Publishing Partner: Cambridge University Press CUP Extra Wiley-Blackwell Publisher Login
amazon logo
More Info


New from Oxford University Press!

ad

The Social Origins of Language

By Daniel Dor

Presents a new theoretical framework for the origins of human language and sets key issues in language evolution in their wider context within biological and cultural evolution


New from Cambridge University Press!

ad

Preposition Placement in English: A Usage-Based Approach

By Thomas Hoffmann

This is the first study that empirically investigates preposition placement across all clause types. The study compares first-language (British English) and second-language (Kenyan English) data and will therefore appeal to readers interested in world Englishes. Over 100 authentic corpus examples are discussed in the text, which will appeal to those who want to see 'real data'


New from Brill!

ad

Free Access 4 You

Free access to several Brill linguistics journals, such as Journal of Jewish Languages, Language Dynamics and Change, and Brill’s Annual of Afroasiatic Languages and Linguistics.


Email this page
E-mail this page

Review of  Verb Clusters


Reviewer: Michael Wagner
Book Title: Verb Clusters
Book Author: Katalin É. Kiss Henk van Riemsdijk
Publisher: John Benjamins
Linguistic Field(s): Syntax
Subject Language(s): Dutch
German
Hungarian
Language Family(ies): Uralic
Germanic
Book Announcement: 16.2747

Discuss this Review
Help on Posting
Review:


Date: Sat, 17 Sep 2005 16:31:14 -0400 (EDT)
From: Michael Wagner <chael@MIT.EDU>
Subject: Verb Clusters: A study of Hungarian, German and Dutch

EDITORS: Kiss, Katalin É.; Riemsdijk, Henk van
TITLE: Verb Clusters
SUBTITLE: A study of Hungarian, German and Dutch.
SERIES: Linguistik Aktuell/Linguistics Today 69
PUBLISHER: John Benjamins
YEAR: 2004

INTRODUCTION

This volume comprises 14 articles on verb clustering and related
phenomena in Hungarian and West Germanic and an introduction to
the topic by the editors. It contributes to the substantial body of work
on verb clustering within generative grammar in the 30 years since
Evers (1975) suggested that verbal complexes are constituents
formed by adjoining embedded verbs to matrix verbs. The articles
report on the results of a project funded in collaboration by the
Netherlands and Hungary.

Verb clusters consist of sequences of predicates that form a
selectional chain: The highest predicate or auxiliary (Pred-1) takes an
infinitival or participial complement (Pred-2) which in turn may take a
predicate as its complement (Pred-3) and so on. Clusters may also
contain a particle, which are in general the complement (or within the
complement) of the 'deepest' predicate in the selectional chain.

In a language like English, predicates are ordered left-to-right
according to the selectional sequence:

(1) English
(that) we should-1 want-2 to have-3 to let-4 him solve-5 this problem

Verb-Clustering languages such as Dutch, German, and Hungarian
differ from English in that the predicates seemingly form a constituent.
While in English, arguments intersperse with the predicates (e.g. 'him'
in the example above), in Dutch and German, all arguments precede
the entire predicate complex (Kiss and van Riemsdijk, Introduction). I
coindex the arguments with the predicates they are selected by:

(2) Dutch (1-2-3-4-5 'English Order' as in ex. (1))
dat wij hem-4 dit probleem zouden-1 willen-2 moeten-3 laten-4
oplossen-5.
that we him the problem should want have.to let solve

In this example, the predicates within the cluster are ordered with
respect to each other just like in English: 1-2-3-4-5. Also, the
arguments are ordered with respect to each other like in English. But
in Dutch, both nominal arguments precede the entire predicate
complex, while in English the nominal arguments directly follow the
selecting predicate.

The ordering in Dutch results in cross-serial dependencies (cf.
Bresnan et al 1982). This type of structure that entered the spotlight
of computational linguistics in the 80s since it cannot (if unbounded)
be generated by context-free grammars (see, e.g., discussion in
Shieber 1985).

German, also a clustering language, differs from Dutch in showing the
inverse linear order among the predicates. This order (5-4-3-2-1) is
called 'Roll-up Order' by many contributions in the volume:

(3) German (5-4-3-2-1 'Roll-Up Order')
dass wir ihn dieses Problem lösen-5 lassen-4 müssen-3 wollen-2
können-1
that we him this problem solve let have.to want can

The third type of linear order apart from the 'English Order' and
the 'Roll-up Order' that plays a major role in the volume is the 'Particle
Climbing order' illustrated in (4), where the deepest constituent
precedes all other predicates, which in turn are ordered in the 'English
Order':

(4) Dutch (3-1-2 'Particle Climbing' Order)
dat zij hem op wilde bellen
that she him up wanted call

The three word order options observed in Dutch and German
clusters -- 'English Order', 'Roll-up Order', and 'Particle Climbing' --
are all attested in Hungarian. Part of the goal of the volume is to
explore parallels and differences between clusters in West Germanic
and Hungarian.

In Hungarian, only non-finite verbs can roll up, so that the roll-up in
(5b) is not complete and the finite verb is ordered first (Szendrői &
Tóth, this volume):

(5) a. Hungarian 1-2-3-4 ('English Order')
Kedden fog-1 tudni-2 jarni-3 dezeni-4.
Tuesday will can go train
'He will be able to go training on Tuesdays.'

(5) b. Hungarian 1-4-3-2 ('Partial Roll-Up')
Kedden fog-1 dezeni-4 jarni-3 tudni-2.
Tuesday will train go can

(5) c. Hungarian 3-1-2 ('Particle-Climbing'; from Kiss and van
Riemsdijk)
Janos fel-3 szeretne-1 hivni-2 Marit
Janos up would.like call Marit

The introduction by Katalin É. Kiss and Henk van Riemsdijk
summarizes the main properties of clusters (e.g. adjacency effects,
clause union effects, possible/impossible linear permutations,
morphological properties) and lays out some of the main strands of
ideas in the theoretical literature on the topic, many of which will come
up in the discussion of the individual contributions below.

I will henceforth use numbers to refer to the predicates in a cluster,
labeling the highest predicate in the selectional chain as 1, the next
one down 2 and so on.

DISCUSSION

I structure the discussion section of this review according to the
subheadings used in the volume: DATA, THEORIES, PROSODY,
ASPECT, VO/OV and MORPHOLOGY. For reasons of space I only
report on some salient points for each article, I cannot go into an
equal amount of detail for each one. This selective attention is not
intended as a comment on the value of the contributions, but simply
due to the fact that the length and depth of the volume make a
thorough review of each piece impossible in this format.

DATA

Susi Wurmbrand: West Germanic Clusters: The Empirical Domain
This article summarizes results from a study of variation in linear order
in predicate clusters across a number of dialects of German (classified
as 'German', 'Austrian', 'Vorarlberg', and 'Swiss'). In this study
participants were asked (i) for their favored order of certain predicate
clusters with 3 or 4 members, and (ii) for their acceptability rating of a
choice of orders. It is one the few existing empirical studies on
variation in clustering (see also Hsiao (1999), Schmid and Vogel
(2004), Seiler (2004)).

Several results of this study directly bear on generalizations taken as
a given in some of the earlier literature. First, the results show that the
IPP (infinitivus pro participio) cannot be causally linked to verb-
clustering and verb-reordering. This claim made was made in some
earlier studies based on the suggestive fact that Frisian shows
rigid 'roll-up' order (i.e. 3-2-1) and lacks IPP-effects.

Wurmbrand argues that the IPP neither depends on reordering of
predicates (there are IPP-effects both in the order 3-2-1 and 1-2-3, at
least one of which is the uninverted order in any theory), nor does
reordering necessarily go along with IPP effects. Reordering is also
possible in some auxiliary-modal constructions which involve infinitives
in the first place (see also Kathol 1998 for this point).

Second, the issue of which predicate orders are possible/impossible is
taken up. It seems that each permutation of three predicates is
possible at least in some dialects. Cases of 2-1-3, however, are rare,
and Wurmbrand proposes that existing cases of 2-1-3 involve
extraposition of 3. Extraposition constructions certainly do create 2-1-
3 orders, as is illustrated in the following example:

(6) German 2-1-3 (Extraposition)
dass sie versucht-2 hat-1 zu schweigen-3
that she tried has to be.silent
'that she has tried to be silent'

Wurmbrand proposes to derive possible predicate orders using a post-
syntactic reordering operation, following Haegeman and Riemsdijk
(1986). In particular, she proposes to invoke a linear order 'flip' (cf.
Williams, the same volume, and Kathol 1998 for an HPSG proposal
with similar properties) which inverts linear orders between sisters.

This operation is capable of generating the orders 3-2-1, 2-3-1, 1-2-3,
and 1-3-2. But it cannot generate the orders 2-1-3 and 3-1-2, which
hence must be derived by some other reordering process in narrow
syntax -- according to Wurmbrand extraposition of 3 in the case of 2-1-
3 and phrasal leftward movement of 3 in the case of 3-2-1.

