You are on page 1of 16

17 April 2000 Andrew Kaplan-Myrth

The Movement Rule Formerly Known as Richard

0 Introduction
In this paper, I am interested in English sentences of the following form:

(1) John seems like he is ill

These sentences are of special interest because of their relationship to general processes of
raising. When these sentences were first studied by Postal (Postal, 1971) and Rogers (Rogers,
1971; Rogers, 1973), a movement rule was sought that would raise the subject of a clause headed
by like to the matrix subject position leaving behind a copy. Rogers calls this raising rule
“Richard”, but it has also been known as “doubling”, “subject-copying”, and “copy-raising”.

According to the raising analysis, (1) and (2) both seem to be related to (3) in similar ways:

(2) John seems to be ill


(3) John is ill

The standard GB analysis for (2) is that it is derived from the DS given in (4) by the process of
raising.

(4) [IP seems [IP to be [SC John ill ]]]]

In GB theory, ill assigns a theta-role to John in the Small Clause. John then moves to the specifier
position of the embedded phrase, seeking Case. This A-movement results in (5):

(5) [IP seems [IP Johni [I’ to be [SC ti ill ]]]]

This specifier position, however, is not a Case position – it is not head governed since the
embedded IP is [-finite], and since seem is not an ECM verb John cannot get ECM Case. John must
therefore move to a higher position to have Case assigned. In the Minimalist framework, John
must move to a higher A-position in order for its Case features to be checked – otherwise, the
derivation will crash. In both frameworks, this movement results in the observed word order:

(6) [IP Johni [I’ seems [IP ti [I’ to be [SC ti ill ]]]]]

One crucial difference between (1) and (2) is the tense of the embedded phrase. Specifically, (1)
has a [+tense] phrase and (2) has a [-tense] phrase. How does this difference affect the
derivation? To answer this, consider (7), a sentence that differs from (2) only in the [tense]
feature of the embedded phrase:

(7) It seems John is ill

The accepted derivation for (7) is the same as that for (2), except the movement from the
embedded subject position to the matrix subject position is not motivated by Case requirements
-2- Andrew Kaplan-Myrth

since John can get Case (or, in the Minimalist framework, can have its Case features checked)
from the [+tense] T (or I) head. We therefore get a structure like (8).

(8) [IP seems [CP Ø [IP Johni [I’ is [SC ti ill ]]]]

Finally, an expletive subject it is required in the matrix subject position – this is required by the
Extended Projection Principle (EPP). In short, it appears that when the embedded phrase is
[+tense], raising is not motivated by Case requirements, and therefore does not occur.

This also accounts for the contrast between (1) and sentences like (9).

(9) It seems (that) John is ill


(10) * John seems (that) is ill

The embedded IP is [+tense] so raising is not motivated, and the raised version of the sentence,
(10), is ungrammatical. Note also that the presence of the complementiser that blocks raising by
the comp-trace effect. I will return to this below in §5.

Sentences like (1) are therefore interesting because, when approached from the perspective of a
raising analysis, we are faced with at least three problems: The first is that the subject of the
embedded phrase seems to raise despite being in a Case position; the second is that it seems to
leave behind an overt pronoun, perhaps a copy of some sort; finally, we must provide an account
of theta-role assignment. As we shall see, rejecting the raising analysis outright is not a
straightforward solution, as this creates other problems.

The structure of the remainder of this paper is as follows: in §1, I will present the basic facts of
copy-raising and suggest three possible analyses in §2 through §4, the Language-Particular Rule
analysis, the Small Clause analysis, and the Base Generation analysis. Finally, a Minimalist analysis
of the phenomenon will be presented in §5.

1 Some facts of Copy-Raising


Any analysis we eventually choose to explain copy-raising must account for the facts that we will
lay out in this section. Therefore, we will deal here only with word order phenomena, and not
with syntactic structure.

To begin with, copy-raising in English apparently only occurs with physical perception verbs1 and
three lexical forms that appear to be complementisers. Specifically, as far as I have been able to
determine, only the matrix verbs seem, appear, and look definitely show this pattern, together with
the complementisers like, as if, and as though. Note however that one combination here is not
possible – we never get *appear like. With this one exception, therefore, the following sentences

1 While Rogers (Rogers, 1971; Rogers, 1973) implicates physical perception verbs (including sound, feel, and smell) in

this effect, only seem and appear clearly assign no thematic role to their external argument. As we shall see, this is an
important part of the problem, so I will consider only those verbs here. If it turns out that other perception verbs
also assign no thematic role in these constructions then they will be included in the same analysis.
-3- Andrew Kaplan-Myrth

are all possible:

 seems   like 
(11) John appears  as if  he is ill
 looks  as though

For the greater part of this discussion I will assume that like, as if, and as though are
complementisers, but this will eventually be called into question. Evidence for taking them to be
complementisers include the impossibility of a co-occurring complementiser, as in the
ungrammatical (12), and the simple problem of what else it could be. I return to these questions
below.

