.g. overlapwith the attitude time) remains. Enc ¸(1987) proposes two different
mechanisms to licence tense references. In one case, the tense isbound by the matrix’s tense yielding identity of reference (simultaneousreadings). In the other, the tense is licenced by other temporal elementsin the clause so that it can refer to a different time relative to theattitude time (non-vacuous readings). Abusch (1997) proposes a featurepassing mechanism specifically for the case of past embedded underpast whereby the tense information of the matrix verb is transmittedto the embedded tense. The effect of such a mechanism is that thesemantic contribution of the embedded past tense is interpreted asco-indexed with the attitude time. The overlapping reading is thuspredicted (see also Stechow 1995). Finally, Ogihara (1996) proposes anoptional deletion rule applying at LF whereby a tense c-commandedby another morphologically identical tense can be erased. The resultingtenseless clause (a set of world-time pairs) is interpreted as holding at theattitude time via the truth conditions of belief-sentences, thus predicting
the overlapping reading.
As an illustration of this non-uniform solution, consider Ogihara’s
account in more detail. For the case of past embedded under past, thedeletion rule would turn (2), with the surface structure in (2a), intoan LF representation and corresponding semantic translation given in(2b), where t=interval in Ogihara’s notation and st=speech time (all
translations are taken from Ogihara’s proposal):
(2) The secretary believed that the senator was happy.
a. SS: The secretary PAST believe that the senator PAST be
happy.
b. LF
1:The secretary PAST believe that the senator ∅be happy.
∃t[t<st&believe/prime(t,
thesecretary/prime,∧λt1[behappy/prime(t1,thesenator/prime)])]

Silvia Gennari 39
c. LF 2:The secretary PAST believe that the senator PAST be
happy.∃t[t<st&believe
/prime(t,
thesecretary/prime,∧λt1∃t2[t2<t1&behappy/prime(t2,thesenator/prime)])]
Given the optionality of the rule, the two readings available in (2)
are predicted. If the rule does not apply as in (2c), the embedded PASTis interpreted relative to the believer’s attitude time t
1(representing
the believer’s now, counterpart of the actual attitude time) and thebackward shifted reading is obtained. If the rule applies, the resulting LFrepresentation in (2b) gives rise to the overlapping reading. Note thatin (2b), no temporal relation is expressed in the embedded sentence.The embedded sentence in (2b) denotes a set of world-time pairs, i.e.{/angbracketleftw
/prime,t/prime/angbracketright|the senator is happy at t/primeinw/prime}.The overlapping reading is
not obvious in (2b) but is obtained via the truth conditions. The truthconditions say that (2b) is true iff every world compatible with thesecretary’s beliefs in the actual world at the secretary’s believing timebelongs to {/angbracketleftw
/prime,t/prime/angbracketright|the senator is happy at t/primeinw/prime},i.e. if the secretary
locates her/himself at a world and time in which the senator is happy.Thus, if the attitude attribution is true and the secretary’s belief worlds-time pairs contain the ascribed world-time pairs at the attitude time, itfollows that the senator is happy at the secretary’s believing time (seeOgihara 1996: 120).
For the case of embedded present, the deletion mechanism yields
the following surface and LF representations and translation:
(4) The press will think that the president is out of town.
a. SS: The press PRES[woll] think that the president PRES be
out of town.
b. LF
1:The press PRES[woll] think that the president ∅be out
of town.∃t[st<t&think
/prime(t,thepress/prime,∧λt1[beout/prime(t1,thepresident/prime)])]
In (4a), the crucial assumption is the non-overt morphological
analysis of will,paralleled by that of would .willis morphologically
analysed as a combination of present and future morphologyrepresented as Pres[woll],while would is analysed as a combination of past
and future morphology Past[woll].Semantically, wollis interpreted as an
operator shifting the evaluation time into the future. The applicationof the rule under identity of present tense morphology in (4b) resultsin the overlapping reading as in (2b) before.
However, the non-application of the rule in (4a) does not yield the
double access reading because present tense is assumed to be indexical.

40Tense Meanings and Temporal Interpretation
Asimilar situation also obtains in present under past reports. As in
Abusch (1997), this situation triggers other mechanisms needed toobtain the double access reading of embedded present. Because inAbusch’s and Ogihra’s approaches, a general constraint prohibits theoccurrence of indexical tenses within intensional domains, the tenseshould move outside this domain and be interpreted de re,i na way
parallel to de reinterpretations of NPs. The intuition here is that there is
amismatch between the content reported and that intuitively believed.
In (4), for example, the press will surely represent the president as beingout of town at the attitude time, but not necessarily at the ST, a pasttime from the press’ perspective. Likewise for (7) below, where the STis future relative to Bill’s attitude time.
Thede re analysis (taken from Lewis 1979; Cresswell & Stechow
1982) accounts for the content-report mismatch intuition by proposingthat the speaker, rather than representing the de dicto content of the
attitude, refers to some state in the actual world. This state is theentity or resthat the attitude is about. The resmust be related to the
attitude holder via a causal acquaintance relation Rwhich is provided
by context. Intuitively, Rgives the way the attitude holder represents
theresin the belief worlds. The truth conditions for the attitude verb
specify that the attitude holder ascribes a temporal property to the res
state to which he/she must be acquainted. For example, in (7), the mostdiscussed example in the literature, Bill is acquainted with the state ofHillary having a big belly, and he believed this state to be the state ofHillary’s pregnancy. The semantic representations proposed for (4) and(7) are as follows:
(4) The press will think that the president is out of town.
c. LF
2:[CPPres 2[Sthe press Pres[woll] think s 2[CP1that [ Sthe
president s 1be out of town]]]]
∃s2[exist(st,s2)&∃t[t>st&think/prime(t,
thepress/prime,s2,∧λs1[beoutoftown/prime(s1,thepresident/prime)])]]
(7) Bill believed that Hillary is pregnant.
a. LF: [ CPPres 2[SBill Past believe s 2[CP1that [ SHillary s 1be
pregnant]]]]∃s
2[exist(st,s2)&∃t[t<st&believe/prime(t,
b,s2,∧λs1[bepregnant/prime(s1,h)])]]
sis a state and exist is an ‘operator’ such that [[exist]](s)(t)=1i f f
tis included in the duration of s.According to the truth conditions
proposed for de reattitude verbs, the translation in (4c) says that there

Silvia Gennari 41
exists a state s2at the ST such that the press ascribes to s2the property
of being a state in which the president is out of town. In Ohigara’sformulation, (4) is true iff (a) there is a resstates s
2at the ST, (b) there
is an acquaintance relation Rsuch that s2is the state to which the
press is connected via Rat the time of saying tin the actual world w,
and (c) for all worlds /angbracketleftw/prime,t/prime/angbracketrightcompatible with what the press thinks
inwatt,the state to which the press is acquainted via Rinw/primeat
t/primehas the property λs1[beoutoftown/prime(s1,thepresident/prime)].Similar truth
conditions are proposed for (7). The truth conditions thus require bothan acquaintance relation Rand the existence of the resstate in the actual
world.
The double access reading comes about because on the one hand,
Rrequires the attitude holder to be acquainted with the resat the time
of the attitude t(condition (b) above). This automatically guarantees
that the resstate in the actual world overlaps with the attitude time t.
On the other hand, the truth conditions also require that the resstate
picked up by Robtains in the actual world wat the ST (condition (a)
above). This ensures that the resstate also overlaps with the ST. The
double access reading is thus achieved through the fact that whateverstate the attitude holder is acquainted with via Rat the attitude time is
also the state that obtains at the ST in the actual world.
1.2.1 Some problematic consequences One problem with the non-
uniform approach is that, for every embedded sentence, the proposedsyntactic mechanisms predict a systematic ambiguity of the tensemorphology: an embedded past tense should receive both anoverlapping and a backward shifted reading. However, this is not thecorrect empirical prediction for eventive embedded sentences:
(8) The president believed that the senator called him.(9) The president denied that he hung up.
The overlapping interpretation is not possible in these cases. As
discussed in detail in the next section, past sentences of eventiveaktionsart only receive one possible interpretation in a consistent andsystematic fashion, while stative sentences are the only ones that allowan overlapping interpretation.
Consider also Ogihara’s analysis of would -sentences, where would=
PAST[woll]:
(10) The president said that he would clarify his position.