The motivation of this distinction between PF-reordering vs. narrow-
syntax reordering remains somewhat unclear. Wurmbrand (2004)
takes up this issue and argues on independent grounds for the post-
syntactic nature of the 'Flipping' of predicates on the one hand, and
the syntactic nature of the two additional reorderings that generate 3-
1-2 and 2-1-3 on the other.

Wurmbrand's proposal captures all possible word orders in the
respective dialects (except the distribution of the 3-1-2 and 2-1-3
orders) by lexically specifying which types of predicates (auxiliaries,
modals) trigger a 'flip' with their complement, and whether or not
the 'flip' is obligatory or optional for the particular class of embedding
predicate, thus providing an elegant account of the dialectal variation.

The article only discusses the analysis of predicate clusters with 3
elements. An interesting question that is left open for future inquiry is
how the account deals with the data reported in the same article for
predicate clusters with 4 elements. While the results of the survey of
clusters with 4 elements are preliminary, one clear result seems to be
that the order 1-4-2-3 (which cannot be derived by 'flip') is the
preferred neutral word order for certain predicate sequences. This
seems somewhat unexpected, since 1-4-2-3 is predicted to involve a
syntactically motivated reordering by phrasal movement (focus
according to Wurmbrand 2004, following the results of the survey in
Schmid and Vogel (2004)), and should not be derivable as a result of
PF-movement in the neutral case.

A related question that eventually needs to be addressed can be
illustrated by Hungarian: only 3 of the 4 orderings derivable by 'flip'
are attested in predicate clusters with 4 elements in 'roll-ups' where
the highest predicate comes first: 1-2-3-4, 1-2-4-3, 1-4-3-2. The order
1-3-4-2 -- i.e. the order where the 'roll-up' does not start at the lowest
node -- is unattested. Why would 1-3-4-2 be special and be excluded
as opposed to the other three? This issue arises not only in
Hungarian, a similar pattern was reported for a German dialect in
Bech (1955).

Kriszta Szendrői and Ildikó Tóth: Hungarian verbal clusters
Hungarian verb clusters occur in two types of sentences. The first
type is called 'non-neutral', these are sentences which contain
contrastive focus or sentential negation. They typically display a
partial roll-up order (e.g. a cluster with order 1-4-3-2 or 1-2-4-3).

(7) Hungarian Partial Roll-up in Non-Neutral Sentences: 1-4-3-2 and 1-
2-4-3

The rolled-up predicates have to be adjacent and cannot be
separated with adverbials or arguments. The finite predicate cannot
roll up, and is thus always initial. If the highest predicate is nonfinite, a
complete roll-up is possible. According to the judgments in the
literature, the roll-up has to start at the bottom, but need not invert all
predicates.

The second type of sentences are called 'neutral'. These typically
include predicate clusters in the particle climbing order (e.g. a cluster
with order 4-1-2-3 where 4 is the particle). Particle Climbing orders
are e.g. the following ('Particle' is also called 'Verbal Modifier'
or 'Preverb'; (4=Particle)):

(8) Hungarian Particle Climbing in Neutral Sentences: 4-1-2-3, 1-4-2-
3, and 1-2-4-3

The predicates following the particle are in the 'English order'. They
can be interspersed with adverbials and quantifiers, but not with focus
sensitive operators (e.g. 'only') or sentential negation.

This study, again based on a questionnaire, was designed to test the
claims about possible word orders reported in the literature. It seems
to be the first empirical study of word order possibilities in Hungarian
verb clusters.

For 'non-neutral' sentences, the pattern reported in earlier literature
was by and large confirmed:

(9) Judgments reported in the literature
1-2-3-4 (English Order)
1-2-4-3 (Partial Roll-up)
1-4-3-2 (Roll-Up (all predicates except finite one)
*1-3-2-4 (Partial Roll-Up in the middle)
*1-3-4-2 (Partial Roll-Up starting at higher node)
*1-4-2-3 (non-local Roll-Up)

The three orders claimed to be grammatical in the literature were
rejected in less than 40% or the responses, while the orders claimed
to be ungrammatical were rejected in more than 80% of the cases.
The troubling result is the frequency with which participants rejected
certain word orders that are generally assumed to be grammatical, at
least for certain particular predicates. In particular, the 'English order'
does not seem to be accepted by all speakers.

For 'neutral' sentences, orders involving full (e.g. 4-1-2-3 where 4 is
the particle) and partial particle climbing (1-4-2-3) were tested. Partial
climbing was more acceptable in cluster with 3 elements that in
clusters with 4 elements. Other significant factors determining the
acceptability of word orders were the predicate class and the
transparency of the semantic relation between a predicate and its
complement particle.

THEORIES

Jonathan Bobaljik: Clustering Theories
This article summarizes some of the main theoretical problems posed
by predicate clusters. One interesting issue raised is why predicate
clusters occur in Germanic OV but not in VO languages.

For example, cross-serial dependencies with respect to the
distribution of arguments have not been reported (to my knowledge)
for Germanic VO languages. In other words, while (10a) is attested in
Dutch, (10b) has not been observed:

(10)
a. Arg1 Arg2 Arg3 Pred1 Pred2 Pred3
b. *Pred1 Pred2 Pred3 Arg1 Arg2 Arg3

If this gap turns out to be systematic, it poses a challenge to what
Bobaljik calls inheritance-based approaches, which allow the
formation of complex predicates (e.g. Forward Partial Combination in
Steedman (1985, 533), also Williams' 'Reassociate', this volume). The
idea in those approaches is that predicates combine first to form a
complex predicate and then subsequently take their arguments:

(11) Arg1 Arg2 [Pred1 + Pred2]

The question that Bobaljik raises is why, after forming the complex
predicate, it has to take the arguments to its *left*. A language in
which it takes both arguments to the *right* is not attested.

Bobaljik points out that theories that form clusters by reanalysis of a
surface string (e.g. Haegeman and Riemsdijk 1985) might be able to
account for the distribution. The reanalysis approach requires
adjacency before forming the complex predicate. This is only given in
OV order:

(12) Arg-1 Arg-2 Arg-3 Pred-3 Pred-2 Pred-1

After reanalysis of the string of predicates as one constituent, (post-
syntactic) rules of permutations derive different cluster word orders.
Linear Adjacency, according to Bobaljik, might play a crucial role then
in the derivation of predicate clusters, since it is a precondition on
restructuring.

It is not clear however whether the generalization tying verb clustering
to the OV-property is entirely correct -- at least if we also take adverbs
into account, and not just arguments. One of the criteria for cluster
formation is the linearization of adverbs relative to the predicates that
they modify. Haider (2003) notes that adverbs, just like arguments,
precede the entire predicate cluster and cannot be interspersed, while
in English they are adjacent to the modified predicate:

(13) a. The new theory certainly may possibly have indeed been badly
formulated (Quirk et al. 1986, 85)
(13) b.
dass die Theorie wohl tatsächlich schlecht formuliert worden sein mag
that the theory possibly indeed badly formulated been be may

In English, the adverbs intersperse with the predicates, whereas in
Cluster-languages, the adverbs are ordered in the same way with
respect to each other, but they precede the predicate complex:

(14) a. English: Adv-1 Pred-1 Adv-2 Pred-2 Adv-3 Pred-3
(14) b. German: Adv-1 Adv-2 Adv-3 Pred-3 Pred-2 Pred-1
(14) c. Dutch: Adv-1 Adv-2 Adv-3 Pred-1 Pred-2 Pred-3

Once again, Dutch shows cross-serial dependencies.

Nilsen (2002, p. 72) illustrates adverb-related clustering effects in
Norwegian, a VO-language. Adverbs (as opposed to arguments)
sometimes precede a sequence of predicates, yielding cross-serial
dependencies, similar to those observed in Dutch (Adv-1 modifies
Pred-1 etc.):

(14) d. Norwegian: Adv-1 Adv-2 Adv-3 Pred-1 Pred-2 Pred-3

It is unclear then how the reanalysis theory would derive clustering
effects with respect to adverbials in a VO language such as
Norwegian. In addition, Bobaljik's puzzle remains of why clustering of
predicates with respect to arguments seems only to be attested in OV
languages.

Michael Brody: Roll-Up Structures and Morphological Words
Brody's article illustrates how 'mirror theory' (Brody 2000) captures
some of the differences between the roll-up construction on the one
hand and the particle-climbing structure on the other.

The only way to derive a 'Roll-up' order (e.g. 1-4-3-2) in mirror theory
is for the 'rolled-up' predicates to form a single morphological word.
Within words, syntactic complements are specifiers, resulting in the
inverted linearization. This is due to the assumption that specifiers (as
opposed to complements) precede their sister.

The particle climbing order (e.g. 4-1-2-3) involves a long-distance
dependency. Therefore it cannot be a single morphological word in
Brody's theory and must involve a phrasal chain.