(12) * Joan seems like that she is in a bad mood

We can account for these distributional facts in any theory by allowing these verbs to
subcategorise for their particular complementisers. Furthermore, in all these cases, the
complementisers select for a [+tense] feature on the T/I of the embedded phrase. So, taking seems
like to be representational of all the combinations, (13) is clearly ungrammatical.

(13) * Joan seems like to be ill

Now, we must establish what positions the two subjects appear in. Consider first the matrix
subject. The first question to ask is whether it gets a theta-role from the matrix verb. Three
arguments can be given supporting the conclusion that no theta-role is assigned by the matrix
verb to its external argument position: It can take the expletive it; it does not change the
idiomatic reading of an NP raised out of an embedded idiom; and it does not change the
idiomatic reading of a funny NP.

First, the matrix verb can take the expletive pronoun it in the non-raised sentence, as in (14).

(14) It seems like John is ill

Upon first glance, it might also seem as though these constructions can take the expletive there as
well, as in (15). This is mistaken however. Instead, the alleged expletive there is in fact the copy-
raised subject, as can be seen by the agreement in (16). Without agreement between the subjects,
and the subsequent agreement of the non-theta-assigning matrix verb, the sentence is
ungrammatical, as shown in (17).

(15) There looks as if there is a problem


(16) There look as if there are problems
(17) * There looks as if there are problems

Second, idiomatic readings are maintained.

(18) The cat seems like it is out of the bag


(19) The shit looks like it finally hit the fan
-4- Andrew Kaplan-Myrth

(18) and (19) both maintain their idiomatic readings despite the interference by the matrix verb
and the complementiser. In contrast, consider a verb like act, which has a similar syntactic
structure to seem except it assigns an external theta role:

(20) The cat acts like it is out of the bag [not idiomatic]
(21) The shit acts like it hit the fan [not idiomatic]

So whereas act seems to assign a theta-role to its subject, copy-raising verbs do not.

And finally, so-called “funny NPs” can raise into the matrix subject position without losing their
idiomatic readings, as sentences (22) to (24) show.

(22) Careful tabsi seem like theyi were kept on the workers
(23) Good headwayi seems like iti was made on my essay today
(24) Heedi seems like iti was taken of my advice

Having established that the matrix subject position is not assigned a theta-role, I must point out
briefly that both the matrix subject and the embedded subject are in A-positions. Support for this
comes from their binding properties – both can bind anaphors.

(25) Joan looked [PP to herself ] [CP as though [IP she had been in a fight ]]
(26) Joan looked [CP as though [IP she hurt herself ]]

We can therefore be fairly confident that both subjects are in SpecIP positions.

The fact that the matrix subject position gets no theta-role from the matrix verb was a motivation
for Rogers’ introduction of Richard, the copy-raising rule. Unfortunately, as Horn points out
(Horn, 1981), the raised idiomatic NPs and there are all unavailable for some readers. The same is
true for funny NP’s. Since idioms are assumed to be inserted as a unit, and there is not inserted in
D-structure, the fact that the raised versions of these sentences are grammatical is the major
motivation for concluding that a raising operation is active in the first place. We must therefore
search for another NP that is merged under restricted circumstances into the embedded phrase
and yet can be copy-raised by all speakers. Horn calls attention to sentences such as (27) to (29),
which are accepted by all speakers.

(27) It sounds like it’s raining out


(28) It feels like it’s cold out there
(29) It looks like it’s raining/snowing/midnight

As we know, “weather it” appears in very limited cases such as weather and time expressions. If
we could show that the matrix subject in (27) to (29) is actually “weather it” and not the expletive
it, then we would have evidence for raising.

Support for this comes from studying the contradictory readings of certain ambiguous sentences.
Consider the pair of sentences given in (30) and (31).

(30) John seems like he’s sicker than he really is


-5- Andrew Kaplan-Myrth

(31) ?# It seems like John is sicker than he really is

Sentence (30) is ambiguous, with one of its readings actually contradictory, and equal to the only
reading possible from (31). Crucially, however, a non-contradictory reading is possible because
John can take scope over seem. With this contrast in mind, consider (32).

(32) It seems like it’s raining harder than it really is

This sentence has a contradictory reading, but a non-contradictory reading is also available. If the
parallel with (30) is reliable, and there is no reason not to believe it is, then this indicates that the
it in the matrix subject position is in fact the raised “weather it” and not an expletive.

