Anti-Extraction in Québécois French Wh

advertisement
Anti-Extraction in Québécois French
Wh-Interrogatives (DRAFT)
Jean-Philippe Marcotte
University of Minnesota
marco043 *at* umn *dot* edu
November 2006
1
Introduction
Pesetsky (1987) has suggested that the difference in grammaticality between the two
questions in 1 is due to a difference in the discourse properties of wh-pronouns and
wh-NPs:
(1)
a. Which booki did you persuade which man to read ei ?
b. ??Whati did you persuade who to read ei ?
Wh-NPs like “which book” and “which man” limit the range of felicitous answers
to (1a) to a set of ordered pairs of contextually salient books and men, whereas the
wh-pronouns “what” and “who” in (1b) do not. In Pesetsky’s terms, wh-NPs are
D(iscourse)-linked, while wh-pronouns are non-D-linked. Being D-linked allows a whphrase to take scope in-situ,1 absolving it from its obligation to move to an operator
position at LF: “which man” in (1a) can stay put at LF, but “who” in (1b) cannot:
(2)
a. Which booki did you persuade which man to read ei
b. ??Whoj whati did you persuade ej to read ei
What makes (2b) anomalous is the crossed dependencies between the gaps and their
antecedents. In contrast, (2a) features no such crossing, and is more acceptable as a
result.
1
This is accomplished via binding by a question operator, following a proposal by Baker (1970).
The details of this are not important for my purposes in this paper.
1
A distinction similar to D-linking is made by Cinque (1990), who shows that Italian “nonreferential” (i.e. non-D-linked) arguments are restricted in their Ā-extraction
possibilities, while “referential” (i.e. D-linked) arguments are not. According to
Chung (1994), Chamorro displays the same contrast.
In simplified terms, the apparent conclusion is that, due to their discourse properties, wh-NPs can enter in a broader range of extraction configurations than whpronouns. What makes Québécois French (QF) interesting in this context is that the
exact opposite conclusion seems to obtain:2
(3)
a. Quii est-ce que Marie voit ei ?
who is-it that Marie sees
‘Who does Marie see?’
b. Quelle personnei est-ce que Marie voit ei ?
which person is-it that Marie sees
‘Which person does Marie see?’
c. Quii est-ce qui ei voit Paul?
who is-it that sees Paul
‘Who sees Paul?’
d. *Quelle personnei est-ce qui ei voit Paul?
which person is-it that sees Paul
‘Which person sees Paul?’
Both wh-pronouns and wh-NPs can be extracted from object position, as shown in
(3a,b), but only wh-pronouns can be extracted from subject position (3c): example
(3d), in which a wh-NP is extracted from subject position, is ungrammatical. Here, Dlinking seems to be having the opposite effect it has in English, Italian and Chamorro:
2
The judgements in (3) represent the gathered intuitions of four native speakers of the dialect,
including myself and another trained linguist. The two other speakers were presented with a list
of some twenty wh-interrogatives, four of which I judged to feature anti-extraction, and told to
mark those sentences that they felt were unlikely to be heard in informal dialogue on the streets
of Montréal. One of the speakers marked exactly the four anti-extraction examples disseminated
throughout, while the other marked none of the examples, but added that she hesitated on exactly
those examples that the first speaker rejected. This kind of hesitation can be attributed to a conflict
between a speaker’s knowledge of the dialect and their knowledge of the more standard form of the
language; in this case, the latter would have won out. Of course, though this is often forgotten,
linguistic knowledge is a property of individual minds, and a linguistic phenomenon can at best be
claimed to hold of collections of observed idiolects, rather than dialects or languages. As always,
somebody’s asterisk is somebody else’s question mark, with a third person wondering why the other
two find anything wrong with the sentence in the first place. But the fact that all the judgements
I asked for converge leads me to believe that anti-extraction may be a feature of the idiolects of a
significant proportion of QF speakers.
2
D-linked wh-NPs enter in a narrower range of constructions than the non-D-linked
wh-pronouns.
My goal in this paper will be to properly characterize this anti-extraction phenomenon of QF, and provide an analysis for it. It will turn out not only that Dlinking is not sufficient to the task, as the extraction site, the “qui” complementizer
and SpecCP position are all crucial to the phenomenon; but that D-linking may in
fact not be relevant: certain non-D-linked wh-phrases are also anti-extracted.
To allow for a precise formalization of all this, I will develop along the way an
analysis of QF wh-interrogatives, which will address an issue that has fallen by the
wayside since the early days of the Extended Standard Theory, mainly the syntactic
status of “est-ce que”; the discussions in Kayne (1972) and Langacker (1972) have
not been picked up.3 Lexical-functional grammar (LFG Bresnan, 1982; Dalrymple
et al., 1995) will provide the framework in which these analyses will be cast.
The paper is laid out as follows: a few peripheral issues about French syntax that
will pop up throughout and could raise a few eyebrows will be discussed in Section
2, and used to introduce the LFG framework. Section 3 will discuss “est-ce que” and
wh-interrogatives, and present an analysis that will be used in Section 4, where I will
return to anti-extraction and suggest an analysis. Note that all the grammaticality
judgements for the French sentences in this paper are given from the point of view of
a QF speaker, unless explicitly stated.
2
Some Preliminaries: French and LFG
There are two thing about French syntax that warrant discussion. First is verb movement, in the sense of Pollock (1989), a discussion of which will clarify the structural
positioning of verbs throughout this paper. Second is the hotly debated status of
argument pronouns (Rizzi, 1986; Miller, 1991; Auger, 1995; Labelle and Hirschbuhler, 2000), an affixal treatment of which I will briefly justify. These two issues will
be put to use as a helpful way of introducing some key concepts of LFG. Here, and
throughout the paper, I assume the following schematic clause structure:
3
Jones (1996) recapitulates these discussions, but offers no new insight.
3
(4)
CP
C′
SpecCP
C0
TP
T′
SpecTP
T0
VP
V′
SpecVP
V0
2.1
NP
Verb (Non-)Movement
The influential analysis of Pollock (1989) ascribes to language-specific constraints on
verb movement the difference in grammaticality between the following English and
French examples:
(5)
a. *John likes not Mary.
b. Jean (n’)aime pas Marie. (Standard French/QF)
c. John doesn’t like Mary.
d. *Jean (ne) pas aime Marie. (SF/QF)
(6)
a. *John kisses often Mary.
b. Jean embrasse souvent Marie. (SF/QF)
c. John often kisses Mary.
d. *Jean souvent embrasse Marie. (SF/QF)
Assuming, as Pollock does, that both negation and adverbs have a fixed position, and
that this position is lower in the structure than the projections of T0 but higher than
those of V0 , then the difference between the English and French examples above is
explained if verbs are allowed to move from V0 to T0 in French, but not in English.
The English examples in (5a) and (6a) are ungrammatical because the main verb
has moved from V0 to T0 ; those in (5c) and (6c) are grammatical because the main
verb has not moved from V0 at all. As for French, examples (5b) and (6b) are
grammatical because the verb is allowed to move from V0 to T0 ; examples (5d) and
(6d) are ungrammatical because movement from V0 to T0 is obligatory, but has not
taken place.
4
This last conclusion, that verb movement from V0 to T0 is obligatory in French,
must in fact be amended to account for the placement of auxiliaries:
(7)
a. Jean (n’)a pas aimé Marie. (SF/QF)
b. Jean (n’)est pas revenu de Paris. (SF/QF)
What these examples show is that the presence of an auxiliary in T0 absolves the
verb from its obligation to move. The generalization about French can be recast in
the following terms: that T0 must be occupied by a tense auxiliary or, in the absence
of such an auxiliary, by the finite verb. In transformational grammar, this variable
verb positioning is achieved through movement; in LFG, the movement device is
unavailable, and the solution must necessarily be different.
Syntactic structure in LFG is represented at two different levels. The c(onstituent)structure, in tree notation, represents the dominance and precedence relationships
between the words in a sentence. The f(unctional)-structure, in attribute-value matrix notation, represents the deeper grammatical relationships that hold between these
words, taking notions like subject and object as primitive. The c-structure is defined
in terms of phrase structure rules, which bear functional annotations establishing how
each node is mapped to the f-structure. Each node in the c-structure corresponds to
an f-structure;4 the root node always corresponds to the outermost f-structure, and
the f-structures of all other nodes are related to this outermost f-structure via the ↓
and ↑ metavariables in their functional annotations. These metavariables are interpreted locally, relative to the node that introduces them: ↓ refers to the f-structure
of the node introducing the annotation, and ↑ refers to the f-structure of the mother
of that node.
Now imagine that the lower part of the schematic clause structure shown in (4)
is generated by the following, partially annotated c-structure rules:
(8)
TP
−→ SpecTP
(9)
T’
(10)
VP −→
(11)
V’
−→
−→
T’
↑=↓
T0
VP
↑=↓ ↑=↓
SpecVP
V’
↑=↓
V0 NP
↑=↓
4
In a rather confusing bit of terminology, the f-structure of a sentence is composed of the fstructures of its parts.
5
The ↑=↓ annotations in these c-structure rules establish that the f-structures corresponding to the TP, T’, T0 , VP, V’ and V0 nodes are in an equality relationship:
these f-structures are one and the same. Whether the verb is in T0 or in V0 makes no
difference, since it will map its functional information to the same f-structure. But
what could force the verb to appear in T0 in just those instances it needs to?
The solution to this problem is in fact uncomplicated. As Kroeger (1993) points
out, there is a morphosyntactic difference between what appears in T0 and what
appears in V0 : finite verbs in the former, infinitivals in the latter. This corresponds
exactly to the French pattern: a lone verb carries its own tense information, but
a verb co-occurring with an auxiliary is stripped of that information, which occurs
instead on the auxiliary. Consequently, auxiliaries and finite verbs, as tense-bearers,
will appear in T0 , while verbs with auxiliaries, as non-tense-bearers, will occur in V0 .
Following the analysis of Russian in King (1995), I will assume that verb-stems
can be realized as both Ts and Vs (12), that auxiliaries (13) and tense suffixes (14) are
Ts, and that nonfinite suffixes are Vs (15). In the lexical entries that follow, pred is
an attribute whose value is a semantic form with its subcategorization requirements;
tense is just that; annotations which assign a value to an attribute are defining
equations, and annotations which do not are existential equations, which constrain
that the attribute they state to exist have some value. All equations must be satisfied,
or the sentence is ungrammatical.
(12)
tomb- T,V
(↑ pred)=‘fall < subj >’
(↑ tense)
(13)
suis
(14)
-ais T (↑ tense)=imparfait
(15)
-é V (↑ tense)
T (↑ tense)=passé composé
A verb in a sentence with no auxiliary will thus necessarily be realized in T0 : tense
inflections are affixal, and need a stem to attach to. Only verb stems in their Tincarnation can be used, because tense inflection is uniquely categorized as T, and
could not attach to a verb stem in its V-incarnation; this is shown in (16). But
when a tense auxiliary is present, the T0 position is taken by the auxiliary, effectively
confining the verb to V0 , as shown in (17).5
5
Note that suppletive forms present no particular problem: a suppletive tensed verb-form is
uniquely categorized as T, while a suppletive non-tensed verb-form is uniquely categorized as V.
6
(16)

