Icelandic and German

Journal of Comparative Germanic Linguistics 8: 153, 2005.
2005 Kluwer Academic Publishers. Printed in the Netherlands.
How to turn German into Icelandic and derive the OVVO

contrasts*
HUBERT HAIDER
Department of Linguistics, University of Salzburg, Muehlbacherhofweg 6, A 5020 Salzburg
(E-mail: [email protected])
Key words: asymmetry, object shift, OV and VO, quirky subject, scrambling, verbal clusters
Abstract. Icelandic and German dier in the head-complement order (VO vs. OV), but their
morpho-syntactic systems of verbal and nominal inection are similar enough for factoring out the
specic grammatical eects of the OV/VO-property. The analysis of a broad range of constructions
(quirky subjects, expletive subjects, object shift, scrambling, particle constructions, V-clustering)
provides the empirical basis for the following claim: the headedness dierence (OV vs. VO) is the
basic and crucial factor for the systematic dierences between the respective grammars. As for the
theoretic modeling of the OV/VO property, the key concept is argued to be a universal constraint
on the direction of merger interacting with the directionality of licensing by the head.
1. Introduction
A comparison of Icelandic and German is a most advantageous testing

ground for competing grammar theoretic models of the VO and OV
parameterization. Icelandic is consistently head-initial (VO) whereas
German is head-nal in the V projections (OV). The two languages
dier with respect to headedness (OV vs. VO), but their morphosyn-
tactic systems of verbal and nominal inection are similar to an extent
that guarantees a broad enough common background for identifying
the specic eects of the dierent headedness options and their under-
lying grammatical causalities.
The following phenomena will be discussed as examples of pairs of
contrasts that are assumed to result from the reverse headedness as
the single source factor in interaction with common core grammar
principles:
unique and obligatory functional subject position vs. VP-internal
nominative licensing
* The author wishes to extend thanks for their constructive comments to the discussants of the
2002 Reykjavik conference and especially to the three anonymous reviewers. The responsibility for
remaining shortcomings rests on the authors side, of course.
2 HUBERT HAIDER
quirky subjects vs. object-initial base orders

object shift vs. scrambling
particle constructions with multiple vs. single stranding positions
stacked VPs vs. V-clustering and clause union
The answer to the title question will be simple: change the headedness
directionality of the verbal heads in German, and what you get
(and indeed should expect to get) for free is a lot of Icelandic syntax. In
other words, the dierent headedness value and its corollaries in the
system of grammar are shown to cover a wide range of syntactic con-
trasts between Icelandic as a Germanic VO and German as an OV
language.
The paper is organized as follows: Section 2.1 provides background
information on controversial issues of the clausal architecture of Ger-
man, in comparison with Icelandic. In particular, it will be argued that
the architecture of a simple clause in German (OV) and in Icelandic
(VO), for principled reasons, diers to a greater extent than is recog-
nized in current writing. Section 2.2 prepares the grammar theoretic
background for the discussion of head-initial and head-nal projec-
tions. It is argued that VO requires a more articulate VP structure than
OV, namely VP shells, for principled reasons. VP shells are a direct
consequence of implementing the headedness value in a universally
constrained merging system (there is a ban against merger on the right-
hand side). OV does not require shells. This is the basic dierence that
is reected in the construction dierences to be discussed below. Section
2 concludes with a discussion of the existence of apparently hybrid VO
OV languages (as has been claimed for older stages of Germanic lan-
guages, and notably for Icelandic). Yiddish is a good and living
example of this hybrid nature. Its alleged hybrid status stems from the
same source as the OVVO contrast, namely partial VP shell projec-
tions in the course of merger. A language like Yiddish is neither strictly
VO nor strictly OV but rather an example of a third possibility that has
not been fully appreciated hitherto, neither for the analysis of Yiddish
nor for its impact on accounts of the diachronic split of the German
language family into an OV and a VO group. Section 3, the main data
section, discusses the above-mentioned areas of Icelandic-German
contrasts, their analyses, and their theoretical implications. Section 4
summarizes the results and discusses their impact for competing ac-
counts of OV vs. VO.
HOW TO TURN GERMAN INTO ICELANDIC 3
2. Basic issues of the clausal architecture in an OVVO perspective
Current theoretical approaches in the Generative Grammar family tend

to reduce surface structure dierences to derivational dierences. In the
Minimalist Program (Chomsky 1995), surface order dierences are the
result of xing the spell-out point on earlier or later stages of a basically
identical derivation. From this point of view, the dierence between a
VO and VO organization of a language becomes a dierence in terms of
the derivational history of the respective clause structure. In principle,
either OV or VO could be a derivational continuation of the VO or OV
structure, respectively. Currently, VO seems to be the preferred candi-
date for the derivational continuation into OV. The LCA-based ap-
proach (Linear Correspondence Axiom, see Kayne 1994 p. 33 and
Zwart 1997 for Dutch) takes head-nal structures to be derivationally
transformed head-initial structures.
In the LCA approach, the essential dierence between the Icelandic
structure for (1a) and the German structure for (2a) would be the result
of additional movements in German: the German order (1b) would be
derived from a structure like that for (1a) by moving VP-internal
material to the left of the VP. More eorts are needed for deriving the
order of clause-nal auxiliary verbs in German, which is the mirror
image order of the VO order ((1c) vs. (2b) or (1d)): VP-internal material
must be fronted, and then the evacuated VP must change places with the
selecting auxiliary. Proponents of this analysis remain silent on the
trigger issue.1 Why should all these elements move and not stay? In
other words, what triggers the VO-to-VO mutation, and why should
Universal Grammar (UG) admit OV languages at all?
hef ekki ekkt essa konu

1 a. Eg ( by VP-evacuation
I have not known this woman German with Icelandic
words; (2a)
hef essa konui ekki ekkt ei
b. Eg ( by VP-evacuation
VP-fronting2
I have this woman not known German with Icelandic
words; (2b)
c. a eg hef ekki ekkt essa konu
d. a eg essa knou ekki ekkt hef
4 HUBERT HAIDER
2 a. Ich habe diese Frau nicht gekannt

I have this woman not known
b. da ich diese Frau nicht gekannt habe
that I this woman not known have
In this view, the OV properties of German are derived as superimposed

on a VO-type structure. So the basic, and merely masked, VO properties
continue to exist However, the data tell a dierent story (see Haider
2000 for a more detailed discussion of the drawbacks of the LCA ac-
count for OV). The comparison of German and Icelandic will indicate
that the derivational extension approach is not satisfactory. German
does not behave like a derivational extension of Icelandic structures in
any signicant respect.
Contrary to widely held beliefs, German does not provide an oblig-
atory, structural subject position, that is, a functional spec position for
checking subject features, and German does not show (overt) V-to-I
movement (for the dissidents view see Haider 1993 and earlier work
cited there; Broekhuis 1992). This is not a peculiar feature of German
but a general consequence of the OV character in combination with the
morphosyntactic inventory of German. I will briey explain this in
Section 2.1, with two motives. First, one ought to give adequate reasons
if one does not follow the mainstream analyses, and, second, the anal-
ysis proposed as alternative (and required for a satisfactory account of
the data) is highly minimalist in terms of the clausal architecture for
German (see Haider 1997a, b).
I shall try to demonstrate that head-nal projections are more ele-
mentary and that the relation between Icelandic and German on the
level of phrasal and clausal architecture is not simply that of a deriva-
tional extension but rather of the implementation of dierent headed-
ness values in a universal system of phrasal architecture, with dierent
outcomes.
Let me illustrate the issue with a simple question. Why does Ice-
landic, but not German, have quirky subjects?3 Icelandic and German
have the same case system (Nom, Ace, Dat, Gen as cases of verbal
arguments), and both languages allow VP-internal nominatives. The
latter property indicates that in both languages nominative checking
does not depend on a specic structural position in surface structure
but that it depends on an overt agreement relation. A quirky subject is
a non-nominative DP in the functional spec position that otherwise
accommodates the nominative subject. By virtue of being the inhabi-
tant of the functional subject position, the non-nominative phrase ac-

quires subject properties.
A particularly clear instance of a subject property in Icelandic (see
Sigursson 1989, pp. 204f) is, for instance, the alternation with a PRO
subject. Quirky subjects, just like ordinary nominative subjects, are in
complementary distribution with PRO in nite and innitival clauses,
respectively. In other words, the PRO subject of the innitival clause
may correspond not only to a nominative subject of the nite clause but
also to a dative or accusative DP if the construction is a quirky subject
construction.
3 a. a mer lkar essi hakarl Icelandic

that medat likes this sharkNom
vonast til a PRO lka
b. Eg essi hakarl
I hope for to PRODat pleaseinf: this shark
I hope that the shark will please me.
In German, the PRO subject of an innitival clause always corresponds

to a potential nominative argument and not to a non-nominative one.4
This indicates that in German, a dative or an accusative cannot function
as a structural subject. This is unexpected, however, if German has
basically a system like Icelandic with some additional movements to the
left, given that German would then oer all the necessary prerequisites
for this very construction. Take, for instance, the passive of a double
object verb, as in (4a). Nominative checking, as in Icelandic, does not
require fronting, so the VP-internal nominative may be pied-piped by
VP topicalization, as in (4b).
4 a. da Kindern oft Marchen erzahlt werden

that childrenDat often fairy-talesNom told are
b. M
archen erz
ahlt] werden Kindern oft
fairy-talesNom told are childrenDat often
The corresponding passive construction in Icelandic would be one in

which the dative DP can be fronted to a spec position and thereby
turned into a quirky subject. Why is this not the case in German, that is,
why does the dative not acquire subject properties?
The answer to this question cannot be a construction-specic one,
that is, it cannot depend on a peculiar dierence in passive formation or
6 HUBERT HAIDER
on the properties of innitival clauses. It must be given on a general

level of clause structure dierences between Icelandic in German, or else
it would miss a crucial generalization: the Icelandic clause structure
provides an obligatory structural subject position; the German clause
structure arguably does not.5 If the German clause structure is taken to
be merely a derivational continuation of the Icelandic one, ample and
systematic counterevidence casts profound doubts on the potential
success of this enterprise. German diers systematically from Icelandic
in two basic and crucial respects. First, the nite verb in its clause-nal
position stays in situ in its VP-internal head position and is not raised to
a functional head position (no overt V-to-I). Second, there is no evi-
dence for an obligatory functional subject position in the German clause
(there are no EPP eects; see footnote for discussion of Extended
Projection Principle).
This is a clear contrast to Icelandic. In Icelandic the nite verb moves
out of the VP to a functional head position both in main clauses and in
C-introduced embedded clauses. In addition, Icelandic provides a
functional spec position for subjects.
This contrast is not so much language specic as type specic. For
VO, the existence of a functional projection for licensing subjects is
predictable, just as it is predictable for OV that there is no need for this
kind of functional projection: in VO, the so-called spec-of-VP position
precedes V and thus is outside the domain of directional licensing since
in VO, V licenses elements that follow. In OV, however, any position in
the VP (attached to the projection line) is a position in the domain of
directional licensing of V since all positions precede V , and V licenses
preceding elements.
2.1. Evidence against overt V-to-I in German
This subsection provides a single argument for the claim in the para-
graph above: there is no overt V-to-I for nite verbs and no functional
subject position in German. The impact of this result is the following:
neither the verb nor the subject moves to a functional projection, this is
surprisingly unless there is no functional projection available or neces-
sary that could trigger these movements.6 This insight is crucial for
understanding the German-Icelandic contrasts. Why does German
scramble but not Icelandic? The answer will be this: scrambling does not
leave the extended VP. Scrambling data conrm that in German, neither
subjects nor objects move to clause-internal spec positions (see Section

3.3).
With respect to V-to-I raising, German seems to behave like Faroese
(see Vikner 1995, p. 148): the verb does not have to raise.7 The verb
stays in its VP-internal head position, and the niteness features get
checked in situ (or in a derivation after spell-out, in Chomskyan terms).
There is direct positive evidence for this claim, and it comes from verbs
with multiple and separable prexes.
The argument is simple. Whenever a verb has moved to an interme-
diate functional head position, it cannot be prevented from moving to a
higher functional position if movement to this position is required. A
familiar instance of this is V-to-I followed by V-to-C in English: if a
verb is moved by V-to-I, it will move again, if I-to-C applies. There
are no cases in which verbs of an exceptional verb class only move to I
but do not move further. In other words, if a verb is moveable in the
rst step, it moves in further steps too. On the other hand, verbs may
resist movement (to intermediate functional heads) and stay in situ, as
for instance, English main verbs or nite verbs in mainland Scandina-
vian languages. So we note this: there are no exceptional verbs that
allow V-to-I but resist further movement to C.8
With this in mind, let us analyze what the theory predicts for German
verbs with more than one separable prex9 (see Haider 1993, p. 62 and
Vikner 2002). It is easy to nd or to construct them because particle-
verb formation is a productive word formation paradigm in German.
Here is the prediction: verbs with two separable prexes are unable to
move from the position in which the particle is stranded. This is so
because there is no way to strip o both particles without violating a
stranding condition. Let me illustrate this with a verb like mit-an-
kundigen (literally: with-advertise, advertise jointly).
If kundigen moves and an- is correctly stranded, mit- would be still
attached to a morpheme, namely an-, and not to a verbal trace, hence it
would not count as stranded. So it is ill formed since mit- is a separable
particle and thereby requires stranding.10 If, however, mit is stranded,
then an-kundigen must have moved. But then an is not stranded and is
therefore ill formed. So there is no way to meet the demands of both
particles simultaneously. Therefore the theory unambiguously predicts
that verbs with more than one separable particle are bound to stay since
there is no well-formed way of meeting the stranding requirements. The
only option is to avoid stranding from the beginning.
Let us now compare the two competing analyses for a clause-nal
nite verb in German. Hypothesis I the V-to-I hypothesis assumes
8 HUBERT HAIDER
that a clause-nal nite verb has moved to a functional head position.

