head movement and negation in english
TRANSCRIPT
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
HEAD MOVEMENT AND NEGATION IN ENGLISH1
By Annabel Cormack and Neil SmithUniversity College London
Abstract
This paper has two main aims: to argue for a third negation
position (Echo) and to introduce and defend a `soft constraint'
solution to the problem of the `displacement' of a head relative
to negation. We argue that an adequate account of the
position of various heads in relation to negation needs first
to establish the LF positions of these heads and then to
determine the reasons for the PF displacement of some
heads. Scope of modals and adverbs with respect to negation
reveals that there must be three negation/polarity positions:
Pol[neg], one above this (Echo[neg] ) and one below
(Adv[neg] ). There are two modal positions, one above and
one below Pol. PF positioning of the heads is determined
partly by what morphophonological combinations are avail-
able, and partly by a set of soft constraints. Corroborating
evidence is provided from Basque, Catalan and Italian.
Part One: LF-interpretable positions of heads
1. Introduction
In this paper, we reconsider the interaction of negation with verbs
(including modals and auxiliaries) and temporal adverbs in English.
In the first part of the paper, we introduce the problem, and indicate
some of the theoretical considerations which drive our analyses.
These lead to the postulation of three distinct LF negation/polarity
Transactions of the Philological Society Volume 98:1 (2000) 49±85
# The Philological Society 2000. Published by Blackwell Publishers,108 Cowley Road, Oxford OX4 1JF and 350 Main Street, Malden, MA 02148, USA.
1 This paper is based on Cormack and Smith (1998). We are grateful to JohnAnderson, Keith Brown, Robyn Carston, Susanna Aspa Corbeto, John Harris, Hansvan de Koot, Bob Ladd, Rita Manzini, Ad Neeleman, Paul Rowlett, Vieri Samek-Lodovici, Larry Trask, Deirdre Wilson and two reviewers, for advice, comments,criticism and information.
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
positions for English, with the `Echo' position being an innovation.
We also show that an equivalent negation position must exist in
Catalan, Basque and Italian. Readers whose primary interest is in
descriptive semantics or pragmatics, or who are sceptical of our
theoretical position, should be able to read Part One and may ignore
Part Two.
In the second part of the paper, we address the question of where
the PF parts of various verbal and negation heads are merged, and
how these are related to the LF positions of the heads. Here, we
offer a `soft constraint' analysis, within our `Split Signs' framework,
and show that this accounts for the possible surface positions of not
and n't in English. The core data that need an account in English are
familiar:
(1) a. John often snores.
b. *John not snores.
c. John did/will/must not snore.
d. *John snores not.
This array of data gives rise to the following questions:
Q1: Why is (1b) ungrammatical? In particular,
Q2: What accounts for the difference between often and not in (1a)
vs. (1b)?
Q3: Why can an Aux or modal precede not, while a V cannot ( (1c)
vs. (1d) )?2
We couch the discussion in an austere version of Minimalism
(Chomsky 1995), which will enable us ultimately to dispense with
head movement, though for exegetical purposes we begin by
presenting the analysis in terms of movement of the PF-interpret-
able part of a head away from the merge position of the head.
We are going to claim that explanations for these data require
that the LF-interpretable positions of the set of UG-given and
obligatory functional projections above V in a clause merge as
50 transactions of the philological society 98, 2000
2 One of our reviewers raises the question of examples such as I think not. We takethe negation here to be in a lower (largely elided) clause. The legitimacy of thestructure is an interesting issue, which we have no space to discuss.
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
shown in (2), where either of Pol and Echo may contain pos instead
of neg.3,4 C is split, as in Rizzi (1997).
(2)
C FORCE
Echo[NEG]
Pol[NEG] V...
C
T
In (3), possible LF positions for various other heads are shown
(obligatory heads are shown in bold).
(3) CForce Echo C T * (Modal1) * Pol * (Modal2) * (Aux) . . .
* [VP . . . [V0 * V] . . .
The bracketed positions allow for optional adjoined heads.5
Typically, adjuncts may appear in more than one environment,
with possible positions regulated by selection. Modal1 and Modal2are possible positions for modals (see section 3). The `*' marks
positions where temporal adverbs may optionally be adjoined;
Stroik (1990) argues for the possibility of a V0 adverbial adjunct.
Adverbial negation, Adv[neg], may adjoin at any adverb position
below Pol.6 The subject will appear immediately before T, and the
object, we claim, before V0 (Cormack 1999).
51cormack and smith ± head movement and negation
3 We assume without argument that negation is merged in a head position ratherthan a specifier position (see Cormack 1999 for discussion of specifiers).
4 It may be that the head we have called Echo here should be equated with Laka's� projection (Laka 1994). See section 5 below.
5 We take adjuncts to be headed by one- or two-place `minor category' heads(Cormack 1999). Because these are syntactic and semantic non-binding operators,they behave more like functional than lexical heads.
6 What we call Adv[neg] is sometimes referred to as `constituent' negation. Wethink this term is misleading.
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
We will argue below for the placing of sentential negation under
Pol rather than simply under Neg, and for the positioning of Pol
below T. We argue as well for a higher `polarity' position, Echo,
which, like Pol, may host a negation head.
Our answer to the questions posed above is essentially as
follows: displacement of PF parts of items relative to their LF
parts is regulated by a set of soft constraints (section 9). These
require inter alia that a tensed element precede negation (Q1). The
displacements are subject to Checking, with a category-based
minimality requirement. Because of this, a verb cannot be displaced
across Negation, which bears a V-feature (section 10), but a Modal
or Auxiliary may be (Q3). An adverb will not block Tense-related
Checking (Q2).
2. Merge and LF
The Minimalist program has eliminated D-structure and S-struc-
ture. After heads are merged into a structure, movement may apply
either to Formal Features or, if required by morphophonological
conditions, to a head (or phrase). Where head movement takes
place, it is as a result of morphophonological necessity. Concep-
tually, morphophonological conditions apply to the PF-interpret-
able part of a linguistic sign, so that the question arises as to whether
head movement has any semantic (LF-interpretable) effect. Our
hypothesis is that the answer to the question is `No', as made explicit
in the axiom in (4):
(4) A head is merged in the position in which it is LF-interpreted.
Given this axiom, the position of Pol[neg] below T in (2) follows
from sentences such as (5), under the interpretation `It is sometimes
obligatory that John not leave early':
(5) John sometimes mustn't leave early.
Here, at LF, suppose Tense introduces a temporal interval, within
which the temporal adverb may further specify times. The adverb
then must be within the scope of T, and the semantics of the
sentence tells us that sometimes has scope over must, and that the
52 transactions of the philological society 98, 2000
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
modal has scope over negation. By transitivity, T has scope over
sentential negation.7
Relevant examples of analyses ruled out by our hypothesis
include the kind of derivation proposed by Zanuttini (1997a: 101±
102), where in the Milanese sentence in (6), the negation word no is
merged at the trace position, but is interpreted at both PF and LF in
the position above the adverb semper.
(6) l'a no semper di t tuÈscos.
s.cl-has neg always said everything
`He hasn't always said everything.'
Equally, we reject the `Duke of York' account in Roberts (1998:
115), where, in order to account for the `not ± possible' scope order
in (7),
(7) There can't be a unicorn in the garden.
he postulates that some modals, including can, are generated lower
than Neg. The modal must then be raised to Neg before Spell-Out.
Because it is assumed that this movement carries along the LF part,
Roberts states that ``At LF, narrow scope modals are `recon-
structed' into the lower position. This movement is obligatory, for
reasons which are unclear.'' Of course, the requirement would look
somewhat less absurd under the Copy Theory of movement, but the
question as to which copy would be spelled out at which interface
would remain to be dealt with (see Roberts footnote 5, p. 120).8
53cormack and smith ± head movement and negation
7 Alternatively, suppose the temporal adverb (or a default existential) introducesquantification over times, while the Tense node introduces a restriction on these. Sucha restriction cannot have scope under the modal, and so here, too, T must have scopeover Pol. Placing the Tense predicate under the modal would lead to a representationof the form 9t&[present(t) ^ P(t)], where & is the necessity operator and P, apredicate over times. But this entails that there are some times which are obligatorilyin the present, which is presumably false. Further, with Modal1 have to, the have isintuitively within the scope of Tense. John often had to sneeze does not mean `thereexist many times which necessarily are in the past and at which John sneezes', noreven `there exist many times in the past at which necessarily John sneezes'. It means`In the past, there were many times at which necessarily John sneezed'. Thiscorresponds to the scope order T Adv Modal, which is what we are claiming to becorrect.