pred
TP
tense
↑=↓
T′
D
E
‘fall subj ’
imparfait
↑=↓
T0
tomb − ais
(17)

pred
TP
tense
↑=↓
T′
↑=↓
T0
↑=↓
VP
suis
↑=↓
V′
D
E
‘love subj,obj ’
passé composé
↑=↓
V0
tomb-é
The tense suffix “-ais” in (16) needs a verb stem to attach to, forcing “tomb-” to
appear in its T-incarnation. In contrast, the auxiliary “suis” in (17) occupies T0 ,
forcing the appearance of “tomb-” in its V-incarnation. Notice how the c-structure
rules ensure that both attributes are mapped to the same f-structure, though the
words that contribute them are under separate nodes.
2.2
Pronouns As Affixes
The morphological status of French pronouns has long been a point of contention
in the literature. One camp (Barnes, 1985; Rizzi, 1986; Labelle and Hirschbuhler,
2000) maintains that pronouns are independent syntactic elements that cliticize to
the verb, while the other (Miller, 1991; Auger, 1993, 1995) opines that pronouns are
in fact verbal affixes. The second approach is the one I adopt in this paper. The
data in Auger (1993, 1995) is especially appropriate, as it is drawn from QF. Auger
7
shows that pronouns have several of the properties pointed out in Zwicky and Pullum
(1983) as more characteristic of affixes than clitics. Specifically, (i) pronouns show a
high degree of selection for the kind of syntactic entity they are necessarily bound to,
mainly tensed verbal elements; (ii) given certain pairs of homophonous pronoun+host
sequences, one of the sequences but not the other can have an alternate, idiosyncratic
variant;6 (iii) there are arbitrary gaps in the set of possible pronoun sequences, as no
combinations of first or second person accusative pronouns with third person datives
pronoun is licit. In addition the following behavior is also observed: (iv) pronouns
are necessarily repeated when two verbs are coordinated; (v) some pronouns differ in
their pre- and post-verbal forms; (vi) pronouns can co-occur with nouns or emphatic
pronouns in canonical argument position, in such a way that they appear to function
as agreement markers.7
This accumulation of evidence thus seems to support an affixal analysis of pronouns in QF over a syntactic one, under which pronouns would be generated as
independent syntactic entities, and cliticized through some other mechansm. This
independent status is reserved here for emphatic pronouns like “moi”:
(18)
je- T
(19)
moi N
(↑ subj)=↓
((↓ pred)=‘pro’)
(↓ pers)=1
(↓ num)=sg
(↑ pred)=‘pro’
(↑ pers)=1
(↑ num)=sg
Every occurrence of “je-” defines a first person singular subj for the clause, but the
optionality of pred ensures that the same prefix can be used as an agreement marker.
Imagine that the revised TP rule in (20) is used to generate the structure in (21):
(20)
TP
−→
NP
T’
(↑ subj)=↓
6
For example, the sequence of a first person singular pronoun followed by either the periphrastic
future auxiliary or the main verb aller both have the pronunciation [Z@ vE], but only the sequence
whose second member is the auxiliary can be realized as [mA].
7
For example “Pierre il vient; Moi j’ai vu Pierre”, where “il” and “j’ ” agree with “Pierre” and
“moi”, respectively. Auger’s evidence that these are indeed cases of doubling rather than dislocation
is compelling. An important piece is that, when these sentences occur in embedded clauses, the
suspect NP follows the complementizer: “Il a fallu que Pierre il vienne”.
8
(21)

