Andrew Radford
(radford@essex.ac.uk)
1. Introduction
Pesetsky
and Torrego 2000 (henceforth: PT) present a novel minimalist analysis of the
syntax of clauses, primarily in relation to present-day English. This paper
presents a brief outline of key ideas in their paper, discusses theoretical and
descriptive shortcomings of their account, and presents an alternative analysis
which captures their intuitions and insights while avoiding some of the
pitfalls of their work. Sections 2-5 provide a brief outline of relevant
aspects of PT’s analysis; sections 6-8 highlight specific problems in their
analysis, and section 9 suggests an alternative which is consistent with many
(though not all) of the assumptions PT make. Section 10 highlights other
potential problems with PT’s paper to be explored in future research.
2. Wh-questions
PT
posit that in root wh-questions such as that in (1) below:
(1) Where has Mary gone?
the
head C of CP carries an uninterpretable EPP tense-feature [uT] which drives movement of the auxiliary has to C, and an uninterpretable EPP wh-feature [uWh] which drives movement of where to spec-CP; has carries an interpretable tense feature and where carries an interpretable wh-feature. Under their analysis, (1) has the structure shown in simplified
form in (2) below at the point where C merges with its TP complement[2]:
(2) [CP [C uT, uWh] [TP Mary [T has]
gone where]]
Given
the assumption that EPP-features can only be deleted via movement, the [uT] feature of C will trigger movement
of the present tense auxiliary has
from T to C[3] (thereby
erasing the uninterpretable [uT]
feature on C), and the [uWh] feature
of C will trigger movement of where
to spec-CP (thereby deleting the uninterpretable [uWh] feature on C). The structure derived after both movement
operations have applied is therefore (3) below (strikethrough indicating
material to be deleted at the end of the relevant phase, and traces being shown
in a smaller typeface):
(3) [CP Where [C has uT, uWh] [TP
Mary [T has] gone where]]
Since
both uninterpretable EPP-features of C have been deleted, the structure
converges at LF.
PT
note the observation by Koopman (1983) that in questions in which the
wh-expression
is the subject of the interrogative clause, T-to-C movement (and hence do-support) is barred: cf.
(4)(a) Who bought the book?
(b)
*Who did buy the book [*unless did
is focused]
They
present evidence that the wh-subject raises to spec-CP in wh-subject questions
like (4), from the distribution of expressions like the hell which attach to wh-phrases. They note that the hell can only attach to a wh-moved
expression in spec-CP, not to an in situ expression[4]:
cf.
(5)(a) What
the hell did Sue give to who?
(b)
*What did Sue give to who the hell?
And
they further note that wh-subjects (like who
in (4a) above) can be modified by the
hell: cf.
(6) Who
the hell bought the book?
From
this they conclude that wh-subjects like who
in (4a) move to spec-CP. Under their analysis, movement of the wh-subject who to spec-CP in (4a) is driven by the
[uWh] EPP-feature of C.
However,
if we assume (as PT do) that a root interrogative C always contains both a [uWh] and a [uT] feature, this will mean that (4a) has the structure (7) below
at the point where C is merged with its TP complement:
(7) [CP [C uT, uWh] [TP who [T ø]
bought the book]]
Movement
of who to spec-CP will erase the [uWh] EPP-feature of C: but how is the [uT] feature of C erased?
PT
argue that who erases not only the [uWh] feature of C but also its [uT] feature. More precisely, they claim
that subjects of tensed clauses carry an uninterpretable tense feature [uT]. They
argue that the uninterpretable nominative case feature of finite subjects is
anomalous within the framework of Chomsky (1998, 1999), and propose to replace
it by an uninterpretable tense
feature. Uninterpretable features are canonically erased via agreement with
corresponding interpretable features (e.g. the uninterpretable person/number
features of T are erased via agreement with a subject carrying interpretable
person/number features): however, the fact that the uninterpretable nominative
case feature of the subject of a finite clause is erased by a T carrying an
interpretable tense feature suggests that nominative case should be reanalysed
as an uninterpretable tense feature – thereby leading to a symmetrical theory
of feature erasure in which an uninterpretable feature [uF] can only be erased via an agreement relation with a
corresponding interpretable feature [F].
Given PT’s assumptions, who in (7)
will contain not only an interpretable
wh-feature
(by virtue of its interrogative operator status), but also an uninterpretable
tense feature (by virtue of being the subject of a tensed clause). Hence,
movement of who to spec-CP will erase
both of the uninterpretable EPP-features carried by C, so deriving the
structure (8) below:
(8) [CP Who [C uT, uWh]
[TP [who] [T ø T] bought the book]]
The
alternative possibility of using do-support
(i.e. merging do in T and raising it
to C) to erase the [uT] feature of C
in (7), and then moving who to
spec-CP to erase the [uWh] feature of
C is ruled out by economy considerations as being more costly than simply
moving who to spec-CP – thereby
accounting for the ungrammaticality of (4b).
3. Wh-exclamatives and
wh-interpretation
PT
extend their analysis of root wh-questions to root wh-exclamatives such as:
(9) What a silly book Mary has bought!
They
suggest that (like root wh-questions) these too are CPs headed by a C which has
uninterpretable tense and wh-features. Given this assumption, (9) will have the
structure (10) below at the point where C merges with its TP complement:
(10) [CP [C uWh, uT] [TP Mary [T has]
bought what a silly book]]
If
the [uWh] feature of C attracts what a silly book and its [uT] feature attracts the tensed subject Mary, we derive (11) below:
(11) [CP What a silly book Mary [C uWh, uT] [TP Mary [T has] bought what a
silly book]]
And
(11) is the structure associated with (9).