42Tense Meanings and Temporal Interpretation
a. LF 1:The president say PAST that he ∅[woll] clarify his
position.
b. LF 2:The president say PAST that he PAST[woll] clarify his
position.
The application of the rule in (10a) predicts a future reading relative
to the attitude time, as expected. However, the non-application of therule in (10b) incorrectly predicts other possible readings which are notattested. For example, it predicts a backward shifted interpretation inwhich the time of clarifying is prior to the time of the president’s saying.The situation is that of PAST
1…[PAST 2…[FUT…]] where the operators
are relative. Since the time denoted by PAST 2may be located long
before PAST 1,some future times relative to the time of PAST 2could
still be located in the past of PAST 1,thus yielding an incorrect reading.
Another problem with the non-uniform accounts (particularly
those of Ogihara and Abusch) relates to the treatment of the doubleaccess readings. The proposed analyses treat all double access readingsas involving a de re interpretation. Such an interpretation truth-
conditionally requires the following conditions:
(a) the existence of a state in the actual world overlapping both the ST
and the believing time;
(b) the existence of an acquaintance relation causally connecting the
believer with the actual resstate the belief is about.
Because the truth conditions for double access sentences must satisfy
these two requirements, the proposals imply that when these conditionsdo not obtain, the sentences are false or perhaps infelicitous. However,neither of these requirements is necessary to yield an obviouslytrue double-access sentence. The examples below show the intuitiveinadequacy of conditions (a) and (b) above.
Consider the following. Imagine a situation in which Bill sees
Hillary wearing a pretty loose dress at a party that made her lookpregnant. Now, the party is over, Bill is in a business trip and Hillary ofcourse does not look pregnant any more, as in fact, she never was. Inthis context, it is perfectly fine to utter (7):
(7) Bill believed that Hillary is pregnant.
Here, Bill is only acquainted with Hillary’s loose-dress state in the past
and this state does not obtain in the actual world at the ST. Of course,because Bill was deceived, it follows from his belief worlds that Hillaryis pregnant at an interval including the ST (some future time from Bill’s

Silvia Gennari 43
perspective). The state to which the believer is acquainted need not be
true in the actual world at the ST as required by the de reaccount in
the factuality condition (a). Consider also the following cases:
(11) Betty told little Bill that an angel is watching him.1
(12) The detective reasoned (concluded) that the murderer is still in
town.
(13) After another suspicious excuse, Hillary believed (concluded) that
her husband is having an affair.
(14) Socrates believed that the soul is located in the stomach.
In (11)–(14), the belief worlds entail the truth of the embedded state
at the ST, but the attitude holder need not intuitively be acquaintedwith any particular actual state that overlaps with both the attitude timeand the ST. The situations in which these reports could be true requireneither the existence of the embedded state nor the acquaintancerelation (condition (a) and (b) above). This is clear in (11). For (12)–(14), one can imagine situations that led the attitude holders to makecertain conclusions, but such situations are not necessarily the actual res
states the attitudes are about.In(13), for example, Hillary does not
seem to believe ofsome state or event (e.g. the excuse) that it has the
property of being the state/event of her husband having an affair. Theexcuse and the affair are two different things in her mind. She simplymakes a conclusion from other beliefs previously acquired or from herknowledge of her husband. Similarly, Socrates may have believed thecomplement of (14) as a statement compatible with his system of beliefs.This belief, a belief true of all times, may have followed from others hehad, without requiring an acquaintance relation with any particular orgeneric state that also obtains at the ST.
In all these cases, intuition suggest that the existence of an
acquaintance relation and/or an actual state overlapping with both theST and the believing time is not truth-conditionally required. Rather,the states in question may only exist de dicto -like in the belief worlds.
The belief worlds entail the truth of the embedded state at the ST inthe belief worlds (a future time from the perspective of the believer)but the factuality of this state is not required. This thus challenges theadequacy of the temporal de reanalysis.
1Note that it does not matter that Betty lies to little Bill. From the perspective of the speaker, the
attribution could be true.

44Tense Meanings and Temporal Interpretation
1.3 Anew approach
Given the difficulties of the non-uniform approach, I will maintain the
following: (a) tenses can have a single uniform meaning in all embeddedand independent occurrences, provided that this meaning is properlycharacterized, and (b) the temporal interpretation of a given sentencedepends on the interaction of the tenses’ meanings with the sentence’saktionsart. The latter is an old observation that I intend to systematicallyintegrate into a uniform theory of tense interpretation (see for exampleHinrichs 1986; Hornstein 1990; Dowty 1986; Parsons 1990; Kamp &Reyle 1993; Ter Meulen 1995). To motivate this latter point, considerthe following:
(15) The secretary said that
a. the president was sad.
b. the president was arriving.c. the president arrived at 9 am.d. the president prepared a speech.e. the president worked hard.
(16) The president decided that in ten days he would say to his party
that
a. he was in trouble.
b. they were having their last meal together.c. his lawyers left him.d. his friends asked for more money.e. his executive committee talked too much.
Note that contrary to current theories’ predictions, the overlapping
reading is not possible in the cases (c) and (d) of (15). Similarly, theoverlapping reading between the interval denoted by the innermostpast tense and that of would in (16) is not available in cases (c) and (d).
The difference between these examples and those in (a) and (b) is theaktionsart of the embedded sentence. The examples in (a) and (b) havestative complements while those in cases (c) and (d) have eventive ones(an achievement and an accomplishment, according to Vendler’s (1967)classification).
Note that (15e) and (16e) could receive two possible interpretations.
If they are interpreted as a single occurrence of the activity (sometimescalled an episodic reading), only the backward shifted reading obtains.One would think that the overlapping interpretation should be possiblebecause activities are atelic predicates and no pragmatic informationprecludes such a reading. For example, the president could have been

Silvia Gennari 45
working hard at the time the secretary said so in (15e). However, if
(15e) and (16e) are interpreted as a single occurrence of the activity,the overlapping interpretation is not available. The overlapping readingonly arises if the embedded sentence in (15e) and (16e) is understoodas generic or as a habitual activity, and it is well known that genericsentences are stative.
2
Note also that sentences (15b) and (16b) with progressive aspect may
receive an overlapping interpretation. Progressive aspect, which sharessome properties with stative aktionsart (unlike activities per se), coerces
any type of event into what Moens & Steedman (1988) call progressivestates. Thus, examples in (15) and (16) suggest that aktionsart haspredictable effects on temporal interpretation. Stative properties (sharedby habitual and progressive sentences) are associated with overlappingreadings while eventive sentences only generate sequential readings.
Asimilar influence of stativity on the temporal reading can be
observed in future and modal sentences. Such sentences may receivean overlapping reading relative to the ST as in (17), particularlyin a context in which the speaker makes a supposition about thepresident’s current whereabouts (see Boyd & Thorne 1969; Wekker1976; Steedman 1997; Palmer 1969; Condoravdi 2001):
(17) The president will/may/must be at home (now).
(18) The president will/may/must leave (now).(19) The president thought that the secretary would/may/must be at
home (and decided to call her).
(20) The president thought that the secretary would/may/must leave.
In contrast, the overlapping reading is not available in (18), despite
the fact that the adverbial modification is now.Asbefore, overlapping
readings only arise with statives, while eventive sentences receivesequential readings. The same generalization holds for (19) and (20).
3
Independent evidence that aktionsart determines the temporal
readings comes from the interpretations of perfect and imperfect pasttenses in Romance. While perfect aspect in embedded sentencescorrelates with sequential readings, imperfect aspect is usually correlatedwith overlapping readings. Imperfect in Romance, like the progressivein English, shares entailments and semantic properties with statives(Cipria & Roberts 2001; Gennari 1999a). Consider the following:
2Generic sentences satisfy all the tests of stative predicates so that as far as aktionsart properties are
concerned, generic sentences are statives (see Carlson & Pelletier 1995).
3See also Klein 1994, for observations along these lines regarding the interpretation of German
present tense.

46Tense Meanings and Temporal Interpretation
(21) a. Juan dijo que Mar ´ıa estaba enferma.
John say-past-perf that Mary be-past- imperf sick.
b. Juan dijo que Mar ´ıa estuvo enferma.
John say-past-perf. that Mary be-past- perfsick.
(21a) receives an overlapping reading, like (2) in English, although a
backward shifted reading may also be available with an appropriatecontext. In contrast, (21b) only has the backward shifted reading.Thus, in Spanish, stative properties (manifested through aspect) arealso determinant of the temporal readings available in embeddedsentences.
4
More importantly, independent evidence for the influence of
aktionsart on temporal interpretations also comes from discoursecontexts, although the relation affected in this domain is inter-sentential, i.e. that between the interval in which a sentence is trueand the truth interval of the previous sentence in discourse. As noted inDiscourse Representation Theory (Hinrichs 1986; Partee 1984; Kamp&Reyle 1993, see also Ter Meulen 1995), stative and progressive
sentences usually yield overlapping readings relative to the truth intervalof the previous sentence while event sentences generate sequentialreadings.
Given these facts, the question arises as to why and how aktionsart
properties have an effect on temporal interpretations. To answer this,Iwill argue that stative predicates and sentences (including habitual
and progressive ones) all share the property of logically entailing orlexically specifying a superinterval which contains the truth intervalof the sentence provided by the tense operator. It follows from thisthat the superinterval of stative sentences (if large enough) may overlapwith the ST or other times contextually provided. In contrast, eventverbs and sentences have the property of lexically specifying or logicallyentailing that the event in question is maximally contained withintheir intervals of truth. This excludes the possibility of an overlappinginterpretation with the ST or other times contextually provided, giventhe meaning of the temporal operators. Thus, rather than non-uniformmeanings, tenses have the same meaning in all contexts but aktionsartproperties lead to the overlapping readings. This would be, in principle,asimpler and more desirable theory, since (a) it would not claim
vacuous temporal morphology, (b) it would not require any syntacticmechanism insensitive to aktionsart; and (c) it would explain tensemeanings in a uniform way.
4See also Kamp & Rohrer’s (1983) analysis of simple past and past imperfect in French.