D
E 
pred
‘fall subj ’


tense passé composé 






pred ‘pro’ 



subj

pers 1


num sg
TP
(↑ subj)=↓
NP
↑=↓
T′
moi ↑=↓
T0
je-suis
↑=↓
VP
↑=↓
V′
↑=↓
V0
tomb-é
In (21), both “je-” in T0 and “moi” in SpecTP contribute pers and num to the
f-structure of the subject: the values assigned to these attributes match, or unify.
“Moi” also defines the pred value of the subject, but “je-” does not: pred values are
uniquely indexed semantic forms that cannot unify. If both “moi” and “je-” defined
a pred value for the subject, pred would have two different semantic forms as its
value, and would be in violation of Functional Uniqueness: every attribute has a
single value.
3
The Analysis of Wh-Interrogatives
Québécois French is particularly rich in possibilities when it comes to the formulation
of wh-interrogatives. The full range is shown in (22). Apart from (22a), which has
a clear discourse function, there is no clear difference in meaning between these constructions, though there are differences in register: (22b,e) are usable in the broadset
range of social contexts, (22c) is the most formal, and the rest have a looser, colloquial
feel.
(22)
a. Tu vois qui?
you.sg sees who
‘You see who?’
b. Qui tu vois?
‘Who do you see?’
9
c. Qui vois-tu?
d. Qui que tu vois?
e. Qui est-ce (que) tu vois?
f. Qui est-ce que c’est (que) tu vois?
g. Qui c’est (que) tu vois?
h. Qui c’est que c’est (que) tu vois?
i. Qui que c’est (que) tu vois?
Example (22a) shows an interrogative with the wh-pronoun in situ in object position,
and is most felicitous in echo questions.8
In (22b), the wh-pronoun has been dislocated to sentence-initial position. This
is the construction which would allow complex inversion9 in Standard French
(23)
Qui Pierre voit-il? (SF/*QF)
Complex inversion is not possible in QF, and indeed the use of subject-verb inversion
is extremely constrained: only second person pronouns can be inverted (see Picard
(1991, 1992) for discussion).10 The label “inversion” is in keeping with the standard
nomenclature, but does not characterize the phenomenon properly: recall from section 2.2 that pronouns are affixes, and that so-called inverted pronouns are merely
occurrences of pronoun suffixes. This kind of pronoun suffixation is shown (22c).
In all the remaining constructions, lexical material has been added. (22d) adds
what looks like a complementizer preceding the verb “vois”. In (22e), “est-ce que”
makes its first appearance, though the parentheses indicate that the “que” can be
omitted. Example (22f) shows the fullest extent to which wh-interrogatives can be
stretched, with “c’est que” embedded under “est-ce que”; the bottom “que” is again
optional. Examples (22g,h) show that “est-ce” in the predecing two examples can be
replaced by “c’est”.11 (22i) is like (22d) in having a complementizer precede the verb
“vois”, though here “c’est que”, where “que” is optional, intervenes between the two
as well.
8
Of the other constructions in (22), only (e) also allows the wh-pronoun to be in situ: “Est-ce
que tu vois qui?”
9
That is, the cooccurrence of a lexical NP subject and a post-verbal pronoun; see Rizzi and
Roberts (1989).
10
A form like est-ce would seem to be a counterexample to this claim, as it seems to feature an
inverted third person expletive pronoun, and can occur in wh-interrogatives as well as polar interrogatives. However, I will show later on in this section that est-ce cannot be analyzed synchronically
as featuring inversion, though it retains some of the properties of inverted forms.
11
This is not true of “est-ce” as it appears in polar interrogatives. Nor is it the case that “est-ce”
can replace “c’est” in all its occurrences here; this is discussed below.
10
The challenge for any analysis of QF wh-interrogatives is to cover the full range
of possible formulations. Because anti-extraction only happens in extraction configurations (by definition), I will limit my attention to (22b-k), which feature such
configurations.
Let me suggest that determining the syntactic status of “est-ce que” and “est-ce
qui” should be the starting point of this analysis: all of the relevant issues can be
tied to constructions that contain these sequences. I will use the terms core and
complementizer to refer to “est-ce” and “que/qui”, respectively. Complementizer is
not quite theory neutral, but since it will turn out that “que” and “qui” are exactly
that, I see no reason to introduce terminological confusion to an already explosive mix:
when the wh-pronouns are brought into the picture, we are left with two forms of both
“que” and “qui”. Each of these words has a pronominal form and a complementizer
one; the subscripts wh and c will serve to distinguish between the two.12
The discussion of the separability of core and complementizer will lead into a
discussion of the separability of wh-pronoun and core, which should not be taken for
granted. Then I will assign each of the words isolated to a syntactic category and
position, saving the last section for a formalization of the analysis.
3.1
Separability
A first clue that core+complementizer sequences are not lexicalized is that, as shown
in (22e-h), the core can be “c’est” as well as “est-ce”:
(24)
a. Qui c’est qui voit Pierre?
who it-is that sees P.
‘Who sees Pierre?’
b. Qui c’est que Pierre voit?
who it-is that P. sees
‘Who sees Pierre?’
If “est-ce que” and “est-ce qui” were lexical items, then a different set of lexical entries
would have to be posited to account for “c’est que” and “c’est qui”: either these two
sequences are themselves lexicalized, of a more analytic set of entries that could be
used to synthesize them. The latter option is given some credence by the optionality
patterns observed:
12
The pronouns and complementizers alternate according to different criteria. The alternation
in the form of the wh-pronouns is humanness-driven: human are referred to with quiwh , and nonhumans with quewh . The alternation in the form of the complementizers is discussed in detail in
Section 4.1. For now, let it be said that quic forces the dislocation of the subject of its clause, and
that quec forces the non-dislocation of the subject of its clause.
11
(25)
a. Qui (est-ce qui)/(qui) voit Pierre?
b. Qui (est-ce) (que) Pierre voit?
(26)
a. Qui (c’est qui)/(qui) voit Pierre?
b. Qui (c’est) (que) Pierre voit?
The complementizer is always optional when the core is absent, indicating at least
that quec and quic must have their own lexical entries. But (25a,26a) show that the
opposite is not always true: quic must be present when the core is. This would seem
to support the conclusion that “est-ce”, though it can be used independently of quec
when the object is extracted, must be supported by quic when the subject is extracted.
Thus “est-ce qui” would be lexicalized, but coexist with an independent lexical entry
for “est-ce”, itself restricted to sentences with object extraction. But I will show later
on that “est-ce” and quic can in fact be separated by parentheticals, and that the
hypothetical conclusion hinted at here is invalid: “est-ce qui” is decomposable in the
same way that “est-ce que” is.
This leaves the cooccurrence restriction between “est-ce” and quic in need of an
explanation; but let me suggest that quic is not the complementizer whose behavior
must be explained here. First I anticipate a conclusion to be reached in Section
3.2, that quec and quic in core+complementizer sequences in fact introduce finite
complement clauses subcategorized by the core. It is a well-known property of French
that the complementizer introducing finite complement clauses, unlike its English