Hypothesis II the checking in situ hypothesis assumes that clause-
nal nite verbs in German are checked in the VP-internal lexical head
position and remain in situ.
What are the respective predictions? Hypothesis I predicts that verbs
with two particles cannot appear in nite form at all because they would
incur a stranding conict for the particles when the bare verb moves.
Only forms that do not involve stranding (due to movement of the bare
verb to a functional head position) would be admitted.
Hypothesis II predicts a dierent outcome. Since Hypothesis II pre-
supposes that clause-nal verbs stay in situ, the stranding conict does
not arise. It only arises if the verb moves to the clause-initial functional
head position in a verb second or verb rst clause. So the prediction is
this: verbs with two separable particles can be used as nite verbs but
only in clauses that do not involve fronting of the nite verb, that is,
only in verb-nal clauses. What do the data tell us? Hypothesis II is
correct; Hypothesis I lacks empirical support (see (5))
5 a: wenn du uns voranmeldest

if you us preregister lit:pre-on-register
b. Du meldesti uns voran-ei
you register us pre-on
c. *Du anmeldesti uns vor-ei
d. *Du voranmeldesti uns ei
As (5bd) shows, a verb with two separable prexes may occur as nite
verb, but it is well formed only in the clause-nal position (5a), and
there is no way to derive a well-formed version with fronting.
Hypothesis I fails because it predicts that the stranding conict already
arises in the nal position when the verb is allegedly raised to the
hypothetical postverbal functional head position.
In (6), more of these verbs are listed, for the sake of illustration. The
crucial point is that this verb format is productive, and the fronting
failure is easy to understand. So there is no room for the kind of doubts
raised against the original argument (Haider 1993, p. 62), based on
verbs that arise through back formation11 (see (7)), that there might be
some ill-understood property of back formation verbs that blocks
fronting (see Koopman 1995 for Dutch).
6 a. ab-drucken - vor-ab-drucken lit. pre-print

b. an-melden - vor-an-melden lit. pre-an-nounce; preregister
c. ein-teilen - um-ein-teilen lit. re-in-deal;reorganize
d. ein-steigen - mit-ein-steigen lit. with-in-step; get ontogether
e. aus-drucken - mit-aus-drucken lit. with-out-print; print out jointly
7 a. auf-f
uhren lit. up-lead; perform, put on stage
b. Auff
uhrung performance- ur-auff
uhrung ur-performance
premiere
c. ur-auf-f
uhren show for the first time
A verb like urauuhren is a back formation from the prexed deverbal

noun Urauuhrung ((7b)) by stripping the nominalizing sux and
reanalyzing the remnant as a verb. The result is a verb prexed with a
word formation prex for nouns, namely ur-. This, again, produces a
clash with the stranding requirements for the sandwiched particle auf-.
In this case, the grammatical causality is not so immediately evident as
in the case of doubly prexed verbs above.
Note that these data do not only decide the controversy on potential
V-to-I raising for nite verbs but also for innitival constructions. In
German, the innitival markercorresponding to English tois not a
functional head but an inectional morpheme prexed to the verb. As
illustrated in (8), verbs with two separable prexes are perfect also in
innitival constructions.
8 a. ohne sich (vor)anzumelden
without oneself to preannounce
b. anstatt es (mit)auszudrucken
instead-of it to print-out jointly
What the examples in (8) conrm is that sentential innitival comple-

ments in German do not require V-to-I raising of the innitival verb,
and they conrm that the innitival marker zu is a morpheme attached
to the verb and not a separate functional head. If it were an innitival
functional head like English to, the verbs in (8) would have to raise,
and this would cause a stranding conict for the particles.
The conclusions of the above discussion for the German
clause structure are as follows: (i) there is no overt V movement to a
10 HUBERT HAIDER
clause-nal nite functional head position nor to a clause-nal innitival

functional head; (ii) there is no overt V movement to a clause-medial
functional head either;12 (iii) the nite verb remains in situ, hence par-
ticle stranding is not at issue. Movement applies only when the nite
verb moves to the V-second position, that is, the top functional head
position (i.e., the V2 position).
This state of aairs has an obvious implication for the identication
of functional head positions in German too. Since one cannot decide by
simple inspection of data whether an empty functional head exists at all
or whether it follows or precedes the VP, it is not justied to insist on a
postverbal position of an empty functional head rather than a preverbal
one. At any rate, circumstantial evidence plays a crucial role in nding
out whether there is a functional head at all13 .
One important area of circumstantial evidence for the functional
architecture of the clause are the results on extraction domains gained in
the past two decades of research in this area, in spite of the fact that they
are presently not a focal research area. One robust generalization was
that a phrase in the functional spec position of the subject (formerly
Spec-I) or in any higher (and thereby preverbal in VO languages) spec
position is an absolute extraction island. As a test criterion for the status
of a position in clause structure, this descriptive generalization is easy to
apply. If a given position is a functional spec position of the relevant
kind, a phrase in this position must be opaque. If the phrase is not
opaque, the position cannot be a functional spec position of the subject
type or a higher one. German subjects (see(9)) and phrases preceding
them (see (10)) are not opaque.
If the subject clause in (9) is VP internal, as assumed here, extraction
is expected to be possible. If, however, the subject clause would be, as
assumed by the competing analysis, in a spec position, extraction is
predicted to be blocked.14
9 a. Mit wemi hatte (*es) denn ei speisen zu d
u rfendichmehrgefreut?
with whom had it PRT dine to be-allowed youAcc more pleased
b. *With whomi would [to have dinner ei please you more?
c. With whomi would it please you more [to have dinner ei ?
The contrast between (9a) and the ungrammatical English construction
(9b) is sharp and damaging for analyses that situate the innitival
subject clause in German in a functional spec position. A clause in a
functional spec position corresponding to the English subject position,
or in a higher one, is opaque for extraction. The straightforward
alternative is a subject-in situ analysis. The clause remains in its VP-

internal position, and extraction is unproblematic. The contrast between
(9c) and (9b) is one between a clausal subject in a functional spec po-
sition and in a VP-internal one, respectively.
The in situ analysis for (9a) raises a question as to why there is no
noticeable EPP violation in (9a). The obvious answer that there is an
empty pronominal corresponding to English it in (9c) in the functional
subject position is untenable, however. An overt expletive pronominal
subject or object never co-occurs with a non-extraposed clause
(see(9a),(10a)), but it is optionally present with extraposed clauses.
Given that the pronominal quality of the overt and the covert instance
of the pronominal is identical, the covert pronominal expletive is ruled
out in the very same context that rules out the overt item, namely in the
context without extraposition. The less obvious answer seems to be the
correct one: there is no EPP eect because there is no functional spec
position that needs to be lexicalized. For principled reasons, EPP comes
into play only in VO-type clause structures, (see Section 3, the subsec-
tion on quirky subjects).
The argument against a clause-medial functional subject position in
German is strengthened by the following data. Object clauses remain
transparent for extraction when scrambled (10a) across the subject but
not if they are moved to a functional spec position like Spec-C (10b).
(10b) is an embedded V2 clause, with the innitival object clause in
Spec-C. If the subject in (10a) would be in a spec position, the scrambled
innitival clause would be either in a higher spec position or adjoined to
a functional projection. In any case, opacity for extraction is guaranteed
and therefore expected. The data, however, do not support this expec-
tation.
10 a. Wasi hat (*es) [ihr ei zu verraten] keiner riskiert?

what has her to reveal nobody risked
b. *Wasi hat sie behauptet CP ihr ei zu erklarenj
C0 habe [keiner ej riskiert
what has she claimed her to reveal had nobody risked
So we have to dismiss the premise that the overt subject position in

German is a functional spec position. The simplest account for the lack
of opacity of scrambled phrases is obvious: scrambling is a VP-internal
phenomenon. Scrambling across the subject in German is scrambling
across the VP-internal subject (see Haider and Rosengren 1998, 2003).
12 HUBERT HAIDER
At this point of the discussion it is worthwhile commenting on the

claim that a clause-initial subject in a German verb-second clause is not
in Spec-C but rather in the position that used to be called the Spec-I
position (see Travis 1991, Zwart 1993). If this analysis is correct, the
existence of an overt functional spec position for the subject could not
be denied. Travis (1991, p. 359) takes as crucial evidence an apparent
distributional restriction for the weak pronoun es in German: it may
appear in the clause-initial position as subject, but as an object it is said
to be deviant as illustrated by (11a). There are data, however, that call
for a more dierentiated account, namely (11bd).
11 a. Es hat den/*der Hund erschreckt

itNom= Acc has theAcc= Nom dog frightened
b. Ihr Geld ist nicht verloren: Es hat jetzt nur jemand anderer:15
your money is not lost it has now only someone else
Your money did not get lost. It is only in the hands of
someone else:
c. Ihr schmutziges Geld ist noch da. Es hat bis jetzt niemand
your dirty money is still there it has until now nobody
beansprucht:
claimed
d. Dieses Schild k
onnen Sie genauso gut weglassen. Es hat
this sign could you just-as-well remove it has anyway nobody
ohnehin keiner beachtet:
observed
The Travis-Zwart account for (11a) in brief is this: if a clause with a

clause-initial subject may be analyzed as an IP, a clause-initial subject
has not left Spec-I, but a clause-initial object has of course moved to
Spec-C. Thus, the dierence can be attributed to the failure of es to
undergo A movement. If the initial es is subject, it has been fronted by
A movement, and not by A movement.
But this is not the only way, nor the most straightforward one, to
capture the dierence. The primary property of es as a weak pronoun is
its stress avoidance. A weak pronoun by denition is unstressed.
Stressing of es is deviant in any position, and if fronting an object
induces stress, es cannot be fronted. Why is the subject not stressed
when fronted? The appropriate descriptive generalization is one in terms

of the information structure (topic focus organization) of the clause: an
element moved to Spec-C is not stressed if it is the highest element in the
complement of C. Moving a lower element changes the information
structure unless the fronted element re-instantiates the topic of the
preceding utterance. This is illustrated by (11bd). In these sequences,
the fronting of a weak object pronoun is possible because it remains
unstressed.
Note that the sentences with object es in initial position in (11bd)
would be rated deviant if presented in isolation. What accounts for the
contrast or the parallel between clause-initial subject es and object es,
respectively, is stressing and de-stressing rather than a structural dif-
ference. In (11b and c), de-stressing is a consequence of the information
structure.
The evidence discussed so far is evidence for two major properties of
the German clause structure (for a more detailed presentation see Ha-
ider 1997a). First, there is no evidence for functional head positions to
the right of the VP, and second, there is no evidence for movement of
the nite verb or the subject to intermediate functional head or spec
positions, respectively.16 From a comparative perspective, the lack of
verb movement has a parallel in Faroese, but the lack of EPP eects is
unparalleled, given that German, unlike Icelandic, does not provide a
null expletive (see Section 3.2). This set of circumstances becomes
understandable, however, if the stronger assumption holds. The subject
resists moving to a spec position not because it must not move but
because it cannot. There is no functional subject position to host it, and
there is no functional head to trigger movement of the nite verb to this
intermediate position. This result may become less appalling once the
reason for the need of a functional subject position in VO clauses is
appreciated (see Section 3, the subsection on quirky subjects).
2.2. The OVVO parameterization directionality and asymmetry
VO and OV are but abbreviations for head-initial and head-nal

structures, respectively, and the concomitant implications for phrase
structure in general. The foremost and fundamental dierence is this: a
complex17 head-initial projection, in contrast to a head-nal one, con-
sists of projection shells (Larson 1988). In Haider (1992/2000), this
property of head-initial projections has been shown to be a corollary of
three axioms: (i) projections are endocentric, (ii) heads license their
14 HUBERT HAIDER
complements directionally, and (iii) projections are universally right

branching.18 The rst two axioms are widely accepted; the third one (the
basic branching constraint (BBC))19 is argued for at length in the
aforementioned paper and in subsequent publications (Haider 1993,
1997c, 2000).
VP shell structures for head-initial phrases are an immediate corollary
of the three axioms. As illustrated in (12a), the verbal head in its foot
position cannot license more than one complement directionally.
Directional merger of V (i.e., merger with a complement on the right
hand side) with a second complement would produce a left-branching
structure in violation of (iii). As a consequence of this restriction, the
only admissible option is a re-instantiation of a head-complement
merger with the same verb. The lower verb thus becomes a copy of the
head of the higher V projection. The result is a VP shell structure. Note
that the two options in (12) are a direct result of consistently applying
the three axioms for the two directionality values. Licensing or merger
to the left is consistent with (iii). Therefore a shell structure is not
required. In sum, the shell structure is a purely structural requirement
for head-initial projections. There is no need for postulating a light verb
for the higher shells (Chomsky 1995).20 The arrows in (12) indicate the
licensing directionality of the (extended) head.
(12) a. head-initial VP b. head-nal VP
Evidently, in (12b), any VP-internal phrase that is immediately domi-

nated by the projection of the head c-commands the surface head po-
sition of the VP, and the phrases are all on the directionally licensed side
of the head or one of its projections. In VO projections, on the other
hand, pre-head positions are not directionally licensed unless there is a
higher VP shell with its head as the required directional licenser.
Licensing in the shell structure is strictly local. The licensed XP is in a
strictly local relation to the projecting head and to the licensing head.21
As a consequence, the licensed XP is adjacent both to the licensing head
and the projecting head position:
13 a. . . . V XPV e . . .
b. . . . V a YPXP ZPV e . . .
In (13b), YP blocks a licensing relation, and so does ZP, for dierent
reasons though. The licensing relation is the directional identication of
an argument by its head. The relation is minimal, mutual c-command
under directionality. Mutuality requires sisterhood or a chain. In (13a),
XP is minimally c-commanded by V, and XP minimally c-commands V,
by c-commanding a member of the V chain.
The YP in (13b) prevents XP from being minimally c-commanded by
the preceding V. The ZP prevents XP from minimally c-commanding
the lower V, and thereby YP or ZP destroy mutual, minimal, directional
c-command between V and XP.
Note that in OV projections, each projecting node of the verbal
projection minimally directionally c-commands its sister argument, be-
cause the argument precedes the projecting node. Hence, in OV, inter-
veners do not matter.
These considerations are sucient for deriving two essential properties
of head-initial projections that contrast with head-nal projections.
First, head initial-projections are compact, that is, they do not allow
intervening material in the domain of argument structure projection (see
13b). This follows from the mismatch of licensing directionality and
merger directionality. In the resulting shell structure, minimal mutual
c-command entails compactness.
Second, head-initial projections possess a single, local argument po-
sition that is not in the licensing domain of the head. This is the so-called
spec position of a head-initial lexical projection. Spec-VP is just the
highest projection position in the VP. It is the highest position that meets
the locality requirement, and it is to the left of the head, so it is outside the
licensing domain of the head and in need of external licensing.22 This is an
essential dierence between VO and OV projections. VO projections need
an external licenser for the highest projection position (see(14a)). In OV
projections, every projection position of the head is within the licensing
domain of the head or its projecting nodes (see(14(b)).
14 a. VO:VP DPV0 V ! . . .
b. OV:VP DPV0 . . . V
An illustration of (14) is given in (15). The verb projects its argument
structure by stepwise merger (see Chomsky 1995) with the appropriate
argument phrase. The result of each step is a partial projection with the
16 HUBERT HAIDER
unsaturated remnant of the argument structure. The projection is

