Raising Constructions

How raising works

Some verbs and adjectives — seem, appear, happen, tend, be likely — take infinitival complements but assign no thematic role to their subject. They predicate over the whole event, not over a participant in it. So the matrix subject position is structurally required (English requires a subject in finite clauses — the Extended Projection Principle, EPP) but semantically empty.

The grammar fills the empty slot by moving an argument up from the embedded clause. In John seems to leave, John originates as the subject of leave (where he gets the leaver role) and raises to matrix subject:

Underlying:    [ ___  seems [ John  to leave ]]
After raising: [ John  seems [ <t>  to leave ]]

The trace <t> marks where John used to be. He gets his thematic role from leave, his case from the matrix tense, and his surface position from the EPP-driven movement. He is — strictly speaking — a syntactic argument of the embedded clause that has migrated to the matrix.

This is called A-movement (argument movement, to an argument position), the same kind of movement that operates in the passive. It contrasts with A-bar movement, the movement that builds questions and relative clauses (where arguments move to non-argument positions like Spec-CP).

Raising vs control — the central contrast

On the surface, raising and control look identical:

(a) John seems to leave.    [RAISING]
(b) John tried to leave.    [CONTROL]

Both have a matrix verb, an infinitival complement, and a single overt noun phrase. But the underlying structures are completely different. In raising, John has only one thematic role (leaver). In control, John has two (trier and leaver — with the embedded leaver role going to a silent PRO that John controls).

	Raising (seem, appear)	Control (try, want)
θ-roles for surface subject	One — assigned in embedded clause	Two — one matrix, one to PRO
Matrix verb assigns role?	No	Yes
Idiomatic subject possible?	Yes — "The cat seems to be out of the bag"	No — *"The cat tried to be out of the bag"
Expletive subject possible?	Yes — "It seems to be raining"	No — *"It tried to be raining"
Selectional restrictions on subject	None from matrix; embedded verb constrains	Matrix verb constrains (sentient, agent, etc.)
Embedded passive preserves meaning?	Yes — "Mary seems to be liked by John" ≈ "John seems to like Mary"	No — meaning shifts
Empty embedded subject	Trace of moved NP	PRO (silent pronoun)

The diagnostics are sharp because they exploit the theta-role difference. Try requires its subject to be a sentient agent capable of trying — that's incompatible with idioms (whose pieces have no independent referents) and with meaningless expletives. Seem requires nothing of its subject; whatever the embedded clause needs is fine.

The three classic diagnostics in detail

1. The idiom test. English has idioms whose pieces don't refer independently — "the cat is out of the bag" means "the secret is revealed," and the cat doesn't refer to a real cat. Try inserting the idiom under a candidate verb:

The cat seems to be out of the bag.   ✓ idiom preserved (RAISING)
The cat tried to be out of the bag.   ✗ idiom broken (CONTROL — try assigns a real role to "the cat")

2. The expletive test. Weather predicates (rain, snow, drizzle) take meaningless expletive subjects ("it"). Try embedding under a candidate verb:

It seems to be raining.   ✓ (RAISING)
It tried to be raining.   ✗ (CONTROL — try needs a real subject; "it" is meaningless)

3. The embedded-passive test. Compare an active embedded clause with its passive counterpart:

John seems to have hit Mary.    ≈    Mary seems to have been hit by John.   ✓ same meaning (RAISING)
John tried to hit Mary.         ≠    Mary tried to have been hit by John.    ✗ meaning shifts (CONTROL — trying changes hands)

All three diagnostics align for any given verb. Seem, appear, happen, tend pass all three raising tests; try, want, hope, persuade fail all three (they pass control tests instead).

Raising-to-object and ECM

The second flavor of raising involves the embedded subject ending up as the matrix object (or being case-marked by the matrix verb without moving):

I believe John to be honest.
I expect Mary to leave early.
I want him to call.

In each, the noun phrase after the matrix verb (John, Mary, him) is interpreted as the subject of the embedded clause but receives accusative case from the matrix verb. This is called Exceptional Case Marking (ECM) because the matrix verb assigns case across a clause boundary — exceptional in that case-marking normally happens within a clause.

