Alexiadou, Artemis, Paul Law, Andr� Meinunger and Chris
Wilder, eds. (2000) The Syntax of Relative Clauses,
Amsterdam and Philadelphia: John Benjamins, hardback, vi,
395 pp., Linguistics Today #32, ISBN 90 272 2753 5
(Europe) / 1 55619 916 3 (USA).
Georges Rebuschi, Sorbonne Nouvelle & CNRS (Paris)
This book is an outgrowth of a conference on Relative
Clauses (RCs) organized by the editors and hosted by the
Zentrum f�r Allgemeine Sprachwissenchaft (ZAS) in Berlin
in 1996: a first version of six of its nine chapters were
first presented there. Most of the contributions address
the Head Raising Analysis (HRA), originally due to
Vergnaud and readvocated in Kayne (1994), according to
which the nominal "head" (note 1) of the DP raises from
within the embedded IP to the Spec,CP position. From this
point of view, the book under review has two remarkable
features. First, a long introduction by the editors (1-51)
provides an excellent and detailed summary of most of the
pros and cons of the HRA as compared to the more classical
analysis according to which a restrictive relative clause
(RRC) is a CP right-adjoined to the NP that is itself the
determiner's complement (Bianchi's and Grosu's chapters
also contain lucid accounts of Kayne's theory). Second, a
wide variety of natural languages is covered: Brazilian
Portuguese, Dutch, English, Hindi, Japanese, Latin,
Romanian, Swahili, Swedish and Turkish, to mention the
main ones from which the various chapters borrow their
crucial data.
I will first summarize each of the individual chapters,
and will next discuss some salient aspects of the various
treatments of RCs proposed therein.
1. THE BOOK'S CONTENTS
1.1. Valentina Bianchi, in 'Some Issues in the Syntax of
Relative Determiners', 53-81, addresses two questions
which are particularly conspicuous in Latin, Ancient Greek
and quite a few other Indo-European languages, namely, (i)
the analysis of what is descriptively known as
"Correlative Constructions" in certain grammatical
traditions, and (ii) the dual phenomena of Case Attraction
and Inverse Case Attraction. V.B. clearly shows that the
Kaynean approach to RC construction renders the second
problem much easier to deal with, because, on the standard
analysis, there is no independently established
relationship between the specifier of an adjunct (what the
relative pronoun is in that standard theory) and the XP
within which it is adjoined to a lower YP. More
specifically, she suggests that the Relative Determiner
(Rel.D) undergoes a further raising operation from within
Spec,CP (the landing position of the relative "head" under
the HRA) to the Specifier position of the full DP,
wherefrom case information can be exchanged with D� under
Spec-Head Agreement (note 2).
In correlative sentences (cf. Srivastav 1991, and
Mahajan's chapter reviewed in 1.5 below), a CP which
contains either a relative pronoun or a Rel.D associated
with a "head" NP may appear in a left peripheric position,
and be coindexed with a correlative pronoun or a full DP
(i.e. a DP that may also contain the nominal "head"). V.B.
reinterprets Haudry's (1973) diachronic work on Ancient
Indo-European correlatives as the source of standard RRCs
as follows: the internal structure of the correlative
protasis is that of a relative clause � la Kayne, i.e. its
"head" NP occupies the Spec position of the Rel(ative) DP
which has itself raised to Spec,CP (see above). The
reanalysis then simply requires that a functional nominal
projection be created above the dislocated CP, "turning it
into a nominal phrase which can directly occupy an
argument position within the main clause (whence the
elimination of the correlative pronoun in the latter)" (p.
66).
1.2. In 'Type-Resolution in Relative Constructions;
Featural marking and dependency encoding', 83-119,
Alexander Grosu argues "that the search for a purely
configurational characterization of the various semantic
subtypes of relatives is likely to prove quixotic even
with respect to individual languages, and a fortiori with
respect to UG" (p. 115). Instead, the author seeks to
characterize the three basic types of relative
constructions: RRCs, Appositive RCs (ARCs), and
Maximalizing RC (MaxRCs) -- such as described in Grosu &
Landman (1998) -- in terms of both structure and featural
content of the C� head of these clauses. More
specifically, this head would always contain a feature
[REL] whose semantic import is (i) that the complement IP
include at least one unbound variable, and (ii) that this
variable "be consonant with" the syntactic category and
logical type of the phrase that contains the RC (where
"consonant with" basically means "identical", but also
allows for categorial differences limited to the level of
the pertinent extended projection and/or those concerned
with logical "equivalent classes" such as individuals and
generalized quantifiers).
The three basic types, besides sharing the same [REL]
feature, would also differ in featural content. The MaxRCs
(among which, following Grosu & Landman (1998), A.G.
includes correlative protases) would be characterized by a
feature [MAX] -- the simple encoding of the fact that "in
the absence of a theory of maximalization (which, to the
best of my [=A.G.'s] knowledge, does not exist at the
moment), a left-adjoined configuration rules out any
conceivable construal other than maximalization" (p. 102).
