Raising vs.

Subject control predicates

Introduction
1. a. John is likely to leave.
b. John is reluctant to leave.
What are the theta grids of these predicates? Are they the same?

Is likely selects one internal argument <proposition>

Is reluctant selects two arguments: one external argument <experiencer> and one internal argument
<proposition>.

This means that John receives the theta role of experiencer and to leave receives theta role of proposition.
ask students:
However, are is reluctant and is likely the only predicates present in those examples? What’s the DS of each of
them?
2.a. DS: __ is likely John to leave.

We also have the predicate leave, with the following theta grid: leave selects an Agent.
What is this theta role assigned to? It also appears to be assigned to the DP John but what’s the problem with
that?

2. b. DS: John is reluctant John to leave

The DP “John” receives two theta roles.

The theta criterion only allows one theta role per DP. This sentence seems to be a violation of the theta
criterion, as its subject DP gets two theta roles. How do we solve this problem? The theta criterion says that
there must be a one-to-one mapping between the number of theta roles and the number of arguments in a
sentence and this sentence has three theta roles (agent, experiencer, and proposition), but only two arguments.

Actually, there is a third DP here (getting the surplus agent theta role); a DP that it is not pronounced. This DP
argument is called PRO. PRO only appears in the subject positions of non-finite clauses.
(Carnie, 2009. p 400)
So, can we now write the DS of the second sentence?
2.c. John is reluctant PRO to leave

1
This shows that the two sentences presented in (1) are different constructions with respect to the semantic
properties of the subject position of the main clause.

1. a. John is likely to leave. RAISING CONSTRUCTION

b. John is reluctant to leave. CONTROL CONSTRUCTION

1. Semantic and syntactic properties.

The main predicate in a raising construction does not assign an external theta role (the spec position of the TP
is empty at D-structure). The subject of the embedded clause is caseless and raises to spec position to get Case.

DS: ____ is likely John-k to leave.

SS: Johni is likely ti to leave.

In control constructions the main clause predicate does assign an external argument. There is no raising; the
external theta role of the embedded predicate is assigned to a null caseless PRO.

a. Johnexp is reluctant PROagent to leave.  Control predicate

Now let’s pay a look at the following sentence:

b. *John is reluctant Peter to leave.  Why is this ungrammatical?

Why do we need PRO? So as not to violate the theta Criterion
What’s the difference between (b) and (c) below:
c. John believed Peter to leave early. “Believe” assigns acc k to the DP “Peter” - ECM

Control vs. raising constructions

As superficially raising and control predicates look similar, we can apply some tests to distinguish raising from
control predicates.
The most reliable way to distinguish raising constructions from control constructions is to work out the theta
grids associated with the matrix predicates.
2
By means of these tests, we will see that if the matrix predicate assigns an external theta role then it is not a
raising construction.

3. a. Jane is likely to dance.

b. Jane is reluctant to dance.

How can we test this?

If we contrast the role of Jane in these sentences, we can see that in (b) is reluctant is a property we attribute to
Jane. But in (a), there is nothing about Jane that is likely. Instead what is likely is Jane’s dancing.

2. Semantic Test to distinguish raising vs. control predicates

a. Idioms
One test that works well to show this difference is the behaviour of idioms, for instance, the idiom the cat is out
of the bag which means the secret is widely known. This expression gets its idiomatic meaning when the
expression is a whole. When it is broken up, it gets the literal interpretation: the feline is out of the sack.
Let’s see the following examples and their interpretations.

4. a. The cat is likely to be out of the bag. (idiomatic meaning)

b. The cat is eager to be out of the bag. (non-idiomatic meaning)

Which one shows the idiomatic meaning?

If D-structure is the level at which we interpret idiomatic meaning, then we should get idiomatic
meanings with raising constructions. With control constructions, on the other hand, the subject of the main
clause is never in the embedded clause, so we don’t expect to get idiomatic readings.
So, we can use this test to distinguish between raising and control predicates. If you get an idiomatic
interpretation, then raising is involved.