counterpart, is obligatory. In this light, it is the optionality of quec that is deviant. But
only in appearance, as can be demonstrated by adding to the evidence the following
sentence, in which quec is obligatory:
(27)
Qui est-ce *(que) il a vu?
who is-it that he has seen
‘Who did he see?’
The difference between (25,26) and (27) is that quec is followed by a consonant in
each of the former, but by a vowel in the latter. On its own, this observation is
inconsequential, but to it must be added the liberality with which French, including
QF, allows optional schwa deletion (Anderson, 1982). Suppose that schwa deletion
has indeed applied to quec in all instances,13 creating a [kp] stop cluster in (25,26),
13
There is a simplification here, because the schwa of quec is in fact obligatorily absent in (27).
This is an effect of elision (Schane, 1968) whereby the final vowel of certain words is deleted when
those words are followed by vowel-initial words. Elision does not always target schwa — for instance
the [æ] of the feminine definite article is elided — but the crucial distinction here is that the [i] of
quic is not so targeted.
12
but a [kI] sequence in (27). The [k] is conserved when it is followed by a vowel, but the
stop cluster is (optionally) simplified to [p], effectively permitting the eradication of
all phonological evidence of the complementizer’s presence in just those cases where
it appears to be optional. In contrast, quic does not end in a schwa, and by definition
does not undergo schwa deletion; its [k] is therefore never in a cluster with the initial
stop of a following word, and is never deleted. Thus both quec and quic are obligatory
when they introduce the finite clausal complements required by cores, but the former
is such that phonological processes can delete all its segments in the appropriate
contexts.14
The main piece of evidence that core+complementizer sequences are separable
is that parentheticals can be inserted between the two:15
(28)
a. Qui est-ce, à ton avis, qui a vu Pierre?
who is-it to your opinion who has seen P.
‘Who, in your opinion, saw Pierre?’
b. Qui est-ce, selon toi, que Pierre a vu?
who is-it according you who P. has seen
‘Who, in your opinion, did Pierre see?’
(29)
a. ?*Qu’ est-ce, à ton avis, qui a frappé Pierre?
what is-it in your opinion that has hit P.
‘What, according to you, hit Pierre?’
b. ?*Qu’ est-ce, selon toi, que Pierre a fait?
what is-it according you that P. has done
‘What, according to you, did Pierre do?’
Whereas the example in (28) are fully grammatical examples of QF radio-speak,
those in (29) sound stilted and very Standard-French.16 The examples in (28) seem
14
This explanation predicts that quec will appear to be optional wherever it can be followed by a
stop consonant. In fact, the so-called optionality patterns shown for core+complementizer sequences
in (25,26) are exactly replicated when the core in “c’est”. Moreover, a declarative sentence like “Je
pense (que) tu dors trop” reveals that quec is subject to deletion in declarative sentences as well,
provided that the phonological conditions are met. That this phonological condition involves initial
stops, as opposed to initial consonants, is confirmed by the topmost quec in (22f,h,i): the initial
fricative of “c’est” is not sufficient.
15
Jones (1996:473) claims that insertion is impossible in this position. This claim is contrary to
my native speaker intuitions, but understandable given the very restricted character of the insertable
parentheticals. Here, the parentheticals transform the interrogatives from demands for information
about a particular situation to demands for the addressee’s opinion about that situation. Other
types of parentheticals, like temporal or locational adverbials and phrases, sound much less natural.
16
Though I make no claim as to their grammaticality in Standard French.
13
to establish conclusively that the quic and quec complementizers can be separated from
“est-ce” when the extracted wh-pronoun is quiwh . Why the nature of the extractum
should matter here is not transparent, but the following examples will perhaps help
clarify the matter:
(30)
a. ?Que c’est, à ton avis, qui a tué Pierre?
what it-is in your opinion that has killed P.
‘What, according to you, killed Pierre?’
b. ?Que c’est, selon toi, que Pierre a fait?
what it-is according you that P. has seen
‘What, according to you, did Pierre see?’
These parentheticals, with “est-ce” replaced by “c’est”, are repeated from (29), and
are also much better. Here is the outline of an explanation. Parentheticals have
special prosodic properties; in particular, they interrupt the prosodic contour of the
sentence in which they are inserted. What the better examples of parenthetical
insertion (28,30) have in common is that the pre-parenthetical bit of the sentence is
bisyllabic: [ki.jEs] for (28) and [k@.se] for (30). In contrast, the bad examples in (29)
only have the monosyllabic bit [kEjs] in front of the parenthetical. My suggestion is
that the prosodic requirements for parenthetical insertion are simply not met in (29):
the pre-parenthetical part of the sentence simply does not have sufficient prosodic
weight. This explanation is supported by changing the extractum in (29b) to a whNP:17
(31)
Quelle personne est-ce, selon toi, que Pierre a vu?
what is-it according you that P. has done
‘What, according to you, did Pierre do?’
An extracted wh-NP causes the pre-parenthetical part of the sentence to have sufficient prosodic weight, and to meet the prosodic requirements for parenthetical insertion.
Let me now turn to the separability of the wh-pronouns and the core. The
first point I want to make is that quiwh and quewh are not the only possible extracta.
Non-argument wh-pronouns (32), wh-NPs (33), and XPs with arguments or modifiers
containing wh-pronouns or -NPs (34), can also be extracted:
(32)
17
a. Où est-ce que Pierre a vu Marie?
where is-it that P. has see M.
‘Where did Pierre see Marie?’
A similar operation cannot be performed for (29a), since the wh-NP would be anti-extracted.
14
b. Quand est-ce que Pierre va revenir?
when is-it that P. go return
‘When will Pierre come back?’
(33)
a. Quelle personne est-ce que Pierre voit?
which person is-it that P. sees
‘Which person does Pierre see?’
b. Quel étudiant est-ce que l’ours a senti?
which student is-it that the bear has smelled
‘Which student did the bear smell?’
(34)
a. Pour qui est-ce que Marie a préparé un gâteau?
for who is-it that M. has prepared a cake
‘For whom did Marie prepare a cake?’
b. Le frère de qui est-ce qui est venu?
the brother of who is-it that is come
‘Whose brother came?’
c. L’ homme que Marie a présenté à qui est-ce qui est venu?
the man that M. has introduced to who is-it that is come
‘The man that Marie introduced to who came?’
The wh-pronouns in (32) show at the very least that there would have to be a larger set
of lexicalized sequences containing “est-ce” than the examples so far have suggested.
They can also be interpreted as requiring the existence of an entry for “est-ce” separate
from non-argument wh-pronouns, though they provide no basis for a conclusion to be
reached about the separability of “est-ce” and argument wh-pronons. Unsurprisingly,
(33) shows that “est-ce” can be independent of all kinds of wh-pronouns, since none
of those appear in the example sentences. Because wh-NPs are prohibitively productive, this establishes conclusively that there is a separate entry for “est-ce”. The
complement of “pour” in (34a), the PP modifying “frère” in (34b) and the relative
