Chapter 1
On the Temporal Composition
of Infinitives
Dorit Abusch
1.1 Introduction
Nontensed intensional complements in English often have a future interpretation.
(1a–g) illustrate this, respectively, for a control to-infinitive, raising-to-object to-
infinitive, raising-to-passive-subject to-infinitive, passive participle small clause, inten-
sional NP, raising to-infinitive, and accusative-ing argument of a predicative noun
phrase.
(1) a. Solange hopes to be in Stockholm next week.
b. Guido expects Solange to be in Stockholm next week.
c. Barak is predicted to win.
d. I want th is dispute resolved.
e. Guido needs a bike.
f. Guido is likely to flunk out of college.
g. Guido flunking out of college is a certainty.
This chapter investigates the logical form (compositional structure ) of to-
complements. Much of what I say, though, seems to apply to the other types of
futurate complements too. One issue I will address is the interaction between futurate
verbs and embedded tenses. In (2), the verb intends a¤ec ts not only the temporal
location of the answering events corresponding to the untensed head answer of its
complement, but also the location of the sending events corresponding to the present
tense verb sends in the relative clause.
(2) Solange intends to answer every e-mail Guido sends.
A related issue I will address is whether to-infinitive complements have a proposi-
tional structure similar to that of tensed complements. Does the complement in (1b)
have the same denotation as the complement in (3)? If so, is the compositional
structure of the two complements isomorphic, so that (1b) includes an element
semantically similar to will?
(3) Guido expects that Solange will be in Stockholm next week.
The chapter is organized as follows. Section 1.2 considers the interpretations avail-
able for to-complements. Section 1.3 reviews a semantic-interpretive framework and
an analysis of the English future auxiliary will developed in Abusch 1998, which
involves temporal substitution. Section 1.4 shows that arguments that motivate tem-
poral substitution with will also apply to future-oriented infinitives. Section 1.5 gives
an argument that meaning elements identified in section 1.4 are syntactically overt.
1.2 Possible Readings of To-Complements
1.2.1 Diagnostics for B-Verbs
In this section, I consider the interpretation of to-infinitive complements in one syn-
tactic class, with the aim of characterizing the available readings and establishing
diagnostics for them. In (4), sixteen verbs taking a raising-to-passive-subject comple-
ment structure are listed in their past participle forms. The examples in (5) illustrate
the raising-to-passive-subject complement structure.
(4) I II
asserted anticipated
believed expected
claimed forecast
confessed intended
known meant
reported planned
said predicted
thought projected
(5) a. Barak is believed to be in the lead.
b. Barak is forecast to win by about 8 percentage points.
Criteria such as the possibility of an expletive subject indicate that these are raising
structures, with the subject filling no argument position of the higher verb (e.g.,
believed or predicted). (6) gives examples with an expletive there subject.
(6) a. There is believed to be a linguistics department in Geneva.
b. There is predicted to be a volcanic eruption in Oregon next year.
This and other syntactic diagnostics for raising structures are discussed in Postal
1974. Turning to s emantics, the verbs in column I of (4) allow only for simul taneous
readings of the complement infinitive. This is reflected in three propert ies:
1. incompatibility with past and future frame adverbs modifying the top-level predi-
cate in the complement,
28 Dorit Abusch
2. incompatibility with nonstative complements, and
3. equivalence with present tense or sequence-of-tense past tense tensed complements.
Property 1 is illustrated in (7).
(7) a. *Guido is believed to be at Monique’s place last night.
b. */OK Guido is believed to be at Monique’s place tomorrow night.
c. Guido is believed to be at Monique’s place (now).
As used in (7a), last night is a past-denoting temporal frame adverb in the sense that
last night precedes the attitude time, which in our example is the believing time. In a
similar sense, tomorrow night in (7b) is future denoting. Last night in (7a) modifies
the top-level predicate be at Monique’s place in the complement, and so the star in
(7a) illustrates property 1. Notice that (7a) contrasts with the tensed complement (8a)
and with the version with an infinitival temporal have in (8b).
(8) a. It is believed that Guido was at Monique’s place last night.
b. Guido is believed to have been at Monique’s place last night.
In (8b), last night does not modify the top-level predicate have, but instead the
embedded predicate be at Monique’s place. Therefore, (8b) does not violate prop-
erty 1.
In (7b) with the future adverb, there is a twist: the example is good on a scheduling
interpretation of the complement. This interpretation also shows up with present
tense nonstative verbs.
(9) Guido is at Monique’s place tomorrow night.
In (9), it is understood that Guido has a schedule, perhaps a regular one, for where
he is to stay. The sch eduling interpretation can be controlled for by adjusting the
content. While (10a) is fine as a description of the scheduled start of a party, (10b) is
a bit odd as break up suggests an unscheduled end. The same distinction carries over
to the to-infinitives in (10c, d). This supports the claim that (10c) involves a distinct
scheduling interpretation.
(10) a. The party starts at 10 p.m.
b. ?#The party breaks up at 4 a.m.
c. The party is thought to start at 10 p.m.
d. ?#The party is thought to break up around 4 a.m.
The examples in (11) and (12) make the same point. Many electronic components
have a fixed probability of failing over any year starting at a time when the compo-
nent has not failed yet. Such components will fail sooner or later, but one does not
know when it will happen. (11a) is a good description of this situation, but (11b) is
odd, presumably because of an incompatibility between the scheduling modality and
Temporal Composition of Infinitives 29
the indeterminateness of the time of failure. Keeping the scenario the same, (12c) is
odd in the same way as (11b), because the scenario makes the scheduling interpreta -
tion implausible.
(11) a. The component will sooner or later fail.
b. #The component sooner or later fails.
(12) a. It is believed that the component will sooner or later fail.
b. #It is believed that the component sooner or later fail s.
c. #The component is believed to sooner or later fail.
With the scheduling interpretation eliminated, the hash marks in (10d) and (12c)
illustrate the part of property 1 having to do with future adverbs: column I verbs
with future time adverbs modifying the to-infinitive do not have an ordinary, non-
scheduling interpretation.1 Column I verbs like believe and thought and column II
verbs like forecast and predict will be referred to as B-verbs and F-verbs, respectively.