completed when the argument structure is saturated. (15b) is an infor-
mal representation of the set-based structure that results from the merge
operation in Chomskys (1995) system. (15c) illustrates the discharge
and licensing relations for the same projection. The arrows indicate the
licensing direction. (15d) is the German counterpart of (15c).
The principal dierence between a head-initial and a head-nal
V projection is the fact that in the head-nal projection all discharged
(or merged) argument positions are within the directional licensing
domain of the head or its projections. But in the head-initial structure,
the highest argument position is not in the licensing domain since it is a
left sister, and the verbs licensing direction is to the right.
(15)
Let me repeat the results of the discussion above that will be needed
for the systematic comparison of Icelandic and German in the fol-
lowing section: VO projections are compact, and the highest argument
position in the projection (subject argument) needs an external iden-
tier for this argument since is not within the verbs directional
licensing domain.
2.3. Hybrid systems VO and OV, or neither?
The existence of languages that exhibit both VO and OV surface orders

is a crucial and potentially fatal problem for the authors approach, a
reviewer commented. Examples for this type of language are historic
variants of Germanic languages, such as Old Icelandic (e.g., Rognv-
aldsson 1996; Hroarsdottir 2000) and Old English (e.g. Pintzuk 1991).
But let me add that there is also a present day specimen, namely Yiddish.
The reviewers point is well taken, given that the current discussions
in grammar theory tend to focus on rigid OV or VO languages and
overlook a variety of languages that are neither strictly OV nor strictly
VO in their surface word order patterns. Yiddish is a well-known case,
and Slavic languages should be listed in this group as well.
Let me illustrate this property with the examples in (16), taken from
Diesing (1997, p. 402). One of the four variants is a perfect English
order (namely (16a)), and two are perfect German orders (namely (16c
and d)). In German, (16c and d) would be scrambling variants. The
order (16b) is ungrammatical in both English and German. It is not
immediately obvious what the base order for the variants in (16) is.
16 a. Maks hot nit gegebn Rifken das bukh. (Yiddish)
Max has not given Rebecca the book:
Max did not give Rebecca the book.
b. Maks hot Rifken nit gegebn dos bukh.
Max has Rebecca not given the book:
c. Maks hot Rifken dos bukh nit gegebn.
Max has Rebecca the book not given:
d. Maks hot dos bukh Rifken nit gegebn:
Max has the book Rebecca not given:
For Diesing, (16a) reects the base order of the VP, with (16bd) as
scrambling variants. But, if this is true, Yiddish is a puzzling exception.
It would be the only Germanic VO language that allows scrambling.
Moreover, ling Yiddish as a Germanic VO language would make it
exceptional in many more respects (see below).
Is Yiddish an (exceptional) VO or an (exceptional) OV language? It is
neither. Yiddish combines properties of both OV and VO languages. It
is basically OV but with one additional property, namely the possibility
of V-f fronting within a VP shell structure. This property has
independently been ascertained for Hindi (Mahajan 1997). What
18 HUBERT HAIDER
supercially looks like scrambling to the right in an OV language as in

(17a) turns out to be the result of (VP-internal) V movement to the left,
that is, optional VP shell formation (as in 17b).
17 a. XPej ei VYPj ZPi
b. XPVi YPei ZP
In grammar theoretic terms, this state of aairs represents a genuine
third option besides strict OV and VO, namely a language with
adjustable directionality values. It combines the properties of an OV
projection with those of a VO projection. In other words, the Yiddish
verb may license to the left and project a head-nal structure like
German, but alternatively it may license to the right and project a shell,
and moreover both options may be combined. What this amounts to is
this: merger may start as in VO and then continue as in OV, producing
VP shells rst and then optionally switching to OV and continuing to
produce a layered V projection familiar from OV. Let me illustrate this
as follows. (18a) and (18b) are the result of strict licensing to the left or
to the right, respectively. (18c), however, results if the directionality
value is adjustable in the course of merging. It combines the possibilities
of OV and VO, thus yielding a third quality.
18 a. XPYPZP V
b. XPVi YPVi ZP
c. XPYPZP V ; XPYPV ZP; XPVi YPVi ZPg adjustable
The OV and the VO option are the result of merger with the value either
as left or as right, just as in (18a) and (18b). The third possibility is the
result of changing the directionality in the course of merging. The
middle variant in (18c) starts out with directionality=right, licensing a
complement to the right, and then the value switches to left, proceeding
in an OV manner.
Why should there be a language with this property, that is, how could
UG allow for a language with licensing alternations? In brief, what we
have to admit is this: the value of the licensing parameter can be un-
derspecied. So, in the course of building a projection, either value may
be instantiated, and moreover the actualized value may be switched.
This is the right place, it seems, to point out a crucial implication for
diachronic syntax: if the historic variants of Germanic languages are
regarded from this perspective, the development of the Germanic lan-
guages is much easier to understand. The basic change was one from an
adjustable underspecied headedness value to a rigid one. The rigid
directionality allows either of two values. The choice of the value is in

principle free. One branch of Germanic languages xed it in the VO way
(West Germanic), the other in the OV way (North Germanic). So two
very similar languages in terms of their morphosyntactic make-up,
namely Icelandic and German, ended up in dierent systems by acci-
dent. The accident is the choice of the directionality value when giving
up the adjustability option.
The rest of this section is devoted to skeptical readers desire for solid
empirical evidence for the claim that the Yiddish VP clearly shows OV
properties (besides the obvious VO-like orders), and that it is contra
Diesing (1997) not simply a VO language with scrambling options. I
shall rely mainly on Chapter 2 of Vikner (2001), which is a painstakingly
argued demonstration that Yiddish has typical OV properties that are
never found in Germanic VO languages. I shall just briey review the
main data areas and refer to Vikner (2001) for an extensive discussion and
for the data sources. First, he notes that scrambling (see (16b and c)) is a
common property of OV- Germanic languages and not of VO-Germanic
languages. Then he reviews a variety of independent and contrasting
constructions, in all of which Yiddish patterns with OV and not with VO:
Null objects under coordination

Following Sadock (1998), Vikner points to coordination construc-
tions with an empty second object in Yiddish that are most easily
accounted for if the licensing relations in Yiddish are the same as in
German, namely OV. Note that Yiddish patterns like OV (e.g.,
German) and not like VO languages (e.g., English or Danish).
19 a. Di yidene hot aroysgenumen eyn gandz . . . Yiddish

the woman has out-taken one goose . . .
. . . un zi avekgeleygt afn tish
. . . and it down-put on-the table
The woman took out one goose and put it down on the table;
cited in Sadock 1998; pp: 222225:
b. Die Frau hat eine Gans herausgenommen . . . German
the woman has one goose out-taken . . .
. . . und (sie) auf den Tisch gestellt
. . . and it on the table put
c. The woman has taken out a goose and put *(it) on the table
20 HUBERT HAIDER
Verb particles (separable and non-separable ones)

Vikner (2001, p. 37) characterizes the relevant facts as follows: In
Yiddish and the (other) Germanic OV languages, particle verbs
whose particles are postverbal under V2 (separate) nevertheless al-
ways have preverbal particles in non-V2 contexts, whereas in the
Germanic VO-languages, particle verbs whose particles have to be
stranded under V2 never have preverbal particles in non-V2 con-
texts.
Only if Yiddish is an OV language like German and Dutch, and not
a VO language like English or Danish, does it follow straightfor-
wardly that Yiddish is like German and unlike Scandinavian or
English in allowing even such particles to occur preverbally in non-V2
constructions that do not incorporate [see the contrast between (20b
and c) and between (20d and f)].
20 a. Wann kommt der Zug an? German

when comes the train on
When does the train arrive?
b. Ven kumt der tsug on? Yiddish
c. *Hvorn
ar kommer toget an? Danish
d. *Wann ankommt der Zug? German
e. *Ven onkumt der tsug? Yiddish
f. Hvorn
ar ankommer toget? Danish
Obligatory lack of agreement on predicative adjectives
Yiddish, like the OV languages German, Frisian, and Dutch, has in-
ected attributive adjectives but uninected predicative adjectives
whereas those VO languages which have inected attributive adjectives
(that is, all the Scandinavian languages and all the Romance ones) also
have inected predicative adjectives. None of the present day OV
languages have predicative agreement. Vikner argues that the direc-
tionality in the VP (OV vs. VO) corresponds to the directionality in the
AP and that Yiddish forms a group with the (other) Germanic OV
languages.23
21 a. een groene bus / twee groene bussen Dutch

a green:M=F bus two green:PL buses
b. een groen huis / twee groene huizen
a greenN:Sg house two green:Pl houses
c. Een bus is groen
one bus is green
d. Twee bussen zijn groen
two buses are green
The Dutch examples in (21a and b) are representative for agreement
with an attribute and no agreement with a predicative adjective (21c and
d) in OV. Yiddish patterns like Dutch and not like Swedish. The
Swedish examples in (22ef) correspond to (22ad), respectively.
22 a. a griner oytobus / tsvey grine oytobusn Yiddish

a greenM:Sg:Nom bus two greenPl buses
b. a grin hoyz / tsvey grine hayzer
a greenN:Sg house two greenPl houses
c. Eyn oytobus iz grin
one bus is green
d. Tsvey oytobusn zaynen grin
two buses are green
e. en gr
on bus / tv
a gr
onabussar Swedish
ont hus / tv
f. ett gr a gr
ona hus
g. en bus
ar gr
on
h. tva bussar
ar gr
ona
Variation in (non-nite) verb sequences (verb clustering, verb
projection raising)
In terms of word order patterns for sequences of non-nite verbs, the

VO languages show no variation whatsoever, but the OV languages
vary very much, with Frisian being the only one of nine Germanic OV
languages that shows no variation at all. Yiddish would be exceptional
within the VO group but ts very well into the picture of the OV ones. It
22 HUBERT HAIDER
shares the verb word order of German plus the variation that derives the
Dutch order from the German basic order (see Haider 2003). Note
however that there is no Germanic VO language that shows anything
similar. Hence, Yiddish again is a well-behaved OV Germanic language
in this respect. (23) and (24) illustrate the verb-auxiliary order in passive,
and (25) and (26) present the contrasts for causative constructions. For
an exhaustive overview over all verb construction see Vikner (2002).
23 a. *Di shrub iz gevorn opgebrent Yiddish
the house is been up-burned
the house has been burnt down:
b. Das Haus ist abgebrannt worden German
the house is up-burned been
c. Di shrub iz opgebrent gevorn Yiddish
the house is up-burned been
24 a. The book will be bought English
b. *The book will bought be
c. Bogen vil blive kbt Danish
d. *Bogen vil kbt blive
The patterns found with the passive auxiliary are representative for
other auxiliaries and constructions like the causative construction with
let. The pairs of examples in (25c and d) and (26) are the counterparts
of (25a and b) with respect to the relative order of the causative verb and
the innitive.
25 a. He has let us wait English
b. *He has wait let us
c. Han har ladet os vente Danish
d. *Han har os vente ladet
26 a. Er hot undz gelozt vartn Yiddish
b. Er hot undz vartn gelozt
c. Hij heeft ons laten wachten Dutch
d. *Hij heeft ons wachten laten
e. *Er hat uns lassen warten German
f. Er hat uns warten lassen
On the basis of this extensive, robust, and diverse enough evidence, in

combination with the scrambling property of Yiddish, Vikner (2001,
p. 86) concludes that it is obvious that an account of Yiddish as an OV
language will have far less problems to deal with than an account of
Yiddish as a VO language would.
With Yiddish as a representative case for a property common to
historic stages of Germanic languages, we arrive at this result: OV
and VO are but the opposite settings in a system of merger with xed
directionality values. But crucially, these settings are not exhaustive.
There is a third possibility, namely an adjustable directionality value.
This property is responsible for the mixed appearance of linear order
in languages like Yiddish or Hindi, and it is the key for under-
standing the diachronic development of Germanic languages from a
common stage with adjustable directionality values to the split when
the values got xed. Fixing meant the choice of one of two available
implementations, namely right or left, with VO and OV as the
resulting manifestation. The split seems to be independent of the
morphosyntactic set-up. So Icelandic and German ended up with
dierent directionality values in spite of their common morphosyn-
tactic inventories.
3. Some Icelandic-German contrasts and their OV/VO based

grammatical source
Having prepared the background, let us now analyze in more detail

some prominent areas of syntactic contrasts between Icelandic and
German in order to check whether they indeed follow from a single
common source, namely the opposite organization of the V projection
and its extensions in terms of the headedness value, that is, in terms of
the basic OV vs. VO organization of the clause.
3.1. Quirky subjects
Quirky subjects are the joint result of four factors, namely the need of
directional licensing of arguments by a head, a head-initial V projection,
relational (and not strictly positional) agreement checking, and verbs
whose highest ranked argument in the argument structure is not the
nominative candidate. If the highest argument is not the candidate for
nominative, the result is a quirky subject construction in Icelandic but
24 HUBERT HAIDER
not in German. In German, these verbs are just verbs with a non-
nominative DP preceding a nominative DP in the VP.
27 a. a henni=stelpunum lkuu hestarnir:

that herDat =girls-theDat lked3:Pl horses-theNom
that she/the girls liked the horses Sigursson 2002
b. da ihr=den Madchen
die Pferde gefielen
that herDat =theDat girls the horses pleased
that the horses pleased her/the girls
Arguments are merged in the order of their ranking in the argument
structure. The highest ranked argument will end up in the top-most
argument position of the V projection (see 15). In this position, it is not
directionally licensed by the verb if the verb licenses to the right (VO).
So it needs a functional head as licenser in this position. The required
functional layer introduces a head and a spec position. The unchecked
DP moves to this functional spec position. Here it is licensed under the
spec-head relation. This is the structural subject property.
German and Icelandic share three of the four factors and dier in a
single factor, namely the licensing directionality of the verb. In German
(see (14b)), all argument are directionally licensed already in their VP-
internal positions. Hence there is no grammatical trigger for moving a
particular unlicensed argument to a functional spec position.
Crucial dierences between German and Icelandic are already high-
lighted in Zaenen et al. (1985), who concluded that German does not
have quirky subjects. Sigursson (1989, pp. 204205) discusses in detail
a wide range of contrasts (reexivization, PRO-subjects, conjunction
reduction, subject position in ECM innitives, raising) and re-empha-
sized this conclusion. Fanselow (2002) and Bayer (2003) analyze the
corresponding data in German and conrm the conclusion that German
does not show quirky subject eects.
Let us summarize: A head-initial VP is a VP whose licensing direc-
tionality is to the right. This is the opposite of the universal direction-
ality of merging, namely to the left.24 So the position preceding the verb
in the head-initial VP is not within the directional licensing domain. It is
functionally licensed. The need of functional licensing is the indirect
source of the EPP25 property of VO languages (Chomsky 1995, p.428).
The grammar provides functional means for licensing the external
argument of the VP. If the verb does not provide a suitable candidate
for the functional spec position, an expletive is used, or, if there is no
suitable expletive, the construction is ill formed (as in the case of the
passive of a single argument verb in English). Note that this is a
structural issue and not a matter of supply on demand. The projection
of the functional layer is automatic and not conditioned by the argu-
ment structure of the head of the VP. The spec must be lled.
Head-nal V projections dier in an essential respect from head-
initial ones: the direction of merger harmonizes with the licensing
directionality. No functional projection is needed for licensing a phrase
that is merged with a V projection but cannot be licensed VP-internally,
simply because there is no unlicensed phrase. So, in OV languages, there
is no grammatical necessity for a functional projection that licenses the
subject of a clause.26 Hence, contrasts between Icelandic and German in
terms of EPP eects are expected. This is the topic of the following
subsection. The rest of this subsection is a critical commentary on two
recent proposals that relate to the above analysis, namely one by Si-
gursson (2002a, pp. 126f.) and one by Bardals (2002) as this was
recommended by a reviewer.
Sigursson (2002a, pp. 126f.) suggests that the crucial grammatical
dierence that accounts for the presence or absence of quirky subject
constructions is one of agreement: Icelandic, but not German, is argued
to provide morphologically silent person agreement for the quirky
subjects and therefore a trigger for movement to the spec position. In
other words, a quirky subject is attracted by silent agreement features.
In this view, the dierence between German and Icelandic is a purely
accidental one (presence or absence of silent features). Moreover, it fails
to capture the generalization that a nite verb checks the agreement
features of the functional heads it is associated with: if the verb checks
the features of the functional head whose spec hosts the quirky subject,
it ought to agree with the quirky subject if there is agreement at all.
Since Icelandic has person agreement with nominative subjects in this
position, it ought to agree with the quirky subject in person features, but
it does not. The predicted outcome is ungrammatical (28):
28 Mer hafa= h
ofum ekki lka hestarnir:
meDAT have3Pl= 1Sg not liked horses:the:NOM
I did not like the horses:
According to Sigursson (2002a, pp.125f.), the person feature in a
clause like (28) is engaged in (invisibly) matching the person of the
dative subject and is thus blocked from agreeing in person with the
nominative object.27 But, on the other hand, the VP-internal
26 HUBERT HAIDER
nominative obligatorily agrees in number (27a), and only in number.

Consequently, the nite verb in (28) could be associated with a person
feature, namely 1st person, and a number feature, namely plural.
Therefore, the predicted morphological form would be 1st person plural.
But this is ungrammatical, of course.28 So agreement does not seem to
be the key for the dierences.
The second comment concerns Bardals (2002) claim that German,
contrary to the consensus among grammarians working in this area (see
the reference cited above, under example (16), allows oblique subjects.
The problem with this claim is this: it is based, on the one hand, on
ambiguous diagnostics for subjecthood and, on the other hand, on
deviant German data, gathered on internet sites. I do not see any reason
to abandon the established consensus that German does not allow
quirky subjects. Let me give examples for each of the two types of
shortcomings.29 As for reexives and coordination reduction, the
arguments are not compelling, and as for oblique PRO (illustrated with
web citations), the data are unreliable.
The fact that a (strong) reexive in a PP may be bound by a dative or
an accusative in an impersonal construction in German does not
contrary Bardals (2002, pp. 72) claim show anything about subject-
hood since objects may generally bind PP-internal reexives, as in (29).
29 a. Beim Quadrieren mu man eine Zahl mit sich multiplizieren

when squaring; one must multiply a number with itself
b. Wir haben ihnen f
ur sich selbst Kleider beschafft
we have them for themselves clothes procured
Only if the reexive is a co-argument (and not part of a PP), can a direct
object anaphor not be bound by a dative, but it must be bound by the
nominative subject as illustrated in (30). Note, however, that passive
does not change the situation. But, unlike Icelandic, the dative object
cannot be promoted to the subject function.
30 a. Wir stellten den Leutenj sichj = einanderj vor

we introduced the peopleDat themselves=each otherPRT
b. *Den Leutenj wurde sichj =einanderj vorgestellt
the peopleDat was themselves=each other introduced
Coordination reduction is not a reliable subject diagnostic except for

asymmetric coordinations. Symmetric coordination allows the ellipsis of
positionally matching constituents in the second conjunct (independent

of their grammatical functions, as long as they are identical):
31 a. IhmDat hat kein Rat geholfen und ih
m wird keiner schaden k
onnen
him has no advice helped and him shall no-one be-able-to harm
b. SieAcc hat keiner angerufen und sie wird auch keiner erreichen konnen
her has no-one phoned-up and her shall also no-one reach be-able-to
So an example like (32), from Bardal (2002, pp. 73), does not bear on
the subject issue since it is but an instance of symmetric coordination
that applies to subjects as well as to objects just as in (31).
32 MichAcc hungert nach Brot und m
ichAcc d
urstet nach Wasser
me longs for bread and me thirsts for water
What seems to be sensitive to subjecthood is asymmetric coordination,
but note that it crucially does not tolerate oblique arguments of
impersonal constructions. As (33c) shows, the dative in the impersonal
construction does not behave like the subject in (33a).
33 a. In den Wald ging der Jager und [er schoss einen Hasen]
into the wood went a hunter and he shot a rabbit
b. In den Wald schickte ich den Jager und er scho einen Hasen
into the wood sent I the hunter and he shot a rabbit
c. Im Zoo schauderte mir vor Baren und mir w
urde auch im Wald
in-the zoo shuddered meDat at bears and me would also in-the wood
davor schaudern
at-it shudder
A conclusive case of an oblique subject would be the systematic avail-
ability of a quirky subject PRO as in Icelandic. As already mentioned,
German does not allow this. This is uncontroversial not only among
theoretic syntacticians but also among descriptive grammarians.
Bardal (2002, pp. 94f.) contests this with data gathered on web sites (all
of which are deviant for me) as illustrated for instance in (34):
34 Christus selber, der w
urdig ist, geliebt und gefolgt zu werden
Christ himself; who worthy is loved and followed to be
Christ himself, who is worthy of being loved and followed
As for this particular case, a standard descriptive grammar of German
(Drosdowski 1984, p. 183) has a note on folgen (follow) that says that
28 HUBERT HAIDER
there are rare cases in vernacular usage in which folgen in passive is used
with nominative instead of dative.30 Note, however, that in this case the
innitival subject is not quirky, but the dative has been replaced by a
nominative. This, and not a quirky dative subject, is the source for the
PRO construction with folgen as in (34).31
In general, the value of unanalyzed data of unclear language com-
petence (web citations) is at best heuristic. If oblique subjects existed (in
varieties of) German, they should be detectable in corpora of German
authors, they would have been noticed in one or the other descriptive
grammar, and they should be detectable with systematic informant
questionnaires. Unfortunately, this is not the case. Hence I fail to see
any evidence for hitherto undiscovered quirky subjects in German in
general and in innitival constructions in particular.
3.2. Subjectless clauses
For a VO language such as Icelandic or Faroese (Vikner 1995, p. 227),

the existence of a functional subject position is uncontroversial. This
position is open for an expletive if the position is not targeted by
movement. German, on the other hand, seems to provide no room for
an expletive in these cases:
35 a. a a hefur veri dansa Icelandic
that EXPL has been danced
b. da es getanzt wurde
that EXPL danced was
c. I dag er a komin ein drongur Faroese
today is EXPL come a boy
d. Heute ist (*es) ein Junge gekommen
today is EXPL a boy come
In German, expletives are systematically ungrammatical in contexts in

which they are obligatory in Icelandic or Faroese.32 What these con-
trasts illustrate is what we expect: if there is no functional spec position
for the subject, there is no need for an expletive subject. If EPP is a VO-
specic side eect, it is not justied to elevate it to the rank of a uni-
versal requirement.
At rst glance, it seems easy to shield the universal EPP hypothesis
against this particular counterevidence with an ancillary hypothesis,
namely an expletive empty pronoun hypothesis. However, this is

clearly an unsatisfactory move. First, it would be ad hoc, and second, if
one admitted an expletive pro in cases like (35b and d) in German, one
would invite massive overgeneration.
As for the rst objection, the examples in (36a, and b) show that
German requires overt expletives in uncontroversial spec positions,
namely in the topmost spec position in a V2 clause. So if an expletive
for a spec position is available, it should be available also for the spec
position in (36c), given that Icelandic requires (or at least admits)33 an
overt expletive (35a) even though it requires pro-drop for quasi
arguments, such as for instance the subjects of weather verbs, as in
(36d).
36 a: Es wurde nicht getanzt

EMPL was not danced
b. Ich glaube CP es wurde nicht getanzt

embedded V2-clause complement
I think EXPL was not danced
c. Ich glaube CP da es nicht getanzt wurde

I think that EXPL not danced was
d. Hefur a rignt n
ott? Icelandic
has EXPL rained tonight
e. Hat (es) geregnet in der Nacht?

has it rained in the night
In German, non-referential subjects cannot be dropped. This is true

for the quasi arguments of weather verbs (36e) but also for the
expletive argument in the intransitive middle construction (37a)
and for the impersonal variant of a class of transitive verbs. (37b) In
Icelandic, the non-referential subject of these verbs is dropped (37c).34
In German it is obligatorily present (see Haider 2001). Finally,
(37d) illustrates an extraposed clause that relates to an obligatorily
overt pronominal subject. In all these cases, the pro-drop hypo-
thesis makes the wrong prediction for German since it would
overgenerate.
30 HUBERT HAIDER
37 a. Hier lebt *(es) sich gut intransitive middle construction

here lives it itself well
b. Das Boot hat *(es) in Stucke zerrissen
see Paul 1919, vol. 3, Section 24
the boatAcc has it into pieces broken
c. Batana hefur pa broti sp
on
boat-theAcc has it broken in pieces
d. War *(es) sehr peinlich, da er betrunken war?
was it very embarassing that he drunk was
The so-called stylistic fronting of participles (or innitives or various
other elements) in Icelandic is another piece of contrasting evidence. In
Icelandic, participles or innitives may be fronted to the spec position.
In German, fronting to the spec position in a V2 clause ((38c)) is pos-
sible but not fronting to a position below CP Vembu((38b)).
38 a. efi engi er eftir Laugaveginum
if walked is along the Laugavegur
b. *wenn gewandelti entlang des Laugavegur ei wird
if walked along the Laugavegur is
c. [Spec-F Gewandelti wird entlang des Laugavegur ei
walked is along the Laugavegur
If stylistic fronting in Icelandic targets the spec position (Holmberg
2000), this satises whatever principle is responsible for the EPP. In
German, the absence of the corresponding functional projection ex-
plains the absence of fronting participles and innitives to this
position.35
In sum, the evidence from impersonal constructions and the distribu-
tion of expletives conrms the analysis proposed for the quirky-subject
phenomenon: Icelandic has a functional layer for the subject, German
does not, and the contrast is predictable as a consequence of the OV/VO
distinctions, with OV as a necessary (but, as it seems, unfortunately not as
a sucient) condition.36
3.3. Scrambling and object shift
Scrambling applies in German, object shift in Icelandic. The precondi-

tions and the results are suciently dierent. The analysis for scrambling
I adopt in this section is presented in detail in Haider and Rosengren

(2003). It is an adjunction-to-VP analysis. The result is an extended VP.
What is important for the present discussion is that the scrambled
arguments are adjoined to, but remain within, the VP and therefore
within the licensing domain of the verb. In a VO language, adjunction to
the left (the only admitted adjunction direction) would remove the ad-
joined elements from the directional licensing domain of the verb.
A dening property of scrambling is a change in the relative order of
arguments. Since scrambling is found even in topicalized VPs (39) in
German, a VP must be a possible scrambling domain. Moreover, the
order alternations like those in (39a, and b) in German are movement
eects (see Haider and Rosengren 2003, pp. 223227 for arguments
based on focus and scope data) rather than alternative base orders (see
Fanselow 2001).
39 a: VP Einem kleinen Kind einen groen Hund anvertrauen
wurde ich niemals
aDat little childDat a big dogAcc entrust would I never
entrust a little child with a big dog, I would never
b: VP Einen groen Hund einem kleinen Kind anvertrauen
w
urde ich niemals
a big dogAcc a little childDat entrust would I never
entrust a little child with a big dog, I would never
The topicalized phrase in (39) cannot be a higher functional projection
that contains the VP. If it were a higher projection, it would contain
functional heads whose features need to be checked by the nite verb.
Hence, the fronted constituent would contain traces of the nite verb.
This is ungrammatical, however:
40 a: VP Einem kleinen Kind einen groen Hund an-ei vertrautei ich niemals
a little childDat a big dogAcc en-37 trusted I never
b. Einem kleinen Kind vertrautei ich niemals einen groen Hund an-ei
The verb an-vertrauen (entrust) is a verb with a separable particle prex.

