Remus Gergel

Modal Syntax: Detecting its Parameters with VP-Ellipsis

Abstract

Recent syntactic approaches unconventionally transfer semantic modal distinctions to

syntactic structure or syntax-relevant observations with the evidence usually
concentrating on scope (relative to negation, quantifiers, etc). This paper discusses why
such interesting preliminary indications are less than fully satisfactory. Moreover, it
argues for at least a local co-encoding of modal parametels in English syntax with
corroborating support based on licensing ofVP-ellipsis. Evidence is presented in which
dichotomies befween (i) differently grammaticalized modals, (ii) epistemic/deontic
modals, and (iii) universaVexistential modal force, respectively, interfere with ellipsis in
diagnostically telling ways.-

1. Introduction and structure ofthe article

Part of ongoing 'rvork, this syntactic paper presupposes the essentials of

modal mechanics developed in the semantic tradition of Lewis, Kratzer,
and others (cf., e.g., Kratzer 1981). These are taken as the null-hypothesis
mechanism regulating modal interpretation at LF. Any additional,
particularly pre-spellout involvement of modality is a departure from the
null-hypothesis and needs syntactic evidence (Reis 2001, Vy'urmbrand
2001). This is where we start in conceptual terms, heading for English.

*
For discussions at various stages, I am grateful to C.Baatz, B. Drubig, J. Hartmann' A.
Kroch, L. López,W. Stemefeld, S. Winkler, the audiences in Tübingen and Tromsø,
where parts of this material were presented, and to two reviewers. I wish I could have
more ñrlly explored all their suggestions. special thanks go to the organizers and the
audience of Approaches to Historical syntax 2002 (Linguistic Association ofFinland),
particularly to J. Rostila and A. Warner. The research for this study has been supported
by the Deutsche Forschungsgemeinschaft through SFB 4411813, which is hereby
acknowledged. The usual disclaimers apply.

SKY Journal of Línguistics I 6 (2003), 27-56

28 REMUS GERGEL

Section 2 introduces a classical held among the attempts to investigate

discriminatory syntactic effects in the domain of modality: the outlandish
status of epistemic modals. By scrutinizing recent approaches, we will see
that there is interesting motivation to consider encoding epistemicity as a
syntactic factor amongst the computations operative in natural language,
but we will retum to the issue later. In section 3, a different type of
argument building on diachronic data is introduced. It illushates distinct
(nonJexical) modal positions relevant to the reanalysis of the English
Modals. The link between this section and the others consists in the method
used: licensing properties for VP-ellipsis through functional material
instantiated by modals (cf. Lobeck 1995).
In sections 4 and 5, we refurn to the synchronic concerns, homing in
to the issues of division of labor, particularly between epistemics vs.
deontics on the one hand, and between strong vs. weak (or universal vs.
existential) on the other.' We discuss the relevant aspects of two specific
approaches attempting to come to terms with such properties: one as a rigid
cartographic arrangement of functional heads, the other as a condition on
quantifiers for epistemic modals. I use the quantificational approach as a
term of comparison, and because I think it is a potentially interesting
alternative venue directly pertinent to the well-trodden approaches on
modal syntax of the last decade, but I dwell more on cartography arguing
that pending further evidence it is still the tool with the wider data
coverage. At least up to date it seems to be an ampler machinery, capable
of handling more dimensions of modality. Nonetheless it should be taken
with a grain of salt. This may entail relative c-command relations and
possibly language-particular telescopic structures, rather than wholesale
universal representational cartography, as suggested in the final sections.
Section 6 in particular includes strong additional evidence from VP-ellipsis
both for the epistemic/deontic, and the necessity/possibility distinctions as
being co-encoded in the syntax.

I The field of English studies often uses the strong/weak terminology (cf., e.g.,
Huddleston & Pullum 2002), whereas the formal semantics school in the Lewis-K¡atzer
tradition the universaVexistential one. Nothing relevant should hinge on notation for the
purposes ofthis paper.
 MODAL SYNTAX: DETECTING ITS PARAMETERS WITH VP-ELLIPSIS 29

2. Epistemics go farther

To see epistemicify provoking havoc consider (l), for which williams

(lgg4:2i) notes that it "does not mean 'It is not probable that John left'
even though this is a sensible thing to say."

(1) John didn't ProbablY leave.

(* negation > epistemic marker, where ">" denotes "takes scope over")

'what we see in (1) then is an epistemic adverb obligatorily taking scope

over negation, though the relationship could be otherwise-both
semantically, as Williams' paraphrase shows, and syntactically, as seen
from the reverse overt linearization.
According to Cinque (1999,2001), adverbs and functional heads stand
in a tight specifier-head relationship. If the recent claims hinting at a
tendency foi functional epistemic elements to take wider scope in the
clause over other scope bearing elements afe correct, as for instance
illustrated with negation above, then we would expect it to be identifiable
with the modals (the heads) in particular. There is indication that this is the
case in view ofdata such as (2).

(2) Maryann might not sleep well these days'

(*neg > epi)

This cannot be interpreted with the negation scoping over the epistemic
modal.
Further elements that lend themselves to scope comparisons with
epistemics are tense and deontic modals, prima facie both not too practical
choices due to the morphosyntax of standard English registers, which
precludes the modals from directly combining with tenses (though not with
ãspect, cf. Jack might be driving to his ffice) and with each other-the
laiter restriction being not cross-dialectal. Drubig (2001), however,
observes an instance where tense-modal interaction can be directly
wiûressed in English. The diagrostic originates in Hoffinann's (1976) rule
of past tense replacement. This states that, in a set of certain delimited
contexts, a past tense morpheme (-ed) is tealized as have. Through this
replacement, the past tense morpheme can override the dilemma of getting
30 REMUS GERGEL

bound to an inhospitable host such as the modal (cf. *musted, or the

absence of maytei' which cannot generally be equated with might, as is
well known). This phenomenon is clearly distinct from the (unambiguous)
perfect interpretation, which would be ungrammatical with temporal
adverbs such as yesterday,or last year in nonexistential perfect
constructions; cf. by contrast the grammatical sentence (3).

(3) Sue may have bought the book yesterday/last year

(epi > past)

Drubig works out a concept of epistemicity aligning its particular syntactic

form to that of more fully developed evidential systems. Under this
approach the scope relation 'evidentiality ) tense' is explicitly predicted by
the syntactic form. But what about examples like (a) and (5)? Are they not
contrary evidence, viz. with tense scoping out?

(4) At that point, he could/might still have won the game.

(past > alethic possibility)

(5) In October, Gore should still have won the election.

(past > alethic necessitf)
(Stowell, forth: (20a, b))

On closer semantic and pragmatic inspection, it turns out that they are not.
Assuming a somewhat similarly restrictive, i.e. mostly an evidentiality-
based concept of epistemicity, as the one explicitly argued for by Drubig,
Condoravdi (2001) and Stowell (forth.) show that counterexamples which
allow have to scope over the modal are not bona fide epistemics (though
traditionally classified as such), but rather metaphysicals or alethics.3

2 Huddleston & Pullum (2002) classify should as medium modality given its
peculiarities. In this paper I will not deal with the specifics of this particular type of
modality. Cf. also de Harm (1997).
i An interesting question sometimes explored in the literature is to what extent scope
reversal relates to the morphology ofthe modal itself. For discussion on the particular
case ofEnglish see Condoravdi (2001) and Stowell (forth.), who suggest semantic and
syntactic explanations, respectively, why scope reversal, i.e., counterfactual alethics,
should be only possible for non-present modals such as might, could, should.
 MoDAL SYNTAX: DETECTING ITS PAR.AMETERS WITH VP-ELLIPSIS 3l

