Clause Selection

Phillip Potamites
August 8, 2007
Contents
1 Introduction 2
2 Method & goals (& assumptions) 3
2.1 Long-distance selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 The variable-free program . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Chain=semantic identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3 CP/IP 15
4 Entailment classes 18
5 P(w) 23
5.1 Propositionals, desideratives, and implicatives: P(w) . . . . . . . . . . . . . 23
5.1.1 Sequence-of-tense . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.2 Factives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.3 Interrogatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
6 P(i)(w) 32
6.1 n-comp non-n-comp . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
7 T 40
7.1 C
0
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
7.2 Independent tense and partial control . . . . . . . . . . . . . . . . . . . . . 43
7.3 Non-stativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
1
8 Case 48
8.1 Chomsky and Lasnik (1993) . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
8.1.1 Adverb interventions . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
8.1.2 Adjective Caselessness . . . . . . . . . . . . . . . . . . . . . . . . . . 49
8.1.3 Nominative environments . . . . . . . . . . . . . . . . . . . . . . . . 50
8.2 EPP: movement to Caseless positions . . . . . . . . . . . . . . . . . . . . . 51
9 P(x)(i)(w) 53
9.1 Binding: g & z, Jacobson (1999b) . . . . . . . . . . . . . . . . . . . . . . . . 54
9.2 Reexivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
9.3 Raising . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
10 Landaus critique of Hornstein 58
10.1 Expletives and Idioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
10.2 Complementizers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
10.3 Unaccusatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
10.3.1 en . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
10.3.2 si . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
10.4 Each . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
10.5 Case concord . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
11 Additional overt/covert dierences 62
11.1 De se/re . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
11.2 Missing strict readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
12 Conclusion 64
1 Introduction
This paper analyzes the thematic requirements of embedded clauses. It is part of a larger
research agenda which hopes to characterize predicate classes, their potential alternations,
and how dierent arguments depend on each other.
Clause selectors are known to restrict the availability of complementizers, subject man-
ifestation, tense, aspect, and aktionsart. This ability runs contrary to implementations of
compositionality, which would restrict selection to immediate constituents. However,
such long-distance selection is immediately provided for in type-driven function application.
This immediate resolution to a theoretical problem is the prime motivation for adopting a
semantic-like typing over the more traditional syntactic classes. Below is a list of some of
the types which I will argue for:
1
1
Due to confusing overlaps in the usual alphabetic choices, and because I want to use e for events
2
(1) a. e
b. an event
c. < w, t >
d. a function which, given a world, returns a truth value.
e. < i, < w, t >>
2
f. a function which, given an interval and a world, returns a truth value.
g. < x, < i, < w, t >>>
h. a function which, given a participant, an interval, and a world, returns a truth
value.
The traditional GB analysis of clause selection divided predicates among those that
selected CP and those that selected IP. Building on Pesetsky (1991)s analysis, I will argue
that this distinction, while characterizing important phenomena, is not sucient. More
recent theories of clause selection are locked in a struggle over the proper contribution
of the feature [+/-T/tense] (Landau (2000), Martin (2001), Pesetsky and Torrego (2001),
Pesetsky and Torrego (2002), Pesetsky and Torrego (2004), Pesetsky and Torrego (2006)).
While commendable in avoiding a proliferation of features, their over-lapping accounts
of disparate phenomena suggest that additional distinctions are also necessary here to
maintain any formal characterization of the relevant data. The semantic skeleton I adopt
will be responsible for thematic assignments, while syntactic properties, such as Case (cf.
Jean-Roger Vergnaud), and the concept of range assignment (cf. Hagit Borer), will be
necessary to begin to characterize the more subtle clause distinctions. The way that these
formal properties reect deeper interpretive implications will be suggested, but not yet
fully resolved.
2 Method & goals (& assumptions)
(3) I will argue in this section:
a. Type-matching is a necessary and appropriate tool for long-distance selection.
instead of entities, I have chosen to use type letters which correspond exactly to their typical representation
as variables. Given the specic contexts of their use, there should be no danger of uncertainty.
2
Abusch (1999), p. 24, actually argues for the following correlation:
(2) (a) VP: < i, < w, t >>: tenseless clause
(b) IP: < w, t >: tensed clause
(c) CP: < i, < w, t >>: complement clause
Given that she designates Barak is predicted to be in the lead now as embedding a CP, I do not think
we should assume these terms have the implications that they did in the GB framework. Since my types
describe the overt appearance of clauses, before their more subtle dierences, they also fail to have any
direct correlation with that framework.
3
b. Methods like the variable-free program (cf. Pauline Jacobson) are necessary
to avoid the confusing and illicit mixture of discourse and binding in assignment
functions.
c. The concept of a chain is only unied in a semantic sense, and therefore must
be expanded to allow multiple and Case assignments.
2.1 Long-distance selection
I claimed above that clause selectors restrict their complementizer, subject manifestation,
tense, aspect, and aktionsart. Below, I provide evidence for each of these.
(4) complementizer selection:
a. Jack hopes that Jill runs.
b. Jack hopes for Jill to run.
c. *Jack wants that Jill runs.
d. Jack wants for Jill to run.
e. Jack believes that Jill runs.
f. *Jack believes for Jill to run.
Although hope and want are usually classied together as desideratives, hope can
take that- and for-clauses (correlated with niteness, of course), while want fails with
nite that, and believe fails with for.
(5) subject manifestation:
a. *Jack believes to be intelligent.
b. Jack believes himself to be intelligent.
c. Jack tried to win.
d. *Jack tried himself to win.
e. Jack expects to win.
f. Jack expects himself to win.
The complement of believe requires an overt subject, while try requires a silent
subject, and expect takes either.
Tense, aspect, and aktionsart involve complex semantic, syntactic, and morphological
issues. Below, I divide them somewhat supercially. For present purposes, these crude
distinctions should suce to substantiate my claims about selection. In this crude sense,
tense selection can be demonstrated by the selection of to/-ing.
4
(6) tense:
a. Jack wants to run.
b. *Jack wants running.
c. *Jack nished to run.
d. Jack nished running.
e. Jack likes to run.
f. Jack likes running.
As shown, non-nite selectors vary in allowing one, the other, or both.
Aspect, here, means one temporal orientation embedded in another. The aspect of
clausal complements also varies with the selecting predicate.
(7) aspect:
a. Jack expected to run.
b. Jack managed to run.
While expect imposes a future oriented interpretation on its complement, managed
seems to imply a co-occurrent interpretation.
Finally, certain propositional selectors (also associated with ECM and potential pas-
sivization) are known to require stative complements.
(8) aktionsart: stative selection
a. *Jack believed Jill to fall down.
b. Jack believed Jill to be intelligent.
c. Jack believed Jill to have fallen down.
d. Jack believed Jill to be running.
e. Jack believed Jill to fall down often.
f. Jack believed sharks to attack humans.
As shown, stative apparently must include perfects and progressives, and, at least
marginally, habituals and generics. As (a) shows, predicates like believe do not allow
simple eventives.
This paper will not be able to resolve all the distinctions presented above, but it will
attempt to clarify methods which I hope may ultimately lead to their solutions. For the
moment, I have merely tried to show that the selection of all these properties is a necessary
ability of clause-selecting predicates.
5
Regardless of this evidence of long-distance selection, theories are sometimes still pro-
moted on the basis of restricting sub-categorization to immediate constituents. David
Adgers usually clear and up-to-date text book, Core Syntax (2003), reects the problem
with a rather confusing shift that leaves his ultimate position unclear. He rst allows
try to select sub-features of its sister category, and considers extending that method to
ECM verbs like believe, but then rejects this proposal on the basis of avoiding further
structure (p. 311). Hasnt he then also rejected his own account of try ?
(9) p. 310:
We now propose that a verb like try selects a CP headed by a complementizer with
this [null] feature.
3
(10) p. 311:
An obvious approach [[to ECM predicates]] would be to say that the embedding
verb selects a complementizer in the same way that try does. However, this comple-
mentizer has accusative features rather than a [null] case feature.
(11) p. 311:
However, as our theory stands, we simply have no way of stating the lexical entry ...
C-selection features are simply uninterpretable categorical features with no further
structure ...
(12) no way of stating: (p. 311)
believe, [V, uC[acc]]
He ultimately argues for a little v bearing [accusative], and apparently avoids AgrO,
by long-distance checking of Case. Such distance checking is just another way of achieving
long-distance selection, and I will also assume that something in the verbal complex that
is pronounced believe must satisfy its accusative Case.
We have already seen that Case is not the only embedded-selection, and that many of
these selected properties are highly interpretive, and not purely formal.
In a type-driven -calculus, these sub-selectional properties are immediately manifest
(for -roles, though not Case). Examples to come will reect some of the chains required by
the above dependencies. For the moment, a simple example of a lifted subject, may serve
as a better introduction, while demonstrating the sub-selection property of type-matching
function application.
(13) a. Jill ran.
3
In section 8.1, Ill discuss the further problem of selecting a phrase with a feature which should have
already been deleted. I believe the problem that Adger is discussing applies more to [null] Case, which
should be checked and deleted internal to a non-nite CP, than to accusative Case, which is reasonably
associated with the external selector, and not necessarily deleted internally.
6
b. [ x . x ran ]
<x,t>
(Jill
x
)
c. [P
<x,t>
. P(Jill
x
)]
<<x,t>,t>
(x . x ran)
<x,t>
d. Every boy ran.
e. [ P
<x,t>
. x s.t. boy(x): P(x) ]
<<x,t>,t>
(y . y ran)
<x,t>
(b) is an extremely simple example of decomposing (a) into a function of type <x,t>
and an argument of type x. The function applies to the entity (type x) Jill, and returns a
t=truth value. (c) is equivalent, but Jill has been lifted, so that it is the selecting function,
and the predicate becomes its argument. After lifting, Jill is of type << x, t >, t >: it
takes an argument of type < x, t >, i.e. the predicate, and still returns a t. Given the
equivalence of (b) and (c), lifting may seem pointless, but it plays an important role in
relating argument positions, and the more complex types that sometimes occupy them; for
instance, the quantier phrase shown in (d,e). Furthermore, it will play a necessary role
in the variable-free binding system to be promoted here.
The crucial point is that (c), and (e), insure that the selected argument is also in need
of a specic kind of argument. This kind of transitive restrictiveness would seem to be
exactly the kind of tool we need to deal with the properties in the data presented above.
Perhaps, we could allow lexical entries, like the one Adger excludes, but I would argue that
that approach lacks the explicit formalism of the type-driven semantic formulas.
I have tried, in this section, to show that long-distance selection is necessary, and that
type-matching provides a reasonable means to its formalization.
2.2 The variable-free program
The work of Pauline Jacobson provides numerous theoretical and empirical motivations
for the variable-free program.
4
Though the name is catchy, I will continue to use the term
variable for exactly those arguments which are bound by -operators, and which may
be identied, through linguistic binding operations, with other arguments. I would here
like to add some additional motivation for insisting that the use of variables be speci-
cally constrained in this way, and that discourse reference not be freely intermixed with
syntactically constrained binding.
Traditional approaches to binding make use of assignment functions, their modication,
and the addition of a -operator binding the variable connected to that modication.
(14) Heim and Kratzer (1998), p. 186: Predicate Abstraction Rule (PA):
Let be a branching node with daughter and , where dominates only a numerical
index i. Then, for any variable assignment a,
a
= xD .
a
x/i
4
The crucial nature of this program is that all variables are bound, and thus, for representational
purposes only (Jacobson (1999a)). I think I would prefer to call it the no free variables program. I
am not as concerned with excluding variables as much as I want explicit semantic binding without illicit
binding of free variables.
7
(15) Heim and Kratzer (1998), p. 112:
Let a be an assignment, i N, and x D. Then a
x/i
(read: a modied so as to assign
x to i
5
) is the unique assignment which fullls the following conditions:
a. dom(a
x/i
) = dom(a) {i},
b. a
x/i
(i) = x, and
c. for every j dom(a
x/i
) such that j = i: a
x/i
(j) = a(j).
(a) insures that the domain of the assignment function a
x/i
includes i; (b) insures that
x will be substituted for any embedded is; and (c) insures that a
x/i
is identical to a in
every way except for the interpretation of i. The rule assumes that every binder sits above
a numerical index indicating its binding.
(17) a.
Who

did Jill see 1

b.[ Who [x . did Jill see 1
a
x/1
] ]
The rule in (14) insures that (17a) produces (17b) and the rules in (15) then insure that
(17b) means [ Who x[ did Jill see x ]]. The abstraction rule connes itself to a context
like that represented in (18a), but it should be clear that its primary function focuses on
situations like that in (18b), where the daughter index is matched to one more deeply
embedded by its sister.
(18) iN:
a.

b.
i
... i ...
We could even represent the meaning of such an index i, equivalently, as shown in (19a).
It takes any kind of argument, adds a -operator, and modies the assignment function
accordingly. In (b), this function is shown applying to an argument that is some function
over i, like that shown in (18b).
5
Explicitly representing, in a single notation, this super-superscript might also help to expose the problem
I am talking about. We want to substitute i with x, and thats what -operators are for, so lets write it
that way:
(16) a. i f(i)
a
=x . f(i)
a
x/i
=
b. [ix . f(i)
a
](x)=
c. x . f(x)
a
8
(19) a. i=Px . P
x/i
b. [Px . P
x/i
]((i))
c. [Px . P]((x))
d. x . (x)
Essentially, the interpretation of (b) is just the same as saying that (c) is equivalent
to (d). However, such a move involves a fundamentally illegal operation. To maintain
equivalence, uninstantiated operators should maintain the number and relative position of
all variables. For instance, (20a), should not be considered equivalent to (20b) (see Chris
Barkers excellent -tutorial, on his webpage).
(20) a. [x[y[x y]]](y)
b. [y[y y]]
c. [z[y z]]
Instead a process known as -reduction is designed to prevent such illicit bindings. All
variables must be replaced by non-identical designations before formulas can be substituted
into one another. Thus, (20a) could be equivalently represented as (c), but not (b). The
use of assignment functions to achieve syntactic binding then seems to function as a means
to circumventing this essential rule and to allow temporarily unbound elements to become
bound.
Allowing these kinds of modications to our assignment functions raises several ques-
tions about their proper representation and interpretation. They are most simply accepted
as references to discourse:
(21) a. He
1
saw her
2
b.

1 Jack
2 Jill

c. Every girl
2
said he
1
saw her
2
.
d.

1 Jack
2 x

Their use in binding assumes that they can also return variables, as shown in (d). But,
if we allow functions of that form, and dont impose a rule of -reduction, we can derive
improper co-bindings.
(22) a. The boy that likes every girl
1
called her
2
.
b. The boy that likes Jill
3
called her
3
.
c. Every boy
1
said he
1
likes him
2
.
9
d. ?