8 For discussion of the `Duke of York' gambit, see Pullum (1976).
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
Similarly, Laka's (1994) proposals for Basque cannot be accepted
as they stand. She argues (1994: 21±23) that the Negation head in
Basque is above T (to account inter alia for the grammaticality of
Negative Polarity Items such as inork `anyone' in subject position),
but then claims (1994: 23±25) that, as a principle of UG, all clausal
operators (including negation) must fall within the scope of T at LF
(to account for the surface order Neg ± finite verb). This entails that
Neg has two distinct positions at which it has LF-interpretable
properties: the higher one for scope with respect to the subject, and
the lower one for scope with respect to T.9
As we see from these examples, adopting (4) as an axiom reduces
the options available to the syntactician. Equally, and desirably, it
reduces the options available to the child learner. Once the relative
semantic scope of the various heads is known, the relative order in
which they are merged at LF is known.
3. LF positions for negation and modals
On the basis of the axiom in (4), we argue for three broadly clausal
LF-positions for negation in English.
First, uncontroversially, we distinguish Pol[neg] from Adv[neg].
The latter is always realised as not, whereas Pol[neg] may be
instantiated by not or n't. Adv[neg] occurs lower down the tree
than Pol, giving rise to the contrast between (8) and (9), where the
shared reading arises from Pol[neg], and the extra reading in (9)
from Adv[neg] (Klima 1964, Ernst 1992, Williams 1994: 49,
Zanuttini 1996, de Haan 1997, and other authors).
(8) Gerry couldn't swim the channel. not [poss [ . . .
(9) Gerry could not swim the channel. not [poss [ . . . or
poss [not [ . . .
Like Haegeman (1995), we take it that Pol may have a positive
instantiation, Pol[pos]. Evidence for this is given in section 10
54 transactions of the philological society 98, 2000
9 Laka does however suggest that an alternative, where negation is necessarilybelow T at LF as well as at surface structure, should be considered (1994: 53±55).
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
(example (79) ). However, unlike Haegeman (1995) and Zanuttini
(1996), we claim that Pol is below T (as does Pollock 1989 ).
Our analysis of negation has been influenced by the work of
Picallo (1990) on Catalan, which prompted us to use the English
modals as a probe for negation positions. Picallo argues that, in
Catalan, simple sentences with modals are monoclausal, and that a
modal with an epistemic interpretation is always merged above
negation, and one with a deontic interpretation, below negation, as
in examples (10) and (11). We take it that the negation here is
Pol[neg]; for Echo[neg], see section 4.
(10) En Jordi pot no haver sortit.
may not have left
`It is possible that Jordi hasn't left yet.' may not (epistemic)
(11) En Jordi no ha pogut sortir.
not has could leave
`Jordi hasn't been able to leave.' not may (deontic)
By contrast, it turns out that in English there is a class of modals
(Modal2) which must be LF-merged below Pol[neg], irrespective of
the epistemic or deontic reading (e.g. could, cf. the lack of ambiguity
of (8) ). Another class (Modal1), including should, must be merged
above Pol, again irrespective of the epistemic or deontic reading, as
shown by (12).
(12) Paula shouldn't/should not be at home now.
should [not [ . . .
`It is necessary/obligatory that Paula not be at home now.'
6� `It is not necessary/obligatory that Paula be at home now.'
May and might, however, behave like the Catalan modals, so we
have the modal distribution shown in Table 1, which corresponds to
the readings given in Quirk et al. (1972) or Palmer (1990).10
Given the usual array of heads in (13), the modals fit into the
structure of the clause as shown in (14). Note that the terms Modal1and Modal2 are for mnemonic convenience. All English modals
55cormack and smith ± head movement and negation
10 For further discussion of modals in relation to negation, in Basque and Italian,as well as in English, see Cormack and Smith (forthcoming).
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
have the same syntactic category (perhaps Aux), with the Modal1subgroup distinguished by selection for Pol.11
(13) C (subject) [T [Pol [V . . .
(14)
T
(Mod1)
(Aux) V...
Pol
(Mod2)
Both modal positions may be filled simultaneously, as predicted by
this configuration, as in (15) (Catalan, Picallo 1990: 294), and (16)
(Hawick Scots, Brown 1991: 98).
(15) En Pere deu poder tocar el piano.
must can play the piano
must (epistemic) [ Pol pos [can (deontic) . . .
`It must be the case that Peter is able/allowed to play
the piano.'
56 transactions of the philological society 98, 2000
Table 1
Pre-Pol (Modal1) necessity shall, should, must, will, would,ought + to, is + to,
possibility epistemic readings only: may, mightPost-Pol (Modal2) necessity need
possibility can, could, daredeontic readings only: may, might
11 The variation in meaning according to selection seen with may/might in English,and with the Catalan modals is not unique. Another English instance is forget, whereJohn forgot that he had locked the door and John forgot to lock the door show distinctbut related meanings.
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
(16) He must no can do it.
must (epistemic) [ Pol not [can (deontic) . . .
`I conclude he is unable to do it.'
4. Echo
Despite the robust generalisations captured in Table 1 and (14),
there are certain contexts in which a Modal1 may be interpreted
within the scope of negation, as in (17). We argue that such contexts
are pragmatically `echoic' (Carston 1994, Wilson and Sperber
1988).12
(17) Shouldn't Susie be at school? ? [not [oblig [ . . .
We illustrate such echoic contexts with questions of various sorts,
and denial, using should. Of course, Modal2 and auxiliaries may also
appear in echoic contexts, but they exhibit no useful scope effects.
A: questions
(18) A to B: Shouldn't you be at work? ? [not [oblig`Is it not the case that you (B) should be at work?'
(19) Should you not be at work? Possible interpretation as (18)
including tags
(20) You should eat more vegetables, shouldn't you?
and rhetorical questions
(21) Who shouldn't eat more vegetables? ?x not [oblig . . . x
B: denial:
(22) A: You should eat more vegetables.
B: No I shouldn't. not [obligInterpretation: `No, it is not the case that I (B) should eat more
vegetables.'
Since the presence of an Echoic reading cannot plausibly change
the selection properties of a Modal1, such as should, we must assume
57cormack and smith ± head movement and negation
12 For discussion of negation in questions and tags, see also Cattell (1973) andLadd (1981).
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
that the negation in (18) to (22) is not Pol, but arises from a further
negation position, which we call Echo. The LF position of Echo is
higher than that of Pol, and above the Merge position of should,
giving the required scope for the LF in these examples. We assume it
lies between CForce and the lower C, as in (23).
(23) CForce Echo C [subject] T (Modal1) Pol V. . .
The PF of Echo[neg], however, may appear in the same
relative positions that Pol[neg] does; note especially the position
of Echo[neg] lower than the subject in (19). Emphatic affirmation,
assigned to Pol/Sigma by Chomsky (1955/75: 448 and 1957: 65) and
Laka (1994), is the converse of denial, and belongs under Echo[pos].
American English so, discussed by Laka, is probably LF-merged at
(focused) Echo[pos], too.
We take it that the emphasis in sentences such as (24) is a reflex of
the focusing of Echo[pos], so that, here again, there is evidence that
the PF of Echo is lowered to a position near T.13
(24) The policeman did bite the dog.
Horn (1989: 366) expresses surprise that if there is ambiguity
(between our Pol[neg] and Echo[neg] ), the same morpheme is used
in both cases. However, given that the meaning is the same, this
is hardly surprising: even related but distinct meanings frequently
share a PF form.14 Moreover, in the Piedmontese dialect of
Italian, discussed in section 5.2, Echo is expressed with a distinct
morpheme.
Given the evidence for Echo[neg] in English, the question
naturally arises as to whether there is comparable evidence in
Catalan. The answer is `Yes': in echoic contexts, it is indeed possible
for negation to be construed as having scope over an epistemic
modal, as shown in (25).15
58 transactions of the philological society 98, 2000
13 It is relevant that McCloskey (1996) has recently argued convincingly for Irishthat C has to be lowered, as has Henry (1992) for Belfast English.