3. VP-ellipsis in Middte English

This section starts the discussion of the interaction between modals and
ellipsis in English. we will particularly see that the diachronic treatment of
the English modals makes (a minimum of) two distinct positions necessary,
specifically one preceding and one following their reanalysis. The
discussion is less refined than in the other sections, mainly due to the fact
that one seems to be better advised to discuss various readings and their
syntactic correlations in synchronic terms with an accessible language.
Nonetheless, the diachronic data prove the point of looking at modals and
ellipsis in conjunction. We will see that the modals were picked up by the
process ofdiachronic reanalysis from a projection different from V (cf. the
"standard" reanalysis theory developed in the wake oflightfoot 1979)'
We tum our attention to the most relevant period for the reanalysis,
i.e. the time span ranging over the later periods of Middle English (ME) to
early Modem English (ModE). It will be argued that an appropriate
approximation equates the post-reanalysis position of the modals indeed
with T, but the pre-reanalysis one with a lower projection, quasi-functional
in nature, but crucially still able to license VP-ellipsis.
The fact that the reanalysis stranded the emerging ModE modals
roughly in T is generally accepted and convincingly demonstrated by
Roberts (1985, 1993). However, based on the quantitative and theoretical
analysis in Gergel (2002), drawing on the second version of the Pmn
Helsinki Parsed Corpus of Middle Englßh (PPCME2), I argue that the
existence of ellipsis makes the standard account untenable with respect to
where the reanaþsis picked the modals up.o While the standard view of the
reanalysis assumes that the premodals were fully lexical verbs up into the
sixteenth century, when they allegedly have cataclysmically reanalyzed, it

Denison (1992) observes a perhaps less well-knowntse of may in counterfactuals,

where on strict morphological mapping one would expect might, as illustrated by (i).
(i) Swift launch may have saved Penlee lives. (Gdn, 15 Mar 1983' cited from
Denison 1992)
Context: all lives have been lost, the possibility doesn't exist any longer.
a
see especially wamer (1992) lor complementary discussion on ellipsis forms possibly
extant in old English. Moreover, besides ellipsis there are also further problems with
what is here referred to as the standard accounts ofthe reanalysis, oflesser relevance for
present pwposes; cf. Harris & Campbell (1995) for discussion.
)z RIMUS GERGEL

will be argued here that they were quasi-functional, in the configurational

sense defined below, by that time already. More specifically, instead of
reanalyzing from V to T they only underwent the shorter distance (in the
syntactic metric) from the quasi-functional projection to T. Note that while
functional material going by different names, say, in the Infl domain
(AUX, I, T, etc.) may yield similar results, e.g., in the NICE contexts, or
similarly in the so called low IP domain (where, e.9., v, Tr, Pr, Asp have
been recently operated with), we can also phrase the crucial difference at
hand in more general conceptual terms. The reanalysis ended with
functional modals (this is where we agree with the standard accounts) but it
started quasi-functional and not lexical (this being the part where we
disagree with it).
While there is little doubt that volitional and perhaps further adjunct
theta-assignments-in the sense of htbizarreta (1982f-co-existed with the
late Middle English modals, (as is still the case today, e.g., notably with
negated wil[), there is also strong evidence that in many cases they elided
their VP complements. Interestingly this type of elision is not random but
falls in place with all the major diagnostics of VP ellipsis, which in tum are
a typical indicator of functional status. Consider (6f(1 1) for some of these
feafures.

(6) But for he couþe not selle and undo his clooþ as a womman
but since he could not sell and undo his clothes as â woman
schulde _, he was...
should he was ...
(POLYCH,VIII,l 05.367 7s)

(7) Bie war se ðe wile

be cautious who that (whoever)will
(vrcEsl,l39.r725)

(8) telle who þat wil _ what it myghte mene.

tell who that will what it might mean
(POLYCH,VrIr,89.3584)

s
The notation follows the PPCME2 standa¡ds.
 Moo¡l Svnrax: DETECTING lrs PARAMETERS wlrH VP-ELLIPSIS 33

(9) 'My doghty , why wol þou not schew me þyhert ?

my daughter why will you not show me your heart
(MIRK,90.2421)

(10) ' Lord , 'quod scho , ' I maY not

- for schame . '
lord said she I maY not for shame
(MIß.K,90.2422)

(ll) þei sent hir fro þe lryngß coferes what þei wold-
they sent her from the king's coffers what they would./ wanted to
(CAPCHR,150.35l6)

What these examples immediately show is that we find linguistic

antecedents for the ellipsis sites, and that ellipsis can range over phfasal
categories and need not be flanked, as opposed to gapping. Furthermore
therè are cases of sloppy identity between antecedent-related referents and
the elided ones (6), evasion of the complex NP-constraint (7), the
relationship between antecedent and ellipsis can function across
embeddings (8), and also across utterances as the sentence-pair in (9þ(10)
illustrates. There are cases of w¿-extraction from the elliptical sites, like in
(11).6 The existence of wå-movement is a sfrong discriminating indicator of
VP-ellipsis, as opposed to null-complement anaphora due to the fact that
null-complement anaphora verbs do not allow it, as the contrast to the null-
complement verb in (12) shows (cf. Depiante 2000: 35f. for discussion).

(12) * Mary wondered which conference talk Tommy refused to attend and Susan
wonders which colloquium talk Mary refused
-.
Furthermore, from the corpus-theoretic perspective, it is noteworthy that
virtually all elliptical examples are parsed as M in the PPCME2-and not
as V, which is an altemative annotational variant used in the corpus, e.g,
especially for earlier instances of modals. In terms of numerical
developments, after a collapse in the early Middle English period, the
relativè frequency of verbal elliptical structures in relation to the total
number of tokens increases with approximately 50o/o per cor?us-segment, a

6 cf. Gergel (2002) for further empirical evidence, discussion of some pitfalls, and
additional corpus-theoretic and numerical argumentation.
34 REMUS GERGEL

trend that may be taken as significant for an internal gmmmar change in the
sense of Kroch (1989, 2002).
Before investigating how the reanalysis might have worked on the
syntactic component diachronically, wê need to sketch some facts
underlying VP ellipsis in slmchronic terms. With Harley (1995), Emonds
(2000), among others (and contra, e.g., the classical analysis of Emonds,
1970),1assume that ModE verbs like have and be are merged directly in a
projection structurally higher than V, which predicts the properties setting
them apart from main verbs today. For the purposes of ellipsis licensing
this has an interesting consequence. Consider (l3F(16), (cf. Emonds
2000).

(13) * Since in the past you have represented us well, I'm sure you must _ yesterday.
(14) Since in the past you have represented us well, I'm sure you must have _
yesterday.
(15) * I makeMary be examined often even though her brother refuses to _.
(16) I make Mary be examined often even though her brother refuses to be_.

The ungrammaticality of examples like (13) and (15) suggests that the
verbs have and be cannot be generally deleted under a T-element such as
musÍ oÍ to-by contrast, it is well-known that under the appropriate
parallelism requirements in terms of feature subsets (e.g. Oku 2001) VPs
headed by genuine main verbs can. An explanation of these facts is that
deletion of the VP depends on a set of two licensers in order for it to safely
occur. Consider (17).