I have established that the matrix subject position is an non-θ position, that both subjects are
actually in subject positions, and provided some support for the assumption that these sentences
are the results of a raising operation. I now turn to some possible analyses of the copy-raising
phenomenon.

2 Language-Particular rules
To account for the presence of the extra pronominal in copy-raising in Turkish and Ìgbo, Moore
(Moore, 1998) and Ura (Ura, 1998) respectively propose language-particular rules that insert a
pronominal under certain circumstances.

Moore follows the analysis provided for Modern Greek by Perlmutter and Soames (Perlmutter &
Soames, 1979). In both Greek and Turkish, the difficulty for the analysis is that the NP seems to
move out of a finite ([+tense]) embedded phrase. For instance, consider (33) and (34) (from
Perlmutter and Soames).

(33) Fenome na ime fliaros simera


seem/1Sg SUBJUNCTIVE be/1Sg talkative today
‘I seem to be talkative today’
(literally: ‘I seem that I am talkative today’)

(34) O filos mu fenete na kerdizi to pegnidi mono aftos


the friend my seem/3Sg SUBJUNCTIVE be-winning the game only he/NOM
‘Only my friend seems to be winning the game’
(literally: ‘My friend seems that only he is winning the game’)

Of course, the fact that Modern Greek is a pro-drop language can make it difficult to find the
pronominal. In (33) we see what appears to be regular raising, except that the NP (which is not
overt) has raised out of a [+tense] phrase. Note, however, that only pronouns that are “not
emphasized or contrasted” are dropped (163). Using a term like mono ‘only’ causes the pronoun
to appear at the surface, as in (34).

The story in Turkish looks very similar. Raising seems to draw NPs out of [+tense] embedded
phrases, leaving behind a pronominal form. In Turkish, however, this pronominal is never overt
for some speakers. Still, Moore proposes that raising out of finite phrases is copy-raising in which
-6- Andrew Kaplan-Myrth

silent pro is left in the embedded subject position. This is accomplished by the “copy-chain
parameter”:

(35) COPY-CHAIN PARAMETER: the tail of an A-chain may be pronominal (168).

In this way, Moore essentially stipulates that in some languages raising is copy-raising. This does
not, however, address the problems introduced above. Specifically, we still have no account of
where theta-roles are assigned, and where Case is checked.

Ura picks up the argument here. Studying Ìgbo, a language with copy-raising out of finite phrases,
Ura proposes what amounts to a slight modification of Moore’s analysis. While Ura cannot get
away from proposing a language specific rule or parameter to account for the distribution of this
phenomenon, he does provide an explanation of the part of the problem related to Case.

Recall that in regular movement-raising, the movement of the embedded NP is motivated by


Case requirements. That is, it cannot be assigned Case in the embedded subject position. Ura
proposes that we should consider Case checking to be an operation, and as such it is subject to
the same principles that affect other operations. Specifically, checking is subject to the Last
Resort Condition. This means that checking must only proceed if the derivation would otherwise
crash.

He provides the following analysis. Consider the target structure, the grammatical sentence

(36) Ézèi di m [ kà oi hũ-rũ Adá ]


Eze seems to me COMP he see-asp Ada
Ezei seems to me [ that hei saw Ada ]

Consider the structure at the stage of the derivation where the embedded T is merged with the
embedded vP:

(37) [TP T [vP Eze v [VP V Ada ]]]

The subject is forced to move to SpecTP to satisfy EPP requirements. We then have (38):

(38) [TP Ezei T [vP ti v [VP V Ada ]]]

Now, while it is true that Ézè is in a configuration in which it could check Case features with T,
this only occurs if it is required in order for the derivation to converge. Instead, suppose Case
features are not checked at this point. After the higher structure is merged with the tree there will
be another EPP requirement to satisfy in the higher subject position. Ézè moves once again to
satisfy the EPP. It is in this higher position that Case features are checked.

(39) [TP Ezei T … [CP COMP [TP ti T [vP ti v [VP V Ada ]]]
× Case (checked) × Case (unchecked)

There remain unchecked Case features in the embedded phrase. If these are not checked, the
derivation will crash. Luckily, however, Ìgbo has a language-particular rule that can “supply an
-7- Andrew Kaplan-Myrth

intermediate position of the A-chain with a pronominal copy of the head of the A-chain” (74). As
Ura points out, this is the rule that is responsible for copy-raising in all of the languages in which
it is found. This pronoun has Case features (and agreement features, as Ura discusses), and is
therefore in a configuration to have its features checked by the embedded T:

(40) [TP Ezei T … [CP COMP [TP hei T [vP ti v [VP V Ada ]]]
× Case (checked) × Case (checked)

This accounts for Case (and similarly agreement) features, so it is progress over Moore’s
proposal. It does not, however, address theta-role assignment. There are at least two possible
solutions to this problem. If we accept that the matrix raising verb does not assign an external
theta-role then we only have one theta-role to assign, and two overt NPs. Clearly, only one of
them gets a theta-role assigned in the usual manner. Therefore, either the raised NP gets a theta-
role in its base position and transmits it down to the inserted pronoun, or else the pronoun
somehow gets the theta-role after being inserted and transmits it up to its antecedent. Neither of
these solutions is perfect, but the former is clearly the lesser of two evils – it, at least, was merged
into a theta-position in Spec of vP earlier in the derivation.