Silvia Gennari 47
2THEORETICAL FRAMEWORK
2.1 Basic definitions
In this paper, I assume the traditional view of tenses as quantifiers, and
of embedded sentences as sets of world-interval pairs (within a singletemporal structure). Thus, abelieves Q atwand interval iis true iff for
allw
/primeandi/primecompatible with a’sbeliefs at wandi,Q(w/prime)(i/prime)=1(as
in Stechow 1995).5Because I will not be concerned with de se orde
rereadings here, the traditional definition of belief reports will do. The
putative de re interpretations of tenses are treated below as reports of
implicit attitudes (as in Gennari 1999b), which do not require any de
remechanism. Also, the notion of local evaluation time is understood
as in the traditional logical sense: the evaluation time (or interval) ofaquantifier tense is the time with respect to which the truth of the
sentence is evaluated. For example, in an expression such as ∃i[i<
i
0&Q(i)],i0is the local evaluation time. This time can be bound by
alambda operator or indexically refer to the ST. Finally, I will refer to
the event time or interval (ET) of a sentence and occasionally, to itsreference time (RT). By event time ,Imean the interval specified by the
tense operator in which a sentence is true. In ∃i[i<i
0&Q(i)],the ET
isi.Following Dowty (1986), Hinrichs (1986), Nerbonne (1986) and
others, I assume that the reference time of a sentence is given either byadverbs that temporally locate the sentence or by the event time of theprevious sentence in discourse.
Toillustrate these concepts, consider some examples:
(22) David received a letter yesterday.
∃i[i<st&receive
/prime(i,d,aletter/prime)&i⊂yesterday/prime]
(23) Anna believed that David received a letter yesterday.
∃i/prime[i/prime<st believe/prime(i/prime,a,∧λi0∃i[i<i0&receive/prime(i,d,aletter/prime)
&i⊂yesterday/prime]
(22) is true iff at an interval before the ST (and included in yesterday),
David received a letter. The ET is the interval iat which David
received the letter, some (short) interval within the RT yesterday .The
ST is the local evaluation time for the past operator, because the truth ofthe sentence is evaluated relative to the ST (non-embedded sentences ingeneral take the ST as their evaluation time). In an embedded sentence,the evaluation time will be the time representing the attitude time in
5Little hinges on these assumptions because the proposal can be recast in any equivalent
framework. For example, in a framework treating tenses as variables as in Schlenker (1999), ortreating belief reports as self-ascription of properties, as in Ogihara (1996).

48Tense Meanings and Temporal Interpretation
the belief worlds, as i0in (23). Sentence (23) is true at the ST iff at
atime i/primeprevious to the ST, Anna has a belief and for all /angbracketleftw0,i0/angbracketright
compatible with Anna’s beliefs at wandi/prime,there is a time prior to i0
at which David received a letter in w0.The world-time pairs /angbracketleftw0,i0/angbracketright
represent Anna’s belief worlds (Anna’s here and now) at the actual timeof the belief. Therefore, the time of these pairs are the evaluation timerelative to which the truth of the embedded sentence is evaluated.
2.2 The effect of aktionsart on temporal readings
In this paper, I adopt Taylor’s (1977) and Dowty’s (1979, 1986) defining
criteria of aktionsart classes. The criteria are logical entailments of asentence Qat any interval iifQis a sentence of a given class as
determined by the internal semantic composition (Verkuyl 1993). Asentence Q(i)is stative iff it entails that Qis true at all instants within
i.Asentence Q(i)is an accomplishment/achievement (or a telic event)
iff it entails that Qis false at all subintervals of i(where iis the minimal
interval at which the change of state takes place).
6Q(i)is an activity
(or an atelic event) iff it entails that Qis true at all subintervals of i
down to a certain limit of size. Note that in general, the duration ofthe interval or subinterval at which any sentence is true depends onworld knowledge about the event or state in question (Dowty 1986).For example, activities such as gardening may not hold at very small
subintervals. Similarly, the intervals of being sad orwriting a letter are
typically shorter than those of being German orliving a corrupt life .
Ialso adopt Dowty’s (1986) general treatment of the relationship
between aktionsart and temporal interpretation, but while Dowty’sproposal referred to the effect of aktionsart on discourse interpretation(specifically, to the relation between the sentence’s ET and thediscourse RT), I interpret it to apply to intra-sentential domains, i.e.to the relation between a sentence’s ET and its local evaluation time.As an illustration, consider the interpretation of independent sentencesof different aktionsart with future tense:
(24) John will leave (now).
FUT
/prime(leave/prime(j)(i))= ∃i[i>st&leave/prime(j)(i)]
6This entailment follows from Dowty’s truth conditions of telic events represented with the
BECOME operator. BECOME [Q]is true at an interval iiff (a) Qis false at an interval jcontaining
the initial moment of i,(b)Qis true at an interval kcontaining the final instant of i,a n d( c )t h e r ei s
no subinterval of iin which conditions (a) and (b) hold. It follows that the event is not true at any
subinterval of iand tha tt h i si nterval is the minimal interval at which the relevant change of state
could take place.

Silvia Gennari 49
Under the standard analysis of future tense, (24) is true at the ST iff
there is an interval iafter the ST such that John leaves at i.Note that
(24) does not mean that John is leaving at the ST but that he is aboutto leave (near future). To understand why an overlapping reading withthe ST is not possible, consider the aktionsart entailments. First, a telicsentence Q(i)entails that Qis false at all subintervals of i.Second, this
entailment in turn entails that if Qis true at an interval i,Qis false at
all superintervals of ias well (Dowty 1986), for if Qwere true at some
superinterval i
/primeofiand at iitself, it would be false at all subintervals
ofi/prime,including i,according to the telicity entailment, contradicting
the assumption. Thus, if John leaves is true at some future interval i,
the sentence entails that John’s leaving does not hold at any subintervalwithin i,and therefore, it does not hold at any superinterval of ieither.
Given this and the standard meaning of future, it follows that John will
leave now crucially cannot overlap with the ST, despite the adverbial
modification.
7
Now consider a stative sentence:
(25) John will be at home (tomorrow/now).
FUT/prime(beathome/prime(j)(i)) = ∃i[i>st&beathome/prime(j)(i)]
(25) can have a future or an overlapping interpretation relative to
the ST, as each of possible temporal adverbs indicates. In the presentreading, willis inferential: it requires a context in which the speaker
makes an inference from certain background assumptions (cf. Kratzer1981), and is roughly equivalent to the use of mustin the same context.
As indicated earlier, the overlapping reading only arises with stativesentences but not eventive ones. The state/event distinction also causesadifferential effect on temporal interpretation with other expressions
that are clearly modal, as in (17) and (18) above. This indicates thatthe possible temporal location of the sentence in question depends onaktionsart, but not on modality per se.
Leaving aside the problem of whether modality is conveyed with
future tense,
8the question of interest here is why the effect of aktionsart
7What I have just said seems incompatible with the occurrence of nowin (24). If the adverb
modifies the ET of the sentence (the leaving time), this information contradicts the future tense.The solution to this puzzle is that nowdenotes an interval and not an instant. Such an interval could
be extended enough to include both the interval in which John leaves and the ST. The punctualityof the event still obtains as in the case without the adverbial now,because nowis an interval that may
encompass both the ST and the (punctual) event of leaving.
8Several arguments have been put forward to treat willas conveying both temporal and modal
information. Since the purpose of this article is not to elucidate its modal meaning, I will assumethatwillconveys future temporal information and that its modality can be independently explained
(following Gennari 1997, 2000; Condoravdi 2001; and others).

50Tense Meanings and Temporal Interpretation
on temporal interpretation occurs. The answer, suggested in different
forms in the literature (Dowty 1986; Kamp & Reyle 1993; Lascarides&Asher 1993), is that stative sentences have temporal properties that
distinguish them from event sentences. Note that (25) entails its truthat all subinstants of i.Incontrast to (24), this does not exclude the
possibility that the sentence is actually true at a larger interval thatproperly includes its ET, the future interval i.Infact, when states
are asserted, the normal assumption is that they are true at a largerinterval surrounding their ETs and their RTs, if any (Dowty 1986). Icall this assumption the superinterval property. This property conveysthe information that be
athome(i) in (25) holds at a superinterval
surrounding i.Itfollows that this superinterval can include the future
interval iand the ST. This inference is schematically represented below,
where Prepresents the modal operator possibly ,isis a superinterval and
bold material highlights the key statements of the inference:
(26) John will be at home.
(1)∃i[i>st&beathome/prime(j)(i)]
(2)∃is[∃i[i>st&i⊂is&beathome/prime(j)(is)]]
∃is[∃i[i>st&i⊂is&beathome/prime(j)(is)&Pst⊂is]]
The meaning of (26) in (1) and the assumption of the superinterval
in (2) entail the possibility that the ST is included (or overlaps with)the superinterval i
ssurrounding i.The overlapping reading comes
about because the superinterval isat which the state of being at home
holds is able to overlap with the evaluation time of the entire sentence.However, this is a mere possibility. Whether the overlapping or futurereading obtains depends on the RT and our knowledge of the typicalduration of the state involved.
When an adverb such as nowortoday is added to (26), the
state’s superinterval is interpreted as surrounding this RT. With thisassumption plus the lexical meaning of now,itfollows that John will be
at home now entails that John will be at home for a superinterval that
includes the interval now,i.e. John is and will be at home for a while.
This is schematically represented in (27). Line (3) below assumes thatnow’denotes an interval including the ST:
(27) John will be at home now.
(1)∃i[i>st&beathome/prime(j)(i)&i⊂now/prime]
(2)∃is[∃i[i>st&i⊂is&beathome/prime(j)(is)&now/prime⊂is]]
(3)st⊂now/prime
∃is[∃i[i>st&i⊂is&beathome/prime(j)(is)&now/prime⊂is&st⊂is]]