14 However, it is arguable that Echo and Pol, and perhaps the Adv of Adv[neg],should all have the same syntactic category, with the different occurrences beingdistinguished by selection properties only (see Cormack and Smith 1998).
15 The Catalan data below were given to us by Susanna Aspa Corbeto.
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
(25) No deu haver pas marxat en Joan?
not must have pas left the John
`Mustn't John have left?' [His car isn't there.]
In this section and the previous one, we have established that there
are three positions for negation in English. Negation in the highest
position, Echo, has scope over any modal. It also necessarily has
scope over the subject of the clause. In contrast, negation in any of
the possible Adv[neg] positions is within the scope of any modal.
Negation appearing at Pol[neg] has wide scope with respect to
Modal2 elements or auxiliaries, but narrow scope with respect to
Modal1 elements. The subject most naturally has scope over either
Pol[neg] or Adv[neg], but may have lower scope by virtue of
`reconstruction to the VP internal position', or some semantic
equivalent (Lechner 1998).
In the next section, we consider evidence for diverse negation
positions in Basque and Italian.
5. Evidence for Echo[neg], Pol[neg] and Adv[neg] in Basqueand Italian
5.1. Basque
We mentioned earlier that we could plausibly equate our Echo head
with the � head Laka proposes for Basque (Laka 1994). The
question then arises as to whether or not Basque has Pol in addition,
and in particular a lower Pol[neg]. If it does not, then our hypo-
theses would be at risk for two reasons. First, if the position of Pol is
not given by UG, and it can appear above instead of below T, the
severity of the learnability problem is increased. Second, if there
were no Pol at all, then the principles ± whatever they are ± that
dictate when material in an LF structure may or must be interpreted
as echoic will also have to vary from language to language, with
input from LF in some cases, but not others. This seems to us to be
implausible.
Laka (1994) argues for two negation positions for Basque. One is
the pre-VP position (1994: 71±73), for `constituent negation', which
we equate to our Adv[neg]. This is seen in (26), which also shows
`affirmative fronting'.
59cormack and smith ± head movement and negation
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
(26) Nik diot Mariari/*inork trikota ez eman.
I have to-Mary/anybody sweater-the not given
`I have not given the sweater to Mary/anybody.'
The second is Laka's � projection, which hosts either Neg or the
emphatic positive counterpart Aff (1994: 74±78). Laka argues that �has scope over T and over the subject. The structure she argues for
(1994: 13±23) is shown in (27a), in which some projections are head-
final. Our notation, in contrast, would give (27b), where we have
added the VP adjoined Adv[neg].
(27) a. [�P [� Neg/Aff ] [IP [AspP VP Asp ] I ] ]
b. [EchoP [Echo neg/pos ] [TP [AuxP [( Adv[neg] ) VP ] Aux ] T ] ]
Fortunately for our thesis, there is evidence for a lower Pol[neg],
arising from the interaction of negation with the deontic obligation
modal.16 The modal corresponding to deontic `must' is constructed
by combining an item homophonous with the noun behar `necessity'
with an auxiliary, almost always the defective transitive auxiliary
*edun.17
(28) Joan behar dut.
go-PerfPart behar AuxTr-1Sg
`I have to go.'
(29) Hori egin behar dut.
that do-PerfPart behar AuxTr-1Sg
`I have to do that.'
These can be negated with the regular negative particle ez `not', and
show the usual inverted word order. The result is ambiguous, with
the reading chosen according to context.
60 transactions of the philological society 98, 2000
16 The data and description here were supplied by Larry Trask.17 There is a good deal of variation among different dialects in Basque. In the
Western varieties, there is a variant (and preferred) construction, where the nounbehar functions as a noun in the structure, as can be seen from the partitive -ik onbehar, which is the direct object of the verb eduki `have', in (i):
(i) Ez daukat joan beharrik.Neg have-1Sg go-PerfPart behar-Partitive`I don't have to go.'
In the examples in the text, it does not function as a noun.
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
(30) Ez dut joan behar.
Neg AuxTr-1Sg go-PerfPart behar
`I don't have to go' or `I must not go.'
not [must or must [not
(31) Ez dut hori egin behar.
Neg AuxTr-1Sg that do-PerfPart behar
`I don't have to do that' or `I mustn't do that.'
not [must or must [not
According to our proposals then, Basque must have a negation
position lower than the modal element behar. Since the structure
with Adv[neg], as in (26), is quite different, it seems that this is
indeed Pol[neg]. There is no reason, then, to suppose that the LF
scope positions of the functional heads, including Echo and Pol, are
any different in Basque (though the linearisation might be different,
as Laka argues). We are left, however, with an interesting problem
about the PF order, since a non-clitic subject can occur between ez
and the auxiliary: apparently, Pol[neg] is attracted to some head in
the region of Echo.
5.2. Italian
Italian is particularly interesting for us because of the range of
negation strategies available in different dialects (Zanuttini 1997a,
b), which provide potential evidence of the different negation
positions. We consider first Standard Italian.
It can be shown that in Standard Italian, the modals dovere and
potere pattern with respect to negation like the English must and
can.18 Consider the following sentences, where capitalisation indi-
cates the main sentence stress, and the comma marks an intona-
tional break and an optional pause.
(32) Gianni non puoÁ canTAre. not [canGianni not can sing
`Gianni can't sing.'
61cormack and smith ± head movement and negation
18 We are indebted to Vieri Samek-Lodovici for all the data relating to StandardItalian in this section. See also Cormack and Smith (forthcoming) for discussion ofStandard Italian in relation to Cinque's (1999) claims about modals.
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
(33) Gianni non PUOÁ , cantare. not [can`Gianni can't sing.'
(34) Gianni puoÁ non risPONdere. can [notGianni can not answer
`Gianni can not answer.'
(35) Gianni PUOÁ , non rispondere. can [not`Gianni can not answer.'
The analysis here is clear. Non before the main verb takes scope
where it appears, and is either Adv[neg], or Pol[neg] in a lower
clause (see below). Non before the modal is Pol[neg] (or Echo[neg] ),
which has scope over the modal puoÁ in its deontic interpretation.
The `necessity' modal dovere behaves somewhat differently, as we
see in the equivalent examples.
(36) Gianni non deve canTAre. must [notGianni not must sing
`Gianni must not sing.'
(37) Gianni non DEve, cantare. not [must19
Gianni not must sing
`Gianni does not have to sing.'
(38) Gianni deve non risPONdere. must [not`Gianni must not reply.'
(39) Gianni DEve, non rispondere.20 must [not`Gianni must not reply.'
The instances of non before the main verb are interpreted as before,
and (37) parallels (33) in having wide scope negation over the modal.
However, (36) unexpectedly has narrow scope negation with respect
to the modal. The obvious explanation of the difference between
(36) and (37) is that the former arises from Pol[neg], and the latter,
from Echo[neg]. We may now assume that the same is true of the
comparably stressed examples (32) and (33).
62 transactions of the philological society 98, 2000
19 If this is pronounced protracting `DEve', then it may be interpreted as echoic. Inthat case, both the not [must . . . and the must [not. . . scopes are possible.
20 This sentence is said to be quite unnatural, and is probably usable only in veryspecific contexts.
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
Epistemic uses pattern the same way, as the translations for the
clause in (40) indicate.
(40) (If the forecast is right), it shouldn't rain
. . ., non puoÁ PIOvere. not [possiblenot may rain
. . ., non dovrebbe PIOvere. necessary [notnot mustconditional rain
The conclusion, then, is that in Standard Italian, potere `can' is a
Modal2, while dovere `must' is a Modal1, like their English equiva-
lents; and Italian has Pol[neg] and Echo[neg].
Note that the unmarked sentence stress (towards the end of the
sentence) always gives a Pol[neg] reading. Stress on the modal tends
to correlate with the occurrence of Echo[neg], but can, as follows
from the information in footnote 19, still correlate with Pol[neg].
Our claim that what we have in (32) and (36) is indeed Echo[neg] is
reinforced by the fact that the post-verbal particle mica occurs only
with negation which has wide scope over the modal.21 According to
Zanuttini (1997a, b) and others, mica is associated with `presupposi-
tional', that is, plausibly echoic, use.22
It can also be shown that Italian has Adv[neg]. Consider a double
modal sentence like (41):
(41) Maria puoÁ non dovere andare a lavorare oggi.