Turning to the difference between RRCs and ARCs, the
discussion includes an interesting typology of various
analyses, which can be subsumed under the four classes
obtained by the combination of two independent features,
namely [+/- Overt] and [+/-Constituent]: "Overt" refers to
the locus of configuration distinction (before s-
s/Spellout or at LF) between RRCs and ARCs, and "Const."
to whether an ARC is supposed to form a constituent with
its antecedent or not. Thus, Jackendoff's (1977) treatment
of RCs is [+O,+C], Emonds's (1979) is [+O,-C], Kayne's
(1994) is [-O,+C] and, finally, Bianchi's (1995) revision
of Kayne's analysis is [-O,-C], since the RC must
eventually leave the DP and adjoin to some functional head
of the main clause -- thereby (hopefully) circumventing
scopal problems.
In his conclusion, the author, like the editors in their
Introduction, does not really takes sides for or against
the HRA, but notes that, if, on the one hand, certain
examples of reconstruction exist for which a movement
analysis seems implausible (as in pseudo-clefts like 'the
individuals that John and Mary dislike most are each
other'), on the other hand, some contrasts between
comparative constructions and degree relatives, already
noted by Carlson (1977) would be easily accounted for in
terms of the HRA.
1.3. 'Some Syntactic and Morphological Properties of
Relative Clauses in Turkish', 121-159, by Jacklin
Kornfilt, addresses the question of the distribution of
the two distinct nominalizers that appear in (standard
modern) Turkish RCs, /-(y)An/ and /-DIK-/ in more
"traditional" terms, those of the GB theory. Contrary to
traditional approaches, it is not those nominalizers as
such that are considered to be at stake, but rather the
fact that the first one disallows subject/possessor
agreement, whereas the second requires it: J.K.'s strategy
consists in identifying the factor that determines that
subject relativisation is incompatible with agreement,
whilst non-subject agreement demands it. Form this point
of view, she notes that resumptive pronouns are impossible
in simple RCs (whether it is their subject or object that
is the target of relativisation), but appear (sometimes as
realized pronouns, sometimes as pro) in complex relatives
such as the (grammatical) Turkish counterpart of 'the
captain<x> (such) that I bought a house on the island<y>
such that he<x> invited me (to y)' (J.K's ex. (8a)).
The solution lies in the exploitation of Generalized
Binding (after work by Aoun, Li and others), which
constrains pronouns with respect to non-argumental (or A')
potential (anti-)antecedents as well as with respect to A
positions: if pronouns must be A'-free in their minimal
Complete Functional Complex -- a CP by hypothesis (note
3), then resumptives will be excluded in simple RCs, since
an abstract operator would A'-bind them. Next, borrowing
from Jaeggli the idea that "if an empty category is
licensed and identified by AGR, it must be pro" (p. 134),
the author is able to nicely tie the absence of agreement
in subject-relativized RCs to the barring of resumptive
pronouns in the Generalised Binding framework -- whence
the appearance of the nominalizer /-(y)An/ which, as
noted above, disallows agreement.
The rest of the chapter is devoted to two other types of
RC configurations: complex RCs with targets within larger
subjects (in which case, as might be expected, it is also
the nominalizer /-(y)An/ which appears), cf. 'the
director<x> (such) that (it) was heard that (he<x>) was
going to fire the teacher' -- and RCs with non-subject
relativized positions in clauses with non-thematic, or
expletive, subjects, where, contrary to expectation now,
it is again /-(y)An/ which turns up, as in 'the stop
(such) that it-is-boarded on-the-bus (there)'. To deal
with the first case, the author must assume the existence
of another functional projection, between the
nominal(ized) AgrP and the nominal(ized) CP, viz. TopP:
the whole subject would move to the Spec position of this
projection, and from there, the operator would raise to
Spec,CP. But if the nominalizer is /-DIK-/, a violation of
the generalized Binding Condition B will ensue: just as in
the first part of the paper, the /-(y)An/ morpheme is
selected to salvage the construction (no agreement, hence
no pro -- consequently the empty A position must be a
variable). In the second case, J.K. has to postulate the
coindexation of the abstract OP and C� -- a mechanism
reminiscent of Pesetsky's treatment of the 'que'/'qui'
alternation in French -- so that the C� will govern the
(empty) subject position: a chain will be formed which,
since the construction is impersonal, will result in
reanalyzing the targeted object DP as a subject, so that
the marked nominalizer must be used again.
1.4. Paul Law, in 'On Relative Clauses and the DP/PP
Adjunction Asymmetry', 161-199, deals in a highly original
manner with the long-standing question raised by the
obligatoriness of preposition piedpiping illustrated by
the pair *'the man who(m) to discuss linguistics with' vs.
'the man with whom to discuss linguistics'. If it can be
shown that 'whom' and 'with whom' in these examples do not
sit in a Spec,CP position, but are adjoined somewhere
lower, the account will follow naturally (compare the
grammatical 'I wondered who(m) to discuss linguistics
with', where the Wh-expressions are interrogative, and
where, crucially, the CP is selected by the governing verb
- a situation which never obtains with relative clauses,
as the author stresses). P.L.'s idea is that relative
pronouns (i.e. DPs) and PPs containing them occupy an
adjoined position: either to IP, if the clause is finite,
or to an infinitival VP (headed by 'to' analyzed as a V�,
following arguments by Pullum 1982 and others), in the
cases under discussion. A slight extension of Emonds'
well-known Structure Preserving Constraint, whereby "what
is required of [a] landing site is that it be able to host
a phrase by base-generation independently" (p. 188) would
then suffice to allow PPs (even when they contain a
relative pronoun) to be adjoined to a VP, and to prevent
DPs from undergoing such a movement, since they are always
excluded from such a position. Naturally, the force of his
arguments in favor of a V� analysis of infinitival 'to'
is crucial (note 4).