Moving the verb to the V2 position strands the particle in the base
position of the verb ((40b)). (40a) clearly shows that the fronted VP
cannot contain a trace of the nite verb, for an independent reason,
namely a crossing violation, since the antecedent does not c-command
its trace. If, however, the fronted constituent in (39) were a higher
projection, this would apply to (40a) as well. The nite verb raised
32 HUBERT HAIDER
through these functional projections would leave traces. Fronting one of

these projections would produce the same violation as in (40a).
(41a and b) are the passive variants of (39a and b). The fronted VP
contains a nominative argument. So nominative agreement is denitely
involved, and Agr-S would denitely qualify as a higher functional
projection. Nevertheless, the dative DP and the nominative DP may be
scrambled. A parallel case is (41b and c). The verb is an unaccusative
verb, and the arguments may scramble.
41 a: VP Einem kleinen Kind ein groer Hund anvertraut] wurde hier noch nie
a little childDat a big dogNom entrusted was never ever
Never ever was a little child entrusted with a big dog:
b: VP Ein groer Hund einem kleinen Kind anvertraut wurde hier noch nie
c: VP Einem Linguisten ein Gespenst erschienen ist hier noch nie
a linguistDat a ghostNom appeared is here yet never
A ghost has never ever appeared to a linguist here:
d: VP Ein Gespenst einem Linguisten erschienen ist hier noch nie
a ghostNom aDat linguistDat appeared is here yet never
The examples are illustrations both of VP-internal scrambling and of

VP-internal nominatives. Icelandic allows VP-internal nominatives, but,
unlike German, it does not allow scrambling to a position preceding the
subject ((42c)) or to the left edge of the VP ((42d)), nor does it allow
freely scrambling arguments VP-internally (42c and d).
tlaa gefa b
42 a. Eg oksafninu b
okina
I want to give library-theDat book-theAcc
tla a gefa b
b. *Eg okina b
oksafninu
(Holmberg and Platzack 1995, p. 201)38
I want to give book-theAcc library-theDat
c. *I gr lasj bkurnari J
on ekki VP ej ei
Collins and Thrainsson 1996, p. 410
yesterday read books-the John not
d: I gr hefur J
on bkurnari VP lesi ei
yesterday has John books-the read
e: I gr hefur J
on VP lesi bkurnar
yesterday has John read books-the
The source of the contrast is the dierence in the VP structure,

namely the dierence between a head-initial and a head-nal V pro-
jection. As discussed already in Section 2.2, the licensing directionality
of the verb in a head-nal VP harmonizes with the universal direc-
tionality of merger. Therefore, all positions merged with V or one of
its projections are within the directional licensing domain of V (see
(43a)). A head-initial VP has a more complex structure (see Section
2.2). Since the licensing direction is the converse of the direction
of merger, a complex head-initial VP must have a shell structure
((43b)).
43 a: Vmax XP YP ZP V
b: Vmax XPVi ! YPVi ! ZP
The shell structure is a straightforward consequence of the mismatch of

merger and licensing directionality: the YP in (43b) cannot be licensed
directionally unless the verb is re-instantiated in a shell structure. Each
shell licenses one position and opens another one. In (43b), YP and ZP
are in dierent licensing domains. ZP is licensed by V in the bottom
projection; YP is licensed by the V of the higher V projection. In (43),
however, the position of YP, just like that of XP or ZP, are within the
same licensing domain.
This principal dierence with respect to the licensing of positions in a
lexical projection is the key to understanding the correlation between
scrambling and the VO/OV property. (43a) is a homogeneously layered
structure. (43b), however, is a composite structure, consisting of two
shells. The argument positions are uniquely identied by their licensing
head. Directional licensing is a one-to-one relation in (43b). In (43a),
however, the verb in its base position has all arguments in its directional
licensing domain.
A comparison of Dutch and German reveals that phrases must be
morphologically identiable. Dutch lacks case morphology. Direct
and indirect objects do not dier by overt case morphology. They do
not scramble ((44b)) unlike German ((44c)). PP objects ((44e)), how-
ever, scramble (Geerts, Haeseryn, de Rooij and van den Toorn 1984,
pp. 989), just as they do in German.
34 HUBERT HAIDER
44 a. Toen hebben de autoriteiten de moeder het kind

teruggegeven Dutch
then have the authorities the mother the child back-given
b: * Toen hebben de autoriteiten het kindi de moeder ei
teruggegeven Dutch
then have the authorities the child the mother back-given
c: Dann hat die Beh orde das Kindi der Mutter ei ;
zuriickgegeben German
then has the authority the mother the child back-given
d: Toen hebben de autoriteiten het kind aan de moeder
teruggegeven (Dutch)
then have the authorities the child to the mother back-given
e: Toen hebben de autoriteiten aan de moedeni het kind ei
teruggegeven Dutch
The contrast between (44b) and (44e) is instructive. It shows that a

scrambled order must be detectable.39 Dutch does not provide the nec-
essary morphosyntactic means for identifying a scrambled order.40 The
order in (44b) is a possible base order for a sequence of two DPs in Dutch,
and there is no structural or morphosyntactic indicator of scrambling.
On the other hand, the order in (44e) is clearly a scrambling order, given
that the base order for a verb like teruggeven is DP-PP-V ((44d)).
Obviously, scrambling does not target a structurally identiable position
(such as for instance a specic spec position) because then the need for
overt morphological distinctions would not matter since a spec position
could be identied by the functional head. But if scrambling is reordering
within a single domain of identication, the morphosyntactic identi-
cation matters. Otherwise, a scrambled order cannot be identied.41
Let us now turn to Icelandic. As a consequence of its head-initial VP
structure, Icelandic arguments cannot scramble to the left of the verbal
head. Left-adjunction to the VP would create a position outside the
directional licensing domain. Order variation within the VP is possible
to a certain extent, however. Collins and Thrainsson (1996, p. 417) claim
that for ditransitive verbs with a [VP V-Dat-Acc] base order only,42 the
order [VP V-Acc-Dat] is possible if the Dat-DP is stressed.
Icelandic does not scramble to the left of the VP, but it shifts objects.
Object shift and scrambling dier clearly. First, object shift does not
apply across any phonological material in the predicate except predicate
adjuncts such as the negation word and other adverbs (Holmberg and
Platzack 1995, p. 165). In other words, object shift applies only in a
clause with a single verb that has moved to the topmost functional head
position. Second, object shift does not change the relative order of
arguments. Scrambling changes the relative order, and the surface po-
sition of the verb in the clause does not interact with scrambling.
Icelandic is unique among the Scandinavian languages, insofar as in
the other Scandinavian languages object shift is restricted to pronouns.
Holmberg and Platzack (1995, pp. 172f.), however, report that to a
limited extent non-pronominal arguments may shift: in Norwegian
varieties of Swedish and Faroese, a non-pronominal DP of a double
object construction may be shifted, again without change of the relative
order. Single, non-pronominal objects do not shift (see (45c and d)).
45 a. De ga Marit ikke/gjerne blomstene (Norwegian) H&P, p. 172
they gave Marit not=gladly flowers-the
b. Vi ger barnen altid/inte vad de vil han (Swedish) H&P, p. 172
we gave children-the always=not what they want have
c. Eg visi brnunum fegin myndir mnar (Faroese) H&P, p. 174n.
I show the children happily my pictures
ogvan kennir hana= Siggu ikki (Faroese) H&P, p. 172
d. J
John knows her=Sigga not
Let us recapitulate: An adequate account of object shift should capture
at least three generalizations. First, in all Scandinavian languages,
object shift is contingent on V movement. Second, in all Scandinavian
languages, object shift preserves the relative order of arguments. Third,
Icelandic is a Scandinavian language with overt V-to-I movement, and it
is a Scandinavian language with full DP object shift. None of these
generalizations is characteristic of scrambling. But nevertheless,
scrambling and object shift have one property in common, namely the
availability of an extended licensing domain of the main verb.
In a head-nal V projection, the extension is simply a consequence of
the fact that a phrase that is merged on the left-hand side of the V
projection remains within the directional licensing domain of the verbal
head. In a head-initial VP, the extension requires additional shell
structures. In the case of object shift structures, the domain of the main
verb qualies as extended by virtue of V movement.43 This is conrmed
by the fact that the structural relation between the verb and the shifted
object obeys the same constraint as the relation between the verb and
the object in the base position. Elements that cannot intervene in the
base structure cannot intervene in the shifted structure. This is illus-
trated in (46) with data from Vikner (1994, p. 507).
36 HUBERT HAIDER
46 a. Petur lasi eflaust ekki ei b

okina
Peter read doubtlessly not book-the
b. Petur las b
okina eflaust ekki

c: Petur las eflaust b
okina ekki
d: Petur hefur lesi eflaust b
okina
Peter has read doubtlessly book-the
In sum, again the directionality of licensing is the crucial factor.
Scrambling is dependent on head-nal projections. Object shift is a
particularly constrained option of extending the licensing domain of a
head-initial VP.
3.4. Particle constructions
Verb + particle constructions are a good indicator of verb movement if

in the given language particles are stranded. In Haider (1997c), the
distribution of particles in English and continental Scandinavian lan-
guages is analyzed as the eect of particle stranding in a VP shell
structure. The positions of the particles in English are V positions in the
VP shell structure. The verb positions in (47) are possible particle
positions.
47 a: sendi someone Vi something
b: sendi someone Vi somethingVi to some place
In English, a particle may be V-adjacent (48a) or stranded. In a double
object construction like (47a) the stranded particle may be sandwiched
by the objects (48b), but it must not follow the objects. This is possible
only in a 3-place construction like (48c). The distribution follows
immediately from the shell structure of the V projection.
48 a. send out the clients the mail (to their holiday resorts)
b. send the clients out the mail (to their holiday resorts)
c. send the clients the mail out (to their holiday resorts)
English is a language that allows particles to be stranded or to remain
V-adjacent44 as in (48a). Icelandic particles must be stranded as Collins
and Thrainsson (1996, pp. 435f.) illustrate. The V-adjacent position
(49a) is ungrammatical in Icelandic.
49 a. I gr hafa peir sent upp strakunum peningana

yesterday have they sent up the boys the money
b. ?I gr hafa beir sent strakunum upp peningana
c. I gr hafa peir sent strakunum peningana upp
In the analysis sketched above, clause-nal particles must be analyzed as
result predicates, that is, as part of a complex predicate (Haider 1997c).
Den Dikken (1992) emphasizes that in English (see also Quirk et al.
1985, p. 1152), a particle in nal position accepts a modier but not in
the medial position. The Icelandic distribution is parallel (Neeleman
and Weerman 1999, p. 98). This shows that a clause-nal particle may
be a syntactically separate unit, but the medial one must be a stranded
part of a complex verbal predicate.
50 a. I gr hafa eir sent strakunum peningana beint upp

yesterday have they sent the boys the money straight up
b. I gr hafa peir sent beint upp strakunum peningana

In German, there is a single position for particles, namely the V-adja-

cent one. Particles are stranded only when the nite verb is fronted
(51b). Topicalization, of course, does not strand a particle (51b). A
strandable particle must be stranded (51c), and a particle cannot intrude
into the VP (51d).
51 a. Sie stelltei ihm Gaste vor-ei
she put himDat guestsAcc before
vor-stellen; lit. before-put introduce
She introduced guests to him.

b. Stelleni wird sie ihm Gaste vor-ei
c. Sie vorstellte ihm Gaste
d. Sie stelltei ihm vor Gaste
The contrasts in particle distribution between Icelandic and German are
a direct reex of the organization of the respective VPs. In a head-nal
V projection, there is a single head position and therefore only a single
particle position. The shell structure of a complex head-initial VP,
however, provides more than one position for stranding a particle since
it provides more than one verb position.
38 HUBERT HAIDER
The serialization (particle + verb in VO, verb + particle in VO) is a

reex of the converse directionality requirements of the verb. This is not
directly honored by an analysis that derives OV from VO. The fact that
in Germanic OV languages, particle verbs whose particles are postverbal
under V2, nevertheless always have preverbal particles in non-V2 con-
texts whereas the fact that in the Germanic VO languages particle verbs
whose particles have to be stranded under V2 never have preverbal
particles in non-V2 contexts (Vikner 2002, p. 37) does not follow in the
derivational account of OV (Kayne 1994), but it follows immediately in
the directionality account.
3.5. VP stacking vs. V clustering
In languages with head-initial VPs, verbs that select a VP are projected/