Licensers VPE
111
(17) site

o
lrp Subjr To [¡p l¡ F

t Giu"n that at the latest on minimalist assumptions categories are but feature bundles,
the issue of what F is does not affect the positional argument of this section. We have a
distinct projecting head accommodating auxiliaries. Regarding the set of solutions for F
one has various candidates following recent options ofinvestigation in this domain. Fox
& Lasnik (2003: 151) operate with aspect. Thus if we replace F in (17) with Asp we
essentially get the Fox & Lasnik proposal for licensing ofVP-ellipsis. Though I haven't
been able to fully evaluate this recent approach, as Fox and Lasnik þ. 151) seem to
concede, in their framework not much hinges on the label of the projection below T -
 Mooel Svur¡x: DETECTING ITS PARAMETERS wlrH VP-ELLIPSIS 35

Thus in addition to the well-known licensing function of T, we have a co-

licensing element F. This hypothesis in fact neatly captures some features
of both ME and ModE modals and their respective behavior in elliptical
contexts. The modals in synchronic ModE terms are merged in T given that
they do not check tense (cf. Roberts 1985, 1993 for additional justihcation
of this finding). The auxiliaries havelbe do check tense, but they are also
above the deletion line as aforementioned'
Now, for ME the situation is different. Auxiliary deletion was
available in cases in which it would be illicit today (Warner 1995). That is
havelbe were below the deletion line (as also indicated by corpus findings).
The modals in ME were able to overtly check tense and they were right at
the critical licensing line of VP deletion, but on the licensing, not the
licensee side. A possible generalization for ME then from these facts is the
following: Fill T with the modals through merge (in complementary
distribution with the tense morpheme -ed), and F with theteanalyzed verbs
høvelbe.8

perhaps due to the different focus oftheirpaper. Gergel (2002) argues for a syntactic Pr
head in the position occupied by F, partly following Bowers (2001) and additional
diachronic argumentation.
8
The relative positions ofthe modals on the one hand, and belhave on the other in pre-
and post-reanalysis grammars are surprising at first sight. There is, however, rather
strong evidence for this rough modals/"aspectuals" dichotomy for the data at hand.
First, note that there is indication that roughly 99%o of the verbal elliptical structures in
ME are licensed by modals solely - notby havelbe, not by verbs (cf. Gergel 2002' vtith
the estimate based on studies on the PPCME2). Second, the peculiarities of have, and
especially of åe with respect to vP-ellipsis are late changes, as is well-known. This fact
was noted and seriously discussed by Anthony TVamer, and explored in various
diachronic and synchronic studies; cf. Warner (1995) among others. For time and space
reasons, the changes axe not discussed here, but the relevant approximating
generalization is clear: Preceding these late changes, the "aspectuals" could establish
antecedent-ellipsis dependencies following the same or similar retrieval rules as used in
the parallelism-requirements for main verbs, whereas today they no longer can under
the same conditions.
Note also that I am not ruling out that aspectuals and main verbs be further
distinguished following various criteria, e.g., semantic, typology-oriented ones (e.g., in
the spirit ofHeine 1993), or even further syntactic ones. All that I am saying is that for
the course run by English the ellipsis-facts strongly suggest taking havelbe all in all
below the critical double-line of co-licensers before the reanalysis, and that they are
merged higher up later on. A third piece ofevidence can be adduced from linearization
36 Rptrtus Gnncel

Finally in this section, we look at adverb placement, which offers a

corroborating diagnostic for the structural distinction between the VPE
licensers in the PPCME2 data, viz. the modals vs. lexical verbs. Since all
finite verbs move to T in ME, in order to detect anything at all, it is useful
to use this diagnostic with the verb to be tested in situ. Consider (18) and
(le).

(18) And whan thß creatur was fþud [gracyowslyJ comen a-geyn ..
and when this creature AUX thus graciously come.PRT again...
(KEMPE,9, 139)

( 19) hemuste ofte and manytymes redein thys bokeand

he had to often and manytimes read in this book and
[ernestly and diligentlyJ marke [welJ that he redeth
eamestly and diligently mark well that (what) he reads
(REYNAR,6.7)

The intermediate position F has an effect in these cases by providing an

intermediate hosting site for a second adverb between T and in-situ V. Thus
both configurations with two adverbs, (20), and (21), are available for
lexical verbs, asjust exemplified by (18) and (19).

(20) T [Adv] [Adv] [non-fin V], where non-fin V is a non-finite in-situverb

(21) T [Adv] [non-lin v] [Adv]

For modals, the comparable configurations are not available in the entire
PPCME2 independently of adverb classes. Given that ME is the flourishing
period of non-finite modals when compared to the other stages of English,
the failure of the test can indeed not be blamed on a lack of in-situ modals.
The phenomenon can straightforwardly be explained by means of the
intermediate projection. With modals, the intermediate position cannot be

facts. Modals productively co-occur with havelbe in Modem English. However the use
of such sequences is late (cf., e.g., van Gelderen 2003: 35). One way to go would be that
having found a licit position to be first-merged (namely in F in (17), after F has been
evacuated by the modals thorough the reanalysis), have/be have now found a more
natural position to be licensed in cases where modals are present in the initial
numeration of a given sentence, i.e. in the quasi-functional F in post-reanalysis
-
grammars they are not blocked by the presence ofthe modals from that position.
 MoDAL SYNTAX: DETECTING ITS PARAMETERS WITH VP.ELLIPSIS 37

made visible due to the fact that it most probably does not exist, as shown
by the unattested pattems in (22), and (23)-the premodals occupying F
themselves.

(22) * ? T [Adv] [Adv] [non-fin M],

where non-fin M is a non-finite in-situ modal

(23) * ? T [Adv] [non-fin M] [Adv]

As one reviewer notes, complications may lurk behind adverb s)'ntax'

Following this lead, the first, fundamental question would be whether we
have traditional adjunction or more intricate mechanisms involved, e.g', to
take one option from the literature, say, with correspondences between
syntactic position and semantic content (Cinque 1999). However, from
recent research the issue seems to hardly be settled for any of the two
diametric options just mentioned, i.e., (A) syntactically random and
potentially stacked adjunction, or (B) structurally determined and rigid
insertion sites, e.g., as specifiers. Note, however, that the discriminating
findings of (20)(23), i.e. crucially that one set of configurations is attested
and the other is not, already make an unrestricted adjunction theory highly
unlikely by themselves, no circularity involved. This, of course, still does
not tell us what an appropriate theory, or perhaps theories, of adverbs
should look like in general, that also being far beyond the ellipsis-based
concems of this paper.e What the data simply corroborate instead is that an

e While the contrast (20) vs. (22) may speak for itselt the reviewer furthermore
perspicuously points out that firther complications arise from the attested type I gave in
(21) (and cross-comparisons between this type and the three others). For instance, what
is the projection where the final adverb in (21) is merged to. While a definite take on the
issue indeed has to await furthe¡ research on adverbial syntax, I think the data are less
confusing than they appear. First, adjunction is out ofthe question for the data we are
concemed with as mentioned above. Then we are essentially left with two options:
. either move the non-finite verb in configurations such as (21) over the
adverb,
, or merge the adverb to a relatively low position (perhaps in a vP-shell
architecture).
Given that there is no modal in the numeration in data reflecting (21), this being the
crucial point in the pair (20)-(21), I am not sure whether the first option should be
discarded in order to keep the "base-generating" modal position free. It is, however,
 38 R¡rr¿us Genc¡l

intermediate position may be detectable with verbs, but not with modals. If
the modals are structurally higher then verbs, as we have seen based on the
discriminating ellipsis evidence above, then this fact is explained.

