A riddle on negation in Haitian1

MICHEL DEGRAFF

Abstract
As its phonetic resemblance with French pas suggests, Haitian Creole pa marks
sentential negation, like French pas. Yet, this paper establishes a phrase-structural
distinction between pa and pas, their cognation notwithstanding: I argue that
Haitian pa heads NegP while French pas is in Spec of NegP. In so doing, I explore
the syntax and semantics ofsentential negation in Haitian, especially the structural
basis of negative concord in the presence of pa, compared with the double
(cancelled) negation in the presence of pas in Standard French. I then explore the
implications of my analysis of Haitian pa vis-a-vis the syntax of predication. I
conclude with a sampling of the diachronic puzzle posed by pa, as Haitian is
compared to two of its source languages.

Haitian Creole emerged in the 17th century primarily from the contact between
French and a few West-African languages. This paper can be motivated from

I. I wrote this paper in room 1406 at the CUNY Graduate Center, while on an exciting
post-doctoral appointment. Over the year 1992-1993, this room has seen me grow as a linguist,
and I thank Richard Kayne and uncountable CUNY people for making it all so very special. I
now have, to cherish, a roomful of memories.
For help toward solving the riddle that is the topic of this paper, I thank Beatrice
Santorini, Bill Stewart, Claire Lefebvre, Enoh Titilayo Ebong, Gillian Sankoff, Jean Nicolas,
Julie Auger, John Lumsden, Liliane Haegeman, Maxime da Cruz, Michael Hegarty, Mitch
Marcus, Pieter Muysken, Raffaella Zanuttini, Richard Kayne, Ronel Perrault, Rose-Marie
Dechaine, Sabine Iatridou, Salikoko Mufwene, Tonjes Veenstra, Tony Kroch, Victor Manfredi,
Viviane Deprez, two anonymous Probus reviewers, and the wonderful participants in the meeting
of the Society of Caribbean Linguistics in Barbados, in the Going Romance symposium in
Utrecht and in colloquia at CUNY, Georgetown University and UMass Amherst. I am more than
grateful to Yves Dejean, of Haiti, for extensive and illuminating written comments (dated
6/21193) and intense telephone debates - Jv monche, mesi anpil!

Probus 5 (1993), 63-93 0921 - 44 71193/0005-0063

© Walter de Gruyter
64 M. DeGraff

two perspectives. On the one hand, I provide an analysis for one aspect of
Haitian Creole syntax, namely negation. On the other hand, I use this analysis
in comparing Haitian Creole with French, the language from which Haitian
Creole derives the phonetic shapes of most of its morphemes. I also take a brief
look at negation in Fon, a Kwa language spoken mostly in Benin, West Africa.
Fon, along with a few Kwa neighbors, played a key role in the genesis of
Haitian Creole. Altogether, dialects of Fon (and of other Kwa languages) and of
French are perhaps the most influential languages implicated in the development
of Haitian Creole. Interestingly, with respect to negation, Haitian Creole seems
different from both (standard) French and Fon.
Because of the audible parallel, the comparison between Haitian Creole and
French is the most alluring. As its phonetic resemblance with the French mor-
pheme pas suggests, pa in Haitian Creole - like French pas- marks sentential
negation. In what follows, I will establish a phrase-structural distinction between
Haitian Creole pa and French pas, their cognation notwithstanding.
Specifically, I argue that Haitian Creole pa heads NegP while French pas is
in the specifier of NegP. In order to derive this distinction, I develop the
specifics of the syntax and semantics of Haitian Creole sentential negation,
especially the phenomenon of negative concord in presence of pa. I draw
relevant comparisons with (standard) French pas, which blocks negative concord
and induces double (cancelled) negation. I then explore consequences of my
analysis of Haitian Creole pa for one other area of the grammar - predication.
I conclude by considering the diachronic puzzle posed by the properties of
Haitian Creole pa.

1. Pa and pas in their surface strings

It is quite straightforward to characterize the position of Haitian Creole pa in the

surface string: when expressing sentential negation, pa generally precedes the
sequence of tense, mood and aspect (TMA) markers and main verb. 2 This is
shown in (la). Typically, it is ungrammatical for pa to follow a TMA marker or
a main verb occurring in its clause; see (lb)-(ld). 3.4

2. In DeGraff (1992d, e), I argue that TMA markers should be syntactically analyzed as verbs.
3. The following abbreviations are used in the glosses and in the examples:
ANT anterior PROG progressive
FUT future SG singular
IRREAL irrealis PL plural
NUM number REfi reflexive
NEG negation TMA tense-mood-aspect
#pause (comma intonation)
0 phonetically null element
 A riddle on negation in Haitian 65

(1) a. Jan pa t::_ av- ale nan mache. (Haitian Creole)

Jan NEG ANT IRREAL go in market
'Jan would not have gone to the market.'
b. *Jan if pa (ill0 ale nan mac he.
Jan ANT NEG !RREAL go in market
c. *Jan (£::) ava pa ale nan mac he.
Jan ANT IRREAL NEG go in market
d. *Jan(£::) (av-) ale pa nan mache.
Jan ANT IRREAL go NEG in market
Compare Haitian Creole pa in (1) with French pas and English not in (2). Both
pas and not may occur after a finite verb. Also, in English, not can follow a
modal, as in the translation of (2a). 5
(2) a. Jean (ne) serait pas alle au marche. (French)
Jean (ne) would-be pas gone to-the market
'John would not have gone to the market.'
b. Jean (n ') etait pas alle au marche.
Jean (ne) was pas gone to-the market
'John has not gone to the market.'
c. Jean (n ') ira pas au cmema.
Jean (n ') go+FUT pas at-the movies
'John will not go to the movies.'
The sentences in (2) show that, in French and English, the sentential negation
markers pas and not can interrupt the sequence composed of one or more
auxiliaries and of the main verb. This is not possible with Haitian Creole pa, as
witnessed by (1).

4. Apparent counter-examples to this generalization are constructions like (i) where pa occurs
between the epistemic modal ka 'might' and the tense marker te 'ANT':
(i) Jan ka pa te vini.
Jan might NEG ANT come
'Jan might not have come.'
However, Magloire-Holly (1982) analyzes such modals as "EQUI-verbs" subcategorizing for
a clausal complement. (DeGraff 1992e reinterprets Magloire-Holly's arguments within a control
structure with an embedded PRO subject.) Such analyses defuse the counter-example.
(ii) is another potential counter-example:
(ii) Jan pa -p pa vini.
Jan NEG IRREAL NEG come
'Jan wouldn't (won't) not come.'
However, it can be argued that the second negation in (ii) only takes scope over the VP headed
by vini 'come', not over the whole clause; see the English translation and French Jean peut ne
pas venir 'Jean might not come' (I thank Yves Dejean for data and discussion).
5. All characters used in this paper's examples are fictional. Any resemblance with living, dead or
legendary figures is purely accidental.
66 M. DeGraff

Does the contrast between (1) and (2) constitute sufficient evidence to posit
a phrase-structural distinction between Haitian Creole pa and French pas?
Probably not. Within the principles-and-parameters framework, (1) and (2) are
accounted for by possibility of verb movement in French and English (as per
Emonds's 1978 and Pollock's 1989 accounts) and by impossibility of verb
movement in Haitian Creole. At D-structure in all three languages, the negation
markers pa, pas and not would govern and precede the VP node that dominates
the verbal sequence. However, it is only in French and English that the finite
auxiliary verb raises at S-structure out of its VP, to the left of the negation
markers pas and not; see (8) below for a schematization of verb raising in
French. The absence of verb raising in Haitian Creole sets (1) apart from (2).
In Pollock's theory, verb movement in French derives in syntax the inflection-
al morphology of the tensed verb: the verb stem undergoes cyclic-successive
head movement to various inflectional heads in order to collect its inflectional
suffixes. Given that Haitian Creole has virtually no inflectional morphology, it
is not surprising that the language has no verb movement: in Haitian Creole,
there is no tense affixes to be collected and the verb stays in place. Recall that
Haitian Creole tense-mood-aspect markers are independent morphemes which
precede the semantically main verb; see (la).
Given the above verb raising hypothesis, the patterns in (1) and (2) do not
suffice to phrase-structurally differentiate Haitian Creole pa from French pas.
Indeed, the differences between (1) and (2) may be the sole result of verb
(non-)movement and are most likely not the result of an eventual phrase-
structural difference between the Haitian Creole and French negation markers.
If anything, the above patterns do liken Haitian Creole pa to French pas: at
D-structurc they both govern anu pn:ccuc VP (assuming along with DeGraff
1992d that Haitian Creole TMA markers are auxiliary verbs). This means that
the differences in (1) and (2) are compatible with the assumption that Haitian
Creole pa and French pas are phrase-structural homologues.
In fact, if no other discriminating evidence were available in addition to (1)
and (2), one could very well assume that Haitian Creole pa and French pas
occupy identical position in their syntactic trees - perhaps in Spec of NegP.
This would seem a natural assumption given that Haitian Creole pa and French
pas are cognates. However, I will argue on empirical and theoretical grounds that
Haitian Creole pa and French pas are systematically differentiated.
 A riddle on negation in Haitian 67

2. Pa versus pas: negative concord and double negation

There is one well-defined class of facts distinguishing Haitian Creole pa from

French pas. 6 One conspicuous dissimilarity between Haitian Creole pa and
French pas involves the interpretation of negative quantifiers. The essential point
is that, semantically, negative quantifiers interact differently with Haitian Creole
pa than they do with French pas. Compare (3) and (4): the interpretation of each
acceptable French sentence in (3) is truth-conditionally opposite to that of its
Haitian Creole counterpart in (4). 7
(3) a. Personne n' est pas venu. (French)
nobody ne+is pas come
'Everybody came' (lit. 'Nobody has not come.')
b. ??Je n'ai pas vu personne.
lsG ne+have pas seen nobody
c. Ce n' est pas rien.
3sG ne+is pas nothing
'This is something.' (lit. 'This is not nothing.')
(4) a. Pesonn pa vini. (Haitian Creole)
nobody pa come
'Nobody has come.'
b. Mwen pa we pesonn.
1SG pa see nobody
'I haven't seen anybody.'
c. Sa pa anyen.
3sc; pa nothing
'This is nothing.'