Ura goes on to specifically address some cases of English copy-raising, likening the above
language-specific rule to a resumptive pronoun strategy. That is, since the resumptive pronoun
strategy is active to some extent in English, it can fulfil the same role that the language particular
rule does in Ìgbo, Greek, Turkish, and Haitian Creole, as we see below. Ura does not address
entirely grammatical cases of English copy-raising such as (1), but rather focuses on more
borderline cases:

(41) ??? Johni seems that hei has hit Bill


(42) * Johni seems that Bill has hit himi

Ura suggests that (41) is marginally grammatical because the derivation can proceed as above up
to the language particular rule. At that point, “the vacated subject position in the embedded
clause is supplied with a pronominal copy of the moved element by the resumptive pronoun
strategy, which is marginal in English” (82). On the other hand, (42) is judged to be
ungrammatical because the movement of John out of the embedded phrase is not permitted since
raising out of the object position is not licensed by seem.

Ura’s footnote 81 briefly addresses sentences with as-if or like as complementisers, but states that
sentences in which the embedded object has apparently raised are “not so good”. He provides
(43) as an example.

(43) ?? Johni seems as if Bill has hit himi

It seems, however, that this may have simply been a badly chosen example. Based on my
judgement and an informal survey, (44) and (45) are both grammatical.

(44) John looks like Mary finally kissed him


(45) Joan seems like a car just ran her over
-8- Andrew Kaplan-Myrth

(46) Tedi looks like Jane has been hassling himi again (Rogers, 1971)

In fact, there are other positions from which copy-raising is possible although the implied A-
movement would be illicit. Consider, for instance, NPs that originate in possessive phrases.
These can be either in the subject or the object of the embedded phrase, and in both cases “copy-
raising” seems to be possible.

(47) Johni looked like [ hisi cat ] had died


(48) Joani appeared as if [ heri investments ] had been lost on the stock market
(49) The presidenti seemed as if the senate has finally passed [ heri bill ]
(50) Johni looked like there was a fly in [ hisi soup ]

As Rogers points out, copy-raising also affects noun phrases within PP’s, as in (51):

(51) This violini sounds like Max has been hammering nails with iti (Rogers, 1971)

A-movement is not permitted from any of these positions, and yet the sentences are grammatical.
Even a resumptive pronoun strategy will not help, since the original movement is illegal. Thus,
despite Ura’s explanation of feature checking, something else must be responsible for this
phenomenon. Moore’s and Ura’s theories were already undesirable because they were essentially
stipulative given the language-specific parameter, but they are empirically insufficient to account
for the data.

We therefore turn our attention to an alternative explanation, derived from an investigation of


Haitian Creole.

3 The Small Clause analysis


Haitian Creole (HC) raising constructions have the same pattern as these special cases in English.
In fact, as Deprez shows (Deprez, 1992), copy-raising always “occurs with predicates which
assign no external theta-role and take tensed sentential complements” (198). Therefore, one
important difference between HC and English is that all raising is copy-raising in HC but not in
English. Another difference is that there is no (overt) complementiser in these constructions in
HC. For example, (53) is related to (52) in HC in the same way as (55) is related to (54) in
English.

(52) sanble Jan pati


seem John left
It seems that John leaves

(53) Jan sanble li pati


John seem he left
John seems he leaves

(54) It seems like Joan left

(55) Joan seems like she left


-9- Andrew Kaplan-Myrth

Despite the differences between these phenomena, most conspicuously the fact that in HC the
pronominal copy appears much more freely than in English, there are striking similarities
between these two constructions. Most important for us is that in both cases there is an extra
element, the pronoun, whose presence we must explain.

Deprez proposes that the proper analysis of “pronominal raising”, as she calls it, involves a small
clause with an extra specifier position, SPEC2. This small clause has a “subject” in SPEC2 and the
pronominal in the specifier of the predicate of the small clause. Under this analysis, the derivation
of (53) at the step before raising looks like (56).