However,
an apparent problem posed by their analysis is that there is another (equally
economical) way of erasing the [uT]
EPP-feature of C – namely by preposing the tensed auxiliary has (rather than the tensed subject Mary), so deriving:
(12) [CP What a silly book [C has uWh, uT] [TP Mary [T has] bought what a
silly book]]
Following
Chomsky (1995, p.296) PT maintain that feature-attraction is subject to a
locality constraint to the effect that a head H which attracts a feature F
attracts the closest constituent
carrying F. They posit that in a structure of the form
X...
[YP SPEC Y...] where X c-commands YP, both Y (= head-YP) and spec-YP
are equally close to X[5].
This means that the [uT] feature of C
in (10) can attract either T (= has)
or spec-TP (= Mary). If C attracts Mary, we derive the grammatical sentence
(9) What a silly book Mary has bought!
But if C attracts has we derive the
ungrammatical sentence *What a silly book
has Mary bought! It would therefore seem that the syntactic component of
PT’s grammar overgenerates (i.e.
generates a class of structures which are ill-formed).
Moreover,
the converse problem arises with interrogatives; after all, in a structure such
as (13) below (repeated from (2) above):
(13) [CP [C uT, uWh] [TP Mary [T has]
gone where]]
the
[uT] feature of C could attract the
tensed subject Mary rather than the
tensed auxiliary have, so deriving:
(14) [CP Where Mary [C uT, uWh]
[TP Mary [T has] gone where]]
But
once again, the resulting sentence *Where
Mary has gone? is ungrammatical.
The
nature of the problem in both cases is the following: if all finite root
wh-clauses
are projections of a C carrying a [uT]
EPP-feature, we should expect that C can freely attract either a tensed
auxiliary or a tensed subject; in other words, we should expect auxiliary
inversion to be optional in root wh-interrogatives and wh-exclamatives alike.
But this is not the case. Instead (on PT’s assumptions) C attracts a tensed
auxiliary in root wh-questions and a tensed subject in root wh-exclamatives.
How can this be accounted for?
PT
argue that the unwanted derivations are ruled out by principles of
interpretation which map syntactic into semantic representations. More
specifically, they posit an interpretive principle which can be paraphrased
informally as follows:
(15) Wh-Interpretation
Principle/WHIP[6]
A root CP with a wh-specifier is
interpreted as
(i) exclamative if it also has a non-wh
specifier
(ii) interrogative
otherwise
Consider
how WHIP handles the set of
structures in (16a/b/c/d) below (repeated from (3/14/11/12) above):
(16)(a) [CP Where [C has uT, uWh]
[TP Mary [T has] gone where]]
(b)
[CP Where Mary [C uT, uWh] [TP Mary [T has] gone
where]]
(c)
[CP What a silly book Mary [C uWh,
uT] [TP Mary [T has]
bought what
a silly book]]
(d)
[CP What a silly book [C has uWh,
uT] [TP Mary [T has]
bought what a
silly
book]]
Since
CP in (16a) has a wh-specifier but no other specifier, it is correctly
interpreted as interrogative by (15ii). Since CP in (16b) has both a
wh-specifier and a non-wh specifier, it is interpreted as exclamative by (15i):
if it is a lexical property of where
that it cannot have an exclamative interpretation, even the possibility of an
exclamative interpretation for (16b) will be ruled out, leaving (16b) as an
uninterpretable structure. Since CP in (16c) has both a wh-specifier and a
non-wh specifier, it is correctly interpreted as exclamative by (15i). Since CP
in (16d) has a wh-specifier but no other specifier, it is interpreted as
interrogative by (15ii); but if it is a lexical property of what that it cannot have an
interrogative interpretation when it modifies a singular indefinite expression
like a silly book, even the
possibility of an interrogative interpretation will be ruled out, leaving (16d)
as an uninterpretable structure. So, overall WHIP correctly specifies which of the structures in (16) are interpretable
(and what interpretation they have) and which are not.
Moreover,
as PT note, WHIP correctly predicts
that wh-subject structures like (17) below are uninterpretable:
(17) *What a silly person just called me on
the phone!
Given
PT’s assumptions, (17) will have the structure (18) at the point where C merges
with its TP complement:
(18) [CP [C uWh, uT] [TP what a silly
person [T ø] just called me on the
phone]]
The
most economical way of erasing both the [uWh]
and the [uT] features on C is to move
the tensed wh-expression what a silly
person (which is tensed by virtue of being the subject of a tensed clause)
into spec-CP as in (19) below:
(19) [CP What a silly person [C uWh, uT] [TP what a silly person [T ø] just
called me on the phone]]
Since
the CP in (19) has a wh-specifier but no other specifier, (19) will be assigned
an interrogative interpretation in accordance with (15ii). However, since it is
a lexical property of what that it
cannot have an interrogative interpretation when it modifies a singular
indefinite expression like a silly person,
the possibility of an interrogative interpretation is ruled out, leaving the
resulting sentence (17) *What a silly
person just called me on the phone uninterpretable.