The fact that the predicate sequence in the particle climbing order can
be interrupted by adverbs, but cannot in the roll-up structure (cf. also
Koopman and Szabolci 1998) jibes well with mirror theory, since only
the latter constitutes a morphological word, as can be seen in its linear
ordering ('roll-up'). The article gives further evidence that particle-
climbing involves a phrasal chain, but roll-up does not.

Brody only discusses Hungarian data. Dutch predicate clusters show
similar adjacency effects as Hungarian roll-up clusters, and are
also 'word-like'. However, they have the 'English' word order. It is not
clear then whether mirror theory might not be too restrictive and
specifically designed to account for the Hungarian data.

Edwin Williams: The structure of clusters
Williams presents the set of structures CAT, from Williams (2002),
which is argued to comprise all permutations of functional elements
that are possible in natural language -- at least all those permutations
that do not involve phrasal movement. CAT takes the Cinquean
hierarchy of functional elements (F-seq) as a given.

The set of permutations that CAT allows can be described in the
following way. Assuming that the basic order is Pred1 > Pred2 >
Pred3 ..., the set of permutations can be derived by the
operations 'flip' and 'reassociate'.

'Flip' swaps the linear order of two sisters, but keeps the functor-
argument relation constant, thus deriving 'left-branching' structures
such as:
A > B ----------> B < A

'Reassociate' can re-bracket predicate sequences:
A > [ B > C] ---------> [ A > B] > C

Reassociate can feed flip, deriving [ B < A ] > C from A > [ B > C ]. But
flip blocks any further application of Reassociate (i.e. Predicates
combining by '<' cannot reassociate). Combinations of Flip and
Reassociate can generate long-distance dependencies.

CAT is similar to combinatory categorial grammar (e.g. Steedman
1985 for an application to predicate clusters) in that actual orderings
that a particular language chooses are not derived by movement, but
are due to lexical linear order information. The lexical entry specifies
whether a predicate takes its complement to left/right. But Williams
assumes phrasal movement in addition to the permutations provided
by CAT, thus departing from the CCG program.

CAT comprises too many structures -- language specific constraints
further restrict the set of possible permutations. Williams illustrates
that CAT allows Verb Projection Raising constructions (Haegeman
and Riemsdijk 1986), and how it accounts for predicate orders in
Hungarian.

Restrictions on the role up-construction in Hungarian predicate
clusters are explained by making a number of assumptions:
(i) roll-ups are single words (similar to Brody);
(ii) words contain root-level elements, but not other words or phrases;
(iii) all auxiliaries are ambiguous between root-level and word-level;
(iv) root level lexical entries are specified to take their complement to
the left, word level are specified to take their complement to the right;
(v) finite verbs can only be word-level;
(vi) complex words cannot head phrases. Once these assumptions
have been granted the right linear orders follow.

But this is hardly surprising, since essentially every linear precedence
relation is stipulated in a lexical entry. CAT is only relevant in that
each of the output linear orders in Hungarian (there are three: 1-2-3-
4, 1-2-4-3, 1-4-3-2) have to lie within the combinatorial power of CAT.
This is not a strong argument for CAT, however. With verb clusters of
4 elements, CAT predicts that 22 out of 24 orders are possible
(according to Williams). So if one randomly picks three word orders
out of the 24, the probability that it falls outside of CAT is just 0.24.
Also, once one assumes with Williams that the highest predicate
cannot flip, CAT actually generates *all* 6 possible permutations of the
lower three predicates.

It also seems that there is no deep reason in the analysis why words
should be linearized in one way and phrases in another. A mirror
image of Hungarian would seem to lie in the scope of CAT: If all roots
take their complements to the right, words take their complement to
the left, then only predicate clusters of the following form would be
allowed, where roll-up has to start from the top:

(15)
2-3-4-5-1
3-4-5-2-1
4-5-3-2-1

These orders can be derived by flipping the higher predicates with
their sisters and maintaining the basic order between the lower
predicates. This Hungarian*, it seems to me, could be derived simply
by minimally changing Williams account by specifying roots as taking
their complement to the right and words as taking their complement to
the left. The theory thus seems much more powerful than e.g. Brody's,
in which the difference of linearization of words vs. phrases is
universally fixed.

Williams goes on to show that CAT cannot explain why particle-
climbing is incompatible with any amount of roll-up and can only occur
if all predicates to the right of the particle are linearized in the 'English'
order (e.g. 4-1-2-3). He concludes with many other authors in the
volume that particle climbing must involve phrasal movement. Phrasal
movement, however, falls outside of the domain of CAT.

The article concludes with an incisive discussion of Koopman and
Szabolcsi's (2000) phrasal movement account of predicate clustering.
This is the only discussion of an approach to verb clustering involving
only phrasal movement.

PROSODY

Kriszta Szendrői: A stress based approach to climbing
Szendrői proposes that particle climbing is driven by the prosodic
deficiency of certain auxiliaries and other stress-avoiding predicates.
The idea is that stress-avoiding predicates resist the assignment of
main stress in the cluster, the particle climbs to the position where
main stress is assigned.

This operation of particle climbing is assumed to involve phrasal
movement. Phrasal movement is generally costly, so a derivation with
particle-climbing would in general be ruled out. However, these more
costly derivations win out if a stress-avoiding predicate can end up
destressed if the movement applies but not otherwise. In this case, the
less economical derivation wins out. The system thus operates within
a global economy approach, using transderivational comparisons (cf.
Chomsky 1995).

A phonological motivation for particle climbing is also assumed in
many other contributions in the volume (Csirmaz, Kiss, Gabor,
Ackema, Olsvay).

Anikó Csirmaz: Particles and phonologically defective predicates
This article presents an insight into the deeper parallels between
Dutch and Hungarian Predicate Clusters. Csirmaz shows evidence
that Hungarian distinguishes phrasal particles from particles that are
heads, and draws a parallel to a similar observation about Dutch in
(Koster 1994) (see also Kiss, same volume, p. 346 for discussion of
this observation). Only head-particles can occur within a verb cluster;
phrasal particles (e.g. particles with a modifier) have to climb to the
initial position, in Csirmaz' analysis the specifier of a PredP
projection. 'Phrasal' and 'Head' are defined in terms of bare phrase
structure, such that a single element that does not further project has
both X0 and XP status. In both languages, Particles and other verbal
modifiers move to a specifier position in a projection PredP. Heads, in
addition, have the alternative option of incorporating into the predicate.

ASPECT

Gabor Alberti: Climbing for Aspect
Alberti sees a different force as the trigger for particle climbing:
particles are taken to be aspectualizers, and they move to check an
aspectual feature in AspP (following Pinon 1995). The feature driven
grammar of aspectuality interacts with independent factors, e.g.
phonological constraints disfavoring certain kinds of complex
phonological words.

The contributions by Csaba Olsvay, Kriszta Szendrői and Anikó
Csirmaz all take auxiliaries (and certain other predicates) to be
phonologically defective; Olsvay suggests that they are also
aspectually defective. Part of the motivation of particle climbing is the
stress-avoiding property of auxiliaries (and certain other predicates),
just as in Szendrői and Csirmaz, but the generalization about what
can act as a particle according to Olsvay is the following: the stressed
element preceding the finite auxiliary must always be the closest
constituent having an aspectual feature.

Furthermore, predicates that avoid stress are characterized as
elements that cannot 'represent the main assertion'. According to Kiss
(this volume), the 'Main assertion' is defined as the 'leftmost element
of the predicate phrase'. This restriction is held responsible for the
fact that auxiliaries are stress-avoiding.

These correlations prompt the question why there would be such a
close connection between aspect and prosody, which unfortunately
none of the papers in the volume addresses.

VO/OV

Katalin É. Kiss: Parallel Strategies of verbal complex formation in
Hungarian and West Germanic.
This contribution points out interesting parallels between clustering in
West Germanic and Hungarian, e.g. the observation that focus
sensitive operators and negation are generally ruled out in
the 'particle climbing' order (e.g. Part-1-2-3). This is true even in those
WG languages that generally allow to break up clusters by such
elements in other orders (West Flemish and the Swiss German
spoken in Zürich).

The paper argues that underlyingly, predicate clusters should be
taken to have the 'English order'. One argument for this view is that In
Hungarian, Particle-Climbing is possible from embedded clauses
separated by a complementizer from the matrix clause. In those cases,
Kiss argues, leftward movement of the particle has to be involved. By
analogy, the same analysis can be given for particle climbing more
generally. This is taken to argue for a base with the 'English word
order', at least in Hungarian. (This argument is countered in Ackema
(same volume) by the claim that in Hungarian, a complementizer can
incorporate into a predicate.)

Kiss concludes that the predicate clusters in West Germanic and
Hungarian can thus only be derived in a parallel way if the base for
both is assumed to be VO and not OV.

Peter Ackema: Do preverbs climb?
Ackema discusses the claim that 'particle-climbing' (e.g. orders of the
form 4-1-2-3) is derived by leftward movement assuming a VO-base
and shows various problems with this kind of approach.