4. Cartography

Having illushated licensing as a function of modals in conjunction with

ellipsis through the diachronic excursus, we now prepare the ground for
using VP-ellipsis licensing in some more intricate cases of modalized data.
Thus Cinque (1999, 2001) argues for an articulate division of the functional
structure of the clause into various projections correlating with semantic
functions relating to tense, aspect, mood, and in particular distinct kinds of
modality. One of the main claims of these and related studies is the rigid
order of the functional material. In this section and the subsequent ones we
will see evidence supporting some ordering relations. I will, however,
remain agnostic as to whether one needs a rich Cinque-style hierarchy.
What I will argue for are local relationships between heads hosting the
modals. For now, to put the discussion on a concrete footing, consider the
syntactic representatio n in (24) (Butler 2003 ). 0
I

more than anything an independent matter of empirical finding whether non-finites have
the ability to undergo at least some shorter movements in ME, and what principles this
movement should exactly obey. Cross-linguistically this option is certainly available
(see, e.g., Bok-Bennema 2002 for interesting facts regarding short V-movement in
Romance, partly foreshadowed by Pollock's 1989 system), and it is not unlikely to find
it, e.g., for reasons of participle-morphology licensing in English varieties either, as
recently assumed by Roberts & Roussou (2003:2.1). Otherwise the second option or a
variant thereof, needed for what appear to be cases of apparent low adverbs in any
event, may be involved.
'0 Butler capitalizes on the necessity-possibility distinction drawn in Cormack & Smith
(2002). Cinque also has distinctions befween necessity and possibility, although he
argues against this split in the epistemic domain. I focus on Butler's structure, rat¡er
than Cormack & Smith's for the purposes of this paper, first, because it allows a more
straightforward comparison with Cinque's well-known cartography; second because it
is worked out in a close relationship with the interface interpretation; and third, because
it has roughly the usual assumptions on the grammatical model, whereas, e.g., Cormack
and Smith's introduces additional assumptions, (for instance, on Merge) and the split-
sign hypothesis, which would need an additional layer oftheorizing.
 MoDAL SYNTAX: DETECTING ITS PARAMETERS WITH VP.ELLIPSIS 39

(24)

FocP

[neg] FinP

Iposs] TP

T'

T P

FocP

lnegl

VP

(cf. Butler 2003: 988)

Let's start with the theoretical implications. First, rather complex at first
glance, this phrase marker is on the one hand overall much less articulate
and hence perhaps more learning adequate than Cinque (1999)'s influential
cartogaphy of the functional domain. This comparison holds even if the
latter were truncated to the modality-relevant projections, for more
appropriate conÍast. At the same time, the structure is symmetrical in
interesting ways. Note that necessity c-commands possibility twice, once in
the higher domain of epistemicity, and once again in the lower one in
charge of root modality. We mainly concentrate on modality, though Butler
assumes connections between the heads hosting the modals and functional
projection in the sense of Rizzi's (1997) Split-CP hypothesis, as the
labelling in (24) shows.
Second, the reduplicative character of many of Cinque's projections
has been recently critiqued as a way-out solution designed to fit scope facts
otherwise not accommodated. Recursion might, however, be simply a
descriptive necessity, in view of the wide fypological view argued by
Cinque-style approaches. How is then the tension to be solved? In fact,
40 Rrvus Gpncel

once one assumes a structure along the lines of (24) and, crucially,
substantiates the evidence for it, the alleged weakness ofrecursion can tum
into a virtue. For instance, this structure has recursion in a// nodes
responsible for modal interpretation. Cinque's original formulation, where,
for instance, evidentiality only appears once, does not. What I consider the
main effect of this line of thought is the ability to descriptively generate the
wide gamut that Cinque aims at in the domain of modality; but at a lower
cost. Incidentally, it is also very well known that similar phenomena exist
in other functional domains as well; e.g., in the domain of focalization.rr It
is no little achievement to make rich structures fit explanatory and
acquisition-based principles. In this line of reasoning, leaming one segment
of the structure and re-applying the same part at a later age in a higher
domain of the clause would facilitate the task of the language learner
considerably. The reapplication might be motivated by meta-
representational interpretative impact as approaches such as Papafragou
(1998) would suggest. Furthermore, if the recent phase-theoretical concems
(Chomsky 2001) are on the right track, then a structure rounding off the vP
and then ending symmetrically in a reduplicated CP would also be in line
with it.
Consider now the main empirical support Butler adduces to underpin
the structure proposed, e.g., through examples like the ones rendered as
(zsYQÐ below.

(25) The regishar mustn'lmightn't have got my letter

(epi > neg)
(26) The children mustn't do that in here.
(root necessity > neg)
(27) The children can't do that in here.
(neg > root possibility)

The set ofsentences above can be used for two purposes. On the one hand,
the relative order epistemic vs. root comes to the fore when considering
subject scope relative to the modal in (25) vs. (26) and (27). While an

" From this kind of research, Butler adopts the labelling ForceP, FocP, FinP, as
aforementioned; see Rizzi (1997), and Drubig (2003) for ampler discussion of layered
effects in the domain of focalization. I leave information-structural concems aside in
this paper.
 Mooel Swr¡.x: DETECTING ITs PAR{METERS viITH VP-ELLIPSIS 4l

epistemic doesn't allow the subject to scope over it, as in (25), the two
dèontics clearly do. The scope position for subjects assumed in this context
is the (generatively) traditional Spec, TP. On the other hand, the possibility
vs. necessity divide is deducible from the comparison between (26) and
(27) and the indicated scope relations. These sentences suggest that while
possibility scopes below, necessity scopes above negation.

5. Quantilier containment: insights on epistemics from intervention

effects

In this section, we consider an altemative view of phrasing at least the

outlandish properfy of epistemic modals. Instead of phrasing the high
position of epistemic modals in cartographic, and thus phrase-structural
terms, von Fintel & Iatridou (2003), F&I herafter, formulate it as an
intervention effect by investigating the interaction between quantificational
expressions and epistemic modals. According to F&I, a quantifier carulot
take scope over an epistemic modal, an observation dubbed the epistemic
containment principle (ECP).

(28) ECP: At LF, a quantifier cannot bind its trace across an epistemic modal.
* Q¡... [Epistemic Modat (...t¡...) (cf. F&I 2003: (31))

(29) Most of our students must be home by now.

(*rrosf ...> epistemic must)

Suppose the modal in (29) is epistemic. Then it also must get scope over
the quantifier. As a counterfoil to the observation on epistemics it is useful
to observe that, by contrast, deontics and tenses do not obey it. For
instance, on a deontic reading (30) allows both scope atrangements modal
> quantifier ('it is necessary that most students get outside funding') on the
reading facilitated by continuation in a., and quantifier > modal (very
roughly 'for most students it is the case that they necessitate outside
funding') on the reading induced by b.
42 RIMUS GERGEL

(30) Most of our students must get outside funding-

a. for the department budget to work out.
(must> most ...)
b. the others have already been given university fellowships.
(most...> must)

In the same vein, the tense in (31) displays both options with respect to
relative scope with the quantifier.

(31) a. Most of our students will be professors in a few years.

(most....>will)
b. Most of our students will be foreigners in a few years.
(wíll> most...)

As the contrast between (29) on the one hand, and (30F(31) on the other
illustrates, quantification constitutes a vivid domain to observe
discriminating effects of epistemic modals. While it might be too early to
fully assess F&I's approach, there is initial evidence presented in section 6
that to the extent that it is corect, its main syntactic effect can be derived
from the syntax and semantics of epistemic modals in general.

6. Ellipses in the approaches. Approaches through ellipsis

First, note that for some of the examples above, e.g., (1) or (29), it could be
counterargued that not only do they not parallel semantic paraphrases
which would allow the modals to scope lower but, put simply, they also
override overt syntax, so that the entire matter becomes a somewhat
impalpable "scope matter," in many syntacticians' view not a satisfactory
situation. This alone would make additional evidence necessary.
Second, though appealing, Butler's s¡rmmetric approach has some
loops in the argumentation. It distinguishes strong from weak modality in
both phases. But the evidence is not as well-balanced and symmetrical as
the tree-structure proposed. Consider again the negation-based evidence,
used as a crucial argument. The bottom line is the assumption that negation
resides in a distinct head sandwiched between modals in both phases, thus
serving as the scope element relative to each necessity and possibility can
be tested, both in the epistemic and the root domain. However, the main
support adduced with respect to negation is of the type rendered above in
 MoDAL SYNTAX: DETECTING ITS PARAMETERS WITH VP-ELLIPSIS 43

(25YQ7), i.e., in which only root necessity and possibility can be told
apart. But for epistemic modalify, negation only indicates the ordering
modal > negation, not the relative scope of strong and weak modality; cf.
(32), and (33).12

(32) Jack must not have had a very good education.

(epistemic necessity > negation)

(33) John may not have finished dinner.