4. Optional complementisers
and the that-trace effect
PT
present an interesting analysis of the seeming optionality of complementisers
in complement clauses like those in (20) below:
(20)(a) Mary thinks [Sue will buy the book]
(b)
Mary thinks [that Sue will buy the book]
They
follow Chomsky (1998, 1999) in positing that complement clauses like those
bracketed in (20) are CPs. They further posit that C in such clauses has an
uninterpretable tense feature [uT]
with the EPP-property. Hence, at the point where the embedded C is merged with
its TP complement, (20a/b) will have the simplified structure (21) below:
(21) [CP [C uT] [TP Sue [T will] buy the book]]
There
are two (equally economical) ways of erasing the uninterpretable [uT] feature of C. One is by raising the
tensed subject Sue to spec-CP, so
deriving (22) below:
(22) [CP Sue [C uT] [TP Sue [T will] buy the book]]
The
other is by moving the tensed auxiliary will
to C. However, PT posit that when T raises to C in embedded declarative
clauses, the preposed auxiliary is spelled out as that, and the trace copy it leaves behind is not erased, so that
T-to-C raising produces the derived structure (23) below (the subscripts used
by PT indicating that the first occurrence of will is spelled out as that):
(23) [CP [C thatj uT] [TP Sue [T willj]
buy the book]]
On
this view, that is not in fact a
complementiser, but rather is a double
of a preposed auxiliary in a declarative complement clause: hence, for
succinctness, we can refer to the relevant phenomenon as that-doubling. Since will
and Sue are equally close to C,
T-to-C movement and spec/TP-to-spec/CP movement are equally economical – hence
the fact that both types of bracketed complement clause in (20) are equally
grammatical.
A
particular advantage which PT claim for their analysis is that the so-called
that-trace effect can be attributed to
general economy requirements rather than to an ad hoc surface structure
constraint (as in Perlmutter 1971) or to a local (ECP) binding requirement on
surface traces (as in Kayne 1980 or Rizzi 1990). To see this, consider how they
account for the contrast below:
(24)(a) What did John say Mary
will buy?
(b)
What did John say that Mary will
buy?
(25)(a) Who did John say will buy the book?
(b)
*Who did John say that will buy
the book?
Given
the assumptions PT make, each of the italicised complement clauses in (24/25)
will be a CP headed by a C which carries two EPP-features – a [uT] feature which attracts a tensed
expression, and a [uWh] feature which
attracts a
wh-expression
(so allowing wh-movement to apply in a successive cyclic fashion). At the stage
where the embedded C is merged with its TP complement, (24a/b) will have the
simplified structure (26) below:
(26) [CP [C uT, uWh] [TP Mary [T
will] buy what]]
The
[uWh] EPP-feature of C will attract
(and trigger movement of) what; its [uT] feature will attract either the
tensed auxiliary will (which
undergoes that-doubling) so deriving
(27a) below, or the tensed subject Mary
so deriving (27b):
(27)(a) [CP what [C thatj uT, uWh] [TP Mary [T willj] buy what]]
(b)
[CP what Mary [C uT, uWh] [TP Mary [T will] buy
what]]
Subsequent
merger and movement operations will derive (24b) What did John say that Mary will buy from (27a) and (24b) What did John say Mary will buy? from
(27a).
Now
consider the derivation of (25a/b). At the point where the embedded C is merged
with its TP complement, both will have the structure (28) below:
(28) [CP [C uT, uWh] [TP who [T will]
buy the book]]
Since
who carries an interpretable
wh-feature and an uninterpretable tense feature, the [uT, uWh] EPP-features of
C can be erased by moving who to
spec-CP, so deriving:
(29) [CP who [C uT, uWh]
[TP who [T will] buy the book]]
Subsequent
merger and movement operations will derive the structure associated with (25a) Who did John say will buy the book?
However, an alternative way of erasing the EPP-features of C in (28) might seem
to be for who to move to spec-CP in
order to erase the [uWh] feature of C
and for the tensed auxiliary will to
move to C (being doubled as that) in
order to erase the [uT] feature of C,
so deriving:
(30) [CP who [C thatj uT, uWh] [TP who [T willj] buy the book]]
Subsequent
merger and movement operations will derive the structure associated with (25b)
*Who did John say that will buy the book?
But the resulting structure is ungrammatical. Why? PT’s answer is that moving who on its own (as in (29) above) is
more economical than moving both who
and will (as in (30) above), so that
the that-trace effect in (25b) is
simply an artefact of more general economy considerations.
5. Wh+that structures
PT
extend their analysis to offer an account of the syntax of wh+that structures such as the following (found in Belfast English:
see Henry 1995, p.107):
(31) They didn’t know which model that we had discussed
They
note that such structures are ungrammatical when the wh-word is the subject of
the that clause, as in (32) below:
(32) *I wonder which author that wrote this book
Under
their analysis, the wh-clause in (31) will have the structure (33) below at the
stage of derivation where C merges with its TP complement:
(33) [CP [C uT, uWh] [TP we [T had]
discussed which model]]
The
[uWh] EPP-feature of C attracts which model to move to spec-CP; the [uT] feature can attract either the
subject we (ultimately deriving They didn’t know which model we had
discussed) or the past tense auxiliary had.
In the latter case, the auxiliary moved to C can either be spelled out as had (deriving They didn’t know which model had we discussed) or as that (thereby deriving (31) They didn’t know which model that we had
discussed).
However,
a different situation arises in (32). At the stage of derivation where C merges
with its TP complement, we have the structure (34) below:
(34) [CP [C uT, uWh] [TP which author
[T ø] wrote this book]]
Here,
which author carries both a
wh-feature and an uninterpretable tense feature (by virtue of being the subject
of a tensed T); hence the most economical way of erasing both the [uT] and the [uWh] EPP-features of C is to move which author into spec-CP as in (35) below:
(35) [CP which author [C uT, uWh]
[TP which author [T ø] wrote this book]]
The
alternative possibility of moving which
author to spec-CP to erase the [uWh]
feature of C and using do-support
(i.e. merging do in T and raising it
to C, where it can be spelled out as either did
or as that) is less economical and
therefore ruled out – thereby accounting for the ungrammaticality of *I
wonder which author did write this book (with did unfocused) and of (32) *I
wonder which author that wrote this book.
In
footnote 42, PT note that ‘Much the same facts were observed by Keyser (1975)
in Middle English relative clauses. Nominative who (an innovation in the middle period) was never followed by that, while non-nominative whom often was.’