1 x
2 x

I assume that one variable could have been in the assignment function to begin with,
and the other would be introduced by the binding operations. But this reading, where
every girl binds her, is not truly available in (a). We could, of course, impose a rule
such that only binding could introduce variables to the assignment function, but this
move would only highlight the oddity of their presence there. We would still need to
impose a rule of -reduction, within the assignment function, so that multiple independent
bindings dont overlap, but this only increases the number of computational stipulations
that we would impose on our discourse references.
6
In fact, overlaps in the assignment
function are necessary to allow for the phenomenon known as accidental co-reference. In
other words, putting binding into the assignment function requires also moving additional
composition rules which serve no other purpose there. Necessary alphabetic variation is
already implemented within the compositional framework implied by the lambda calculus;
extending it to deal in discourse with a problem previously perceived as syntactic just reeks
of manipulating inappropriate tools to push aside certain dilemmas.
Furthermore, following Reinhart (1983) and others, we would like to maintain essen-
tial dierences between co-reference and binding. Thus, (b) easily achieves a co-referent
interpretation while (a) cannot. If we allow our assignment functions to return bindable
variables, it is no longer clear why (b) is any dierent than (a).
However, (c) raises more debatable issues. Any intended co-indexing between the 2
masculine pronouns would clearly violate rules on the use of reexives, but I think it
might be possible to claim that there is a de re reading where every boy likes, unwittingly,
someone who turns out to be himself. This reading, if available, would seem to be exactly
what our accidently over-lapping variables would produce, and thus, might be seen as
an argument for these derivations. However, I believe it would be preferable to derive
this reading from some sort of choice function mechanism, and not to wholesale mingle
reference and binding to achieve this kind of singular, semi-marginal reading. I cannot
pretend to fully resolve the reference-binding issues here, but it is my hope that the links
between syntactically-constrained antecedent-anaphor pairs can be fully represented in the
compositional semantics without appeal to external functions.
Thus, following Jacobson, I assume that unsaturated arguments can survive to the end
of the derivation, just as long as no formal, syntactic features go unsatised. Discourse
will then be the mechanism appealed to, to resolve these unsaturated arguments.
(23) a. He ran.
b. x . x ran
c. Its necessary to study hard for this test.
6
Wed practically be implying that our syntactic composition rules dont adhere to -reduction, but our
discourse function does.
10
d. x . necessary(x to study hard for the test)
Simplifying extensively, for the moment, in ignoring gender, etc., a proposition em-
bedding a pronoun is not fully saturated until that pronoun receives reference. Being
unbound, it must be embedded in discourse before the proposition can be properly inter-
preted. The ultimate interpretation of the subject of study also seems dependent on some
extra-syntactic function (i.e. to provide it with an argument).
7
Jacobsons combinatorics also allow for the following identity, while making use of
additional syntactic apparatus to distinguish the overt manifestations.
(25) a. x . ran(x) = ran
b. ran, NP/
L
S
c. He ran, S
NP
I prefer to expect the presence (or absence) of -operators to represent some signicant
information, as well as making using of Case to insure or prohibit syntactic projection.
(26) a. ran: x . ran(x),[nom]
b. He ran: x . ran(x)
c. x . ran(x)
Px . P(x),[nom] y . ran(y),[nom]
2.3 Chain=semantic identity
Two traditional

A chains are presented in the following examples. The simplied -formulas
show how the wh-term binds its trace, while the trees represent cyclic movements which
syntacticians have argued for.
(27) a. Who
1
[ t
1
ran ]
8
b.
Who
t
1
ran
7
Allowing unsaturated semantic slots might also help represent passive/unaccusative distinctions (though
I should refrain from hastily over-simplifying this matter):
(24) a. His crown was broken (by the naughty children).
b. [ xy . y broke x ](his crown)
c. His crown broke (*by the naughty children).
d. [ x . x broke ](his crown)
8
Following the roughly isomorphic notation of Heim (1998)
11
c. Who
1
[ did John think t
1
ran ]
d.
Who
did
John
think
t
1
ran
In the system advocated here, the chain doesnt necessarily project (by which I mean,
leave traces in silent terminal nodes), but it is passed up as a -binder in the left margin,
at every node, in fact, as shown below. The question of barriers, or phases, then becomes
a question, not of where the chain needs temporary docks, but where it fails to pass up.
(28)
Who z . did John think z ran
did z . John think z ran
John yz . y think z ran
Py . y think P z . z ran
z . z x . x ran
It is important to note that many of the mergers above violate our type constraints on
mergers. In many cases, passing will be motivated by exactly these type mismatches. Spe-
cic operations allow these mismatches to be resolved. These operations will be presented
in section 9.1, where Jacobsons analysis will be presented in greater detail. Barriers, then,
can also be seen as specic prohibitions on such operations.
Both traditional and variable-free approaches ultimately try to derive roughly the fol-
lowing parallel representations for the various constructions below.
(29) a. Every boy thought he was smart.
b. Every boy (x . thought(smart(x))(x))
9
c. Which boy did Jill think was smart?
9
I infer this kind of representation directly from Jacobson (2000), p. 111, ex. 60, where she represents
thinks he lost; S/LNP; x[thinks(lost(x))(x)], which is semantically identical to the standard account
which she presents on p. 106, ex. 53.
12
d. Which boy (x . thought(smart(x))(Jill))
e. Every boy wanted to be smart.
f. Every boy (x . wanted(smart(x))(x))
g. Every boy seemed to be smart.
h. Every boy (x . seemed(smart(x)))
While the manifest expression of pronouns, quantiers, questions, and control (a,c,e) are
all very dierent, the semantic formulas postulate similar structures which are derived by
identical methods (either modication of the assignment function or passing operations).
Raising and control (e, g) are overtly identical, but distinguishable in structure by the
dierence of a single -role. In section 9.3, however, we will see that the variable-free
approach requires dierent operations for raising and control, though their results will
equally be chains in the sense shown above.
10
In a syntactic sense, the concept of chain must fragment into at least the following
categories:
(30) a. A
(a) unaccusative
(b) passive
(c) raising
(d) control
b.

A
(a) wh
(b) QR
c. pronominal binding
A-chains (usually excluding control) are normally claimed to receive -roles at their
tails, and Case at their heads.

A-chains are normally claimed to receive both -roles and
Case at their tails. Control is usually not considered a chain, primarily because it relates
2 -positions; also, perhaps, because control predicates present evidence of CP-selection,
and CPs seem to block A-movement, though not

A-movement; also, because arbitrarily-
interpreted silent subjects are assumed to be the same thing. Binding is normally not
considered a chain, since it relates positions hosting their own independent Case- and -
satisfactions, though it is often claimed to at least have a c-command requirement
11
, as all
chains have been argued to.
Though grouped under the A-movement category, unaccusatives, passives, and raisers
need some kind of analytic terms to characterize their overt dierences:
10
Though it is not entirely clear to me, yet, how these dierent operations would derive all the phenomena
which distinguish raising and control, some suggestions will at least be proposed.
11
Though both of the following kinds of sentences create complications for this claim, particularly the
second:
13
(32) a. The table broke. (unaccusative)
b. The table was broken. (passive)
c. The table seems to wobble. (raising)
Wh-movement and QR are usually grouped as