To the extent that the French sentences in (3) are interpretable, they give rise
to instances of double negation. In double negation, co-occurring negative
elements cancel each other, giving rise to a net positive statement - duplex
negatio affirmat. In (3a) for instance, personne n 'est pas venu - like nobody
has not come in standard varieties of English - actually means 'everyone has
come', a positive statement, and (3c) Ce n 'est pas rien means 'This is some-
thing'. On the contrary, the perfectly grammatical Haitian Creole sentences in
(4), with two negative elements each, are immediately construed as net
negative statements. Pesonn pa vini in (4a) means 'nobody has come' and Sa

6. These facts were also noted in Deprez (1992), although they lead her to conclusions very
different from mine. Deprez's analysis is discussed in section 6.2.
7. Native speakers vary considerably as to their acceptance of (3b). (3a) seems more readily
acceptable, specially with stress on perso1zne. (See Kayne 1984: 39, note 4 for relevant
comments.)
68 M. DeGraff

pa anyen means 'This is nothing'. 8 In the linguistic literature, for example

Labov (1972), the phenomena illustrated in (4) has been called "negative
concord". The semantic contrast between double negation in (3) and negative
concord in (4) is a robust empirical distinction between Haitian Creole pa and
French pas. 9•10
Now, with respect to negative concord, Haitian Creole pa behaves very much like
another French negation marker, that is to say, ne. Each clause in (5), like in (4),
produces a single instance of negation in the presence of two negative markers: 11
(5) a. Personne n' est venu. (French)
nobody ne+is come
'Nobody has come.'
b. Je n'ai vu personne.
lsG ne+have seen nobody
'I haven't seen anybody.'
c. Ce n' est rien.
3SG ne+is nothing
'This is nothing.'
There are cases in Haitian Creole and in French where several negative
elements combine into a single instance of sentence negation (in the presence of
Haitian Creole pa and French ne).

8. Haitian Creole negative quantifiers, when used in full clauses, require the presence of a negative
marker such as pa 'not' or poko 'not yet'. The negative head in Haitian Creole must always be
overt, like ne in Classical (and literary) French; see note II. For the use of pesonn in isolation,
see (ill) and surrounding comments.
9. At this point, I am setting aside French varieties which do allow negative concord with pas. I
come back to these varieties in Section 8 where I address diachronic implications.
10. With respect to negative concord and double negation, one can approximately say that Haitian
Creole is to French what African-American English is to standard English; see Labov ( 1972) for
data from several English dialects along with insightful comments. An ideal segue to this paper
would extend the forthcoming analysis to the paradigms noted by Labov (and to all instances
of negative concord and double negation).
II. The negation clitic ne tends to disappear in colloquial French. But there are contexts where ne
tends to remain present. Ashby ( 1981), for example, notes that ne is retained categorically when
the grammatical subject is a negative noun phrase. Three of Ashby's examples of obligatory ne
are shown below:
(i) Rien ne me surprend.
nothing ne I so surprise
'Nothing surprises me.'
(ii) Personne n' est ici.
nobody ne+is here
'Nobody is here.'
(iii) Aucun erudiant n' arrive ii l'heure.
no student ne+arrive on the-time
'No student arrives on time.'
 A riddle on negation in Haitian 69

(6) Nan katye sa-a pesonn pa di pesonn anyen. (Haitian C.)

in neighborhood DEM-SG nobody pa say nobody nothing
Dans ce quartier; personne ne dit rien apersonne. (French)
'In this neighborhood, nobody says anything to anybody.'
(6) strongly suggests that Haitian Creole (like French) has a rule of negative
concord involving pa (as the counterpart of French ne). Further note the absence
of pas in the French translation of (6).
Based on the above data, Haitian Creole pa seems to have much more in common
with French ne than with French pas - modulo, of course, the fact that French ne
is a morphologically weak morpheme which is disappearing in certain registers (see
notes 8 and 11) whereas Haitian Creole pa is a morphologically strong morpheme
which is generally not optional where it occurs (but see note 22). Semantically,
pa is also stronger than ne: it can express negation on its own, unlike French ne;
contrast (1) with Jeanne serait *(pas) aile au marche. Now, how can the above
data be mapped onto recent theoretical developments? As a prerequisite, a brief
examination of the structure of French negated clauses is in order.

3. Sentential negation in French

Negation in French has been at the limelight of generative literature, due in great
part to seminal papers by Emonds (1978) and Pollock ( 1989). French typically
expresses sentential negation with two markers: a eli tic ne (which is disappearing
in colloquial French; but see note 11) and an independent morpheme pas.
Descriptively, in finite declarative clauses, 11e and pas embrace the tensed verb,
as in (7).
(7) a. Pierre n' est pas venu.
Pierre ne+is pas come
'Pie11'e has not come.'
b. Marie n'aime pas Pierre.
Marie ne+loves pas Pierre
'Marie doesn't love Pierre.'
I adopt without discussion the assumption that sentential negation markers
project their own phrases in syntax, according to the templates of X-bar theory,
that is, with (i) a specifier under XP and sister to X'; (ii) a head X0; and (iii) a
complement under X' and sister to X0 ; see for example Pollock (1989) and
Zanuttini (1991) for detailed motivations.
Following Pollock, many syntacticians have analyzed French pas as occurring
in the Spec of NegP (specifier of negation phrase), with ne as the head of NegP.
As shown in (8), Pollock derives the order ne ... V0 pas by head-movement of
ne, along with the finite verb, into a head which is higher than NegP. For
70 M. DeGraff

Pollock, the landing site of this movement is explicitly the head of TP (tense
phrase). Movement of ne is motivated by its clitic nature, among other reasons.

(8) DS: [TP · · · T0 · · · [NegP pas [Neg' [Nego ne l [VP ·· · yo ··· Jll]
SS: [TP ... nei+V~+T 0 ... [NegPpas [Nego ei [vp ... ej ... ]]]]
Pollock's hypothesis is reinforced by the following constraint: only the tensed
verb may precede pas. This is instantiated in (9).
(9) a. *Pierre n' est venu pas.
Pierre ne+is come pas
'Pierre has not come.'
b. *Marie n'aime Pierre pas.
Marie ne+loves Pierre pas
'Marie doesn't like Pierre.'
In (8), the tensed verb and the negation head ne move to a position higher than
pas, up to Tense (or to Infl, pre-Pollock) leaving behind the remnant of the verb
phrase. Thus pas always precedes past participles and the verb's complements,
as shown in (7) and (9).
Verb movement into Comp also shows that ne forms a complex head with the
verb. In yes/no questions, ne is fronted, piggybacking on the verb:
(10) a. N'est il pas venu?
ne+is 3sa pas come
'Has he not come?'
b. *Est il ne pas venu?
is 3SG ne pas come
c. *Est il pas ne venu?
is 3SG pas ne come
In (10), ne obligatorily moves with the tensed verb into Comp. (lOb) and (lOc),
where ne remains stranded, are ungrammatical.
In (7)-(10), movement of ne along with the tensed verb is also made
necessary by the Empty Category Principle: as a head, ne, if it were to remain
in place, would intervene between the raised verb and the trace of V0 , preventing
antecedent-government of the trace; cf. Chomsky (1986). That pas, unlike ne,
may separate the raised verb from its trace in the VP further supports Pollock's
hypothesis that pas is in the specifier position: by relativized minimality,
elements in specifier position do not interfere with antecedent-government of
heads (Rizzi 1990).
This overview of French sentential negation will suffice for our purposes, the
main import being that French pas resides in Spec of NegP with ne acting as the
head of NegP.
 A riddle on negation in Haitian 71