(56) [ sanble [SC Jan [PRED li pati ]]]

The assignment of theta-roles is critical to Deprez’s argument. In the configuration in (56), the
pronoun li gets both case and a theta-role from the head, pati. According to Deprez, li may fail to
satisfy the theta-requirement due to underspecificity that is particular to HC (Deprez 215). It
therefore assigns its theta-role to its maximal projection by a process called vertical binding, in
which an “unassigned external theta-role is transmitted to the immediate maximal projection of
the theta-assigning head” (213). This theta-role is then assigned to Jan via another process of
theta-assignment, predication, which “is a relation by which a maximal projection assigns a theta
role inherited by vertical binding to its sister” (213). Thus, Jan gets a theta-role assigned via a
rather complex sequence of events involving three different types of theta-assignment. It does
not, however, get Case in this position. The result is that Jan raises to the matrix subject position
and checks Case features with the matrix I0 head.

The extension of this proposal to the English situation is straight-forward. An earlier stage of the
derivation of (55) looks like (57), before raising:

(57) [ seems [CP like [SC Joan [PRED she left ]]]]

One obvious problem at this point is the question of how the theta-role is assigned by left to Joan.
In HC it is facilitated by underspecificity on the part of the pronominal, but this does not hold in
English. With no theta-role assigned to Joan, it is not clear that the derivation will converge.

Another problem surfaces when we consider sentences in which it is the object of the embedded
phrase that is coindexed with the matrix subject, such as (44) and (45) above. Deprez points out
that structures such as (58) (Deprez’s (31a)) are ungrammatical in HC (217):

(58) * Jan sanble Mari renmen li


John seems Mary to like him

This result is in fact predicted by the small clause analysis provided. This is because theta-role
assignment to the SPEC2 position of the SC is impossible since the pronominal no longer
intervenes in order to facilitate theta-role assignment. Therefore, since Jan can not get a theta-
role, the derivation crashes.

In English, however, these sentences are grammatical as we saw above. Worse still, un-governed
pronouns within possessive phrases are possible, as shown in (47) to (50). Since theta-role
- 10 - Andrew Kaplan-Myrth

assignment will not work out in these cases as described by Deprez, the small clause analysis will
not hold. So, while Deprez’s analysis is preferable to those provided by Moore and Ura, it is still
empirically insufficient to account for the English data. Furthermore, the idea of the subject
raising from an extra position posited only for this analysis is theoretically undesirable. If there is
no more motivation for proposing the SPEC2 position, then this explanation of copy-raising
amounts to a stipulation as well.

We therefore turn our attention to the remaining possible analysis, that in which, despite
appearances, there is in fact no raising operation.

4 Base Generation
Having described two possible analyses of English copy-raising and focussed attention on their
shortcomings, we turn briefly to the remaining possibility. This is the idea that while (1) and
sentences like it seem to be derived from (3) in the same way that (2) is, they are in fact not. That
is, no raising is involved. The strongest argument for this position comes from the observation
that the purported movement targets non-subject positions, as was indicated above in (44) to
(51).

Instead, we can liken copy-raised constructions to tough-movement constructions such as (59)


which, under one analysis (Chomsky, 1981), has the underlying structure in (60).

(59) Joan is easy to please


(60) Joani is easy [CP OPi [C’ [IP PROarb to please ti ]]]

Case checking is simple in this derivation: PRO checks Case in its merged position as the
argument of please and then moves to Spec IP to satisfy the EPP. Separately, Joan has Case
checked in the matrix subject position.

Still, assignment of theta-roles is not unproblematic. While it seems clear that the operator is
assigned the theta-role of the argument of please, it is unclear whether or not Joan gets a theta-role.
If it does not, as Chomsky argues, then we have a theta-criterion violation since every overt NP
must be assigned a theta-role.

Turning back to copy-raising constructions now, we could propose that the apparently raised NP
is actually merged into the matrix subject position rather than moved there. Recall however that
this was thought to be a non-thematic position. As a result, the investigation must turn to theta-
role assignment to the overt NP.

In pursuing the small clause analysis of HC raising, Deprez makes use of this tough-movement-
style derivation to explain raising from the object position of the embedded phrase. Recall that
sentences like (58), in which sanble raises the object of the embedded phrase, are ungrammatical.
Interestingly, another HC verb, rete ‘remains’, allows raising from either the subject or the object
position of the embedded phrase. For instance, (61) is entirely grammatical:

(61) Jan rete pou Mari renmen li


John remain for Mary to like him
- 11 - Andrew Kaplan-Myrth

‘It remains for Mary to like John’

When the verb was sanble, the problem with this derivation was that the pronoun is no longer in a
position to facilitate theta-role assignment to the SPEC2 position of the small clause. This is easily
solved, however – if an empty operator, OP, were in the position to assign a theta-role to the
subject of the small clause by the process of predication. The structure of (61) is given in (62).