A-movement, though it would seem that
one lexicalizes its head, and one its tail.
(33) a. Who did Jack see?
b. Wh x: did Jack see x?
c. Jack saw everyone.
d. Every x: Jack saw x
Furthermore, they demonstrate dierent extraction properties, as shown in the following
(classic) examples:
(34) a. *Who did a friend of call Jack?
b. *Wh x: did a friend of x call Jack?
c. A man from every city loves it.
12
d. Every x: A man from x loves x
13
In this section, I have argued that chains are unied semantic concepts, but distinct
under a variety of syntactic restrictions and licensing conditions. Pronominal binding, con-
trol, and raising interpretations all require identications derived through chain operations.
Therefore, claims that chains receive only one -role, or one Case, must be seen as only
referring to certain kinds of chains. Pronominal binding chains necessarily receive multiple
Case and -assignments, and (obligatory) control is even more easily seen as an example
of chain formation.
Finally, let us note, as an additional argument in favor of the variable-free method, that,
while the traditional semantic approach, with its assignment functions, oers no reason to
project anything that would correspond to the cyclic movements required by past syntactic
proposals, the variable-free method represents binders whose movements up the tree are
specically required, and licensed or restricted in specic contexts.
(31) a. Somebody from every city despises it. (May (1985))
b. The woman who every man admires (the most) is his mother. (Geach (1972))
See Jacobson (1999b) for a choice-function analysis of (b), and the main text here, (34), for a quick cover-up
of the controversy surrounding (a).
12
Whitewashing Mays original a bit.
13
As well as rejecting his noun adjunction revision.
14
3 CP/IP
Pesetsky (1991) characterizes the analysis of (non-nite) clause selection in the LGB frame-
work according to the following paradigm:
(35) IP Complementation:
14
a. Jack believed Jill to be fast. (ECM)
b. Jill was believed to be fast. (passive)
c. *Jack believed to be fast. (PRO)
(36) CP Complementation:
a. *Jack decided Jill to run. (ECM)
b. *Jill was decided to run. (passive)
c. Jack decided to run. (PRO)
If all predicates fell into just one of these classes, without ambiguity, we could make a
prediction like the following:
(37) (incomplete):
ECM PRO
Passivizes *
Cant passivize *
Assuming that some predicates are ambiguous, we would also derive predicates like
expect.
15
(40) a. Jack expected Jill to run.
b. Jill was expected to run.
14
Which also explains why we cant say Jack believed for Jill to be fast.
15
There is actually a lot of confusion about the proper analysis of expect.
(38) a. Jill expects Jack to leave.
b. Jill expects Jack.
Since (a) does not entail (b), it could be argued that Jack, in (a), does not receive a -role from expect.
(39) a. Jill expected Jack to fetch the pail.
b. Jill expected the pail to be fetched by Jack.
The sense that these might not mean the same thing suggests that expect might assign a patient -role
after all; which would then oer an easy account of its passive ability.
15
c. Jack expected to run.
In fact, the LGB framework also distinguishes predicates like dare or persuade, where
the matrix accusative Case really is associated with a matrix -role. These types also need
to be distinguished from each other, since only one allows subject-control.
(41) a. Jack dared Jill to run. ( Jack dared Jill.)
b. Jill was dared to run.
c. Jack dared to run.
(42) a. Jack persuaded Jill to run. ( Jack persuaded Jill.)
b. Jill was persuaded to run.
c. *Jack persuaded to run.
Predicates like want, however, still require additional stipulations, since they do not
t the mold above:
(43) a. Jack wanted Jill to run.
b. *Jill was wanted to run.
c. Jack wanted to run.
Its failure to passivize argues for CP-selection, while the ECM-like construction is
attributed to a Case-assigning silent complementizer.
Other predicates that might be categorized with want actually require overt comple-
mentizers when appearing with overt embedded subjects (which can be taken to support
a CP analysis, at least for those overt subject clauses, but leaves many questions unan-
swered
16
).
(44) a. Jack hoped *(for) Jill to run.
b. *Jill was hoped to run.
c. Jack hoped to run.
The potential instantiations of the the 2 initial categories (CP & IP) can be summed up
as follows (where the crossed out items indicate prohibited elements, PRO is the subject
of control clauses, lex stands for a lexical item, and NP-t for the trace of passivization, or
raising).
16
Such as when or why the silent complementizers are available, or what determines niteness, ...
16
(45) CP
C
for/that/
TP
PRO/lex/NP-t T
T
to/-ed/(-ing?/...?)
VP
(46) TP
PRO/lex/NP-t T
T
to/(-ing?/...?)
VP
Additional principles must then explain the further restrictions in co-occurence between
these terms. For and that have been prohibited from occuring with NP-trace, but we
also need some additional rule to exclude them with PRO. Restrictions on selecting that,
for, or are left to arbitrary selection. The exclusion of lexicalized subjects in many
control contexts also appears to be unexplained, unless one additionally proposes dierent
kinds of null complementizers, which is exactly what Pesetsky (1991) eventually does.
(47) a. Jack tried to run.
b. *Jack tried Jill to run.
Pesetsky (1991), instead, assumes that all clauses should be CPs, and begins with the
type categorization diagrammed below.
17
(48)
-PRO, +NP-trace +PRO, -NP-trace
+ECM believe want
-ECM wager demand
Pesetsky then goes on to develop a theory of null complementizer axes which must
move from C to the selecting V. His treatment of the topic is thorough and highly recom-
mended, but also requires many kinds of silent elements making many movements. I will
17
Try falls in the demand class.
17
try to characterize the properties of these predicates more directly with 3 independently
justied tools: 1.) -roles, as formalized in the -calculus; 2.) Case, which I will generally
associate with a necessary, overt reex; and 3.) range, which links these arguments and po-
sitions with event qualities. Speaking in an informal way, if the rst is purely interpretive,
and the second purely formal, then I think the third is somewhere in the middle.
I have tried to show in this section that the CP/IP contrast is insucient, and will
probably not provide the most ecient path to a minimal characterization of clause select-
ing properties. In fact, even if the CP/IP distinction is necessary, or were sucient, I am
not really sure by what standard it would count as an explanation. I would still like some
better sense of what further properties of these predicates cause them to fall into whatever
class (see Pesetsky (1991), for compelling support of this analytic ethic). I have put aside
Pesetskys proposals, with insucient discussion, in favor of what I believe will be more
direct descriptions of the phenomena.
4 Entailment classes
Presumably, the classic framework would treat all nite clauses as CPs. However, the pred-
icates which select them often impose signicant dierences in interpretation. Non-nite
clauses also participate in these dierences, as well as introducing additional complications.
Following Landaus division of clause-complement selectors, we can consider the fol-
lowing predicate classications. In fact, many of these items vary their entailments across
their lexical alternations.
(49) Landau (2000), p. 38, ex. 29:
a. Implicative: dare, manage, make sure, bother, remember, get, see t, conde-
scend, avoid, forget, fail, refrain, decline, neglect, force, compel.
b. Aspectual: begin, start, continue, nish, stop, resume.
c. Modal: have, need
18
, may, should, is able
19
, must.
d. Factive: glad, sad, regret, like, dislike, hate, loath, surprised, shocked, sorry.
e. Propositional: believe, think, suppose, imagine, say, claim, assert, arm, de-
clare, deny.
f. Desiderative: want, prefer, yearn, arrange, hope, afraid, refuse, agree, plan,
aspire, decide, mean, intend, resolve, strive, demand, promise, choose, oer,
eager, ready.
g. Interrogative: wonder, ask, nd out, interrogate, inquire, contemplate, deliber-
ate, guess, grasp, understand, know, unclear.
18
I guess that he puts this predicate here because of its status in other languages. It could be more
appropriate as a desiderative.
19
This item also seems odd here.
18
Implicatives can imply or deny the occurrence of their complements (see Karttunen
(1970), Karttunen (1971)), and their negation can reverse the implication. Given that
this reversal only occurs in some of the above classied predicates (in some of their con-
structions), they might require reclassication. However, many of them would still fail to
classify as factives in the sense to be described further along.
(50) a. Jack made sure that Jill ran. (Jill denitely ran.)
b. Jack didnt make sure that Jill ran. (uncertain)
c. Jack dared to run. (denitely ran)
d. Jack didnt dare to run.
20
(denitely didnt run)
e. Jack failed to succeed. (no success)
f. Jack did not fail to impress. (lots of impress)
Aspectuals seem to share this property, but also imply some kind of temporal properties.
(52) a. Jack started to run. (some running, but not necessarily complete)
b. Jack didnt start to run. (no running)
Modals have disparate entailments, but, if we recategorized need and is able, we could
at least say that they impose specic inections (possibly none) on their complements which
overtly appear as no more than a basic VP.
(53) a. Jack has fallen. (did fall)
b. *Jack has fall.
c. *Jack has to fall.
d. *Jack has Jill to fall.
e. *Jack has for Jill to fall.
f. *Jack has that Jill fell.
(54) a. *Jack should fallen.
b. Jack should fall. (This fall hasnt occurred yet.)
c. *Jack should to fall.
d. *Jack should Jill to fall.
20
Notice that the object-control alternation loses this implicative property.
(51) (a) Jack dared Jill to run. (running uncertain)
(b) Jack didnt dare Jill to run. (running uncertain)
19
e. *Jack should for Jill to fall.
f. *Jack should that Jill fell.
Factives seem to presume the occurrence of their complements, regardless of negation
(see Kiparsky and Kiparsky (1971)), at least in nite contexts. The proper treatment of
their non-nite forms is less clear.
(55) a. Jack is glad that Jill ran. (Jill denitely ran.)
b. Jack is not glad that Jill ran. (Jill still ran.)
c. Jack is glad to run. (sounds hypothetical)
d. Jack is not glad to run. (same)
(56) a. Jack likes that Jill ran.
b. Jack doesnt like that Jill ran.
c. Jack likes to run. (Doesnt that presume that he runs sometimes?)
d. Jack doesnt like to run. (Do you have to have tried it to dislike it?)
In so far as Im glad to go to the store can indicate a willingness to participate in a
yet-to-occur event, glad seems to lose its factivity in the innitive, while like has some
partial sense that it should be retained.
Imaginary NPs, however, cast doubt on the claim that factives presume the occurrence
of their complement in the matrix world. Evaluating imaginary NPs, in so far as we accept
statements about them as true, requires contextualizing them to imaginary worlds, as
shown by (a-b).
(57) a. Santa Clause brings presents on Christmas.
b. In all worlds where Santa Clause exists, he brings presents on Christmas.
c. Kids love that Harry Potter is a wizard.
d. In this world, kids love that Harry Potter is a wizard in his worlds.
(c), on the other hand, would be misconstrued if we assumed that Potter is a wizard
in this world, or if we assumed that children only love Harry Potter in worlds where he
is a real wizard. I, therefore, accept that factives can contextualize their complements to
appropriate discourse worlds.
21
Propositionals do not verify anything about their complements, but their complements
can have truth or falsity predicated of them.
21
I was forced to this conclusion by a number of friends who refused to accept that me saying Jack loves
that Santa brings presents had to imply that I believe in Santa, although they accepted that Jack loves
that Jill ran implied that I believed Jill ran.
Anankastic conditionals also oer unexpected examples of pragmatic world availability.
(58) a. Theres some biscuits in the cupboard, if you want some.
20
(59) a. Jack believes Jill to be a spy, which is true/false.
b. Jack believes that Jill is a spy, which is true/false.
This property seems to contrast with desideratives, which are more emotive.
(60) a. *Jack wanted to run, which is true.
b. *Jack planned to be right, which is true.
However, it also seems clear that temporal-aspectual, and (non)-indicative mood re-
strictions interfere with a full comparison. Many desideratives prohibit nite, indicative
complements
22
, and those that allow them dont seem quite as bad. Use of individual-level
predicates, or shifting the aspect of the truth predication also seems to improve quality.
(62) a. (?)Jack hoped that Jill is a genius, which is true.
b. (?)Jack is afraid that Jill is a spy, which is true.
c. (?)Jack wanted Jill to be a genius, which is true.
d. (?)She hoped he won, which turned out to be true.
e. (?)She intended to win, which came true.
b. If you want to go to Harlem, you have to take the A train.
The special characteristic of the biscuit sentence in (a), is the fact that it is, typically, not relativized to
the antecedent world. We can imagine a reading where the speaker only asserts the reality of the biscuits
in the cupboard in those worlds where you want some, but it is more likely to assume that they are there
regardless of our desires.
In (b), want is again disregarded, on the usual interpretation, and the train-taking should be restricted
to just those worlds where one actually goes to Harlem. Again, we can imagine a reading where you
(have to) take the A train in all worlds where you want to go, but it seems more reasonable to assume
that the sentence only implies A-train-taking in all worlds where you actually go to Harlem. Alternative
proposals have stumbled in dening the set of train-taking worlds as all those where one wants to. Then
contradictory desires or restrictive desires within multiple options can result in unintended falsications.
Getting the consequent relativized to worlds of going, in the case of the train sentence would seem to resolve
at least some of these issues.
The combination of the biscuit and train sentences suggest that the consequent can be relativized to the
real/speech world (as in the case where the biscuits are assumed to exist in any case), to the wanted
worlds (where the biscuits manifest themselves according to my desires, and I have to take the A train,
even if I want to go to Brooklyn more), or to the worlds of actually going (where I (have to) take the train
in all those worlds where I go).
22
Desideratives also vary in their ability to take the subjunctive:
(61) a. *Jack hopes that Jill run.
b. Jack demands that Jill run.
I will not attempt to include any account of licensing the subjunctive in what follows. Hopefully, this
licensing will eventually be derived from certain world relationships. However, it would certainly seem odd
to say that the complement of hope must be less hypothetical than the complement of demand.
21
Nonetheless, the contrast between certain, supercially similar non-nites can be quite
strong, and requires some account:
(63) a. Jack believes Jill to be a spy, which is true.
b. *Jack chose Jill to be a spy, which is true.
I do not believe that calling them irrealis provides a sucient account of this oddity,
in so far as the complement of believe must also be asserted in hypothetical worlds.
Finally, interrogatives embed questions, and they seem to share the factive property.
(64) a. Jack said who ran.
b. Jack didnt say who ran.
c. Jack asked who ran.
d. Jack didnt ask who ran.
These sentences presume that someone ran, probably in the matrix world, but maybe
in some other.
(65) a. Jack asked where Santa lives.
b. Jack said where Santa lives.
Again, I assert the asking and saying in this world, while I presume Santas existence
in some other(s).
In the following sections, I will consider formal characterizations intended to capture
at least some of the above distinctions, rst as they apply to nite clauses, then overt
subject innitives, and nally silent subject innitives. Many questions about world-time
interactions, their possibilities, and meanings will be considered, but often postponed for
further research.
5 P(w)
5.1 Propositionals, desideratives, and implicatives: P(w)
Heim and Kratzer (1998), in their contemporary classic, propose exactly this complement
type for intensional attitude verbs, like believe:
(66) a=p. 306a:
23
a. believe
w
= pD
<w,t>
. [xD . p(w) =1 for all w W that are compatible
with what x believes in w]
23
I have taken the liberty of changing their s type to my w.
22
b. x believes p means that p is true in all worlds that are compatible with what
x believes in w.
Since, desideratives also require hypothical worlds, we can consider an exactly parallel
denition for hope. Heim and Kratzer (1998) also make this proposal. Obviously, this
fails to account for any dierence sensed in the which is true test. In fact, I will also
have to leave this problem as a topic for future research.
(67) a=p. 306a:
under consideration:
a. hope
w
= pD
<w,t>
. [xD . p(w) =1 for all w W that are compatible
with what x hopes in w]
b. x hopes p means that p is true in all worlds that are compatible with what x
hopes in w.
I, however, am a little uncomfortable with the notion that a specic hope must be