Leftward dislocation of predicates in other cases does not generally
seem to obey the same restrictions. É.g. V2 movement can strand a
particle, while the fronting of a participle within the cluster cannot:

(15)
a. John called his lawyer up. (Transliterated from Dutch)
b. *that John his lawyer called has up. (Transliterated from Dutch)

Ackema's point is that the two types of dislocation (Preverb/Particle
Climbing and V2-movement) should not both be analyzed in the same
way (i.e. leftward head-movement). Indeed, particle climbing violates
the Head-Movement Constraint, in that it targets the lowest instead of
the highest head.

Most analyses in the volume would agree with the position that
(15a,b) involve different kinds of dislocation. But while the other
approaches take Particle-Climbing to be phrasal movement, Ackema
defends an alternative analysis, which assumes an OV-base, and
derives apparent preverb-climbing as preverb-stranding. In other
words, it is not the particle that moves leftward but it is everything else
that moves rightward.

Ackema adds the stipulation that the preverb cannot move on its own.
This stipulation is independently motivated by the fact that particles in
Dutch are also excluded from scrambling. This restriction against the
scrambling or particles is not explained in those proposals that treat
particle climbing as phrasal movement (e.g. most of the other
proposals in the volume).

This article is the most comprehensive comparative treatment in the
volume, and very meticulous in trying to work out the parallels and
differences between Hungarian and Dutch particle-climbing
constructions.

MORPHOLOGY

Huba Bartos: Verbal complexes and morphosyntactic merger
This paper presents predicate sequences that are scopally
ambiguous, but both readings are linearized in the same way. The tool
used to deal with this apparent mismatch in morphological bracketing
and syntactic scope is morpho-syntactic merger (based
on 'morphological merger' in Halle and Marantz 1993). The
assumption is that each step of merge in syntax can be followed by
morphological operations. The particular linear orders are achieved by
stipulating predicates to be underlying specified as [+ suffix], thus
triggering a reordering in morphology.

Ildikó Tóth: Infinitival complements of modals in Hungarian and in
German
The paper addresses clause-union effects, a property closely linked to
verb clustering. Tóth observes that German modals can embed both
personal and impersonal passives; Hungarian modals cannot embed
impersonal constructions. Another difference relates to case marking.
Hungarian subjects embedded under modals receive dative case,
while in German they can receive nominative case if they are the
highest argument in the sentence.

Interestingly, raising predicates in Hungarian and German pattern like
German modals. Impersonal constructions can be embedded under
raising verbs such as 'seem', where nominative case is assigned by
the matrix predicate to the embedded subject. Tóth analyzes the
difference between the modals in the two languages by positing that
German but not Hungarian modals are raising predicates. German
modals are thus restructuring predicates and involve a mono-clausal
structure.

Marcel den Dikken: Agreement and 'Clause Union'
This last contribution also addresses clause-union effects. Den Dikken
starts out by illustrating that 'Clause Union' is not unitary phenomenon
but involves a scale of properties, and depending on the embedding
verb some but not other clause-union effects occur:

Clause Union Effects:
1. Preverb-climbing to the matrix clause
2. Definiteness agreement with embedded object (definiteness
agreement is evidenced by the choice of a different set of subject
agreement markers in the presence of a 3rd person definite object).
3. Person agreement with embedded object (e.g. choice of special
agreement markers in the case of first person subjects and second
person objects)

den Dikken links these three effects to three different syntactic factors,
which are independently argued for:

1. the location of AspP (preverbs move to the specifier of AspP)
2. the location of vP/AgrOP (if this projection is high, then there will be
definiteness agreement)
3. the presence/absence of an IP boundary between the matrix and
the embedded clause (which blocks person agreement )

The analysis is based on an intricate web of observations about
agreement in Hungarian, and manages to relate them to other
syntactic facts. The proposal leads to the novel proposal that
Hungarian has object clitics.

CONCLUSION

The volume constitutes an important contribution to the understanding
of verb clusters and related phenomena, and serves to familiarize the
reader with the state of the art with respect to the empirical evidence
and the main theoretical issues.

The main lessons to be learned is that the field is far from converging
on a solution to the problem, and that the phenomena that factor into
the explanation (possible/impossible linear order permutations, clause-
union effects) are more complex and necessitate finer grained
analyses than previous studies were assuming.

Overall, the discussion of the theoretical issues would have been
sharpened by a more direct confrontation of different proposals and
different theoretical frameworks, rather than (as is mostly the case,
one exception is Bobaljik's contribution) merely juxtaposing very
different frameworks.

A notable gap in the coverage of theoretical approaches is the lack of
a discussion of the treatment of verbal clusters in categorial grammar
(e.g. Steedman 1985) and in HPSG (e.g. Kathol 2000, Mueller 2002),
although individual articles make reference to results from both lines
of research. Another issue that is not addressed is that one underlying
assumption made in the volume -- namely that predicate clusters form
constituents -- has been challenged in Kroch and Santorini (1991).

REFERENCES

Bech, Gunnar (1955/1957). Studien zum deutschen Verbum infinitum.
Tübingen: Max Niemeyer Verlag.

Bresnan, Joan, Ronald Kaplan, Stanley Peters and Annie Zaenen
(1982). Cross-serial dependencies in Dutch, Linguistic Inquiry 13: 613-
635.

Brody, Michael (2000). Mirror theory. Linguistic Inquiry 31:29-56.

Cinque, G. (1993). A null theory of phrase and compound stress.
Linguistic Inquiry 24, 239-298.

Evers, Arnold (1975). The Transformational Cycle of Dutch and
German, Ph.D. Dissertation, University of Utrecht.

Evers, Arnold (2003). Verb clusters and cluster creepers. In Verb
Constructions in German and Dutch, ed. by Pieter Seuren and Gerard
Kempen, 43-89. Amsterdam/Philadelphia: John Benjamins Publishing
Company.

Haider, Hubert (2003). V-clustering and clause union: Causes and
effects. In Verb Constructions in German and Dutch, ed. by Pieter
Seuren and Gerard Kempen, 91-126. Amsterdam/Philadelphia: John
Benjamins Publishing Company.

Hsiao, Franny P. (2000). Relating syntactic variation in West
Germanic to checking parameters. In Proceedings of ConSOLE 8, ed.
by Christine Czinglar, Katharina Koehler, Erica Thrift, Erik Jan van der
Torre, and Malte Zimmermann, 155-169. Leiden: SOLÉ.

Kathol, Andreas (1998). Linearization of verb clusters in West
Germanic. In Proceedings of the 16th Western Conference on
Linguistics (WECOL 16, 1996), ed. by Vida Samiian, 149-161. Fresno:
California State University at Fresno.

Kathol, Andreas (2000). Linear Syntax. New York/Oxford: Oxford
University Press.

Koopman, Hilda, and Anna Szabolcsi (2000). Verbal Complexes.
Cambridge, MA: MIT Press.

Koster, Jan. 1994. Predicate incorporation and the word order of
Dutch. In Paths towards Universal Grammar: Studies in honor of
Richard S. Kayne, ed. by Guglielmo Cinque, Jan Koster, Jean-Yves
Pollock, Luigi Rizzi, and Raffaella Zanuttini, 255-277. Washington,
D.C.: Georgetown University Press.

Kroch, Anthony S. and Beatrice Santorini (1991). The derived
structure of the West Germanic verb raising construction. In Robert
Freidin, ed., Principles and parameters in comparative grammar
(Current studies in linguistics 20). Cambridge, MA: MIT Press. 269-
338.

Mueller, Stefan (2002). Complex Predicates: Verbal Complexes,
resultative constructions, and particle verbs in German. Stanford: CSLI.

Nilsen, Oystein (2002). Eliminating Positions. Syntax and semantics of
sentence modification. PhD dissertation, Utrecht University.

Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan
Svartvik (1986). A Comprehensive Grammar of the English Language.
London: Longman.

Schmid, Tanja and Ralf Vogel (2004). Dialectal variation in German 3-
verb clusters: A surface-oriented OT-account. Journal of Comparative
Germanic Linguistics 7: 235-274. http://www.ling.uni-
potsdam.de/~rvogel/schmid-vogel04.pdf

Seiler, Guido (2004). On three types of dialect variation, and their
implications for linguistic theory. Evidence from verb clusters in Swiss
German dialects. In: Dialectology meets Typology, ed. by Bernd
Kortmann Berlin: Mouton de Gruyter. Draft dated May 15, 2003
available at http://www.ds.unizh.ch/gseiler/downloads/seiler3types.pdf

Shieber, Stuart M. (1985). Evidence Against the Context-freeness of
Natural Language, Linguistics and Philosophy 8: 333-343.

Steedman, Mark (1985). Dependency and coordination in Dutch and
English. Language 61.3. 523--568.