(epistemic possibility > negation)

As for the quantificational approach, F&I refer to a number of problematic

cases themselves (some of which they eventually attempt to accommodate),
and further ones could be added. Suppose however, for the sake of the
argument, that the quantihcational observation were accurate. While an
interesting fact, then, it would still be desirable to find an explanation for it.
However, the 'þrinciple" supposedly driving it given in (4), viz. the. ECP,
tums out to be equally descriptive-mainly a restatement of the facts.l3
There are two major venues to achieve more. One would be the
following. F&I clearly dehne the ECP as a condition on QR and would like
to see it in the general context ofresearch ofthis type. The integration into
this type of investigation does, however, not actually happen, despite the
insightful discussion. For instance, some of the essential recent work on
QR accounts for scope (im)possibilities as an economy condition on a
reference set of logical forms, i.e. as a process motivated by concrete,
ideally derivational options at each step.ra For brief illustration consider

12
Butler (2003: 984) presents evidence with respect to negation in the higher domain
þhase); however, with suppletive forms. That is, instead of capturing the syntax of
epistemic may and must, we leam interesting things about suppletive forms like can't
and needn't, which in logical terms should do. For the purposes of natural language
analysis, however, this rests on the assumption that they are the exact linguistic
equivalents of the modals nay and must in negated contexts. This would need
investigation first. If, based on empirical inquiry, it tums out that they are, then this
paper may provide strong empirical support to the suggestions based on negation..
13
The ECP as rendered above is a more accurate statement than a preliminary version
F&I offe¡ and which for lack of space I omit.
to Cf., e.g., Fox (1995), and Fox (2000) with updates-which F&I also cite and wo¡k
with for other purposes.
44 R¡røus GBncsL

(34), an ambiguous sentence with respect to the relative scope of the

quantifiers, and then (35), a variant with a continuation including VP-
ellipsis.

(34) Some boy admires every teacher.

(35) Some boy admi¡es every teacher and every girl does _ too.
(3ó) *every teacherl
[some boy admires t¡] and every teacher2 [every girl admires t2]

However, in (35), only the overt scope relation is possible, viz. no scope
reversal and thus no ambiguity is possible given that in logical terms
movement of the quantifier in the elliptical conjunct would be vacuous.
The laws of Quantifier Independence predict this lack of movement (Partee
et. al. 1993: 148). In other words, the universal quantifier from the elided
site in (35) has a right to stay in situ, because moving over another
universal would not make an interpretative difference. This indeed predicts
the ungrammaticality of the superfluous derivation in (36). The
generalization beyond this kind ofbehavior is, according to Fox, along the
lines of (37).

(37) Ellipsis Scope Generalization (ESG): The relative scope of two quantifiers, one
of which is an antecedent VP of an ellipsis construction, may differ from the
surface c-command relation, only if the parallel difference will have semantic
effects in the elided VP. (cf. Fox, 1995:149)

Nonetheless, retuming to the ECP, no similar or altemative explanatory

condition is offered and what remains is the carefully stated and well-
thought descriptive observation in (4).
A second way to gain more insight into the phenomenon is to note that
quantification is just one among various properties setting epistemics apart.
Then the cartographic approach has some of the makings to derive the
consequence of their observation, in that the epistemic modal, operating
from a topologically high position c-commanding the other scope positions
ofthe clause also takes scope, inter alia, over quantifiers, even after these
have QRed. The phrase takes scope over in the perspicuous formulation of
the ECP reflects the fact that the containment may not apply to quantifiers
directly merged in a superordinate clause of the epistemic modal (though
F&I, e.g., p. 182, frr. 15, once more diligently caution against the
possibility of being on somewhat shaþ judgments ground).
 MoDAL SYNTAX: DETECTING ITS PARAMETERS WITH VP.ELLIPSIS 45

Some of the main observations from a cartographic thrust are that

epistemic modals tend to avoid co-occurrence with other inhabitants of the
high clausal domain, roughly the traditional C area, a fact visible from
question formation, only-preposing, possibly focus structure more
generally, and at least for some speakers also for wh-exÍaction (cf.
Jackendoff 1972:103, Drubig 2001:2.2,3.3; and Butler 2003, for some
suggestions on sources of variation).

(38) Must / Should/ May Max leave?

(39) Only three people musl should/ may Max see.
(40) * Where must he have been going?

The sentences above all lack the epistemic inte¡pretation according to the
aforesaid studies. Leaving wl¡-movement per se aside, we return to our
more central concems in this study. It is thus possible that the cartogaphic
approach gains consolidating support from further syntax-oriented studies.
Without getting into the issue of how quantificational elements move and
get into scope positions, note that albeit being able to cross nonfinites, by
and large they do not cross finite clause-boundaries, as opposed for
instance to topicalization (cf. Johnson, 2000 for discussion). By way of
simplified midway conclusion, given the following conditions:
. when merged into a derivation epistemics close off the finite
clause,
r (epistemic) modals only appear in hnite clauses in (standard)
English,
. quantification generally stays within the finite clause,
then the ECP follows. If correct in some form, then this line of thought
makes the interesting quantificational facts derivable and aligns them in the
more general context of research on modal syntax, ideally towards
explanation.r5
15
F&I 1e.g., section 3.2) also mention the quest for explanation. They contend that one
should explain why epistemics should be so high. Though not too often asked,
requirements asking not only for proofthat a node X c-commands a node Y, but, after
having presented syntactic evidence that it does so, also asking why this is the case may
have an interesting appeal fiom a more general perspective on language. From what we
know on the specific case of epistemics, language acquisition, and the way mental
representations work are things that come to mind, but essentially such requirements fall
outside the scope of syntax proper. As a matter of fact, if seriously asked, the issue may
46 R¡rr¿us GrncBr-

Note that F&I foreshadow a venue roughly along the cartographical

lines described above, and present purported arguments against it, as
follows:
. justi$, that the one case discussed, where the ECP fares worse
than the cartographic (topological) account, should not be
considered (F&I: 3.3. 1),
¡ attempt to show that syntactic claims on epistemic modality are
in trouble (F &l: 3.3.2),
r argue that modals cannot move to epistemic positions (4.1).
We turn our attention to each of these arguments. The first argument
implicitly acknowledges that a cartography-like approach might do better
on cases of antecedent-contained deletion (ACD), and is then concerned
with an interesting discussion of why QR and ACD-QR are different,
against traditional beliefs. Due to this difference, the implication then goes,
the latter should not be considered quantiher movement any longer. By that
reasoning, given that the ECP's domain of competence is specifically
restricted to QR, and not to explaining epistemics in other terms, notably
not in other quantihcational terms either, the ACD cases are discarded as
irrelevant. This does not prove cartography wrong-it rather nicely justifies
that the ECP itself is at least descriptively not incorrect once one narrows
down its domain accordingly.
Let's now proceed to the second part of the argument (F&I: 3.3.2).
The actual piece ofevidence, and the essential portion ofdiscussion against
cartography F&I present, centres on what they assert to be "low epistemic
modals."l6 Three sentences are given:

need an investigation of the standard theory about C¡1¡ as being operative in some
version of the syntactocentric T-model, in which phrase-structure is rather an axiom
than a theorem - an exploration that would take us too far afield here.
16
This is a somewhat puzzlingly declared argument when fending off cartography. Is
the set {ror+ have to, need+not, can't} of (41) representing the low modals, and the
others are still "the high ones"? Then one could simply restrict the domain of the
topological account to the complement set, e.g., due to morphosyntactic reasons, and
then still derive the ECP. It seems a dubious move to discard topology while taking
topological relations for granted. I take it that the proper logic of the ECP argument
should involve an additional step of a reductio ad absurdum, and continue the argument
from there.
 MoDAL SYNTAX: DETECTING ITS PARAMETERS WITH VP-ELLIPSIS 47

(41) a. John does not have to be at home.

b. John need not be home.
c. John can't be at home.