(62) Jani rete [SC ti [PRED OPi pou Mari renmen lii ]]

The corresponding analysis of the English sentence given in (45) is shown as (63).

(63) Joani seems like [SC ti [PRED OPi a car just ran heri over ]]

Clearly, we must now ask what the required relationship is between OP and the coindexed
pronoun. Deprez states that “empty operators can be related either to subjects or to objects,
since they bind a variable which receives its own theta-role” (227). Even if we accept this,
however, it is not clear that OP can be related to a non-argument position. That is, while Deprez
considers cases in which OP is coindexed with the embedded object, she does not consider cases
in which it must be coindexed with possessive-phrase-internal pronouns, as we know we must
have in the English cases. If coindexation is enough then we would expect there to be no
structural restrictions on the relative locations of OP and the pronoun. This may in fact be the
case – I return to a similar question when I consider the relation between the matrix subject and
the embedded pronoun in §5.

Besides the problem with the relationship between OP and the pronoun is the fact that this
solution seems stipulative – encoding the desired subject in a position where it can slip a theta-
role to the matrix subject creates the impression of anticipating the outcome. We would clearly
like a solution with more predictive power.

A more Minimalist solution is that the matrix subject does not get a theta-role. In this view, its
reference comes from coindexation, but it has no theta-role assigned. If this were possible within
the theory, we would have no need for movement or empty operators in order for the derivation
to converge. The structures would simply look like this:

(64) Johni seems [CP like [IP hei is ill ]]


(65) Joani looks [CP like [IP a car hit heri ]]
(66) Johni looked [CP like [IP [DP hisi cat ] had died ]]
(67) Joani looked [CP like [IP there was a fly [PP in [DP heri soup ]]]]

Note that coindexation is required for these sentences to be grammatical. When its removal is
forced by changing the gender of the pronoun, the sentences are no longer acceptable.

(68) * Johni seems [CP like [IP shei is ill ]]


(69) * Joani looks [CP like [IP a car hit himi ]]
(70) * Johni looked [CP like [IP [DP heri cat ] had died ]]
(71) * Joani looked [CP like [IP there was a fly [PP in [DP hisi soup ]]]]
- 12 - Andrew Kaplan-Myrth

This might be evidence that the matrix subject does not get a theta-role in its “raised” position,
but rather through coindexation. We must leave the investigation of the feasibility of this
proposal for another time.

5 A Minimalist Analysis
At this point, it is useful to stand back from the problem and evaluate the information we have
gathered so far. The most significant result we have is that movement from the overt embedded
position is impossible in so-called copy-raising sentences. This is a direct result of the availability
of sentences such as (47) to (50), in which the apparently raised subject comes from a position in
the embedded clause from which movement is impossible, such as a possessive phrase. This
movement must be ruled out by the ECP or some other Minimalist principle in order to account
for the impossibility of this movement in general.

This does not rule out a movement analysis altogether, however, since movement from a
different position is possible in principle. Here, I refer to movement from a special specifier
position following seems like, as in Deprez’s analysis of copy-raising in Haitian Creole. In that
particular analysis, the subject moves from the specifier of a small clause, as in the D-structure
represented in (72).

(72) [ sanble [SC Jan [PRED li pati ]]]


seems Jan he left

Recall that in this analysis, theta-role assignment to Jan is mediated by the pronoun li, and then
Jan moves to the matrix subject position to satisfy the Case filter. Deprez’s analysis included the
possibility for a subject to relate to an object position in the embedded clause, although she
provides no examples for which the subject is related to an NP within a possessive phrase. In
order to get assignment of a theta-role to the subject when it is related to the embedded object,
an additional operator, OP, is used as an intermediary, as in the D-structure in

(73) rete [SC Jan [PRED OPi pou Mari renmen lii ]]
remains Jan for Mari to like him

Two main difficulties for this analysis are discussed above. The first is that the relationship
between OP and the pronoun is unexplained, making it unclear whether the analysis could
account for the pronoun being located in a possessive phrase. The second difficulty for the
analysis is that it has a flavour of stipulation. For these reasons, this account was abandoned.

Indeed, any analysis involving movement from an extra position, whether it is an extra specifier
position on a small clause or the specifier position of the complement phrase itself, will have a
flavour of stipulation and will therefore be undesirable. An analysis that does not appeal to extra
structural positions will be preferred.