1.5. In 'Relative Asymmetries and Hindi Correlatives',
201-229, Anoop Mahajan endeavours to show that the three
types of RCs to be found in this language: those that
appear within the restricted DP, the right-extraposed
ones, and the correlatives (or left-adjoined RCs), are
compatible with the HRA, provided it is supplemented by
the copy theory of movement due to Chomsky (1995), and
provided that what is elsewhere treated as Wh-movement is
taken to be an instance of Scrambling -- an independently
justified property of Hindi grammar. Among others, the
author discusses Srivastav's (1991) influential paper that
reduces the three types to only two -- the base-generated
correlatives on the one hand, and the "canonical" RRCs,
which she considers to be the source of the postposed
variant. Much more explicitly than many of the other
contributors to the volume who advocate some version or
other of Kayne's views, A.M. holds that, "on purely
conceptual grounds, we [should] prefer theories that adopt
one single mechanism to account for discontinuous
dependencies rather than allowing for base generation and
movement as two coexistent mechanisms for this purpose."
This methodological stance entails that those correlatives
which contain two Wh-expressions (note 5), must be left
outside of the scope of the study (p. 212), and that if
the sort of deletion associated with Chomsky's theory of
movement as copy plus deletion must be made under phonetic
identity, it must also be done independently of
constituent structure.
1.6. Keiko S. Murasugi's chapter, 'An Antisymmetry
Analysis of Japanese Relative Clauses', 231-263, is
another effort to deal with the properties of an "exotic"
language within the Kaynean frame. The results may,
however, look paradoxical. First, the author maintains her
view (cf. Murasugi 1991) that Japanese RCs are IPs rather
than CPs, a non-movement analysis which is consonant with
Kayne's remark that Japanese is unique among languages
that have prenominal RCs in that those RCs are fully
inflected (note 6); second, she shows that an analysis
according to which an IP would raise to Spec,DP after the
"head" NP has been raised to Spec,CP would lead to
problems with respect to the trace within IP, whence the
hypothesis that, under a relaxed version of Kayne's theory,
which only proposes a Det + CP hypothesis (the DCPH hereafter
- note 7), the NP should be base-generated in Spec,CP, the
trace being replaced by a pro, as in (1), her (66).
(1) [dp [ip... pro<i>...]<j> [d' D�[cp NP<i>
[c' C� t<j>]]]]
Next, K.S. suggests more simplifications: (i) instead of
IP being raised to Spec,DP, it could also be base-
generated there; (ii) the CP complement of D would now
contain the NP in its base-generated position, and the
(unrealized) C having have no complement, "[its] structure
plays no role" (p. 256), whence a further reduction, to
(2):
(2) [dp [ip... pro<i>...]<j> [d' D� NP<i>]]
(iii) Finally, since the "aboutness" relation (typical of
topicalized (wa) DPs) is also sufficient to license
Japanese RCs, as in (lit.) 'physics, which is hard to get
a job', and since Japanese allows (preposed) complement
clauses much more freely than e.g. English, as in (lit.)
'the sound (that) a door shuts', K.S. suggests that
Japanese might, after all, have no RCs at all, and would
be satisfied to use "pure complex NPs" [or DPs] where
other languages distinguish them radically from complement
clauses (note 8).
Another very interesting suggestion, quite in harmony with
this general result, is that Japanese would not either
have Internally "Headed" RCs (or IHRCs), the author's main
argument being that those phrases are in fact adverbial
clauses, as shown (between others) by the facts that (i)
they can usually be paraphrased by constructions which
contain the (partially delexicalized) word 'tokoro',
'place', which contributes a circumstantial reading, and
(ii) that, just like those explicitly adverbial phrases,
they cannot be passivized.
1.7. Christer Platzack is another scholar who seeks to
defend Kayne's general approach without adopting the
specific HRA advocated in the 1994 book. In 'A Complement
of N� Account of Restrictive and Non-Restrictive
Relatives; The case of Swedish', 265-308, he rejects a
priori any analysis that would not abide by Kayne's Linear
Correspondance Axiom (LCA) and would therefore allow the
traditional adjunction-to-NP description (under the DP
hypothesis). It follows that if some arguments can be
given against the HRA, the only solution left will be to
treat RRCs as N� complements -- and C.P. does have one
argument, provided by the fact that the reflexive
possessive 'sin(a)' does not show reconstruction effects
in RCs as it does in the case of topicalisation, which is
more uncontroversially considered an instance of movement.
As a consequence, the "head" of the construction would not
be an NP (as in Kayne's book) or a functional projection
thereof (as in Bianchi 1995 -- see Zwart's chapter below),
but a bare N�; moreover, an empty operator would have to
be reintroduced, which would move from within the embedded
IP/AgrSP to Spec,CP, the structural complement of the
"head".