true in all worlds that are compatible with anything I hope.
24
It seems to rule out the
possibility of contradictory hopes or disparate hopes. For instance, hoping that I nd a
good job seems to rule out the possibility that I hope even more that I win the lottery and
never work again. Of course, the statement of hope should only be interpreted within the
exact parameters that it describes, but it is the universal quantication that seems to go
beyond these parameters. Therefore, I think it is worthwhile to at least consider another
possibility. The hoped-for event must occur in a hypothetical world, but this world is
otherwise assumed to be identical to this one. I make use of a second order equation to
achieve this eect.
25
(68) under consideration:
a. hope
w
= pD
<w,t>
. [xD . x hopes w=wW: q
<w,t>
=p, q(w)=q(w),
and p(w)=1
b. x hopes p means that x hopes that w is the unique world of the current
discourse with all the characteristics of w, but also where p is true.
26
24
I also wonder how these equations survive under negation. Presumably, we need to make sure that the
negation results in rather than .
25
A potentially simpler alternative is to treat worlds as sets of statements which are true in them and
propose w=wp, but then we would have to reconsider how to appropriately relativize statements across
multiple worlds.
26
It might seem more reasonable to say that I hope you know means that I hope the world is such that
it includes an event of you knowing, but this would entail the problems mentioned in the previous footnote.
23
In general, however, I would like to adopt a notation which divides propositions up
into constituent conjuncts as much as possible, thus making it easier to see fundamental
implications and allowing us to represent the event arguments which will eventually become
necessary. -operators will bind all uninstantiated variables, and all other letter symbols
should be considered skolem constants, that is to say, items dened just by the properties
attributed to them in the formulas. Skolem constants normally function in a way equivalent
to existentially bound variables, but here it should be understood that they are allowing
us to postpone resolution of the questions raised above about the proper quantication
of these variables, their proper context, accessibility, etc. These are important questions,
which I cannot resolve at this time, but hopefully the proposals I advocate will be able to
survive their resolution in some form or another.
(69) a. hope = pxw . hope(e,w) ag(e,x) pat(e,w) p(w)
b. x hopes p means that there is an event of hoping where x was the agent, w
was the patient, and p is true in w.
Notice both e and w are like constants, or implicitly bound variables, and w is, for the
moment, dened just as being the world(s) of xs hope where ps truth is asserted. Being
the patient of hope is assumed to imply that one hopes w=w. Also, I have discarded the
assignment-function-like superscripting of world arguments, to represent them in the more
explicit, and consistent, variable-free form.
To adopt this conjunctive, event-oriented notation, I have also found it most convenient
to make use of proto-roles, like agent and patient. This step also disregards signicant
theoretical issues about their existence, character, and relation to structural congurations.
For the moment, however, they are a reasonably simple way of representing the intent of
the formulas. I will not place too much stock in their specic characters nor say much here
about how they might be derived.
Returning to the issue at hand, and how it applies to implicatives: In so far as implica-
tives have been claimed to imply the truth of their complements, it may not be clear why
they would be relativized to any world other than the matrix discourse. The crucial issue
is that their negations do not imply events in the matrix discourse. Recall that make sure
(one of the (few?) nite implicatives) implies uncertainty in its negation. To capture the
positive implication, we can add the assertion that the event implies that the complement
world is equal to the matrix world.
(70) a. make sure = pxw . make-sure(e,w) ag(e,x) pat(e,w) p(w)
(make-sure(e,w)(w=w))
b. not make-sure = pxw . make-sure(e,w) ag(e,x) pat(e,w) p(w)
(make-sure(e,w)(w=w))
Playing rather roughly with theories of negation, in (b), since there is not an event of
making sure in w, we have no idea if w=w.
24
Although I have not discussed tense, and non-nites yet, I note that the denite negative
implications can be achieved by strengenthing the conditional to a bi-conditional.
(71) a. manage = pxw . manage(e,w) ag(e,x) pat(e,w) p(x)(i)(w)
(manage(e,w)(w=w))
b. not manage = pxw . manage(e,w) ag(e,x) pat(e,w) p(x)(i)(w)
(manage(e,w)(w=w))
Since manage(e,w) is false, so is w=w. I have already argued w should be dened as
identical to w, except that p(w) is true, and therefore the failure of equality must be due
to p(w) being false. I admit the formulation of these inferences requires better precision
and eciency, but they hopefully, at least, approximate a possible resolution.
27
Up to this point, we have ignored temporal eects. Below, I will begin to remedy this
deciency.
5.1.1 Sequence-of-tense
Other researchers have claimed that embedded past nites relativize to the time of the
embedder event (Abusch (1999), Altshuler (2006), Gennari (1999), Higginbotham (2001)).
These proposals might seem to contradict the proposal here, which only supplies nites
with a world argument. In fact, the system here still allows relativization of time, but
restricts it to mediation through the world argument. This shift provides a novel approach
to well-known problems with relativization.
28
Pasts and futures are crucially distinct in
their modal/intensional representations, and this dierence is used to account for their
distinct relativization properties.
29
We can consider an LF like (b) for a sentence like (a). (c) gives an English approxi-
mation, which (d) more concisely summarizes. (e) and (f) show part of its composition, in
repeating our basic pattern for thought which merges with that Jill won.
(72) a. Jack thought that Jill won.
27
Since they, mostly, just intend to replicate Karttunens claims, with the provision that there are pred-
icates which exist in a gap between the full denite implicatives and factives.
28
Although I have claimed that implicatives, factives, and interrogatives are not specically assigned
world arguments, the same argument should hold of the discourse satisfaction that they have been argued
to undergo.
29
I will not attempt to include an explanation of the additional double access issue here, but I hope it
may fall prey to a similar analysis. In so far as the double access reading implies anything about the speech
present as a relative future of a past event, it poses a serious threat to the theory here, which implies the
impossibility of relative future modalities without future operators. However, a developed analysis of the
simple present will be required to ultimately resolve the issue. Generics and habituals have been claimed
to require intensional world modalities, and thus, it might not be surprising to nd simple present statives
licensing reference to times outside the domain of the specic, discourse world.
25
b. think(e,w) t(e)<now(w) ag(e,Jack) pat(e,w) win(e,w) t(e)<now(w ?)
ag(e,Jill) to be reconsidered
c. There was an event e of thinking, in world w, which preceded now, and Jack
was the agent
30
of e, and the patient of e is a world w such that there was an
event e of winning, in w, which preceded now in the world of Jacks thought,
and Jill was the agent of e.
d. In the world of Jacks thought, Jill won.
e. thought =Pxw . think(e,w) ag(e,x) t(e)<now(w) pat(e,w) P(w)
f. that Jill won =w . win(e,w) ag(e,Jill) t(e)<now(w) to be recon-
sidered
I assume that now(w)=t(e), since w is the patient of e. Therefore, the thinking
preceded the utterance time, and the winning preceded that thinking time. We certainly
have to admit that the winning that Jack is thinking of could not have preceded the
utterance time, and yet followed the thinking time (see the table below). That is why we
say that the past tense relativizes to the embedder time. We will see, however, that there
is a surprising complication.
Although the past seems to undergo something like this kind of relativization, the future
behaves dierently. Jack insisted that Jill will win cannot mean that Jack spoke of an
event of winning which followed his insistence, but preceded the utterance time. That is
why the future might, instead, be called absolute. Would, however, seems able to ll
in the gap. The tables below diagram my judgments about the potential timing of the
embedded event. t(think) and t(speech) are intended to indicate the thought time and the
time of asserting the complete utterance.
(73) does winning at time i, in the world of Jacks thought, guarantee the sentence is
true?
31
i: past future
t(think) t(speech)
a. Jack thought Jill won. X X
b. Jack thought Jill would win. X
c. Jack thought Jill will win. X X
The above characterization of an absolute future, and a relative past runs into problems
when examined under a future matrix clause. When embedded under a future operator,
30
Its not entirely clear to me whether the subject of thinking should be an agent or an experiencer, but
I will stress again that I am not taking these tenative roles too seriously.
31
How can we also formalize the way that futures under past eventives are so much better than futures
under past statives, as in (c)?
26
the past seems to continue its relativization to the embedder, while it now appears that
the future also relativizes. In (b), the winning cannot be future to the utterance time, but
past to the thinking time.
(74) does winning at time i, in the world of Jacks thought, guarantee the sentence is true?
i: past future
t(speech) t(think)
a. Jack will think Jill won. X
b. Jack will think Jill will win X X
Another problem for simple relativization centers around a phenomena called sequence
of tense. In the more complete diagram of the table below, we see (with !!!) that
embedded pasts can produce co-occurant readings. (76) is presented for some semblance
of completeness.
(75) does being pregnant at time i, in the world of Jacks thought, guarantee the sentence
is true?
i: past future
t(think) t(speech)
a. Jack thought Jill was pregnant !!! X X X
b. Jack thought Jill will be pregnant X X X X
(76) does being pregnant at time i, in the world of Jacks thought, guarantee the sentence
is true?
i: past future
t(speech) t(think)
a. Jack will think Jill was pregnant. X X
b. Jack will think Jill will be pregnant X X X X
Gennari (1999) and Altshuler (2006) have claimed that the simultaneous interpretation
is restricted to stative contexts. I think that this is only true if we again use the word,
stative, so broadly as to include perfects, progressives, habituals, generics, and even
activities. If someone says Jack thought the horse trotted, but Jill said it galloped, I
think it is quite likely, or at least easy to imagine, that they are making judgments that
are co-occurant with the horses movements. She might have even said something like
Cant you see that its galloping right now?! If the earlier examples lack this eect, I
27
would suggest that it has something to do with telicity and the dierent implications that
bounded or unbounded events force on potential interpretations. A full account of this
interference is another problem that I must leave to future investigation.
I suggest that the sequence-of-tense phenomenon can be resolved by allowing the em-
bedded past to relativize to the matrix time, while assuming that events in the embedded
world are undened for events occurring beyond the temporal existence of that embedded
world.
32
As Bridget Copley puts it, we adopt the common assumption that the actual
world only exists up to the time of utterance (Copley (2004), p. 12). Well, I dont really
have any assumptions about the actual world, but a strict division should be maintained
between a given discourse world and events which are future oriented to that world.
(77b) spells out this assumption a little more formally, in regards to (a), while (c) shows
that the result allows Jills pregnancy to occur at any time that is included in the world
of Jacks thought. That, of course, should include the present of his thought, and also any
previous time, since there is no restriction on a worlds past. The event of her pregnancy
is occurring, in his thought, some time prior to the utterance time, but also necessarily
within times where his thought actually existed, t(w). This utterance time is referred to
as now(SW) to indicate that it is the now of the speech world.
(77) a. Jack thought Jill was pregnant.
b. there is no e (here, the pregnancy) in w (here, Jacks thought) such that e is
dened for t(w)<t(e)
c. since t(w)<now(SW), ((t(e)<now(SW)) (t(e)t(w)))=t(w)
33
Taken too strictly, what Ive said might lead one to conclude that it is impossible to
think about the future. That, however, is clearly not the case. When a future oriented
operator is present, we immediately postulate hypothetical worlds which are dened for
future times (see Copleys work for good presentation and elaboration of interesting issues).
One might wonder why worlds of thought cant already be dened for future times, but
we are simply maintaining consistency. Would (in the sense used here) then appears to
x on the embedder worlds time, as in (78). I tentatively suggest that will, on the other
hand, apparently, must be xed relative to the maximum of any time which embeds it, as
in (79, 80). This claim seems necessary to account for its absolute nature under a past, as
32
In fact, it worries me a great deal to suggest that a single predicate like won could assert its event
in one world and its tense in another, but the data, combined with my framework, almost begs for this
explanation.
33
This argument borrows heavily from Abuschs derivation of the potential simultaneity in Barak is
predicted to be in the lead now. It is also reminiscent of Gennaris argument that sequence-of-tense
relies on a super-interval property of states. Both of them work with intersections on acceptable times.
However, both also assume that embedded events are relativized to the time of matrix events, rather than
worlds.
28
in (79, also 73c), and its relative nature under another future, as in (80, also 74b).
34
Both
would and will allow e to be dened for times greater than t(w), which is determined
by t(e).
(78) a. Jack thought Jill would win.
b. (t(e)<now(SW)) (t(w)<t(e))
(79) a. Jack thought Jill will win.
b. (t(e)<now(SW)) (now(SW)<t(e))
(80) a. Jack will think Jill will win.
b. (now(SW)<t(e)) (t(w)<t(e))
(81) a. Jack will think Jill won.
b. (now(SW)<t(e)) (t(e)<t(w))
Other approaches to the sequence-of-tense problem, have had to assume ambiguity for
tense morphemes. The approach here also seems to imply ambiguity, in so far as t(w) and
t(SW) vary for both the past (77c, 81b) and the future (79b, 80b). However, I suggested
above that the indicative future should be understood as the maximum of the world-times
which embed it. I think it may also be possible to derive the pasts behavior with the exact
same assumption (qualied by the constraint outlined in (77b)). The resolution of this
issue will certainly be contingent on how people evaluate more complex constructions:
35
(82) a. Jack said Jill will think Jane went.
b. Jack will say Jill thought Jane will go.
c. t(e) orients to t=max(t(w), t(w), ...) over all the w that embed e?
I have tried, in this section, to argue that simple world arguments can still provide for
necessary temporal relativizations. They, in fact, suggest new approaches to old problems.
Furthermore, where other systems have been indierent to the intensional type proposed
by Heim and Kratzer (1998), I have tried to derive temporal properties while maintaining
the basic type.
34
Hopefully this eect could be derived from its contrast with would, similar, perhaps, to the way that
Pancheva and von Stechow (2004) derive idiosyncrasies of the English perfect from its competition with the
past.
35
Aspectual eects also need fuller consideration in regards to this hypothesis.
29
5.2 Factives
It would be pleasantly simple if we could analyze all nite embedders as that same basic
type, but we have already seen that factives require a dierent interpretation. Furthermore,
even if extra meaning postulates could produce the extra implication, I worry that, as with
desideratives, universal quantication over a contextual set of worlds would make wrong
predictions.
(83) under consideration:
a. love
w
= pD
<w,t>
. [xD . p(w) =1 for all w W that are compatible
with what x loves in w]
b. x hopes p means that p is true in all worlds that are compatible with what x
loves in w.
I utilize the following simplied notation to explicate what I nd wrong with the above
proposal:
(84) a. Jack loves that Jill ran.
b. false: w Jack-loves(w) Jill-ran(w)
c. false: w Jill-ran(w) Jack-loves(w)
Both of the above entailments are false because Jack should still be able to love worlds
where he has a million dollars, but Jill didnt run, and hate worlds where she ran, but he
has leprosy, for instance.
Instead, I would like factives to simply refer to event arguments occurring in some
appropriate discourse world. Loves would take an event argument as complement and
that Jill ran would mean something like the current discourses unique event of Jill
running that occurred in the past.
(85) a. loves =e