clause modifying “homme” in (34c) all contain wh-pronouns, though the examples
are inconclusive as to separability. But it is also possible to construct examples in
which the wh-pronoun is not adjacent to “est-ce”:
(35)
a. L’ homme que qui à vu est-ce qui est venu?
the man that who has seen is-it that is come
‘The man that who has seen came?’
b. L’ homme que quoi/*que à mordu est-ce qui est venu?
the man that what has bitten is-it that is come
‘The man that what has bitten came?’
15
In (35a), quiwh is the subject of the relative clause modifying “homme”, and the
remainder of that clause separates it from “est-ce”. This establishes that quiwh and
“est-ce” are independent from one another.
The example in (35b) reveals that quewh has a different status, since the appropriate wh-pronoun in this construction is in fact “quoi”. The standard observation
regarding the distribution of quewh and “quoi” is that the latter is the in-situ version of the former. The wh-pronouns in (35) are in subject position, SpecTP, which
explains the appearance of “quoi”.18 “Quep ” cannot replace “quoi” in these constructions, and cannot be shown through this test to be independent from “est-ce”,
though the conclusion that “est-ce” can be independent from the wh-pronouns is
further supported.
Some of the alternative formulations of wh-interrogatives shown in (22) also
provide relevant evidence:
(36)
quelle personne/qui/*que Pierre voit?
which person/who/what that P. sees
‘Which person/who/what does Pierre see?’
Example (36) shows that “est-ce” is not necessary in object questions with quiwh
or wh-NP extracta. But the absence of “est-ce” when the extractum is quewh is
ungrammatical.19 Again, this fails to establish that quewh is separable from “est-ce”.
Some evidence in favor of a wh-pronoun quewh independent from “est-ce” comes
from another possible formulation of wh-interrogatives:
18
Subject wh-pronouns are not compatible with agreement prefixes,
i. l’homme [CP que qui (*il) a vu ]
and cannot actually be shown to be in SpecTP as opposed to being affixes on T0 . But it is also true
that they cannot co-occur with lexical NPs or non-affix non-wh-pronouns, which would themselves
be in SpecTP and prove the affixal status of the in situ wh-pronouns:
ii l’homme [CP que [T P (*Pierre/lui) qui a vu ] ]
I suggest the following explanations for these incompatibilities: wh-pronouns have pred values,
which means they cannot occupy the same grammatical function, in this case subj, as other words
with pred values, such as lexical NPs, non-affix pronouns, or affix pronouns: different pred values cannot be assigned to the same f-structure, by Functional Uniqueness. Furthermore, non-whpronouns are more definite than wh-pronouns, a feature which is shared by agreement affixes derived
from non-wh-pronouns. Whether in-situ wh-pronouns subject are in SpecTP remains undetermined,
but I assume in this paper that they are.
19
Note that “quoi” is especially unacceptable in this sentence: “*Quoi (que) Pierre voit?”. My
explanation for this is that “quoi” has properties that prevent it from occurring in SpecCP. These
will be discussed further down.
16
(37)
a. Que c’est que Pierre voit?
what it-is that P. sees
‘What is it that Pierre sees?’
b. Que c’est qui voit Pierre?
what it-is that sees P.
‘What is-it that sees Pierre?’
Although I have regularized the orthography in (37),20 it appears as though quewh is
not entirely dependent on “est-ce”, since it is possible to find it with “c’est” as well.
Although this might be sufficient to conclude that quewh is an independent word, such
a conclusion would leave unexplained the impossibility of having quewh in (36). An
explanation for this is forthcoming, however, and I conclude in anticipation of this
that “quep is indeed an independent syntactic entity.
To recapitulate, the following syntactic units were isolated: quewh and a quiwh whpronouns, a quec and a quic complementizer, and finally “est-ce” and “c’est”. These
latter two items will be discussed extensively in the next section, and a discussion of
the rest will follow.
3.2
The Category of c’est and est-ce
The key to establishing the structure of wh-interrogatives is establishing the category
of “c’est” and “est-ce”. Superficially, these look a lot like the present tense form of
the copula, each with an expletive pronoun affix. I want to argue that this is the
proper analysis of “c’est”, but not of “est-ce”; rather “est-ce” is an inherently tensed
interrogative verb, with an inherent expletive subject.
It should be sufficient, to provide proof of the verb-hood of “c’est”, to present
evidence that it can inflect following the regular paradigms:
(38)
a. Qui c’était que Fred appelait?
b. ?Qui c’a été que Fred a appelé?
c. ?Qui c’avait été que Fred avait appelé?
d. Qui çà va être que Fred va appeler?
The strangeness of (38b,c) can be eliminated by neutralizing the strangeness of the
tense sequencing:
20
This is a more colloquial formulation of wh-interrogatives. The spelling one might find in plays
and novels would most likely be either “quessé” or “quossé” (although initial “k” is possible as well),
with the actual pronoun seemingly ambiguous between quewh and “quoi”.
17
(39)
b. Qui c’ a été qui a mal interprété ton commentaire, déjà?
who it has been that has badly interpreted your comment again
‘Who was it that misinterpreted your comment, again?’
c. Où c’ avait été que Fred nous avait emmenés, déjà?
where it had been that F. us.acc had brought again
‘Where was it that Fred had brought us, again?’
“C’est” can thus inflect following the regular paradigms. The necessary conclusions
are that (i) “c’ ” and “est” can be separated, (ii) “est” is just one of the inflected
forms of the copula verb “être”, which can take a sentential complement, and (iii)
“c’ ” is the reduced form of the expletive subject pronoun prefix “ce-”. From my
earlier conclusions, it follows that any tensed form of the copula (“est”, “était”), and
any auxiliaties that occur with its non-tensed forms, are in T0 . Non-tensed form of
the copula are in V0 , and “ce-” is a pronoun prefix:
(40)
est T (↑ pred)=‘be < comp > subj ’
(↑ tense)=present
((↑ int)=+)
(↑ comp int)=−
(41)
être/été V
(42)
ce- T
(↑ pred)=‘be < comp > subj ’
((↑ int)=+)
(↑ comp int)=−
(↑ subj)=↓
(↓ pred)=‘pro’
(↓ pers)=3
(↓ num)=sg
Note that both finite and nonfinite forms of the copula verb are optionally interrogative, and that their complements are defined to be non-interrogative via the
assignment of a negative int feature. Both of these properties will turn out to be
crucial to the analysis.
As for “est-ce”, it is ungrammatical to put it in the imparfait (43a), the passé
composé (43b), the plus-que-parfait (43c), or the periphrastic future (43d), the only
indicative inflectional paradigms in QF beyond the present tense:21
(43)
a. *Qui était-ce que Fred appelait?
who was-it that F. called
‘Who was Fred calling?’
21
The passé simple, passé antérieur, futur simple and futur antérieur have essentially disappeared
from the dialectal usage.
18
b. *Qui a-ce été que Fred a appelé?
who has-it been that F. has called.
‘Who has Fred called?’
c. *Qui avait-ce été que Fred avait appelé?
who had-it been that F. had called
‘Who had Fred called?’
d. *Qui va-ce être que Fred va appeler?
who go-it be that F. go call
‘Who is Fred going to call?’
Because “est-ce” does not inflect, it is not possible to determine whether “-ce” is a
pronoun suffix or merely a grammaticalized subject marker: it simply cannot occur