As for ARCs, the author also choses an alternative
analysis to Kayne's, which, recall, is supposedly
structurally identical to that of RRCs, the only
difference lying in the presence of an (ill-identified)
feature that would both trigger the further raising of the
"head" to Spec,DP, and determine the special prosodic
contour of the sentence that contains an appositive RC.
Platzack suggests that, in this case, the "head" is now a
DP sitting in the Specifier position of the NP which is
the complement of the D, the N� itself being empty. The
difference between the two structures, that of RRCs and
that of ARCs, draws support, according to the author, from
an asymmetry in extraction facts: since the "head" in the
former case is an N�, the Spec,NP position can function as
an escape hatch, but this very same position being filled
by the "head" of the ARC, no such escape hatch is
available.
1.8. Cristina Schmitt's empirical problem in, 'Some
Consequences of the Complement Analysis for Relative
Clauses, Demonstratives, and the Wrong Adjectives', 309-
348 is constituted by the fact that certain definite DPs
require a RC to be grammatical, as in (3):
(3) a I bought the type of bread *(you like)
b Mary made all the headway *(she could make)
c John bought the *(wrong) type of house
The author's solution (contrary to Murasugi's, for
instance) is not even based on a non-movement
implementation of the DCPH, but more directly on Kayne's
LCA itself, which simply imposes binary branching towards
the right, thereby excluding right-adjunction and reducing
left-adjunction to a single Specifier position in any
phrase. C.S. thus proposes that the "head" is not IP
internal (and moved to Spec,CP), but is an indefinite
Number Phrase (NumP) base-generated in Spec,AgrP, a
functional layer located in between D(P) and C(P) --
hence, the movement of an empty operator must also be
postulated.
What this complex syntactic set up does is to allow one
to derive what the author calls the "Determiner Transparency"
that characterizes the examples above. Borrowing from
Higginbotham the idea that Determiners theta-bind N's (NP
within the DP framework) or discharge the <R> role of
those DPs (note 9), the structure thus built allows
external constraints to be respected, whatever they are;
in particular, an idiom chunk like 'headway' in (3b) must
be indefinite, but since the D may bind the CP (given that
finite clauses may fulfil this role, owing to the neo-
Davidsonian event role "e" they contain), its definiteness
is satisfied, and so is the indefiniteness requirement on
the NumP 'headway', which is bypassed by the theta-binding
relation under discussion.
An extension of the system is proposed wrt. aspect; as is
well-kown, aspect is compositional, in the sense that
certain properties of the direct object are directly
relevant to its computation; if the NumP raises out of DP
to Spec,AgrOP, it will then be able to contribute an
indefinite reading to the verb, this contribution being
necessary to account for the grammaticality, and durative
interpretation, of sentences like 'Peter killed the
rabbits that ate his plants for two years' in Brazilian
Portuguese (of course, with the adverbial PP 'in one
hour', the whole DP contributes its definiteness to the
terminative reading with which such an adjunct is
compatible).
1.9. Jan-Wouter Zwart's contribution, 'A Head Raising
Analysis of Relative Clauses in Dutch', 349-385, builds on
Bianchi's modification of Kayne's analysis. First, given
that complementizer agreement is a "morphological reflex
of movement to C of a lower functional head" (p. 363) and
that Dutch dialects display distinct types of Comp, and
Comp agreement, he shows that the left-peripheral field
should be split into three (heads and) projections, first
noncommittally labeled C1�, C2� and C3�. The lower heads
would correspond, respectively, to W-specifiers, and D-
specifiers (note 10). Swahili is next proposed as another
instance of a natural language that shows that at least
two distinct functional heads can be realized, as in the
'amba-+Relative suffix + kwamba' constructions described
by Barrett-Keach (1985). In both languages, the Rel.DP
would raise to Spec,CP3 or spec,CP2, depending on the
syntactic feature (D- or W-) of its Rel.D.
As for the highest head (C�1) the author suggests that it
is semantically motivated: the NP "head" would leave the
relative DP and undergo a further operation of raising to
Spec,CP1; a configuration would ensue in which this "head"
would in fact be interpreted as the Restrictor of the
domain quantified by the D head of the whole DP, thereby
allowing for a compositional semantic interpretation of
the whole structure (note 11).
2. CRITICAL EVALUATION
2.1. The reader will have noticed that the technical
solutions presented in the various chapters of this volume
are not altogether compatible with each other. However, in
my opinion, this is not a short-coming of the book: it is
rather a reflex of the inherent methodological
difficulties of the task of theoretical linguists, if we
view their enterprise not as a one global and rigid
"scientific program" or "paradigm", but as a series of
partly independent and competing ones, each of them
resting on many common assumptions, but not exactly the
same ones, and/or ranking them more or less differently.