yw . love(e,w) ag(e,y) nowt(e) pat(e,e)

b. that Jill ran =e

: run(e,w) ag(e,Jill) t(e)<now

36
Use of the denite operator would immediately derive the presupposition associated
with these predicates, but how these would compose together is not entirely clear. Simply
inserting the full denite denition in the pat predicate certainly would make a mess of
our formulas. It might be preferable to consider a lifting operation, but then it also seems
more appropriate to use the unique existential, and we would lose the presupposition.
36
w is presumed to already be xed by discourse, and not necessarily the same as the matrix world.
Furthermore, now is represented without any world argument, pending resolution of the issues discussed
in section 5.1.1.
30
(86) that Jill ran =P . !e

: run(e,w) ag(e,Jill) t(e)<now P(e)

The complement would then take the embedder as argument and hand it the event
argument that it is looking for. There is still the internal composition problem of deriving
the event denotation for that Jill ran. I have generally adopted an ethic of maintaining
the event argument as a constant strictly dened within the predicates lexical entry. The
e

would then have to be derived through some kind of co-referent relationship paralleling
maneuvers I have criticized in traditional binding. Thus, this ethic of strictly local denition
may have to be discarded. Furthermore, these kinds of formulations will run into even
bigger obstacles when we attempt to deal with control/raising. Resolving these issues
will require further research, and ultimately may entail radical revisions to the system
advocated here.
Additionally, we would also have to handle plural events, which I would assume to be
present in both generics and other quantied constructions.
(87) a. Jack loves that birds y.
b. Jack loves that every girl likes him.
c. Every boy loves that some girl likes him.
Also note, clause subjects also frequently fall into the factive class ((88c) is the lifted,
existential version of (88b) and produces a cleaner nal equation, but loses the presuppo-
sition eect.):
(88) a. That Jill ran pleased Jack.
b. [e . please(e,w) t(e)<now pat(e,Jack) ag(e,e)](e: run(e,w) t(e)<now
ag(e,Jill))
c. [P . !e

: run(e,w) ag(e,Jill) t(e)<now P(e)](x . please(e,w)

t(e)<now pat(e,Jack) ag(e,x))
5.3 Interrogatives
As with factives, I assume that the world argument of the complement is satised by
discourse. It is frequently presumed to be the same as the matrix discourse world, but
is not necessarily so. The selector takes a question quantier as argument. As a rough
preliminary approximation, we might use something like Karttunens proposal for questions
(1977).
(89) a. Who ran?
b. Wh x: x ran
c. {P|x P(x)=x ran=1}
31
d. {P|x P(x)=(run(e,w) t(e)<now ag(e,x))=1}
However, I am currently uncertain whether it is (90c) or (90d) that correlates better
with intuitions or any relevant tests.
(90) a. Jack asked who ran.
b. Jack asked wh x s.t. x ran
c. ask(e,w) t(e)<now ag(e,Jack) pat(e,P) {P|x P(x)=(run(e,w) t(e)<now
ag(e,x))}
d. ask(e,w) t(e)<now ag(e,Jack) pat(e,x) x s.t. {P|P(x)=(run(e,w)
t(e)<now ag(e,x))} =
I have tried in these last two subsections to acknowledge specic and important ex-
ceptions to the more general pattern I would like to describe. They present signicant
problems and suggest potential revisions, but I will have to leave these as topics for future
research.
6 P(i)(w)
Unlike the temporal complications observed in the sequence-of-tense phenomenon for nite
complements, non-nites necessarily relativize to the embedder time. However, as shown
in section 2.1, they relativize dierently for dierent selectors.
37
(92) a. Jack believes Jill to be nice.
b. Jack wants Jill to win.
While (92a) receives a co-occurrent interpretation, (92b) appears to be strongly future
oriented. However, it would be incorrect to say that (92b)s complement must have a
37
Its also clear that relativization, sometimes, depends on the kind of non-nite selected. For instance:
(91) a. Jack remembered to buy milk.
b. Jack remembered buying milk.
c. Jack likes to run.
d. Jack likes running.
e. Jack likes Jill to run.
f. (?)Jack likes Jill running.
While in (a), the remembering must have preceded the buying, and results in a factive interpretation (odd
to think that a future oriented predicate would be factive, isnt it?), in (b), the buying must have preceded
the remembering, though with a hypothetical implication (just as ironic). On the other hand, I sense no
intuitive dierence between (c) and (d), though some kind of dierence seems to arise when an overt subject
is introduced (e, f). These topics, I also have to leave to future research.
32
future interpretation.
38
(93) shows some future orienters, which also allow co-occurent
interpretations.
(93) a. Jack really wants to be on holiday right now.
b. Jill is likely to be running now.
c. Jack demands to be let in now!
I will continue to call these future oriented, in so far as they certainly allow the reading,
but they are not quite identical to those predicates which almost seem to lose all meaning
with co-occurant complements.
(94) a. ?Jack intends to be here now.
b. ?Jill hopes to be working now.
c. ?Jack expects to live in L.A. now.
Therefore, while the predicates in (94) would probably just need something like (t(e)<t(e)),
for want-like predicates, Ill use (t(e)t(e)).
39
Because the interpretation of the comple-
ment varies with the selector, I assume the temporal orientation is imposed by the selector
(see Stowell (1982), Abusch (1999) for more discussion). For instance,
(95) a. want =iPxw . want(e,w) ag(e,x) i(e) pat(e,w) P(e

. t(e)t(e))(w)
b. Jill to win =iw . win(e,w) ag(e,Jill) i(e)
Believe, on the other hand, needs something like (96a), using t(e)=t(e).
(96) a. believe =iPxw . believe(e,w) ag(e,x) i(e) pat(e,w) P(e

.
t(e)=t(e))(w)
b. Jill to be nice =iw . nice(e,w) ag(e,Jill) i(e)
Notice that since t(e)=t(e) and e is stative, and e has no other reference time than
t(e), e must also be stative. That is, it must have duration and be atelic (i.e. lacking
completion). I have not yet specically shown how these aktionsart properties would need
to be represented.
We can also consider the hypothesis, that for is only associated with t(e)t(e)/t(e)<t(e),
and thus does not appear with stative selectors.
(97) a. *Jack believes for Jill to run.
b. *Jack imagines for Jill to run.
38
Thanks to Abusch (1999) for reminding me.
39
Abuschs notation: [t,).
33
c. Jack decided for Jill to run.
Pesetsky and Torrego (2000-2006) basically claim that for is only associated with ir-
realis selecting predicates, further associating this with tense, and future orientation, as
I will discuss more fully in section 7.1. First, we need the qualication about potential
over-lapping discussed above. Furthermore, we would have to exclude factives from con-
sideration.
(98) a. Jack likes for Jill to run.
b. Jack hates for Jill to fall.
It would also depend highly on how one judged the quality and interpretation of the
predicates like the following:
(99) a. ?Jack made sure for Jill to run.
b. Jack saw t for Jill to run.
I nd (a) much worse than (b), but hesitate to classify either as truly unacceptable.
That said, I think the matrix predicates do suggest a temporal precedence, so the future
orientation of non-factive for might be tenable.
The denotations in (95) and (96), however, leave a signicant generalization unex-
plained. Why do most stative selectors allow passivization, while future orienters usually
dont?
(100) a. Jack was considered to be intelligent.
b. *Jack was wanted to be intelligent.
The behavior of these passivizers (not to be confused with object-controllers which
also passivize) is particularly odd for 2 reasons: 1.) Passives are usually problematic for
statives, as shown by (101a), and 2.) They dont even assign a -role to the passivized
subject, as shown by (101b) though that is, by denition, the nature of raisers.
(101) a. ?Jill was liked.
40
40
There is an appreciable dierence, when the by-phrase is used, and I would guess that this is exactly
because if its ability to assign range indirectly. See the discussion on locatives in Ch. 10, Borer (2005b), or
more generally, for various means of satisfying range.
(102) (a) ?Jack was liked.
(b) Jack was liked by all.
(c) ?Jill was wanted.
(d) Jill was wanted by the police.
34
b. Jack was believed to be a liar.
I believe that the key to both these puzzles lies in the concept of range assignment
(Borer (2005a), Borer (2005b)). Borer derives the telicity of unaccusatives, by proposing
they have two open values (e, in her tree, having, perhaps, a more open-ended im-
plication than just event). Non-telic transitives will also need to ll in 2 values, but
the internal one will have a non-quantity implication, and agree with non-quantity DPs.
Unergatives, on the other hand, will only need 1 (e)
E
. Notice that this system will give us
a way to talk about subjects in a way that is distinct from both and Case.
(103) Borer (2005b), p. 84, ch. 3, ex. 15a (slightly modied):
The arrows indicate the range assignments.
EP
the ower.nom
(e)
E
T
max
the ower.nom
T Asp
max
Q
the ower.nom
(e
2
)
#
VP
wilt
In fact, we can say that a necessary property of passives and unaccusatives is that
range((e)
E
)=range((e
2
)), since they were both assigned by the same argument. When I
use e to refer specically to events, as I normally do, and assuming that a predicate is just
one event (which may be a questionable assumption, but will jive more with a laymans
intuitions), I will, henceforth, generalize this passive property as r
ex
(e)=r
in
(e). That is to
say, the external range of the event is equal to its internal range. The argument that I am
going to present below will, in fact, rely on the idea that passives are specically licensed
by intuitively satisfying this condition, r
ex
(e)=r
in
(e).
Given that many languages assign dierent cases to quantity and non-quantity (i.e. of
unspecied quantity) objects, Borer argues that dierent functional projections are associ-
ated with these Cases. AspQ above is associated with accusative Case, but this accusative
Case is dependent on nominative also being assigned (Borer (2005b), p. 81). In the unac-
cusative construction above, AspQ is projected, but accusative Case is not assigned. Thus,
35
we see that we cannot reduce one to the other. For partitive objects, she still postulates a
distinct functional projection, F
s
P, (p. 109).
Id like to suggest that stative events are characterized by being in a superset relation
to their internal arguments. Id also suggest that we return to the assumption that the
objects of activities and telics are hosted in the same functional category, regardless of
their Case dierences in some languages. That will allow us to associate them both with
a relation, and derive the argument eect straight away.
41
(104) a. stative: r
in
(e) arg
in
(e)
b. activity: r
in
(e) open-arg
in
(e)
c. telic: r
in
(e) closed-arg
in
(e)
While the telicity of non-stative predicates is, frequently, dependent on their internal
arguments, this property is not characteristic of stative predicates:
(105) a. Jack ate hamburgers (*in an hour). (atelic)
b. Jack ate a hamburger (in an hour). (telic)
c. Jack likes girls (*in an hour). (atelic)
d. Jack likes a girl (*in an hour). (atelic)
The proposal in (104) captures this distinction. The same subset relation is at work,
in ate, but an argument with a closed quantity will insure that the internal range of the
event is also closed, while an argument with open, unspecied quantity will leave the issue
open. Because stative events, like like, maintain a superset relation with their internal
arguments, they are not closed either way.
I furthermore suggest that external ranges are necessarily subsets of some argument,
though whether their range becomes denitively restricted is not insured, in English.
42
(106) For some X, r
ex
(e)X
Thats why even atelics can passivize, but statives dont like to (but see fn. 40).
(107) a. Naughty students will be severely punished.
b. r
ex
(e)=r
in
(e) open-arg
in
(e)
c. ?Polite students will be really liked.
d. ?r
ex
(e)=r
in
(e) arg
in
(e)
41
It does not account for inherent telics.
42
Apparently not in other languages. See Borer, and others, for discussions on bare subject restrictions
in other languages.
36
Although, (b)s argument is open, having some -relation seems to suce. In (d), the
subject condition is not satised, and thus, polite students do not qualify as a subject of
the event.
Now, we are prepared to look at ECM passives, and other raisers. I characterize those
that passivize and those that dont as follows:
(108) a. passive: r
in
(e) = r
ex
(e) (for instance, believe)
b. passive: r
in
(e) r
ex
(e) (for instance, want)
Crucially, (a) doesnt mean a predicate is not a stative event, even if the subject of my
belief is closed. Note that it still satises the condition in (104a).
Trying to speak intuitively, it seems reasonable to say that the internal range of an
event of my belief is limited to the external range of whats believed. For instance, I know
that leprechauns dont exist, but I believe that they are stealing my socks, makes no sense
(though we should always make room for subjective contradictions). Their range is not
within my belief world, so there can be no such belief-event of sock-stealing. On the other
hand, the internal range of my want is in no way restricted by the subject. I think that
leprechauns dont exist, but I still want them to.
(109) a. Jack is believed to be smart.
b. r
ex
(e)=r
in
(e) = r
ex
(e)X
c. *Jack is wanted to be smart.
d. *r
ex
(e)=r
in
(e) r
ex
(e)X
X is whatever assigns the external range of the embedded event (in this case Jack). It
qualies as a subject of the matrix event in (109a,b) because the formula in (b) is consitent
with (106); in other words, because the event is suciently characterized. In (109c,d), the
external argument of the embedded event has no sucient relation to the matrix event.
There is an intuitive appeal to the notion that stative selectors imply r
in
(e) = r
ex
(e),
and passivize, because t(e)=t(e). However, it would be a mistake to assume that these
were identical. Though uncommon, there are future orienters, such as expect, which
also allow passivization, and raisers like likely, which also allow non- assigned subjects.
Furthermore, their passive/raised alternations are not in any way restricted to t(e)=t(e).
I also admit that I am highly tempted to associate r
in
(e) = r
ex
(e) with some greater
likelihood of e, in so far as expect is much more likely than want and, of course, likely
is more likely than possible, which does not allow raising. But then, what would I say
about probable ?
In fact, I need to claim that possible and probable have no external range. Iatridou
(1990) goes even further, claiming that possible has no tense. She oers the following
judgment:
37
(110) Iatridous judgment, p. 123, ex. 3:
#It was/will be possible/probable that John stole the tapes.
I do not agree with this judgment. I think embedding possible/probable in appropriate
circumstances gives reasonable saliency to the use of a past tense.
(111) Jack thought that it was possible/probable that John stole the tapes.
However, by claiming that complement selecting possible and probable have no exter-
nal range, we can both account for her intuition, and explain why they never take thematic
subjects in their complement taking forms.
(112) a. It is possible/probable that Jack ran.
b. It is possible/probable for Jack to run.
c. *Jack is possible/probable to run.
d. For Jack to run is possible.
Possible, though having both nite and non-nite forms, has no alternation where
r
ex
(e)=r
in
(e)=r
ex
(e).
Having an additional property beyond Case and also makes it possible for us to divide
the varying clause ellipsis properties of these kinds of predicates.
(113) a. Did Jack win?
b. He tried.
c. *He wanted.
d. *He believed. (i.e. cannot mean believed that he won)
I have argued that unsaturated s do not cause a derivation to crash. Furthermore,
Case has never, to my knowledge, really been considered an option to enforce complement
clause selection. However, now we can say that try has an alternation which projects
without internal range, while want and believe do not.
To summarize, I have proposed the following characterizations:
(114) a. For some X, r
ex
(e)X
b. passive: r
ex
(e)=r
in
(e)
c. stative: r
in
(e) arg
in
(e)
d. activity: r
in
(e) open-arg
in
(e)
e. telic: r
in
(e) closed-arg
in
(e)
f. clause selectors that can passivize: r
in
(e) = r
ex
(e)
38
g. clause selectors that cant: r
in
(e) r
ex
(e)
If these arguments are on the right track, they suggest a unied means, using Borers
independently justied mechanism, to characterize stativity, passivization, its exceptional
behavior and how that is partially related to stative selection, surprising raising failures, and
clause ellipsis. However, a more in-depth examination of the data suggests that the current
generalizations may be too broad, and the eects of tense, DP-quantity, and possibly other
factors must be given more prominence to ultimately capture the subtle shifts of linguistic
acceptability.
In this section, I have also distinguished future orienters from stative selectors, and
discussed the partial future property of for.
6.1 n-comp non-n-comp
In section 9.1, I am going to present some of the operations which Jacobson argues are at
work in binding, and which I have also assumed to be relevant to control and raising. In
so far as they relate those constructions, and those constructions appear within patterns
of lexical alternations, we will be able to see them as specically describing, in a highly
analytic sense, those lexical alternations. In that spirit, I would like to formally dene the
function that relates a nite-selector to a non-nite selector (at this point, with an overt
embedded subject; the silent subject alternations will be described when a more complete
presentation of Jacobsons theory is given in section 9.1).
I have generally assumed an ethic of xing the event as close to the predicate as possible.
Then, because the temporal relativizations above make use of (e