on forms other than “est”. What evidence is there, then, that “-ce” has been grammaticalized into “est”, and that the resulting “est-ce” form is a verb? The relevant
evidence comes from the distribution of suffix pronoun subjects.
Suffixation of subject pronouns is much less productive in QF than it is in
Standard French: only second person subjects can be pronoun suffixes at all:22
(44)
a. Sais -tu si Pierre est venu?
know you.sg whether P. is come
‘Do you know whether Pierre came?’
b. Venez -vous ce soir?
come you.pl this evening
‘Are you coming this evening?’
c. *Sort -il les déchets chaque vendredi?
take-out he the garbage each Friday
‘Does he take out the garbage each Friday?’
Those suffix pronouns that do occur are barred from embedded clauses:
(45)
a. *Pierre pense que es-tu invité.
P. thinks that are-you.sg invited
‘Pierre thinks that are you invited?’
b. *Marie se demande si attendez-vous quelqu’un.
M. refl. asks whether wait-you.pl someone
‘Marie wonders whether are you waiting for someone?’
22
See Picard (1991, 1992) for a diachronic account of this phenomenon.
19
Suffix pronouns cannot occur in embedded declaratives (45a) or embedded interrogatives (45b). I want to argue that there is a restriction on suffix pronouns, that they
cannot appear in embedded clauses; a historical residue, present in both the Standard and Québécois dialects of French, from an earlier state of the language in which
inversion of subject pronouns and verbs was a productive syntactic process, as opposed to the restricted (at least in QF) morphological phenomenon it is today. Much
like subject-auxiliary inversion in English today, this phenomenon was restricted to
matrix clauses.
Embedding tests can now be performed to determine the status of “est-ce”, and
of its “-ce” component: if “est-ce” is an embedding predicate, it should follow that
suffix pronouns cannot occur in its complement; if “-ce” is a suffix pronoun, it should
never occur in embedded clauses.
(46)
a. *Qui est-ce que as-tu vu?
who it-it that have-you.sg seen
‘Who did you see?’
b. *Qu’ est-ce que voyez-vous?
what is-it that see-you.pl
‘What do you see?’
These examples demonstrate that the verbs bearing “-tu” and “-vous” are embedded
under some predicate.23 The only likely candidate for that role is “est-ce”. Note
that wh-interrogatives with embedding under “c’est” or “est-ce” do not differ at all
in meaning from wh-interrogatives without such embedding.24 Because of this, I will
refer to interrogative embedding under these verbs as dummy embedding.
On the other hand, “-ce” is demonstrably not a suffix pronoun because “est-ce”
can occur in embedded clauses:
(47)
a. Je me demande qu’ est-ce que Pierre fait.
I self ask what it-is that P does
‘I wonder what Pierre is doing?’
b. Sais -tu qu’ est-ce que Pierre fait?
know you.sg what it-is that P. does
‘Do you know what Pierre is doing?’
23
Suffix pronouns are also ungrammatical in a matrix question if this question has a complementizer: “*Qui que vois-tu?” This does not indicate that the source of the ungrammaticality of the
sentences in (46) is an incompatibility between suffix pronouns and wh-questions: (22c), which has
both of the putatively incompatible elements, is grammatical. Another explanation must be sought
for for a sentence like the above. I have none at the moment.
24
Formally, this would follow if the lambda term for “est” were λP.P , where P is an expression
of type t. This is thus a function that returns its input intact as an output.
20
c. Qui est-ce qui se demande qu’ est-ce que Pierre fait?
who is-it that self asks what is-it that P. does
‘Who wonders what P. does?’
It is possible to embed “est-ce” under any verb that can take a wh-interrogative
complement, like “se demander” (47a) and “savoir” (47b), even when this verb is
itself embedded (47c). This indicates that “-ce” is not a suffix pronoun on par with
“-tu” and “-vous”. My suggestion is that it has been grammaticalized into its former
stem “est”, and lost some of its previous properties, such as the restriction against
embedding, but kept some others, like the ability to define an expletive subject.
However, “est-ce” cannot be embedded just anywhere:
(48)
a. *Qui est-ce que est-ce que Pierre a vu?
who is-it that is-it that P. has seen
‘Who did Pierre see?’
b. *Qui c’ est qu’ est-ce que Pierre a vu?
who it is that is-it that P. has seen
‘Who did Pierre see?’
It is not possible to embed “est-ce” directly under dummy embedding verbs like itself
(48a) and “c’est” (48b). I suggest that this is due to restrictions on the possible
complements of dummy embedding verbs: they cannot be interrogative. “Est-ce”
cannot be dummy embedded because it is inherently interrogative; “c’est” can because
“est” is only optionally interrogative. I suggest the following lexical entry for “est-ce”:
(49)
est-ce
T (↑
(↑
(↑
(↑
pred)=‘be < comp > subj ’
tense)=present
int)=+
subj)=↓
(↓ pred)=‘’
(↓ pers)=3
(↓ num)=sg
(↑ comp int)=−
Synchronically, “est-ce” is an interrogative, present tense (and thus T0 ) verb, which is
inherently specified for an expletive subject and requires a non-interrogative clausal
complement.25
25
It may seem strange for a verb to define its own subject, as opposed to this being accomplished
by, for example, verbal morphology. However, compare the argument in Morin (1985) that voici and
voilà are subjectless verbs in French, and the GB analysis of this phenomenon in Bouchard (1988).
21
3.3
Structure
Now that it has been established that “est-ce” and “c’est” are in T0 , their occurrence
can be used as a diagnostic for the left edge of an T’ constituent. Moreover, a sentence
with a lexical subject, such as “Pierre voit Marie” is rooted in TP. Therefore, the
following bracketing can be determined for a sentence like “Quand est-ce que Pierre
voit Marie” (“When is Pierre seeing Marie?”), where only the crucial constituent
boundaries are indicated:
(50)
Quand [T ′ est-ce que [T P Pierre voit Marie? ] ]
It is possible to determine from this that the extractum, “quand”, is outside of the
topmost T’ constituent, and that quec is outside of the lower TP. The occurrence of
these words in a monoclausal interrogative sheds further light on their location in the
structure:
(51)
Quand que [T P Pierre voit Marie? ]
If “quand” and quec are outside of TP, but still within the matrix clause, then they
must be somewhere in CP. The two available positions are SpecCP and C0 . Because
the former linearly precedes the latter and “quand” linearly precedes quec , it must be
the case that “quand” is in SpecCP, and quec in C0 . These tests can be replicated
for all possible extracta, whether they are wh-NPs (52a), wh-NPs in object position
of prepositions (52b), NPs with a modifier containing any wh-expression (52c,d), and
all wh-pronouns except “quoi” and quewh (52e-j):
(52)
a. Quelle homme que Pierre a vu?
which man that P. has seen
‘Which man did Pierre see?’
b. À quel endroit que Pierre a vu Marie?
to which place that P. has seen M.
‘In which place did Pierre see Marie?’
c. Le frère de qui que Pierre a vu?
the brother of who that P. has seen
‘Whose brother did Pierre see?’
d. L’ homme qui a dit quoi que Pierre a vu?
the man that has said what that P. has seen
‘The man who said what did Pierre see?’
e. Qui que Pierre a vu?
who that P. has seen
‘Who did Pierre see?’
22
f. *Que que Pierre a vu?
what that P. has seen
‘What did Pierre see?’
g. *Quoi que Pierre a vu?
what that P. has seen
‘What did Pierre see?’
h. Où que Pierre a vu Marie?
where that P. has seen M.
‘Where did Pierre see Marie?’
i. Quand que Pierre a vu Marie?
when that P. has seen M.
‘When did Pierre see Marie?’