Postal (1974) called this raising-to-object: the embedded subject raises into the matrix clause as an object. Chomsky (1981) reanalyzed it as case-assignment-without-movement (the noun phrase stays in the embedded clause, case is assigned across). Modern minimalist syntax has gone back and forth. What's uncontroversial is that John in "I believe John to be honest" is interpreted as the subject of be honest (he is the honest one, not the believer), gets accusative case (use him: "I believe him to be honest"), and behaves syntactically as a matrix object for some tests (passivization, binding) and as an embedded subject for others.

Worked example: classifying ten verbs

Apply the diagnostics to each candidate verb. ✓ means the test favors raising; ✗ means it favors control.

Verb	Idiom test	Expletive test	Verdict
seem	✓ (cat ok)	✓ (it ok)	Raising
appear	✓	✓	Raising
tend	✓	✓	Raising
happen	✓	✓	Raising
be likely	✓	✓	Raising
try	✗	✗	Control
want	✗	✗	Control (but ECM with object)
persuade	✗	✗	Object control
believe	(N/A — takes object)	"It is believed to be raining" ✓	ECM (raising-to-object)
force	✗	✗	Object control

The diagnostics pattern cleanly. Verbs that pass all raising tests assign no role to their surface subject; verbs that fail them do.

Variants and related cases

• Adjectival raising. "John is likely to leave" — likely is a raising adjective, like certain, sure, apt, prone, bound. Same A-movement story.
• Hyper-raising. Some languages (Brazilian Portuguese, some African languages) allow raising out of a finite clause, which most languages disallow. A topic of active research.
• Raising of expletives. "There seems to be a problem" — the expletive there raises from the embedded clause, leaving its associate noun phrase ("a problem") behind.
• Multiple-clause raising chains. "John seems to be likely to tend to leave" — three nested raising verbs, John raises the whole way up.
• Tough movement. "John is tough to please" — superficially looks like raising (John is interpreted as the object of please), but is now usually analyzed as a different mechanism (null operator + base-generation), distinct from A-movement raising.
• Cross-clausal binding. Raising chains feed binding: "John₁ seems to himself₁ to be honest" — the trace position lets himself find its antecedent.
• Modal-like raising verbs. "John is going to leave," "John is about to leave" — semi-grammaticalized raising structures that have evolved into future markers in many languages.

Why raising matters

• A-movement theory. Raising and passive are the two paradigm cases of A-movement; the theory of subject promotion centers on them.
• EPP and clause structure. The Extended Projection Principle (every clause has a subject) is the engine driving raising; without it, the empty matrix subject would not need filling.
• Semantics-syntax interface. Raising clearly separates surface position from thematic role — a textbook case for arguing that surface form underdetermines meaning.
• Acquisition. Children acquire raising verbs (especially seem) later than control verbs and sometimes treat them as control through pre-school years; the eventual reanalysis is a much-studied developmental milestone.
• Aphasia. Patients with Broca's aphasia struggle with movement-derived structures; raising sentences are a probe alongside passives.
• Computational parsing. NLP systems must distinguish raising from control to assign correct logical-form arguments — "John seems to leave" should give leave(John), not seem(John, leave).

Common pitfalls

• Treating raising verbs as control verbs. Children do this until late in development; adult learners often do too. The diagnostics matter.
• Assuming idiom-survival is the only test. Some idioms are partial or have edge cases; combine the idiom, expletive, and embedded-passive tests for confidence.
• Confusing raising with passive. Both are A-movement, but passive moves a patient over a suppressed agent; raising moves an embedded argument over a thematically empty matrix predicate.
• Calling all infinitival complements raising. Most aren't — control is far more common with verbs like try, want, decide. Run the diagnostics rather than assuming.
• Misanalyzing ECM as control. "I believe John to be honest" — John is not a controller of PRO, he's the embedded subject. ECM verbs allow expletives ("I believe it to be raining"), control verbs of the same surface shape do not.
• Forgetting that want is special. "I want to leave" is control; "I want him to leave" is ECM (raising-to-object). Same verb, different structures depending on whether the embedded subject is overt.
• Mixing up tough-movement with raising. "John is easy to please" looks like raising but isn't a textbook A-movement structure — most theories now treat tough as a separate construction with a null operator.