merged in stacked VPs (52a). In languages with head-nal VPs, they are
merged in clusters (52b). Stacked head-initial VPs are right-branching
structures. Stacked head-nal VPs would be left-branching (52c). This
may be taken to be in violation of the branching constraint (BBC:
branching nodes of the functionally or lexically extended projection line
follow their sister node; see note 20). Icelandic stacks, German clusters.
In a nutshell, this is the topic of this subsection. For a more extensive
discussion the reader is referred to Haider (2003).
The bracketing in (52) already reveals the crucial factor. In (52c), the
parser would need to guess how many brackets to open; in (52a), the
parser encounters the head of the VP rst, with the maximal projection
of the head as the immediately dominating node. Clustering (52c) re-
duces the indeterminacy for the parser to the local environment of a
single V projection. Clustering is UGs contribution to parser friendli-
ness.45
52 a: VP V1 VP V2 VP V3 . . .
b: VP . . . V V V1 V2 V3
c: VP VP VP . . . V1 V2 V3
Icelandic verbs, like those in English and the other Scandinavian lan-
guages, do not cluster, neither with auxiliaries nor with other verbs.
Why do German and Dutch verbs cluster? The answer must again be
sought in the headedness dierence: the constraint against left-branch-
ing (extended) projections not only rules out structures with head-nal
functional heads (see Section 2.1) but also lexically extended46 head-nal
V projections, that is, a V selecting a VP as its left-hand complement.
(52a) is the structure with stacked, head-initial VPs. (52b) is the corre-
sponding structure with head-nal VPs. If we regard these structures as
lexically extended V projections, the branching constraint admits (52a)
but excludes (52b): a head-initial (extended) projection is well formed
with respect to the branching constraint; there are no left brackets
adjacent to each other. This is just another way of expressing the fact
that there are no right-hand side sisters to nodes on the main projection
line. The right daughter always is the node on the projection line. (52c)
is the grammatical alternative to (52a) that does not violate the
branching constraint. The projection line of the VP starts with the
highest V node. It is the projection line of a simple VP. The cluster is a
head-to-head merger structure (see Haider 2003 for details).
Let me rst sketch some empirical evidence for clustering in German,
then briey sketch how theta management works in the two types of
merger (phrase to head, head to head), and then readdress the general
issue of why German and Dutch cluster and why VO language never
cluster. The answer will shed light on an additional issue, namely the
Dutch-German dierence with respect to linearization in the cluster.
In English, auxiliaries are projected in separate VPs. The fact that it is
possible to separate auxiliaries by intervening adverbials tells us that the
structure for (53a) is one with stacked phrasal projections. Each auxil-
iary projects a VP and selects a VP (or a functional extension thereof) as
its complement.
53 a. The new theory certainly may possibly have indeed

been badly formulated
(Quirk et al. 1985, Section 8.20, p. 495)
b. da die Theorie wohl tatsachlich
schlecht formuliert
(*) worden (*) sein(*)mag
that the theory possibly indeed badly formulated been be may
In German, the series of verbs is the mirror image of the English pat-
tern, but intervening material is strictly excluded. A robust and char-
acteristic property of German (and Dutch) auxiliaries is the adjacency
property of the verbs in the cluster. The bracketed asterisks in (53b)
mark positions that do not allow any intervening non-verbal material.47
This is not at all a peculiarity of adverbs, but it holds for any potentially
post verbal material as potential intervener. A comparison with topi-
calized (remnant) VPs is instructive. These data show that (remnant)
40 HUBERT HAIDER
VPs may contain extraposed material, hence potential postverbal

interveners in stacked head-nal VPs.
54 a: Gesprocheni mit uns haben sie nicht ei
spoken with us have they not
b: da sie nicht [gesprochen mit uns haben
that they not spoken with us have
c: da sie nicht gesprochen haben (werden) mit uns
that they not spoken have will with us
d: da sie nicht gesprochen haben mit uns werden
that they not spoken with us have will
What these data show is, rst, that the cluster structure of auxiliaries
and the main verb does not leave any room for intervening material.
There would be enough room, however, if the clusters consisted of
stacked (remnant) projections because on the one hand a (remnant) VP
allows extraposed material (see VP topicalization with extraposition in
(54a), and on the other hand English shows that V projections (plus
their functional extensions) allow intervening material. So the no
intervener property in German or Dutch is not reducible to a constraint
on verbs selecting V projections.48
A serious obstacle for derivational accounts of clustering is the clause
union property of clustering constructions. This can be best observed
with binding relations (Chomsky 1981) in sentential innitival com-
plements that alternate with a clustering construction. (55) shows that
the binding of a reexive (Binding Principle A) is aected by clustering,
and (56) illustrates that the binding environment for a pronominal
(Binding Principle B) changes. The grammatical cause is identical. The
clustering construction is a simple sentence (clause union). Hence an
anaphor must be bound in this domain, but a pronominal must be free.
In (55a), the antecedent of the anaphor is PRO and is controlled by
the matrix dative object. In the cluster construction (55b),49 there is no
embedded sentence, hence no PRO.50 In (54b), the mono-sentential
clustering construction, the reexive would have to be directly bound by
the dative object, but co-argument reexives cannot be bound by datives
in German. Thus the reexive ends up without a suitable binder, whence
its ungrammaticality.
55 a. da PROi sichi =es zu entfernen ihneni nicht erlaubt wurde

that themselves=it to remove themDat not allowed was
They were not allowed to go-away=remove it
b: VC Zu entfernen erlaubtwurde es= sich ihnenDat nicht

Given the described eect for anaphors, there is a straightforward

prediction for pronominals: if clustering is a mono-sentential con-
struction, a pronoun bound by a matrix element in the bisentential
control construction will be ungrammatical in the clustering construc-
tion. In (56a and b), the pronoun can be bound by the matrix subject; in
(56c and d) it must be free. (56d) shows that the position of the pro-
nouns fronted or not does not matter. The binding dierence be-
tween (56a and b) and (56c and d) follows immediately from the
respective structures.
56 a. Mariai hat unsj PROj siei zu besuchen] nicht erlaubt
Mary has us her to visit not allowed
b. daPROj siei zu besuchen] unsj Mariai nicht erlaubt hat
that her to visit us Mary not allowed has
c. VC Zu besuchen erlaubt] hat siei uns Mariai nicht
to visit allowed has her us Mary not
d. VC Zu besuchen erlaubt] hat Mariai siei uns nicht51
to visit allowed has Mary her us not
How are the theta requirements satised in the cluster? In the cluster, a
head selects another head with appropriate morphosyntactic features
(e.g., bare innitive, participle, verb with innitival prex). The result is
the merger of two heads into a head-head complex. The theta grids of
the two elements are merged too. The theta slot in the theta grid of the
selecting element is replaced by the theta grid of the selected verb. The
result is a verbal cluster whose theta grid is a pooled theta grid.
57 a: erlauben permit :< A1 ; A2Dat ; A3 >
b: zu besuchen tovisit :< B1 ; B2 >52
c: zu besuchen erlauben :< A1 ; A2Dat ; < B1 ; B2 >>
)< A1 B1 ; A2Dat ; B2 >
If the complement slot A3 replaced by the theta grid of the selected
innitival verb, and the theta grids are merged, the result is non-distinct
42 HUBERT HAIDER
from the theta grid of a simple verb except for the theta identication of
the controlled subject argument of the innitive.
Imagine now what will happen if the clause union variant is passiv-
ized. Just as in the case of a simple verb, the primary subject theta role
gets blocked, and the direct object ends up with nominative case instead.
This is exactly what happens in German in the case of the so-called long
passive, noted rst by Hohle (1978) and illustrated in (58a). (58b) is the
sentential complement variant. The verbs are separated by a non-verbal
element (the negation), hence clause union is not possible, that is, the
verbs are not clustering. Note that these innitival constructions are
optional clause union constructions.
58 a. da sieNom uns nicht zu besuchen erlaubt wurden3rdPl:
that they us not to visit allowed were
b. da uns [sie zu besuchen] nicht erlaubt wurde= wurden3rdPl:
that us them to visit not allowed was=were
Let me nally return to the general issue. Why is V clustering an OV

phenomenon, and why is it not found in VO languages? The answer is
now straightforward. In VO languages, V clustering would not improve
parser friendliness, and, what is worse, it would violate the branching
constraint. In (59b), the clustering variant, the complex V cluster, is a
branching node as the left branch on the projection line of V . This is
ruled out by the branching constraint. Structure (59a), the stacked
projections, is optimal for parsing since the rst element of each new
sub-tree is the head of the sub-constituent, immediately dominated by
its maximal projection.
59 a: VP V1 VP V2 VP V3 . . .
b: VP V V1 V2 V3 . . .
Note that in this perspective a notorious puzzle nds an answer,
namely this: why is clustering obligatory only for in situ sequences of
verbs (with VP projection as an option for non-local environments, as
in the case of VP topicalization), As demonstrated in Haider (2003),
rst the branching issue does not arise in the extraction side if the
trace of the VP is taken to be atomic (and not a foil copy), and,
second, the fronted VP itself is still subject to the clustering
requirements for the verbs it contains. In sum, if clustering is ana-
lyzed as a BBC eect, the environments for obligatory clustering are
correctly identiable.
The nal comment concerns the dierent word orders in the cluster in
Dutch and German, respectively. As already mentioned, the V projec-
tion with the cluster (53c) reduces to a single VP the potential VP stacks
that the parser could not anticipate. The structural complexity is shifted
from the phrasal projections to that of a V cluster. This is a local
domain, which makes backtracking easy.53
(60c), the German cluster, is still a sub-optimal solution.54 The
optimal structure is a completely right-branching cluster as in (61). This
is the Dutch solution to the problem. In Dutch, the verbs not only
cluster (as in (60c)) but are also raised within the cluster. The resulting
structure of a fully inverted cluster (traces omitted) is (61). (Note that
the subscripts are merely notational devices for indicating the verb that
is the head of the respective VP.)
(60) a. VO b. *stacked OV c. OV+clustering
(61) Dutch (fully inverted) cluster
In sum, for OV languages, UG provides the clustering option as a

means of enhancing parser friendliness by avoiding center embedding in
the form of stacked head-nal VPs. This option is not found in VO
languages because the stacked structures with head-initial VPs are
parser friendly. Hence German shows clustering, but Icelandic does not.
4. Discussion and summary
The answer to the title question is as follows: change the directionality

value of the German verbal heads, and you get a lot of Icelandic syntax.
Let me briey compare the results summarized above with a competing
model before I summarize the main points of the paper.
44 HUBERT HAIDER
The discussion of the Icelandic-German contrasts in this paper pre-

supposes that the basic dierence between OV and VO languages is the
directionality of licensing in combination with the branching constraint
(BBC); see Haider (1992/2000, 2001). The shell structure for VO is an
immediate consequence of the directional mismatch of the licensing
directionality and the branching/merger directionality.
A competing hypothesis on the relation between OV ad VO languages
has been proposed by Kayne (1994) and for Dutch by Zwart (1993). In
these models, OV is derived from a basic VO structure: the postverbal
material is fronted to preverbal positions. To turn German into Ice-
landic would amount to undoing movements and re-establishing the VO
structure. In other words, German would turn out as a derivational
continuation of Icelandic. Let us ask therefore what the predictions of
this approach are, ceteris paribus, for the relation between Icelandic and
German.
First, if Icelandic allows quirky subjects, German would allow them
too. German and Icelandic would just dier with respect to the post-
verbal arguments in Icelandic. These would be fronted to preverbal
positions.
Second, German would show EPP eects like those in Icelandic. In
particular, it would require or at least tolerate expletives for impersonal
passives or there constructions just as in Icelandic. However German
does not require them; it forbids them.
Third, particle distribution is easy to understand if directionality (and
stranding) is acknowledged as the distinguishing factor. For the deri-
vational scenario of VO-to-OV, particle distribution is a complicating
factor.
Fourth, scrambled phrases would be phrases moved to preverbal spec
positions. Hence they would be expected to display the behavior of
phrases in preverbal spec positions. In particular, these phrases would
be predicted to be as opaque for extraction as preverbal subjects are.
This is not the case, however. It is uncontroversial that scrambled
clauses do not become opaque in German (see the discussion in Section
2.1, Example (9)).
Fifth, there is no trigger for clustering or clause union. It is unclear
why German should have to invert the word order in the cascade of
stacked VPs in auxiliary constructions.
Sixth, the universal VO theory lacks an identiable trigger for
mapping Icelandic into German, that is, for turning a well-formed VO
base structure into an OV structure. In this perspective, an OV language
appears to be an unexpected and redundant distortion of a well-behaved

VO system.
Seventh, in a diachronic perspective, an OV language is predicted to
be the continuation of a VO stage. What the diachrony of Germanic
languages tells is dierent. Strict VO systems are an innovation. A strict
VO or a strict OV system is the consequence of xed directionality
values. This is better captured as the result of a diachronic change in a
system with adjustable values than in a model that characterizes OV as a
transformational extension of VO.
The model defended in this paper provides insights on principled
grounds into the following major contrasts between Icelandic and
German:
EPP eects: head-initial VPs require an external licenser (Agr-S)55 for
the highest argument position in the VP. The licenser is a functional
head, whose spec position accommodates the licensee. In head-nal
VPs, any argument position within the VP is in the licensing domain
of the verbal head. Hence no external licenser is needed.
Expletive subject: The spec position of Agr-S is the source of the EPP
eects, and it is the position for an expletive subject. OV-type clauses
with relational nominative checking do not require a functional Agr-
S projection. If there is no obligatory functional spec position, there is
no room for an expletive for this kind of position. They do have
expletive arguments (quasi arguments), however.
Quirky subjects: In VO languages with relational nominative check-
ing, the spec position of Agr-S is not restricted to nominatives. It
accommodates the highest argument in the VP irrespective of its case
properties. The highest argument (and the relative hierarchy of
arguments in the VP in general) is determined by the ranking of
arguments in the lexical argument structure and its mapping onto the
phrase structure. In other words, the ranking determines the order of
merger. If the highest ranking argument is not the nominative can-
didate, the result is a quirky subject construction.
No scrambling to preverbal positions in VO: Clause-bound scram-
bling takes place within the licensing domain of the verb. Hence a VO
language such as Icelandic cannot scramble to a preverbal position.56
Multiple particle positions and particle serialization: Complex head-
initial VPs have a shell structure, that is, a structure with more than
one verb position. Strandable particles can be stranded in these
positions. In head-initial VPs, particles are postverbal. Head-nal VPs
46 HUBERT HAIDER
have a single verb position and hence a single preverbal particle po-
sition.
Stacked VPs, no clusters: Verb clustering is an OV property. Clusters
replace stacked, left-branching, extended V projections. Left-
branching VP stacks would violate the universal branching con-
straint. VO structures yield right-branching VP stacks. So clustering
is not at issue in a VO language.
Finally, a language-specic property of Icelandic distinguishes it from
German: Quasi arguments are not lexicalized in Icelandic. With this
property, the German variants (62a and b) would map perfectly onto
the Icelandic patterns in (62c and d). Drop the es in (62a), and you get
(62c), the Icelandic stray accusative construction (see Haider 2001),
and the puzzling accusative57 in (62c) has a straightforward account:
both in (southern) German and in Icelandic, there is a class of
transitive verbs with an impersonal variant (Jonsson 2001). The quasi
argument gets dropped in the impersonal variant in Icelandic.
62 a. Den Schornstein hat es vom Haus geweht