In contrast with the B-verbs in (4), the F-verbs are compatible with future adverbs
in the complement.2
(13) a. Bibi is predicted to be in the lead next week.
b. A solar eclipse is forecast to occur in Wu
¨
rttemberg in August 1999.
c. The meeting is meant to end at 3 p.m.
(The utterance time for example (13b) is in 1999, before August.)
Property 2 refers to the following paradigm:
(14) a. */gen Guido is believed to visit Stockholm.
b. */gen Guido is claimed to visit Stockholm.
c. */gen Guido is reported to visit Stockholm.
d. */gen Guido is said to visit Stockholm.
(15) a. Guido is expected to visit Stockholm.
b. Guido is meant to visit Stockholm.
c. Guido is predicted to visit Stockholm.
d. Guido is projected to visit Stockholm.
(16) a. Guido is believed to be in Stockholm.
b. Guido is claimed to be in Stockholm.
c. Guido is reported to be in Stockholm.
d. Guido is said to be in Stockholm.
(17) a. Guido is expected to be in Stockholm.
b. Guido is meant to be in Stockholm.
c. Guido is predicted to be in Stockholm.
d. Guido is projected to be in Stockholm.
30 Dorit Abusch
The predicate visit Stockholm is nonstative, while the predicate be in Stockholm is
stative. This is evidenced by the fact th at (18) has no episodic interpretation—it has
only the generic/habitual interpretation ‘Guido has the habit or practice of visiting
Stockholm’. (19) does have an episodic interp retation, which describes a simple fact
about location.
(18) */gen Guido visits Stockholm.
(19) Guido is in Stockholm.
In (14) and (16), the contrast between (18) and (19) is duplicated in the complement
infinitive of B-verbs. In the sentences in (14), the complements have only generic
interpretations. We can attribute this to B-verbs allowing only stative complements.3
I assume the generic reading involves a top-level covert generic operator, which
makes the complem ent stative (as evidenced by the possibility of the generic reading
of (18)). It is easy to control for the generic interpretation by consi dering meaning.
When we put this reading aside, the stars in (14) illustrate the stativity restriction 2.
The examples in (15) of F-verbs with nonstative complements have episodic future
interpretations, generic simultaneous interpretations, and also generic future inter-
pretations. They have no episodic simultaneous interpretations. The stativity restric-
tion is in fact a general restriction on simultaneous readings, which is also observed
with tensed complements.
(20) a. */gen It is thought that Solange sleeps.
b. It is thought that Solange is sleeping.
Because sleep is nonstative, (20a) has no epis odic interpretation approximately
equivalent to (20b). Since all simultaneous readings are stative, we can attribute the
stativity restriction on B-verbs (i.e., the restriction that B-verbs’ complements are
always stative) to B-verbs having only simultaneous readings.
Property 3 refers to the equivalence of pairs such as these:
(21) a. Guido was thought to be in Stockholm.
b. It was thought that Guido was in Stockholm.
(22) a. Guido is claimed to be in Stockholm.
b. It is claimed that Guido is in Stockholm.
This diagnostic can be seen to derive from the fact (or assumption) that the present
tense complement (22b) and the sequence-of-tense past tense complement (21b) have
simultaneous interpretations. The test cannot be applied totally mechanically, be-
cause the tensed complement may have other interpretations. When an antecedent is
set up for the embedded past tense verb as in (23), a reading is possible where even-
tualities corresponding to the embedded clause (in this case the finishing) precede the
attitude time.
Temporal Composition of Infinitives 31
(23) As demonstrated in the present study, Nabokov was in Flims when he finished
Laughter in the Dark . In a talk given last year, Prof. Schmetterling incorrectly
claimed that he was in Montreux.
Also, with a small set of verbs, a present tense complement has a future interpretation.
(24) a. Guido hopes Monique wins.
b. I pray that the snow stops soon.
c. ?I predict Barak wins by at least 10 percentage points.
So actually, there is something circular about diagnostic 3, on a weakened under-
standing that allows the exceptions above. It in e¤ect says that the infinitival
complements of B-verbs are equivalent to tensed complements with simul taneous
interpretations.
Summing up, properties 1–3 distinguish B-verbs fro m F-verbs. Especially in the
case of property 1, the connection with simultaneous interpretations should be clear.
1.2.2 Past Readings
We have seen that F-verbs with to-infinitive complements are compatible with fu-
ture frame adverbs. As illustrated in (25), such F-verbs are incompatible with past
adverbs.
(25) *Guido is predicted (by almost everyo ne) to spend the night of last Friday’s
party at Monique’s place.
Curiously, the intended reading can be expressed by F-verbs with past tense
complements.
(26) a. I predict that Guido spent the night of last Friday’s party at Monique’s
place.
b. It is predicted (by almost everyone) that Guido spent the night of last
Friday’s party at Monique’s place.
Note that there is a special pragmatics for these examples: it is suggested that it is not
known at the predicting time where Guido spent that night.
The same data are observed with other F-verbs: anticipated, forecast, planned, and
projected.4
(27) a. It is anticipated that a meteor impact took place yesterday afternoon in a
remote part of Quebec. Scientists have not arrived at the scene yet.
b. *A meteor impact is anticipated to take place yesterday afternoon in a
remote part of Quebec.
Past readings are also impossible for B-verbs with to-infinitive complements.
(28) *Guido is thought to be at Monique’s place last night.
32 Dorit Abusch
This perhaps suggests that the compositional semantics of infinitives should exclude
past readings.
The examples in (26) and (27a) indicate that the obvious element of futurity in pre-
dict has nothing immediately to do with the temporal location of the event described
by the main verb in the complement. Note that (26a) is roughly paraphrasable as
follows:
(29) I say that the proposition that Guido spent the night of last Friday’s party at
Monique’s place will turn out true.