j. Comment que Pierre a vu Marie?
how that P. has seen M.
‘How did Pierre see Marie?’
I conclude from this that quec is in C0 , and that all extracta are in SpecCP, though
special measures must be taken to account for the ungrammaticality of “quoi” and
quewh in this position. Assume the following lexical entry for “quoi”:
(53)
quoi
N (↑
(↑
(↑
(↑
pred)=‘pro’
type)=nonhuman
extr)=−
int)=+
“Quoi” is negatively specified for the extr feature, barring it from positions that
require a positively specified extr feature. SpecCP will be such a position, but
argument positions will not.
As for quewh , I have shown in examples (35-37) above that it can only occur
immediately preceding “est-ce” or “c’est”. This would be accounted for if quewh was
in SpecTP, but this cannot be the case since (35) showed that quewh cannot be in-situ.
A possible solution is that quewh is really a prefix on T0 , and is thus bound to that
position. It would follows from this that quewh cannot precede the C0 position. But
this would also predict that quewh could follow the C0 and SpecTP positions, which
is decidedly not the case:
(54)
a. *quec Pierre quep a vu?
b. *Pierre quep a vu?
23
Let me suggest instead that quewh is indeed an independent word, with the following
lexical entry:
(55)
quep
N (↑
(↑
(↑
(↑
pred)=‘pro’
type)=nonhuman
extr)=c +
int)=c +
The last two equations, with the subscript “c” after the equality sign, are constraining
equations. Much like the existential equations of (12,15), they constrain that there be
an attribute of the kind stated, but on top of that they state that this attribute must
have a particular value. “Quep ” is barred from argument positions because it needs
a positive extr feature, which is provided only by SpecCP. Further, quewh needs a
positive specification for the int feature, which will be provided by “est-ce” or the
interrogative version of “est”, effectively forcing quewh to occur only when those verbs
are present.26
Contrast these last two lexical entries with that of a wh-pronoun like quiwh ,
which has no restrictions on its distribution:
(56)
qui
N
(↑ pred)=‘pro’
(↑ type)=human
(↑ int)=+
“Quip ” is not restricted to any particular extr value; it can thus occur in both
SpecCP and argument positions. Furthermore, it carries its own positive int feature,
and does not need to cooccur with an interrogative word, though it can.
Of the words isolated in Section 3.1, only quic remains unassigned to a category.
In contrast to quec , quic cannot be shown to necessarily occur in C0 , or even outside
of TP. “Quic ” signals subject extraction, and the subject status of the extractum.
Because of this, no positive evidence that quic is not in SpecTP can be adduced,
though no positive evidence that is is necessarily in SpecTP can either. In (57),
either the X bracket or the Y bracket could mark the edge of TP:
(57)
a. *[CP Qui [X qui [Y Pierre voit Marie? ] ] ]
b. *[CP Qui [X qui [Y [T ′ il-voit Marie? ] ] ] ]
c. [CP Qui [X qui [Y [T ′ voit Marie? ] ] ] ]
26
This leaves the problem of in situ wh-pronouns, which are in principle able to make a clause
interrogative at a distance: “Pierre a dit que quoi a mordu le chien?” is fine, while “*Que Pierre
a dit que quoi a mordu ?” is awful. Also, nothing so far prevents “*Quep quec est-ce que Pierre
a vu?” “*Quep quic est-ce...” is ruled out because quic forces the subject of its clause to be the
extractum, thereby causes a clash between quewh and the inherent subject specified by “est-ce”.
24
The ungrammaticality of example (57a) is explained away by the impossibility of
having both the wh-pronoun and “Pierre” be the subject of the same sentence. This
only provides evidence that an explanation under which quic is in SpecTP, thereby
preventing the occurrence of a lexical subject NP in that position, is redundant. It
does not settle whether the edge of TP is at X or at Y.
Example (57b) is ungrammatical if the pronoun use of the “il-” subject prefix is
involved, because “il-” and quiwh cannot both be subject. The same example is also
ungrammatical if “il-” merely marks agreement with the fronted subject, because ‘il-”
and the wh-pronoun do not match in definiteness.27 In either instance, the edge of
TP could be at X or at Y.
The possibility that quic occurs in SpecTP thus remains open. In fact, I am
aware of no tests that could settle the issue: the final decision must be based on
purely theoretical factors. The theoretical button I want to press is the categorial
difference between the SpecTP and C0 positions: the former is associated with nominal expressions, or at least things that can be subjects of sentences, while the latter
is associated with function words. It must therefore be possible to determine the
position of quic by determining whether it functions more like a nominal, or more
like a function word, say by determining whether or not it is referential, or whether
it can occur in contexts in which it would be unlawfully bound. This sounds all well
and good in principle, but in practice things are a bit different: since there is a referential lexical item homophonous with quic , mainly quiwh , any referential instance
of quic could in fact be an instance of quiwh . And, of course, to determine whether
something is bound, one must know whether that thing is referential, and what it
refers to. Even if that could be determined for quiwh , the fact that it is a third person
pronoun would ensure that it could always be coreferential with some entity outside
of its binding domain, that it could always be construed as unbound, and that its
occurrence would always be grammatical under some reading.
I do, however, have one argument, a phonological one, for treating quic as a
function word: it involves the oft-made claim, justified at length in Selkirk (1972),
that function words are more often phonologically reduced than lexical words. Take
the following two sentences:
(58)
a.
b.
Qui est venu?
Qui qui est venu?
[ki.je.vny]
[kje.vny]
*[kje.vny]
*[ki.je.vny]
The boldfaced sequences in (58) are those that interest us. When their words are produced in isolation, these sequences are perfectly homophonous. In running speech,
a difference emerges: the boldfaced sequence in (58a), whose initial word is quiwh ,
is pronounced [ki.je.vny], while the one in (58b), whose initial word is pronounced
27
See fn. 18.
25
[kje.vny]. Moreover, the two pronunciations cannot be exchanged: it is ungrammatical to pronounce (58a) [kje.vny], or to pronounce (58b) [ki.ki.je.vny]. Thus, when
they are put in a comparable phonological context, quiwh and quic behave differently:
the former cannot be reduced, while the latter necessarily is. I take this as evidence
that quiwh is lexical, and that quic is functional.
3.4
Formalization
All these findings now need to be put together. I have provided lexical entries along
the way, but c-structure rules still need to be formulated. It will perhaps help to start
with a discussion of the kind of f-structure that should be generated.
The LFG analysis generally assumed to hold for wh-interrogatives is that of Kaplan and Bresnan (1982). This analysis assumes that the f-structure of the extractum
is the value of a focus attribute in the outermost f-structure, and that the f-structure
of the question-word is the value of the q attribute, also in the outermost f-structure.
The second assumption is not redundant: in instances of pied piping, it ensures that
the f-structure of the question word is the value of at least one attribute, mainly q,
in the outermost f-structure:
(59)
Le frère de qui est-ce qui est venu?
“Whose brother