However, as we have seen, if we discard the idea of movement altogether and proceed with an
analysis in which the matrix subject is base generated in that position then, having concluded that
the matrix subject position is not a theta position, we would have to face a violation of the theta
criterion. However, the theta criterion is not present in Minimalist analyses, its work instead being
- 13 - Andrew Kaplan-Myrth

accomplished for the most part by Full Interpretation. Specifically, theta role assignment is
feature checking, and unchecked theta features are uninterpretable at LF. The result of this is that
all theta features must be checked by LF. However, there is no requirement that an argument
must have theta features at all. If an argument were to have no theta feature then it could inhabit
a non-theta-checking position, such as the external argument position of a verb like seem.

In short, I suggest that in apparently copy-raised constructions the matrix subject is base
generated in its surface position, and it has no theta features to be checked. As stated, this still
incurs a violation of Full Interpretation at LF, since all DP’s must be semantically interpretable as
well. So while it is not fatal for the matrix subject to not get a thematic role, it is fatal if it does not
get a semantic role.

In light of this interpretation of thematic role assignment and semantic role requirements, we can
see a pattern in the grammaticality and ungrammaticality of copy-raising sentences. Consider the
critical cases:

(74) * John seems that there is no tomorrow


(75) * Johni seems that hei is ill
(76) ?? John seems like there is no tomorrow
(77) Johni seems like hei is ill
(78) John eats like there is no tomorrow
(79) Johni eats like hei is on a diet

In sentences (74) to (77), John does not receive a thematic role. In sentences (74) and (75), John
also receives no semantic role, resulting in the ungrammaticality of these sentences. In (74), there
is no co-referent from which John can receive a semantic role. In contrast, John is coindexed with
he in (75), but co-reference is blocked. Similarly, John has no co-referent and therefore receives no
semantic role in (76)2. Crucially, in sentence (77), while by hypothesis no thematic role is
assigned, John is coindexed with he and co-reference is not blocked, resulting in John inheriting a
semantic role from he. In fact, as we saw from both the weather-it sentences and the there-
agreement sentences, the features of he are somehow transmitted to John through coindexation.
Finally, (78) and (79) are included to fill out the paradigm, providing cases in which thematic
roles are assigned directly to John from the matrix verb, eat.

We are left with the question of why co-reference is blocked in (75) but not in (77). If that blocks
co-reference while like does not, then the results are modelled. It remains to elaborate on the
that-blocking effect or, more likely, the “like/as if/as though”-transparency effect. In order to do
this, we must consider more data, comparing the grammaticality and acceptability of sentences
with that-CP’s and those with like.

2 Sentence (76) is not judged to be as ungrammatical as sentence (74), and it is unclear why this is the case. A hint
may be that the possible reading of (76) requires that John be visibly depressed or otherwise displaying his sense that
the world is going to end before tomorrow. This suggests that, under this reading, John gets a thematic role from seem.
Perhaps this reading is impossible in (74) because the thematic-role assigning verb seem does not take a CP headed by
that.
- 14 - Andrew Kaplan-Myrth

In this discussion, I follow Brody in taking a representational approach to syntax, evaluating


complete structures based on certain constraints (Brody, 1995). This is in contrast to a
derivational approach in which certain movement operations would be disallowed because the
movement itself violates some condition. In contrast, in a representational framework the
structure that results from apparent movement (rather than the movement itself) is disallowed
because the resulting structure violates a condition or constraint.

Given this framework and the conclusions derived so far, a simple analysis presents itself:

(80) a) There is some correspondence involving coindexation which licences the


transmission of semantic properties
b) that blocks this correspondence
c) like does not block this correspondence

The analysis of a typical copy-raised sentence is therefore quite simple. For an example, consider
(1). John has no theta feature so the fact that seem does not check theta features is unproblematic.
However it is still a ment by Full Interpretation at LF that John get a semantic interpretation.
Since it cannot do this by theta feature checking, it must get its semantic role by coreference. In
this case, coreference by the coindexation correspondence is not blocked, and so the structure is
legal. If instead of like, the complementiser had been that, then coreference would be blocked,
John would get no semantic role, and the structure would be ruled out.

We must now ask, what is the correspondence that licences transmission of semantic information
from the embedded pronoun to the coindexed matrix subject, and why does that block the
correspondence while like does not.

I suggest that the only structural correspondence that will capture the data is a simple c-command
relation between the matrix subject and the pronoun in the embedded clause. Specifically, an NP
is in correspondence with a pronoun if it c-commands the pronoun. The remainder of this paper
is concerned with specifying the relevant distinction between complementisers like that and those
like like.

One possibility is that like does not trigger the comp-trace effect while that does. This would be a
desirable solution since the comp-trace effect is independently motivated. Furthermore, if we
assume the existence of a null complementiser Ø, then we have a precedent for a complementiser
that does not trigger the comp-trace effect, as in (81).

(81) Whoi does it seem Ø ti is ill?