This goes alo ng with the implicature that at the predicting time, it is not known
whether this proposition is true or not. The curious possibility such examples suggest
is that in (30), futurity is represented twice: once by will in the complement, and once
internal to the lexical meaning of predict. This results in the paraphrase (31), with
two occurrences of will.
(30) It is predicted that Barak will win.
(31) I say that the proposition that Barak will win will turn out true.
It is conceivable that predict with an infinitival complement has a similar composi-
tional structure. This would be the case if the to-complement contained a discrete
element FUT whose compositional role is similar to that of will. A past reading for
to-infinitive complements of predict wou ld be impossible for the same reason that
(30) has no past reading.
1.2.3 Simultaneous Readings of F-Verbs
We have seen that infinitival complements of F-verbs have no past interpretations.
However, contrary to the impression that the infinitival complements of these verbs
are uniformly futurate, simultaneous readings are possible.
Consider the following scenario. A petition for a ballot initiative is being circu-
lated. A lot of signatures have been collected, but not yet summed up. In (32), what is
at issue is how many signatures have already been obtained, and in this sense the com-
plement has a simultaneous interpretation. In view of the pragmatics mentioned in
the previous subsection and the fact that the present number of signatures is unknown,
the use of projected in this context should make perfect sense. As indeed it does.
(32) The petition is projected to have over 20,000 signatures now.
Now consider (33).
(33) Monique is predicted to already be pregnant.
This is another example of an F-verb with a simultaneous reading. Monique is trying
to get pregnant with the new technology. Her doctor is confident of the e‰cacy of his
Temporal Composition of Infinitives 33
treatments and is pretty sure she is pregnant now. They will not find out definitely
until tests are possible in a few days. Again, what is at issue is her being pregnant
now, and because of the epistemic situation, a use of predicted with a simultaneous
complement is pragmatically licensed.
Raising adjectives such as likely and certain are also compatible with both simul-
taneous and future scenarios.
(34) a. Monique is likely/certain to be in Stockholm now.
b. Monique is likely/certain to be in Stockholm next weekend.
c. Monique is likely/certain to win.
Note that with the event verb win in (34c), the aspectual restriction on simultaneous
readings is observed, and only a future interpretation is possible.
A comprehensive analysis of verb classes is beyond the scope of this chapter.
However, it is relevant to ask whether there are any exclusively future-oriented
infinitive-embedding verbs—in other words, ones that exclude a simultaneous inter-
pretation. This is the case with a control use of promise. While promised with a tensed
complement in (35a) has a simultaneous reading, the version with a to-infinitive in
(35b) does not.5
(35) a. In her phone call to Guido, Monique promised that she was in the o‰ce,
not at Paul’s place.
b. In her phone call to Guido, Monique promised to be in the o‰ce, not at
Paul’s place.
It is relevant to ask whether promise in (35a) and ordinary examples of promise with
an infinitival complement like (36) involve the same root word sense.
(36) Monique promised to be home before midnight.
If they do, then there has to be something structural about (35b) (such as an addi-
tional futurity morpheme being present in the complem ent) that excludes the simul-
taneous reading. Notice that future-oriented to-complements can be conjoined with
simultaneous that-complements.
(37) In her phone call to Guido, Monique promised to be home before midnight,
and that she was in her o‰ce.
This supp orts the hypothesis of a single word sense for promise in (35a) and (36).
However, promise with a to-complement seems to have a narrower range of meaning
than with the corresponding that-complement.
(38) a. Monique promised that she would eventually fall asleep tonight.
b. Monique promised to eventually fall asleep tonight.
There is a way of reading (38a) that does not imply the same kind of lasting com-
mitment that (38b) implies. Assume Monique realizes she is tired and, knowing
34 Dorit Abusch
herself well, is sure she will fall asleep tonight. (38a) can describe her making an em-
phatic statement, whose truth she is committed to when she makes the statement,
without making a promise in the ordinary sense. Suppose that after all, Monique
does not fall asleep. If she made a promise, she would have to do something to fall
asleep, such as take a sleeping pill. If she merely made a statement, she need not be
committed in the same way to making it turn out true. The interesting point now is
that (38b) with the to-infinitive can only describe a speech act of making a prom ise.
So, promise with a to-infinitive has a narrower range of meaning. This might suggest
a lexical ambiguity in the tensed version of promise between a commitment reading
and a s tatement reading.
The argument can be clarified by substituting make a promise, which seems not
to be ambiguous in the same way. Intuitively, (39a) is not ambiguous in its force,
and (39a) is equivalent to (39b).
(39) a. Monique made a promise that she would fall asleep tonight.
b. Monique made a promise to fall asleep tonight.
The crucial contrast is the one in (40). Example (40a) with a tensed complement has
a simultaneous reading, while example (40b) with a to-complement does not.
(40) a. In her phone call to Guido, Monique made a promise that she was in the
o‰ce, not at Paul’s place.
b. In her phone call to Guido, Monique made a promise to be in the o‰ce,
not at Paul’s place.
It is perhaps puzzling what kind of promise (40a) on a simultaneous reading describes.
It is clear, though, that (40b) with the to-infinitive cannot be read as equivalent to (40a).
The conclusion is that with some verbs, to-infinitives have strictly future inter-
pretations. Another verb with this property is decide, where contrasts similar to the
one in (35) are observed.
(41) a. Sitting on the train and looking at the landscape, Monique decided that
she was in France.
b. #Sitting on the train and looking at the landscape, Monique decided to be
in France.
(42) a. Monique decided that she would not see Paul again.
b. Monique decided not to see Paul again.
(41a) with the tensed complement has a simultaneous reading that (41b) lacks, even if
an adverb like now or already is inserted. As with promise , correlated with or in ad-
dition to the temporal di¤erence there is a di¤erence in the kind of act that can be
described. (41a) describes a mental act of drawing a conclusion: observing the land-
scape, Monique concludes that she is in France. Example (42a), where decide embeds
would, can describe an act either of drawing a conclusion or of entering a state of
Temporal Composition of Infinitives 35
intention. In the first interpretation, Monique might want to see Paul and try to see
him, but know that it is not going to happen. If the second interpretation entailing
intention is true, then Monique would not try to see Paul. Example (42b) with the to-
infinitive can describe only a mental act of entering a state of intention.