Sentence a. contains have to, a "more verbal" element as acknowledged by

the authors. This can indeed be an interesting counterfoil for discussing
English modals in general, morphosyntactically it is, however, notoriously
too different to be seen on a par with them (Roberts 1993, Stowell forth.)'
Taking negated need as an illustration as in b. may be in need of more
discussion as well, since negation scoping over epistemic necessity
expressed through need-the purported crucial argument of this sentence-
"is only acceptable to a small fraction of speakers" (Butler 2003: 985).
Suppose we shove acceptability and morphosyntactic concerns aside, and
focus on the remaining stock. First, we are still dealing with suppletive
forms not with the may's and must's F&l (p. 184) started out with (cf. the
discussion in fn. l2). Furthermore, for this argument to refute the phrase-
structure approach it presupposes that all the latter hinges on is negation.lT
It is in fact perhaps no accident that the putative counter-examples involve
negation. First note that a modal llke need is well-known to be a polarity-
sensitive item (cf., e.g., Cormack & Smith 2002).tB I extend this approach
and claim that epistemic can't is also but a polarity item. By and large,
there is no epistemic affirmative can ínBnglish (Stowell, forth). This may
then represent additional and independent evidence why the examples in
point involve negation (under the proviso that negation is a good tester).
Third, F&I (p. 185f.) claim that on a topological account, epistemic
modals would be moved to a high position, but they note that on stacked
modal structures it is rather the surface structure that ends up being

tt While negation often plays a central role in cartography oriented approaches, it is less
than accurate to assume that it is the only argument ofthe topological account; see for
other arguments Jackendoff (1972), Cinque (1999), Drubig (2001), Butler (2003), and
the discussion ofthis paper, among others.
r8
One may slightly disagree with the terms of Cormack & Smith here, who claim that
need is a negative polarity item, although they give the right contexts of its occurrence
elsewhere. Particularly need as a modal is also licensed by, e.g., interrogatives. By
contrast, there are languages that have indeed developed modal uses more restrictedly
licensed by negation. Thus German brauchen'need' is a case in point. This verb is not
standardly licensed in intenogatives in its modal use, but it is by negative polarity
contexts.
48 Rnrøus Genc¡r

interpreted as epistemic. This argument contains some assumptions tn

syntactic terms. I will only briefly illustrate why it would present
drawbacks as a counter-argument placed in a syntactic discussion. To
begin, it is based on unresfficted feature checking, or movement. One
could, for instance, imagine economy conditions why modals check
epistemic feafures paralleling surface structure, at least as a general rule,
rather than crossing paths. It would seem dubious instead to expect that
modals should freely move diverging from first-merged structures given
their interpretable impact. In languages with multiple modal sequences, it is
the structurally higher modal that gets an epistemic reading. This by itself
is in fact a strong argument for positing epistemicity high in relative terms.
The modals presented by F&I þ. 185) are, moreover, lexical verbs.
Furthermore, it is not the case that modals are totally precluded from scope
reversal situations, it is rather the case that modal expressions usually are,
perhaps regulated by requirements of avoiding gibberish at LF, economy,
and perhaps indeed also by phrase strucfure as aforementioned. There are,
however, even cases of scope reversal known with modal expressions like
in (42), in which a lexical epistemic raising expression like seem scopes
over the rootmodal can't.

(42) John can't seem to run very fast. (Quirk 1965: 217 , cited from Langedoen 1970:
2s)

This might, however, happen in rather exceptional constructions only.

More productively, in cases when it comes to the interaction not of pairs of
modals, but of one modal and one auxiliary, scope reversal is again
potentially possible, however obeying modal nuances (cf., e.g., Condoravdi
2002).
Concluding this comparison, we may note that the quantificational
approach is not orthogonal to cartography, though the domains of
application do not overlap. There are significant intersections; e.g., both
approaches try to captue aspects in the behavior of epistemic modals.
Despite the fact that the proposal is not entirely problem-free, recall that
F&I also correctly account for an impressing range of data involving
epistemic modals (sketched in section 4). It's also their merit to have
opened up ways of viewing modality, in a manner reminiscent of the
 MoDAL SYNTAX: DETECTING ITS PARAMETERS WITH VP-ELLIPSIS 49

research on intervention effects and different from the well-trodden

-
syntactic paths, at the very least.
Retuming to cartography, the idea of the phase and paficularly the
sort of "reduplication with a purpose" might bring some alleviation to the
heavy style research on rich structures over the last years. There is in fact
evidence from VP-ellipsis that partly corroborates and partly complements
some of the structural relations proposed by Butler. McDowell (1987)
offers some interesting views relating modality and elliptical processes
such VP-ellipsis in (43Y94) (cited from Drubig 2001: 30). She also
discusses the related pseudo-gapping construction illustrated in (45) (for
which she uses the term gapping), whose discussion I will leave aside for
space reasons and for its more complex character, due to the additional
syntactic operations involved known on independent grounds (Lasnik
tgee).

(43) John must wash his car every day and Peter must
-
too (*epi)
(44) John will often sit there and do nothing and Bill will
-
too (*epi)
(45) John may not obey his mother, but he must his
-
father (??epi)

McDowell's main claim is that ellipsis is incompatible with an epistemic

reading of the modal. Originally phrased in govemment theory, in very
general cartographic terms her proposal translates into having an epistemic
modal in a scope position higher than the deontic modal, and thus too
distant from the ellipsis site for licensing to occur. The
traditional
assumption is that VP-ellipsis depends on the presence of a licensing head
along the lines of Infl (cf. Lobeck 1995 and Drubig 2001 for explanatory
oriented exploration). Consequently, the epistemic modal position is
predicted to be ungrammatical with ellipsis.
Note that all the examples rendered above include the universal
quantifiers must and wll/. However, with existential quantifiers like, say,
might, could, or may ellipsis licensing becomes possible on both readings,
to wit the epistemic reading will not be excluded.

(46) Jane may wash her car and Mary may/mighlcould - too.
(47) Mary will talk to her boss and John might to his.
-
(48) John will fly to London and Mary may too. (Boðkovic, 1994: 280)
(49) You have to be a real masochist to
-
want to direct," he says with a smile. But
Fearheiley does, and Smith might- , too. (Gazette.Nel, Maryland, Ãttg.29,2002)
 50 REMUS GERGEL

An assumption McDowell has to stipulate is that only the ambiguous

modals should be reluctant to license VP-ellipsis on an epistemic reading.
This way she would perhaps be able to discard the instances of might in the
examples above, given that this modal is arguably only epistemic in present
day usage. But this cannot discard could from the list of potential licenser,
which has various productive uses. More important is, however, the fact
that this ad hoc condition on ambiguity is conspicuously at odds with her
clause-structural theory of a general incompatibility between epistemic
modals and VP-ellipsis.
Next, let's examine how an epistemic modal such as may, as in (46),
would be discarded in McDowell's framework, incidentally, an enterprise
conducted more explicitly. First, it is clear that may is generally
ambiguous, as McDowell agrees, and the aforesaid assumption does not go
through. A second assumption is needed: The purported reason in this case
is that readings with a future shift are simply not epistemic. For instance
epistemic and future oriented readings of may, would exclude each other
under her approach. However, under the standard concept on epistemic
modality, to which she subscribes, this is a stipulation she is forced to make
for her claim to be applicable. Drubig (2001), on a more focused line of
research on evidentiality also allows epistemic readings with the scheduled
future; Declerck (1991: 8617) observes evidence-evaluation at a future-
shifted time (e.g., for what traditionally are known as epistemic will cases
of the type That will be the mailman tttered, when hearing a ringing bell);
and Stowell (forth.) crediting Karen Zagona (1990) argues more generally
that unless they have a stative complement, epistemics in fact must shift the
eventuality time. Building on the empirical insight of these studies, I argue
that the problem in logical terms has been caused by an illicit use of a
biconditional where no more than one-way implication would have been
motivated. That is, deontics tend to shift the eventuality time to the
future-given the well-known presupposition that if something is
allowed/required to be done, then it has not yet been done. But conversely,
it is not the case that future-shift with a modal automatically implies root
modality.re There is thereby no reason to generally discard épistanics due
le
ln fact note that even the reverse implication is not absolute. There are root modals
which seem to make use of their deontic force with a past-shift of the eventuality time
especially in some formalized registers, e.g., as in (i)-(ii).
(Ð The plaintiff must have filled in a complaint form.
 MoDAL SYNTAX: DETECTINC ITS PARAMETERS WITH VP-ELLIPSIS 51

to temporal properties (which is not to say that there are no correlations

between mood, tense and predicate type; see Stowell for a perspicuous
review and some new observations on these topics). Moreover, consider
(50), for which consultants gave the epistemic reading either as the strongly
preferred or the only one.