4. Haitian pa as head of NegP

The next questions are: (i) How does the configuration in (8) interact with
negative concord and double negation? (ii) Is Haitian Creole pa in Spec of
NegP, like French pas? The answer to (i) will suggest an answer to (ii). Question
(i) itself brings me directly to Zanuttini's (1991) seminal work on negation, and
to a lesser extent to the work of Haegeman (1991). These are very important
works because they contribute to a more precise definition of the interface
between syntax and semantics. Zanuttini studies the distribution of negative
elements in a variety of languages. She sets, inter alia, the configurational
conditions for their semantic interpretation. For the purpose at hand, the crux of
Zanuttini's observations is her reliance on phrase structure configurations in
order to distinguish double negation versus negative concord. The configurations
she relies on significantly bear on whether Haitian Creole pa is in the head
position or in the specifier position of NegP.
Let us return to instances of negative concord in Haitian Creole and in French.
In particular, recall that Haitian Creole pain (4), like French ne in (5) and unlike
French pas in (3), may co-occur with the negative quantifiers pesonn and anyen
in either pre- or post-verbal position without inducing cancelled negation. For
example, in (4) and (5), the a-sentences translate as 'Nobody has come' and the
b-sentences translate as 'I didn't see anybody'. Both of these are negative
statements. Notably, the negative force of the markers pa in Haitian Creole and
ne in French does not cancel the negative force of the quantifiers pesonn and
personne, respectively.
In (4), (5) and (6), two or more negative elements occur in the same clauses.
Each of these elements can by itself express sentence negation. Yet, when put
together, their collective interpretation is not the sum of their individual negative
forces, but rather a single instance of sentence negation. The intuition behind the
phrase-structural underpinnings of negative concord is that it results from a
configuration where a negative marker and a negative quantifier enter into
agreement. According to Zanuttini, this sort of agreement (like other phenomena
of grammatical agreement, for example, subject-verb agreement) is realized via
a Spec-head relationship. At Logical Form (LF), the (trace of the) negative
marker in head position and the negative quantifier in specifier position share
their negative values under NegP.
Thus, in (4) and (5), Haitian Creole pa and French ne are heads of NegP.
They enter at LF into Spec-head agreement with a negative quantifier in Spec
of NegP. Quoting Zanuttini (p. 144), "when such a configuration occurs, the
semantic contribution to the interpretation of the sentence is the same as that of
the head." 12 To sum up, I take the negative concord data in (4), (5) and (6) in

12. Example (ii) in note 4, Jan pa-p pa vini, illustrates double negation in Haitian Creole. There we
have two instances of pa (thus two NegPs) cancelling each other (but see note 20).
72 M. DeGraff

Haitian Creole and French to be symptomatic of the head status of Haitian

Creole pa and of French ne. 13 · 14
However, the head status of pa is not agreed upon in the literature on Haitian
Creole. Previous work on negation in Haitian Creole has put pa in Spec of
NegP, assuming that it was structurally similar to French pas; see Lefebvre and
Lumsden (1992). But, given my reasoning thus far, if Haitian Creole pa were in
Spec of NegP, it would prevent the negative quantifier in (4) and (6) from
occurring there at LF. In absence of the Spec-head agreement configuration,
negative concord in these sentences would remain unexplained (assuming
Zanuttini's framework and the quantifier status of pesonn). Indeed, Zanuttini's
framework predicts that pa in Spec of NegP would impose a reading of double
negation instead of one of negative concord. This is disconfirmed by the data in
(4) and (6). By reductio ad absurdum, pa is not generated in Spec of NegP.
The above logic finds typological support. Compare Haitian Creole pa with
French pas. As noted by Zanuttini, when it occurs with a negative quantifier,
French pas does entail double negation, as in (3). Crucially, French pas, unlike
Haitian Creole pa, is generated in Spec of NegP, as argued by Pollock for
independent reasons (see Section 3). Because it is in Spec of NegP, French pas
prevents negative concord by blocking (LF) movement of the negative quantifier

13. Haegeman's (1991) framework, based on data from Standard Dutch, West Flemish, French and
Italian, makes identical predictions when applied to the status of Haitian Creole pa. Haegeman
writes (p. 16):
We might propose that in languages with NC [Negative Concord] readings the
head of NegP is "strong": it is autonomously licensed: it has its NEG feature
in Lhe base. The NEG niLerion is mel by a "strong" slalic agreement
configuration. In non-NC languages, on the other hand, Neg is "weak" and
would be assigned the NEG feature by its specifier by virtue of Spec-head
agreement. ... What is crucial for NC ... is that the NEG feature on Neg0 is
independently licensed, i.e., that Neg0 is a strong head. In languages where the
NEG feature on Neg 0 can only be achieved via dynamic agreement the negative
head is not strong and NC is not possible ....
Consider (6) for example in light of Haegeman's hypothesis. Given that negative concord
obtains in (6), the negative head must be "strong". Thus pa must be heading NegP. If pa were
in Spec of NegP, it would license a "weak" Neg0 and negative concord would not obtain.
Note though that French data such as J'ai donne rien apersonne 'I haven't given anything to
anybody' would be problematic for Haegeman; but see note 11 for a subset of negative concord
cases in French where ne seems not to be optional. Haegeman herself toys with the idea that
French ne is deleted only at PF.
14. In the LF representation of (6), with three negative quantifiers (pre-verbal pesonn, post-verbal
pesonn and anyen), one quantifier would be in Spec of NegP and the other two would be
adjoined to Spec of NegP, as in Haegeman (1991). There they would undergo a process of
absorption somewhat reminiscent to wh-absorption in certain questions with multiple
wh-elements (Higginbotham and May 1981). Presumably, French pas in Spec of NegP, not being
a quantifier of the sort of personne, rien, etc. cannot undergo absorption alongside the latter; see
note 15.
 A riddle on negation in Haitian 73

into Spec of NegP. Therefore, in (3), the negative quantifier, personne or rien,
cannot enter into agreement with the negative head. Therefore, the negation
marker and the quantifier each contribute separately their negative force to the
interpretation of the sentence. This results in double-negation readings, as
expected. 15 · 16

5. More on pa versus pas

What other facts distinguish Haitian Creole pa and French pas, besides negative
concord and double negation? (11) and (12) present two further kinds of
distributional evidence in favor of a structural distinction between the two
pa(s)'s.l7.I8
Firstly, in (lla), French pas may occur at the periphery of the clause it
modifies, to the left of the complementizer. In the Haitian Creole clause in (11 b),
pa must occur clause-internally between subject and predicate.

15. At this stage, one could ask: Why docsn 't pas undergo absorption at LF when co-occurring with
negative quantifiers? Such absorption would induce negative concord, contra the readings in (3);
see note 14. However, note that negative quantifiers are distinct from pas because of their
quantifying properties. The former have both a quantifier component and a negative component
whereas the latter only has a negative component. Pas is akin to yes/no operators, like whethet;
which also hlock absorption: compare i•mnder •rho lol'cs whom and */wonder whether Mat)'
lm•es whom. It thus seems that absorbed operators must quantify over certain sets. Pas inverts
the truth-value of its propositional argument and does not quantity over sets, at least not over
sets of the sorts that personne and rien quantify over.
16. Liliane Haegeman (pers. comm. November I ~93) alerts me to the fact that nie, the West Flemish
equivalent of French pas, does enter into negative concord as in (i). (The head of West Flemish
NegP is en.)
(i) T-ee doa niemand nie over gek/aapt.
it-has there no-nne not about talked
'No one talked about that.'
With respect to negative quantifiers, there might be one (perhaps, crucial) difference between
Haitian Creole and French on one hand and West Flemish on the other. It seems to be the case
that the latter must obligatorily scramble the negative quantifier which enters into negative
concord. I hypothesize that this obligatory scrambling (adjunction to Spec of NegP at
S-structure) is what permits the negative quantifier to be absorbed with nie. S-structure
adjunction of niemand to nie in (i) would give quantifier-status to nie, allowing it to undergo
absorption at LF. LF adjunction of negative quantifiers to French pas would occur too late for
pas to become endowed of quantifier status; pas would remain a yes-no operator throughout the
derivation and would not be able to undergo quantifier absorption. (See Haegeman 1993 for a
different and fully-fledged analysis of West Flemish negative concord based on the
Neg-criterion.)
17. I thank Richard Kayne for indicating to me the relevance of the data in this section.
I 8. Gaatone (1971) contains additional pertinent data from French.
74 M. DeGraff

(11) a. Boukifait le clown pour (pas) qu'ils s'ennuient.

Bouki makes the clown for pas that+they bore+REfl
b. Bouki ap je komik pou (*pa) yo (pa) anniye.
Bouki PROG make comic for pa 3PL pa bore
'Bouki is clowning around so that they (don't) get bored.'

Secondly, French pas may modify adjectival phrases, as in (12a). This is

unlike Haitian Creole pa in (12b).

(12) a. Voila un type pas betel

there a fellow pas stupid
b. *Men yon moun pa sot!
here/there a fellow pa stupid
'There goes a man who is not stupid!'
It is not clear to me how to account for the contrast in (11) and (12) in a precise
manner. Suffice to say that this contrast could be the indirect result of the
different structural statuses of Haitian Creole pa and French pas. 19 ·20

6. Alternatives to pa as head of NegP

Let us now go back to the interaction of Haitian Creole pa with what I have
assumed to be negative quantifiers. In particular, are there alternatives to my
hypothesis that pa heads NegP and enters at LF into a Spec-head agreement

19. The contrasts in (II) and (12) might be related to Cardinaletti and Guasti's (1992) observation
that French pas may function as an adverbial projection and, as such, adjoin to, for example,
AP. Haitian Creole pa, a head, would not adjoin to a maximal projection (dito for French ne:
*Boukifait le clown pour ne qu'i/ s'ennuie and *Voila un type ne bete!).
20. Yves Dejean brings to my attention an optional use of pa, which doesn't seem to affect the
truth-conditions of its clause:
(i) Mwen pa we (pa) yon grenn moun.
ISG NEG see NEG a single person
'I haven't seen (anybody/not one person}.'
(ii) Se pa ti kauri mwen (pa) kauri.
se NEG little run ISG NEG run
'l REALLY ran.' (i.e. 'I ran a whole lot.')
In (i), pa yon seems to be a relic of French pas un (= aucun, as in Je n 'ai vu pas une seule femme
and Pas une seulefemme n 'est venue; for more data, see Gaatone 1971: 49, 176ff.). Structurally,
the second pa would be part of a larger nominal phrase (Neg0 taking DP as a complement and
being an extended projection of NP?). LF movement of the thus formed nominal phrase into the
specifier position of the matrix NegP would then derive apparent negative concord.
As of (ii), I can only say for now that its interpretation is suggestive of expletive (optional) ne
in French Je era ins qu 'it (ne) parte 'I fear that he leave'. Space and time prevents me from
addressing many of Dejean's fascinating data; their treatment require future papers.
 A riddle on negation in Haitian 75

relationship with negative quantifiers, whence the negative concord facts in (4)
and (6)? One other possibility is that pa is in a position adjoined to VP. I must
also address the hypothesis adopted by Deprez (1992) that pesonn (and perhaps
anyen) 21 in (4) and (6) may manifest properties of negative polarity items
instead of negative quantifiers.