Returning to F-predicates that are consistent with both simultaneous and future
scenarios, we may ask whether this is a matter of nonspecificity or ambiguity. The
following example mixes simultaneous and future interpretations:
(43) Monique is likely to be in Stockholm both now and next Saturday.
This suggests that likely to has a single representation that is com patible with both
simultaneous and future scenarios. This will be my working hypothesis. A possible
problem is that it is hard or impossible to read the following example as being non-
committal about whether Monique is in Stockholm now or will be in Stockholm in
the future:
(44) Monique is likely to be in Stockholm.
But the reading becomes possible if one makes it explicit, by inserting either now or
later.
1.2.4 Summary
Three classes of interpretations for verbs taking to -infinitive complements have been
identified:
1. B-verbs such as believed that permit only simultaneous interpretations of their
to-infinitive complements,
2. futurate verbs like predicted where the to-infinitive complement is interpreted in a
manner consistent with both simultaneous and future scenarios, and
3. futurate verbs like promised where the to-infinitive has only a future interpretation.
Representations for these readings will be discussed later. We first turn to an inde-
pendent line of evidence on the logical form of infinitives having to do with inter-
pretive interactions with embedded tenses.
1.3 Temporal Substitution
1.3.1 Interpretive Framework
This section and the next use a grammatical framework with the following properties:
0
Meaning is represented explicitly at LF, and semantic composition is limited to
function application, variable binding, and type raising.
0
Temporal aspects of meaning are modeled with semantic objects built using a dis-
tinct type i of time intervals. Tenses denote time intervals or type-raised versions of
them, rather than being proposit ional operators.
36 Dorit Abusch
Applying this framework to (45), we arrive at representation (46), which is an LF
tree annotated with terms naming the denotations of phrases.
(45) John believes Monique loves Solange.
(46)
The verb loves in (46) denotes a function that maps two individuals (type e) and a
time interval (type i) to a proposition. I assume a possible-worlds construction, and
I take propositions to be characteristic functions of sets of worlds, with type label
wt. Under these circumstances, the type label for the verb loves is eeiwt. I use Link’s
(1979) notation for type labels, with right association. The type eeiwt written with
commas and brackets is he, he, hi, hw, tiiii.
In this system, a VP including its subject is a tenseless clause and has type iwt.A
tense fills the interval argument, giving an IP with the proposition type wt. A com -
plement CP has a l-binder in C or Spec,CP of the interval variable and therefore
denotes a property of times (type iwt). In general, the type of a CP depends on the
operator in C; for instance, a relative clause might have type ewt. The type and cate-
gory labels are summarized here.6
VP iwt tenseless clause
IP wt tensed clause
CP iwt complement clause
In order to achieve simpler types, interpretations, and syntactic structures, I am
assuming that the subject is in VP at LF.7
Temporal Composition of Infinitives 37
The interpretation of the term n is straightforward—in the logic, it is a variable. Its
particular role comes from being used in particular ways at the syntax-semantics
interface in the grammars of particular languages, in the general theory of natural
language (Un iversal Grammar), and in the theory of language use (pragmatics). The
most relevant points for the purposes of this chapter are that an intensional clausal
complement starts with a l-binder of n that creates a property of times, and that the
lexical entries for tenses have free occurrences of n. This allow s for the characteriza-
tion of interactions between tense and intensionality.
1.3.2 Semantics of the Future Auxiliary
In Abusch 1998, I proposed an account where the logical form for the future aux-
iliary will/would expresses a substitution for n. The analysis is exemplified in (47)
and (48), with adjustments in notation relative to the earlier paper.8
(47) Mary will answer every e-mail Bill sends next year.
(48)
38 Dorit Abusch
The surface form will corresponds to the four heads Pres: n,[
V
lPltP((t, y))], ln, n.
The subject Mary is represented as reconstructed into the VP headed by answer,
which corresponds to the overt base form VP. The top n is the present tense on will.
[
V
lPltP((t, y))] is a temporal substitution operator that is the core meaning of will.
In the substitution operator, t is a bound variable that corresponds to the tense
argument of will. For a top-level occurrence of will, the e¤ect is to substitute (n, y)
for n. One consequence of this is that eventualities corresponding to the main verb
complement of will are located in the interval (n, y).
Notice that the complement structure in (48) is the same as in the tensed comple-
ment in (46); the CP/IP layer between the two VP nodes is syntactically covert, since
will has a base form complement. The extra structure has semantic motiva tion in
the interaction between futurity and tense: the ln on the complem ent of will binds
two occurrences of n. One corresponds to the present tense on sends. The other is the
temporal argument of the base form verb answer; this temporal argument is treated
as a covert I heading IP. The property of times denoted by CP, together with the top-
level present tense, are arguments of a core meaning lPltP ((t, y)) for will.(t, y)is
an interval stretching from the bound time variable t to positive infinity. In the given
configuration, the denotation for will substitutes a future interval (t, y) for both
occurrences of n in the complement.
Shifted present tenses. The event time for an overt present tense verb in the argu-
ment of will/would falls in the future, rather than at the utterance time. This accounts
for the interpretation of sends in (47), where the possible sending events follow the
utterance time.
Shifted past tenses. Past tenses in the scope of will/would measure back from a time
within the future interval (u, y), rather than from the utterance time.9 On February
1, at the beginning of the spring semester, I say (49). The contemplated turning-in
events are understood as ordered before May 21, rather than February 1. (Though
if some student had turned in a term paper satisfying the required length before the
start of the semester on February 1, he should also get an A according to what I
said.)
(49) On May 21, I will give an automatic A to the first student who turned in a
term paper at least fifteen pages long.
Noncomplementarity. Fixing the location of described events, past and present
tense under the scope of will / would are not in complementary distribution. Consider
the following scenario. In November 1999, the members of a program committee
discuss procedures for reviewing abstracts that are to be submitted in the first two
months of the year 2000. The abstract deadline is February 21, 2000. In this scenario,
the committee members can use either (50a) or (50b).