. t(e)=t(e)) or (e

.
t(e)t(e)), dening a totally general function to relate them runs into diculties. With
the functions in section 9.1, it seems to make little dierence whether we dene them in the
lexicon or in the syntax. Perhaps because of my potentially erroneous assumptions, I
will have to make a claim like this function must be dened in the lexicion. That is to
say, since I need to introduce another instantiation of a pre-existing constant, it appears
almost necessary, and at least most convenient, to dene this function as applying before
that constant, the event argument, is resolved.
43
(115) a. d
R
(f)=eQ . f(e)(Q(e

. t(e)Rt(e)))
b. It is likely that Jill won.
c. ePiw . likely(e,w) pat(e,w) i(e) P(w)
d. It is likely for Jill to win.
e. eQiw . likely(e,w) pat(e,w) i(e) Q(e

. t(e)t(e))(w)
43
d stands for diachronic, since all the other good letters are taken.
39
The reader is invited to ascertain that d

(c)=e
44
. When, why, and with which lexical
items, an alternation like this is permitted is the more challenging, exciting, and mysterious
question. It is my hope that precise formulations may, eventually, shed some light on, at
least, some of the underlying causes.
7 T
I have assumed that all complement innitives receive i-type arguments from their selec-
tors.
45
Various syntacticians have, on the other hand, argued that non-nites must be
distinguised by the presense or absence of T features. At least 3 recent proposals associate
these T features with both dierent interpretative results and dierent syntactic licensing.
In the 3 subsections that follow, we can observe the phenomena these researchers associate
with T, see how we can distinguish these properties, and try to see what other accounts
may be available.
7.1 C
0
Since the turn of the century, David Pesetsky and Esther Torrego have been building a
body of works which construct a theory of the that-trace eect (shown in (117)) around
the parallel behavior of tense-inversion (shown in (116)).
(116) Pesetsky and Torrego (2004), ex. 21: (Koopman 1983):
a. What a nice book Mary read !
b. What did Mary read ?
c. Who read the book?
d. *Who did read the book?/*What a nice person did read the book!
Although (116a) and (116b) have dierent meanings, they are argued to show that
tense-inversion is optional in the case of object extractions. In the case of subject ex-
tractions, (116c,d), however, tense-inversion is prohibited. The use of that in embedded
complements shows a similar asymmetry.
(117) Pesetsky and Torrego (2004), ex. 23: (Perlmutter 1971):
a. What do you think [ Mary read ] ?
44
Notice that this introduces a future oriented complement, which I would intuitively associate with
greater than. It appears that this is because I have maintained a notion where precedence is always on
the left; thus, the embedded argument is on the right, since we have cause to talk about future orientations
most. Weve also seen constructions where R is = or <, but other relations might also be available.
45
Notice that I do not touch on small clauses anywhere within this paper, as well as setting aside the
issues of -ing.
40
b. What do you think [ that Mary read ] ?
c. Who do you think [ read the book ] ?
d. *Who do you think [ that read the book ] ?
They, therefore, argue that that is just another manifestation of T-to-C movement,
and that subject movement to or through C satises a (u)T feature in C (because of its
nominative case), excluding tense movement (by economy, i.e. why move both?).
46
In the
case of object extraction, either Tense or the nominative subject also must move to C,
because the objects (u)T feature (argued for in the later development) has already been
deleted. Notice that in (117b), they then must claim that tense is phonologizing in 2 places,
like a resumptive pronoun (2000/01, p. 12).
Since there are also for-trace eects, they extend this explanation to for-clauses:
(118) a. What do you want Mary to read ?
b. What do you want for Mary to read ?
c. Who do you want to read the book?
d. *Who do you want for to read the book?
Notice that there is just as much a for-PRO eect:
(119) a. I want to read the book.
b. *I want for to read the book.
47
Thus, they require a separate account for this phenomenon and propose that the uT
feature on C is [-EPP] in a declarative clause whose subject is PRO (2000/01, p. 28). In
other words, that uT doesnt cause T-to-C movement. Neither the cause nor the means
of this dependency between the PRO subject and Cs uT are entirely clear, as they admit
(2000/01, p. 28).
Id like to modify their proposal, in some specic ways, which I hope will provide for
a more direct description of our intuitions. We want to insist that that and for always
have overt DPs directly to their right (excluding parenthetical insertion). While it appears
that for assigns accusative Case, this checking relationship is always immediately adjacent
(within the potential checking relations), as shown in (a) and (b).
46
Dont they need to assume that wh-movement precedes Tense movement, so that C knows if its uT has
been checked by the subject before moving Tense? Also notice that the subject seems to enter 2 T-checking
relations. As the theory develops, what is iT or uT or when/why features are deleted shifts. On p. 5
(2000/01), they give an example of Attraction to Spec,TP as [TP [DP subject, uT, ]i [T, u] [t-subjecti
bought the book]], clearly marking the subjects uT for deletion, and then seem to imply that delaying
deletion will play some signicant role, before switching to the closeness issue.
47
Such sentences are acceptable in Belfast English, and therefore, any theory needs to attribute some
slightly dierent status to the Belfast for. In fact, these x-ec eects are quite language specic.
41
(120) a. *Who did you want for to come?
b. *Jack wants for usually Jill to run.
c. Whom did you speak to ?
d. I usually drink coee in the morning.
e. *I drink usually coee in the morning.
Accusative case, as well as nominative, is usually assumed to be transmitted by wh-
chains, as in (c). It also appears that nominative Case allows adverbial intervention (d), and
accusative Case doesnt, as in (e).
48
However, because we need to allow for wh-movement,
we cant say that the verbs accusative feature is strong, so Ill assume that the adverb
problem is strictly prohibited because adverbs merge outside of the basic VP (as most
researchers have ultimately concluded). For-accusative, on the other hand, is strong, in
the sense that it forces an overt checking adjacency.
To be more specic, avoid confusion with morphological Case, and account for these
special properties, I propose that overt C always has a uD* feature. This feature neces-
sarily checks itself against an overt, adjacent DP on its right. With for, this also bears
[accusative], thus satisfying the embedded subject DPs case requirement, when this type
of uD feature on C gets checked. This feature, without accusative Case, applies to that,
just as much as to for, since they both imply overt, adjacent DPs (n.b. the qualication
above). In fact, it is the same feature which transitive verbs appear to bear, except that it
is not strong in the verb case, uD, thus allowing wh-movement. Presumably DPs never
loose their D feature, so there is nothing wrong with letting an object check uD for both
its transitive verb and that/for. Of course, an overt subject can also check Cs uD.
Do, on the other hand, I would argue, arises to support a uT*feature on the question
particle, in Spec, CP, which drives wh-movement. Since T on tense is undeletable, there is
nothing wrong with letting it check both the subject and the manifestation of a question
word.
Notice that I have now separated what Pesetsky and Torrego attempted to put together.
I believe that this is correct exactly because (116a) and (116b) do not mean the same thing.
Id argue that (116a) is really just a simple topicalization like Such a wonderful book
Mary read!, and should not be mixed with the issue of tense inversion which accompanies
question formation.
48
Chomsky attributes this suggestion to Stowell (while protesting). The argument was that certain word
order variations might be accounted for by adjacency conditions (Chomsky (1995)). In other words, French
and English dier, not in the Strength of Tense, but rather, the adjacency requirement on Case being at
work in just one of the languages.
(121) a. *John reads often books.
b. Jean lire souvent les livres.
42
Thus, I have argued that Cs need their own uD to be checked, while do checks Qus
uT. They appear partially identical just because overt subjects, with undeletable D, can
check Cs uD, and, when merged with Qu, allow Qu direct adjacency to an undeletable
T.
49
That is to say, manifestations of T and C have somewhat similar eects because they
either license or insure the presence of DPs at the left margin of clauses. That does not
mean they should be completely identied.
(122) Summarizing:
a. for: uD*(acc)
b. that: uD*
c. transitive verb: uD(acc)
50
d. do: T
e. Qu: uT*
51
, ...?
7.2 Independent tense and partial control
Landau (2000) proposes an entirely dierent function for T features. He observes an
extremely interesting correlation between the licensing of distinctive, embedded, temporal
adverbials and what he calls partial control.
He divides his previously discussed predicate groups into 2 classes:
52
(124) Untensed: Implicative, Aspectual, Modal
Tensed: Factive, Propositional, Desiderative, Interrogative
He argues that these tense properties are made manifest by the possibility of embedded
adverbs representing distinct time frames:
(125) a. *Yesterday, Jack tried to run tomorrow.
b. Yesterday, Jack wanted to run tomorrow.
49
In fact, Who did see John? just has the emphatic separation of tense and the predicate.
50
Perhaps, really, transmitted from little v. Also notice that given Strength and Case, uD appears
with one, the other, or both. Could neither exist? What about the other direction?
51
In fact, if we made this feature directionless, then, in embedded contexts, we could let it check against
a selecting verb on the left (that is, if we accepted Pesetsky and Torregos later theories that verbs also
bear a kind of T feature), and thus account for the lack of T-to-C movement in embedded questions.
(123) (a) Who did Jack see?
(b) I know who Jack saw.
52
In fact, many of these items are not control predicates, in English, and might require nite complements,
bare verbs, etc.
43
He claims the exact same classes correlate with the ability to license partial control.
Partial control is the ability of a singular argument to control a collective predicate:
(126) a. *Jack tried to meet.
b. Jack wanted to meet.
While the general correlation is compelling, I worry about the minor exceptions. Some-
times, the predicates shift classes in their various alternations, but still maintain the cor-
relation. For instance, dare is implicative in subject control (127a), but, perhaps more
desiderative with object control (b). Just as the theory would predict, only its object
control alternation can take temporally independent adverbs, and partial control (c,d).
Other times, the correlation, itself, is not entirely consistent. For instance, like, whether
a factive or not, cant take an independent adverb (e), while it does license collectives, (f).
The behavior of decide, as shown in (g,h), also questions the correlation.
(127) a. Jack dared to run (denitely ran)
b. Jack dared Jill to run. (maybe didnt)
c. ?Jack dared to meet tomorrow..
d. Jack dared Jill to meet tomorrow.
e. *Yesterday, I liked to meet you tomorrow.
f. Jill likes to meet on Fridays.
g. Yesterday, Jill decided to run tomorrow.
h. ?Jill decided to kiss.
More generally, dierent collectives seem to be coercible under dierent circumstances:
(128) a. She can kiss.
53
b. ?She can work together.
(129) a. Jill wanted to kiss.
b. ?Jill wanted to gather.
Landau assumes that the controlled subject (PRO) must be plural and that dierent
Agree mechanisms, based on the presence or absence of T, provide for PRO either equaling
the overt subject or merely including it. If Landau is correct, this potential lack of identity
poses a serious threat to any theory of control as either raising or binding.
Due to the murkiness of the data and the complexity of reciprocity, I will not be able
to reply with a truly formal analysis. However, I assume that there are many situations
53
Though its interesting to note how this only seems to have an ability reading.
44
where predicates are licensed to project with more or less Case features; passives, nominals,
intrinsic reexives, and non-controlled collectives are a few examples. Collective predicates
are normally licensed to appear in such an alternation, not by any process of syntactic
Agreement, but instead through pure semantic plurality.
(130) a. *The boss met.
b. The committee met.
I also suspect that those predicates which fail to license partial control are ones that
imply overlapping events, thus entailing identical times and subjects.
(131) a. Jack tried to run.
b. *Jack tried Jill to run.
c. *Yesterday, Jack tried to run tomorrow.
d. *Jack tried to meet.
e. Jack managed to run.
f. *Jack managed Jill to run.
g. *Yesterday, Jack managed to run tomorrow.
h. *Jack managed to meet.
Thus, there is some connection between the semantic-pragmatic understanding of open
participants, and leaving argument satisfaction to discourse. Specically singular argu-
ments, in a semantic sense, and inherently overlapping events disallow these vagaries.
However, even sustaining these generalizations requires excluding -ing-complements from
consideration.
(132) a. Jill has tried kissing, but she didnt really like it.
b. I avoid meeting on holidays.
-ing-complements seem to be even closer to nominals than to-innitives. That is
to say, I believe the phenomenon should be related to the way that nominalizations can
abstain from argument projection, while maintaining their implicit understanding.
(133) Williams (1987), ex. (1): (also in 1985)
a. Control: The attempt to leave
b. (attempter=leaver)
c. Binding: Respect for oneself is important
d. (respecter=respectee)
This comparison to nominals is not intended to claim that non-nites are nominaliza-
tions, but simply that they might share similar Case abstinence properties. Nominative
case (in English) is never assigned to subjects in non-nite constructions. Thus, I leave the
topic, in just such a mess.
45
7.3 Non-stativity
Martin (2001), following Chomsky and Lasnik (1993), in fact, does argue that silent control
subjects are assigned null Case. This null Case crucially distinguishes raising and control,
and is, as we should expect by now, argued to be assigned by a [+tense], non-nite T.
Raising predicates are argued to select [-tense] complements, and the [-tense] property is
said to be reected in a stativity requirement. In other words, ... eventive predicates
are impossible in raising innitivals (Martin (2001), p. 150). He oers examples like the
following:
(134) Martin (2001), p. 150, ex. 39:
a. *Geno believed Rebecca to win the game.
b. *The doctor showed Bill to take the wrong medicine.
c. *The defendant seems to the DA to steal the car.
Martin claims these ideas also account for the passivization contrast between believe
and want-type predicates, because he actually lets lexical items receive null Case, and
thus they cannot raise.
(135) a. He was believed to be intelligent.
b. *He was wanted to be intelligent.
This theory also forces Martin to recategorize predicates traditionally assumed to be
raisers. However, he admits that (136b) shows that likely and certain need not have
external arguments (p. 159).
(136) Martin (2001), p. 159, ex. 72-3:
a. Sarah is likely/certain to leave.
b. It is likely that Sarah will leave.
Still, he argues that only agreeing functional categories ... permit ellipsis of their
complement,
54
and that control predicates allow ellipsis, but not raisers, which his null
Case checking thus also accounts for. By this standard, likely and certain also appear
as control predicates.
(137) Martin (2001), p. 159, ex. 74:
a. Kim may leave but Sarah is likely/certain to [
V P
e ]
54
See his paper, p. 153, for numerous researchers working on this claim.
46
b. *Bill believes Sarah to be [
AP
honest ], and he believes [ Kim [
T
to ] [
V P
e ]] as
well.
Though the oddity of ellipsis in these stative selectors, under most circumstances,
is certainly noteworthy, Im not entirely convinced that traditional raisers never allow
ellipsis:
55
(138) a. Does he have his coat?
b. He appears to.
c. I know that Jack has his vest, and Jill appears to.
d. Does he like raisins?
e. He seems to.
f. I know that Jack likes raisins, and Jill seems to.
Likely and certain are distinct from most raisers in taking for-complements, though
these predicates adhere to the generalization proposed in this paper that non-factive for
is always future-oriented:
(139) a. It is likely for Jill to win.
b. It is certain for Jill to win.
c. *It seems for Jill to be nice.
I have already argued that control and raising share fundamental chain mechanisms,
so I have no intention of bickering about predicate classications in those terms. The
data above shows that there is not an exact correlation between stativity and -less Case
satisfaction, which are predicate properties I consider signicant. In section 6, I treated
stative selection as a lexical fact, and considered an alternative account for its association
with passivization.
8 Case
In this section, I continue defending the traditional proposal, and apparent fact, that
nominative case is checked in the presence of nite tense, and that silent subjects are
licensed in non-nite clauses specically by its absence. I will furthermore argue against
distinguishing the EPP from Case. In the following subsection, I question the arguments
which originally motivated the theory of null Case. Given that I argue for a complete
lack of projection, I clearly have no use for null Case.
55
How signicant is prosody? Does it change across varying temporal relations? Can it account for the
unacceptable sentences which have been advanced?
47
8.1 Chomsky and Lasnik (1993)
Chomsky and Lasnik (1993) discuss three environments where the distribution of PRO
cannot be accounted for by Caselessness (p. 555). These environments can be characterized
as, 1.) adverb interventions, 2.) adjective environments, 3.) nominative environments. I
will attempt to qualify each of these observations in turn.
Most importantly, notice that its not entirely clear how null Case would actually resolve
any of the problems discussed below. This point returns to the one discussed by Adger
in section 2.1. In each case, there is a problem with why silent subject innitives (with
dierent interpretations) can or cant merge with dierent elements. The problem, in each
environment, is external to the innitive clause, but they then draw the conclusion that an
internal modication (PRO Case checking with non-nite tense) will resolve it. However,
if Case is simply checked and deleted, an external element really should be indierent to
the presence or absence of PRO. The external element should have no way of Agreeing or
selecting innitive phrases with or without null Case, since these Case properties should
be invisible by the time of the external merger. The real problem lies in the assumption
that PRO even exists, and once we remove this assumption, we can begin to consider real
solutions to some of the following problems.
8.1.1 Adverb interventions
As they observe on p. 557, Case generally conforms to an adjacency requirement, (in
English, and I believe we should add: at least for accusative Case). They oer the following
examples:
(140) Chomsky and Lasnik (1993), p. 557, ex. 279:
a. Bill sincerely believed Sam.
b. *Bill believed sincerely Sam.
The logic extends to ECM constructions, which they discuss on p.555:
(141) Chomsky and Lasnik (1993), p. 555, ex. 253:
*John believes sincerely [Mary to be clever].
They argue that if PRO merely avoids Case, the following sentence should be possible.
(142) Chomsky and Lasnik (1993), p. 555, ex. 254:
*John believes sincerely [PRO to be clever].
I simply argue that believe must check accusative Case (as does Pesetsky (1991)).
There is nothing wrong with the internal composition of the bracketed phrase. It is the
48
external composition which causes the derivation to crash. No one would be silly enough
to argue that any and all constructions containing a Caseless PRO should automatically
converge. The current situation really is no dierent than the construction without the
adverb.
(143) a. John believes her to be clever.
b. *John believes (PRO) to be clever.
In (b), regardless of adverb intervention, believes causes a crash whether PRO does
or not. Predicates that possess non-accusative alternations do permit the construction.
(144) a. Jack wanted intensely to win.
b. Jack hoped sincerely to be accepted.
8.1.2 Adjective Caselessness
Chomsky and Lasnik (1993) use the following examples to illustrate the failure of adjectival
Caselessness in accounting for the distribution of PRO.
(145) Chomsky and Lasnik (1993), p. 555, ex. 255-6:
a. *It is likely [PRO to solve the problem].
b. It is important [PRO to solve the problem].
The (a) example is confusing in a couple of ways. For one, it is important to under-
stand the meaning conveyed by PRO, which, I think, is intended to convey an arbitrary
interpretation, as in the (b) example. There is nothing wrong with the overt string itself.
(146) Thanks for the advice. Its likely to solve the problem.
In fact, most theories claim that complements can not license arbitrary control (see
Landau (2000), Manzini (1983), Manzini (1983), Mohanan (1983), Mohanan (1985), Bres-
nan (1982), Borer (1989)). Therefore, the more important question is why important fails
to either control (a) or raise (a =b).
(147) a. *John is important to go.
b. It is important for John to go.
As I argued in section 6, we can also distinguish these predicates with the concept of
range. Although important has nite and non-nite selecting forms, it is only capable
of assigning one range. Likely, though it lacks an external -role, can project external
range, which is determined by the external argument of its internal complement. In fact,
most speakers believe that when it takes a silent subject complement, it must be specied
for this external range.
56
56
I have become less and less comfortable with treating this as a fact of the English language. I have no
problem with the following sentences:
49
8.1.3 Nominative environments
Chomsky and Lasnik (1993), additionally point out that nominal environments have also
been argued to lack Case, and thus should also allow PRO-innitivals.
(149) Chomsky and Lasnik (1993), p. 555, ex. 259-60:
a. *My belief [Harry to be intelligent]
b. *My belief [PRO to be intelligent]
I again claim the silent subject is fully licensed, and furthermore, that all nouns can
take innitival (relative) clause adjuncts, but there are often problems with the thematic,
selectional properties of the items involved. The question is not why PRO cant appear
(because it, or rather, a silent subject can) but why it doesnt produce a reading of believ-
ing in my intelligence. I argue that the following sentences show innitival adjuncts are
available to nouns.
(150) a. The man to call lives in LA now.
b. The method to solve this problem hasnt yet been discovered.
The question remains why, or how, dierent nominals allow dierent thematic assign-
ments.
(151) from Pesetsky and Torrego (2006), ex. 10,11:
a. Marys desire to win
b. Marys need to intervene
c. *Sues love to solve problems
d. *Johns hatred to have to leave
Pesetsky and Torrego (2006) claim that nominals of irrealis selectors dier from those
which select innitival complements without the characteristic semantics of for-clauses
(i.e. factive and implicative clauses) (p. 5).
57
They claim that this is due to the way
that the nominal head and an irrealis complement can check each others T features. I
believe that they are on the right track, but question if this basic generalization can really
be sustained.
(152) a. Jacks failure to attend the meeting resulted in his dismissal.
(148) a. Do you think its likely to win the war in Iraq?
b. Its pretty likely to get killed over there.
57
Of course, weve observed that factives do allow for-clauses, but they are not necessarily future-oriented.
50
b. Jacks ability to resolve problems resulted in his promotion.
Therefore, we still lack a truly formal and consistent way of representing the crucial
properties. I believe, however, that the type-driven system advocated here, at least, pro-
vides formal, axiomatic mechanisms which should be useful in their nal characterization.
For now, Ill just say some nominalizations retain their verbal thematic frames, and others
dont.
8.2 EPP: movement to Caseless positions
An EPP-feature forcing movement to Caseless positions has been argued to be necessary
to derive the cyclic movement that is assumed in sentences like the following.
(153) Jack seemed t to be likely t to win.
t is in a Caseless position, so the traditional theory requires some other feature to force
movement there. Movement to this intermediate position is said to be necessary in so far
as long leaps are not possible. Examples like (154) are taken to provide evidence for this
assumption.
(154) *Jack seemed it was likely t to win.
Manzini and Roussou (2000) argue that this is an unfortunate consequence of the
category theory of A-movement, and replace this theory, rst, with a theory of features,
as in (155a, without any intermediate pitstop), and then even more radically, with a bare
theory of Attract, as indicated by the italics in (155b). They claim that the latter is
consistent with Hale and Keysers (1993) conguration-based theory of predicate-argument
assignment, and argue that the Minimal Link Condition, as applied to Attract, will insure
that sentences like (154) will not be derived.
(155) a. John -I [ seems [ to work () ] ] (Manzini and Roussou (2000), ex. 18)
b. [ John I [ seems [ to work ] ] ] (Manzini and Roussou (2000), ex. 45)
c. *John seems [ that it was told [ that Mary left ] ] (Manzini and Roussou (2000),
ex. 47)
According to their theory, John, in (b), attracts both seems and work. In (c), they
argue that John also attracts seems, but the interpretive requirements of John cannot
be satised by a predicate like seem, and there is no way to attract told, because told
is in the immediate scope of it (p. 430). In more traditional terms, the derivation crashes
because John has no -role.
Thus, they actually replace cyclicity with Attract Closet/MLC. As shown in (28),
the variable-free system that I am advocating represents an even stricter notion of cyclicity,
51
by imposing the representation of potential bindees at the margin of every intermediate
node. The system being developed here is much more explicit, and elaborate, than the
system advocated by Manzini and Roussou (2000). In that sense, if the simple equations
above turned out to an eective means to accurate predictions, their system would certainly
be preferable. However, their system makes no comment about the potential candidates
for merger, determined by selectional restrictions, and derives no association with binding
(which some might consider a good thing). Therefore, I presently maintain the current
methods, even for all the complexity that they impose on the system.
In the system being advocated here, we can easily formalize the intuition behind At-
tract Closest, by imposing a rule of Immediate Match. While pronouns often skip past
potential binders, silent subjects, generally do not.
(156) a. Every boy
1
thought he
2
liked him
1
.
b. Every boy
1
thought he
2
wanted to win. (denitely not every boy winning)
In section 9.1, we will see that similar, though not identical, operations derive both
chains. We should assume that operations licensed by overt pronouns have the ability to
continue passing, even where type matches are available. By assuming that silent subject
chains lack this ability to skip matchers, we can immediately derive the results of Attract
Closest, as it applies to silent subject interpretations.
58
Because there are other means available to account for cyclic intervention eects, there
is no need to postulate an EPP that is distinct from Case. Furthermore, as discussed in the
previous subsection, I dont believe null Case can help shed any light on silent subjects and
the selection of clauses that they are embedded in. Thus, we should maintain the traditional
generalization that nite clauses must check nominative Case, and non-nites simply dont;
that is to say, there is no Case distinction between raising and control complements. I now
turn to my own analysis of silent subject selectors.
9 P(x)(i)(w)
I claim that an innitive complement with a silent subject, at least one which is ultimately
bound by a higher element, necessarily requires 3 arguments: the subject reference, a tem-
poral reference, and a world of interpretation. I assume the following order of satisfaction
so that an identical denotation of to win can be used in Jack wants to win and Jack
wants Jill to win.
(157) a. to win =xiw . win(e,w) ag(e,x) i(e)
58
This comment may need to be qualied to deal with the interaction of arbitrary control constructions
and binding. I will spare the reader the large number of relevant, but highly debatable issues, until a later
date.
52
b. wants =Pxw . want(e,w) ag(e,x) nowt(e) pat(e,w) P(x)(e

.
t(e)t(e))(w)
c. Jill wants to win =w . want(e,w) ag(e,Jill) nowt(e) pat(e,w)
win(e,w) ag(e,Jill) t(e)t(e)
In fact, I have made no assumptions about any kind of verb-raising, and since tense
usually appears directly on the verb stem, I have assumed that verbs usually take their
interval argument rst. Even in the case of present and past participles, we can assume that
this interval argument is satised by some temporal relativization function which would
then put out a new interval argument seeking satisfaction. I will not formally present this
analysis of participles here, but we should consider how patient arguments will merge in
the case of bare verbs, such as those under to, should, etc.. Under such circumstances,
there will be a type mismatch and the object will have to pass the other arguments, over
itself to repair it. Thus, Jill would have to take on the denotation in (b).
(158) a. kiss =ixyw . kiss(e,w) ag(e,y) pat(e,x) i(e)
b. Jill =Piyw . P(i)(Jill)(y)(w)
c. kiss Jill =iyw . kiss(e,w) ag(e,y) pat(e,Jill) i(e)
Furthermore, since to is assumed to merge with predicates seeking temporal satis-
faction (as in 159a), and yet, both postpone this satisfaction, and create a phrase which
would be able to merge immediately with a subject (as in 157a, repeated as 159b), we can
attribute to to a denotation as in (c).
(159) a. win =ixw . win(e,w) i(e) ag(e,x)
b. to win =xiw . win(e,w) i(e) ag(e,x)
c. to =Pxiw . P(i)(x)(w)
9.1 Binding: g & z, Jacobson (1999b)
Pauline Jacobson denes 3 fundamental operations which are crucial to binding in a
variable-free system.
The rst allows certain mismatches to be resolved, by passing the interfering element
up over the selecting function. In other words, suppose that I am a function which takes
elements of type a and returns elements of type b. That means I am a function of type
< a, b >. What if I encounter an element of type < c, a >? Then, the Geach rule (g)
allows me to shift my functional denition in such a way that I can take elements of type
< c, a > and return an element of type < c, b >. Thus, I shift to an element of type
<< c, a >, < c, b >>.
When c is an argument that has been introduced by an overt pronoun, we will let this
operation apply freely, to pass the argument up just as far as any speaker can remember
53
what the hell that they are actually talking about, but in other cases, it becomes an
extremely important theoretical question when and where this operation is licensed or
restricted.
Love, in the example below, would normally take an argument of type x, but, by
shifting with the g rule, it is able to take an argument of type < x, x >. Jacobson uses a
rather reduced notion, which I simply copy for the moment. See footnote 60 and the end
of section 2.2 for more comment.
(160) The Geach Rule: Passes a variable up over.
a. g
c
(f)=P[x[f(P(x))]]
b. < a, b >, << c, a >, < c, b >>
c. love
g
( her )=
d. P[x[love(P(x))]](y[y])
59
=
e. x[love(x)]
60
Jacobsons reduced notation doesnt necessarily make it clear that if we g over a predi-
cate with an argument slot, we actually end up with an additional slot. The next operation
is necessary to bind 2 slots together, by identifying the item that has caused the mismatch
with the argument that the function normally would receive next.
(161) Jacobsons binding rule: Identies the next arg up as the embeddees desired arg.
(see Jacobson (1999b), p. 134, ex. 23, for the properly generalized version; here we
have the introductory version).
a. z
b
(f)=P[x[f(P(x))(x)]]
b. < a, < e, b >>, << e, a >, < e, b >>
c. love
z
( herself )=
d. P[x[love(P(x))(x)]](y[y])=
e. x[love(x)(x)]
It is important to note that these operations only apply to functions, and not arguments.
If we want to pass an argument over an argument, then we must rst lift the latter argument
before applying g. I already introduced this operation in section 2.1. As noted there, it
also denes a relationship between entities and quantiers.
59
Following Jacobson, we here treat pronouns as a basic identity function from individuals to individuals,
and abstract from gender, number, etc.. Note that [y[y]](Jill)=Jill.
60
Jacobson adopts this kind of combinatory semantics without much discussion of the implication on
the congurational-lexicalist debates. Recall that, for her, x[love(x)]=love. Remember that I usually put
more informative value in the presence or absence of these operators.
54
(162) Lift: Are all subjects lifted? g requires a lifted subject.
a. l
b
(a)=F[F(a)]
b. a, << a, b >, b >
c. Jill
l
( smokes )=
d. F[F(Jill)](x[x smokes])
An attentive reader may have caught the scent of a subtle inconsistency. I suggested
that g was for resolving type mismatches, but here suggested that lifting is useful to pass
arguments over arguments. As mentioned in section 8.2, we should be able to derive the
eects of Attract Closest by not allowing this kind of skipping for variables related to
specic chains. The behavior of overt pronouns, on the other hand, seems to require such
a capability.
We are now prepared to observe that the following lexical alternation is nothing but an
application of z. However, we return to the intentional notation that I have also advocated.
We see that expect
2
=z(expect
1
). That is to say, (d) is z applied to (b).
(163)
61
a. Jill expected Jack to win.
b. expected
1
=P
<i,<w,t>>
xw . expect(e) t(e)<now(w) ex(e,x) pat(e,w)
P(e