6. Illicit use of the WHIP
One
particular aspect of PT’s analysis which seems to me to be problematic concerns
the illicit (as I shall argue) use they make of the Wh-Interpretation Principle/WHIP (15). Such a principle seems
questionable from both a descriptive and a theoretical stance. At a descriptive
level, questions arise about its observational adequacy in that (e.g.) contrary
to what WHIP would lead us to
expect, wh-subject structures like those in (36) below are interpreted as
exclamatives:
(36)(a) What a lot of people get drunk at
parties!
(b) What a lot of alcohol gets drunk at parties!
(c) What a lot of weirdos attend Minimalism workshops!
(d) How few people understand Minimalism!
(e)
How many students fall
asleep in syntax classes!
Moreover,
unlike the INT Principle posited in Chomsky (1999) under
which an expression moved to spec-vP via Object Shift is assigned a specific
interpretation INT (e.g. as
definite), WHIP clearly cannot be
universal. We can see this by looking at the syntax of wh-exclamatives in
Elizabethan English/EE, as illustrated by the following examples taken from
various plays by Shakespeare[7]:
(37)(a) How have they baffled thee! (Olivia, Twelfth Night, V.i)
(b)
How hath he been baited! (Princess, Love's
Labour's Lost, V.ii)
(c) How bright and goodly
shines the moon! (Petruchio, Taming of
the
Shrew,
IV.v)
(d) How sound is she asleep!
(Nurse, Romeo and Juliet, IV.v)
(38)(a) How my bones ache! (Nurse, Romeo and Juliet, II.v)
(b) How oddly thou repliest!
(Juliet, Romeo and Juliet, II.v)
(c) How I love thee! How I
dote on thee! (Titania, Midsummer Night's
Dream,
4.i)
(d)
How well my comfort is reviv’d by this! (Romeo, Romeo and Juliet,
III.iii)
As
these examples illustrate, wh-exclamatives in EE involved obligatory
wh-movement
but optional T-to-C movement. What would seem to be the most natural way of
accounting for the optionality of T-to-C movement in EE exclamatives within
PT’s framework is to suppose that C in such structures contains uninterpretable
[uT, uWh] EPP-features, so that (37a) has the structure (39) below at
the point where C merges with it TP complement:
(39)
[CP [C uT, uWh] [TP they [T have] baffled thee
how]]
The [uWh]
feature of C will attract how, and
the [uT] feature of C will attract
either the tensed auxiliary have
(deriving (40a) below) or the tensed subject they (deriving (40b) below)[8]:
(40)(a)
[CP How [C have uT, uWh] [TP they [T have]
baffled thee how]]
(b) [CP How they [C uT, uWh] [TP they [F have] baffled thee how]]
By contrast, root wh-questions in Elizabethan
English require obligatory movement of a finite auxiliary or verb to C, as
illustrated in (41) below:
(41)(a)
What dost thou say? (Othello, Othello,
III.iii)
(b) What sayst thou?
(Olivia, Twelfth Night, III.iv)
(c) Why do you look on me?
Why look you so upon me? (Rosalind, As
You
Like It,
III.v)
It therefore seems clear that the version of the
Wh-Interpretation Principle/ WHIP given
in (15) above cannot have operated in Shakespearean English in the same form as
in present-day English, since otherwise we would wrongly expect structures like
(40a) to be interpreted as interrogative only. Rather, it seems as if WHIP must have had the amended form
(42) below (the amendment being italicised):
(42)
Wh-Interpretation Principle for Elizabethan English/WHIPEE
A root CP with a wh-specifier is interpreted as
(i)
exclamative if it also has a non-wh specifier
(ii) interrogative or exclamative otherwise
Arguably, (42ii) is redundant since it in effect
says that there are no structural constraints on how a root wh-CP with a non-wh
specifier is interpreted; it is added here for clarity of exposition. It would
then follow that a structure like (40a) would correctly be predicted by (42ii)
to be ambiguous between an interrogative or exclamative interpretation, and
that a structure like (40b) would be correctly predicted to be interpretable
only as an exclamative[9].
What the discussion here illustrates is that
WHIP cannot be a universal interpretive principle. Indeed, at a theoretical
level, it could be argued that far from being a principle, WHIP is
simply an ad hoc LF-filter which
filters out overgenerated
LF-representations in much the same way as the
ad hoc surface filters posited in Chomsky and Lasnik (1977) filtered out
overgenerated surface structures; and resorting to filters of any kind could be
argued to reduce the computational efficiency of the grammar in the sense of
Frampton and Guttmann (1999)[10].
In the next section, we explore the possibility of developing an alternative filter-free account of the syntax of
root wh-clauses.
7. Abolishing the WHIP
An
alternative filter-free (WHIP-less) account would be the following. Let’s
suppose that the head T of TP in finite clauses contains not only tense features but also mood features (differentiating e.g.
between forms which are in the indicative/subjunctive/ imperative mood)[11].
Let’s further suppose that C has an uninterpretable EPP mood feature [uM] (but no [uT] feature) in root clause questions in present-day (and
Elizabethan) English, so that C attracts a constituent carrying a mood feature[12].
If T carries a mood feature but subjects do not, it follows that the
constituent attracted to C will be T, not the subject: we can thereby account
for the obligatoriness of auxiliary inversion in present-day English questions.