Silvia Gennari 51
Instead, if the RT is tomorrow night ,the superinterval is not necessarily
interpreted to overlap with the ST, although the possibility is openas in (26). The superinterval of being at home would include the
interval denoted by tomorrow night .But because being at home denotes
atemporary state that does not typically hold for a long period, the
superinterval of the state is most likely entirely located in the futureof the ST. Knowledge of the contextual situation can also supportaparticular interpretation, such as the time of the day in which the
assertion is made, the knowledge of John’s home habits, etc. Thus, theknowledge of the typical duration of the state in question plus discourseconsiderations will determine the actual temporal duration and locationof the superinterval relative to the evaluation time.
Finally, consider what happens if an activity sentence is used:
(28) John will run (now).
FUT
/prime(run/prime(j)(i)) = ∃i[i>st&run/prime(j)(i)]
(28) entails its truth at all subintervals of i.Aswith states, but unlike
telic events, this entailment is in principle logically compatible withan implication of a superinterval so that the overlapping reading withthe ST should be possible. However, this reading is not available,so that activities pattern with telic events in this respect. This isbecause activities have temporal and causal internal developments (i.e.lexically specified aktionsart properties) that negate the possibility ofasuperinterval surrounding the ET. The kind of events that activities
denote is such that they can be sustained for a while, but because theirduration is contingent on a causal force initiating and sustaining therelevant action, they stop when the causal force does. So, they are nottypically assumed to obtain indefinitely. On the contrary, activities arenormally assumed to be contained within their interval of evaluations.In contrast to states, they typically implicate that they take place withinsome arbitrary initial and end point of their interval of truth (Smith1991). Consider the following (where /squigglerightmeans implicates ):
(29) John danced (this morning). /squigglerightHe started and stopped dancing at
some point (this morning).
(30) John was sick (this morning).
∗/squigglerightHe started and stopped being
sick at some point (this morning).
As expected on the basis of the superinterval property, states are not
implied to be wholly contained within their ETs (whichever intervalspecified by the past tense) or their RTs. But this inference is availablewith episodic activities (see next section for more details). This explains

52Tense Meanings and Temporal Interpretation
why (28) only receives a future reading (the running obtains within
the future interval i)and indicates that the superinterval property is not
operative with activities. Thus, episodic activities are understood to bemaximally contained within the limits of their ETs and RTs. Therefore,activities do not yield overlapping readings relative to the evaluationtime but sequential ones.
In sum, different aktionsarten have different effects on the temporal
interpretations. Telic events and activities have internal temporaldevelopments that are incompatible with the possibility of their truthat larger intervals. Therefore, they do not overlap with the localevaluation time. In contrast, states have the superinterval property.This property is an inference that there is a larger interval (containingthe state’s ET and/or RT) at which the state holds. Given this, theinferred superinterval is able to overlap with the local evaluationtime.
2.3 The superinterval property
Since my analysis of temporal interpretations heavily relies on the
superinterval implication of states, in this section, I discuss in detailsits presence with states, as opposed to events, based on their lexicalaktionsart and discourse interpretation. To start with, note that typically,states are facts: they characterize the static structure of the worlddescribing locations and physical or psychological properties of entities.Tostart or stop being in a state depends on other events (e.g. being
German ), but the state’s persistence does not. States remain true without
the aid of an external force and their persistence is independent ofwhatever caused them. In principle, if a stative sentence holds at aninterval i,and no intervening event occurs between iandi+1that
changes this state, then the sentence also holds at i+1(Katz 1994;
TerMeulen 1995). This is the intuition underlying the superinterval
property of states. Dowty (1986) calls it the principle of inertia.
9
This situation radically contrasts with that of events. Eventive
verbs in general have the change-of-state entailment (Dowty 1979).Telic events entail one definite change, while activities entail a
sequence of small changes. These changes are caused by some agentiveforce, which may be sustained for a while. Given this, we do notassume that eventive sentences hold at a superinterval because thereis an intervening cause-event bringing about the changes, and the
9See also Lascarides & Asher’s (1993) principle of States Overlap ,which they interpret to express
Grice’s maxim of relevance.

Silvia Gennari 53
cause-effect relation is sequential. For example, if an activity is true
at interval i,intuitively, it does not necessarily hold at a later time
i+1because the causal force applying at imay not apply at i+1.
The persistence of activities contingently depends on other events.These observations suggest that stative sentences are fundamentallydifferent from eventive ones.
10States have the property of obtaining
for relatively long periods, while events do not.
Evidence for this temporal contrast between states and events
comes from discourse interpretation. Dowty (1986) observes thatthe assumption that states persist in time makes (31) infelicitous, nomatter how many events intervene between the sentences given, unlessinformation is given otherwise. Also, as noted in DRT, a stativesentence in a narrative is most usually interpreted to obtain before andafter the event denoted by the previous sentence as in (32). As suggestedby examples (29) and (30) above, the superinterval of a state can obtaineven beyond the RT or the temporal location given in discourse (in(32), beyond this morning ):
(31) ?? The book was on the table at t
0……..Mary put the book on the
table at tn.
(32) Mary went to see the president this morning. He was sick.
This contrasts with activities and events in general which normally
receive sequential interpretations as in (33). Some authors (e.g.Hinrichs 1986; Dowty 1986), claim that activities, by virtue of beingatelic events, may overlap with other events in discourse. This ispossible as in (34), because the temporal relations between events indiscourse ultimately depends on the contingency and causal relationsestablished between them.
(33) Mary went to see the president (this morning). She asked him
questions about the project.
(34) Mary talked to the president (this morning). She asked him
questions about the project.
Even when overlapping readings are possible, it is critical that activities
are not implied to obtain beyond their RTs. In (34), for example,interpreters do not assume that the questions lasted for longer than the
10For more about the contrast between states and events, see Katz (1994).

54Tense Meanings and Temporal Interpretation
morning or the duration of the meeting. This indicates that, even when
activities may overlap other events in discourse, they do not behave likestates. In sum, lexical aktionsart properties and discourse interpretationsupport the claim that states are assumed to hold for periods largerthan those specified by the tense and adverbials (if any). This contrastswith events in general which are normally interpreted to be maximallycontained within a given spatio-temporal domain.
11
2.4 The superinterval property: semantics or pragmatics?
An important question that needs to be addressed is the exact nature of
the superinterval property. Dowty (1986) argues that the superintervalis a pragmatic implication because it can be cancelled when stativesreceive inceptive readings. Such readings are exemplified in (35), takenfrom Dowty (1986):
(35) Away in front, the engine whistled. Trees, hills and road, slid
sideways and were gone.
This reading differs from other readings in that each state (the
trees, the hills and the road being gone) cannot obtain before each
corresponding thing has slid sideways, i.e. they do not overlap withthese events. This is because each event of sliding sideways entails itsresulting state (the things being gone). Therefore, the only possibleinterpretation of the states is one in which they start when the eventbringing them about has occurred. However, this does not mean thatthe superinterval property is cancelled. It means that the states of being
gonecannot possibly obtain before their causing events in discourse.
But once the states start obtaining, the property that they persist intime according to their typical duration is still part of their meanings(from the standpoint of the observer, the trees are gone for ever). Thus,the superinterval property is not cancelled. Rather, it surrounds the ETin a way compatible with the discourse (by extending into the future).
If the superinterval of state does not really disappear but surrounds
the ET of the sentence in a way compatible with the discoursecontext, the source of the superinterval property is the lexicalmeaning of the stative verb or phrase involved, rather than pragmaticknowledge. Pragmatic knowledge determines the ultimate location ofthe superinterval but not its existence. Contrast for example, being
crazy tobeing sad ,knowing tobelieving .Orconsider the distinction
11See Lascarides & Asher (1993) for a similar approach to discourse interpretations based on world
knowledge assumptions.

Silvia Gennari 55
between individual-level and stage-level predicates, which is taken
to be lexical (cf. Kratzer 1989; Chierchia 1995). Lexical meaningsconvey prototypical information about their denotations (Rosch &Mervis 1975), and part of this information in verbs concerns the typicalduration of events and states. When these lexical meanings combineinto sentences, the sentence carries this information with it. Thus,rather than claiming that the superinterval implication is a pragmaticcancelable implication, I claim that it is an inference derived from thelexical meaning of verbs, and therefore, part of the computation of themeaning of the sentence.
3TENSE MEANINGS AND TEMPORAL
INTERPRETATIONS
3.1 The interpretation of past tenses
3.1.1 Simple Past Recall that past under past was problematic for a
uniform theory because two readings were available and no postulatedmeaning of the simple past tense could account for both of themappropriately. In my approach, past tense has a single meaning, equaltoλQλi
0[∃i[i<i0&Q(i)]],which requires the temporal property
it modifies to be evaluated at a time prior to the evaluation time.Thus, past tense is treated as evaluation time sensitive. When past tenseoccurs in an independent sentence, the evaluation time will be the ST.When past tense is embedded under another tense, the past propositionshould be evaluated relative to the local evaluation time representingthe believer’s now or attitude time. This is clear in the case of pastembedded under future:
(36) Bill will tell you that Mary’s exam went well.
The temporal interpretation of the embedded complement is in relation
to Bill’s future attitude time. The interval specified by the past tense isprior to the future attitude time but can be located either after, beforeor overlapping with the ST. The embedded past tense does not specifyany relation relative to ST.
Consider now how the problematic temporal reading of past
under past come about. The semantic composition of the sentence(37) (omitting the outermost world variable) first combines the tensemeaning with a tenseless clause:
past
/prime(marybesick/prime)=λQ/angbracketlefti,t/angbracketrightλi0∃i[i<i0&Q(i)](λi/prime[besick/prime(m)(i/prime)])