Wurmbrand, Susi (2004). Syntactic vs. post-syntactic movement In:
Proceedings of the 2003 Annual Meeting of the Canadian Linguistic
Association (CLA), ed. by Sophie Burelle and Stanca Somesfalean,
284-295. http://wurmbrand.uconn.edu/research/files/CLA-SW.pdf

Wurmbrand, Susi (2005). Online Bibliography on Verb Clusters and
verb (projection) raising.
http://wurmbrand.uconn.edu/research/Bibliographies/verbclusterbib.ht
ml

ABOUT THE REVIEWER

Michael Wagner is currently a post-doc in the Department of Brain
and Cognitive Sciences at MIT. He has done work on segmental
phonology, the syntax and semantics of focus, and the prosody of
predication and modification. His dissertation looks at the relation
Date: Sat, 17 Sep 2005 16:31:14 -0400 (EDT)
From: Michael Wagner <chael@MIT.EDU>
Subject: Verb Clusters: A study of Hungarian, German and Dutch

EDITORS: Kiss, Katalin É.; Riemsdijk, Henk van
TITLE: Verb Clusters
SUBTITLE: A study of Hungarian, German and Dutch.
SERIES: Linguistik Aktuell/Linguistics Today 69
PUBLISHER: John Benjamins
YEAR: 2004

INTRODUCTION

This volume comprises 14 articles on verb clustering and related
phenomena in Hungarian and West Germanic and an introduction to
the topic by the editors. It contributes to the substantial body of work
on verb clustering within generative grammar in the 30 years since
Evers (1975) suggested that verbal complexes are constituents
formed by adjoining embedded verbs to matrix verbs. The articles
report on the results of a project funded in collaboration by the
Netherlands and Hungary.

Verb clusters consist of sequences of predicates that form a
selectional chain: The highest predicate or auxiliary (Pred-1) takes an
infinitival or participial complement (Pred-2) which in turn may take a
predicate as its complement (Pred-3) and so on. Clusters may also
contain a particle, which are in general the complement (or within the
complement) of the 'deepest' predicate in the selectional chain.

In a language like English, predicates are ordered left-to-right
according to the selectional sequence:

(1) English
(that) we should-1 want-2 to have-3 to let-4 him solve-5 this problem

Verb-Clustering languages such as Dutch, German, and Hungarian
differ from English in that the predicates seemingly form a constituent.
While in English, arguments intersperse with the predicates (e.g. 'him'
in the example above), in Dutch and German, all arguments precede
the entire predicate complex (Kiss and van Riemsdijk, Introduction). I
coindex the arguments with the predicates they are selected by:

(2) Dutch (1-2-3-4-5 'English Order' as in ex. (1))
dat wij hem-4 dit probleem zouden-1 willen-2 moeten-3 laten-4
oplossen-5.
that we him the problem should want have.to let solve

In this example, the predicates within the cluster are ordered with
respect to each other just like in English: 1-2-3-4-5. Also, the
arguments are ordered with respect to each other like in English. But
in Dutch, both nominal arguments precede the entire predicate
complex, while in English the nominal arguments directly follow the
selecting predicate.

The ordering in Dutch results in cross-serial dependencies (cf.
Bresnan et al 1982). This type of structure that entered the spotlight
of computational linguistics in the 80s since it cannot (if unbounded)
be generated by context-free grammars (see, e.g., discussion in
Shieber 1985).

German, also a clustering language, differs from Dutch in showing the
inverse linear order among the predicates. This order (5-4-3-2-1) is
called 'Roll-up Order' by many contributions in the volume:

(3) German (5-4-3-2-1 'Roll-Up Order')
dass wir ihn dieses Problem lösen-5 lassen-4 müssen-3 wollen-2
können-1
that we him this problem solve let have.to want can

The third type of linear order apart from the 'English Order' and
the 'Roll-up Order' that plays a major role in the volume is the 'Particle
Climbing order' illustrated in (4), where the deepest constituent
precedes all other predicates, which in turn are ordered in the 'English
Order':

(4) Dutch (3-1-2 'Particle Climbing' Order)
dat zij hem op wilde bellen
that she him up wanted call

The three word order options observed in Dutch and German
clusters -- 'English Order', 'Roll-up Order', and 'Particle Climbing' --
are all attested in Hungarian. Part of the goal of the volume is to
explore parallels and differences between clusters in West Germanic
and Hungarian.

In Hungarian, only non-finite verbs can roll up, so that the roll-up in
(5b) is not complete and the finite verb is ordered first (Szendrői &
Tóth, this volume):

(5) a. Hungarian 1-2-3-4 ('English Order')
Kedden fog-1 tudni-2 jarni-3 dezeni-4.
Tuesday will can go train
'He will be able to go training on Tuesdays.'

(5) b. Hungarian 1-4-3-2 ('Partial Roll-Up')
Kedden fog-1 dezeni-4 jarni-3 tudni-2.
Tuesday will train go can

(5) c. Hungarian 3-1-2 ('Particle-Climbing'; from Kiss and van
Riemsdijk)
Janos fel-3 szeretne-1 hivni-2 Marit
Janos up would.like call Marit

The introduction by Katalin É. Kiss and Henk van Riemsdijk
summarizes the main properties of clusters (e.g. adjacency effects,
clause union effects, possible/impossible linear permutations,
morphological properties) and lays out some of the main strands of
ideas in the theoretical literature on the topic, many of which will come
up in the discussion of the individual contributions below.

I will henceforth use numbers to refer to the predicates in a cluster,
labeling the highest predicate in the selectional chain as 1, the next
one down 2 and so on.

DISCUSSION

I structure the discussion section of this review according to the
subheadings used in the volume: DATA, THEORIES, PROSODY,
ASPECT, VO/OV and MORPHOLOGY. For reasons of space I only
report on some salient points for each article, I cannot go into an
equal amount of detail for each one. This selective attention is not
intended as a comment on the value of the contributions, but simply
due to the fact that the length and depth of the volume make a
thorough review of each piece impossible in this format.

DATA

Susi Wurmbrand: West Germanic Clusters: The Empirical Domain
This article summarizes results from a study of variation in linear order
in predicate clusters across a number of dialects of German (classified
as 'German', 'Austrian', 'Vorarlberg', and 'Swiss'). In this study
participants were asked (i) for their favored order of certain predicate
clusters with 3 or 4 members, and (ii) for their acceptability rating of a
choice of orders. It is one the few existing empirical studies on
variation in clustering (see also Hsiao (1999), Schmid and Vogel
(2004), Seiler (2004)).

Several results of this study directly bear on generalizations taken as
a given in some of the earlier literature. First, the results show that the
IPP (infinitivus pro participio) cannot be causally linked to verb-
clustering and verb-reordering. This claim made was made in some
earlier studies based on the suggestive fact that Frisian shows
rigid 'roll-up' order (i.e. 3-2-1) and lacks IPP-effects.

Wurmbrand argues that the IPP neither depends on reordering of
predicates (there are IPP-effects both in the order 3-2-1 and 1-2-3, at
least one of which is the uninverted order in any theory), nor does
reordering necessarily go along with IPP effects. Reordering is also
possible in some auxiliary-modal constructions which involve infinitives
in the first place (see also Kathol 1998 for this point).

Second, the issue of which predicate orders are possible/impossible is
taken up. It seems that each permutation of three predicates is
possible at least in some dialects. Cases of 2-1-3, however, are rare,
and Wurmbrand proposes that existing cases of 2-1-3 involve
extraposition of 3. Extraposition constructions certainly do create 2-1-
3 orders, as is illustrated in the following example:

(6) German 2-1-3 (Extraposition)
dass sie versucht-2 hat-1 zu schweigen-3
that she tried has to be.silent
'that she has tried to be silent'

Wurmbrand proposes to derive possible predicate orders using a post-
syntactic reordering operation, following Haegeman and Riemsdijk
(1986). In particular, she proposes to invoke a linear order 'flip' (cf.
Williams, the same volume, and Kathol 1998 for an HPSG proposal
with similar properties) which inverts linear orders between sisters.

This operation is capable of generating the orders 3-2-1, 2-3-1, 1-2-3,
and 1-3-2. But it cannot generate the orders 2-1-3 and 3-1-2, which
hence must be derived by some other reordering process in narrow
syntax -- according to Wurmbrand extraposition of 3 in the case of 2-1-
3 and phrasal leftward movement of 3 in the case of 3-2-1.

The motivation of this distinction between PF-reordering vs. narrow-
syntax reordering remains somewhat unclear. Wurmbrand (2004)
takes up this issue and argues on independent grounds for the post-
syntactic nature of the 'Flipping' of predicates on the one hand, and
the syntactic nature of the two additional reorderings that generate 3-
1-2 and 2-1-3 on the other.

Wurmbrand's proposal captures all possible word orders in the
respective dialects (except the distribution of the 3-1-2 and 2-1-3
orders) by lexically specifying which types of predicates (auxiliaries,
modals) trigger a 'flip' with their complement, and whether or not
the 'flip' is obligatory or optional for the particular class of embedding
predicate, thus providing an elegant account of the dialectal variation.