Under this account, a sentence such as Where
has Mary gone? would have the structure (43) below at the point where C is
merged with its TP complement:
(43) [CP [C uM, uWh] [TP Mary [T has]
gone where]]
The
[uWh] EPP-feature of C would attract
the wh-feature of where, and the [uM] EPP-feature of C would attract the
mood feature of has, so deriving (44)
below:
(44) [CP Where [C uM, uWh
has] [TP Mary [T has] gone where]]
By
contrast, if C carries neither a [uT]
nor a [uM] feature in present-day
English exclamatives, there will be no auxiliary inversion or movement of the
subject to spec-CP. An exclamative such as (9) What a silly book Mary has bought! will have the structure (45)
below at the point where C merges with its TP complement:
(45) [CP [C uWh] [TP Mary [T has] bought what a silly book]]
Since
the only EPP-feature carried by C is a [uWh]
feature (not a tense or mood feature), only the wh-phrase what a silly book will be preposed, not the auxiliary has or the subject Mary, so deriving:
(46) [CP What a silly book [C uWh] [TP Mary [T has] bought what
a silly
book]]
Conversely,
however, if a root C carries a [uT]
feature in Elizabethan exclamatives, there will either be T-to-C movement or
movement of the subject to spec-CP – as in (40) above[13].
The
account sketched here offers the advantage over PT’s analysis of entirely
dispensing with patently ad hoc
interpretive filters like WHIP and WHIPEE – at the cost of positing a set
of mood features which are independently motivated by morphological and
syntactic differences between indicative, subjunctive and imperative verb forms[14].
8. Another look at wh+that
structures
As
noted in §5, an interesting prediction made by PT’s analysis is that we do not
find wh+that structures in which the
wh-word is the subject of the that
clause. However, the generality of this conclusion is called into question by
examples such as the following (from Radford 1988, p.500):
(47)(a) It’ll probably be evident from the field
which of the players that are
feeling the heat most (Jimmy
Hill, BBC1 TV)
(b)
Jeg forfalte Jan hvem som var
kommet (Norwegian)
I asked Jan who that
had come
Moreover,
PT’s claim that Middle English relative clauses allowed wh+that structures only where the wh-word was not the subject of
the relative clause seems to be questionable. In order to verify their claim, I
checked the occurrence of relative clauses in the Chaucer’s Troilus and Criseyde (in the edition
edited by Brewer and Brewer 1969). By far the most frequently use relative
pronoun was which[15]:
in 13 of the relative clauses containing which
followed by that, which was the subject of the relative that clause, in 2 it was the direct
object of a transitive verb, and a further 2 it was a prepositional object
(with the preposition being pied-piped along with which). The relevant examples are given below (with a reference to
the book in question in Roman numerals and to the line in question in Arabic
numerals):
(48)
which as the subject of the relative
clause
(a) every peril which that is to drede (I.84)
(b)
his doughter, which that was
in gret penaunce (I.94)
(c)
This lady, which that alday herd at ere Hire fadres
shame... (I.106)
(d)
every wight (= person) which that to Rome went... (II.36)
(e)
he which that is my lord so deere (II.330)
(f)
Criseyde, which that herde hym in this wise...
(II.386)
(g)
Criseyde, which that well nigh
starf (= died) for feere... (II.449)
(h)
Criseyde, which that koude as
much good As half a world... (III.638)
(i)
Criseyde, which that al this wonder herde (III.799)
(j) For his love, which
that us bothe made... (II.500)
(k)
...save only Ector, which that
is the beste (II.740)
(l)
He which that nothing
undertaketh Nothing n’acheveth (II.808)
(m)
I love oon which that is most
ententif (= eager) To serven well
(II.838)
(49)
which as the object of a transitive
verb
(a)
every word which that she of hire herde... (II.899)
(b)
every tere which that Criseyde asterte (= shed)...
(III.1070)
(50)
which as the object of a pied-piped
preposition
(a)
swich a people...thorugh which that Troie most ben fordo (I.74)
(h)
a mirror...in which that ye me
se youre face a-morwe (II.404)
From
the examples given above, there seems little doubt that which is a relative pronoun since (when used as the object of a
preposition) it pied-pipes a preposition along with it under wh-movement. The
examples suggest that far from never occurring as the subject of a relative that-clause, this was by far the most
frequent use of which in Chaucer’s which that clauses.
Under
PT’s analysis, the relative clause in (48a) would have the simplified structure
(51) below at the point where C merges with its TP complement:
(51) [CP [C uT, uWh] [TP which [T is]
to drede]]
In
order to generate which that is to drede,
the [uT] feature of C would have to
attract is to move to C (where it
would eventually be spelled out as that),
and the [uWh] feature of C attract which to move to spec-CP, so deriving:
(52) [CP which [C uT, uWh that]
[TP which [T is] to drede]]
But
the problem posed by (52) is that such a derivation is ruled out by the Economy
Condition by virtue of the fact that there is a more economical derivation
available whereby movement of the tensed wh-pronoun which alone to spec-CP can erase both the [uT] and the [uWh]
features on C, as in (53) below:
(53) [CP which [C uT, uWh]
[TP which [T is] to drede]]
Hence,
(53) is the sole expected outcome from (51). However, while (53) would have
been grammatical for Chaucer, what remains unaccounted for is the fact that which that is to drede is also
grammatical. It would seem that a rather different set of assumptions are going
to have to be made from those made by PT.
For
example, a more traditional analysis would be to suppose that relative that is a complementiser directly merged
in C which carries a [uWh]
EPP-feature (but no [uT] feature) and
so attracts a (relative) wh-pronoun to move to spec-CP. On this view, the
relative clause in (48a) would have the structure (54) below at the point where
C merges with its TP complement:
(54) [CP [C uWh that] [TP which [T is] to drede]]
Movement
of which to spec-CP would then
derive:
(55) [CP which [C uWh that] [TP which [T is] to drede]]
and
(55) would be the structure of the relative clause in (48a)[16].
Of
course, an analysis such as (55) amounts to simply abandoning the key PT claim that the complementiser that moves to C rather than being
directly generated in C. In the next section, I explore the possibility of
developing an alternative analysis which would retain the idea that that moves to C while abandoning the
idea that that is a copy of a moved
auxiliary.