Temporal Composition of Infinitives 39
(50) a. On March 1, we will discuss the abstracts which are submitted by e-mail.
b. On March 1, we will discuss the abstracts which were submitted by e-mail.
The above-listed three consequences follow from logical forms along the lines of
(48), with l-binding of n and a node introducing the interval (t, y), and from specific
denotations for tenses. In the case of present tense, the analysis is simple: present
tense denotes n. (51) is a semantic derivation corresponding to the LF (48). The
important thing about the result is that the sending events are ordered inside the in-
terval (n, y) D year(1, u), which, given that next year falls in the future interval
(n, y), amounts to just next year. This is a result of the substitution for n performed
by the material corresponding to will.
(51) VP
1
ltbe[e M tb send(e, b, x
3
)]
Tadv lQlt[Q(t D year(1, u))]
VP
2
ltbe[e M t D year(1, u)b send(e , b, x
3
)]
IP
3
be[eM n D year(1, u)b send(e, b, x
3
)]
N
1
lx[e-mail(x) b be[eM n D year(1, u) b send(e, b, x)]]
VP
4
ltevery
lx[e-mail(x) b be[eM n D year(1, u)b send(e, b, x)]],
lxbe[eM tb answer(e, m, x)]
IP
5
every
lx[e-mail(x) b be[eM n D year(1, u)b send(e, b, x)]],
lxbe[e M nb answer(e, m, x)]
IP
6
every
lx[e-mail(x) b be[eM (n, y) D year(1, u)b send(e, b, x)]],
lxbe[e M (n, y)b answer (e, m, x)]
Note that at the node VP
4
, a generalized quantifier is quantified into a property of
times, producing a property of times.
Tree (52) is the LF for the past example (50b). As before, the important point is
that the l-operator on the complement of the subst itution operator associated with
will binds a variable associated with the relative clause tense, which in this case is the
past tense on submitted . Semantically, this has the e¤ect of shifting the time that
the past tense measures back from to March 1. See Abusch 1998 for details about the
semantics of past tense and how it interacts via indexing with other elements of
an LF.
40 Dorit Abusch
(52)
Summing up, an LF for will involving a temporal substitution operator and l-
binding of n results in shifted present and past tenses, and the noncomplementarity
of present and past tenses in future contexts. These properties will be used as diag-
nostics for an LF with a substitution operator.
1.4 Tense Interactions in To-Complements
1.4.1 Data
Future-oriented infinitives interact with tense in the same way as the future auxiliary
will/would.
Shifted present tenses. In (53), the possible submission times follow the utterance
time.
Temporal Composition of Infinitives 41
(53) Mary intends to give an automatic A to every stu dent who submits a term
paper at least fifteen pages long.
Shifted past tenses. In (54), the past tense can be understood as measuring back
from a time following the utterance time (June 1), rather than measuring back from
the utterance time.
(54) On June 1, Mary intends to give an automatic A to every student who
submitted a term paper at least fifteen pages long.
Noncomplementarity. (55) and (56) uttered on New Year’s Eve quantify possible
sending, receiving, and ans wering events that are distributed throughout the year
following the utterance time in an interleaved fashion. In (56), the present tense
receive and the past tense received correspond to the very same eventualities.
(55) This coming year, I intend to immediately answer every e-mail I receive which
was sent by a friend.
(56) I intend to imm ediately answer every e-mail I receive which was received from
a friend.
Other future-oriented infinitives behave in the same way. This is illustrated for pres-
ent tenses in (57) and for past tenses in (58).
(57) a. Solange is predicted to win most of the races she enters.
b. Monique hopes to live in a house her parents buy her.
(58) a. Solange is predicted to win all of the matches she enters which she had
adequate time to prepare for.
b. Monique hopes to tell Paul about something outrageous she did on the trip.
1.4.2 Analysis
These data suggest that the compositional rep resentations of future-oriented infin-
itives include operators that make a substitution for n by means of binding and
function application. As a starting point, it is useful to compare future-oriented
infinitives with tensed complements headed by will. The tensed and infinitival ver-
sions in (59) and (60) are equivalent.10
(59) a. Solange hopes to visit Bjo
¨
rn next week.
b. Solange hopes that she will visit Bjo
¨
rn next week.
(60) a. Barak is predicted to win.
b. It is predicted that Barak will win.
(61) is the representation of the tensed complement in (60b) on the theory reviewed in
section 1.3.
42 Dorit Abusch
(61)
The clausal nodes are numbered from the top; temporal substitution is performed by
the head of VP
3
, which is th e root of will. The idea is to use this tree as a starting
point for the representation of (60a). Given the mechanics of temporal substitution,
something that must be included in the LF of (60a) is the l-operator heading CP
4
,
which binds occurrences of n in the complement, in particular ones coming from
tenses. Above this level, there are some choices to make regarding what structure is
to be maintained. First, notice that the CP
1
/IP
2
layer in (61) is semantically redun-
dant, because of the identity
lnf(n) ¼ f if f contains no free occurrences of n.
In (61), and in fact in all the representations for tensed complements with will
considered above, VP
3
has no free occurrences of n. This suggests the possibility of
dropping the CP
1
/IP
2
layer in the LF for the to-infinitive.
Second, for some embedding predicates, the temporal substituti on operator should
be changed. We saw in section 1.3 that predict and be likely allow simultaneous sce-
narios for the eventualities corresponding to their infinitival complements, in addi-
tion to future ones. A direct way of dealing with this is to use the interval [ t, y)in
the substitution operator in place of the interval (t, y). The interval [t, y) is an in-
terval that includes the left boundary t. We will see below that this gives the right
results for simultaneous scenarios.
Third, there is the question of what layers of structure are overt in the LF of (60a).