The article only discusses the analysis of predicate clusters with 3
elements. An interesting question that is left open for future inquiry is
how the account deals with the data reported in the same article for
predicate clusters with 4 elements. While the results of the survey of
clusters with 4 elements are preliminary, one clear result seems to be
that the order 1-4-2-3 (which cannot be derived by 'flip') is the
preferred neutral word order for certain predicate sequences. This
seems somewhat unexpected, since 1-4-2-3 is predicted to involve a
syntactically motivated reordering by phrasal movement (focus
according to Wurmbrand 2004, following the results of the survey in
Schmid and Vogel (2004)), and should not be derivable as a result of
PF-movement in the neutral case.

A related question that eventually needs to be addressed can be
illustrated by Hungarian: only 3 of the 4 orderings derivable by 'flip'
are attested in predicate clusters with 4 elements in 'roll-ups' where
the highest predicate comes first: 1-2-3-4, 1-2-4-3, 1-4-3-2. The order
1-3-4-2 -- i.e. the order where the 'roll-up' does not start at the lowest
node -- is unattested. Why would 1-3-4-2 be special and be excluded
as opposed to the other three? This issue arises not only in
Hungarian, a similar pattern was reported for a German dialect in
Bech (1955).

Kriszta Szendrői and Ildikó Tóth: Hungarian verbal clusters
Hungarian verb clusters occur in two types of sentences. The first
type is called 'non-neutral', these are sentences which contain
contrastive focus or sentential negation. They typically display a
partial roll-up order (e.g. a cluster with order 1-4-3-2 or 1-2-4-3).

(7) Hungarian Partial Roll-up in Non-Neutral Sentences: 1-4-3-2 and 1-
2-4-3

The rolled-up predicates have to be adjacent and cannot be
separated with adverbials or arguments. The finite predicate cannot
roll up, and is thus always initial. If the highest predicate is nonfinite, a
complete roll-up is possible. According to the judgments in the
literature, the roll-up has to start at the bottom, but need not invert all
predicates.

The second type of sentences are called 'neutral'. These typically
include predicate clusters in the particle climbing order (e.g. a cluster
with order 4-1-2-3 where 4 is the particle). Particle Climbing orders
are e.g. the following ('Particle' is also called 'Verbal Modifier'
or 'Preverb'; (4=Particle)):

(8) Hungarian Particle Climbing in Neutral Sentences: 4-1-2-3, 1-4-2-
3, and 1-2-4-3

The predicates following the particle are in the 'English order'. They
can be interspersed with adverbials and quantifiers, but not with focus
sensitive operators (e.g. 'only') or sentential negation.

This study, again based on a questionnaire, was designed to test the
claims about possible word orders reported in the literature. It seems
to be the first empirical study of word order possibilities in Hungarian
verb clusters.

For 'non-neutral' sentences, the pattern reported in earlier literature
was by and large confirmed:

(9) Judgments reported in the literature
1-2-3-4 (English Order)
1-2-4-3 (Partial Roll-up)
1-4-3-2 (Roll-Up (all predicates except finite one)
*1-3-2-4 (Partial Roll-Up in the middle)
*1-3-4-2 (Partial Roll-Up starting at higher node)
*1-4-2-3 (non-local Roll-Up)

The three orders claimed to be grammatical in the literature were
rejected in less than 40% or the responses, while the orders claimed
to be ungrammatical were rejected in more than 80% of the cases.
The troubling result is the frequency with which participants rejected
certain word orders that are generally assumed to be grammatical, at
least for certain particular predicates. In particular, the 'English order'
does not seem to be accepted by all speakers.

For 'neutral' sentences, orders involving full (e.g. 4-1-2-3 where 4 is
the particle) and partial particle climbing (1-4-2-3) were tested. Partial
climbing was more acceptable in cluster with 3 elements that in
clusters with 4 elements. Other significant factors determining the
acceptability of word orders were the predicate class and the
transparency of the semantic relation between a predicate and its
complement particle.

THEORIES

Jonathan Bobaljik: Clustering Theories
This article summarizes some of the main theoretical problems posed
by predicate clusters. One interesting issue raised is why predicate
clusters occur in Germanic OV but not in VO languages.

For example, cross-serial dependencies with respect to the
distribution of arguments have not been reported (to my knowledge)
for Germanic VO languages. In other words, while (10a) is attested in
Dutch, (10b) has not been observed:

(10)
a. Arg1 Arg2 Arg3 Pred1 Pred2 Pred3
b. *Pred1 Pred2 Pred3 Arg1 Arg2 Arg3

If this gap turns out to be systematic, it poses a challenge to what
Bobaljik calls inheritance-based approaches, which allow the
formation of complex predicates (e.g. Forward Partial Combination in
Steedman (1985, 533), also Williams' 'Reassociate', this volume). The
idea in those approaches is that predicates combine first to form a
complex predicate and then subsequently take their arguments:

(11) Arg1 Arg2 [Pred1 + Pred2]

The question that Bobaljik raises is why, after forming the complex
predicate, it has to take the arguments to its *left*. A language in
which it takes both arguments to the *right* is not attested.

Bobaljik points out that theories that form clusters by reanalysis of a
surface string (e.g. Haegeman and Riemsdijk 1985) might be able to
account for the distribution. The reanalysis approach requires
adjacency before forming the complex predicate. This is only given in
OV order:

(12) Arg-1 Arg-2 Arg-3 Pred-3 Pred-2 Pred-1

After reanalysis of the string of predicates as one constituent, (post-
syntactic) rules of permutations derive different cluster word orders.
Linear Adjacency, according to Bobaljik, might play a crucial role then
in the derivation of predicate clusters, since it is a precondition on
restructuring.

It is not clear however whether the generalization tying verb clustering
to the OV-property is entirely correct -- at least if we also take adverbs
into account, and not just arguments. One of the criteria for cluster
formation is the linearization of adverbs relative to the predicates that
they modify. Haider (2003) notes that adverbs, just like arguments,
precede the entire predicate cluster and cannot be interspersed, while
in English they are adjacent to the modified predicate:

(13) a. The new theory certainly may possibly have indeed been badly
formulated (Quirk et al. 1986, 85)
(13) b.
dass die Theorie wohl tatsächlich schlecht formuliert worden sein mag
that the theory possibly indeed badly formulated been be may

In English, the adverbs intersperse with the predicates, whereas in
Cluster-languages, the adverbs are ordered in the same way with
respect to each other, but they precede the predicate complex:

(14) a. English: Adv-1 Pred-1 Adv-2 Pred-2 Adv-3 Pred-3
(14) b. German: Adv-1 Adv-2 Adv-3 Pred-3 Pred-2 Pred-1
(14) c. Dutch: Adv-1 Adv-2 Adv-3 Pred-1 Pred-2 Pred-3

Once again, Dutch shows cross-serial dependencies.

Nilsen (2002, p. 72) illustrates adverb-related clustering effects in
Norwegian, a VO-language. Adverbs (as opposed to arguments)
sometimes precede a sequence of predicates, yielding cross-serial
dependencies, similar to those observed in Dutch (Adv-1 modifies
Pred-1 etc.):

(14) d. Norwegian: Adv-1 Adv-2 Adv-3 Pred-1 Pred-2 Pred-3

It is unclear then how the reanalysis theory would derive clustering
effects with respect to adverbials in a VO language such as
Norwegian. In addition, Bobaljik's puzzle remains of why clustering of
predicates with respect to arguments seems only to be attested in OV
languages.

Michael Brody: Roll-Up Structures and Morphological Words
Brody's article illustrates how 'mirror theory' (Brody 2000) captures
some of the differences between the roll-up construction on the one
hand and the particle-climbing structure on the other.

The only way to derive a 'Roll-up' order (e.g. 1-4-3-2) in mirror theory
is for the 'rolled-up' predicates to form a single morphological word.
Within words, syntactic complements are specifiers, resulting in the
inverted linearization. This is due to the assumption that specifiers (as
opposed to complements) precede their sister.

The particle climbing order (e.g. 4-1-2-3) involves a long-distance
dependency. Therefore it cannot be a single morphological word in
Brody's theory and must involve a phrasal chain.

The fact that the predicate sequence in the particle climbing order can
be interrupted by adverbs, but cannot in the roll-up structure (cf. also
Koopman and Szabolci 1998) jibes well with mirror theory, since only
the latter constitutes a morphological word, as can be seen in its linear
ordering ('roll-up'). The article gives further evidence that particle-
climbing involves a phrasal chain, but roll-up does not.

Brody only discusses Hungarian data. Dutch predicate clusters show
similar adjacency effects as Hungarian roll-up clusters, and are
also 'word-like'. However, they have the 'English' word order. It is not
clear then whether mirror theory might not be too restrictive and
specifically designed to account for the Hungarian data.