2.1.1. Moreover, the mere description of relative clause
types across languages is probably more complex than Grosu
chose to spell out in his chapter: if there is a least one
fairly general and clear-cut case of non-correlative
"sortal external" relativization, to use the latter's
expression (but this relativization type is not universal,
since certain languages only have IHRCs), it is the case
of subject relativization (as shown by various Malayo-
Polynesian languages, such as Malay and Malagasy, which
only permit to target the subject position or argument);
but there are at least two main dimensions along which it
is possible to enrich the grammar of a given language. On
the one hand, we've all learned from Keenan and Comrie's
work that the functions of the relativised element are
ranked along a (more or less) universal hierarchy, SUBJ. >
D.O. > I.O. > ADJUNCT > Standard of Comparison -- and
Kornfilt's study clearly shows that taking this dimension
into account is unavoidable, since the Turkish nominalizer
is different according as the target is the subject of the
RC or not. On the other hand, Generative Grammarians
working on Subjacency, Islands, and Resumptive Pronouns,
the licensing of Parasitic Gaps, etc., have also shown
that the relationship between the (at least superficial)
external "head" or (possibly invisible) Det of the DP that
contains a RC, and the RC itself, is an important factor
of variation: besides entertaining a local relationship
with the head (note 12), the RC may a priori be embedded
in the clausal complement of a bridge verb, the sentential
complement of a non-bridge verb (?), or else in an adjunct
clause, or even in another relative clause, as, again,
illustrated in Kornfilt's paper -- see her ex. (8a), cited
in 1.3.
Moreover, a careful analysis should take into account:
(a) the nature of the relativized position: an empty
category (whence a pro or a trace, see Murasugi's chapter
or Kornfilt's again), a resumptive pronoun, or possible an
"intrusive" one (cf. Sells (1984));
(b) the type of complementizer: visible or not, and, if
visible, specific or not;
(c) the type of relative pronoun: visible or not again,
and, if visible, specific, or akin (or identical) to
interrogative pronouns, etc. (note that points (ii) and
(iii) also offer the option of allowing, or forbidding, a
doubly-filled "COMP").
Finally, there also are, among RRCs, IHRCs and
correlative protases -- and possibly, as suggested by Grosu,
"existential relatives", as in those languages which
display constructions like "I don't have what to eat", if
they are not a subtype of free relatives.
2.1.2. Returning to the theoretical issues, it is clearly
difficult to contrast papers which differ in their
responses to the following questions:
(a) Should movement be considered costless, or more
costly than base-generation?
(b) Should the multiplication of functional heads
(cf. the Split CP hypothesis, entertained by several
contributors) be preferred to the multiplication of
specifiers (as suggested by Chomsky)?
(c) What role, if any, is a *compositional*
constraint on the interpretation of linguistic structure
(be it s-structure or LF) relevant to an author's choice
between alternative analyses?
(d) Should empirical coverage in one language rank
higher than coverage of a narrower domain in a wider
selection of languages?
Trying to locate each individual contribution in such
a multi-dimensional space would be otiose -- hopefully,
the brief summaries provided in section 1 (or, much
preferably, the careful perusal of the texts themselves)
will help the readers classify the chapters along those
lines or others they might prefer. Therefore, in the
remainder of this review, I will concentrate on a few
points that seem to me to be particularly revealing.
2.2. As indirectly indicated in 1.6, 1.8 and 1.9, there
are four basic options if one chooses to deal with RRCs
within a Kaynean frame: besides the HRA itself, it is also
possible to base generate the head in Spec,CP (a view
compatible with the DCPH, as recalled in note 7), to
introduce a functional projection -- for instance a NumP
- as complement to D (this NumP taking the CP as its own
complement, as in Schmitt's paper), and, finally, to
consider that RRCs are structural complements of N� --the
last two options respecting at least the general
Antisymmetry requirement.
When *counterparts* of (English, Italian, etc.) RRCs are
analysed as Noun complements, there remains, however, the
problem of the syntax-semantic interface, since it is
difficult to see how a head and a complement could
possible yield an intersective interpretation. In the case
of Murasugi, however, some intuitive arguments are given
according to which, at least as far as Japanese is
concerned, some clausal "expansions" of nouns may well
blur the divide between completive and relative clauses --
but nothing of the sort appears in Platzack's paper. In
fact, if this latter author had tried a base-generated
approach along the more general DCPH, instead of sticking
to the strict HRA, he might possibly have found it easier
to deal with the absence of reconstruction effects with
anaphoric possessives (see 1.8).
2.3. In a sense, one might regard Bianchi's implicit
description of correlatives clauses as base-generated
outside of the main clause as a fifth way of building
"sortal external" relatives (note 13). In this connection,
however, note that, from an Antisymmetry point of view,
the correlative relative cannot be left-adjoined in the
traditional sense, but must be analyzed as either base-
generated in, or raised to, the Specifier of some
functional projection. From this point of view, it is
interesting to note that complex sentences incorporating a
correlative protasis and an apodosis on the right of it
may, in some languages, globally function like completive
clauses -- in which case the correlative clause cannot sit
in Spec,CP; now, some of these languages happen to display
"conjunctions" which are prosodically, and sometimes
morphologically, *integrated* in the apodosis or "main"
clause (e.g. Northern Basque, Hittite, or Gothic -- see
Rebuschi 1999 for details): in such cases, the LCA is
totally respected, at least if these conjunctions are
regarded as the head of the complex construction.