6.1. Pa in adjoined position?

I first look at the possibility that Haitian Creole pa is generated adjoined to VP.
Adjunction, obviously, does not lend itself to a typical configuration of
Spec-head agreement. Such a configuration is assumed to be necessary for
negative concord. The possibility of adjoining pa (to VP, say) is thus excluded,
in principle.

6.2. Pesonn and an yen: quantifiers or polarity items?

There is one other alternate analysis to consider. This analysis would obviate the
need for pesonn and anyen in (4) and (6) to move at LF into Spec of NegP.
Deprez ( 1992), for one, argues that Haitian Creole pesonn, contrarily to French
personne (and English nobody), is not a true negative quantifier. According to
her, pesonn manifests in certain contexts properties of negative polarity items.
These negative polarity item properties would account for (a few of) the cases
of negative concord in (4) and (6).
Deprez (1992: 38) writes: "Sentences [similar to (3a) ami (Jb)] are usually
judged by speakers either as uninterpretable or as involving canceled negation
[double negation] which produces a positive statement. Clearly this is not the
case in [Haitian Creole]." I agree wholeheartedly with Deprez that there is a
clear semantic difference between (3) in French and (4) in Haitian Creole: the
former instantiates double negation and the latter negative concord. But I
disagree with her as to locating the reasons for that difference on Haitian Creole
pesonn versus French personne - Deprez considers pesonn to potentially have
negative polarity item properties and personne to be a true negative quantifier.
As stated earlier, I believe that the comparison of (4) in Haitian Creole with (3)
and (5) in French, coupled with Zanuttini's insights, indicates that Haitian Creole
pais the equivalent of French ne and is not the equivalent of French pas. Hence,
Haitian Creole (4) is the counterpart of French (5). 22

21. Deprez doesn't explicitly address the status of anyen.

22. Although some (colloquial) varieties of French do allow (3) to be interpreted on a par with (4);
see Section 8.
76 M. DeGraff

Recall that in my own account, the direct opposition between (3) - double
negation in French - and (4) - negative concord in Haitian Creole - does not
stem from the different properties of personne and pesonn. I claim that both
pesonn and personne act as negative quantifiers in the relevant cases. However,
unlike French pas, which is in Spec of NegP, Haitian Creole pa, like French ne,
heads NegP. This phrase structure distinction between French pas and Haitian
Creole pa is what explains the contrast between (3) and (4).
But by Deprez's arguments, the negative concord facts in (4) follow instead
frompesonn's and anyen's negative polarity item properties (while she implicitly
maintains that pa and pas are homologous). Deprez directly compares Haitian
Creole sentences similar to (4a) and (4b) to French sentences similar to (3a) and
(3b) and concludes that Haitian Creole pesonn must be different from French
personne, the former having negative polarity item properties in certain
environments and the latter being always a negative quantifier.
Negative polarity items are elements like English anybody. Negative polarity
items enter into phenomena superficially resembling negative concord:
(13) I didn't see anybody.
However, negative polarity items display properties quite different from those
of negative quantifiers. One such property is that negative polarity items are not
inherently negative. For example, (14) in English has no negative import:
(I 4) Did you see anybody?
Another distinction between negative quantifiers and negative polarity items
is that negative polarity items must be in the scope of an appropriate trigger.
Compar~ (15) and (16):

(15) Nobody saw me.

( 16) a. *Anybody didn't see me.
b. *Anybody saw me.
(15) is acceptable with the negative quantifier nobody in subject position. Both
sentences in ( 16) are ungrammatical because anybody is not in the scope of an
appropriate trigger. 23
The above distinction between anybody and nobody in English succinctly
exemplifies the division between negative polarity items and negative quantifiers.
What about pesonn and anyen?
Deprez argues that pesonn, if a negative polarity item, would not need to

23. There are occurrences of any in matrix subject position, for example, anybody can do that. Any
in such a context is not a negative polarity item, but a "possibility polarity item" (Lawler 1972)
or "free-choice" any (Carlson 1981), and does not require a negative trigger. See Labov (1972),
Hom (1989), Kadmon and Landman (to appear) and references therein for further remarks on
various uses of any.
 A riddle on negation in Haitian 77

undergo movement at LF and could be interpreted in situ. (Recall that Deprez's

examples of true negative quantifiers are French personne and English nobody.)
Contra Deprez, I now show that pesonn and anyen are not negative polarity
items.
There are two tests which have been commonly used in order to distinguish
negative quantifiers from negative polarity items. (See Zanuttini 1991 and
Haegeman 1991) for application of these tests to Italian ness uno and West
Flemish niemand.) I apply these tests to pesonn and anyen.

6.2.1. Modification by preske 'almost'

Firstly, pesonn and anyen may be modified by preske 'almost', as in (17). Modifi-
cation by almost is ruled out in the case of negative polarity items such as English
anything.Z4 Witness the possible and impossible English translations for (17b). 25
(17) a. Preske pesonn pa vote pou Manigat. (Haitian Creole)
almost nobody pa votefor Manigat
Presque personne n 'a vote pour Manigat. (French)
'Almost nobody voted for Manigat.'
b. Mwen pa manje preske an yen jodi-a. (Haitian Creole)
I SG pa eat almost nothing today
Je n 'ai presque rien mange aujourd'hui. (French)
'I have eaten almost nothing today.'
*I haven't eaten almost anything today. 26

6.2.2. Use in isolation

Secondly, negative polarity items like English anything and anybody cannot
occur in isolation. This is because they must be licensed within the scope of an
appropriate trigger. Unlike negative polarity items, pesonn and anyen do occur
in isolation, for example, as a negative answer to a question:

24. When modifying a negative quantifier, almost is interpreted relative to the cardinality of the set
over which the quantifier ranges. Presumably, negative polarity items, unlike negative quantifiers,
are not related to sets, whence the inadmissibility of almost as modifier of negative polarity
items. (I thank Michael Hegarty for this observation.)
(As noted in Carlson 1981, among others, free-choice any may be modified by almost: I could
eat almost anything!; cf. note 23.)
25. English negative polarity items do not occur in pre-verbal position, so the issue does not arise as to
whether (l7a) can be translated using a negative polarity item. Actually, the sheer occurrence
of pesonn in subject position distinguishes it from English negative polarity item anybody.
26. A few speakers do accept I haven't eaten almost anything today, although I didn't see almost
anybody at the party seems worse to most informants. I have nothing to say about this variation.
78 M. DeGraff

(18) a. Kimoun ki we ou? - Pesonn (Haitian Creole)

who ki see 2sa nobody
Qui t'a vu?- Personne (French)
'Who saw you?- Nobody/* Anybody'
b. Kisa ou manje? - Any en (Haitian Creole)
what 2sa eat nothing
Qu 'as tu mange? - Rien (French)
'What did you eat? - Nothing/* Anything'
Therefore, the data in (17) and (18) indicate that pesonn and anyen are indeed
negative quantifiers.

6.2.3. Non-licensing by conditionals and comparatives

Further distributional facts distinguish Haitian Creole pesonn from negative

polarity item anybody. These facts implicate conditionals and comparatives.
Linebarger ( 1987), among others, has noted that English conditionals and
comparatives license anybody in their scope. Laka has noted similar facts in
Spanish, with ningun for example. Interestingly, in (19), pesonn is not licensed
by conditionals and comparatives. This provides one additional reason for
denying the status of negative polarity item to pesonn. 27

(19) a. *Si ou touye pesonn, ou pral nan prison. (Haitian Creole)

if 2SG kill ? 2SG go in jail
'If you kill anybody. you go to jail.'
b. *Bouki pi wo pase pesonn.
Bouki more tall pass ?
'Bouki is taller than anybody (I have ever known).'

27. This argument is somewhat muddled by Carlson's (1981) intuition that the English translations
in (19) actually instantiate "free choice" any and not negative polarity item any, as indicated
by modification by almost: If you kill almost anybody, you go to jail and Mary is taller than
almost anybody I have ever known. Is it coincidental that English any is ambiguous between
negative polarity item and "free choice"? If this polysemy is not accidental, but based on
intrinsic properties of any and/or deeper principles of grammar (as alluded, but rejected, in
Carlson p. 18; see also Kadmon and Landman, to appear), then one would expect pesonn, if an
negative polarity item, to double as a "free choice" item (like any), which it doesn't.
In this regard, note the following French examples: Je /'ai vue de plus pres que perso1111e 'l
saw her closer than anyboby' and Y a-t-il personne qui veuille venir? 'Is there anybody who
wants to come?' There, personne is licensed in comparatives and interrogatives, much like
English anybody, contra Deprez's tacit supposition that personne and nobody pattern more alike
than pesonn and nobody, with pesonn in some environments equated to anybody; see note 29.
(Kayne 1984: 39, note 4 observes further correspondences between perso1me and anybody; see
also Gaatone 1971.)
 A riddle on negation in Haitian 79

6.2.5. Further data

While arguing that pesonn is an negative polarity item under certain conditions,
Deprez (1992: 38f.) rightly notes a distinction between (20a) and (20b). (20a)
has a single pa, in the matrix clause, and gives rise to a negative expectation
(the party will be deserted). (20b) has two pa's, one in the matrix and the other
in the embedded clause, and gives rise to a positive expectation (at least some
people will come to the party).