Edwin Williams: The structure of clusters
Williams presents the set of structures CAT, from Williams (2002),
which is argued to comprise all permutations of functional elements
that are possible in natural language -- at least all those permutations
that do not involve phrasal movement. CAT takes the Cinquean
hierarchy of functional elements (F-seq) as a given.

The set of permutations that CAT allows can be described in the
following way. Assuming that the basic order is Pred1 > Pred2 >
Pred3 ..., the set of permutations can be derived by the
operations 'flip' and 'reassociate'.

'Flip' swaps the linear order of two sisters, but keeps the functor-
argument relation constant, thus deriving 'left-branching' structures
such as:
A > B ----------> B < A

'Reassociate' can re-bracket predicate sequences:
A > [ B > C] ---------> [ A > B] > C

Reassociate can feed flip, deriving [ B < A ] > C from A > [ B > C ]. But
flip blocks any further application of Reassociate (i.e. Predicates
combining by '<' cannot reassociate). Combinations of Flip and
Reassociate can generate long-distance dependencies.

CAT is similar to combinatory categorial grammar (e.g. Steedman
1985 for an application to predicate clusters) in that actual orderings
that a particular language chooses are not derived by movement, but
are due to lexical linear order information. The lexical entry specifies
whether a predicate takes its complement to left/right. But Williams
assumes phrasal movement in addition to the permutations provided
by CAT, thus departing from the CCG program.

CAT comprises too many structures -- language specific constraints
further restrict the set of possible permutations. Williams illustrates
that CAT allows Verb Projection Raising constructions (Haegeman
and Riemsdijk 1986), and how it accounts for predicate orders in
Hungarian.

Restrictions on the role up-construction in Hungarian predicate
clusters are explained by making a number of assumptions:
(i) roll-ups are single words (similar to Brody);
(ii) words contain root-level elements, but not other words or phrases;
(iii) all auxiliaries are ambiguous between root-level and word-level;
(iv) root level lexical entries are specified to take their complement to
the left, word level are specified to take their complement to the right;
(v) finite verbs can only be word-level;
(vi) complex words cannot head phrases. Once these assumptions
have been granted the right linear orders follow.

But this is hardly surprising, since essentially every linear precedence
relation is stipulated in a lexical entry. CAT is only relevant in that
each of the output linear orders in Hungarian (there are three: 1-2-3-
4, 1-2-4-3, 1-4-3-2) have to lie within the combinatorial power of CAT.
This is not a strong argument for CAT, however. With verb clusters of
4 elements, CAT predicts that 22 out of 24 orders are possible
(according to Williams). So if one randomly picks three word orders
out of the 24, the probability that it falls outside of CAT is just 0.24.
Also, once one assumes with Williams that the highest predicate
cannot flip, CAT actually generates *all* 6 possible permutations of the
lower three predicates.

It also seems that there is no deep reason in the analysis why words
should be linearized in one way and phrases in another. A mirror
image of Hungarian would seem to lie in the scope of CAT: If all roots
take their complements to the right, words take their complement to
the left, then only predicate clusters of the following form would be
allowed, where roll-up has to start from the top:

(15)
2-3-4-5-1
3-4-5-2-1
4-5-3-2-1

These orders can be derived by flipping the higher predicates with
their sisters and maintaining the basic order between the lower
predicates. This Hungarian*, it seems to me, could be derived simply
by minimally changing Williams account by specifying roots as taking
their complement to the right and words as taking their complement to
the left. The theory thus seems much more powerful than e.g. Brody's,
in which the difference of linearization of words vs. phrases is
universally fixed.

Williams goes on to show that CAT cannot explain why particle-
climbing is incompatible with any amount of roll-up and can only occur
if all predicates to the right of the particle are linearized in the 'English'
order (e.g. 4-1-2-3). He concludes with many other authors in the
volume that particle climbing must involve phrasal movement. Phrasal
movement, however, falls outside of the domain of CAT.

The article concludes with an incisive discussion of Koopman and
Szabolcsi's (2000) phrasal movement account of predicate clustering.
This is the only discussion of an approach to verb clustering involving
only phrasal movement.

PROSODY

Kriszta Szendrői: A stress based approach to climbing
Szendrői proposes that particle climbing is driven by the prosodic
deficiency of certain auxiliaries and other stress-avoiding predicates.
The idea is that stress-avoiding predicates resist the assignment of
main stress in the cluster, the particle climbs to the position where
main stress is assigned.

This operation of particle climbing is assumed to involve phrasal
movement. Phrasal movement is generally costly, so a derivation with
particle-climbing would in general be ruled out. However, these more
costly derivations win out if a stress-avoiding predicate can end up
destressed if the movement applies but not otherwise. In this case, the
less economical derivation wins out. The system thus operates within
a global economy approach, using transderivational comparisons (cf.
Chomsky 1995).

A phonological motivation for particle climbing is also assumed in
many other contributions in the volume (Csirmaz, Kiss, Gabor,
Ackema, Olsvay).

Anikó Csirmaz: Particles and phonologically defective predicates
This article presents an insight into the deeper parallels between
Dutch and Hungarian Predicate Clusters. Csirmaz shows evidence
that Hungarian distinguishes phrasal particles from particles that are
heads, and draws a parallel to a similar observation about Dutch in
(Koster 1994) (see also Kiss, same volume, p. 346 for discussion of
this observation). Only head-particles can occur within a verb cluster;
phrasal particles (e.g. particles with a modifier) have to climb to the
initial position, in Csirmaz' analysis the specifier of a PredP
projection. 'Phrasal' and 'Head' are defined in terms of bare phrase
structure, such that a single element that does not further project has
both X0 and XP status. In both languages, Particles and other verbal
modifiers move to a specifier position in a projection PredP. Heads, in
addition, have the alternative option of incorporating into the predicate.

ASPECT

Gabor Alberti: Climbing for Aspect
Alberti sees a different force as the trigger for particle climbing:
particles are taken to be aspectualizers, and they move to check an
aspectual feature in AspP (following Pinon 1995). The feature driven
grammar of aspectuality interacts with independent factors, e.g.
phonological constraints disfavoring certain kinds of complex
phonological words.

The contributions by Csaba Olsvay, Kriszta Szendrői and Anikó
Csirmaz all take auxiliaries (and certain other predicates) to be
phonologically defective; Olsvay suggests that they are also
aspectually defective. Part of the motivation of particle climbing is the
stress-avoiding property of auxiliaries (and certain other predicates),
just as in Szendrői and Csirmaz, but the generalization about what
can act as a particle according to Olsvay is the following: the stressed
element preceding the finite auxiliary must always be the closest
constituent having an aspectual feature.

Furthermore, predicates that avoid stress are characterized as
elements that cannot 'represent the main assertion'. According to Kiss
(this volume), the 'Main assertion' is defined as the 'leftmost element
of the predicate phrase'. This restriction is held responsible for the
fact that auxiliaries are stress-avoiding.

These correlations prompt the question why there would be such a
close connection between aspect and prosody, which unfortunately
none of the papers in the volume addresses.

VO/OV

Katalin É. Kiss: Parallel Strategies of verbal complex formation in
Hungarian and West Germanic.
This contribution points out interesting parallels between clustering in
West Germanic and Hungarian, e.g. the observation that focus
sensitive operators and negation are generally ruled out in
the 'particle climbing' order (e.g. Part-1-2-3). This is true even in those
WG languages that generally allow to break up clusters by such
elements in other orders (West Flemish and the Swiss German
spoken in Zürich).

The paper argues that underlyingly, predicate clusters should be
taken to have the 'English order'. One argument for this view is that In
Hungarian, Particle-Climbing is possible from embedded clauses
separated by a complementizer from the matrix clause. In those cases,
Kiss argues, leftward movement of the particle has to be involved. By
analogy, the same analysis can be given for particle climbing more
generally. This is taken to argue for a base with the 'English word
order', at least in Hungarian. (This argument is countered in Ackema
(same volume) by the claim that in Hungarian, a complementizer can
incorporate into a predicate.)

Kiss concludes that the predicate clusters in West Germanic and
Hungarian can thus only be derived in a parallel way if the base for
both is assumed to be VO and not OV.

Peter Ackema: Do preverbs climb?
Ackema discusses the claim that 'particle-climbing' (e.g. orders of the
form 4-1-2-3) is derived by leftward movement assuming a VO-base
and shows various problems with this kind of approach.

Leftward dislocation of predicates in other cases does not generally
seem to obey the same restrictions. É.g. V2 movement can strand a
particle, while the fronting of a participle within the cluster cannot:

(15)
a. John called his lawyer up. (Transliterated from Dutch)
b. *that John his lawyer called has up. (Transliterated from Dutch)

Ackema's point is that the two types of dislocation (Preverb/Particle
Climbing and V2-movement) should not both be analyzed in the same
way (i.e. leftward head-movement). Indeed, particle climbing violates
the Head-Movement Constraint, in that it targets the lowest instead of
the highest head.