9. An alternative analysis
Having
outlined some of the problems posed by PT’s analysis, I now turn to outline an
alternative analysis of clause structure which incorporates some of PT’s
insights while avoiding the need to resort to that-doubling. In broad terms, I shall make the set of assumptions
in (56) below (which resemble those made by PT in spirit but differ in
important descriptive details):
(56)(i) Clauses have a CP/FP/TP structure in
which F is a head carrying an
interpretable finiteness
feature [F][17]
(ii)
C carries an uninterpretable finiteness feature [uF] in finite clauses
(iii)
The subject of a finite clause carries an uninterpretable finiteness
feature
[uF]
(iv)
Items like that/for are
finiteness markers which contain an interpretable
finiteness feature [F]
and originate in F
(v)
Finiteness markers like that/for
are spelled out overtly if moved, but have
a null spellout if they remain
in situ
The
assumption that between CP and TP there is an additional functional projection
FP is clearly far from new; for example Belletti (1990) posits an AgrS head
above TP, and Rizzi (1997) posits a Finiteness head below CP. In order to
illustrate how (56) is intended to work, below I look at the derivation of a number
of the sentences discussed earlier.
Consider
first how we might derive the that-clause
in (20b) Mary thinks that Sue will buy
the book. Given the assumptions in (56), this will have the following
structure at the point where C is first-merged with its FP complement (if
spec-FP is the canonical position for finite subjects[18]):
(57) [CP [C uF] [FP Sue [F that] [TP [T will] buy the book]]]
The
[uF] EPP-feature on C will attract
the closest finite constituent to C: given (56), that carries an interpretable finiteness feature, and Sue an uninterpretable finiteness
feature; both are equally close to C, so that C can attract either of them. If that moves to C (thereby erasing the [uF] feature on C), we derive:
(58) [CP [C that uF] [FP Sue [F that] [TP [T will] buy the book]]]
The
finiteness marker that is overtly
spelled out in accordance with (56v). By contrast, if the finite subject Sue moves to spec-CP (erasing [uF] on C) we derive:
(59) [CP Sue [C uF] [FP Sue [F ø] [TP [T will] buy the book]]]
The
finiteness marker in (59) is then given a null spellout in accordance with
(56v)[19].
One
aspect of the above analysis which might at first sight seem ad hoc is the condition in (56v) that
complementisers are given a null spellout unless they undergo movement.
However, (56v) arguably follows from a more general economy condition on
PF-representations to the effect that material is overtly spelled out only
where necessary. If a null head cannot serve as the antecedent of a trace[20],
it follows that if a complementiser moves, it must be spelled out overtly.
There are potential parallels here with do-support,
in that T has a null spellout in declaratives like They like syntax (where there is no T-movement) but is overtly
spelled out as do when T moves to C
in interrogatives like Do they like
syntax?[21] The
analysis outlined above also offers the advantage of obviating the need to
posit two distinct declarative clause complementisers, one realised as that and the other as ø: instead, ø is analysed as a null allomorph of that.
The
FP analysis outlined above can be extended from finite complement clauses like
those bracketed in (20) above to infinitive complements like those bracketed in
(60) below:
(60)(a) I would prefer [for Mary to buy the book]
(b)
I would prefer [Mary to buy the book]
If
infinitival to is a nonfinite T
constituent and the complementiser for
in (60a) and its null allomorph in (60b) are nonfinite F constituents, the
bracketed clauses in (60) will have the structure (61) below at the point where
the embedded C is merged with its TP complement:
(61) [CP [C uF] [FP Mary [F for] [TP [T to] buy the book]]]
For carries an interpretable
finiteness feature indicating that it is nonfinite; C has an uninterpretable (non)finiteness
feature, as does Mary (by virtue of
being the subject of a nonfinite clause). Since for and Mary are equally
close to C, either can be attracted by C. If for raises to C (thereby erasing the [uF] feature on C), the result will be (62) below:
(62) [CP [C for uF] [FP Mary [F for] [TP [T to] buy the book]]]
The
moved nonfiniteness marker will be overtly spelled out as for in accordance with (56v). By contrast, if Mary moves to spec-CP (erasing the [uF] feature on C), we derive:
(63) [CP Mary [C uF]
[FP Mary [F ø] [TP [T to] buy the book]]]
and
the nonfiniteness marker (by virtue of remaining in situ in F) is given a null
spellout in accordance with (56v)[22].
The nonfiniteness feature on the subject Mary
will be spelled out as accusative case
by default if we assume that nominals are spelled out as nominative forms if
finite, genitive forms in relevant (e.g. possessive) contexts, and accusative
forms otherwise (including if nonfinite)[23].
The
FP analysis outlined above avoids the pitfalls of PT’s that-doubling analysis while capturing some of the core intuitions
and insights underlying their analysis. For example, we can account for the that-trace effect in sentences such as
(64) below (repeated from (25) above):
(64)(a) Who did John say will buy the book?
(b)
*Who did John say that will buy
the book?
If
the embedded C carries both a [uF]
and a [uWh] feature (the latter being
required to license successive-cyclic wh-movement), the embedded clause will
have the simplified structure (65) below at the stage of derivation where C is
merged with its TP complement:
(65) [CP [C uF, uWh] [FP who [F that]
[TP [T will] buy the book]]]
The
wh-subject who contains both an
interpretable wh-feature and an uninterpretable finiteness feature; the
finiteness marker that contains an
interpretable finiteness feature. There are two ways of erasing the
uninterpretable [uF, uWh] features on C. One is by moving who on its own to spec-CP, so erasing
both the [uF] and [uWh] EPP-features of C and deriving (66)
below (in which the in-situ finiteness marker has a null spellout (rather than
being spelled out as that) in
accordance with (56v) above):
(66) [CP who [C uF, uWh]
[FP who [F ø] [TP [T will] buy the book]]]
Subsequent
merger and movement operations will eventually derive the structure associated
with (64a) Who did John say will buy the
book? and – as expected – this sentence is grammatical.