Maria may not must go to work today
`Maria may not have to go to work today.'
*may [must [notmay[not [must
Since the interpretation of (41) precludes the must [not . . . reading,
the non cannot be Pol[neg]. There is no reason to suppose that it is
63cormack and smith ± head movement and negation
21 Note that where mica is inserted after the modal in (33) or (37), it is mica ratherthan the modal which is stressed: Gianni non deve MIca, cantare. It is also possible tohave mica in sentences like (36): Gianni non deve mica canTAre. Here the interpreta-tion is `it is not singing that he must do'. We take it that this is derived from an LFwith Echo[neg] together with a contrastively focused constituent [cantare]. Mica canalso occur in non-negative questions.
22 Zanuttini (1997a: 7) says that such an item is `associated with the negation of aproposition assumed in the discourse'.
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
Echo[neg], so we assume that it is an Adv[neg] adjoined to the
infinitival phrase, and that this phrase cannot contain Pol[neg].23,24
The Modena dialect of Italian has alternative post-verbal par-
ticles associated with negation: menga is `presuppositional' and briza
`non-presuppositional'.25 In a negative sentence, one or the other
must occur, as in (42), unless there is some other negative element.
(42) a nn iva menga/briza d'andEr a skola
he not had `not' to-go to school
`He doesn't have to go to school.'
In an echoic context, as in (43), only the `presuppositional' menga
may be used.
(43) n ivEt menga/*briza d' andEr a skola
not had-you `not' to go to school
`Shouldn't you go to school?'
We conclude that the `presuppositional' uses are indeed what we
would term echoic.
Even more interesting is Piedmontese (Zanuttini 1997a, b), where
there is no (overt) preverbal particle, and one of the postverbal
particles nen (`non-presuppositional') or pa (`presuppositional')
carries the negative meaning. The examples are from Zanuttini
(1997a: 69±70).
(44) a l'avai giaÁ nen salutami cul di la.
s.cl. s.cl'had already not greeted-me that day there
`Already on that day he had not greeted me.'
(45) a l'eÁ pa giaÁ parti.
s.cl s.cl'is neg already left
`He hasn't already left.'
We are led to speculate that Piedmontese is a language where
64 transactions of the philological society 98, 2000
23 Zanuttini (1996) argues on different grounds that sentential negation isassociated with Tense (i.e. finite tense).
24 The absence of Pol entails that dovere in (41) is not a Modal1, but a main verb,selecting either for a VP or for a non-finite TP.
25 The Modena data were provided by Rita Manzini. Unfortunately, the dialectdoes not use dovere, so confirmatory scope data are unavailable.
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
Pol[neg] and Echo[neg] have distinct phonological realisations: nen
corresponds to Pol[neg] and pa to Echo[neg].
We do not have any data on the scope of modals with respect to
the negation particles in Piedmontese, but the scope of each particle
with respect to giaÁ `already' is as we would predict. Zanuttini (1997a:
67±71, 1997b: 228±231) argues that the merge position of pa is above
that of giaÁ `already', which in turn is above that of nen. We see these
ordering restrictions as a reflex of the fact that giaÁ , like English
already, is a Positive Polarity Item and, hence, cannot fall within the
scope of Pol[neg] nen. It may fall within the scope of Echo[neg], as
in (45), but the Scope Constraint (section 9) will ensure that PF-
Echo[neg] pa would not appear below giaÁ .
We have used these Italian examples simply to show that the
distinction between Echo[neg] and Pol[neg] is supported by data
from other languages. We see no reason to doubt that the rest of the
data can be accounted for using a combination of general and
particular lexical rules, together with soft constraints, but to
demonstrate this is beyond the remit of this paper.
Part two: PF-interpretable positions of heads
6. Introduction
In this part of the paper, we consider not just the LF-positions of the
various heads, but their PF-positions, too. It is not easy to give a
principled account of the relative positions of negation with respect
to auxiliaries and main verbs in English. Baker (1991) went so far as
to argue that it was impossible within `core grammar'. Lasnik (1995)
proposes quite different properties for main verbs and auxiliaries
with respect to their relation to inflection. Roberts (1998) assigns to
auxiliaries (including our Modal2 auxiliaries) the curious property
of consisting of nothing but Formal Features. The solution we
propose, relying as it does on soft constraints, is in a sense outside
core grammar as usually conceived. It is distinct from previous
solutions in three main ways. First, the different behaviour of
auxiliaries and main verbs with respect to negation is regulated
through the uncontroversial devices of categorial and inflectional
features, as is the relevant difference between a negation head and
65cormack and smith ± head movement and negation
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
an adverb. Second, the soft constraints proposed are very natural ±
several have equivalents in other OT analyses ± and the one
responsible for displacement across negation is independently
needed. Third, the flexibility given by the soft constraints appears
to be needed and, as we have argued in Cormack and Smith (1998,
section 11), can account for some surprising data with no extra
apparatus. These data, which limitations of space preclude our
discussing here, involve several gaps in the paradigms of inversion
with n't.26
We begin by considering the theoretical status of PF items in the
grammar, and introducing our `Split Signs' hypothesis (see also
Cormack and Smith 1999).
7. Split Signs
On a lexicalist hypothesis, the lexicon may contain morphologically
complex phonological words. We see no reason to suppose that
these are merged anywhere other than where they are heard.
Minimalist considerations then dictate that the PF part of a sign
is merged where it is heard. If there is apparent movement, this is
because the sign is split, and the PF-interpretable part of the sign is
displaced relative to the LF-interpretable part.27
For example, in (46) the morphophonological word could relates
to the LF interpretable can and past. If it is heard at past, then
there is no point assuming that it is inserted at the position of can,
rather than past. It follows that there is no such thing as head
movement in the usual sense, and that there are no intermediate or
indeed final head traces, nor a Copy Theory of movement.
(46) The dog could not reach the ball.
PF (temporal) order: could not
LF (scope) order: past [not [ can [ . . .
66 transactions of the philological society 98, 2000
26 Lasnik's solution also accounted for some interesting data concerning VPellipsis, but we think there are alternative explanations (see Cormack and Smith1997, especially footnote 43).
27 We do not discuss quantifier scope in this paper, but assume that the sameprinciple holds of quantifier phrases.
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
The discrepancy between LF and PF order in (46) could in principle
be due to the splitting of either the negation or the modal, as in (47).
(47) a. LF [T[past] [ Pol not [ Mod can . . .
PF could not j" jjÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ
b. LF [ Pol not [ Mod can [ X . . .
PF j could notjj "ÿÿÿÿÿÿÿÿÿÿÿÿÿÿ
That it is the modal which is split can be seen from the interaction
with adverbs, as in the examples shown in (48), under the scope
readings given.
(48) a. John can not come home late. Polneg [canb. Sylvia can seldom climb that tree. Polpos [seldom [canc. You could not often bribe officials, then.
Polneg [often [can
All these are consistent with the positioning of PF-can at some
position preceding Pol. In particular, in (48c), the scope orderings
relating to can are disturbed in PF, but not that relating not to a
temporal adverb.
With the Split Signs hypothesis in place, we begin our explanation
of the PF ordering of the various heads by introducing Infl heads,
where these determine the PF inflections on verbs and auxiliaries.
We argue that T is not an Infl head.
8. Infl positions
Consider the array of heads in (49a).
(49) a. Rosa might have been given an armadillo.
b. Aux[may] Aux[have] Aux[be] V[give]
c. T[past] Inflpast Aux[may] Inflbare Aux[have] Inflperf Aux[be]
Inflpass V[give]
We assume, as in Cormack and Smith (1997), following Chomsky
(1957), that each of the LF heads indicated in (49b) must be
67cormack and smith ± head movement and negation
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
associated with an Infl head, as in (49c). We assume that like any
other heads, these Infl must have LF-interpretable content, which
must be determined for each kind of Infl. For example, the auxiliary
may selects for a `bare' inflection, Inflbare. Since this Infl makes no
non-trivial contribution to meaning, the content will be id, the
identity function, so that in (49c), we should have Inflbare[id].
For the pair Aux[be] and Inflpass, we established in Cormack and
Smith (1997) that the passive content lies in the Infl, while the Aux
has the trivial LF-interpretation id. The lexicon provides a complex
morphophonological PF-word given, associated with the LF V[give]
and the LF Inflpass[pass]. All this, as in (50), would enter the
numeration.