Most analyses in the volume would agree with the position that
(15a,b) involve different kinds of dislocation. But while the other
approaches take Particle-Climbing to be phrasal movement, Ackema
defends an alternative analysis, which assumes an OV-base, and
derives apparent preverb-climbing as preverb-stranding. In other
words, it is not the particle that moves leftward but it is everything else
that moves rightward.

Ackema adds the stipulation that the preverb cannot move on its own.
This stipulation is independently motivated by the fact that particles in
Dutch are also excluded from scrambling. This restriction against the
scrambling or particles is not explained in those proposals that treat
particle climbing as phrasal movement (e.g. most of the other
proposals in the volume).

This article is the most comprehensive comparative treatment in the
volume, and very meticulous in trying to work out the parallels and
differences between Hungarian and Dutch particle-climbing
constructions.

MORPHOLOGY

Huba Bartos: Verbal complexes and morphosyntactic merger
This paper presents predicate sequences that are scopally
ambiguous, but both readings are linearized in the same way. The tool
used to deal with this apparent mismatch in morphological bracketing
and syntactic scope is morpho-syntactic merger (based
on 'morphological merger' in Halle and Marantz 1993). The
assumption is that each step of merge in syntax can be followed by
morphological operations. The particular linear orders are achieved by
stipulating predicates to be underlying specified as [+ suffix], thus
triggering a reordering in morphology.

Ildikó Tóth: Infinitival complements of modals in Hungarian and in
German
The paper addresses clause-union effects, a property closely linked to
verb clustering. Tóth observes that German modals can embed both
personal and impersonal passives; Hungarian modals cannot embed
impersonal constructions. Another difference relates to case marking.
Hungarian subjects embedded under modals receive dative case,
while in German they can receive nominative case if they are the
highest argument in the sentence.

Interestingly, raising predicates in Hungarian and German pattern like
German modals. Impersonal constructions can be embedded under
raising verbs such as 'seem', where nominative case is assigned by
the matrix predicate to the embedded subject. Tóth analyzes the
difference between the modals in the two languages by positing that
German but not Hungarian modals are raising predicates. German
modals are thus restructuring predicates and involve a mono-clausal
structure.

Marcel den Dikken: Agreement and 'Clause Union'
This last contribution also addresses clause-union effects. Den Dikken
starts out by illustrating that 'Clause Union' is not unitary phenomenon
but involves a scale of properties, and depending on the embedding
verb some but not other clause-union effects occur:

Clause Union Effects:
1. Preverb-climbing to the matrix clause
2. Definiteness agreement with embedded object (definiteness
agreement is evidenced by the choice of a different set of subject
agreement markers in the presence of a 3rd person definite object).
3. Person agreement with embedded object (e.g. choice of special
agreement markers in the case of first person subjects and second
person objects)

den Dikken links these three effects to three different syntactic factors,
which are independently argued for:

1. the location of AspP (preverbs move to the specifier of AspP)
2. the location of vP/AgrOP (if this projection is high, then there will be
definiteness agreement)
3. the presence/absence of an IP boundary between the matrix and
the embedded clause (which blocks person agreement )

The analysis is based on an intricate web of observations about
agreement in Hungarian, and manages to relate them to other
syntactic facts. The proposal leads to the novel proposal that
Hungarian has object clitics.

CONCLUSION

The volume constitutes an important contribution to the understanding
of verb clusters and related phenomena, and serves to familiarize the
reader with the state of the art with respect to the empirical evidence
and the main theoretical issues.

The main lessons to be learned is that the field is far from converging
on a solution to the problem, and that the phenomena that factor into
the explanation (possible/impossible linear order permutations, clause-
union effects) are more complex and necessitate finer grained
analyses than previous studies were assuming.

Overall, the discussion of the theoretical issues would have been
sharpened by a more direct confrontation of different proposals and
different theoretical frameworks, rather than (as is mostly the case,
one exception is Bobaljik's contribution) merely juxtaposing very
different frameworks.

A notable gap in the coverage of theoretical approaches is the lack of
a discussion of the treatment of verbal clusters in categorial grammar
(e.g. Steedman 1985) and in HPSG (e.g. Kathol 2000, Mueller 2002),
although individual articles make reference to results from both lines
of research. Another issue that is not addressed is that one underlying
assumption made in the volume -- namely that predicate clusters form
constituents -- has been challenged in Kroch and Santorini (1991).

REFERENCES

Bech, Gunnar (1955/1957). Studien zum deutschen Verbum infinitum.
Tübingen: Max Niemeyer Verlag.

Bresnan, Joan, Ronald Kaplan, Stanley Peters and Annie Zaenen
(1982). Cross-serial dependencies in Dutch, Linguistic Inquiry 13: 613-
635.

Brody, Michael (2000). Mirror theory. Linguistic Inquiry 31:29-56.

Cinque, G. (1993). A null theory of phrase and compound stress.
Linguistic Inquiry 24, 239-298.

Evers, Arnold (1975). The Transformational Cycle of Dutch and
German, Ph.D. Dissertation, University of Utrecht.

Evers, Arnold (2003). Verb clusters and cluster creepers. In Verb
Constructions in German and Dutch, ed. by Pieter Seuren and Gerard
Kempen, 43-89. Amsterdam/Philadelphia: John Benjamins Publishing
Company.

Haider, Hubert (2003). V-clustering and clause union: Causes and
effects. In Verb Constructions in German and Dutch, ed. by Pieter
Seuren and Gerard Kempen, 91-126. Amsterdam/Philadelphia: John
Benjamins Publishing Company.

Hsiao, Franny P. (2000). Relating syntactic variation in West
Germanic to checking parameters. In Proceedings of ConSOLE 8, ed.
by Christine Czinglar, Katharina Koehler, Erica Thrift, Erik Jan van der
Torre, and Malte Zimmermann, 155-169. Leiden: SOLÉ.

Kathol, Andreas (1998). Linearization of verb clusters in West
Germanic. In Proceedings of the 16th Western Conference on
Linguistics (WECOL 16, 1996), ed. by Vida Samiian, 149-161. Fresno:
California State University at Fresno.

Kathol, Andreas (2000). Linear Syntax. New York/Oxford: Oxford
University Press.

Koopman, Hilda, and Anna Szabolcsi (2000). Verbal Complexes.
Cambridge, MA: MIT Press.

Koster, Jan. 1994. Predicate incorporation and the word order of
Dutch. In Paths towards Universal Grammar: Studies in honor of
Richard S. Kayne, ed. by Guglielmo Cinque, Jan Koster, Jean-Yves
Pollock, Luigi Rizzi, and Raffaella Zanuttini, 255-277. Washington,
D.C.: Georgetown University Press.

Kroch, Anthony S. and Beatrice Santorini (1991). The derived
structure of the West Germanic verb raising construction. In Robert
Freidin, ed., Principles and parameters in comparative grammar
(Current studies in linguistics 20). Cambridge, MA: MIT Press. 269-
338.

Mueller, Stefan (2002). Complex Predicates: Verbal Complexes,
resultative constructions, and particle verbs in German. Stanford: CSLI.

Nilsen, Oystein (2002). Eliminating Positions. Syntax and semantics of
sentence modification. PhD dissertation, Utrecht University.

Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan
Svartvik (1986). A Comprehensive Grammar of the English Language.
London: Longman.

Schmid, Tanja and Ralf Vogel (2004). Dialectal variation in German 3-
verb clusters: A surface-oriented OT-account. Journal of Comparative
Germanic Linguistics 7: 235-274. http://www.ling.uni-
potsdam.de/~rvogel/schmid-vogel04.pdf

Seiler, Guido (2004). On three types of dialect variation, and their
implications for linguistic theory. Evidence from verb clusters in Swiss
German dialects. In: Dialectology meets Typology, ed. by Bernd
Kortmann Berlin: Mouton de Gruyter. Draft dated May 15, 2003
available at http://www.ds.unizh.ch/gseiler/downloads/seiler3types.pdf

Shieber, Stuart M. (1985). Evidence Against the Context-freeness of
Natural Language, Linguistics and Philosophy 8: 333-343.

Steedman, Mark (1985). Dependency and coordination in Dutch and
English. Language 61.3. 523--568.

Wurmbrand, Susi (2004). Syntactic vs. post-syntactic movement In:
Proceedings of the 2003 Annual Meeting of the Canadian Linguistic
Association (CLA), ed. by Sophie Burelle and Stanca Somesfalean,
284-295. http://wurmbrand.uconn.edu/research/files/CLA-SW.pdf

Wurmbrand, Susi (2005). Online Bibliography on Verb Clusters and
verb (projection) raising.
http://wurmbrand.uconn.edu/research/Bibliographies/verbclusterbib.ht
ml




 
ABOUT THE REVIEWER:
ABOUT THE REVIEWER


Michael Wagner is currently a post-doc in the Department of Brain
and Cognitive Sciences at MIT. He has done work on segmental
phonology, the syntax and semantics of focus, and the prosody of
predication and modification. His dissertation looks at the relation
between syntactic recursion and prosodic structure.