. t(e)t(e))(w)
c. Jill expected to win.
d. expected
2
=P
<e,<i,<w,t>>>
xw . expect(e) t(e)<now(w) ex(e,x)
pat(e,w) P(x)(e

. t(e)t(e))(w)
e. z(expected
1
) =P
<e,<i,<w,t>>>
xw . expect(e) t(e)<now(w) ex(e,x)
pat(e,w) P(x)(e

. t(e)t(e))(w)
(164) shows the work a little more explicitly:
(164) z(expect):
a. P[Q[y[P(Q(y))(y)]]](R
<i,<w,t>>
xw . expect(e) t(e)<now(w) ex(e,x)
pat(e,w) R(e

. t(e)t(e))(w))=
b. Q[y[w . expect(e) t(e)<now(w) ex(e,y) pat(e,w) Q(y)(e

. t(e)t(e))(w))]]
For convenience, I repeat examples of the kinds of complements each function could,
respectively, take as an argument.
61
Notice that I am applying this operation in the syntax, after tense has been assigned, so that I can
claim that it is type-driven. This might be a little goofy, and it may, ultimately make more sense to consider
it a lexical operation. The formulas would only require minor shifts.
55
(165) a. Jack to win =iw . win(e,w) i(e) ag(e,Jack)
b. to win =xiw . win(e,w) i(e) ag(e,x)
There really is nothing especially profound about this relationship, in so far as I, with
pretty much all semanticists, have maintained that binding, control, and raising are funda-
mentally identical operations. However, it is nice to be able to name a specically dened,
and more generally applicable operation as a fundamental component of a common lexical
alternation. Of course, I have also acknowledged that accusative Case must be varied as
well, in appearing in the former version, and being discarded where z, as licensed by control
predicates, applies.
9.2 Reexivity
Naming the operation in this way, furthermore, allows us to shed light on a debate which
raged throughout the GB era: the anaphoric (i.e. reexive) nature of control predicates.
Although attempts were made to reduce the behavior of PRO to that of a silent reexive
(Manzini (1983)), just as NP-trace was already assumed to be, these eorts were also
met with vehement criticism (Mohanan (1985)), while other inventive researchers found
reasonable middle grounds (Borer (1989)). Within the present system, I dont even claim
that there is any projected element which is or isnt reexive. Rather, both reexives and
control can be seen as distinct syntactic-lexical
62
environments licensing applications of z.
Following Reinhart and Reuland (1993), I assume that overt reexives merely serve to
mark their predicates as reexive. Furthermore, some predicates are intrinsically reex-
ive, and some predicates alternate (p. 666). Applying these comments to the alternations
of predicates like shave or embarrass, we see further evidence of applications of z which
license the discarding of accusative Case. However, it should be clear that this ability is
somehow restricted to specic lexical items.
(166) (representing just the relevant elements)
a. Jill shaved Jack.
b. shave =xy . shave(e,w) ag(e,y) pat(e,x), [acc]
c. Jack shaved.
d. shave =y . shave(e,w) ag(e,y) pat(e,y)
9.3 Raising
Crucially, however, raising predicates do not allow applications of z. Raisers do not have
argument slots which are identied with the embedded element being passed up. They
never returned < e, b > to begin with. Trying to apply z to them would result in a kind of
62
As noted, I slipped in the perhaps-overly-radical position that these shifts occur after insertion.
56
over-saturated formula. We would have doubled our argument, and it would expecting
an additional -role. In that sense, we can say that such an operation would violate the
-principle. Therefore, raisers must be restricted to applications of g.
I assume non-nite selecting likely is future-oriented (in the present-inclusive sense).
(167) a. It was likely for Jill to win.
b. likely =Piw . likely(e,w) pat(e,w) i(e) P(e

. t(e)t(e))(w)
63
c. Jill was likely to win.
d. g(likely) =Pxiw . likely(e,w) pat(e,w) i(e) P(x)(e