Correlatives should be taken more seriously for two
more reasons. First, as written texts and personal
fieldwork has taught me, Basque correlatives do not
necessarily trigger maximalization, since they can be
adverbially quantified by 'always', often', etc., and
therefore seem to falsify Grosu's specific claim
concerning them (see 1.2). More importantly, most of the
languages that I know of that do have correlatives allow
multiple Wh-phrases in them: Russian (Izvorski 1996),
Serbo-Croatian (Boskovic 1997), Hindi (as acknowledged by
Mohajan himself), and, outside the Indo-European domain,
Hungarian (Liptak 2000) or Basque (personal work, in
prep.). It follows from this fact that a movement analysis
cannot be the right analysis, since there is no way that I
can think of to raise a Rel.P from inside another, the two
"heads" eventually turning up adjacent to each other.
Returning to Bianchi's diachronic reconstruction, let
me add that she could have been more explicit: once the
correlative protasis is reinterpreted as the complement of
a (possible null) D�, it functions like a left-dislocated
DP (or nominal Hanging Topic) -- an independently
attested structure -- so that the correlative pronoun in
the apodosis may be reanalyzed as an ordinary resumptive
pronoun, whence the possibility for the "newly-created"
dislocated DP to also occupy an argumental position within
the main clause.
2.4. The allusion to resumptive pronouns also raises the
question of the relatives embedded in relatives mentioned
above. I must first underlie that they are by no means
rare cross-linguistically: Barrett-Keach's dissertation
(1985: 71), cited by Zwart for other purposes, also
contains examples of such complex relatives in Swahili.
See also Demirdache (1997: 202, 204) on Hebrew and
Palestinian Arabic, Kaplan & Whitman (1993) and Takahashi
(1997) on Japanese, or Oyhar�abal (1987) on Basque. Pollet
Samvellian has informed me (p.c.) that Modern Persian also
has such complex relatives. Little attention has generally
been paid to these constructions, except in the context of
a broader study of resumptive pronouns, as in Sells (1984)
and Demirdache (1997).
In any case, here again, there is no hope of deriving
a sentence like Kornfilt's ex. (8b), cf. 1.3, through
movement and, even if Sells was right in distinguishing
between resumptive pronouns proper and "intrusive
pronouns", a lot of work remains to be done in this area
- to take but one example, consider the fact that some
pronominal anaphors sometimes turn up in relativized
positions, e.g. 'kendisi' in Turkish (Kornfilt's ex.
(31)), and 'zibun' in Japanese (Kaplan & Whitman 1993, ex.
(22b,c)). Besides, "logophoricity" -- the term J.K. uses
to account for this fact -- is a notion that should
certainly deserve a detailed discussion in this context.
2.5. The fact that some languages have no visible relative
pronouns, that others have specific ones, and that yet
others have but one paradigm for relative and
interrogative pronouns raises many questions. First, Law
explicitly distinguishes between the interrogative, and
relative, *uses* of those pronouns, whereas, according to
Zwart, D-pronouns (which are always relatives) end up in
Spec,CP3, whilst W-pronouns (which may be either
interrogative or relative), raise higher to Spec,CP2: it
would be interesting to see whether such a divide can be
correlated with other properties of the relevant
languages.
Second, Zwart is led to establishing a configurational
hierarchy of three functional heads (when Kayne had only
one, and Bianchi, two), where pure relative pronouns (the
D-words) occupy the lower position, in sharp contrast with
Rizzi's (1997) results, which place relatives higher in
the tree than interogatives -- incidentally, note that
Liptak's independent work on Hungarian correlatives
directly corroborates Rizzi's findings: here again, then,
one would be very interested to see what really is at
stake parameter-wise (note 14).
Third, concerning the invisible rel. pronouns (and C
heads), two contributions have adopted opposite attitudes:
as mentioned above, Kornfilt argues that Turkish, which
has neither C� nor visible relative pronouns, RCs are CPs,
and has recourse to abstract operators, but Murasugi is
satisfied with treating Japanese RCs as simple IPs; one
would have liked to see what counter-arguments this author
has to oppose to Kaplan & Whitman's (1993) study.
2.6. Finally, consider now the question of whether
relative constructions exist as such. Ever since the
emergence of the Principles and Parameters framework
twenty years ago, Chomsky has been defending the view that
constructions are at best a descriptive, pre-theoretical
notion (this is not deny that a feature like Grosu's [REL]
probably belongs to a universal alphabet of semantically
interpretable features); what is more, the idea is now
well-established that semantics can only be
interpretative. Consequently, Mahajan's chapter, which
explicitly adopts the stance that a unitary underlying
structure of the three types of Hindi RCs means progress
with regard to Srivastav-Dayal's earlier work does not
necessarily take us in the right direction: in fact,
Srivastav-Dayal may well have been more "Chomskyan" that
A.M. is, since, on her approach, it is not before LF that
syntactic objects receive any interpretation. In other
words, Mahajan's paper, although it incorporates many
recent hypotheses (such as Chomsky's own copy theory of
movement, and the related hypothesis that deletion is a
"pure" PF process), may be considered as resting on more
old-fashioned general hypotheses concerning the structure
of grammar.