(50) Mary may know about the meeting and Melissa may too.
-
V/ith know it is even clearer that temporal forward-shift for
a predicate 71ke
may isnot compulsory and we have an epistemic reading on the modal.
What the examples (43F(50) may indicate instead is that the
necessity/possibility (or universal/ existential) divide also plays a part in
the licensing of elliptical phenomena. They thus offer additional evidence
for an essential subset of the relations proposed. The epistemic/root divide
is, however, also crucial. Note that it is only for epistemics that necessity
and possibility part ways with respect to ellipsis. Ellipsis then first takes
epistemic and root modals apart in that only the former show the variation
in licensing, and furthermore takes apart the epistemic class in that, within
it, a higher head of strong existential quantification (e.g., must) is not an
appropriate licenser. It is worth noting that this neatly complements the
evidence from negation presented by Butler. While ellipsis operates as just
mentioned, negation hrst distinguishes must and may with respect to
epistemic vs. root status (in that only the latter shows scope variation) and
then fine-grainedly takes apart the deontics into necessity and possibility,
though not epistemics, as we have seen.

7. Conclusion and outlook

This discussion has drawn on various strands of interaction between

modality an VP-ellipsis. Specifically, we have argued that ellipsis is a
profitable share in three main domains:
. diachronic reanalysis in the sense ofRoberts (1993),
. modal force, and in particular the special concems on epistemic
modality,
. the distinction between necessity and possibility.

(iÐ You must have acquired all the credits to get your diploma.
52 RBtr¡us Gencsl

While the arguments have been put forward for English, there is to my
mind no immediate reason to overgeneralize to other languages. Rich
functional hierarchies are perhaps best thought of as telescopic in that
particular languages may pull out certain subsets of them and
grammaticalize them accordingly, or may not. Moreover, elliptical
processes quite in the same vein may function in a number of various ways
and may be more or less developed in certain areas in different varieties. In
general, English has a well-developed fype of VP-ellipsis, as numerous
attempts to compare other systems to its criteria have shown (cf. Holmberg
2001, Ngonyani 1996, among others). Without getting into the exploration
of further elliptical systems, let's just note that there may be parallels in
subdomains. For instance, German allows VP-like elision only under very
restricted lexical and information-structural conditions (Klein 1993,
Winkler 2003). It has, however, an overt anaphoric form es with some of
the hallmarks of VP-ellipsis if not all (López & Winkler 2000).
Interestingly this form of VP-replacement follows the epistemic-root
dichotomy, in that it is only licensed with non-epistemic modals (Ross
1969). This would then again confirm the intuition expressed in Drubig
(2001) that epistemic modality is not an appropriate licenser. Romanian, to
take a typologically different example, behaves----once more with the
proviso on elliptical structures and modal grammaticalization-primarily
sensitive to the other parameter, viz. necessity vs. possibility.

(51) DacãMariapoate sã-çi creascã copüi singurã, çi

If Maria can comp-dat.cl grow children.det alone, too
eu pot / *trebuie
_.
I can-lsg/*must_.
'If Mary can raise her children alone, I can/ must _ too.'

In an antecedent-ellipsis configuration as in (51) the strong modal does not

allow elision while the weak one does. One of the hypotheses argued for in
this paper, and with difierent diagonostics in Cormack & Smith (2002) and
Butler (2003), was that strong modality is structurally higher relative to
weak modality. Based on it we might have an explanation for the
Romanian asymmetry too. An intersting task of research would be to
investigate how far such claims fall in place with the mechanism of
agreement (Chomsky 2001). For instance, López (2002) argues on the basis
of variation within expletive constructions that probe-goal agreement
 Mooel SvNrex: DETECTING lrs PARAMETERS w¡rg VP-Elupsls 53

works cyclically and not at a distance as in Chomsþ. This falls then rn

place with the modality observations. The higher modal (depending on the
parameter chosen by language X, ê.g., say, strength by Romanian) is more
likely to fall outside the domain of agreement configurations (in this case
with the subject), given that there are more steps where things might go
awry in a successive cyclic process. Romanian conñrms this upper limit
assessment through its lack of agreement on the necessity modal trebuie,
'must', and through its presence on the possibility modalpøter'can' .
While it is for further research to more fu1ly investigate some of the
latter ideas laid out above, all in all this paper has hopefully shed additional
light on the fact that the syntax of modality displays an interesting
sensitivity to certain semantic and fi.rnctional parameters which I have
investigated from the perspective of elliptical structures.

References

Bok-Bennem4 Reineke (2002) On short V-movement in Romance. Paper presented at

Chronos 5, Groningen.
Boskovic, imyo çe}q D-structure, Theta-Criterion, and movement into theta-
positions. Linguistic Analysis 24: 247-86.
Bowers, John (2001) Predication. ln Mark Baltin & Chris Collins (eds.), The Handbook
of Contemporary Syntactic Theory, pp. 299-333. Oxford: Blackwell.
Butler, Jonny (2003) A minimalist treatment of modality. Lingua ll3:967-996.
Chomsky Noam (2001) Derivation by Phase. In Michael Kenstowicz (ed.), Ken Hale. A
Liþ in Language,pp. l-52. Cambridge, MA: MIT Press.
Cinque, Guglielmo (1999) Adverbs and Functional Heads: A
Cross-Linguistic
Perspective. Oxford: Oxford University Press.
(2001) 'Restnrcturing' and the Order of Aspectual and Root Modal Heads. In
Guglielmo Cinque & Giampaolo Salvi (eds.), Current Studies in ltalian Syntax.
- Essays offered to Lorenzo Renzi,pp. 137-55. Amsterdam: Elsevier.
Condoravdi, Cleo (2001) Temporal Interpretation of Modals: Modals for the Present
and for the Past. ln David Beaver, Luis D. Casillas Marzinez,Brady Z. Clark, &
Stefan Kaufrnann (eds.), The Construction of Meaning, pp. 59-87. Palo Alto:
CSLI Publications.
Cormack, Annabel & Smith, Neil (2002) Modals and Negation in English. In Sjef
Barbiers, Frits Beukema, & Wim van der Wurff (eds.), Modality and its
Interactionwith the Verbal System, pp.l33-63. Amsterdam: John Benjamins.
Declerck, Renaat (1 991) Tense in English. London: Routledge.
54 REMUS GERGEL