56Tense Meanings and Temporal Interpretation
to yield λi0[∃i[i<i0&besick/prime(m)(i)]].The embedding verb of type
/angbracketleft/angbracketleftw,/angbracketlefti,t/angbracketright/angbracketright,/angbracketlefte,/angbracketlefti,t/angbracketright/angbracketright/angbracketrighttakes as arguments a function from a world to a
temporal abstract, an individual and an interval:
think/prime(∧λi0[∃i[i<i0&besick/prime(m)(i)]])
=λxλi/prime[think/prime(∧λi0[∃i[i<i0&besick/prime(m)(i)]])(x)(i/prime)].
After combining with the subject john’ and the matrix past tense, the
result is the following:
(37) John thought that Mary was sick.
=λi1∃i/prime[i/prime<i1&think/prime(i/prime,j,∧λi0∃i[i<i0&besick/prime(m)(i)])]st
=∃i/prime[i/prime<st&think/prime(i/prime,j,∧λi0∃i[i<i0&besick/prime(i,m)])]
At the last step of the derivation, the temporal abstract resulting from
the sentential composition applies to the ST of the utterance context.The truth conditions say that (37) is true iff there is a past time
i
/primeat which John thinks, and for all his cognitive worlds and times
/angbracketleftw0,i0/angbracketrightcompatible with his beliefs at i/prime,Mary is sick at a time earlier
than i0.Thus, if John’s thinking time i/primein (37) is part of the word-
time pairs compatible with his cognitive state (as it is under normalcircumstances), it follows that Mary is sick at a time earlier than John’sthinking time i
/prime.Inother words, for some belief world-time pairs,
/angbracketleftw,i/prime/angbracketrightwould be included in {/angbracketleftw,i/angbracketrightMary is sick earlier than iin
w}.This is similar to Ogihara’s (1996) account of the backward shifted
reading.12
Now, given the superinterval property associated with states, i.e.
that there is a superinterval isthat contains i,and that for all instants
within is,Mary is sick, it follows that both the interval of thinking and
the interval of being sick can overlap. The superinterval of being sickcan extend from an interval earlier than the thinking interval until thethinking interval itself, as in (26). This is indeed a possible reading and,in absence of adverbial modifications or contextual information, themost likely one.
13
Note that the truth conditions themselves do not explicitly require
the overlapping reading. Rather, they underdetermine the actual
12In what follows, I will give truth conditions in a simplified manner, mostly referring at the
temporal relations and omitting the complex definition of belief reports.
13This may be because there is a conversational implicature arising from the choice between simple
past and past perfect. Past perfect is more informative because it only receives the backward shiftedreading. If the speaker does not choose past perfect, he/she implicates that the backward shiftedreading does not obtain. This makes the overlapping reading the default reading in absence of othertemporal specifications.

Silvia Gennari 57
temporal extension of the embedded state and its superinterval. The
temporal operators only determine the relative location of the eventintervals (the ETs). The possible duration of the embedded propositiondepends on the typical duration of sickness, which is available to bothspeakers and interpreters. Thus the speaker not only attributes thethought that Mary was sick a while before the thinking time, butthat she was and would continue to be sick for a while, given typicalknowledge associated with being sick .The speaker’s intended content
includes the knowledge of the superinterval, the availability of whichfurther specifies the truth conditional content.
However, the superinterval property associated with the state of
being sick does not have to overlap with the thinking interval; thisreading is only available. The backward shifted reading is in principlealso possible. Whether one reading or the other obtains, i.e. the specificduration and location assigned to the embedded state, will depend onthe temporal information available in the context. An obvious case ofmore specific temporal information is the occurrence of adverbs as inYesterday, John said that Mary was sick last year .Another case in which
contextual information constrains the temporal interpretation of thecomplement is by means of the RT in discourse:
(38) John went to a private school when he was a child. He said the
school was awful . . .
Since the events mentioned by John are located in John’s childhood
(the RT introduced by the when-clause), the state of the complement is
accordingly located around this time and not around John’s saying time.Note that the saying event is interpreted sequentially relative to theprevious RT, i.e. later than the childhood events. This is consistent withthe observations above regarding the behaviour of events in discourse.Also, under our definition of RT, saiddoes not introduce a new RT
for the embedded sentence. The RT of a sentence is either its temporaladverbs or the ET of the previous sentence in discourse. Thus, both theembedded state and the saying event are located relative to the intervalof the previous sentence. In this respect, complement clauses behaveas independent clauses, i.e. they tend to overlap with the RT fromprevious sentences in discourse. Thus, adverbials and contextual factorsmay constrain the interpretation of the superinterval (its past locationand duration). The superinterval property only makes the overlappingreading available, while the actual reading depends on its compatibilitywith the context.

58Tense Meanings and Temporal Interpretation
Next, consider a case in which an event complement is involved:
(39) John believed that Mary went to the party.
∃i[i<st&believe/prime(i,j,∧λi0∃i/prime[i/prime<i0&gototheparty/prime(i/prime,m)])]
Note that believing is a state and so it may be included in a
superinterval, according to the superinterval property. It thus seemspossible that the time of going to the party could be included within theinterval of believing. However, this reading is not available because theentailment associated with states precludes this reading. The entailmentof the believe -sentence is such that for all instants twithin i,the
believing interval, it is true that John believes p.Since the embedded
complement is a past proposition that is believed at each instant twithin
i,the entailment of the entire sentence amounts to the following: for
each instant twithin the believing state, there is a prior interval in
which Mary went to the party. In other words, the belief is that therewas a single event of going to the party and this belief is true at allinstant twithin the believing interval i.Thus, the believed event of
going to the party precedes all these instants. It follows that the intervalof going to the party cannot overlap with the interval of believing.
The same type of reasoning applies to other aktionsart combinations
of main and complement sentences. An event main sentence such asJohn say p will have to be contained within its interval of evaluation
so that the event complement pwill have to be true prior to the
saying interval. No overlapping reading is possible. Episodic activitiesoccurring with both stative and eventive matrix verbs will behavelike telic events because they are true within their ETs and they donot have the superinterval property. Their intervals could be more orless large but they will be wholly located before the attitude time forsimilar reasons discussed in (39): For each instant within the believingstate, there is a prior interval within which the activity took place.Note that this entailment of the believing state also obtains with stativecomplements: For each instant within the believing state, there is aprior interval (the ET) in which the embedded state is true. However,astative complement could generate an overlapping reading due to the
superinterval property surrounding the ET of the complement ,asin(37).
Thus, the aktionsart properties of the sentences involved determine theavailable temporal readings: Embedded states yield overlapping readingsif context permits while embedded events yield sequential readings.
3.1.2 Past progressive Following Dowty’s (1979) treatment of the
progressive and tense and aspect composition, I assume that prog is

Silvia Gennari 59
treated as a VP modifier of type /angbracketlefte,/angbracketlefti,t/angbracketright/angbracketright,/angbracketlefte,/angbracketlefti,t/angbracketright/angbracketright,rather than as a
sentential operator. Its meaning applies to a VP-meaning, the subjectargument and a temporal argument. prog(Q
/angbracketlefte,/angbracketlefti,t/angbracketright/angbracketright)(x)(i)is true at w
iff there is a superinterval i/primecontaining isuch that for all possible inertia
continuations w/prime,[Q(x)(i/prime)]is true at w/prime.14An important property of
this composition is that the tense operator provides the local evaluationtime for the aspectual operator. For example, in an expression such asPAST
/prime[prog/prime-run/prime(j)],the aspectual operator prog is evaluated relative
to the past interval of PAST.
The standard definition of progressive contains the key to explain
its possible temporal interpretations. First, the definition satisfies thedefining criterion of stative aktionsart: If prog[Q]is true at i,for
any subinstant tofi,there is a superinterval containing t(and i)
where Qis true, therefore, prog[Q]is also true at every subinstant
ofi.Thus, prog applies to any other aktionsart and returns a stative
proposition. Second, the definition entails that there is a superinterval
i
/primein which the modified proposition is true. As with other stative
sentences, this property explains why progressive sentences usuallygenerate overlapping readings relative to the local evaluation time.Consider some examples:
(40) John said that Mary was going to the party.
∃i[i<st&say/prime(∧λi0∃i/prime[i/prime<i0&prog goparty/prime(m)(i/prime)])(i)(j)]=
∃i[i<st&say/prime(i,j,∧λi0∃i/prime[i/prime<i0&∃is[i/prime⊂is&goparty/prime(is,m)]])]
(41) John said that he would tell his mother that they were having
their last meal.
∃i[i<st&say/prime(i,j,∧λi0∃i/prime[i/prime>i0&tell/prime(i/prime,j,hismother/prime,∧λi1
∃i/prime/prime[i/prime/prime<i1&prog have(i/prime/prime,meal/prime,they/prime)]])])]
(40) is true iff John said in the past that there was another prior
time surrounded by a superinterval in which Mary was in the processof going to the party. The overlapping reading obtains as in the caseof a past state under past, i.e. the superinterval introduced by the truthrequirements of prog can overlap with John’s saying interval. Similarly
for (41). The superinterval introduced around the past time i
/prime/primebefore
the telling interval may overlap the telling.
14More precisely, prog Q(x)(i)is true at a world wiff (a) there is an superinterval i/primeproperly
containing i,( b )iis not the final or initial subinterval of i/prime,a n d( c )f o rall worlds w/primein the inertia
set of possible continuations of wati,Q(x)(i/prime)is true in w/prime.Recall that Dowty (1986) argues
that the possibility that iis the initial subinterval of i/primeshould be excluded, although the definition
of Dowty (1979) did not incorporate this restriction. This is because progressive sentences do notreceive inceptive interpretations and their ETs are clearly understood in the middle of the eventdenoted (e.g. At 5, I was sleeping ). In the text, I assume this restriction and omit consideration of the
modal component of prog.