(50) LF: {Inflpass[pass], V[give]} PF: {Inflpass, V} [given]
The notation indicates that the LF is complex, consisting of a pair of
categories, Inflpass and V, each associated with a feature giving its
appropriate meaning, pass and give, respectively. The PF-part of
the entry has the phonological feature which is simply indicated here
by the orthography given, associated with the complex category
{Inflpass, V}, i.e. a pair of simple categories. Both the categories are
needed so that a checking relation based on categories can be set up
for each of the categories of the LF part of the sign.
For a pair such as Aux[have] and its associated Inflperf, we
showed that the semantic content `perfect' lay in the Aux, with
the Infl being LF-interpretable simply as the identity function id.
Similarly, we argued that contrary to what is often assumed, the past
tense content lies at T, not at the associated Infl. Since this is going
to be crucial, we offer some independent motivation here.
First, consider sentences like (51), under the natural reading
shown.
(51) Leslie often could not find the cat. often [ not [ could
Above, we suggested that a modal2 raising past Pol[neg] might PF-
merge at T. However, on our assumptions this is inconsistent with
the presence of temporal adverbs such as often in (51) since the
temporal adverb, which must be in the semantic scope of tense,
appears above the inflected modal, putatively in T. With the
introduction of InflTense, the problem disappears. If an Inflpast is
68 transactions of the philological society 98, 2000
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
merged after the temporal adverb, and the PF-word corresponding
to Modal plus Inflpast is PF-merged under Inflpast, we obtain the
correct PF-order.
(52) [T past [Adv often [Inflpast [Pol not [Mod1can
often could not
Merging Infl below the adverb accounts also for the PF order in
(1a), repeated here as (53).
(53) John often snores.
[T past [Adv often [Inflpast [Pol pos [V snoreoften snores é
Our second justification for the T-related Infl comes from serial
verbs, such as the Akan serial in (54) and the English quasi-serial in
(55).
(54) meyEE adwuma memaa Amma.
I-do-PRET work I-give-PRET Amma
`I worked for Amma.' (Byrne 1990, from Schachter 1974)
(55) John ran and bought a newspaper.
(56) John both ran and bought a newspaper.
In serial verb structures, two verbs contribute to the description of
what is seen as a single event (cf. (56), which is non-serial). They are
always interpreted as having the same Tense and Aspect values.
The verbal inflections may be identical, as in (54) and (55), but it
is also possible for the two to differ in inflection. Most commonly in
such a situation, the second verb bears a null `default' inflection, as
with the Nupe example in (57). We notate this with Infldef. In other
paradigms, such as the Akan in (58), the second verb bears a
defective or partial inflection, shown as Infldef-fut.
(57) wu aÁ-du eci gi.
he Inflfut-cook yam Infldef-eat
`He will cook and eat a yam.'
(58) me-ba-fa sekan e-twa.
I-Inflfut-take knife Infldef-fut-cut
`I shall cut with a knife.' (Byrne 1990, from Schachter 1974)
69cormack and smith ± head movement and negation
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
This variation is explicable if both verbs are in the scope of a single
Tense head, which is in a Checking relation with one or more related
Infl heads (Collins 1995; but we envisage that Checking takes place
via percolation). Since the Infl heads are not selected, their LF-
positioning is relatively free. If they bear non-trivial content, then
their position will be fixed according to the requirements of scope.
However, tense-related Infls have trivial id content, and so their
position may (and perhaps must) be determined according to other
constraints, which we will introduce very informally.28
9. Some soft constraints
In this section, we pursue the second half of our theoretical claim
about the position for Merge. In the first part of the paper, we
explored the consequences of the assumption that a head (or, as we
now see it, the LF-interpretable part of a head) is merged in its scope
position. We now show how we substantiate the idea introduced in
section 7 above, that the PF-interpretable part of a head is also
merged at the position at which it is interpreted. The relation
between the two parts of a head is regulated by Checking. We
suggest that the position of the Infls and the PF items is determined
by a system of soft constraints, along the lines of Optimality
Theory.29
The soft constraints as we understand them fulfil two tasks which
arise when an LoT (Language of Thought) expression is to be linked
to a paired LF-interpretable and PF-interpretable representation.
Consider first the LF representation. We assume that the LF
respects all the structure (i.e. at least scope information and s-
selection) given by the LoT expression. However, LF may, and
usually will, contain elements not present in the LoT expression ± in
particular, elements with an id interpretation, such as some Infl
70 transactions of the philological society 98, 2000
28 We do not claim that T is associated with an Infl in every language. In particular,French may have no InflT, but may rather have V to T movement as argued byPollock (1989). If French has no Infl with id content, this would explain the absenceof quasi-serials.
29 For a general introduction to Optimality Theory in syntax see Pesetsky (1997);for particular proposals concerning negation and inversion, see Grimshaw (1997) andBresnan (1997).
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
or auxiliaries such as be, or do. As an example, consider Lucinda
is greedy. Since greedy is a predicate in LoT, it can be predicated
directly of Lucinda. However, English requires an Infl to be
checked by T, and a verb or auxiliary to support this Infl at
PF. Thus, at LF, two extra elements must be present. If the
meaning is to be preserved, these must have id interpretation at
the LF/LoT interface. In this instance, the elements that lead to a
well-formed LF/PF pair are InflT and be. The LF position of such
elements is not fixed by semantics, though it will be at least partly
determined by syntax through the mechanisms of c-selection and
Checking theory. One of the tasks of the soft constraints, then, is to
reduce the underdetermination in the grammar by selecting an
optimal LF position for an id element if it is not already fully
determined.
The other main task of the soft constraints is to determine where
the PF form related to a set of LF heads is to appear on the tree. We
see the constraints we introduce as taking up the slack introduced by
the indeterminacies in structure resulting from the demands of
lexicalisation. For example, a complex morphophonological word
ran, associated with Inflpast and Vrun might be merged anywhere,
but processing will be difficult if PF and LF are not as closely and
tightly related as possible.
For example, omitting irrelevant detail, the structure we require
for (55) is as in (59), where the dotted lines show a checking relation
required by a split sign, and the other lines, the checking of Infls
by T.
(59)
John
[JOHN] T[PAST] ∅
[Inflpast InflpastVRUN [AND VBUY]] [A NEWSPAPER]
ran and bought a newspaper
We will begin exemplifying our soft constraints with respect to
(59) above. For (59), the LoT input supplies all the LF entries and
structure, except for the id elements, the Infls. The soft constraints
have to determine the LF positions of the two id Infls, and have to
determine the PF positions of each of the PF items.
71cormack and smith ± head movement and negation
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
For the purposes of this paper, we take the candidate set of
structures over which the soft constraints operate to be the initial
LoT representation augmented by any number of id elements which
satisfy the obligatory rules of syntax. Surplus id elements will be
eliminated by an overarching pragmatically motivated constraint
rejecting representations in which processing an element leads to no
reward. We will not justify our constraints here, but simply indicate
what we need.30
Although we have argued that signs may be split, the normal,
default, situation is that a sign is unsplit.
(60) Constraint 1: Signs are unsplit.
Most of the signs in (59) are indeed unsplit. However, in the case of
run, the lexicon provides no PF associated solely with this LF item.
Instead, the lexicon provides a morphophonologically complex PF-
word ran, associated not only with the LF-item V[run], but with the
LF-item Inflpast[id], as well (and similarly for buy). It follows that
either the verb or the Infl must be split. That it is the verb is
determined by constraint 2:
(61) Constraint 2: Infl is unsplit.
Split parts will be kept near to each other by constraint 3.
(62) Constraint 3: checker and checkee are adjacent (i.e. there is no
intervening branching node in the structure).31
We see now that each of the Infls in (59) must be adjacent to its
associated verb. Which side of the verb the Infl goes is determined
by constraint 4:
(63) Constraint 4: Raise PF-part.
The analysis of the examples in (57) and (58) is similar to that of
(59), except that here, there is a choice of Infl. The syntax will have
ensured that T checks a full InflTense in each of the candidate
structures. Some sort of phonological laziness constraint (cf.
72 transactions of the philological society 98, 2000
30 More detail is given in Cormack and Smith (1998).31 Null nodes are not an issue, since every terminal node has at least id LF-
interpretable content.