However,
an alternative possibility is moving that
to C in order to erase the [uF]
feature on C and moving who to
spec-CP in order to erase the [uWh]
feature on C, so deriving (67) below (in which the finiteness marker that is given an overt spellout in
accordance with (56v) above):
(67) [CP who [C that uF, uWh]
[FP who [F that] [TP [T will] buy the book]]]
Subsequent
merger and movement operations will eventually derive the structure associated
with (64b) *Who did John say that will
buy the book? However, the resulting sentence is ungrammatical because the
derivation in (66) is more economical than that in (67), since (66) involves
only movement of who to spec-CP,
whereas (67) also involves F-to-C movement. The overall conclusion is that (as
in PT’s own analysis) the that-trace
effect is a direct consequence of economy considerations, and does not require
any additional ad hoc apparatus[24].
10. Potential pitfalls or
fatal flaws?
Having
discussed in some detail three potential problems posed by PT’s analysis
(relating to their account of wh-interpretation,
wh+that structures and that-doubling), and having suggested an
alternative analysis within the spirit (though not the letter) of their
proposals in section 9, in this final section I briefly comment on some further
apparent problems, leaving for future research the question of determining whether
these are potential pitfalls or fatal flaws. One problem touched on briefly in
footnote 5 relates to their inadequate formal definition of closeness. A further problem relates to
their analysis of nominative case as
an uninterpretable tense feature, which raises the question of how nominative
subjects in the Portuguese inflected infinitive structures described by Raposo
(1987) are to be accounted for. Their assumption that the wh-subject moves to
spec-TP but the auxiliary remains in T in sentences like Who is telling the truth? raises questions about how we account for
the possibility of is cliticising
onto who in Who’s telling the truth? if cliticisation requires string-adjacency
and there is a null C intervening between who
(in spec-CP) and is (in T). Their
claim that the subject is in spec-CP (rather than in spec-TP) in exclamatives
raises the question of how we account for the pre-subject position of sometimes in sentences like How many of my lectures on Minimalism
sometimes none of the students seem to understand! if sometimes is a TP-adverb (as it appears to be in sentences like I have the feeling that sometimes none of
the students understand my lectures). Moreover, PT’s analysis of
wh-movement
and T-to-C movement (while purporting to use the technology of Chomsky 1998)
raises the question of what uninterpretable features a moved
wh-phrase
or moved auxiliary carry in order to be active.
I leave the question of whether these are apparent or real imperfections for
future research.
11. References
Belletti,
A (1990) Generalised Verb Movement:
Aspects of Verb Syntax, Rosenberg
& Sellier, Turin
Brewer,
D S & Brewer L E (1969) Geoffrey
Chaucer: Troilus and Criseyde,
Routledge and Kegan Paul, London
Chomsky,
N (1998) Minimalist Inquiries, MIT
Working Papers in Linguistics,
no.15
Chomsky,
N & Lasnik H (1977) ‘Filters and Control’, Linguistic Inquiry 8: 425
-504
Frampton,
J & Gutmann, S (1999) ‘Cyclic computation, a computationally efficient
minimalist syntax’, Syntax 2: 1-27
Kayne,
R (1980) ‘Extensions of binding and case-marking’, Linguistic Inquiry
11:75-96
Keyser,
S J (1975) A partial history of the relative clause in English’ in Grimshaw,
J (ed) Papers
in the History and Structure of English, UMass Occasional Papers
in Linguistics 1, GSLA, Mass
Koopman,
H (1983) ‘ECP effects in main clauses’, Linguistic
Inquiry 15: 211-258
Perlmutter,
D (1971) Deep and Surface Structure
Constraints in Syntax, Holt
Rinehart and Winston, New York
Pesetsky,
D & Torrego E (2000) ‘T-to-C movement: Causes and consequences’, to
appear in M Kenstowicz (ed) Ken Hale: A Life in Language, MIT Press,
Cambridge Mass.
Radford,
A (1988) Transformational Grammar,
Cambridge University Press
Raposo,
E (1987) ‘Case theory and INFL to COMP; the inflected infinitive in
European Portuguese’, Linguistic Inquiry 18: 85-109
Rizzi,
L (1990) Relativised Minimality, MIT
Press, Cambridge Mass
Rizzi,
L (1997) ‘The fine structure of the left periphery’ in Haegeman, L (ed)
Elements
of Grammar, Kluwer, Dordrecht, pp. 281-337
Roberts,
I (1998) ‘Have/Be Raising, Move F and
Procrastinate’ Linguistic Inquiry
29: 113-125
___________________________________________________________________
[1] I am grateful to members of the LAUG Research Group and Minimalist Syntax Research Group at Essex for helpful discussion of an earlier draft of this paper, and in particular to Claudia Felser, Bob Borsley and Zeljka Paunovic. This version appeared in Essex Research Reports in Linguistics (November 2000), vol 33, pp.21-32.
[2] Labelled bracketings are simplified by showing only those categories, traces and features most relevant to the discussion at hand, and by not showing intermediate projections. Features are italicised: features containing a prefixed u are uninterpretable, others are interpretable; in general, only the EPP-features of C are shown.
[3] PT rule out the possibility of [uT] on C triggering second-merge of TP with C by a constraint against mutual merger to the effect that X cannot second-merge with Y if Y first-merged with X.