60Tense Meanings and Temporal Interpretation
Also, as with states in the simple past, the intervals introduced by
prog do not necessarily overlap with the interval that happens to be
the local evaluation time. Consider, for example:
(42) John said (today) that Mary was writing the paper yesterday.(43) John said today that he would tell his friend tomorrow that we
were having a good time yesterday.
Progressive thus allows backward shifted readings and, as other
tenses, the available reading (the assigned location and duration)depends on contextual information:
(44) John was watching television when Mary came in. However, she
said he was sleeping on the sofa.
The interval of sleeping overlaps with Mary’s coming in, and not
Mary’s saying time. This is because the when-clause is the RT for the
evaluation of following discourse. Thus, as with simple past, principlesof discourse interpretation and general knowledge determine the mostlikely reading for past progressive.
3.1.3 Independent and embedded past tenses Given that past statives in
embedded constructions may overlap with the local evaluation time,the superinterval property also predicts that an overlapping readingshould be possible when the local evaluation time is the ST. Althoughasentence such as (45) below, for example, may still be true at the
ST, it does not have the same obvious overlapping interpretation aspast under past. This is because the overlapping reading is blocked bystandard Gricean principles applied to grammatical elements.
(45) Karl was German.
If the speaker wanted to convey that the state of being German obtains
at the ST, he would have used present tense, since the superintervalof states would guarantee its application to present and at least somepast times (it would be more informative). If the speaker does not usepresent, the implication is that he/she does not know whether this pieceof information obtains. Thus, the hearer is less inclined to infer that thestate obtains at the ST, although this may in fact be true.
This implicature is clearly operative in (45). In absence of a discourse
RT,the implication is that Karl is dead or changed his citizenship, in
which case the sentence is a habitual state wholly located in the past.

Silvia Gennari 61
The state does not overlap with the ST. Thus, independent past statives
imply a superinterval in a way consistent with my proposal, but itsduration may not overlap the ST due to a Gricean quantity maxim,hence the contrast with embedded past.
3.2 The interpretation of indexical tenses
Recall from the introduction that future and present tenses’ interpre-
tations cannot be uniformly accounted for in embedded sentences iftheir meanings are considered either simply indexical or evaluationtime sensitive. The problem of these tenses is that most often they areinterpreted as indexical, but, in some embedded contexts, they behaveas evaluation time sensitive. Embedded present tense, for example, mayreceive either a double access reading or a purely simultaneous futureinterpretation. Only the former interpret ation requires reference to the
ST. I take this to mean, contrary to the approaches discussed earlier,that the future and the present tense have complex definitions thatinvolve more than a reference to the ST. Specifically, I will argue onthe basis of the distribution of all possible readings, that indexical tensesare interpreted relative to both the ST and the local evaluation time.These tenses are thus both indexical and evaluation time sensitive. Theyare indexical because they always involve a reference to the ST, andevaluation time sensitive because their interpretations depend on thelocal evaluation time in a way that is constrained by their indexicality.
3.2.1 Present tense Consider again the interpretations available for
embedded present:
(46) Hillary is smart.
(3) The president believed that his party is furious.(4) The press will believe that the president is out of town.
Note that a present complement can be interpreted as uniquely
overlapping the local evaluation time only if this time is future (orpresent). In a case such as (4), the reference to the ST is neithernecessary nor implied. In contrast, if the evaluation time is past, thedouble access reading is the only possible one, as in (3), and thereference to the ST is required. This suggests that the denotation ofpresent tense can either be a future interval or an interval overlappingwith the ST, i.e. the tense in fact denotes any non-past interval. Notealso that the interval denoted in every case overlaps with whicheverhappens to be the evaluation time (either the ST or the attitude

62Tense Meanings and Temporal Interpretation
time). This indicates that the tense is evaluation time sensitive. Taking
these properties together, the meaning of present tense thus requiresthe ET of the modified proposition to (a) overlap with the localevaluation time and (b) not to be wholly located before the ST, i.e.either be later or overlap with it. Formally, the definition of present isλQλi[∃i
/prime[i/prime◦i&¬(i/prime<st)&Q(i/prime)]],where ◦means overlap with .The
tense is thus both evaluation time sensitive and indexical.
Tosee how the proposed meaning predicts the correct interpreta-
tions in both embedded and independent occurrences, consider first thecase of present embedded under future or present:
(47) John thinks that Mary is smart.
λi0[∃i[i◦i0&¬(i<st)&think/prime(∧λi1∃i/prime[i/prime◦i1&¬(i/prime<st)
&besmart/prime(m)(i/prime)])(j)(i)]]st=
∃i[i◦st&think/prime(i,j,∧λi1∃i/prime[i/prime◦i1&¬(i/prime<st)&besmart/prime(i/prime,m)])]
(48) John will say that Mary is his wife.
∃i1[i1>st&say/prime(i1,j,∧λi0∃i2[i2◦i0&¬(i2<st)&bewife/prime(i2,m)])]
In the case of (47), which also exemplifies independent sentences
evaluated relative to the ST, the sentence is true iff there is an interval i
overlapping the ST at which John thinks that there is another interval i/prime
overlapping his now and no earlier than the ST, at which Mary is smart.
The embedded present simply overlaps with the local evaluation time,which in turn, overlaps with the ST. The restriction that the embeddedET be non-past is satisfied. Note that in (47), the outermost temporalquantifier need not specify the condition ¬(i<st),thus reproducing
the traditional definition of present tense. This is because, once thetemporal abstract resulting from the tense composition is evaluatedrelative to the ST, the statement i◦stentails that iis not wholly located
before the ST, and so it can be omitted for simplicity, although there isno harm leaving it in.
For the case of (48), the sentence could be true in the two following
situations: when John says at a future time that Mary is John’s wife atsome interval overlapping with John’s saying time, and when John saysat a future time that Mary currently is and has been John’s wife. In theformer case, the ET of the present complement i
2is a future interval
overlapping with the future local evaluation time. In the latter case, thisinterval is extended enough to overlap with the ST and the future localevaluation time. Whether the embedded interval overlaps with the ST(i.e. the most likely size of this interval) will be determined by contextand pragmatic considerations. This is clear in the following examples:

Silvia Gennari 63
(49) (When John gets home), John will think that Mary is talking on
the phone/is in the kitchen.
(50) John will announce tonight that Mary is writing a new book.
The ET of the embedded sentence in (49) most likely surrounds
the future evaluation time. This is because the superinterval introducedby the progressive tends to be interpreted as surrounding its evaluationtime
15and events such as talking on the phone or states such as being in
the kitchen do not tend to go on for long periods. Unless the distancebetween the ST and the future time is close enough, the overlap withthe ST is not pragmatically plausible. This contrasts with (50) and (48),in which the overlap with the ST is pragmatically available.
Consider now the case of present under past sentences such as (7):
(7) Bill believed that Hillary is pregnant.
∃i[i<st&believe/prime(i,b,∧λi0∃i/prime[i/prime◦i0&¬(i/prime<st)&bepregnant/prime(i/prime,h)])]
This says that the sentence is true iff there is an interval iprior to the ST
at which for all Bill’s belief worlds, there is an interval i/primesuch that (a) it
overlaps Bill’s believing time, (b) it is not an interval before the ST, and(c) Hillary is pregnant at i
/prime.Note that by definition of the before andafter
relations, ¬(i/prime<st)means that i/primeis not wholly located before the ST.
Thus, an interval i/primethat overlaps with the past believing interval and is
not wholly located before the ST, necessarily requires that i/primeoverlaps
with both the past believing interval iand the ST. Thus, the proposed
definition of present tense accounts for all temporal readings availablein both embedded and independent occurrences.
Several reasons suggest that the proposed definition of the present
tense is a sensible one. First, it replicates the effects of the traditionaldefinition (overlap with the ST), while expanding its explanatoryadequacy to embedded domains in which the evaluation time is notthe ST. Second, the definition follows from the minimal assumption ofuniform interpretations in all contexts and from the actual distributionof possible readings. This contrasts with Ogihara’s (1996) proposal,where each reading is explained by a different mechanism. The doubleaccess reading is explained by the temporal de remechanism, while the
future reading is obtained via the deletion rule (section 1.2).
In addition to simplicity and economy considerations, the meaning
agrees in spirit with several proposals in which the temporal perspectiveof present is considered to be non-past (e.g. Kamp & Reyle 1993;Abusch 1997). In particular, Abusch (1988) proposes a definition of
15The evaluation time is in the middle of the superinterval, i.e. it is not final or initial in it (see
definition of progressive).

64Tense Meanings and Temporal Interpretation
present tense similar to mine, in which the interval denoted overlaps
with both the evaluation time and the ST. The novelty of my definitionis that it imposes an indexical non-past condition that can be satisfiedin different ways, depending on the local evaluation time.
Finally, there are generality considerations. Languages differ in the
way lexical tense meanings are specified out of a set of universalpossibilities such as evaluation-time sensitivity and indexicality. Forexample, English present tense is both evaluation time sensitive andindexical, while Japanese, which lacks a morphological future tense,has a non-past tense that is the non-indexical (evaluation time sensitive)formal counterpart of English present. This is what one would expect ifthe proposed lexical meaning had cross-linguistic validity (see Gennari1999a for details).
3.2.2 The semantics of present tense and the content-report mismatch
In this section, I would like to briefly discuss the problem of thecontent-report mismatch, which has motivated the temporal de re
analysis discussed before. The purpose of this section is not to offeracomprehensive analysis of the problem (for this, I refer the reader
to Gennari 1999b), but to show how one might explain apparentinconsistencies between the account of the double access reading andthe intuition of the content-report mismatch.
Note for example that it follows from the truth conditions of (7)
that Bill believed that Hillary is pregnant during an interval overlappingwith both Bill’s now and the ST. The attributed embedded propositionseems to commit the believer to a belief about an interval overlappingwith a future time, the ST from the perspective of the speaker.
(7) Bill believed that Hillary is pregnant.
However, Bill presumably had a belief about Hillary being pregnant
at a past interval, and not necessarily at an interval extending into thefuture or including the ST (the content-report mismatch). To explainthis puzzle, I have argued that the believer need not be committedto such specific temporal contents if the speaker’s report is viewed as areport of an implicit attitude. Reports of implicit attitudes do not intendto represent the believer’s literal belief (or the believer’s ways of rep-resentation). They report a proposition that follows from the explicitlybelieved content plus other assumptions normally taken for granted.
In (7), for example, Bill did not actually have a belief about an
interval extending into the future, but his belief entailed that the
embedded state was true in the past and would be true in the future,given the speaker’s addition of assumptions normally taken for granted.