Support for this analysis comes from the comparison of Wh-movement across like and that. We
find that, while it is not entirely grammatical to some speakers, (83) is more acceptable than (82).

(82) * Whoi does it seem that ti is ill?


(83) ? Whoi does it seem like ti is ill?

Unfortunately, this analysis does not describe the data. The comp-trace effect only affects
movement out of the embedded subject, immediately beside the complementiser. It does not
- 15 - Andrew Kaplan-Myrth

affect movement out of the object position, for instance. So, for example, the comp-trace effect
does not rule out sentences like (84), which are just as acceptable as (85) and (86).

(84) Whoi does it seem that John kissed ti ?


(85) Whoi does it seem Ø John kissed ti ?
(86) Whoi does it seem like John kissed ti ?

If the comp-trace effect were the only thing blocking semantic correspondence, we would expect
that a matrix subject with no theta role would get a semantic role from a pronoun embedded in a
CP headed by that as long as that pronoun is not the subject of the embedded IP. However, as we
can see from (87), this is not the case.

(87) * Joani seems that John hugged heri

It looks like that blocks a wider set of structural relations than we can attribute to the comp-trace
effect3. Perhaps like and that simply affect different types of relations between positions, that is,
non-movement chains. It might be possible to distinguish between the two complementisers by
referring to the type of chain they block or do not block – specifically, I propose that that blocks
A-chains (ie. chains between elements in A-positions), does not block Ā-chains, and prohibits
movement from the position immediately below itself in the tree (the comp-trace effect); in
contrast, like does not block A-chains, Ā-chains, or prevent movement out of its immediate
neighbour. As a result, words like like permit the correspondences we have observed and do not
trigger the comp-trace effect. This is a line of research that must be pursued further.

Another line of research that remains open is the question of why like does not trigger the comp-
trace effect. One possibility is that like is not a complementiser after all, but is in fact part of the
VP with seem, taking as a complement a bare IP. If this were true then the reason there is no
comp-trace effect is that there is simply no complementiser. Unfortunately, though perhaps not
fatally, it can be shown that seem like is not a complex verbal unit by insertion of a PP:

(88) [IP It seems [PP to me ] [CP that [IP John is in trouble ]]]
(89) [IP It seems [PP to me ] [IP John is in trouble ]]
(90) ? [IP It seems [PP to me ] like [IP John is in trouble ]]

For at least some speakers, (90) is just as acceptable as (88) and (89). For me, however, (90) is
marginal in comparison with (88) and (89) which are both perfect. This may be an indication that
those speakers that accept (90) are reanalysing like to be a complementiser. More work needs to
be done along these lines to establish the syntactic category for like, as if, and as though.

3Note that sentences like (a) below may provide some evidence that the correspondence is in fact permitted when it
does not trigger the comp-trace effect, since the sentence seems to be acceptable to at least some speakers. On the
other hand, note that (b) is also acceptable to some degree despite being expected to trigger the comp-trace effect.
(a) ? Johni seems like Mary said that we should vote for himi
(b) ? Johni seems like Mary said that hei voted for Bush
- 16 - Andrew Kaplan-Myrth

6 Works Cited
Brody, M. (1995). Lexico Logical Form: A Radically Minimalist Theory. Cambridge, MA: MIT Press.
Chomsky, N. (1981). Lectures on Government and Binding. Dordrecht: Foris.
Deprez, V. (1992). Raising Constructions in Haitian Creole. Natural Language and Linguistic Theory,
10, 191-231.
Horn, L. (1981). A Pragmatic Approach to Certain Ambiguities. Linguistics and Philosophy, 4, 321-
358.
Massam, D. (1988). Case Theory and the Projection Principle (Dissertation Abstract). The
Linguistic Review, 5, 377-390.
McCawley, J.D. (1988). The Syntactic Phenomena of English (Vol. 1-2). Chicago: University of
Chicago Press.
Moore, J. (1998). Turkish Copy-Raising and A-Chain Locality. Natural Language and Linguistic
Theory, 16, 149-189.
Perlmutter, D.M., & Soames, S. (1979). Syntactic Argumentation an the Structure of English. Berkeley:
University of California Press.
Postal, P.M. (1971). Cross-Over Phenomena. New York: Holt, Rinehart, and Winston.
Quirk, R., Leech, G., & Svartvik, J. (1972). A Grammar of Contemporary English. London: Longman
Group Ltd.
Rogers, A. (1971). Three Kinds of Physical Perception Verbs. Chicago Linguistic Society, 7, 206-222.
Rogers, A. (1973). Physical Perception Verbs in English: A Study in Lexical Relatedness, UCLA.
Ura, H. (1998). Checking, Economy, and Copy-raising in Igbo. Linguistic Analysis, 28(1-2), 67-88.

You might also like