In particular, is the substitution operator lPltP([t, y)) a part of the complement,
perhaps as the semanti cs of the morpheme to? Is it ‘‘part’’ of the model-theoretic
interpretation of the embedding verb? Or is it part of a compositionally interpreted
Temporal Composition of Infinitives 43
structured lexical entry for the verb? It is not easy to answer such questions defini-
tively, because they tie in with general questions of how lexical decomposition is
represented. But in section 1.5, I will mention an argument (based on rather intricate
assumptions) that even the highest levels of structure in (61) are overt in the LF of
(60a), in that they are visible to scope interactions.
1.4.3 LFs for Futurate Infinitives
As a working hypothesis, I will leave out the redundant CP
1
/IP
2
layer in the LF of
futurate infinitives, but include an overt substitution operator. This results in the LF
(62) for (60a). (The substitution operator is written as the head of VP. Its actual
syntactic position might be di¤erent, though (e.g., in I).)
(62)
Assuming an overt substitution operator has the advantage that the predicate predict
in (62) can be treated as being the very same predicate as the predict that embeds the
tensed complement in (61). This works because VP
3
in (62) and CP
1
in (61) denote
the same property of times.
If the structure with a temporal substitution operator is postulated, the interaction
of tense with future-oriented infinitives is accounted for in the same way as the par-
allel data with will. For instance, in example (63) the answering and receiving events
can fall in the future, because the n relative to which the present tense on receive and
the past tense on sent are interpreted is an expanded interval [t, y).
(63) Solange is likely to answer every memo she receives this year which was sent
by the dean.
44 Dorit Abusch
This works because in the LF in (64), both tenses are under the scope of the substi-
tution operator.
(64)
The following simpler example illustrates the recursive semantics of the complement:
(65) Barak is predicted to win.
Temporal Composition of Infinitives 45
(66)
Simultaneous readings. The tree in (68) illustrates a simultaneous reading.
(67) Barak is predicted to be in the lead now.
(68)
The compositional semantics is such that the interval denoted by now gets intersected
with the interval contributed by the substitution operator. I assume that now denotes
the utterance time u. The LF (68) is in fact semantically unsatisfactory, since quite
generally, deictic elements must be assumed to be interpreted de re. An LF scheme
for de re interpretation is presented in Cresswell and von Stechow 1982 and is applied
to temporal data in Abusch 1997. The mechanics of de re interpretation goes beyond
the scope of this chapter, but the bottom line in this example is straightforward.
Employing an acquaintance relation of temp oral identity lxltlt
0
[t ¼ t
0
] has the re-
sult that t (with the interpretation of the internal now of the attitude) is substituted
for u. Since t D [ t, y) ¼ t, this results in the denotation lt in-lead(t, b) for the com-
plement. This is a simultaneous reading.
This approach to simultaneous readings in which a substitution operator is present
is suppo rted by the fact that simultaneous readings can be mixed in various ways
with future ones. Example (69) is similar to (43).
46 Dorit Abusch
(69) Solange is likely to be in Paris now and in Stockholm next Friday.
(70) Solange is likely to already know all of the men she dates next year.
In the gapping sentence (69), the time for the locative predication in the right con-
junct is in the future. According to the analysis developed here, this means that it
must be under the scope of a temporal substitution operator. A standard analysis of
gapping copies the predicate for the right conjunct from the left conjunct. This would
require that the substitution operator be present in the left conjunct also. (70) makes
the same point. On one reading, the dating times are in the future, while the knowing
time is simultaneous.
The frame adverb now is essential in obtaining the simultaneous reading of (67),
because the reading results when the extended future interval is intersected with the
interval contributed by now . However, the sentence also has a simultaneous reading
when now is dropped. In this case, I assume th ere is a null frame adverb having the
same function as now. This is an innocuous assumption, since such null frame adverbs
are quite freely available (Abusch 1997, sec. 8). In (70), either know is modified by a
null frame adverb, or already serves the same function as now in (67).
1.4.4 LFs for Strictly Futurate Readings
In section 1.3, we saw that the to-complement of promise has a strictly future inter-
pretation. If this is to be directly stated in the representation, a substitution operator
with the interval (t, y) rather than the interval [t, y) should be used in the LF of
promise. I will assume for now that promise has an LF just like that of predict in (64),
except for the change in the interval involved in the substitution operator, and except
for di¤erences having to do with control versus raising.
Obviously, the di¤erent substitution operators are lexically conditioned. I am
inclined to assume that the presence of one operator rather than another follows
from the structured lexi cal representation of the embedding verb. This can be real-
ized strictly in the lexicon, or more syntactically by allowing lexical entries to stipu-
late local parts of LF trees.
1.5 Overtness of LF Structure for Infinitives
The argument having to do with tense interactions motivates subst itution for the n
parameter in the representation of future-oriented infinitives. This does not give us
much information about the LF repres entation, though. All we can definitely con-
clude is that the l-binder of n that is the head of node CP
4
in (62) is present. It might
be that the substitution itself is built into the lexical semantics of the verb. However,
I will present evidence that the substitution operator is syntacticall y overt, as are
Temporal Composition of Infinitives 47
higher levels of structu re. The argument suggests that what I call the contemporary
now of the attitude is represented in the LF of future-oriented infinitives.
In section 1.3, I presented a logical form for the complement of believe where the
contemporary now is the temporal argument of the property of times denoted by
the complement. In (71), by virtue of l-binding of n, the complement CP denotes
a property of times (typ e iwt). In the lexical semantics of the embedding attitude
predicate, the time argument functions as a counterpart of the attitude time.11
(71) Guido believes [
CP
ln [
TP
he is lying down]].
In cases where the tensed complement of a verb such as believe is headed by will, the
contemporary now is represented in the same way as in (71). The complement in (72)
has the structure (73), where SUB is the substitution operator that is the core mean-
ing of will.
(72) Paul believes he will have dinner with Monique today.
(73) [
CP
ln [
TP
n [
VP
SUB [
CP
ln [
TP
n [
VP
have dinner with Monique today]]]]]]
This follows from the requirement for a systematic syntax and compositional seman-
tics. For instance, given that the higher ln of (73) is present in (71), an analysis that
maintained that it was absent in (73) would involve complications in syn tax and in
the syntax-semantics map. There is also semantic evidence that the higher levels of
structure in (73) are present. If an NP takes scope inside the higher ln in (73) but
outside the lower one, we would expect a free n in that NP to pick up a contempo-
rary now. Examples where this is so can indeed be constructed. In Abusch 1997,
I suggested that the representation of the modal might includes a free n. This is
designed to account for the fact that in a top-level context such as (74), the temporal
perspective for might is the utterance time, while in an attitude context such as (75),
the perspective is the internal now of the attitude.