(20) a. Mwen pa kwe pesonn ap vini. (Haitian Creole)

lsG pa believe nobody IRREAL come
'I don't think that anybody will come.'
(i.e. 'I think the party will be deserted.')
b. Mwen pa kwe pesonn pa ap vinP 8
1SG pa believe nobody pa IRREAL come
'I don't think that nobody will come.'
(i.e. 'At least some people will come to the party.')

From the contrast in (20) and from the similar contrast in the corresponding
English translations, Deprez concludes that it is only when the negation marker
pa is in the embedded clause, as in (20b), that pesonn can be interpreted as a
negative quantifier (equivalent to nobody). Otherwise, in (20a), pesonn is
interpreted as a negative polarity item, similar to anybody, and is licensed by
matrix negation. 29 It is not clear to me how this alleged distinction between
pesonn qua negative quantifier and pesonn qua negative polarity item would be
derived. In any case, I believe that this distinction is unwarranted. Given my
assumptions, pe.wnn in both sentences in (20) is a negative quantifier and moves
at LF into Spec of NegP. The different readings may be derived as follows: In
(20a), pesonn moves into Spec of matrix NegP. 30 In (20b), pesonn moves into

28. The sequence 'NEG IRREAL', which I choose to write pa ap for expository reasons, is always
pronounced [pap]. See Dejean's (1980) most comprehensive treatise on Haitian Creole
orthography.
29. More precisely and in Deprez's words (p. 38):
When it immediately precedes negation. [pesonn] has the meaning of the
English negative quantifier nobody. When it follows the negation however, it
behaves like the [negative polarity item] anybody. Descriptively, we can say
that the meaning of the sum of [pesonn] and the negation pa is ambiguous
between that of a negative quantifier like nobody and that of an negative
polarity item like anybody depending on the position of [pesonn] with respect
to the overt negation. In this, Haitian Creole [pesonn] differs from the French
quantifier personne, which is always interpreted as a negative quantifier.
30. To be explored is whether such movement is licensed by "Neg-raising" verbs (in the sense of
Horn 1989).
80 M. DeGraff

the specifier position of the embedded NegP. In (20a), negation concord within
the single NegP gives rise to one instance of negation. This explains the negative
expectation. In (20b), negative concord through Spec-head agreement still
obtains within the lower NegP. But the matrix pa is obviously unable to
participate into Spec-head agreement within the lower NegP; on the contrary,
it adds its negative force to the lower pa. This gives rise to two separate
instances of negation which cancel each other and result in the positive
expectation (see [ii] of note 4). 31 Therefore, the data in (20) receive a natural
explanation within my assumptions about the interpretation of negative
quantifiers in Haitian Creole; there is no need to postulate an interpretive
ambiguity for pesonn (negative quantifier or negative polarity item). 32 •33
To recapitulate, it appears that Haitian Creole pa is indeed the head of NegP.
One would expect this property of pa to have consequences in other areas of the
grammar. And it does - fortunately for my analysis. Predication patterns in
Haitian Creole constitute one domain where such consequences are clear. Pa
being a head, its head-government capacities affect the distribution of traces
occurring in the position of base-generated subjects, between surface subjects
and nominal predicates - this contrasts with certain clauses without pa where
a resumptive nominal, se, must spell-out the trace of the deep subject because
of the absence of an appropriate head-governor. This is the topic of the next
section.

7. Negation and predication in Haitian

As extensively argued in DeGraff ( 1992a, b, c, c),34 predication patterns in

Haitian Creole are partly determined by the distribution of traces and of

31. Given the interpretation of (20b), it must also be the case that peson11 must move to specifier
position of the closer, embedded NegP. In other words, peso11n cannot escape the lower NegP
and enter into agreement with matrix pa. This constraint might be implemented via relativized
minimality.
32. In French, LF movement of perso11ne exhibits a subject-object asymmetry: ?Je 11 'ai exige qu 'ils
arretent person11e versus *Je 11 'ai exige que perso11ne soit arrete (Kayne 1984: 23 ff.). Such
asymmetry is absent in Haitian Creole, perhaps because of the mechanisms which allow the
language to be pro-drop, namely identification of a null subject from Infl. Haitian Creole also
permits long-distance subject extraction over overt Comp (absence of Comp-trace effects)
although it, unlike Italian, does not freely allow subject inversion (DeGraff 1992d, g). Italian,
somewhat like French, shows a pre-/post-verbal asymmetry with respect to LF movement of
nessuno (Rizzi 1982: 118ff.).
33. Literary French shows a contrast similar to (20), as in Je 11e veux pas que personne vien11e versus
Je ne veux pas que perso11ne ne vie1me (data from Kayne 1984: 40, notes 4, 5) with the embedded
ne filling the role of Haitian Creole embedded pa in (20b).
34. Actually, this section is a concentrate prepared out of extracts from DeGraff (1992a, b, c, e).
 A riddle on negation in Haitian 81

resumptive pronominals spelling-out traces. Universally, the distribution of traces

must respect the Empty Category Principle (ECP). The ECP requires that all
traces be both licensed and identified. What matters here is the licensing
condition on traces. This condition operates via head-government. In certain
predication contexts, the trace of the base-generated subject is not head-governed
and must be realized as a resumptive pronominal in order to avoid an ECP
violation. However, because it is a head, pa can license a non-overt trace in its
governing domain and the trace of the deep subject need not be overtly realized
as se.

7 .1. Basic data

Haitian Creole predicative structures do not contain an overt verbal copula. (21)
presents the basic data; the crucial fact is that predicates not headed by verbs
may be string-adjacent to their subjects, in (2la)-(2lc). 35
(21) a. Bouki (* se) malad.
Bouki se sick
'Bouki is sick.'
b. Bouki (*se) anba tab la.
Bouki se under table the
'Bouki is under the table.'
c. Bouki (??se) abitan.
Bouki se peasant
'Bouki is a peasant.'
d. Malis *(se) [yon dokte/ dokte a! Aristide}.
Malis se a doctor doctor the Aristide
'Malis is {a/the doctor/Aristide}.'

However, not all kinds of predicates are allowed string-adjacent to their

subjects. In (2ld), the predicate is a proper name (Aristide) or a nominal
occurring with a pre-posed or a post-posed article (yon dokte 'a doctor' or dokte
a 'the doctor'). These predicates are determiner phrases (DP) or number phrases
(NumP). 36 With DP and NumP predicates in simple, present-tense clauses, the
morpheme se must occur between subject and predicate.
However, there are clauses where DP and NumP predicates occur without a

35. When se is present, the grammaticality of (2lc) improves with Bouki left-dislocated and se in
subject position (Spec of IP). Left-dislocated structures will be mostly kept at bay for the
purpose of this discussion; but see (24) and surrounding text, Damoiseau (1987) and DeGraff
(1992e) for more comments.
36. I motivate these labels in DeGraff (1992e).
82 M. DeGraff

preceding se. This happens precisely when the predicate is preceded by

morphemes qualifying as heads: tense, mood or aspect (TMA) markers, 37 or
a complementizer. Significantly, pa also excludes occurrence of se between
predicates and subjects. This similarity between TMA and complementizer heads
on the one hand and pa on the other hand confirms that pa is a head. I elaborate
on this similarity in section 7.4.
Finally, se never occurs between subject and predicate when the predicate is
adjectival or prepositional. Therefore, what seems to demand explanation is the
presence or absence of se with nominal predicates.

7.2. Is sea verb?

What is the nature of se? One possibility that immediately comes to mind is that
se is a verbal copula, the counterpart of English be or French etre. It seems
reasonable to discard that possibility for the following reasons.
All verbs in Haitian Creole follow negation and TMA markers while se does
not. (22) and (23) illustrate the positional difference between se and the verb
chante 'sing'.
(22) Bouki (*IHJ) (*g) [a se] yon dokte.
Bouki NEG ANT se a doctor
'Bouki {was/is}(n't) a doctor.'
(23) Kok la IZQ g [yo chante] maten an.
rooster the NEG ANT sing morning DET
'The rooster didn't sing this morning.'
The ungrammaticality of (22) illustrates a robust generality: under no circum-
stances does se follows negation and TMA markers in Haitian Creole.
In addition, whenever se precedes negation and/or TMA markers, it is in
subject position (Spec of IP), with the pre-se nominal in left-dislocated position,
as indicated by the comma-intonation in (24). Furthermore, se in such environ-
ment can be replaced by li, which is undisputedly pronominal:
(24) Bouki # {se/li} (IZQ) (te) (yon) dokte.
Bouki se 3sG NEG ANT a doctor
'Bouki, he {was/is} (not) a doctor.'
If se may occur in Spec of IP and be replaced by a personal pronoun, than it
is unlikely that it is a verb. The generalization that se before negation and/or
TMA markers is in Spec of IP is reinforced by the observation that any pre-se
nominal occurring in these contexts must necessarily be able to left-dislocate.