2.7. Trying to reconcile Chomsky's Minimalist Program and
Kayne's Antisymmetry hypotheses may thus prove to be
"Quixotic". On the other hand, although no contribution to
the volume has argued for a radically representational
view of syntactic structure, given (as we have seen) that
base-generation of the "head" is in many cases
unavoidable, and in others quite reasonable, and since
some grammarians might also entertain the idea that
movement is more costly than base-generation (note 15), it
is to be wondered whether chain-formation mechanisms such
as those advocated by Rizzi (1990) and more recently by
Brody (1995) might not, in fact, be much more compatible
with Kayne's views, as revised by Zwart's chapter, it
being (at first sight) sufficient to base generate the
"head" in Spec,Restrictor-Phrase. But that is, to be sure,
the matter for a least a long independent article.
NOTES
1. I will follow Grosu's convention and put quotation
marks around the word head when it is not used in the X-
bar theoretic sense, but rather refers to the NP --
semantically a CN -- to which a restrictive relative is,
under the standard analysis, adjoined within a DP. It is,
of course, in this sense that the so-called "Internally
headed relative clauses" are discussed e.g. in Murasugi's
contribution.
2. The compatibility of this approach with Kayne's 1994
analysis of Appositive RCs, summed up in section 1.7, is
not clear, but Grosu's chapter summarizes Bianchi's own
analysis of ARCs (see below): there is no contradiction in
the latter's position.
3. The author must therefore also demonstrate (and she
does) that, contrary to appearances, the pre-nominal
Turkish RCs are clausal rather than reduced (participial).
4. Another point worth discussing is the potential
extension of Law's analysis to Scrambling in German
(possibly as opposed to Scandinavian and Dutch) -- see for
instance Vikner (1994, appendix).
5. Compare the qualification "*at least* one unbound
variable" in Grosu's characterization of his feature [REL]
in 1.2. above (emphasis mine, G.R.).
6. Basque is another "exception", which is more difficult
to deal with under the HRA, since its canonical word order
is the following (with a phonetically realized "final" C�
and a phonetically realized "final" Det -- two segments
which need not particularly bother Japanese students:
(i) [[ OP[[[[...t...] Infl�] C�]] NP] D�]
(see de Rijk 1972 and much ensuing work, among which
Oyhar�abal 1989.)
7. As Kayne himself noted (1994: note 73), "The LCA
[=Linear Correspondance Axiom] itself does not determine
whether Spec,CP must be filled by movement or could
perhaps be filled by "base generation". IF movement is
systematic, THEN new work on island constraints is called
for" (emphasis mine, G.R.).
8. Another study that provides independent evidence for
the lack of traces in (some) Japanese RCs is Heycock
(1993).
9. As far as I've been able to check, the very expression
"Role R" is in fact (originally) due to Edwin Williams.
10. Recall that in Dutch (as well as in German, for
intance), there are two types of relative pronouns, some
of which are etymologically related to demonstratives (D-
words) and the others to interrogatives (W-words).
11. It is unclear, however, what to do (semantically) with
the trace of the "head" NP within Spec,CP2 or Spec,CP3,
even if the Rel. Det is interpreted as an abstraction
operator, as suggested by J.-W.Z., in agreement with the
editors' introduction (p. 43, endnote 3).
12. Needless to say, "locality" requires a technical
definition, which may vary from author to author, as
acknowledged by the editors (Intro., p. 36), Bianchi (p.
63), or Grosu (p. 106) who explicitly contrast Chomsky's
definition of the minimal domain of a head X�, and
Manzini's (1994) radical departure from it (she includes
the Specifier of the head's complement, but excludes the
head's own specifier).
13. Correlatives sentences are also recognized in
Slavic linguistics, cf. Izvorski (1996). In generative
Germanic linguistics, on the other hand, they have
generally not been identified as such, but clearly
correspond to Left-Dislocated Free Relatives or "Hanging
Topics" -- see e.g. Prince (1998: 297), where a
correlative example (25b) is sandwiched between two
examples displaying left-dislocated DPs associated with a
resumptive pronoun. In any case, it is intriguing that the
editors should have restricted the existence of
correlatives constructions to "Hindi (along with related
S. Asian languages)" (p. 26) -- all the more so as J.W.
Zwart, in the last chapter of the book, makes an explicit
allusion to correlative sentences in comparing his
examples (73) and (78) (p. 379-380) -- and as they
themselves provide an example of a left-dislocated
"multiply-headed Free Relative construction in Bulgarian"
(p. 25, (69)).
14. The fact that relatives-in-relatives are cross-
linguistically rarer than relatives embedded in indirect
questions (not to mention those embedded in assertive
completive clauses) may well have something to do with the
fact that the two sorts of pronouns occupy distinct
positions.
15. If numerations contain phonetically empty objects,
two distinct numerations would underlie the base-generation
derivation (resorting to Merge only) on the one hand, and
the transformational derivation on the other (where the
abstract OP is moved to Spec,CP) -- whence the impossibility
of comparing the relative costs in Minimalist terms. From a
more abstract stance, though, one might consider that the
fewer occurrences of Move(-alpha) are resorted to, the more
explanatory they are.