Denison, David. (1992) Counterfactual 'May Have.' In Marinel Gerristen & Dieter Stein
(eds.), Internal and External Factors ín Syntactic Change, pp' 229-56. Berlin:
Mouton de Gruyter.
Depiante, Marcela A. (2000) The Syntax of Deep and Surface Anaphora: A Study of
Null Complement Anaphora and StrippinglBare Argument Ellipsis. Ph.D.
dissertation, University of Connecticut.
Drubig, H. Bernhard (2001) The s¡mtactic form of epistemic modality. Ms., University
ofTübingen.
(2003)Towardatypologyof focusandfocusconstructions. Linguístics 41: 1-50.
Emonds, Joseph E. (1970) Root and Structure Preserving Transformatìons.
- Bloomington: Indiana University Linguistics Club.
(2000) Lexicon and Grammar: The English Syntacticon. Berlin: Mouton de
Gruyter.
-von Fintel, Sabine (2003) Epistemic containment. Linguístic Inquiry 34'.
Kai & Iatridou,
173-98.
Fox, Danny (1995) Economy, scope and semantic interpretation - evidence fiom VP
ellipsis. NEZS 25: 143-57.
(2000) Economy and Semantic Interpretation. Carrrbndge, MA.: MIT Press.
Fox, Danny & Lasnik, Howard (2003) Successive-cyclic movement and island repair:
- the difference between sluicing and VP-ellipsis. Linguistic Inquiry 34:. 143-54.
van Gelderen, Elly (2003) ASP(ect) in English Modal Complements. Studia Linguistica
57:2743.
Gergel, Remus (2002) Short-distance reanalysis of Middle English modals: evidence
from ellipsis. Ms., University of Tübingen.
de Haan, Ferdinand. 1997. The Interaction of Modality and Negation. A Typological
.Slrzdy. New York: Garland.
Harley, Heidi B. (1995) Subjects, Events and Licensing.Ph. D. dissertation, MIT.
Harris, Alice C. & Campbell, Lyle. 7995. Historical Syntax in Cross-Linguistic
Perspective. Cambridge: Cambridge University P¡ess.
Heine, Bemd (1993) Auxiliaries: Cognitive Forces and Grammaticalization. Oxford:
Oxford University Press.
Hofinann, T. Ronald (1976) Past Tense Replacement and the Modal System. ln James
McCawley (ed.), Syntax and Semantics. Notes from the Linguistic Underground,
pp. 85-100. New York: Academic Press.
Holmberg, Anders (2001) The syntax ofyes and no in Finnish. StudÌa Linguistica 55:
14t-7 5.
Huddleston, Rodney & Pullum, Geoffrey K. (2002) The Cambridge Grammar of the
Englßh Language. Cambridge: Cambridge University Press.
Jackendoff, Fiay (1972) Semantic Interpreturion in Generative Grammar. Cambridge,
MA: MIT Press.
Johnson, Kyle. (2000) How Fa¡ Will Quantifiers Go? In Roger Martin & Juan
Uriagereka (eds.), ,lrep by Step Essays On Minimalist Syntax in Honor of Howard
Lasnik, pp. 187-210. Cambridge, MA.: MIT Press.
 MoDAL SYNTAX: DETECTING ITS PARAMETERS WITH VP-ELLIPSIS 55

Klein, Wolfgang (1993) Ellipse. In Joachim Jacobs, Amim von Stechow & Theo
Vennemann (eds.), Syntax: Ein internationales Handbuch zeitgenössischer
F or s chung, pp. 97 8-87 . Berlin: Mouton de Gruyter.
Kratzer, Angelika (1981) The Notional Category of Modality. In Hans-Jürgen
Eikmeyer & Hannes Rieser (eds'), Words, l\lorlds, and Contexts, pp. 38--74'
Berlin: Mouton de GruYter.
Kroch, Anthony ( I 989) Reflexes of grammar in pattems of language change. Language
variation and change 7:199-244.
(2002) Modelling language change. Paper presented at DIGS VII, University of
Girona, June 2002.
-Langendoen, Terence D. (1970) The'can't seem to'construction. Lingußtic Inquiry I:
25-35.
Lasnik, Howard (1999) Pseudogapping Puzzles. In Elabbas Benmamoun & Shalom
Lappin (eds.), Fragments: Studies in Ellipsis and Gapping, pp' l4l-74. Oxford:
Oxford University Press.
Lewis, David (1973) Counterfactuals. Oxford: Blackwell.
Lobeck Anne C. (1995) Ellipsis: Functional Heads, Lícensing and Identification.
Oxford: Oxford University Press.
López, Luis. (2002) On Agreement: Locality and Feature Valuation. In A¡temis
Alexiadou (ed.), Theoretical Approaches to Llniversals, pp. 165-209. Amsterdam:
John Benjamin.
López, Luis & Winkler, Susanne (2000) Focus and Topic in VP-anaphora constructions'
Linguis t i cs 3 8 : 623 464.
Ngonyani, Deo (1996) VP ellipsis in Ndendeule and Swahili applicatives. UCLA
lVorking Papers in Syntax and Semantics 1: 109-128.
Oku, Satoshi (2001) A minimalist Theory of LF-Copy. In Galina M. Alexandrova &
Olga Amaudova (eds.), The Minimalist Parameter, pp. 281-294. Amsterdam:
John Benjamins.
Papafragou, Anna (1998) Inference and word meaning: the case of modal auxiliaries.
Lingua 105: 147.
Partee, Barbara H., ter Meulen, Alice & Wall, Robert E. (1993) Mathematical Methods
in Linguistics. Dordrecht: Kluwer.
Pollock, Jean-Yves (1989) Verb movement, UG, and the structure of lP. Linguistíc
Inquiry 20:365424.
Quirk, Randolph (1965) Descriptive statements and serial relationship. Language 41:
20s-17.
Reis, Marga (2001) Bilden Modalverben im Deutschen eine syntaktische Klasse? ln
Reima¡ Müller & Marga Reis (eds.), Modalität und Modalverben im Deutschen,
pp. 287-318. Hamburg: Buske.
Rizzi, Luigi (1997) The Fine Structure of the Left Periphery. In Liliane Haegeman (ed.),
E leme nts of Gramm ar, pp. 28 1-337 . Dordrecht: Kluwer.
Roberts, Ian G. (1985) Agreement parameters and the development of English modal
auxiliaries. Natural Langtage and Linguistic Theory 3: 2l-58.
56 REMUS GERGEL

(1993) Verbs and Diachronic Syntax. A Comparative Hßtory of English and

French. Dordrecht: Kluwer.
-
Roberts, Ian G. & Roussou, Anna (2003) Syntactic Change: A minimalist Approach to
Grammaticalization. Carrbidge: Cambridge University Press.
Ross, John R. (1969) Auxiliaries as main verbs. Philosophical Linguistics 1:77-102.
Stowell, Tim. (forth.) Tense and Modals. In Jacqueline Guéron & Jacqueline Lecarme
(eds.), The Syntax of Time. Cambridge, MA: to appear with MIT Press.
Wamer, Anthony R. (1992) Elliptical and Impersonal Constructions: Evidence for
Auxiliaries in Old English. ln Fran Colman (ed.), Evidence from Old English:
Material and Theoretical Bases for Reconstructíon, pp. 178-210. Edinburgh: John
Donald.
(1995) Predicting the progressive passive: pararnetric variation within a lexicalist
framework. Language 77:. 533-557 .
-
Williams, Edwin (1994) Tkematic Structure in Syntax. Cambridge, MA: MIT Press.
Winkler, Susanne (2003) Elþsß at the Interfaces: An Information Structural
Investigation of
Sentence-Bound and Discourse-Bound Ellipsis in Englßh.
Habilitationsschrift : University of Tübingen.
Wurmbrand, Susanne (2001). Infinitives: Restructuring and Clause Structur¿. Berlin:
Mouton de Gruyter.
Zagona, Karen (1990) Times as temporal argument structure. Ms., University of
Washington, Seattle.
Ztbizarreta, Maria Luisa (1982) On the Relationship of the Lexicon to Syntax. Ph.D.
dissertation, MIT.

Contact information:

Remus Gergel
SFB 44Il 813
Nauklerstrasse 35/ 3.05
7 207 4 T äbingen, Germany
Tel.: ++49 (0)70711 29-771.63
Fax: ++49 (0\70711 29-5830
Email: [email protected]