(74) Paul married a girl who might become rich.
(75) Paul believed his bride might become rich.
Examples (76a–c) show that might in a relative clause that is syntacticall y below will
can take scope outside will and pick up the internal contemporary now. Some time
ago, Paul misidentified a coworker of Guido’s as Guido’s sister. This dictates an LF
for (76a) in which a sister has narrow scope.
(76) a. Paul
1
believed that Guido had a sister
2
, and that she
2
might have a crush
on him
1
.
b. He believed that eventually he would have a long frank conversation with
the woman who might have a crush on him.
c. But he believed that at that point she would not have a crush on him any
more.
48 Dorit Abusch
In (76b), the definite description the woman who might have a crush on him must also
have narrow scope, because its presuppositional antecedent (which is the indefinite
description a sister in 76a) has scope inside the attitude. If the definite description in
(76b) had scope outside believed, that would imply that there was an actual woman
who might have a crush on Paul, which is not the way we understand the sentence. In
the understood meaning of (76b), the time parameter for might is the contemporary
now; this is made clear by the continuation (76c), which indicates that the time of
the woman having a crush is not the futu re conversation time, but an earlier time.
Assuming that all of this is to be represented structurally, the LF must be as in (77),
where the definite description takes scope right inside believed, and above SUB. It
has to be above SUB so that the n parameter on might picks up the contemporary
now, rather than a time following it.
(77) believed [
CP
ln [
NP
the woman who might(n) have a crush on him]
1
[n [SUB
[
CP
ln [
TP
n [
VP
he eventually have a long conversation with e
1
]]]]]]
The n on might is bound by the ln that represents the contemporary now, and so the
reading under discussion is obtained. None of this is surprising: it is what we would
expect given independently motivated denotations and the independently motivated
scope mechanism. The point of the example is to show how to use might as a diag-
nostic for the presence of a contemporary now in LF. In future-oriented infinitives,
the same kind of reading for might shows up as is observed with believe.
(78) Paul hopes to eventually have a conversation with the woman who might have
a crush on him.
(79) Paul decided to eventually have dinner with the woman who might have a
crush on him.
(80) Paul promised to eventually have dinner with the woman who might have a
crush on him.
Examples (79) and (80) are the significant ones. Because promise excludes a simulta-
neous reading for its complement, I postulated the same substitution operator in the
LF of promise as is used in the LF of will. (The same reasoning applies to decide.)
If this reading is to be represented with promise in the same way as with believe
(and why should the story be any di¤erent?), then the ln outside the substitution
operator in (77) has to be present in the LF for (80), as in (81). Here the n param-
eter in might(n) is captured by a ln that cr eates a property with a simultaneous
representation.
(81) Paul promised/decided [
CP
ln [the woman who might(n) have a crush on him]
le
3
[n [SUB ln [PRO have dinner with e
3
]]]]
Temporal Composition of Infinitives 49
Crucially, for this to work, there must be a ln corresponding to the contemporary
now in the representation. Suppose there were no such binder: suppose the LF had
the form in (82), which corresponds to the structure (62) discussed in section 1.4.
(82) Paul promised/decided [SUB ln [PRO have dinner with e
3
]]
Wherever the noun phrase containing might(n) takes scope, the n gets the wrong in-
terpretation. If it takes scope under the ln inside SUB, then the n in might(n) gets a
future interpretation. If it takes scope outside this l-binder, then the n in might(n)
remains free and is interpreted as the utterance time. Neither of these gives the
desired interpretation, where the n parameter in might is the internal now of the
attitude.12
If this argument is sustainable (and it should be remem bered that the argu-
ment depends on specific assumptions), then it shows that the highest levels of
structure that are present in tensed complements with will are present in futurate to-
complements (at least the ones in the promise/decide class). If so, the lower levels,
and in particular the SUB operator, must also be syntactically overt at the level
where scope is represented.
1.6 Conclusion
This chapter has motivated logical forms for futurate to-complements that contain
operators that make a substitution for the parameter n. This indicates that the rep-
resentation of such predicates is decomposed into a core meaning and an operato r
that introduces futurity. In addition to capturing data having to do wi th the interac-
tion between tense and futurity, this allows us to assume a single core meaning for
verbs such as promise that take both tensed and infinitival complements.
In the course of the discussion, two classes of futurate complement infinitives were
identified: ones that are purely futurate (e.g., promise) and ones that are also consis-
tent with simultaneous scenarios (e.g., predict and be likely). The methodology of
decompositional approaches to predicate meaning would suggest trying to explain
this di¤erence in terms of a motivated account of structured lexical semantics. Con-
ceivably, what makes promise and decide purely futurate has to do with their repre-
sentation in the calculus of causation and change: they describe acts of entering states
of commitment and intention.
The argument for a substitution operato r being syntactically present does not tell
us anything about the specific syntactic location of the operator. But locating th e
substitution operator hi gher than to would agree with the fact that there are three
di¤erent semantic classes of to-complements (simultaneous, purely future, and non-
specific future-simultaneous). (83) gives the LFs proposed for promise, predict, and
believe with to-complements. SUB
1
is the substitution operator using (t, y), and
50 Dorit Abusch
SUB
2
is the substitution operator using [t, y). In the LF of believe, there is no sub-
stitution operator.