REFERENCES
Barrett-Keach, C. 1985. The Syntax and Interpretation of the
Relative Clause Construction in Swahili. New York: Garland.
Bianchi, V. 1995. Consequences of Antisymmetry for the
Syntax of Headed Relative Clauses. Doctoral dissertation,
Scuola Normale Superiore, Pisa.
Boskovic, Z. 1997. 'Superiority Effects with Multiple Wh-
Fronting in Serbo-Croatian'. Lingua 102, 1-20.
Brody, M. 1995. Lexico-Logical Form. A Radically Minimalist
Theory. Cambridge (Mass.): MIT Press.
Carlson, G. 1977. 'Amount Relatives'. Language 53, 520-542.
Chomsky, N. 1995. The Minimalist Program. Cambridge (Mass.):
MIT Press.
Demirdache, H. 1997. Dislocation, Resumption and Weakest
Crossover'. In E. Anagnostopoulou et al. (eds.), Materials
on Left Dislocation (Amsterdam: Benjamins), 193-231.
de Rijk, R.P.G. 1972. 'Relative Clauses in Basque: A guided
tour'. The Chicago Which Hunt, Papers from the Relative
Clause Festival (Chicago: CLS), 115-135.
Emonds, J. 1979. 'Appositive Relatives have no Properties'.
Linguistic Inquiry 10, 211-243.
Grosu, A. & F. Landman. 1998. 'Strange Relatives of the
Third Kind'. Natural Language Semantics 6, 125-170.
Haudry, J. 1973. 'Parataxe, hypotaxe et corr�lation dans la
phrase latine'. Bulletin de la Soci�t� de linguistique 68,
147-186.
Heycock, C. 1993. 'Syntactic Predication in Japanese'.
Journal of East Asian Linguistics 2, 167-211.
Izvorski, R. 1996. 'The Syntax and Semantics of Correlative
Proforms'. NELS 26, 133-147.
Jackendoff, R. 1977. X-bar Syntax. Cambridge (Mass.): MIT
Press.
Kaplan, T. & J. Whitman. 1995. 'The Category of Relative
Clauses in Japanese, with Reference to Korean'. Journal of
East Asian Linguistics 4, 29-58.
Kayne, R. 1994. The Antisymmetry of Syntax. Cambridge
(Mass.): MIT Press.
Liptak, A. 2000. 'Multiple Relatives as Relatives of
Questions�. In G. Alberti & I. Kenesei (eds.), Approaches to
Hungarian, VII: Papers from the P�cs Conference (Szeged:
JATE Press), 151-177.
Manzini, R. 1994. 'Locality, Minimalism, and Parasitic
Gaps'. Linguistic Inquiry 25, 481-508.
Murasugi, K.S. 1991. Noun Phrases in Japanese and English
(...). Doctoral dissertation, Univ. of Connecticut.
Oyharcabal, B. 1987. Etude descriptive de constructions
complexes en basque: propositions relatives, temporelles,
conditionnelles et concessives. Doctoral dissertation, Univ.
Paris 7.
Oyharcabal, B. 1989. 'Pro-drop and the Resumptive Pronoun
Strategy in Basque'. In L. Maracz & P. Muysken (eds.),
Configurationality; The typology of asymmetries (Dordrecht:
Foris), 63-83.
Prince, E. 1998. 'On the Limits of Syntax, with Reference to
Left-Dislocation and Topicalization'. In P. Culicover and L.
McNally (eds.), Syntax and Semantics 29: The Limits of
Syntax (San Diego: Academic Press), 282-302.
Pullum, G. 1982. 'Syncategorematicity and English
infinitival "to"'. Glossa 16, 181-215.
Rebuschi, GK. 1999. 'Types de langues et types de
constructions: le cas des corr�latives'. LINX, Special
issue on Linguistic Typology and Universals, 55-72.
Rizzi, L. 1990. Relativized Minimality. Cambridge (Mass.):
MIT Press.
Rizzi, L. 1997. 'The Fine Structure of the Left Periphery'.
In L. Haegeman (ed.), Elements of Grammar; Handbook of
generative syntax (Dordrecht: Kluwer), 281-337.
Sells, P. 1984. Syntax and Semantics of Resumptive Pronouns.
Doctoral dissertation, Univ. of Massachusetts at Amherst.
Srivastav, V. 1991. 'The Syntax and Semantics of
Correlatives'. Natural Language and Linguistic Theory 9,
637-686.
Takahashi, D. 1997. 'Move-F and Null Operator Movement'.
The Linguistic Review 14, 181-196.
Vikner, S. 1994. 'Scandinavian Object Shift and West
Germanic Scrambling'. In N. Corver & H. van Riemdijk (eds.),
Studies on Scrambling (Berlin: Mouton de Gruyter), 487-517.
BIOGRAPHICAL SKETCH
I have been professor of general linguistics at the
Sorbonne Nouvelle for 12 years. My main interests are
cross-linguistic variation (from a parametric standpoint),
and the syntax-semantics interface. Most of my
publications are devoted to Basque linguistics (both
synchronic and diachronic), but I also work sporadically
on French and Bantu languages. I also recently co-edited
a volume on The Grammar of Focus, published in the
same series as the book reviewed here.
|