the chimneyAcc has it from the house blown
b. Der Schornstein wurde vom Haus geweht
the chimneyNom was from the house blown
c. Strompinn bles af h
usinu
the chimneyAcc blew off the house
d. Strompurinn var blasinn af h
usinu
the chimneyNom= Acc was blown off the house
The accusative moved to the subject positions satises the EPP and
acquires subject properties by virtue of being the structural subject, that
is, the phrase in the functional spec position that other wise accom-
modates the nominative subject (see Svenonius 2002 for an alternative
account).
Notes
1. A reviewer points out that a trigger is readily found: an EPP property of a VP-external F-head
(v, in particular). However, if this were so, the EPP (Extended Projection Principle; see foot-
note 25) property could be satised by inserting an expletive. Moreover, the preverbal elements
in OV are of a mixed bag (arguments, secondary predicates, adverbials of all kind, particles).
The postulate that these elements are each attracted by the EPP features of functional heads is
not evident to me.
2. Or by V raising, that is, head-to-head adjunction (see Zwart 1993; Den Dikken and Hoekstra
1997).
3. The presupposition of this question (no quirky subjects in German), I take to be uncontro-
versial: Zaenen, Maling, and Thrainsson (1985) concluded that German does not have quirky
subjects. Sigursson (1989, pp. 204205) re-emphasized this conclusion. Fanselow (2002) and
Bayer (2003) analyze the corresponding data in German and conrm the conclusion once more.
Only Bardal (2002) has raised objections recently, based on data sentences from the web. In
Section 3.1, I shall try to show that there is no need for revising the majority view.
4. See Section 3.1 for the discussion of apparent counterexamples.
5. The deeper reason for this dierence is the dierence between licensing argument positions in a
head-initial and a head-nal V projection, respectively, as will be shown at the end of the paper.
6. In mainland Scandinavian languages, the nite verb does not move to clause-medial functional
head positions, but the spec position is lexicalized with the subject.
7. In German, the verb must stay, and unlike Faroese, German does not provide evidence for a VP-
extemal functional subject position. The latter is a VOOV-based contrast.
8. This does not exclude the converse, namely verbs that move to C but do not stop in I . One
case is that of mainland Scandinavian Germanic languages, and thanks to a reviewer for this
information another case could be quotative inversion in English as in Blair exaggerated, says
The Times. (See Collins 1997).
9. Separable prex refers to the particle of a particle+verb combination whose particle is ob-
ligatorily stranded if the verb moves to a functional head position (see 5).
a. dass sie es ankundigten
that they it announced (they announced it)
b. Sie kundigten es an-ei
They nounced it an-ei (gloss mimics German particle stranding);
c. *Sie ankundigteni es ei
they announced it
10. For example: mit-tanzen - (lit. with-dance dance jointly)

i Er tanzti jetzt mit-ei ii *Er mit-tanzti jetzt ei
he dances now with he with-dances now
11. Hohle (1991) was the rst to note that these are nite verbs that do not front.
12. If nite verbs had to move to an intermediate functional head position, doubly prexed verbs
would become deviant too. In addition, simple particle-verb combination would be inverted by
fronting and stranding. Needless to say, there is no construction in German or Dutch in which
a particle is stranded by fronting the verb and the verb does not target the C -position, that is,
the position of the nite verb in V2 and VI clauses.
13. I do not endorse the universal clause structure hypothesis. As argued in Haider (1997a, b), I
take clause structure to be the minimal complete structure for the given array of terminals. UG
dictates the functional architecture as a function of the given inventory. If, for instance, a
language does not have any morphosyntactic agreement relations (e.g., Chinese), the core
grammar does not have to project agreement nodes.
14. Note that this cannot be checked in Dutch: sentential innitival complements are ungram-
matical in VP-internal positions. Either clause union or extraposition applies. Hence a clause-
internal innitival clause is licit only in a fronted position preceding the structural subject
position (which Dutch seems to provide, see footnote 36), and it is opaque for extraction, then,
as expected. The fact that German clauses remain transparent shows what is intended to be
shown, namely that there is no functional subject position,
i *Met wiei zou [morgen ei te mogen eten] je meer plezier doen
with whom would tomorrow to be-allowed-to eat you more
pleasure give
48 HUBERT HAIDER
15. This example I gratefully acknowledge Werner Frey. It originally was the cold comfort for
investors who had lost their money in an investment fraud. As a reviewer points out, the
corresponding case in Dutch is not acceptable.
16. An anonymous reviewer points out that the data discussed so far do not rule out the possibility
that there are functional projections in between C and the VP in German but that they are
simply skipped. The discussion of the distribution of expletives in German in Section 3.2
provides reasons for excluding this possibility.
17. A projection is complex if it contains more than one complement phrase.
18. In terms of the Minimalist Program this would read as follows: merger is directionally con-
strained.
19. Basic Branching Constraint (BBC): Projection-internal branching nodes of the (functionally or
lexically extended) projection line follows their sister node. In other words, lexical projections
and their functional extensions are structured as in (i) but not as in (ii):
i . . . . . . . . . . . .
ii : . . . . . . . . . . . .
20. If an inherently dened light verb were necessary for establishing a complex head-initial VP, a
light noun would be necessary for complex head-initial N projections, for which an analogous
shell structure is needed and empirically justied (s Haider 1992/2000). Both attempts are
unnecessary. It is a bare phrase structure eect.
21 An XP is local to a head h i it is either the immediate complement of h or the spec of the
immediate complement (that is, the spec of the subjacent shell).
22. A reviewer justly asks why there could not be another shell such that the verb from its highest
position could license all argument positions. In fact, this is the division line between VSO and
SVO languages. In SVO, projection stops when all arguments are locally projected. The spec
position of a shell is a local position, but it is not within the directional licensing domain. If all
arguments have to be projected within the licensing domain of the head, the result is a VSO
system. In VO languages, shell projection is triggered by argument discharge only; in VSO,
shells are triggered by discharge and licensing.
23. The following examples from French illustrate the Romance agreement with the predicate:
i Les soldatsi sont VPti morts ti French
the soldiers are diedM:Pl the soldiers have died:
ii Les victimesi sont VP ti mortes ti
the victims are diedF:Pl the victims have died
24. The representational version of this restriction is this: projections are right branching (see
footnote 20). Kayne (1994) proposed a linearization constraint with the same result, that is, the
exclusion of merging to the right.
25. EPP (Extended Projection Principle): sentences must have subjects, that is, there is an oblig-
atory spec-position that is lexicalized with the subject (or an expletive).
26. One reviewer wonders whether German could not have VP-internal quirky subjects, referring
to Belletti and Rizzi-type (1988) psych verbs (ditransitive ergative verbs whose highest
argument could be a VP-internal quirky subject). First, in German, these verbs are verbs with
the neutral order Acc-Nom. The accusative precedes, but it shows no subject properties. (For
more details see Haider and Rosengren 2003, p.216; see Pesetsky 1995, pp.1953 for a criticism
of the unaccusative approach to these verbs.) Second, quirky subject behavior is a structural
eect of a functional position. For example, a dative of a passivized double-object verb is not
designated as passive subject by the verb in Icelandic; it becomes subject by virtue of being the
highest argument in the VP, which gets moved to a spec position. It is the spec position that
provides the structural subject properties.
27. This claim is based on the unacceptability of 1st and 2nd person pronouns as VP-internal
nominatives.
28. Note that this is a general problem of the split-In-approach: since the respective features are
associated with separate functional heads, namely Pers and Num, the ungrammatical type of
split agreement in Icelandic is actually predicted, rather than excluded. It does not follow
from any principle that the respective specs cannot be inhabited by dierent phrases providing
independent feature values for each instance of agreement with a functional head and that the
nite verb picks up and spells out these features.
29. I shall not expand on claims as on p. 72: The dative . . . can function as an antecedent for an
unexpressed subject of an innitive, a property also conned to subjects. Control in ohne
without-to constructions is not conned to subjects: Man schickte michi wieder weg, ohne PROi
angehort warden zu sein they sent me away without having been listened-to.
30. One of the Duden examples is this: Die SuxeNom werden von einer Flexionssilbe gefolgt the
suxes are followed by an inection syllable.
31. Several of her citations contain the passive of help. In this respect, it is instructive that
German newspapers made a lot of fun of a popular TV entertainer (Verona Feldbusch) who
had uttered: Hier werden sieNom geholfen Here are you helped instead of the grammatically
correct Hier wird ihnenDat geholfen. This might indicate dialectal variation but not with respect
to oblique PRO, but only with respect to dative-nominative promotion, as in standard Dutch
(with the loss of dative).
32. The patterns of Icelandic are not fully parallel with Faroese. Note that the Icelandic coun-
terpart of (35c) is ungrammatical. What is important for the present argumentation is that
there is unambiguous evidence for a functional subject position in these language in one form
or another.
33. One reviewer points out that the variant of (33a) without the expletive need not be regarded as
fully deviant.
34. Dropping in Icelandic refers to argument structure, not to surface structure, since the
remaining accusative is promoted to a quirky subject.
35. In German, various elements may be fronted by scrambling. Remnant V projections are not
scrambled, so the eects of stylistic fronting could be detected easily.
36. It is not a sucient condition because of Dutch, as the following well-known contrast illus-
trates. An expletive er is required in subjectless constructions without a preceding (locative)
adverbial.
i In deze hoek wird (er) volgens mij gefluisterd (Dutch) (Paardekoper