Silvia Gennari 65
The speaker attributes to Bill typical assumptions and knowledge of
the world from which the implicit content reported logically follows.Thus, if Bill believed at a time before the ST that Hillary was pregnant,the speaker could infer that Bill believed that she was pregnant andwould be pregnant for a while, given that Bill has rational beliefs andnormal knowledge about pregnancy (e.g. typical assumptions about itsduration). Before uttering the sentence in (7), the speaker goes throughan inference schematically represented as follows:
(54) a. Bill believed that Hillary was pregnant at t.
b. Bill’s belief worlds are coherent.c. Bill believed that Hillary had a normal pregnancy.d. Bill knew that pregnancies typically last for an interval i
including t.
e. Bill knew that iincludes a future time t
/prime(the ST from the
speaker’s perspective).→Bill believed that Hillary is pregnant at iincluding tand
t
/prime.
From the speaker’s perspective, the future time t/primein Bill’s worlds is
the ST. Note that the embedded interval denoted by the present tensein (7) exists in the belief worlds, rather than in the actual world. Theinference that Hillary’s pregnancy obtains at a future time t
/primeholds in
the belief worlds. However, the actual belief may only be about a pastinterval (premise (a)). This is what the speaker would have reportedif the inference was not possible in the current common ground.The speaker’s inference concludes the pregnancy at the ST implicitin Bill’s beliefs, given the attribution of normal assumptions and typicalknowledge.
16
Present under future sentences in their double access reading have
similar characteristics to present under past ones. Consider, for example:
(55) The dean will believe that Bill’s records are not good enough.
When the double access reading obtains, the speaker attributes an
implicit future attitude based on an inference including common senseassumptions. The attributed future attitude is such that, given normalassumptions, it will entail something true about the past of the attitude
16Note that premise (c) of (54) schematically represents other premises also implicitly assumed by
the speaker. For example, Bill did not think at the time of the belief that Hillary was about to givebirth, or Bill did not have any reason to think that Hillary would not have a normal and full termpregnancy. These are part of normal assumptions that the speaker takes for granted in the commonground and thus attributes to the believer. This correctly predicts that if the speaker knew that Billthought that Hillary’s situation was somewhat atypical (for example, that Hillary was sick and couldloose the child), the present under past report in (7) would be infelicitous.

66Tense Meanings and Temporal Interpretation
time (the ST), although the attitude holder may not know this at
the ST. The difference with present under past reports is that whatis entailed by the belief worlds looks backward instead of forward,i.e. once the believer acquires certain knowledge, his/her view of thepast will change. For example, in (55), the dean will believe that Bill’srecords are bad at a future time t.But, since the dean will learn that
Bill’s records are generally bad at an interval iincluding t,and since i
includes t−1(the ST), it follows that the dean will believe that Bill’s
records are bad at an interval including t−1.
The notion of implicit report, traditionally acknowledged in
philosophical literature (e.g. Stalnaker 1999) is further constrained hereby the existence of a pragmatic inference, the premises of which (ifany) should be taken for granted in the common ground. In particular,Ihave proposed (following Stalnaker’s (1999) observations on implicit
beliefs) that an implicit report such as that in (7) is felicitous, if there isan inference attainable in the common ground that allows the speakerto infer the attributed content. This is because, by the very nature ofimplicit attitudes, the speaker cannot assume anyproposition as part
of the belief worlds. Rather, he/she may assume those propositionsthat are normally taken for granted, i.e. those that constitute commonknowledge and default assumptions. Consider, for example,
(56) Russell believed that Frege’s ear lobe was smaller than Big Ben.
where Russell presumably did not have this explicit belief but it follows
from the general pragmatic knowledge attributable to him. A beliefattribution may thus take for granted beliefs that the believer is notaware of and does not represent in a linguistic form. This is the crucialdifference that distinguishes report of implicit attitude from other typesof reports.
17
In sum, this account contrasts with current ones in that it only
17Note incidentally, that the existence of implicit belief reports has important implications for the
treatment of indexical tenses within intensional domains. The literature discussed so far assumes thatall indexical tenses are interpreted de re(Ogihara 1996) and therefore, move outside the intensional
domain. These assumptions follow from their very nature: Indexical tenses denote times in theutterance context, therefore, this cannot be part of the believer’s worlds. However, while indexicaltenses undoubtedly refer to the utterance context, it is not a necessary truth that they are alwaysinterpreted de reand moved outside the intensional domain. In particular, if indexical tenses occur in
implicit reports, which are neither explicit de dicto norde rereports, it follows that they are not always
interpreted de re.Rather, they can be interpreted as denoting some attributed implicit content, and
thus be interpreted in situ. That the speaker uses an indexical reference to report the believed contentis no longer problematic because the report is not intended to represent an explicit belief, i.e. the waythe believer would represent it or the referential expressions he/she would use. The speaker reportsan implicit proposition with his/her referential expressions (perhaps, because of cooperation withthe addressee, see Asher 1986). This proposition must be equivalent to the one implicitly believed(e.g. equivalent to the last line of (54)).

Silvia Gennari 67
requires the standard semantic analysis of belief-reports, while the
speaker is responsible for the use of the present tense via his/herown pragmatic inference. What creates the intuition that the speakermisrepresents the original belief is the inference process the speaker goesthrough in the report, which in most cases, attributes a stronger beliefthan the original content, given the premises added to the belief worlds.Thus, the content-report mismatch arises because reports containingindexical tenses involve an attribution of an implicit attitude. Thechoice of the present tense correlates with implicit reports but suchreports are not restricted to indexical uses.
3.2.3 Future tense As with present tense, the future tense in
embedded sentences is sensitive to the local evaluation time in a waythat is constrained by its indexical reference to the ST. Consider againsome examples from the introduction:
(5) In two days, an official will announce that the president will
apologize (
∗tomorrow).
(6) A journalist said that the president will resign (∗yesterday).
Note that there is a contrast between future embedded under past
and future embedded under future. Sensitivity to the local evaluationtime only appears in the later case, when the evaluation time isalready located in the future. This suggests that future tense requiresan interpretation relative to both the ST and the evaluation time if thistime is later than the ST. Following this observation, I propose that thefuture tense specifies that the ET of the modified proposition is bothlater than the ST and later than the local evaluation time.
18Formally,
the meaning of future is λQλi[∃i/prime[i/prime>i&i/prime>st&Q(i/prime)]].This is a
way to express that the temporal perspective from which Qis seen is
always the ST and possibly the local evaluation time, if this time is laterthan the ST. When the local evaluation time is located before the STor is equal to it, the first conjunct of the definition does not have anyeffect on the temporal interpretation.
Tosee this, consider first the case of future embedded under future:
(57) John will say that Mary will come.
λi
0[∃i1[i1>i0&i1>st&say/prime(∧λi∃i2[i2>i&i2>st
&come/prime(m)(i2)])(j)(i1)]]st=
∃i1[i1>st&say/prime(i1,j,∧λi∃i2[i2>i&i2>st&come/prime(i2,m)])]]]
18Support for this idea can be found in Reichenbach’s (1947) and Kamp & Reyle’s (1993) analyses
of simple future. According to them, simple future can have either the ST or another future time asRTor temporal perspective. If one thinks of this duality of point of reference in terms of evaluation
times, my definition agrees with this observation.

68Tense Meanings and Temporal Interpretation
(57) is true iff there is a time i1later than the ST at which John says
that there is another future time i2later than his now and later than
the ST at which Mary comes. The requirement that the ET of theembedded proposition be later than the local evaluation time capturesthe future interpretation relative to the attitude time and excludes theunattested reading suggested by the adverbial modification in (5). Inaddition, when the ST is the evaluation time, the requirement referringto the evaluation time in the meaning of the tense will turn out to beequivalent to that referring to the ST. This is clear in the case of theoutermost quantifier of (57), once the future temporal abstract is appliedto the ST (i.e. if i
1>i0&i1>standi0=st,theni1>st&i1>st,
which is equivalent to i1>st). This shows that the occurrence of future
tense in independent sentences or when embedded under present is notproblematic. In these cases, the local evaluation time is the ST. In suchcases, the definition of willis somewhat redundant but harmless.
Note incidentally, that the occurrence of a stative sentence in the
complement of (57) would not receive an overlapping reading withthe evaluation time, although the superinterval of states would makethis reading possible. Consider for example, John will think that Mary
will be sad .The overlapping reading does not obtain here because there
is a quantity implicature. Present tense in this environment wouldbe more informative than the use of future because by definition, itentails an overlapping reading with the evaluation time. If the speakerdoes not use present, it implies that the overlapping interpretationdoes not obtain. As with independent past sentences, implicaturestriggered by grammatical elements lead to an interpretation in whichthe superinterval of states is not large enough to overlap the localevaluation time.
Consider now the case of future embedded under past:
(58) John said that Mary will come.
∃i
1[i1<st&say/prime(i1,j,∧λi0∃i2[i2>i0&i2>st&come/prime(i2,m)])]]
(58) is true iff there is an interval i1prior to ST in which John says
that there is another interval i2later than the past attitude interval and
later than the ST in which Mary comes. Note that the only possiblereading is one in which the future time is located both after the ST andafter the past attitude time. If the future interval was located after thepast tense but not after the ST, the requirement that this interval be laterthan the ST would not be satisfied. This definition thus amounts to aninterpretation of willas indexical rather than evaluation time sensitive.
Thus, our definition of the meaning of the future accounts for all thepossible readings in both independent and embedded sentences.