37. See DeGraff (1992d) for arguments that TMA markers are verbal heads.
 A riddle on negation in Haitian 83

(25) shows that the pronoun li - which cannot bear stress and cannot be left-
dislocated - renders illicit the sequence of se and negation/TMA markers? 8
(25) Li [a se ] (*12{1) (*!§) yon dokte.
Li se NEG ANT a doctor
'He/she is/was (not) a doctor.'
The data in (24) and (25) illustrate two sites of occurrence for se: (i) when
preceding negation/TMA markers, se fills Spec of IP, forcing any pre-se nominal
to dislocate, as in (24); (ii) when preceded by an atonic pronoun like li, se is in
a non-verbal position, between Spec of IP and the (phonetically realized part of
the) predicate, and excludes negation and TMA markers; hence the
ungrammaticality of *Lise pate yon dokte in (25). I now turn to identifying the
position of se in (25), between Spec of IP and a nominal predicate. 39

7.3. Analysis

In what follows, I summarize my analysis of the paradigm in (21), focusing on

the mechanisms that regulate the (non-)appearance of se. 40 I assume that the
subject of all Haitian Creole predicative clauses is generated internal to a small
clause. This is similar to analyses in Stowell (1978) and Burzio (1986), among
others. In (26), the acronym "SC-SP" denotes the base-generated small clause
subject position. In the case of AP, PP and NP, SC-SP is in specifier position.
In the case of DP and NumP, SC-SP is left-adjoined to DP and NumP. This is
for reasons involving, inter alia, the functional nature of D 0 and Num0 ,
mechanisms of 8-role assignment and predication, and Baker's ( 1988) Universal-
ity of Theta-Role Assignment (UTAH) 41 (see DeGraff 1992a, c, e). Briefly,
Spec of DP and Spec of NumP are reserved for a genitive phrase or its trace,
and, because of UTAH, cannot contain the base-generated subject. 42 I also

38. Se also occurs clause-initial in cleft sentences. There as well, it precedes TMA and negation
markers. I argue in DeGraff (1992e) that clause-initial se in clefts is in Spec of IP.
39. DeGraff (l992b) differentiates between Haitian Creole se and its French cognate c'est.
40. More extensive data implicating se and analyses of its occurrences can be found in Fauchois
(1982), Damoiseau (1987), Kihm (1990), Lumsden (1990), Deprez and Vinet (1991), Manfredi
(1991), etc. Most of these analyses are critiqued in DeGraff (1992e).
41. Universal Theta Assignment Hypothesis: Identical thematic relationships between items are
represented by identical structural relationships between those items at the level of D-structure.
42. Tonjes Veenstra astutely points out that my UTAH argument might do violence to a strict
version of the hypothesis. Especially, the last three lines of (26) locate the subject of a nominal
predicate in three different positions, namely in Spec of NP or adjoined to DP or NumP.
However, it is central to my analysis that the thematic relationships in the three cases be
different. Although DP and NumP are extended projections of NP, the actual predicates in (2ld)
- and their corresponding structures in (26) - are crucially distinct from their inner NPs (and
give rise to distinct interpretations).
84 M. DeGraff

assume that, in general, specifiers of functional heads are not assigned 8-roles
at D-structure. The D-structures of the predicate small clauses are shown in (26):
(26) [AP SC-SP [A' A 0 ... ]]
000 (la)
[pp SC-SP [P' ... P0 ... ]] (lb)
[NP SC-SP [N' ... N° ... ]] (lc)
[0 p SC-SP [0 p ... [ 0 , ... D0 ... ] ... ]] (ld) wldokteaandAristide
[NumP SC-SP [NumP ... [Num' ... Numo ... ] ... ]] (ld) w!yon dokte
Now, the deep subject, generated inside a small clause, does not receive Case
in this position. It must move into Spec of IP in order to get Case. 43 •44 By the
ECP, the trace left in SC-SP by movement of the subject to Spec of IP must be
both identified and head-governed. In all the relevant cases, identification of the
trace in SC-SP is satisfied through antecedent-government from Spec of IP. What
about head-government?
I follow Aoun and Sportiche ( 1983), in assuming that m-command is the
relevant command-relation for head-government. X m-commands Y if and only
if (27) holds:
(27) For all Z, Z a maximal projection, if Z dominates X then Z dominates Y.
In (21) (and [26]), with AP, PP and NP predicates ([2la] through [2lc],
respectively) the trace in SC-SP is head-governed by the lexical head of the
predicate. Infi, being phonetically null, the mapping from D-structure to S-structure
is string-vacuous. However, in case of predication by DP and NumP, as in (2ld), the
subject moves from a position adjoined to DP or NumP. In (28), the trace is not
head-governed because of the intermediate DP or NumP segment (represented
by XP2) causing failure of m··command of SC-SP by the predicate head.

(28) · [xPI SC-SP [xP2 ... X0 ... ]]

In order to save the structure, the trace must be spelled-out as a resumptive
pronominal which, being overt, is not subject to ECP. That resumptive
pronominal is se. 45

43. Haitian Creole Infl being phonetically null should not prevent Case assignment to Spec of IP
through Spec-head agreement. Note that the subject of I love Lucy also gets Case through
agreement with a phonetically-null Infl.
44. One anonymous reviewer asks "Does [Infl] have features? If [so], how are these checked?"
Movement of the subject into Spec of IP would allow the (Case-assigning) features of Infl to
be checked. (One could also imagine that Haitian Creole verbs, being morphologically non-
inflected, procrastinate their moving to Infl until LF, where they would check Infl's weak
features - the subject would still need to move at S-structure, perhaps to provide the
IP-predicate with a syntactic subject.)
45. I am reminded by a Probus reviewer that resumptive nominals are also assumed to require Case.
Presumably, se gets (nominative) Case by being in a chain with the subject that has landed into
Spec of IP.
 A riddle on negation in Haitian 85

7.4. Predictions

This analysis predicts that se need not occur with DP and NumP predicates
whenever there is an alternate head-governor for SC-SP, external to the small
clause. This prediction is upheld in at least the following contexts:
- when the predicate is preceded by a complementizer, as in (32); and
-when the predicate is preceded by a TMA verbal marker, as in (34).
Also the minimal pairs in (29) and in (30) support the idea that it is exactly
the occurrence of a functional head within the nominal phrase that entails the
potential ECP violation, not the nouniness of the predicate.
I consider these data in turn.

7 .4.1. Absence of se with bare NPs

When the noun in predicate position is bare -- occurring without a functional

head - the predicative small clause is NP and the subject is generated in Spec
of NP, cf. (31 a), and its trace, after movement into Spec of IP, is head-governed
by the predicate head. Such trace need not be spelled-out as se. As soon as a
functional head occurs, in (29a) and (30a), head-government of SC-SP in
adjoined position fails, as witnessed in (3lb) and (3lc), and the trace in SC-SP
violates the ECP.
(29) a. *Bouki yon abitan.
Bouki NUM peasant
b. Bouki abitan.
'Bouki is a peasant.'
(30) a. *Preval premye minis la.
Preval prime minister DET
b. Preval premye minis.
'Preval is Prime Minister.'
(31) a. [Np SC-SP [N' N ... ]]
b. [0 p SC-SP [0 p ... [ 0 , ... D 0 ••. ] .•. ]]
c. [NumP SC-SP [NumP ··· [Num' ··· Num 0 ..• ] ... ]]

7.4.2. Absence of se when the subject is questioned

In matrix questions, the complementizer ki must follow the wh-phrase when the
wh-phrase originates in subject position. Furthermore, ki only occurs with subject
extraction. This constraint suggests that the presence of ki is required for head-
government. In (32), the predicate is DP or NumP, SC-SP is not head-governed
from inside the small clause, yet se is absent. It must be the case that ki head-
86 M. DeGraff

governs SC-SP, allowing the trace there to remain phonetically null, as

schematized in (33). 46 .47
(32) Kimoun ki {yon dokte/ dokte a}?
who ki a doctor doctor the
'Who is a/the doctor?'
(33) a. [cp Kimouni [c ki [1p e( [1, [ 0 p ei [0 p dokte a]]]]]]
who ki doctor DET-SG
'Who is the doctor?'
b. [cp Kimouni [c· ki [1p ei' [r [NumP ei [NumP yon dokte]]]]]]
who ki NUM doctor
'Who is the doctor?'