[4] A similar argument might be formulated in relation to wh+ever expressions on the basis of contrasts such as:
(i) Whatever did he say to you?
(ii) *He said whatever to you?
(iii) Whatever happened to Idi Amin?
[5] The precise definition of closeness which they propose is problematic in a number of respects for technical reasons which I will not go into here. For present purposes, it is sufficient to assume that the head and specifier of a subordinate projection are equally close to a superordinate (c-commanding) head, setting aside the issue of how closeness is to be formally defined (though a natural candidate would seem to be a relation based on m-command).
[6] The term and acronym used here are my own; PT do not give any specific name to the principle they invoke.
[7] Claudia Felser points out to me that similar facts obtains for present-day Modern Standard German, so it would seem that WHIP cannot hold for German either (thereby further undermining any hope of positing that it is universal).
[8] It may be that the optionality of T-to-C movement in structures in which a focused constituent moves to spec-CP in Shakespearean English can be handed in much the same way as the optionality of movement to C in exclamatives.
[9] An interesting consequence of (42) is that
it predicts that Shakespearean English allows wh-subject exclamatives such as
those below:
(i) What a number of men eat Timon!
(Apemanthus, Timon of Athens,
I.ii)
(ii) What an unkind hour is guilty of this
lamentable shame (Friar Laurence, Romeo
and Juliet, V.iii)
By contrast, the wh-interpretation principle (15) would rule out wh-subject exclamatives like (50) in present-day English.
[10] Attempting to maintain that (15i/42i) is universal whereas (15ii/42ii) is parameterised does nothing to solve the overgeneration problem or change the essentially stipulative character of the postulated principles – principles which are furthermore specific to root clauses (and not applicable to complement clauses).
[11] In fact, the key point of the analysis outlined below is that T should carry two distinct types of feature, one of which is shared by its subject and the other of which is not. For the logic of the argument to go through, it could equally well be the case (e.g.) that T carries finiteness and tense features, and that its subject carries an uninterpretable finiteness (rather than tense) feature. See §9 for an analysis along these lines.
[12] In this connection, note that Roberts (1998) posits that inversion in root-clause questions in English involves C attracting a T which carries an interrogative mood feature.
[13] An alternative account of Elizabathan exclamatives is to posit that C carries an obligatory [uWh] feature and an optional [uM] or [uT] feature – but I shall not explore these possibilities further here.
[14] Among the syntactic differences between indicative/subjunctive/imperative verb-forms are those illustrated below:
(i) I insisted
[that he was not allowed to return]
(ii) *I insisted
[that he be not allowed to return]
(cf. ...not be allowed to return)
(iii) *Be not
afraid! (cf. Don’t be afraid!)
[15] PT’s footnote refers specifically to who, though the point they make would be expected to generalise to other relative pronouns which can be used as subjects. The following two examples of who that structures occur in the text, both involving who used as a subject free relative pronoun (paraphraseable as ‘he who’) followed by that:
(i) Who that hath
an hed of verre, Fro caste of stones war hym in the were (II.868)
(ii) And who that
giltif is, al quyt goth he (III.1019)
[16] Since present-day English does not allow wh+that relatives, we might suppose that in present-day English the relative complementiser that can only attract a null wh-operator, whereas conversely a null relative complementiser can attract either an overt or covert wh-operator. This would provide one way of accounting for the Multiply Filled COMP filter of Chomsky and Lasnik (1977) – at least, in respect of relative clauses.
[17] It is not crucial to my proposal that the feature F be a finiteness feature rather than e.g. a mood feature.
[18] In the same way as spec-AgrSP was the canonical subject position in Agr-based accounts of clause structure.
[19] As Zeljka Paunovic points out, a question that remains to be answered in relation to (58/59) is how the uninterpretable finiteness feature on Sue is erased. One possibility is that it is marked for deletion by F, though the [uF] feature on Sue remains active until the end of the CP phase, and hence it can be attracted by the [uF] EPP feature of C; another is that it is deleted by C – though this requires in the case of (58) that the [uF] feature of C can attract that and also erase the [uF] feature on Sue. Difficult questions arise, but I will not pursue them here. An analogous problem arises in respect of Mary in (63) and who in (67).
[20] More precisely, a null minimal projection cannot undergo movement and hence cannot serve as the antecedent of a trace. Null operators (by virtue of their status as maximal projections) are therefore exempt from this requirement.
[21] In negatives like They don’t like syntax, it may be that do is inserted to serve as a host for n’t. An alternative possibility is that do originates below the negative particle (perhaps in T) and moves into a superordinate head position (perhaps to F), with NEGP positioned between F and T.
[22] The grammaticality of sentences like I want the boys definitely for to be there in Belfast English (See Henry 1995, p.97) suggests that for does not raise to C (since for cannot cross definitely) but that the infinitive subject the boys does raise to some position above for. It may be that to can cliticise onto for in such cases, preventing for from being deleted (if clitic to requires an overt host). If NEGP is positioned between F and T, to must move across (or through) NEG in sentences like I would prefer them for to not go.
[23] I see no reason to adopt the suggestion made by PT (fn. 65) that ‘we treat the subject of these infinitives as nominative’. The fact that Mary cannot passivise when moved to spec-CP (as we see from the ungrammaticality of *Mary would be preferred to buy the book) can be attributed to improper movement of Mary from spec-CP to
spec-TP – i.e. from an A-bar position to an A-position.
[24] Although offering some of the advantages of PT’s analysis, the revised analysis in section 9 also brings with it some of the disadvantages of their analysis. For example, it is not obvious how Chaucerian wh-subject+that structures like those in (48) are to be handled within the revised analysis; perhaps that remains (and can be spelled out) in situ in Chaucer – but this is clearly a speculative claim and alternatives clearly need to be explored.