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
Pesetsky 1997) will prefer candidates with Infldef, and pragmatic
processing convenience demands that the informative Infl is placed
before the default/defective one.32 These two constraints are spelled
out in (64) and (65).
(64) Constraint 5: Use Infldef in preference to full Infl.
(65) Constraint 6: Infl-precedence constraint. (When more than one
Infl is in the Checking domain of the same head, Infldef must
follow a full Infl.)
In order to account for (51), represented in two alternative ways
in (66), we need another constraint.
(66) a. [T past [Adv often [Inflpast [Pol not [Mod1can
often could not
b. *[T past [Inflpast [Adv often [Pol not [Mod1can
could often not
Merging the Infl before the adverb as in (66b) would satisfy the
Adjacency constraint with respect to the Checking of the Infl by T.
However, the LF scope order of adverb and modal would not be
paralleled by the linear order of the two at PF. We need a further
Faithfulness constraint, which requires that PF linear order respects
LF scope order ± guaranteeing that we get the correct (66a).
(67) Constraint 7: PF order respects LF scope.
One final constraint will be needed, to ensure subject-aux inversion
when the PF-content of C is affixal.
(68) Constraint 8: C is unsplit.
A partial ranking of the soft constraints we have postulated is
given in (69).
(69) a. Infl-precedence � Scope � Adjacency
b. Infl-precedence � Raise
73cormack and smith ± head movement and negation
32 The motivation here suggests a PF-based ordering, but the constraint appears tohave been grammaticised, so that the ordering is based on scope or c-command.Whatever the motivation, some constraint is needed to reduce the underdetermina-tion of the position of the full inflection.
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
c. C � Infl � Unsplit � Adjacency
d. Infldef
This ranking is sufficient for our purposes here, at least for English.
The constraints at the left hand ends of each series are inviolable.
This inviolabilty must be understood in relation to the candidate set,
which already excludes syntactically non-viable alternatives.
10. Negation as V
A perennial problem in English has been why a Modal2 like could, or
an auxiliary, which LF-merges after Pol, must PF-merge at an Infl
preceding Pol, whereas a verb is excluded from doing this, leading to
do-support.
(70) a. *John not swims. b. *John swims not.
c. John does not swim.
(71) a. *John not can swim. b. John can not swim.
Our hypothesis is that Pol, although a functional head, has the
category V (as in some of the languages exemplified in Payne 1985).
As such it must bear inflection features at PF. Let us further assume
that any Infl associated with Pol[neg] is Infldef. This Infl will, along
with the Infl associated with the finite modal or Aux or verb, be
checked by T. When more than one Infl is in the Checking domain
of the same head, the full Infl must appear first, in accordance with
the Infl-precedence constraint.
The examples in (71) now appear as in (72), where the Infl-
precedence constraint favours (72b) over (72a). The constraint is
violated in (73), which corresponds to (70a), John not swims.
(72) a. *John T[pres] Infldef Pol[not] Inflpres Modal2[can] swim
not can
b. John T[pres] Inflpres Infldef Pol[not] Modal2[can] swim
can not
(73) *John T[pres] Infldef Pol[not] Inflpres V[swim]
not swims
The idea that Pol has a categorial feature [V] leads to a simple
74 transactions of the philological society 98, 2000
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
account of the other salient facts about Pol. Consider (74), John
swims not, with the detail of the Checking in (75):
(74) *John T[pres] Inflpres Infldef Pol[not] V[swim]
swims not
(75) Inflpres
{Infl, V}swimsInfldef
{Infl,Pol, V}notPol[V][NOT] V[SWIM]*
Checking
We suppose that there is a category-based Minimality constraint on
Checking. LF-V[swim] is to check for PF-V[swim] at the nearest
commanding PF-V. But the nearest PF including the category V is
not, so the Checking fails. We assume that raising of Modal2 or Aux
past Pol[not], as in (70c) and (71b), is possible because both have a
category, Aux, which is not a subset of the category of PF-not.
(76) Inflpres
{Infl,Aux}canInfldef
{Infl,Pol, V}notPol[V][NOT] Aux[CAN]
Checking: short movement’
There are two other important points about (76). First, if there is
an adverb whose LF-position is between Pol and Aux, short move-
ment Checking will cross this. This accounts for the readings of
example (48c), repeated here as (77).
(77) You could not often bribe officials, then.
Polnot [often [can
(78) You can't often bribe officials nowadays.
Polnot [often [can
Example (78), with scope as in (77), contains the irregular form
can't. This must be given in the lexicon. However, as can be seen
from the associated LF scope information, can+not is not a
constituent of the LF, entailing that the PF item can't is not
paired in the lexicon with a single `meaning' (i.e. LF-interpretable
item). Under our interpretation of Split Signs and Merge, this is
unproblematic: the PF-form can't is simply associated with the
two LF meanings can and not by Checking; how these turn up in
75cormack and smith ± head movement and negation
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
the LF representation is not stipulated in the semantics, but arises
only from selection possibilities.
The other point is that the short movement is mainly determined
by the [V]-feature on Pol and the Infl-Precedence constraint. It has
nothing to do with T as such. Two predictions follow. First, there
should be short movement effects even with Pol[pos]. This is correct,
as can be seen from (48b), repeated here as (79), under the reading
given.
(79) Sylvia can seldom climb that tree. Polpos [seldom [can
Second, we predict that identical effects should be seen even when
Modal1 is present, in front of Pol. This is correct. Consider (80):
(80) a. The patient must not have eaten. (deontic)
b. The patient must have not eaten.
c. must [not [perfect
In the context of some sort of medical test, there is an interpretation
for each of these `it is necessary that it is not the case that, at some
time in the (relevant) past, the patient ate'. That is, both these PF-
orders are compatible with the LF-order must [not [perfect.
Cormack and Smith (1997) showed that the perfect content resides
in the auxiliary have, rather than in the associated Infl -en. This
means that PF-have is displaced over LF-not in (80b). We can see
that the interpretation requires Pol[neg], because Adv[neg] would
give rise to the improbable interpretation `It is necessary that there is
some time in the (relevant) past at which the patient did not eat'.
However, because of uncertainties associated with the interpreta-
tion of the perfect, this evidence is not unequivocal. Clearer evidence
can be obtained from a dialect where modals need not be tensed.
Scottish English has untensed modals, and such sentences as (81a)
and (81b) occur.
(81) a. He must no can do it.
b. He must can no do it.
c. must (epistemic) ± not ± can (deontic)
Brown (1991: 98) states that both these may have the interpretation
`I conclude he is unable to do it'. That is, (81b) may be interpreted as
having the scope order in (81c). If we assume that the deontic can is
76 transactions of the philological society 98, 2000
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
merged below Pol[neg] in Scottish English, and must, above, we can
see that for (81b), PF-can is displaced across Pol[neg] to an Infl
position lower than that of Modal1.
The account for the short-movement version is entirely parallel to
what we saw earlier; the only difference is that the relevant full Infl is
Inflbare, required by Modal1, rather than InflTense, demanded by T.
The optimal array of heads for (81b), `he must can no do it', is as in
(82), where the PF-items are under the heads as shown.
(82) T[pres] InflPres
must
�M1[must] Inflbare
can
�Infldef
no
�Pol[neg] M2[can] . . .
However, when Modal1 is present, short movement is optional, as
we see from the relation between (81a) and (81c). This is what we
expect if the negation in (81a) is Adv[NEG].
We conclude this section with an example showing how the soft
constraints choose between four candidates appropriate for the LoT
structure in (83), where the negation is Pol[neg].
(83) john [past [neg [leave] ]]
In (84), we show four from among the fifty-odd possible candidates,
one leading to John not left, and three (only one of which is optimal)
leading to John did not leave. Notice that all contain the extra idelements Inflpast (demanded for Checking by T[past] ) and Infldef
(demanded for Checking by the PF associated with Pol[neg] ). The
last three also contain an extra id element, the auxiliary do, which in
turn demands the id element Inflbare. No candidate leading to John
left not appears, because all such structures fail Checking (because
of the [V] feature on Pol).