3. Syntactic test to distinguish raising and control predicates

When we look at these predicates superficially, they look alike but we now they are different. Let’s
summarise the semantic and syntactic characteristics of them:

RAISING PREDICATES CONTROL PREDICATES

3
 - Do not select an external argument - They impose selectional restrictions, they select an
- Do not assign a theta role to an external argument. external argument.
- They select an internal argument <proposition> - They select an <experiencer>
- The subject of the non-finite clause proposition is a - Subject of the non-finite clause is a null caseless
trace. PRO.

Raising structures don’t select an external argument, (therefore, they don’t impose selectional restrictions on
the type of subject) or assign a theta-role to an external argument. They select an internal argument which
receives the theta role of proposition. The subject of this non-finite clause is a trace. Expletives can be inserted
in this position. We can have any type of phrase as subject. Raising predicates include verbs and adjectives. On
the other hand, control predicates impose selectional restrictions and do not allow expletives.

So, to answer the usual question about whether a given verb which selects an apparently subjectless
infinitive complement is a raising predicate or a control predicate, we can explain the syntactic
differences between the 2 classes of predicate by means of the following tests:

A. EXPLETIVE INSERTION.
A test we can use to distinguish between raising and control constructions is to see if they allow expletive
insertion. Where are expletives allowed?
Expletives are only allowed in non-thematic positions, which are the hallmark of raising.
One difference between the two is that raising verbs like seem allow a dummy subject like expletive there
whereas control verbs like try do not.

1. There seemed/*tried to be someone living there.

2. There seemed/*tried to be no milk in the fridge.

The derivation of the seem-sentences is straightforward: there originates as the subject of be, and then is raised
in a successive cyclic fashion to become first the subject of to, then the subject of seems and finally the subject
of the abstract T constituent preceding seem.

Why doesn’t a verb like try allow expletive there as its subject: why can’t it have a structure such as the
one below?

3. *There tried to PRO be a strike.

4
One reason is that PRO is a referential pronoun which takes its reference from its controller; hence, the
controller of PRO must be a referential expression. Since the expletive there is a nonreferential pronoun, it
cannot serve as the controller of PRO and hence cannot serve as the subject of a control predicate like try which
selects an infinitive complement with a PRO subject.

So, consider the following sentences:

Thematic considerations:

It is a thematic property of the verb try that it assigns the theta role AGENT to its subject, hence, it
requires as its subject an animate entity. Conversely, the fact that seem allows an expletive in subject
position suggests that it does not theta mark its subject.

a. John seems to have worked hard. It seems that John has worked hard. Expletives are possible in this
position because it is – role; seem doesn’t select an external argument vs * It tried to rain.
b. The kettle seems to be boiling.
c. * The kettle tried / has decided to boil. Not an agent
d. It seems to be raining.
e. Your theoryi seems ti to be foolproof. We can have any type of subject because this is a –role
position. “Foolproof” assigns a role to “your theory” then it moves to get case.
f. Johni is likely ti to pass the exam. “John” gets a role from the verb “pass” and then it moves to get
case. its final position is – role.
g. It is likely that I will pass the exam.
h. The table is likely to break.
i. It is likely to snow / rain.
j. Johni is willing PROi to pass the exam. “willing” < agent, proposition >
k. * Your theory is willing to be foolproof.
l. * The theory tried to be foolproof. Not an agent.

4. Two kinds of control predicates

6. a. Jeani is reluctant [PROi to leave]

b. Jean persuaded Roberti [PROi to leave]

5
There are two major control constructions. (a) is what is called a subject control, because the
subject DP of the main clause is co-referential with PRO. (b) is an object control construction,
where the main clause object is co-referential with PRO.

What are the theta grids of these two sentences?

Is reluctant: <exp,prop>

Leave <agent>

Persuade <agent, theme, proposition>

Leave <agent>

In 6. b. the DP Robert is theta marked by persuade. So in order not to violate the theta criterion
we have to propose a null PRO to take the agent theta role of leave.