7.4.3. Absence of se in TMA-marked clauses

The effect of TMA markers vis-a-vis se is similar to that of the complementizer

ki. When a TMA marker, for example, te 'ANT', occurs, se must not precede the
predicate. In (35), the TMA head te head-governs the trace in SC-SP and se is
not needed.
(34) Bouki te yon abitan.
Bouki ANT NUM peasant
'Bouki was a peasant.'
(35) DS: [1p [r 1° [yp [yo te] [NumP Bouki [NumP yon abitan]]]]]
SS: [,p Boukij [r 1° [yp [yo te] [NumP ei [NumP yon abitan]]]]]

7.4.4. Absence of se in negated clauses: pa head-governs SC-SP

Now, let us go back to pa. Pa in (36), like ki in (32) and te in (34), renders
superfluous the presence of se in SC-SP:
(36) Bouki pa yon dokte.
Bouki NEG a doctor
'Bouki is not a doctor.'
In other words, se need not (and must not) precede the predicate when the latter
is governed by pa. What makes pa similar to the complementizer ki and the
TMA marker te with respect to whether se is needed? This similarity is not at

46. My analysis of ki as being in C0 is incompatible with Law's (1992) analysis where ki occupies
Spec of IP as a resumptive pronoun bound by the wh-operator kimoun. Ki in Spec of IP would
not head-govern the trace in SC-SP, and the absence of se in (32) would remain unexplained.
47. If one assumes with Chomsky (1986: 47f.) that IP is defective as a barrier, then ki in (33) does
head-govern the most embedded subject trace ei.
 A riddle on negation in Haitian 87

all surpnsmg, given that pa, like ki and te, is a head and can head-govern
SC-SP. However, if pa were in Spec of NegP or adjoined to VP, it would not
head-govern SC-SP and the parallel between (33), (34) and (36) would be
mysterious. This further substantiates the claim that Haitian Creole pa heads
NegP. As such, pa does head-govern the trace in the small clause subject
position, as shown in (37). 48 •49

(37) DS: [IP [r I0 [NegP [Nego pa] [NumP Bouki [NumP yon dokte]]]]]
SS: [1p Boukii [r I0 [NegP [Nego pa] [NumP ei [NumP yon dokte]]]]]

8. Diachronic implications

As an epilogue, I contemplate the diachronic implications of my analysis of

Haitian Creole pa as head of NegP (instead of Spec of NegP). In particular, what
do these findings implicate for the genesis of Haitian Creole? If my analysis of
Haitian Creole pa is on the right track, not only would it shed light on negative
concord in Haitian Creole, but it might offer as a bonus one more puzzle with
respect to the genesis of Haitian Creole - at least considering the superstrate,
French, and (too sparse) data from one representative substrate, Fon.
I have argued that Haitian Creole pa and French pas, in spite of being
cognates, occupy different positions in their respective syntactic structures. Now,
I briefly look at one eminent progenitor of Haitian Creole and see how sentential
negation functions there.

8.1. Sentential negation in Fon

For negation in Fon, I rely on work by Lefebvre and Lumsden (1992) and da
Cruz (1992). According to them, Fon is very much like French in having two
negation markers: a
and rna, in head and specifier of NegP, respectively. A
would correspond to French ne and rna would correspond to French pas. Unlike
a,
French though, the negation head in Fon, is post-verbal while the specifier rna
is pre-verbal, as shown in (38). Thus the two negation markers necessarily
bracket the verb phrase, when they co-occur: 50

48. See (24), DeGraff (1992e) and Damoiseau (1987) for occurrences of se in the position preceding
pa/te - se there is in Spec of IP, not in the subject's base-position. As expected, se never
precedes C0 lei.
49. There is a surprising dissimilarity between te/pa and ki as of whether se may succeed them. Te
and pa categorically prohibit a subsequent se whereas some speakers allow se subsequent to ki.
This might be related to the distance between governor and governee: the trace ei of the base-
generated subject is hierarchically further away from ki in (33) than it is from both te and pa
in (35) and (37); see DeGraff (1992e) for more speculations.
50. Da Cruz notes that mti and a cannot co-occur in declarative simplex clauses.
88 M. DeGraff

(38) [NegPmd [Neg' VP [Nego a]]]

(39) a. (Ni) K5ku rna du as5n a.
if Koku md eat crab a
'If Koku has not eaten crab ... '
b. *(Ni) K5ku du ma as5n a
if Koku eat rna crab a
'If Koku has not eaten crab ... '
The data in (39a) suggest that in Fon, there is no verb movement. This is like
Haitian Creole, see (1), and unlike French, see (2).
From the above data, I conclude that Haitian Creole differs not only from
French, but also from Fon, which I take to be one representative Kwa ancestor
of Haitian Creole (pending further data from other relevant Kwa languages).
Remarkably, Neg0 (pa) is head-initial in Haitian Creole while it is head-final in
Fon (a). Looking at it from the pre-verbal perspective, the pre-verbal negation
marker in Haitian Creole heads NegP instead of being in Spec of NegP like it
is in Fon (rna).
Finally, contrary to both French pas and Fon rna, Haitian Creole pa seems
excluded from attributive environments; both pas and rna may be used in such
environments (see [12a] for French and da Cruz (1992) for Fon where rna is
shown to modify adnominal participles). 51

8.2. Relexification?

On the surface, negation in Haitian Creole shares properties with both French
and Fon. The sentential negation marker in Haitian Creole is phonetically
identical to French pas. And like Fon rna, Haitian Creole pa always precedes the
VP, including TMA markers (see note 2). In Haitian Creole, like in Fon, verbs
do not move out of VP. It is these similarities, among other things, that have
enticed Lefebvre and Lumsden into proposing that Haitian Creole pa is one
further instantiation of the relexification process.
In Lefebvre and Lumsden (1992), relexification is the process whereby adult
native speakers of one prominent ancestor language, specifically Fon, created
Haitian Creole by replacing phonetic shapes in their lexicon with forms derived

51. DaCruz (pers. comm., June 1993) has provided me with Fon examples of apparent cases of
negative concord in the presence of mti and ii. To be further investigated are the precise
conditions regulating Fon negative concord with mti (as made clear in da Cruz's 1992 insightful
work, the distribution of mti is subject to subtle semantic nuances). Also, great care must be
taken in distinguishing negative quantifiers from negative polarity items. As indicated in Section
6.2, the latter might give rise to apparent negative concord via distinct interpretive mechanisms.
(See Zanuttini 1991 and references therein.)
 A riddle on negation in Haitian 89

from French while maintaining grammatical and semantic properties of their

native language. Put in a simplistic way, Haitian Creole grammar would actually
be the grammar of one of its West-African source languages (Fon, say), and the
audible part of its lexicon would originate from French.
Regarding negation, Lefebvre and Lumsden argue that Haitian Creole pa
results from relexification of Fon rna into French pas. Consequently, like Fon rna
and French pas, Haitian Creole pa would be in Spec of NegP. In this account,
NegP in Haitian Creole would be head-final, like in Fon, but with a null Neg0 .
In contrast with their position, I hope to have shown in this paper that Haitian
Creole pa, if anything, shares more properties with French ne than with French
pas. I have argued that Haitian Creole pa, like French ne, heads NegP, unlike
Fon ma and French pas, which are in Spec of NegP. Also, like French ne,
Haitian Creole pa may partake in negative concord. As it appears, the phrase-
structural characterization of Haitian Creole pa resists a straightforward
explanation via relexification, at least not in the fashion outlined in Lefebvre and
Lumsden (1992). 52
In this respect, it would be fascinating to compare Haitian Creole with
Romance varieties that allow negative concord with homologues of pas (PA, for
short), for example, Vald6tain (Franco-Provenr;al) and Occitan. In these dialects,
PA when co-occurring with negative quantifiers is compatible with negative
concord; see Zanuttini ( 1991) and references therein. 53 •54 This is unlike

52. As it stands, my analysis of pa is neutral as to whether the substrate languages played some role
in the genesis of Haitian Creole pa. At issue here is a re!exification-based analysis of Haitian
Creole NegP ilia Lefebvre and Lumsden where Haitian Creole NegP is isomorphic to Fon NegP
and where Haitian Creole pa is in Spec of NcgP and results from the relexitication of Fon md
with French pas, both in Spec of NegP. As alluded in the main text, such an analysis is
formulated as part of a larger hypothesis in which Haitian Creole is the product of a somewhat
direct re-analysis of French surface strings through the superimposition of Fon phrase structures.
(See Lefebvre and Lumsden 1992 for details and DeGraff, to appear,for a critique.)
53. Quebec French also allows PA with negative concord readings (Julie Auger, pers. comm.
October 1992).
54. With respect to the structural characterization of PA in these languages, there is one difficulty
that I inherit from Zanuttini 's proposal. Because of negative concord, Yald6tain, Occitan and
Quebec French PA must head NegP. However, contrarily to Haitian Creole pa, PA follows the
finite verb (like French pas). If PA were exactly like Haitian Creole pa, then we would expect
it to block head-movement of Y0 across it and to rule out the sequence Y0+PA, contrary to fact.
Notwithstanding my current unfamiliarity with the data, I will venture that PA is not only a
head, like Haitian Creole pa, but it is also syntactically affixal. As such, it allows a finite verb
to attach to it; the complex Y 0+PA then moves to a higher (Tense) projection, conceivably like
ne+Y 0 in (8).
Adding (slight) credence to this suggestion is the exclusion of PA from certain clauses where
there is no Tense head forcing movement of Y0 - Y-raising would be required in order to
provide affixal PA with morphological support. Tenseless clauses excluding PA include a subset
of "true" imperatives (in the sense ofZanuttini 1991, who has a different account of these data
based on the structural positioning of PA; also see Kayne 1991).
90 M. DeGraff

standard French pas in, for example, (3). This similarity might guide the
historical linguist in looking for the exact varieties of Romance which actually
participated in the genesis of Haitian Creole. Vis-a-vis negation, were these
varieties similar to Valdotain and Occitan or to the French of examples (3) and
(5)? In other words, did pas in the grammars of 17th-century French settlers in
Haiti induce double negation or negative concord?55

8.3. Restructuring of French?

Also fascinating is the scenario which was suggested to me by Bill Stewart

(pers. comm. October 1992). Recall that the Haitian Creole verbal system has
no inflectional morphology. Most verbs occur in uninflected form, and they are
possibly preceded by independent morphemes marking tense, mood or aspect;
see (1). Haitian Creole verb forms were plausibly derived from the correspond-
ing French infinitival or participle forms. But it is crucial to note that in French
both infinitives and past participles occur mostly after pas. In Pollock's
framework, this is because only the finite verb needs to move beyond pas to get
its inflectional morphology. Also of interest is the historical fact noted by
Stewart that, in earlier Haitian Creole texts, nepa and napa - from French
n'es(t) pas and n'a(s) pas, respectively - were used as single morphemes
marking negation. 56 Thus the sequence ne+auxiliary+pas would have been
reanalyzed as a pre-verbal negation marker, with the emergent negation
morphemes nepa/napa, and later pa, inheriting their head status from French ne.
This might sketch the mutation of Fon rna ... aand French ne ... pas into Haitian
Creole pa, on their maiden voyage from Africa and Europe to the Caribbean -
ami from specifier Lo head of Lhe Negation Phrase. As for myself, am I sailing
toward a solution to this riddle on negation in Haitian Creole?