(84) A: T[past] Infldef[id] Pol-V[neg] Inflpast[id] V[leave]
é not left
B: T[past] Inflpast[id] Infldef[id] Pol-V[neg] Auxdo[id] Inflbare[id] [leave]
é did not leave
C: T[past] Inflpast[id] Pol-V[neg] Infldef[id] Auxdo[id] Inflbare[id] [leave]
é did not leave
D: T[past] Inflpast[id] Pol-V[neg] Infldef[id] Auxdo[id] Inflbare[id] [leave]
é did not leave
77cormack and smith ± head movement and negation
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
For convenience, the ranking of the constraints in (69) is given again
here. Since in these candidates, there is no possibility of violating
Infldef or C, these are omitted.
(85) a. Infl-precedence � Scope � Adjacency
b. Infl-precedence � Raise
c. Infl � Unsplit � Adjacency
The arrangement in Table 2 below reflects the partial orderings of
(85) in the three blocks, contrary to the usual practice (and leading
to some repetition). The blocks have equal status. We begin with the
highest ranking constraint in each block. Each violation of a
constraint is marked with `*' in the relevant cell. The sign ` '
marks a cell where a relatively high-ranking constraint is violated
more times by some candidates than by others. Such candidates are
ruled out. The next lower ranking constraint in each block is
considered next, and so on, until all but one candidate has been
eliminated. The remaining, and optimal, candidate is marked with
`('. Of the four candidates selected, the soft constraints correctly
eliminate A, and, of the remainder, select B, which is in fact the
overall optimal candidate.
We realise that our use of soft constraints raises questions we have
not answered here.
11. Pol[pos] and verbs
At this point, we need to consider how sentences like (86) are
permitted.
78 transactions of the philological society 98, 2000
Table 2
Infl-P Scope Adjac Infl-P Raise Infl Unsplit Adjac
A * *33 * ** *B ( ** ** *** **C ** ** * *** **D ** ** * *** **
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
(86) John eats fish.
It is clear that Pol[pos] must have some property which Pol[neg]
does not have, since otherwise we would expect do-support with
Pol[pos] when no auxiliary or modal is present. The crucial property
forcing an auxiliary to precede Pol[neg] is the absence of any PF
related to both Pol[neg] and InflT. To permit (86), then, all we need
is the assumption that there is a PF relating Pol[pos] and InflT. The
required configuration will include the LF- and PF-parts as shown
in (87), with the Checking shown by dotted lines.
(87) [Inflpres] Pol[V][POS]
{Inflpres, Pol[V]}∅
[Inflpres] V[EAT]...
{Inflpres, V}eats
Although there is `short movement' of V `eat', it is extremely local. If
an adverb such as often intervened between Pol[pos] and V[eat], the
Scope Constraint in (67) would ensure that the Inflpres relating to the
verb was preferentially placed adjacent to the verb. This will rule out
a structure giving rise to the PF order in (88).34
(88) John é eats often fish.
The question still arises as to what happens if, in a sentence
without an auxiliary, Infldef is selected for Pol[pos]. We know this is
an option because of sentences like (89), with the interpretation as
shown, which was discussed in section 8 (example (48) ) and again in
section 10 (example (79) ).
(89) Sylvia can seldom climb that tree. Polpos [seldom [can
If there is no LoT auxiliary, the structure will not be viable without
do-support; and do-support is certainly needed for questions where
an auxiliary is not otherwise supplied. The simplest solution would
seem to be to claim that sentences like (90) (pace Baker 1991: 400)
79cormack and smith ± head movement and negation
33 In counting Adjacency violations, unsplit signs are counted as conforming to theAdjacency constraint.
34 The order John eats fish often will be derived from a V0 adjoined adverb,assuming movement of a final verb across the object to AgrO (Larson 1991, Cormack1999).
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
and (91) (pace Pollock 1989: 400, Baker 1991: 398 and Warner 1993:
8), are grammatical (irrespective of stress), but are pragmatically
unacceptable without any extra relevance being obtained from
processing the spurious do.
(90) John does snore.
(91) Lucy does be complaining.
Relevance can arise if do is stressed, with the implication that focus
is attached to one of its associated LF-parts. Focus on the do
itself will not be sufficiently relevant, but focus on an associated
Echo[pos] would be (see the next section), giving rise to a positive
affirmation interpretation in (90). (91) would still be unacceptable,
since the same effect could be achieved with stress on be. If snore is
stressed in (90), we might have contrastive focus on the whole
Pol[pos]P, as in a natural reading of (92).
(92) John [doesn't kick me out of bed], but he [does snore].
There are also dialects in south-west England where unstressed do
seems to occur freely in examples like (90) and (91). We assume that
there is some pragmatically mediated contrast involved (see Ihalai-
nen 1991).35
12. PF-Echo
Given the soft constraints proposed, the PF position of Echo[neg]
can be explained very simply. We assume that Echo[neg], like
Pol[neg], has the categorial feature [V], so that it needs inflection,
and that the inflection it can bear is that associated with Infldef.
Because of the Infl-precedence Constraint in (65), this has the
immediate consequence that the PF word not associated with Infldef
and Echo, which is realised under the Infldef head, must follow the
InflT, where the tensed Modal or Aux is realised, just as the PF
associated with Pol would. We show optimal structures for the
relevant portions of echoic John may not leave and I should not eat
80 transactions of the philological society 98, 2000
35 Additional evidence that do is freely available in principle comes from acquisi-tion studies; see Bohnacker (1998).
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
more vegetables in (93) and (94). The T-related Infls will be
positioned as shown, with the upper Infldef being associated with
Echo, according to the dictates of the soft constraints.
(93) [Echo, V]
NOT
C
ID
∅
T
PRES
∅
InflT
ID
can
Infldef
ID
not
Infldef
ID
∅
[Pol, V]
POS/ID
Modal2/Aux...
CAN
(94) [Echo, V]
NOT
C
ID
∅
T
PRES
∅
InflT
ID
should
Infldef
ID
not
Modal1
SHOULD
Inflbare
ID
eat
Infldef
ID
∅
[Pol, V]
POS/ID
V
EAT
The simplicity of this account supports the architecture we have
postulated.
For the n0t version of Echo[neg], we postulate that the lexicon
contains the relevant PF forms of category {Echo[neg], InflT, Aux/
Modal}, and to accommodate inversion, PF forms of category
{Cinv, Echo[neg], InflT, Aux/Modal}. We illustrate with inversion
for Shouldn't John go?
(95) [Echo, V]
NOT
CForce
Q
Cinv
ID
shouldn’t
T
PRES
InflT
ID
∅
Infldef
ID
Modal1
SHOULD
Inflbare
ID
go
Infldef
ID
∅
[Pol, V]
POS/ID
V
GO
PF-Echo[pos] might be assumed to have the same properties as
PF-Echo[neg]. However, since the equivalent of the not form will be
phonologically null, it would be likely to become affixal. The affixal
equivalent of the n0t form will show no change from PF-{InflT, Aux/
Modal}. If it is required to focus Echo[pos], as in (24), repeated here
as (96), then the affixal form must be available, so that stress may
fall on a non-null PF form ± PF{Inflpast, Aux, Echo[neg]} did, here.
(96) The policeman did bite the dog.
81cormack and smith ± head movement and negation
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
Horn (1989: 461) remarks that if there is a wider scope negation as
well as the ordinary sentential negation, it is surprising that the
negation morpheme for the former never turns up in the expected
position, i.e. at the periphery of the clause (citing Dahl's sample of
240 languages). His observation leads us to infer that Pol[neg] and
Echo[neg] must bear inflectional features in the majority of lan-
guages ± hence being forced to occur in the vicinity of some T, as in
English.36
13. Conclusion
Given our assumptions about the architecture of UG, we think there
is clear evidence for an Echo[neg] position in English, as well as
Pol[neg] and Adv[neg]. Italian, Catalan and the typologically distant
Basque support the distinctions and hypotheses we have put forward.
We have postulated a category-based Checking theory, with V, Aux,
and Adverb having distinct categories, and Pol and Echo bearing a
[V] feature. Our hypothesis concerning Infl features associated with
Pol and Echo, together with the set of soft constraints we have
postulated, provide an explanation for the PF-distribution of
negative elements, including the near-identity of the PF-distribution
of Pol[neg] and Echo[neg], despite their different LF-positions. We
hope that work on other languages will provide corroborating
evidence for the required positions of the functional heads.
Department of Phonetics and Linguistics,
University College London,
London WC1E 6BT
E-mail: [email protected]
References
Baker, C. L., 1991. `The syntax of English not: the limits of core grammar', LinguisticInquiry 22, 387±429.