PRO is not to be identified with trace, even though the two share certain properties.
PRO is distinct from an NP TRACE, at least insofar as the theta criterion is concerned. In
fact PRO will count as a lexical NP for the theta criterion. PRO is generated at DS
whereas traces arise as a consequence of movement in the derivation of SS.

PRO (pronominals, pronouns) is an empty category, it isn’t pronounced but it has an interpretation. The
difference between PRO and pronouns is that the latter have phonetic content and the former does not.
PRO is a non-overt NP. The EPP can be satisfied by non-overt material. The PP, TC and the EPP offer
arguments for postulating PRO. It is syntactically represented and it saturates a theta-role to satisfy the
Theta Criterion and the PP. Depending on the context PRO may be taken to refer to a specific referent or it
may be interpreted as equivalent to the arbitrary pronoun one. In a control structure, PRO picks up the
agreement features of the controller. PRO is like an anaphor; it’s dependent on another NP for its
interpretation. PRO is an anaphor with the feature matrix [+ anaphoric, + pronominal]. When PRO is
6
interpreted as referentially dependent on another NP, we say it’s controlled by that NP. (indicated by
coindexation) PRO can also have an arbitrary reading: arbitrary PRO. PROarb refers to people in general, it
has a generic interpretation. It doesn’t have an antecedent in the same clause but within the discourse. It’s
equivalent to the arbitrary pronoun (one, people). PROarb is singular and it can have either third person or
second person features. We can also talk about subject or object control. Control predicates include verbs
and adjectives.

It can be obligatory (when it needs an antecedent for its interpretation) or optional (when the antecedent
can be in the same clause or in the discourse). Control structures select an external argument and assign it a
theta-role. The subject of the non-finite clause complement is PRO. It doesn’t allow expletives and imposes
selectional restrictions on the subject.

PRO Theorem: PRO appears in caseless ungoverned positions. PRO is admitted or licensed when it is in
ungoverned positions. If an NP is not governed, then it will not have a GC. This entails that PRO and
overt NPs are in complementary distribution: where overt NPS are allowed, PRO is excluded and the
other way round. Overt NPs must be case-marked and case is assigned under government. Without a CP
projection, PRO will be governed by an external governor, specially the finite inflection of the higher
clause, which is a governor. CP is a barrier to government, being a maximal projection. PRO must be
free and bound in its MGC. Solution: NPs with the features [+ anaphor, + pron.] will survive if they are
ungoverned. The fact that PRO must appear in ungoverned positions explains why it cannot appear in
these positions:

PRO can’t appear as:

a. An internal argument: * Peter wonders whether to invite PRO.

b. Subject of a finite clause: * Peter wonders whether PRO should invite.
c. Subject of a non-finite clause in ECM contexts: * Peter believed PRO to be a good student.
d. Subject of a non-finite clause introduced by “for”: * Peter prefers very much for PRO to destroy the
evidence.

In the case of obligatory control, the controller must c-command the controlled element. The c-command
requirement distinguishes obligatory form optional control. It’s not possible for PRO to be controlled by an
expletive there in the higher clause. Control by an expletive is not allowed. PRO must be controlled by an
argument. When PRO is the subject of a declarative complement clause it must be controlled by a NP.
Arbitrary PRO is excluded. When the complement clause is infinitival, subject control or arbitrary control is

7
possible. John asked CP [how to behave himself / oneself]. Verbs of subject control: the DO isn’t required as
a controller. Verbs of object control; the DO must be present in order to control the subject of the
infinitival clause. PRO as the subject of adjunct clauses is also obligatorily controlled. In subject clause PRO
is not obligatorily controlled.

EXAMPLES:

1. Johni tried PROi / arbj to believe himselfi / oneselfj

2. Johni was reluctant PROi / arb j to leave himselfi / oneself j.
3. Johni it was important PROi / arb to believe himselfi / oneself arb.
4. Johni asked how PROi / arb to behave oneself arb / himselfi.

a. expletive insertion

b. selectional restrictions

c. passivization

d. that-clause paraphrase.