Richard Kayne (pers. comm., June 1992) judiciously points out that viewing PA as a syntactic
affix is incompatible with data where material (like adverbs) intervenes between V0 and PA and
where an infinitival verb follows, instead of precedes, PA. Should we then resort to Zanuttini 's
(less restrictive) proposal that PA is in Spec of NegP at S-structure (permitting V-raising), but
in head of NegP at LF (permitting negative concord)? (I owe many thanks to Richie for many
long and enriching discussions on this, and many other, topics.)
55. Stepping beyond the Haitian Creole case into (more debatable) issues of Creole genesis, it is
striking that Bickerton (1981: 65) remarks that negative concord is quite common among Creole
languages. Relevant data are taken from Guyanese Creole, Papia Kristang and Hawaiian Creole
English. Of course, it is necessary to buttress this claim against a wider inspection of Creole
languages; see Holm (1988: 171-174) for an effort in this direction. Perhaps negative concord
is typologically the unmarked case (Labov 1972: 774, 803).
56. Yves Dejean advises caution in using early Haitian Creole texts for historical purpose. These
texts were written by non-native speakers that were often disdainful of the Creole. Moreover,
the language is systematically betrayed by the then-prevalent French-based orthography. In any
case, Dejean corroborates Stewart's findings about the use of nepa!napa as negation markers
(but with much lower frequency than pa).
 A riddle on negation in Haitian 91

8.4. 'Pa' pa 'Pas'

Wherever I may land, one proposition, in this sea of conjectures, remains certain:
synchronically, Haitianpa is systematically different from (standard) French pas.
Given that Haitian has no copula, the riddle now reads:

(40) 'Pa' pa 'Pas' .57

The University of Michigan

City University of New York
e-mail: [email protected]

References

Abney, Steven ( 1987). The English noun phrase in its sentential aspect. Unpublished Ph.D.
dissertation, Massachusetts Institute of Technology, Cambridge.
Aoun, Josef and Dominique Sportiche (1983). On the formal theory of government. The Linguistic
Review 2: 211-236.
Ashby, William (1981). The loss of the negative particle ne in French. Language 57: 674-687.
Baker, Mark C. (1988). Incorporation: A Theory of Grammatical Function Changing. Chicago: The
University of Chicago Press.
Bickerton, Derek (1981 ). Roots of Language. Ann Harbor, Michigan: Karoma.
Burzio. Luigi (1986). Italian Syntax: A Gm•emment-Binding Approach. Dordrecht: Reidel.
Cardinaletti, Anna and Maria Teresa Guasti ( 1992). Negation in small clauses. Paper given at the
Sixth Symposium on Romance Languages and Linguistics. Utrecht.
Carlson, Greg ( 1981 ). Distribution of free-choice any. Papers fmm the Meeting uf the Chicagu
Linguistic Society. Chicago: Chicago Linguistics Society.
Chomsky, Noam (1986). Barriers. Cambridge, MA: MIT Press.
da Cruz, Maxime (1992). Contribution a !'etude de Ia negation en fongbe. Unpublished manuscript,
Universite du Quebec a Montreal.
Damoiseau, Robert (1987). Situation de communication et fonctionnement de Ia langue en Creole
Hai'tien: Approches pour une analyse. Revue Qwfbecoise de Linguistique Theorique et Appliquee
6.
DeGraff, Michel (1992a). On the structure of small clauses and on Creole genesis: Evidence from
Haitian. GLOW Workshop. Lisboa.
- (1992b). On certain differences between Haitian and French predicative constructions.
Unpublished manuscript, University of Pennsylvania. (To appear in Selected Papers from the
22nd Annual Linguistic Symposium on Romance Languages. John Benjamins, Amsterdam).
- (1992c). The syntax of predication in Haitian. Proceedings of NELS 22. Graduate Linguistic
Student Association, University of Amherst at Massachusetts.
- (1992d). Is Haitian Creole a pro-drop language? In Atlantic Meets Pacific: A Global View of
Pidginization and Creolization, Byrne and Holm (eds.), 71-90. Amsterdam: John Benjamins.

57. And I smile, contemplating how I will break the news to my dear father- 'Pa' pa 'pas',
Papa! ...
92 M. DeGraff

- (1992e). Creole grammars and acquisition of syntax. Unpublished Ph.D. dissertation, University
of Pennsylvania.
- (1992f). 'Pa' pa 'pas'. Meeting of the Society of Caribbean Linguistics. Barbados.
- (1992g). Haitian null subjects revisited. Travaux de Recherche sur /e Creole Hai"lien II.
Universite du Quebec a Montreal.
-(To appear). Review of "Travaux de Recherche sur le Creole Hai"tien, I 0". Journal ofPidgin and
Creole Languages.
Dejean, Yves (1980). Comment Ecrire le Creole d'Hai"ti. Quebec: Collectif Paroles.
Deprez, Viviane (1992). Is Haitian Creole really a pro-drop language? Travaux de Recherche sur /e
Creole Hanien II. Universite du Quebec a Montreal.
Deprez, Viviane and Marie-Therese Vinet (1991). On the categorial status of the Haitian Creole
particle se. Paper given at the meeting of the Society of Pidgin and Creole linguistics in
Chicago.
Emonds, Joseph (1978). The verbal complex V'-V in French. Linguistic Inquiry 9: 151-175.
Fauchois, Anne (1982). Nature et Fonction des Monemes se en Creole Hai"tien. Centre de
Linguistique, Universite d'Etat d'Ha"iti.
Gaatone, David (1971). Etude Descriptive du Systeme de Ia Negation en Franrais Contemporain.
Geneva: Droz.
Haegeman, Liliane (1991). Negative heads and negative operators. Unpublished manuscript,
University of Geneva.
- (1993). Syntax of negation. Unpublished manuscript, University of Geneva.
Higginbotham, James and Robert May ( 1982). Questions, quantifiers and crossing. The Linguistic
Review 1: 41-79.
Holm, John (1988). Pidgins and Creoles. Vol. 1. Cambridge: Cambridge University Press.
Horn, Laurence (1989). A Natural Hist01y of Negation. Chicago: The University of Chicago Press.
Kadmon, Nirit and Fred Landman (To appear). Any. Linguistics and Philosophy.
Kayne, Richard (1984). Connectedness and Binary Branching. Dordrecht: Foris.
- - (1991 ). Italian negative infinitival imperatives and eli tic climbing. Unpublished manuscript, City
University of New York Graduate Center.
Kihm, Alain ( 1990). What is it that you said? A study of obligatory focalization in two Creoles and
beyond. Paper given at the Conference on Pocus and Grammatical Relations in Creole
Languages. University of Chicago.
Labov, William (1972). Negative attraction and negative concord in English grammar. Language 48:
773-818.
Laka, Itziar (1990). Negation in syntax: On the nature of functional categories and projections.
Unpublished Ph.D. dissertation, Massachusetts Institute of Technology, Cambridge.
Law, Paul (1992). Subject extraction in Haitian interrogatives. Paper given at the 23rd Annual
Conference on African Linguistics.
Lawler, John (1972). Ample negatives. Papers from the Meeting of the Chicago Linguistic Society.
Chicago: Chicago Linguistic Society.
Lefebvre, Claire and John Lumsden (1992). Precision on word order in relexification. Paper given
at the meeting of the Society of Pidgin & Creole Linguistics in Philadelphia.
Linebarger, Marcia (1987). Negative polarity and grammatical representation. Linguistics and
Philosophy 10: 325-387.
Lumsden, John (1990). The bi-clausal structure of Haitian clefts. Linguistics 28: 741-759.
Magloire-Holly, Helene (1982). Les modaux: Auxiliaires ou verbes? In Syntaxe de /'Hanien,
Lefebvre eta!. (eds.). Ann Harbor: Karoma.
Manfredi, Victor (1991). Agb6 and Ehugbo: Igbo linguistic consciousness, its origins and limits.
Unpublished Ph.D. dissertation, Harvard University.
May, Robert (1985). Lagica/ Form: Its Structure and Derivation. Cambridge, MA: MIT Press.
 A riddle on negation in Haitian 93

Pollock, Je<!n-Yves (1989). Verb movement, universal grammar and the structure of IP. Linguistic
Inqu ii)' 20: 365-424.
Rizzi, Lwigi (1982). Issues in Italian Syntax. Dordrecht: Foris.
- ( 199•0). Relativi:ed Minima/if)•. Cambridge, MA: MIT Press.
Stowell, Tim ( 1978). What was there before there was there. Papers from the Meeting of the Chicago
Linguistic Society. Chicago: Chicago Linguistic Society.
Zanuttini, Raffaella (1991). Syntactic properties of sentential negation: A comparative study of
Romance languages. Unpublished Ph.D. dissertation, University of Pennsylvania.