82 transactions of the philological society 98, 2000
36 Even if neg elements do not occur in the vicinity of T, this may be due to variedheadedness, or alternative orderings of the soft constraints.
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
Bohnacker, Ute, 1998. `Children, tense and auxiliary do', MS., Institutionen foÈrNordiska SpraÊk, University of Lund.
Bresnan, Joan, 1997. `Optimal syntax' MS., Stanford University.Brown, Keith, 1991. `Double modals in Hawick Scots', in Peter Trudgill and
J. K. Chambers (eds.), Dialects of English, London: Longman, 74±103.Byrne, F., 1990. `Tense marking in serial structures', in Brian D. Joseph and Arnold
M. Zwicky (eds.), When Verbs Collide: Papers from the 1990 Mini-Conference on
Serial Verbs, Ohio State University Working Papers in Linguistics 39, Department
of Linguistics, OSU, Columbus OH, 149±177.Carston, Robyn, 1994. `Metalinguistic negation and echoic use', in John Harris
(ed.), Working Papers in Linguistics 6, Phonetics and Linguistics, UniversityCollege London, 321±339.
Cattell, Ray, 1973. `Negative transportation and tag questions', Language 49, 612±
639.Chomsky, Noam, 1955, published 1975. The Logical Structure of Linguistic Theory,
New York: Plenum.Chomsky, Noam, 1957. Syntactic Structures, The Hague: Mouton.Chomsky, Noam, 1995. The Minimalist Program, Cambridge MA: MIT Press.Cinque, Guglielmo, 1999. Adverbs and Functional Heads, Oxford: Oxford University
Press.Collins, Chris, 1995. `Serial verb constructions and the theory of multiple feature
checking', MS. Cornell University.Cormack, Annabel, 1989, published 1998. Definitions: Implications for Syntax,
Semantics and the Language of Thought, New York: Garland.Cormack, Annabel, 1999. `Without specifiers', in David Adger, Susan Pintzuk,
Bernadette Plunkett and George Tsoulas (eds.), Specifiers: Minimalist Approaches,Oxford: Oxford University Press, 46±68.
Cormack, Annabel and Smith, Neil, 1994. `Serial Verbs', in John Harris (ed.),
Working Papers in Linguistics 6, Phonetics and Linguistics, University CollegeLondon, 63±88.
Cormack, Annabel and Smith, Neil, 1996. `Checking theory: features, functionalheads and checking-parameters' in Phillip Backley and John Harris (eds.), Working
Papers in Linguistics 8, Phonetics and Linguistics, University College London, 243±281.
Cormack, Annabel and Smith, Neil, 1997. `Checking features and split signs', in
Phillip Backley and John Harris (eds.), Working Papers in Linguistics 9, Phoneticsand Linguistics, University College London, 223±252.
Cormack, Annabel and Smith, Neil, 1998. `Negation, polarity and verb movement',in John Harris and Corinne Iten (eds.), Working Papers in Linguistics 10, Phonetics
and Linguistics, University College London, 285±322.Cormack, Annabel and Smith, Neil, 1999. `Where is a sign merged?' Glot Inter-
national 4.6, 21.Cormack, Annabel and Smith, Neil, forthcoming. `Modals and negation in
English', in Sjef Barbiers and Frits Beukema (eds.), Modality in Generative
Grammar, Amsterdam: John Benjamins.Ernst, Thomas, 1992. `The phrase structure of English negation', The Linguistic
Review 9, 109±144.Grimshaw, Jane, 1997. `Projection, heads and optimality', Linguistic Inquiry 28, 373±
422.
83cormack and smith ± head movement and negation
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
Haan, Ferdinand de, 1997. The Interaction of Modality and Negation: a TypologicalStudy, New York: Garland.
Haegeman, Liliane, 1995. The Syntax of Negation, Cambridge: Cambridge Uni-versity Press.
Henry, Alison, 1992. `Infinitives in a for-to dialect', Natural Language and LinguisticTheory 10, 279±301.
Horn, Laurence R., 1989. A Natural History of Negation, Chicago: ChicagoUniversity Press.
Ihalainen, Ossi, 1991. `Periphrastic do in affirmative sentences in the dialect of EastSomerset', in Peter Trudgill and J. K. Chambers (eds.), Dialects of English,London: Longman, 148±160.
Klima, Edward S., 1964. `Negation in English', in Jerry A. Fodor and Jerrold J.Katz (eds.), The Structure of Language: Readings in the Philosophy of Language,Englewood Cliffs: Prentice Hall, 246±323.
Ladd, D. Robert, 1981. `A first look at the semantics and pragmatics of negativequestions and tag questions', CLS 17, 164±171.
Laka, Itziar, 1994. On the Syntax of Negation, New York: Garland.Larson, Richard K., 1991. `Some issues in verb serialisation', in Claire Lefebvre
(ed.), Serial Verbs: Grammatical, Comparative and Cognitive Approaches, Amster-dam: John Benjamins, 185±210.
Lasnik, Howard, 1995. `Verbal morphology: Syntactic Structures meets the Minim-alist Program', in Hector Campos and Paula Kempchinsky (eds.), Evolution andRevolution in Linguistic Theory: Essays in Honor of Carlos Otero, Washington,D.C.: Georgetown University Press, 251±275.
Lechner, Winfried, 1998. `Two kinds of reconstruction', Studia Linguistica 52, 276±310.
McCloskey, James, 1996. `On the scope of verb movement in Irish', NaturalLanguage and Linguistic Theory 14, 47±104.
Palmer, Frank R., 1990. Modality and the English Modals, London: Longman.Payne, John R., 1985. `Negation', in Timothy Shopen (ed.), Language Typology and
Syntactic Description, Cambridge: Cambridge University Press, 197±242.Pesetsky, David, 1997. `OT and syntax: movement and pronunciation', in Diane
Archangeli and D. Terrence Langendoen (eds.), Optimality Theory: an Overview,Oxford: Blackwell, 134±170.
Picallo, M. Carme, 1990. `Modal Verbs in Catalan', Natural Language and Lin-guistic Theory 8, 285±312.
Pollock, Jean-Yves, 1989. `Verb movement, Universal Grammar and the structureof IP', Linguistic Inquiry 20, 365±424.
Pullum, Geoffrey K., 1976. `The Duke of York gambit', Journal of Linguistics 12,83±102.
Quirk, Randolph, Greenbaum, Sydney, Leech, Geoffrey and Svartvik, Jan,1972. A Grammar of Contemporary English, London: Longman.
Rizzi, Luigi, 1997. `The fine structure of the left periphery', in Liliane Haegeman(ed.), Elements of Grammar: a Handbook of Generative Syntax, Dordrecht: Kluwer,281±337.
Roberts Ian, 1998. `Have/be Raising, Move F, and Procrastinate', Linguistic Inquiry29, 113±125.
Schachter, Paul, 1974. `A non-transformational account of serial verbs', Studies inAfrican Linguistics supplement 5, 253±270.
Stroik, Thomas, 1990. `Adverbs as V-sisters', Linguistic Inquiry 21, 654±661.
84 transactions of the philological society 98, 2000
c:/Joshua/trps98-1/cormack.3d ± 28/3/0 ± 15:31 ± disk/np
Warner, Anthony R., 1993. English Auxiliaries, Cambridge: Cambridge UniversityPress.
Williams, Edwin, 1994. Thematic Structure in Syntax, Cambridge MA: MIT Press.Wilson, Deirdre and Sperber, Dan, 1988. `Mood and non-declarative sentences', in
J. Dancy, J. M. E. Moravcsik, and C. C. W. Taylor (eds.), Human Agency:Language, Duty and Value: Philosophical Essays in Honor of J. O. Urmson,Stanford University Press, 70±101.
Zanuttini, Raffaella, 1996. `On the relevance of tense for sentential negation', inAdriana Belletti and Luigi Rizzi (eds.), Parameters and Functional Heads: Essays inComparative Syntax, Oxford: Oxford University Press, 181±207.
Zanuttini Raffaella, 1997a. Negation and Clausal Structure: a Comparative Studyof Romance Languages, Oxford: Oxford University Press.
Zanuttini, Raffaella, 1997b. `Negation and verb movement', in Liliane Haegeman(ed.), The New Comparative Syntax, London: Longman, 214±245.
85cormack and smith ± head movement and negation