Silvia Gennari 69
4CONCLUSIONS
This paper has shown that the problematic temporal readings of
English complement clauses can be accounted for on the basis ofuniform definitions of tense meanings. While English simple pasttense is evaluation time sensitive, present and future tenses areboth evaluation time sensitive and indexical. These tense meaningssystematically interact with aktionsart properties, which further specifythe temporal readings expected on the basis of the tense meaningsalone. Overlapping readings are available with statives because thisaktionsart class typically implies the truth of the sentence at asuperinterval. In a subordinate construction, the superinterval associatedwith the embedded state is able to overlap with the time of theattitude, although the size and location of the superinterval ultimatelydepend on discourse considerations. In contrast, eventive sentencesnormally trigger sequential readings in both embedded and discoursecontexts because they typically imply that they are maximally containedwithin their ETs. Thus, the interaction of appropriate tense meaningswith aktionsart and discourse inferences predicts the intended readingsfor both independent and embedded sentences without any specialsyntactic mechanism.
Acknowledgements
Ithank my mentors Pauline Jacobson and Craige Roberts for innumerable suggestions
and wonderful support. Also, I thank Professor Arnim von Stechow for helpfulcomments on a previous version of the manuscript, and Mark Steedman for insightfulsuggestions that improved this paper.
SILVIA P. GENNARI
1401 Marie Mount Hall
University of MarylandCollege Park MD 20742USAe-mail: sgen@wam.umd.eduReceived: 02.04.01
Final version received: 07.08.02
REFERENCES
Abusch, D. (1988) ‘Sequence of tense,
intensionality and scope’. In Proceedings
of the Seventh West Coast Conference inFormal Linguistics .CSLI Publications.
Stanford, CA, 1–14.Abusch, D. (1991) ‘The present under
past as de de interpretation’. In Pro-
ceedings of the Tenth West Coast Confer-ence on Formal Linguistics .CSLI Publi-
cations. Stanford, 1–11.

70Tense Meanings and Temporal Interpretation
Abusch, D. (1997) ‘Sequence of tense
and temporal de re’. Linguistics and
Philosophy 20:1–50.
Asher, N. (1986) ‘Belief in discourse
representation theory’. Journal of Philo-
sophical Logic 15:127–189.
Boyd, J. & Thorne, J. (1969) ‘The
semantics of modal verbs’. Journal of
Linguistics 5:57–74.
Carlson, G. & Pelletier, F. (1995) The
Generic Book .University of Chicago
Press. Chicago, IL.
Chierchia, G. (1995) ‘Individual-level
predicates as inherent generics’. In G.Carlson & F. Pelletier (eds), The
Generic Book .University of Chicago
Press. Chicago, IL, 176–223.
Cipria, A. & Roberts, C. (2001) ‘Span-
ish ‘imperfecto’ and ‘pret ´erito’: mul-
tiple interpretations and ambiguoustruth conditions’. Natural Language
Semantics 8(4):297–347.
Condoravdi, C. (2001) ‘Will’ Talk
presented at Workshop on Temporal-
ity and Discourse Context .University of
Dundee. Scotland.
Cresswell, J. & Stechow, A. Von (1982)
‘De re belief generalized’. Linguistics
and Philosophy 5:503–535.
Dowty, D. (1979) WordMeaning and
Montague Grammar .Reidel. Dordrecht.
Dowty, D. (1982) ‘Tenses, time adverbs
and compositional semantic theory’.Linguistics and Philosophy 5:23–58.
Dowty, D. (1986) ‘The effects of aspec-
tual class on the temporal structure ofdiscourse: Semantics or pragmatics?’Linguistics and Philosophy 9:37–61.
Enc¸,M. (1987) ‘Anchoring conditions
for tense’. Linguistic Inquiry 18:633–
657.
Gennari, S. (1997) ‘Modal and temporal
aspects of the meaning of would ’. In
B. Bruening (ed.), Proceedings of the
Eighth Student Conference in Linguistics .
Vol. 31. MIT Working Papers in
Linguistics. Cambridge, MA, 185–204.
Gennari, S. (1999a) Tense Meaning andTemporal Interpretation .Ph.D. disserta-
tion, Department of Cognitive andLinguistic Sciences, Brown University,Providence, RI.
Gennari, S. (1999b) ‘Embedded present
tense and attitude reports’. In T.Matthews & D. Strolovitch (eds),Semantics and Linguistic Theory 9 .CLC
Publications. Cornell University, NY.
Gennari, S. (2000) ‘Semantics and prag-
matics of future tenses in Spanish’. InH. Campos, E. Herburger, A. Morales-Front, & T. J. Walsh (eds), Hispanic
Linguistics at the Turn of the Millen-nium, Papers from the Third HispanicLinguistics Symposium .Cascadilla Press.
Somerville, MA, 264–281.
Hinrichs, E. (1986) ‘Temporal anaphora
in discourse of English’. Linguistics and
Philosophy 9:63–83.
Hornstein, N. (1990) As Time Goes
By: Tense and Universal Grammar .MIT
Press. Cambridge, MA.
Kamp, H. & Reyle, U. (1993) From
Discourse to Logic: Introduction to ModelTheoretic Semantics of Natural Language .
Formal Logic and Discourse Repre-sentation Theory. Reidel. Dordrecht.
Kamp, H. & Rohrer, C. (1983) ‘Tense
in text’. In R. Bauerle, C. Schwarze,&A.von Stechow (eds), Use and
Interpretation of Language .deGruyter.
Berlin, 250–269.
Katz, G. (1994) Stativity, Genericity and
Temporal Reference .Ph.D. Dissertation,
University of Rochester Rochester,NY.
Klein, W. (1994) Time in Language .
Routledge. London.
Kratzer, A. (1981) ‘The notional category
of modality’. In H. J. Eikmeyer &H. Rieser (eds), Words, Worlds and
Contexts .deGruyter. Berlin, 38–74.
Kratzer, A. (1989) ‘Stage-Level and
Individual-Level Predicates’. In Paper
on Quantification .GLSA. University of
Massachusetts, MA, 147–222.
Ladusaw, W. (1977) ‘Some problems

Silvia Gennari 71
with Tense in PTQ’. Texas Linguistic
Forum 6:89–102.
Lascarides, A. & Asher, N.
(1993) ‘Temporal Interpretation,Discourse Relations and CommonSense Entailments’. Linguistics and
Philosophy 16:437–493.
Lewis, D. (1979) ‘Attitudes de dicto
and de se’. Philosophical Review
LXXXVIII(4) :513–543.
Moens, M. & Steedman, M. (1988)
‘Temporal ontology and temporalreference’. Computational Linguistics
14:15–28.
Nerbonne, J. (1986) ‘Reference time
and time in narration’. Linguistics and
Philosophy 9:84–95.
Ogihara, T. (1996) Tense, Attitudes and
Scope.Reidel. Dordrecht.
Palmer, F. (1969) ‘Will, can, may and
must’. Filologia Moderna 31–32 :203–
212.
Partee, B. (1984) ‘Temporal and nominal
anaphora’. Linguistics and Philosophy
7:243–286.
Parsons, T. (1990) Events in the Semantics
of English .A Study in Subatomic
Semantics. MIT Press. Cambridge,MA.
Reichenbach, H. (1947) Elements of
Symbolic Logic .Macmillan. New York.
Rosch, E. & Mervis, C. B. (1975) ‘Family
resemblance: Studies in the internalstructure of categories’. Cognitivie Psy-chology 7:573–605.
Smith, C. (1991) The Parameter of Aspect .
Reidel. Dordrecht.
Schlenker, P. (1999) Propositional Atti-
tudes and Indexicality: A Cross-CategorialApproach .Ph.D. Dissertation, Depart-
ment of Linguistics and Philosophy,MIT, Boston, MA.
Stalnaker, R. (1999) Context and Content .
Oxford University Press. Oxford, UK.
Stechow, A. von (1995) ‘On the proper
treatment of tense’. In M. Simons & T.Galloway (eds), Proceedings of SALT 5 .
CLC Publications. Cornell UniversityPress. Ithaca, NY.
Steedman, M. (1997) Handbook of Logic
and Language .Temporality. In J. van
Benthem & A. Ter Meulen (eds),Elsevier. North Holland, 895–935.
Taylor, B. (1977) ‘Tense and continuity’.
Linguistics and Philosophy 1:199–220.
TerMeulen, A. (1995) Representing Time
in Natural Language The Dynamic Inter-pretation of Tense and Aspect .MIT Press.
Cambridge, MA.
Vendler, Z. (1967) Linguistics and Philos-
ophy.Cornell University Press. Ithaca,
NY.
Verkuyl, H. (1993) ATheory of Aspec-
tuality .Cambridge University Press.
Cambridge, UK.
Wekker, H. (1976) The Expression of
Future Time in Contemporary English .
North-Holland. Netherlands.