1963; p: 55 in this corner is there according-to me whispered
ii Werd * (er) gefluisterd (in deze hoek) (Dutch) (Paardekoper 1963, p. 55)
If the presence of the expletive is an indicator of a structural subject position in Dutch, then the
contrast between German and Dutch must be the result of a third factor. The long-standing
observation of den Besten (1985) on dative inversion points to a factor that is not related to
subject-hood (see Neeleman and Weerman 1999, pp. 210213).
37. The gloss mimics the stranding of the verbal prex in German.
38. But see also Collins and Thrainsson (1996, pp. 415f.) and the comments below.
39. Note that syntactic recoverability means structural or morphosyntactic identication and
crucially not identication by indirect inferencing based on some eects that scrambling may
bring about, for instance for the information structure. For a more detailed discussion of the
role of information structure in scrambling, especially with reference to Dutch, see Haider and
Rosengren (2003, pp. 238240).
40. A reviewer suggests that A much more straightforward take on why (44b) is out in Dutch is
that Dutch does not allow null-headed dative PPs in situ, to the right of the direct object.
True, but irrelevent, since (44b) would be an in situ dative to the left of the direct object, with
the direct object scrambled across the dative as in German.
50 HUBERT HAIDER
41. Note that this way of looking at the problem presupposes that a grammar is a cognitive
algorithm for mapping linear sequences on hierarchical structures. It does not project ante-
cedent-gap relations for sequences that are possible base orders.
42. Not for verbs with V-Acc-Dat or V-Dat-Dat order.
43. Neeleman and Weerman (1999) proposed a slightly dierent implementation of this approach.
44. Collins and Thrainsson (1996, p.436) star the English examples with the adjacent particle.
However, a V-adjacent particle in a double object constructions is not ruled out for all speakers
as Emonds (1976) and Oehrle (1976) note. Stranding a particle (rather than full pied piping)
seems to be the preferred option.
45. It is presupposed here that the structures provided by the core grammar (as a function of UG)
are parser friendly, given that UG and parsing developed in a process of cognitive co-evolution
(see Haider 2003). The data-to-parser t is optimal if the parser a left corner parser can
simultaneously operate bottom up and top down, that is, with continuous data processing
(bottom up) plus grammar guidance (top down information on possible structures). This is best
implemented with right-branching projection structures.
46. Lexical extension of a VP refers to verbal elements that contribute to the aspectual, modal,
epistemic, or other semantic modications of a VP by selecting the respective VP as comple-
ment. In German and Dutch, these verbs are obligatorily clustering. In addition, a large
subclass of control verb is clustering optionally.
47. Non-verbal material means adverbials, arguments, secondary predicates, extraposed phrases,
etc. Only (stranded) particles (which I consider to be parts of a complex verbal element) are
allowed in the cluster. For the argument to be presented here, the essential property is this: can
the sequence of verbs be split by other elements? In VO languages, this is the cardinal evidence
for separate V projections.
48. Let me emphasize that the no-intervener property is clear counterevidence for any analysis of
cluster constructions that operates with V projections (or higher projections) as minimal
building units of clusters. In Koopman and Szabolcsis (2000) as well as in Wurmbrands (2001)
work, this crucial empirical aspect is not adequately honored. In Koopman and Szabolcsi
(2000) lters are introduced; in Wurmbrand (2001) the fact is neglected.
49. Topicalization of the cluster is chosen in order to make sure that we are indeed dealing with the
clustering variant of the two possible constructions.
50. In the verbal cluster, the subject argument of the innitival verb is not projected. It is identied
with the controller-argument of the selecting verb in the amalgamated argument structures of
the clustering verbs (see Haider 1994, 2003 and the brief exposition to follow below).
51. The acceptability improves with the pronominal replaced by a reexive, but the result is still
marginal, for reasons unclear to me.
52. The brackets indicate that in the expanded innitival form (i.e. zu+V); the would-be nominative is
syntactically inactivated. This can be seen directly in the participial construction (rst analyzed in
Haider 1984):
a. ein [das Problem analysierender] Syntaktiker
a this problem analyze+AGR syntactician
a syntactician analyzing this problem
b. ein [zu analysierender] Syntaktiker
a to analyze+AGR syntactician
a syntactician to be analyzed
53. Note that the indices of the verbs just refer to the relative order in the input, not to the
dependency relations. V1 is the rst verb in the input. In VO, this is the highest one, in OV this
is the lowest one.
54. Sub-optimal is meant as a meta-linguistic qualication: as Dutch shows, the cluster too could
be fully BBC compatible. Note that there are phenomena in German (the so-called Ersatz-
innitiv or IPP constructions) with inverted orders like Dutch. But the fact that German
clusters (may) keep the order corresponding to the directional licensing relation for comple-
ments shows at least that the BBC does not have a full grip on clusters.
55. Agr-S is used here as the label for the functional head that directionally licenses the highest
argument position in a head-initial VP. Licensing may involve agreement (in languages like
English) or not (as in Icelandic).
56. Yiddish does not contradict this claim (see Haider and Rosengren 1998,2003) since it is an OV
language with VP-internal V fronting (see Section 2.3, above).
57. It is puzzling because it apparently violates Burzios generalization (Reuland 2000). (63d)
shows that the accusative of the object of this verb behaves as expected. It is not puzzling if the
accusative in (63c) is both object argument and structural subject.
References
Bardal, Johanna: 2002, Oblique Subjects in Icelandic and German, Working Papers in Scandi-
navian Syntax 70, 6199.
Bayer, Josef: 2003, Non-nominative Subjects in Comparison, ms., University of Konstanz.
Belletti, Adriana and Luigi Rizzi: 1988, Psych-verbs and Theta Theory, Natural Language and
Liguistic Theory 6, 291352.
Broekhuis, Hans: 1992, Chain Government, Ph.D. dissertation, Universiteit van Amsterdam.
Chomsky, Noam: 1981, Lectures on government and Binding, Dordrecht, Foris.
Chomsky, Noam: 1995, Bare Phrase Structure, in G. Webelhuth (ed.), Government and Binding
Theory and the Minimalist Program, Blackwell, Oxford, pp. 383439.
Collins, Chris: 1997, Local Economy, MIT Press, Cambridge.
Collins, Chris and Hoskuldur Thrainsson: 1996, VP-internal Structure and Object Shift in Ice-
landic, Linguistic Inquiry 27, 391444.
Den Besten, Hans: 1985, Some Remarks on the Ergative Hypothesis, in W. Abraham (edition),
Erklarende Syntax des Deutschen, Narr, Tubingen, pp. 5374.
Den Dikken, Marcel: 1992, Particles, ph.D. Dissertation, University of Leiden. [Published in up-
dated form: 1995, Particles: On the Syntax of Verb-particle, Triadic, and Causative constructions,
Oxford University Press, Oxford].
Den Dikken, Marcel and Eric Hoekstra: 1997, Parasitic Participles, Linguistics 35(6), 10571089.
Diesing, Molly: 1997, Yiddish VP order and the Typology of Object Movement in germanic,
Natural Language and Linguistic Theory 15, 369427.
Drosdowski, Gunther (ed.): 1984, Duden - Grammatik der deubchen Gegenwartssprache (4th
edition.), Bibliographisches Institut, Mannheim.
Emonds, Joseph E.: 1976, A Transformational Approach to English Syntax, Academic Press, New
York.
Fanselow, Gisbert: 2001, Features, -roles, and Free Constituent Order, Linguistic Inquiry 32,
405437.
Fanselow, Gisbert: 2002, Quirky Subjects and Other Speciers, in I. Kaufmann and B. Stiebels
(eds.), More than Words, Akademie-Verlag (Studia Grammatica #53), Berlin, pp. 227250.
Geerts, Guido, Walter Haeseryn, Jaap de Rooij and Maarten C. van den Toom: 1984, Algemene
Nederlandse Spraakkunst, Wolters-Noordho, Groningen.
Haider, Hubert: 1984, Was zu haben ist und was zu sein hat Bemerkungen zum Innitiv, Papiere
zur Linguistik 30, 2336.
Haider, Hubert: 1992/2000 Branching and Discharge, Arbeitspapiere des SFB 340, #23, Univ.
Stuttgart; 2000, in P. Coopmans, M. Everaert and J.Grimshaw (eds.), Lexical Specication and
Insertion, Benjamins, Amsterdam, pp. 135164.
Haider, Hubert: 1993, Deutsche Syntax, generativ, Narr, Tubingen.
Haider, Hubert: 1994, Fakultativ koharente Innitkonstruktionen im Deutschen, in A. Steube
and G. Zybatow (eds.), Zur Satzwertigkeit von Innitiven und Small Clauses, Niemeyer,
Tubingen, pp. 75106.
52 HUBERT HAIDER
Haider, Hubert: 1997a, Projective Economy on the Minimal Functional Structure of the German
Clause, in W. Abraham and E. van Gelderen (eds.), German: Syntactic Problems Problematic
Syntax, Niemeyer, Tubingen. pp. 83103.
Haider, Hubert: 1997b. Economy in Syntax is Projective Economy, in C. Wilder. H.-M. Gaertner,
and M. Bierwisch (eds.), The Role of Economy Principles in Linguistic Theory, Akademie Verlag
(Studia Grammatica 40), Berlin, pp. 205226.
Haider, Hubert: 1997c, Precedence among Predicates, Journal of Comparative Germanic Lin-
guistics 1: 341.
Haider, Hubert: 2000, OV is More Basic than VO, in P. Svenonius (ed.), The Derivation of VO and
OV, Benjamins, Amsterdam, pp. 4567.
Haider, Hubert: 2001, How to Stay Accusative in Icelandic and Faroese, Working Papers in
Scandina vian Syntax 68, 114.
Haider, Hubert: 2003, Verb Clustering and Clause Union, in P. Seuren and G. Kempen (eds.),
Verb Constructions in German and Dutch, Benjamins, Amsterdam, pp. 91126.
Haider, Hubert and Inger Rosengren: 1998, Scrambling, Sprache und Pragmatik 49, University of
Lund.
Haider, Hubert and Inger Rosengren: 2003, Scrambling Non-triggered Chain Formation in OV-
Languages, Journal of Germanic Linguistics 15, 203267.
Hohle, Tilman: 1978, Lexikalische Syntax: Die Aktiv-Passiv-Relation und andere Innitkonstruk-
tionen im Deutschen, Niemeyer, Tubingen.
Hohle, Tilman: 1991, Projektionsstufen bei V-Projektionen, ms., University of Tubingen.
Holmberg, Anders: 2000, Scandinavian Stylistic Fronting: How any Category can Become an
Expletive, Linguistic Inquiry 31, 445483.
Holmberg, Anders and Christen Platzack: 1995, The Role of Inection in Scandinavian Syntax,
Oxford University Press, Oxford.
Hroarsdottir, Thorbjorg: 2000, Word Order Change in Icelandic From OV to VO, Benjamins,
Amsterdam.
Jonsson, Johannes Gsli: 2001, Not so Quirky: On Subject Case in Icelandic, ms., University of
Reykjavik.
Kayne, Richard: 1994, The Antisymmetry of Syntax, MIT Press, Cambridge.
Koopman, Hilda: 1995, On Verbs that Fail to Undergo V-second, Linguistic Inquiry 26, 137163.
Koopman, Hilda and Anna Szabolcsi: 2000, Verbal Complexes, MIT Press, Cambridge.
Larson, Richard: 1988, On the Double Object Construction, Linguistic Inquiry 19, 335391.
Mahajan, Anoop: 1997, Against a Rightward Movement Analysis of Extraposition and Rightward
Scrambling in Hindi, in S. Tonoike (ed.), Scrambling, Linguistics Workshop Series #5, Kurosio
Publishers, Tokyo pp. 93124.
Neeleman, Ad and Fred Weerman: 1999, Flexible Syntax, Kluwer, Dordrecht.
Oehrle, Richard: 1976, The Grammatical Status of the English Dative Alternation, Ph.D. disserta-
tion, MIT.
Paardekooper, Pieter C.: 1963, Beknopte ABN-Syntaxis, 7th edition, Uitgave in eigen beheer,
Eindhoven.
Paul, Herman: 1919, Deutsche Grammatik (vol. III), Verlag von Max Niemeyer, Halle a.S.
Pesetsky, David: 1995, Zero Syntax: Experiencers and Cascades, MIT Press, Cambridge.
Pintzuk, Susan: 1991, Phrase Structures in Competition: Variation and Change in Old English Word
Order, Ph.D. dissertation, University of Pennsylvania, Philadelphia.
Quirk, Randolph, Sidney Greenbaum, Georey Leech, and Jan Svartvik: 1985, A Comprehensive
Grammar of the English Language, Longman, London.
Reuland, Eric: 2000, Explaining Burzios generalization, in E. Reuland (ed.), Arguments and Case,
Benjamins, Amsterdam, pp. 19.
Rognvaldsson, Eirkur: 1996, Word Order Variation in the VP in Old English, Working Papers in
Scandinavian Syntax 58, 5586.
Sadock, Jerrold: 1998, A Vestige of Verb Final Syntax in Yiddish, Monatshefte fur deutschsp-
rachige Literatur und Kultur (University of Wisconsin at Madison) 90, 220226.
Sigursson, Halldor Armann: 1989, Verbal Syntax and Case in Icelandic, Ph.D. dissertation,
University of Lund. [Reprinted 1992: Reykjavik, Institute of Linguistics, University of Iceland].
Sigursson, Halldor Armann: 2002a, Non-nominative Subjects in Icelandic, ms., University of
Lund [to appear in P. Bhaskaro and K.V. Subbaro (eds.), Non-nominative Subjects, Benjamins
Amsterdam].
Sigursson, Halldor Armann: 2002b, Agree and Agreement: Evidence from Germanic, Working
Papers in Scandinavian Syntax 70, 101156.
Svenonius, Peter: 2002, Icelandic Case and the Structure of Events, Journal of Comparative
Germanic Linguistics 5, 197225.
Travis, Lisa: 1991, Parameters of Phrase Structure and Verb Second Phenomena, in R. Freidin
(ed.), Principles and Parameters in Comparative Grammar, MIT Press, Cambridge pp. 339364.
Vikner, Sten: 1994 Scandinavian Object Shift and West Germanic Scrambling, in N. Corver and
H. van Riemsdijk (eds.), Studies on Scrambling, Mouton de Gruyter, Berlin, pp. 487517.
Vikner, Sten: 1995, Verb Movement and Expletive Subjects in the Germanic Languages, Oxford
University Press, Oxford.
Vikner, Sten: 2001, Verb Movement Variation in Germanic and Optimality Theory, unpublished
Habilitation Thesis, University of Tubingen.
Wurmbrand, Susanne: 2001, Innitives Restructuring and Clause Structure, Mouton de Gruyter,
Berlin.
Zaenen, Annie, Joan Maling and Hoskuldur Thrainsson: 1985, Case and Grammatical Functions:
The Icelandic Passive, Natural Language and Linguistic Theory 3, 441483.
Zwart, Jan-Wouter: 1993, Dutch Syntax: A Minimalist Approach, Ph.D. dissertation, Groningen.
Zwart, Jan-Wouter: 1997, Morphosyntax of Verb Movement: A Minimalist Approach to Syntax,
Kluwer, Dordrecht.

Icelandic and German

Uploaded by

Copyright:

Available Formats

Icelandic and German

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Icelandic and German

Uploaded by

Copyright:

Available Formats

Journal of Comparative Germanic Linguistics 8: 153, 2005.

2005 Kluwer Academic Publishers. Printed in the Netherlands.

How to turn German into Icelandic and derive the OVVO

A comparison of Icelandic and German is a most advantageous testing

quirky subjects vs. object-initial base orders

2. Basic issues of the clausal architecture in an OVVO perspective

Current theoretical approaches in the Generative Grammar family tend

hef ekki ekkt essa konu

2 a. Ich habe diese Frau nicht gekannt

In this view, the OV properties of German are derived as superimposed

tant of the functional subject position, the non-nominative phrase ac-

3 a. a mer lkar essi hakarl Icelandic

In German, the PRO subject of an innitival clause always corresponds

4 a. da Kindern oft Marchen erzahlt werden

The corresponding passive construction in Icelandic would be one in

on the properties of innitival clauses. It must be given on a general

2.1. Evidence against overt V-to-I in German

subjects nor objects move to clause-internal spec positions (see Section

that a clause-nal nite verb has moved to a functional head position.

5 a: wenn du uns voranmeldest

6 a. ab-drucken - vor-ab-drucken lit. pre-print

A verb like urauuhren is a back formation from the prexed deverbal

What the examples in (8) conrm is that sentential innitival comple-

clause-nal nite functional head position nor to a clause-nal innitival

alternative is a subject-in situ analysis. The clause remains in its VP-

10 a. Wasi hat (*es) [ihr ei zu verraten] keiner riskiert?

So we have to dismiss the premise that the overt subject position in

At this point of the discussion it is worthwhile commenting on the

11 a. Es hat den/*der Hund erschreckt

The Travis-Zwart account for (11a) in brief is this: if a clause with a

when fronted? The appropriate descriptive generalization is one in terms

2.2. The OVVO parameterization directionality and asymmetry

VO and OV are but abbreviations for head-initial and head-nal

complements directionally, and (iii) projections are universally right

(12) a. head-initial VP b. head-nal VP

Evidently, in (12b), any VP-internal phrase that is immediately domi-

unsaturated remnant of the argument structure. The projection is

2.3. Hybrid systems VO and OV, or neither?

The existence of languages that exhibit both VO and OV surface orders

supercially looks like scrambling to the right in an OV language as in

directionality allows either of two values. The choice of the value is in

Null objects under coordination

19 a. Di yidene hot aroysgenumen eyn gandz . . . Yiddish

Verb particles (separable and non-separable ones)

20 a. Wann kommt der Zug an? German

21 a. een groene bus / twee groene bussen Dutch

22 a. a griner oytobus / tsvey grine oytobusn Yiddish

In terms of word order patterns for sequences of non-nite verbs, the

On the basis of this extensive, robust, and diverse enough evidence, in

3. Some Icelandic-German contrasts and their OV/VO based

Having prepared the background, let us now analyze in more detail

3.1. Quirky subjects

27 a. a henni=stelpunum lkuu hestarnir:

nominative obligatorily agrees in number (27a), and only in number.

29 a. Beim Quadrieren mu man eine Zahl mit sich multiplizieren

30 a. Wir stellten den Leutenj sichj = einanderj vor

Coordination reduction is not a reliable subject diagnostic except for

positionally matching constituents in the second conjunct (independent

3.2. Subjectless clauses

For a VO language such as Icelandic or Faroese (Vikner 1995, p. 227),

49 a. I gr hafa peir sent upp strakunum peningana

50 a. I gr hafa eir sent strakunum peningana beint upp