. t(e)t(e))(w)
Notice that to use (d) well then need to lift and g was, because of the type mismatch
with x.
(168) a. g(likely) to win =xiw . likely(e,w) pat(e,w) i(e) win(e,w)
ag(e,x) t(e)t(e)
b. was =e . t(e)<now
c. l(was) =P . P(e . t(e)now)
d. g(l(was)) =Qx . Q(x)(e . t(e)now)
e. g(l(was)) g(likely) to win =xw . likely(e,w) pat(e,w) t(e)<now
win(e,w) ag(e,x) t(e)t(e)
Of course, most raisers dont even allow for-clauses anyways.
(169) a. *It seemed (for) Jill to win.
b. * seem =iPw . seem(e,w) pat(e,w) i(e) P(i)(w)
c. Jill seemed to win.
d. seem =iPxw . seem(e,w) pat(e,w) i(e) P(x)(i)(w)
I have argued that the for-clause is excluded because it must be either factive or future-
oriented. However, I have not oered an explicit, formal characterization of for, which
would properly reect this impossibility. This issue awaits full resolution.
I have also suggested that seem lacks an alternation without for because seem cannot
check accusative Case. Whether there is any deeper, underlying cause for this character
also remains an open question.
63
Here, I assume that the adjective merges with its complement before merging with tense.
57
10 Landaus critique of Hornstein
As with Hornstein (1999) and Manzini and Roussou (2000), I have basically reduced control
and raising to essentially identical chains, though just above, I oered a formal description
of the specic dierence of these chains. In fact, that dierence really was no more nor
less than the dierence of a single -role, i.e. an argument slot, which basically everybody
already knew. I have diered from Hornstein in essentially ignoring even the possibility
of category-movement for A-like chains (see Manzini and Roussou (2000) for supporting
arguments), and also argued for a more elaborate system than the one oered by Manzini
and Roussou (2000).
Thus, though our theories dier in signicant ways, we are all equally susceptible to
the scathing critique which Landau (2000) launches on Hornstein (1999). In an eort to
give full consideration to the issue, I will briey present the important distinctions which
Landau notes, and at least try to suggest that the current account has not given up all
hope of deriving them.
Landau (2003) points out some 6 dierences between raising and control which he
claims Hornstein (1999), mostly, fails to account for. They may be listed as follows:
(170)
raising control
expletives/idioms X
complementizers X
unaccusative tests X
clause-superior each X
Case concord X
partial control X
I have already touched on the topic of partial control in section 7.2, so I will not
repeat myself here. Landau additionally claims that Hornsteins theory would overgenerate
passive constructions, and I have also previously considered this issue, in section 6. Below,
I consider each of the remaining issues in turn.
10.1 Expletives and Idioms
As Landau admits, this is the one (p. 487) distinction which Hornstein points out, and
it is easily handled by the claim that the matrix control predicate assigns a -role which
expletives are unable to receive. I also believe that a range assignment analysis might be
more appropriate, but have yet to work out the application to there-constructions.
64
64
Legalizing surviving -operators at the end of a derivation seems to me to basically violate one half
of the -principle. In fact, I take it for granted that arbitrary control, nominalizations, passives, and the
prolic use of silent arguments in other languages all show that the traditional interpretation of the -
principle is not maintainable. We can still say that all DPs must receive -roles, but we cannot say that all
-roles are necessarily assigned. Instead we must look at how Case and range interact with -assignments
to insure argument projection, or license its absence.
58
Idiom chunks also break if one of their components is separately assigned a -role.
(171) Landau (2003), p. 487, ex. 32 (rearranged):
a. There seems to be a man in the garden.
b. *There expects to be a man in the garden.
c. The shit seems to have hit the fan.
d. *The shit expects to hit the fan.
10.2 Complementizers
Landau (2003) oers the following claim, presumably a universal generalization (p. 488).
(172) Landau (2003), p. 488, ex. 33:
Control complements may be introduced by complementizers; raising complements
are never introduced by complementizers.
He cites de in French, di in Italian, om in Dutch, att in Swedish, aD in Icelandic, me
in Hebrew, for in Belfast English, i in Welsh, and so on (p. 488).
First, this collection of items should not be too hastily grouped with the declarative
English complementizers, for and that. For and that always introduce overt subjects.
Many of these items never do. In that sense, these items often look more like to than
for. Notice that Landau, following SigurDsson, later even glosses Icelandic aD as to,
as shown in (177) (though I am not familiar enough with Icelandic to know why PRO is
positioned below it). Since it appears with another complementizer, til, above it, glossed
as for, Landau must be suggesting that Icelandic is a double C language. I do not know
if this is correct or not. Of course, to does appear with raisers.
Im a little afraid that the reasoning that derives this universal generalization is
circular. Traditional GB theory concluded that control complements were CPs and raisers
were not. Since all of these items introduce control complements, they must be Cs. On
the other hand, items which link raisers, such as English to, would automatically fail to
classify as complementizers.
Landau seems to nd it ridiculous that adding arguments would contribute to adding
words, asking why should the presence of an external -feature on the matrix verb license
a CP projection in the complement? (p. 489).
65
Since Im not trying to account for
CP projections, the notion that links are more likely to appear between predicates with
independent participants does not strike me as so shocking.
65
Is such a proposal really any more ridiculous than the standard interpretation of Burzios generalization
which suggests that an external -feature is necessary to license accusative Case and the projection of an
internal argument.
59
10.3 Unaccusatives
Landau presents 2 pieces of data which seem to indicate that raisers are unaccusative
and controllers are not. Licensing en is said to indicate unaccusativity, and si rejects
it. Landau couches his discussion in a theory of A-movement for passives, unaccusatives,
impersonals, and raisers. However, even in movement theories, all of these are also con-
ditioned by the absence of external and/or agentive arguments. In that sense, there is
nothing surprising about them patterning together to the exclusion of control which is
dened by its external, agentive argument. Thus, the results of these tests also could be
reduced to the dierence of the single, additional -role. That said, I do not understand
all the full intricacies of these phenomena, so this is another topic which I have to place on
my already crowded backburner. I provide Landaus examples, for the interested reader.
10.3.1 en
(173) Landau (2003), p. 490, ex. 35 (Ruwet (1972)?):
a. Le directeur du departement semble etre accepte.
the head of-the department seems to-be accepted
The head of the department seems to be accepted.
b. Le directeur semble en etre accepte.
(174) Landau (2003), p. 490, ex. 36:
a. Le directeur du departement esp` pere etre accepte.
the head of-the department hopes to-be accepted
The head of the department hopes to be accepted.
b. *Le directeur esp`ere en etre accepte.
10.3.2 si
(175) Landau (2003), p. 490, ex. 37 (Rizzi (1986)?):
a. *I due candidati
1
si
1
risutavano [t
1
poter vicere].
the two candidates to-each-other appeared to-be-able to-win
The two candidates appeared to each other to be able to win.
b. I due concorrenti
1
si
1
sono promessi [di PRO
1
essere leali].
the two competitors to-each-other were promised di to-be loyal
The two competitors promised to each other to be loyal.
60
10.4 Each
Landau also cites Burzios argument that control interfers with the link between each and
elements that it is intended to quantify over.
(176) Landau (2003), p. 491, ex. 38 (Burzio (1981)?):
a. One interpreter
1
each was assigned t
1
to the visiting diplomats.
b. One interpreter
1
each seemed [t
1
to have been assigned t
1
to the visiting diplo-
mats].
c. *One interpreter
1
each tried [PRO
1
to be assigned t
1
to the visiting diplomats].
d. *One interpreter
1
each said [he
1
had been assigned t
1
to the visiting diplomats].
Eventually I would like to develop a theory of distributivity that focuses on events,
and the plurality of the arguments assigned to them. With the subject-raisers, each is
not in the domain of a new argument assignment, but with the controllers it is. Thus, the
controllers intervene and the raisers dont. I again lack a fully formal characterization, but
I hope such an analysis would be possible.
Thus, I have again argued that the introduction of additional arguments should make
all the dierence.
10.5 Case concord
Icelandic quirky Case is argued to demonstrate both the raising/control distinction and
the existence of a Case-marked PRO (SigurDsson (1991)). The Case of PRO is assumed to
account for the Case mismatch between the accusative all and the nominative boys, in
(a). Crucially, there is no mismatch in the raising Case, (b).
(177) Landau (2003), p. 492, ex. 40a, 41a (SigurDsson (1991)?):
a. Str akarnir vonast til [aD PRO vanta ekki alla skolann].
the-boys.nom hope for to PRO.acc lack not all.acc in the-school
The boys hope not to be all absent from school.
b. Str akarna virDast [t vanta ekki alla skolann].
the-boys.acc seem to-lack not all.acc in the-school
The boys seem not to be all absent from school.
Although I do not know enough about Icelandic, I would argue that Icelandic Case is
heavily associated with -roles. Clearly, there are 2 Cases in (a) because there are 2 roles,
and there is only 1 Case in (b) because there is only 1. The boys get both -roles in (a),
but they manifest the Case of the closer one. In (b), they manifest the only thing theyve
got. Hopefully, the oated quatiers Case can also be accounted for without PRO.
61
In this section, I have suggested that additional -roles should be considered signicant
enough factors to bear on all the interesting data which Landau has collected, and which
I have repeated above.
11 Additional overt/covert dierences
Having generally collapsed binding, raising, and control, in so far as they are chains, I
must also recognize that there are signicant dierences between overt pronouns and silent
arguments. Below, I take note of two important distinctions which have been discussed
in the literature. The explanations for both should rely on the fact that overt pronouns
have a greater range of interpretations. Whereas silent arguments are often licensed by
specic, lexical identications, pronouns can appeal to discourse referents at any time.
Furthermore, overt pronouns appear to license choice-function interpretations which allow
a greater breadth in their resolution of identity.
11.1 De se/re
De se readings insist that the participant have knowledge of the referent specied in his
intensional world. De re readings allow co-reference even where the participant is unaware
of it.
(178) a. The amnesiac expected that he would receive a medal.
b. Every amnesiac expected that he would receive a medal.
c. The amnesiac expected himself to receive a medal.
d. Every amnesiac expected himself to receive a medal.
e. The amnesiac expected to receive a medal.
f. Every amnesiac expected to receive a medal.
Even the bound pronoun (b,d) and the reexive (c,d) have been claimed to allow the
de re readings. Thus, it is claimed that the amnesiac can fail to recognize himself as
the recipient, although he is making an observation about someone who turns out to be
himself. Such a reading is denied for the control cases (e,f), i.e. the silent subject cases. I
nd this interpretation of the reexive highly questionable, but since I will have to admit
the reexives licensing of strict readings, it is probably in my own interest to maintain the
overt/covert signicance.
Getting this reading in the bound cases, would seem to require identifying a relationship
from individuals to individuals: the person who one is, without knowing. That is to say,
it seems to require a choice function. Thus, I propose, simply, that overt pronouns have
choice function interpretations and silent arguments dont.
66
66
I dont believe that I am the rst to propose this idea, but I have yet to locate specic references.
62
11.2 Missing strict readings
It has been noted that the overt pronouns, as in (a) and (b), license either strict or sloppy
readings of the ellipsis site, whereas the silent subject in (c) only allows the sloppy reading.
The strict reading interprets the argument of the ellipsis site as identical to the argument
in the overt representation, i.e. Jack. The sloppy reading interprets the argument of the
ellipsis site as the argument in its own sentence, i.e. Joe.
(179) a. Jack expects that he will win, and Joe does too.
b. Jack expects himself to win, and Joe does too.
c. Jack expects to win, and Joe does too.
While other accounts have had to proliferate indices to account for this behavior (for
instance, Heim (1998)), such explanations are not available to the variable-free system
advocated here.
67
Since Attract Closest/Immediate Match applies to covert chains, there is only the
sloppy reading, in (c), i.e. Joe to win, and no strict reading. With, (a), wed expect
that any discourse reference is available, so not only is the strict reading, i.e. Jack to
win, available, but also some identical other person that they both think will win. The
odd sentence is (b), since discourse should license accidental co-reference, rather than
a reexive. Isnt there a preference for the sloppy reading ? In any case, the silent
replacement of the reexive still seems to have a choice function-like ability to return a
dierent individual for the individual binding it. A more explicit analysis could perhaps
even correlate with the precise features that undergo vehicle change (Fiengo and May
(1994)).
I have not oered a full characterization of choice functions here, but see Reinhart
(1997), regarding their necessity.
12 Conclusion
Numerous questions remain unanswered. How can we account for the erractic selection
and interpretation of to vs. -ing ? How many meanings does that have?
Numerous phenomena have been overlooked. Small clauses and subjunctive selection
were ignored. Complications introduced by the interpretation of arbitrary silent subjects
were omitted. Purpose clauses, logophors, super-equi, and split antecedence were also left
out.
This paper still maintains a number of radical positions and proposes a number of
original ideas. All hierarchal conguration should be analyzed as type-driven function ap-
plication, and syntactic argument projection should be accounted for with a combination
67
I also think a discourse appeal would be illegitimate, because the discourse appeal needs to license
accidental co-reference, and wouldnt license the reexive in the rst place.
63
of Case and range. All binding is represented in the literal notations, and syntactic modi-
cations to assignment functions are avoided. The sequence-of-tense phenomenon is strictly
restricted to world relativization. Range assignment accounts for stativity, passivization,
raising, and clause ellipsis. Lexical alternations can be identied as generalized formulas.
For is only partially future oriented, and to may be an argument shifter. The that-
trace, for-trace, and for-PRO eects are unied, and tense inversion is distinguished.
The proposed absence of projection denied null Case, and replaced the EPP with a combi-
nation of Case and range assignment. z showed reexivity and (obligatory) control to be
distinct manifestations of the same operation. I have tried to consider the many properties
which distinguish raising and control. The additional properties of overt pronouns may
derive from the combination of discourse access and choice function binding.
I have not necessarily made any negative predictions, but I have tried to work on tools
which can provide for precise, linguistic description.
References
Abusch, Dorit (1999) On the Temporal Composition of Innitives, To appear in The
Syntax of Time, J. Gueron and J. LaCarme (eds), Cambridge: MIT Press, 2004.
Adger, David (2003) Core Syntax, Oxford, Oxford.
Allen, Cynthia L. (1984) On the Dating of Raised Empty Subjects in English, Linguistic
Inquiry, 15(3), 461465.
Altshuler, Daniel (2006) Sequence of tense and double access, Manuscript, Rutgers Uni-
versity.
Bach, Emmon (1979) Control in Montague Grammar, Linguistic Inquiry, 10, 515531.
Baker, Mark, Johnson, Kyle, and Roberts, Ian (1989) Passive Arguments Raised, Lin-
guistic Inquiry, 20(2), 219251.
Battistella, Edwin (1985) On the Distribution of PRO in Chinese, Natural Language and
Linguistic Theory, 3, 417340.
Boeckx, Cedric (2001) Scope Reconstruction and A-Movement, Natural Language and
Linguistic Theory, 19, 503548.
Boeckx, Cedric and Hornstein, Norbert (2004) Movement under Control, Linguistic In-
quiry, 35(3), 431452.
Borer, Hagit (1989) Anaphoric Agr, in The Null Subject Parameter, pp. 69109, Kluwer,
Drudrecht.
64
Borer, Hagit (2005a) Structuring Sense Volume I: In Name Only, Oxford University Press,
Oxford.
Borer, Hagit (2005b) Structuring Sense Volume II: The Normal Course of Events, Oxford
University Press, Oxford.
Bouchard, Denis (1983) On the content of empty categories, Foris, Dordrect.
Bresnan, Joan (1972/9) Theory of Complementation in English Syntax, Ph.D. thesis, MIT.
Bresnan, Joan (1982) Control and Complementation, Linguistic Inquiry, 13(3), 343434.
Brody, Michael (1985) On the Complementary Distribution of Empty Categories, Lin-
guistic Inquiry, 16(4), 505546.
Brody, Michael (1999) Relating syntactic elements, Unpublished.
Chierchia, Gennaro (1984) Topics in the Syntax and Semantics of Innitives and Gerunds,
Ph.D. thesis, UMass, Amherst.
Chierchia, Gennaro (1985) Formal Semantics and the Grammar of Predication, Linguis-
tic Inquiry, 16, 417443.
Chierchia, Gennaro (1992) Functional WH and Weak Crossover, in Bates, D., ed., WC-
CFL 10.
Chomsky, Noam (1995) Bare Phrase Structure, in Webelhuth, Gert, ed., Government
and Binding Theory and the Minimalist Program, Blackwell, Cambridge, USA.
Chomsky, Noam and Lasnik, Howard (1993) The Theory of Principles and Parameters,
in Jacobs, Joachim, von Stechow, Arnim, Sternefeld, Wolfgang, and Vannemann, Theo,
eds., Syntax: An International Handbook of Contemporary Research, pp. 506569, Wal-
ter de Gruyter, Berlin, reprinted in Noam Chomsky (1995), The Minimalist Program,
Cambridge, Mass.: MIT Press, 13127.
Copley, Bridget (2004) Aspect and Scope in Future Conditionals, .
En c, M urvet (1987) Anchoring conditions for tense, Linguistic Inquiry, 18(4), 633657.
Epstein, Samuel David (1984) Quantier-pro and the LF Representation of PRO, Lin-
guistic Inquiry, 15(3), 499505.
Fiengo, Robert and May, Robert C. (1994) Indices and Identity, MIT Press, Cambridge,
MA.
65
Gennari, Silvia (1999) Tense, aktionsart, and sequence of tense, in Corblin, F., Dobrovie-
Sorin, C., and Mandarin, J., eds., Empirical Issues in Formal Syntax and Semantics 2,
pp. 309332, Peter Lang, Berne France.
Heim, Irene (1998) Anaphora and Semantic Interpretation: A Reinterpretation of Rein-
harts Approach, MITWPL, 25, 205246.
Heim, Irene and Kratzer, Angelika (1998) Semantics in Generative Grammar, Blackwell,
Malden, MA.
Higginbotham, James (2001) Why is Sequence of Tense Obligatory? Manuscript, USC
University.
Hirai, Daisuke (2004) Control Innitives and Two Types of CP Phases, English Linguis-
tics, 21, 241264.
Hornstein, Norbert (1999) Movement and Control, Linguistic Inquiry, 30(1), 6996.
Huang, C. T. James (1989) pro-Drop in Chinese: A Generalized Control Theory, in The
Null Subject Parameter, pp. 185214, Kluwer, Drudrecht.
Huang, C.T. James (1984) On the Distribution and Reference of Empty Pronouns, Lin-
guistic Inquiry, 15(4), 531574.
Iatridou, S., Anagnostopoulou, E., and Izvorski, R. (2001) Observations about the Form
and Meaning of the Perfect, in Kenstowicz, M., ed., Ken Hale: A Life in Language, pp.
189238, MIT Press, re-printed in A. Alexiadou, M. Rathert, and A. von Stechow (eds.)
(2003) Perfect Explorations. Mouton de Gruyter.
Iatridou, Sabine (1990) The Past, the Possible, and the Evident, Linguistic Inquiry, 21,
123129.
Jacobson, Pauline (1999a) Binding without pronouns (and pronouns without binding),
To appear in R. Oehrle and G-J. Krui (eds.), Binding and Resource Sensitivity, Kluwer
Academic Press.
Jacobson, Pauline (1999b) Towards a Variable-Free Semantics, Linguistics and Philoso-
phy, 22, 117184.
Jacobson, Pauline (2000) Paycheck Pronouns, Bach-Peters Sentences, and Variable-Free
Semantics, Natural Language Semantic, 8(2), 77155.
Karttunen, Lauri (1970) On the semantics of complement sentences, in CLS 6, pp. 328
339.
Karttunen, Lauri (1971) Implicative Verbs, Language, 47, 340358.
66
Kiparsky, Paul and Kiparsky, Carol (1971) Fact, in Steinberg, D. and Jakobovits, L.,
eds., Semantics, pp. 143173, Cambridge.
Koster, Jan (1984) On Binding and Control, Linguistic Inquiry, 15(3), 417459.
Landau, Idan (2000) Elements of Control, Kluwer, Dordrecht.
Landau, Idan (2001) Control and Extraposition: The Case of Super-Equi, Natural Lan-
guage and Linguistic Theory, 19, 109152.
Landau, Idan (2003) Movement Out of Control, Linguistic Inquiry, 34(3), 471498.
Landau, Idan (2004) The Scale of Finiteness and the Calculus of Control, Natural Lan-
guage and Linguistic Theory, 22, 811877.
Larson, Richard (1991) Promise and the Theory of Control, Linguistic Inquiry, 22, 103
139.
Lasnik, Howard (1992) Two notes on control and binding, in Larson, Richard, Iatridou,
Sabine, Lahiri, Utpal, and Higginbotham, James, eds., Control and grammar, pp. 235
251, Kluwer, Dordrect.
Lasnik, Howard (2006) A Family of Questions, USC, lecture HO.
Lasnik, Howard and Saito, Mamoru (1991) On the subject of innitives, in Dobrin,
L., Nichols, L., and Rodriguez, R., eds., CLS 27, pp. 324343, University of Chicago,
Chicago.
Manzini, Rita M. (1983) On Control and Control Theory, Linguistic Inquiry, 14, 421446.
Manzini, Rita M. and Roussou, Anna (2000) A Minimalist Theory of A-movement and
Control, Lingua, 110(6), 409447.
Manzini, Rita M. and Savoia, Leonardo M (1997) Null subjects without pro, UCLWPL,
9.
Martin, Roger (2001) Null Case and the Distribution of PRO, Linguistic Inquiry, 32(1),
141166.
May, Robert (1985) Logical Form, MIT Press, Cambridge, MA.
Mohanan, K. P. (1983) Functional and Anaphoric Control, Linguistic Inquiry, 14, 641
674.
Mohanan, K. P. (1985) Remarks on Control and Control Theory, Linguistic Inquiry,
16(4), 637648.
67
Pancheva, Roumyana (2003) The Aspectual Makeup of Perfect Participles and the In-
terpretations of the Perfect, in Perfect Explorations, pp. 277306, Mouton de Gruyter,
(Ms. USC, 2002).
Pancheva, Roumyana (2004) Another Perfect Puzzle, in WCCFL 23, pp. 621634, Cas-
cadilla Press.
Pancheva, Roumyana and von Stechow, Armin (2004) On the Present Perfect Puzzle, in
Proceedings of NELS 34, Stony Brook University.
Perlmutter, David M. (1970) The two verbs begin, in Readings in transformational gram-
mar, pp. 10719, Ginn and Co., Waltham, MA.
Pesetsky, David (1987) Wh-in-Situ: Movement and Unselective Binding, in Reuland,
Eric and ter Meulen, Alice, eds., The Representation of (In)deniteness, pp. 98129,
MIT Press, Cambridge, Mass.
Pesetsky, David (1991) Zero Syntax II: Innitives, Unpublished MS.
Pesetsky, David and Torrego, Esther (2001) T-to-C Movement: Causes and Conse-
quences, [Final version in Michael Kenstowicz (ed.), Ken Hale: A Life on Language.
Cambridge, MA: MIT Press. (2001)].
Pesetsky, David and Torrego, Esther (2002) Tense, case, and the nature of syntactic
categories, [Final version in The Syntax of Time, J. Guron and J. Lecarme (eds.), MIT
Press 2004].
Pesetsky, David and Torrego, Esther (2004) The Syntax of Valuation and the Inter-
pretability of Features, To appear in Clever and Right: a Festschrift for Joe Emonds
(S. Karimi, V. Samiian and W.Wilkins, eds).
Pesetsky, David and Torrego, Esther (2006) Probes, Goals and Syntactic Categories, To
appear in Proceedings of the Seventh annual Tokyo Conference on Psycholinguistics (Y.
Otsu, ed.).
Reinhart, T. (1997) Quantier Scope: How labor is divided between QR and choice
functions, Linguistics and Philosophy, 20, 335397.
Reinhart, Tanya (1983) Anaphora and Semantic Interpretation, The University of Chicago,
Chicago, IL.
Reinhart, Tanya and Reuland, Eric (1993) Reexivity, Linguistic Inquiry, 24, 657720.
Roeper, Thomas (1987) Implicit Arguments and the Head-Complement Relation, Lin-
guistic Inquiry, 18, 267310.
68
Rosenbaum, Peter S. (1967) The Grammar of English Predicate Complement Construc-
tions, MIT, Cambridge, MA.
Schein, Barry (1997) Conjunction Reduction Redux, Ms.
Schein, Barry (2002) Events and the semantic content of thematic relations, in Logical
Form & Language, pp. 263344, Oxford University Press, Oxford.
Schein, Barry (2003) Adverbial, descriptive reciprocals, in Hawthorne, John, ed., Lan-
guage & Philosophical Linguistics, Philosophical Perspectives, vol. 17.1, pp. 333367,
abridged in Rachel Hastings, Brendan Jackson & Zsoa Zvolenszky, eds., Proceedings of
Semantics and Linguistic Theory XI. Ithaca: CLC Publications, Department of Linguis-
tics, Cornell University, 2001.
SigurDsson, Halldor

Armann (1991) Icelandic Case-Marked PRO and the Licensing of
Lexical Arguments, Natural Language and Linguistic Theory, 9, 327363.
Stowell, Tim (1982) The Tense of Innitives, Linguistic Inquiry, 13, 561570.
Stowell, Tim (1995) What do the present and past tenses mean? in Bertinetto, P.,
Bianchi, V., Higginbotham, J., and Squartini, M., eds., Temporal Reference, Aspect, and
Actionality: Semantic and Syntactic Perspectives, vol. 1, pp. 381396, Rosenberg and
Sellier, Torino.
Stowell, Tim (2004) Tense and Modals, in Gueron, J. and Lecarme, J., eds., The Syntax
of Time, MIT.
Tanaka, Hidekazu (2004) Raising to Object out of CP, Linguistic Inquiry, 33(4), 637652.
Williams, Edwin (1980) Predication, Linguistic Inquiry, 11, 203238.
Williams, Edwin (1985) PRO and Subject of NP, Natural Language and Linguistic The-
ory, 3, 297315.
Williams, Edwin (1987) Implicit Arguments, The Binding Theory, and Control, Natural
Language and Linguistic Theory, 5, 151180.
Williams, Edwin S. (1977) Discourse and Logical Form, Linguistic Inquiry, 8, 101139.
Wurmbrand, Susi (2001) Innitives: Restructuring and Clause Structure, Walter de
Gruyter, Berlin.
Wurmbrand, Susi (2005) Tense in innitives, Plenary speaker at the
workshop on New Horizons in the Grammar of Raising and Control,
LSA Summer Institute, Cambridge, Mass, July 2005. [last available at
http://depts.washington.edu/lingweb/events/wcc25/abstracts/Wurmbrand.html].
69