(83) a. promise [
CP
ln [
IP
n [SUB
1
[
CP
ln [
IP
n VP]]]]]
b. predict [
CP
ln [
IP
n [SUB
2
[
CP
ln [
IP
n VP]]]]]
c. believe [
CP
ln [
IP
n VP]]
A simple way of matching th ese representations up with syntax is to identify to with
the most embedded n in I. The di¤erent structures above the most embedded CP
could then be treated as lexically stipulated syntactic stru ctures, as implemented for
example via incorporation (Baker 1988).13
The analysis proposed here is consistent with the Clausal Complement Hypothesis
suggested in Abusch 1998: all clausal complements have (either overtly or covertly) a
full CP/IP/VP structure. In the to-infinitive LFs of (83), the complements of promise,
predict, believe,SUB
1
, and SUB
2
all have this structure.
Other examples of this complementation structure are tensed complements (where
the CP and IP are overt) and the LF of will as in (48), which is like the LF (83a) with
a SUB
1
operator and a CP/IP layer embedded under it.
Notes
The research for this chapter was supported by Sonderforschungsbereich 340, funded by the
Deutsche Forschungsgemeinschaft. I would like to thank Mats Rooth for exciting discussions
of the material. The chapter was circulated in 1999 and was presented at Cornell University in
February 2000, at the University of Edinburgh in May 2000, at the University of Paris 7 in
November 2000, and at Rutgers University in April 2002. I would like to thank the audiences
at these presentations for their comments and suggestions. I am solely responsible for any
mistakes or inadequacy.
1. Under plausible assumptions, a representation of the scheduling interpretation also agrees
with property 1. Suppose we analyze this interpretation by means of a modal operator ac-
cording to schedule with top-level scope in the complement. Then in (10c), the future-denoting
adverb at 10 p.m. modifies the embedded predicate start rather than the top-level covert modal
operator. Since the modal is stative, this representation is compatible with property 1.
2. There is a curious point about how property 1 applies to the sentences in (13). It might be
that the LF of (13a) contains a future operator.
(i) predicted [FUT [
VP
[Bibi be in the lead][next week]]]
We want to apply property 1 to (13a) as evidence that predicted is an F-verb. For this to work,
we would have to treat be in the lead rather than FUT in (i) as the top-level predicate in the
complement. Otherwise, structure (i) would not be relevant for my diagnostic, as next week
would not modify the top-level predicate.
3. This is because simultaneous readings involve stative complements. Performatives are an
exception, assuming that in (i), the complement predicate headed by promise is nonstative.
(i) I a‰rm that I promise to clean up.
Temporal Composition of Infinitives 51
4. With expect, where the data in (i)–(ii) are parallel to those with predict in (26), there is some
feeling of a distinct elevated register, and perhaps a distinct word sense. I do not know what to
make of this, since the experienced meaning seems parallel to what is found with predict and its
synonyms, where there is no feeling of an elevated register.
(i) I expect that Guido spent the night of last Friday’s party at Monique’s place.
(ii) *I expect Guido to spend the night of last Friday’s party at Monique’s place.
5. A simultaneous interpretation is possible in (i).
(i) The treatment went so well that Monique promises to be pregnant.
Example (i) does not describe a speech act, since Monique is not making a promise to be
pregnant, but showing promise of being pregnant. In this example, promises does not have the
standard control sense; rather, this is the raising sense noted in Postal 1974.
6. In Abusch 1998, I used the syntactic label S where I now use VP and IP.
7. The system can easily be extended to accommodate a structure with an I
0
node and a subject
raised into IP.
(i) [
IP
NP [
I
0
I[
VP
eV
0
]]]
Rooth (1999) modifies the system in this way in order to theorize about VP-ellipsis.
8. One might want to compare this analysis with the proposal made by Dowty (1982) and
Nerbonne (1986), where a future operator makes a substitution for a tense-logical parameter
known as pseudo-speech time.
9. Technically, in this example the future interval comes out as (n, y), where n is contributed
by the matrix tense on will. A free n is pragmatically interpreted as the utterance time.
10. Apart from the fact that (59a) has only a de se reading for the PRO complement subject
(see Chierchia 1989).
11. This discussion is simplified, because it ignores de re and de se interpretation. In actuality,
I follow Lewis (1979) and take the argument of believe to have an argument position for an
individual (corresponding to the self ), so that the type is eiwt rather than iwt. In the LF of (71),
the individual argument is introduced by de re interpretation of he, using an acquaintance
relation of identity.
12. There is a further possibility of de re interpretation. This also gives the wrong reading,
since the position of the res would be occupied by a counterpart of the utterance time, rather
than a counterpart of the attitude time.
13. It is necessary to assume that in LF, operators and verbs are in the right-branching con-
figurations of (83). An incorporation structure such as the following would be uninterpretable:
(i) [ln [SUB
1
[ln predict]]] [
CP
e[
IP
e[
VP
e[
CP
e[
IP
n VP]]]]]
Since l is a binding operator, no compositional interpretation can be given for the traces of ln.
References
Abusch, D. 1997. Sequence of tense and temporal de re. Linguistics and Philosophy 20, 1– 50.
Abusch, D. 1998. Generalizing tense semantics for future contexts. In S. Rothstein, ed., Events
and grammar. Dordrecht: Kluwer.
Baker, M. 1988. Incorporation: A theory of grammatical function changing. Chicago: Univer-
sity of Chicago Press.
52 Dorit Abusch
Chierchia, G. 1989. Anaphora and attitudes de se. In R. Bartsch, J. van Benthem, and P. van
Emde Boas, eds., Semantics and contextual expression. Dordrecht: Foris.
Cresswell, M., and A. von Stechow. 1982. De re belief generalized. Linguistics and Philosophy
5, 503– 535.
Dowty, D. 1982. Tenses, time adverbs and compositional semantic theory. Linguistics and
Philosophy 5, 405–426.
Lewis, D. 1979. Attitudes de dicto and de se. Philosophical Review 88, 513–543.
Link, G. 1979. Montague Grammatik, die logischen Grundlagen. Munich: Fink.
Nerbonne, J. 1986. Reference time in narration. Linguistics and Philosophy 9, 83–95.
Postal, P. 1974. On raising: One rule of English grammar and its theoretical implications.
Cambridge, Mass.: MIT Press.
Rooth, M. 1999. Handout for talk given at MIT, April 2, 1999.
Temporal Composition of Infinitives 53