D came?”
E
pred


int



q






focus








comp
‘be comp subj ’



+


"
#

pred ‘pro’



int
+




pred
‘brother’

 



pred ‘of’


h
i

oblposs 


obj




D
E


pred
‘come
subj
’



h
i



subj
In (59), the value of focus is identified, or structure-shared, with the value of obl,
as indicated by the connecting line. The f-structure of the question-word “qui” is
embedded in the focus f-structure, but is structure-shared with the value of q in
the outermost f-structure. Under this analysis, a non-pied-piped extractum is merely
a special case in which the values of q and focus are exactly the same. This is an
interesting consequence, but there are empirical and conceptual reasons to think that
26
this analysis is flawed.
The usual assumption is that focal elements constitute new and prominent information in the discourse (Bresnan and Mchombo, 1987; King, 1995; Butt and King,
2000). The idea that focus is the attribute that holds the value of the question-word
is intuitively attractive: after all, the intended referent of question-word can hardly
be part of the asker’s context if the question about it needs to be asked. But the
implementation of Kaplan and Bresnan (1982) comes up short: discourse newness
and prominence are criteria that pied piped extracta, bracketed in (60), do not meet:
(60)
a. [ Pour qui ] est-ce que Marie a préparé un gâteau?
for who is-it that M. has prepared a cake
‘For whom did Marie prepare a cake?’
b. [ Le frère de qui ] est-ce qui est venu?
the brother of who is-it that is come
‘Whose brother came?’
c. [ L’ homme que Marie a présenté à qui ] est-ce qui est venu?
the man that M. has introduced to who is-it that is come
‘The man that Marie introduced to who came?’
In (60a), the fact that Marie baked a cake for someone is assumed, and the question is
asked about who exactly the person is that Marie baked a cake for. In (60b), the fact
that someone’s brother came is assumed, and a question is asked about exactly whose
brother that was. In (60c), it is being assumed that a man that Marie introduced to
someone came, and the question is about who exactly this man was introduced to.
In all these questions, only the boldfaced part of the extractum is being questioned,
and thus only that part really belongs in focus. But I am not suggesting that the
roles of focus and q in the Kaplan and Bresnan analysis merely be switched. What
I want to question is the validity of having these two attributes at all.
The question-word has a double function, that of interrogative focus, and whichever
grammatical function (gf) it fills lower down in the clause. It therefore makes sense
for the f-structure of the question-word to occur twice in the f-structure, once in focus and another wherever it holds a gf. But the pied piped extractum only really
fills some grammatical function down in the clause; its role in interrogation merely
to serve as the vehicle that fronts an otherwise unfrontable word. Witness:
(61)
a. *Qui est-ce que Marie a préparé un gâteau pour
b. *Qui est-ce que/qui le frère de
?
est venu?
c. *Qui est-ce que/qui l’homme que Marie a présenté à
27
est venu?
Constraints on extraction prevent the question-word from being fronted alone in these
sentences; in effect, the pied piped extracta are wh-islands, in the sense of Ross
(1967), out of which question-words cannot be extracted. Fronting the question-word
requires fronting the entire island. I surmise that the f-structure of the pied-piped
extractum is not structure-shared between the value of some focus (in the Kaplan
and Bresnan (1982) sense) or q attribute in the matrix f-structure and the value of
it base grammatical function, but is rather uniquely the value of that grammatical
function. Only the question-word is structure-shared at all. The f-structure of (59)
is modified in (62) to reflect this notion:
(62)
Le frère de qui est-ce qui est venu?
“Whose brother

D came?”
E
pred

int



focus








comp




‘be comp subj ’
+
"
pred
int

pred





subj


#
‘pro’
+
D
E
‘come subj ’

pred
‘brother’



pred ‘of’

h
oblposs 
obj






















i 



The c-structure position of the pied-piped extractum does not correspond to its fstructure position: it is near the top of the former structure, but embedded deep in
the latter. I surmise that an analysis along these lines is not only truer to the data, but
also closer to the spirit of LFG than the analysis of Kaplan and Bresnan (1982), since
it preserves a relationship between the c- and f-structure of wh-interrogatives that
is operative between their surface structure and their deeper grammatical relations,
instead of recasting the c-structure in f-structure terms.28
The question now is how to get this to follow from the c-structure rules. I have
established above that the extractum is in SpecCP, and shown that it can be extracted
from embedded clauses. Furthermore, the question-word can be embedded deep into
the extractum itself. The problem, in a nutshell, is to assign this extractum to an
arbitrarily embedded grammatical function from SpecCP position, and to retrieve
28
The analysis of wh-interrogatives I present here makes predictions about binding and crossover
phenomena, given the standard conceptions of those. I will not concern myself here with determining
whether those predictions are in fact borne out, or whether different analyses of these phenomena
would be required.
28
the question-word from arbitrarily far inside this extractum so that it can be placed
in focus. The attribute path leading down through the f-structure to the gf of
the extractum, and the path leading from that f-structure on down to the gf of the
question-word cannot be fixed, because they can change from sentence to sentence.
Assume this c-structure rule for CP:
(63)
CP
−→
ExtrP
(↑ ExtrPath)=↓
(↑ focus)=(↓ FocusPath)
(↑ int)=c +
(↑ int)=(↑ focus int)
(↑ focus extr)=+
C’
↑=↓
Here I assume that ExtrP is a metacategory that subsumes the c-structure categories
that extracta can occupy. A definition for this metacategory can be found in the
appendix.
ExtrPath and FocusPath stand for the attribute paths through the f-structure
that lead to the extractum and the question-word, respectively. ExtrPath picks out
a grammatical function and assigns the f-structure of the fronted constituent to it,
while FocusPath reaches inside the f-structure of that fronted constituent to find
a second grammatical function to equate with the Focus attribute of the outermost
f-structure. Should the sentence feature no pied piping, this second grammatical
function will be the very same that is picked out by ExtrPath, and FocusPath
will resolve to the empty string. In (62), ExtrPath resolves to comp subj, the
f-structure that corresponds to the fronted constituent “le frère de qui”, and FocusPath starts from comp subj to pick out obl obj, the f-structure that corresponds to
the question-word quiwh . Each of these functional annotations is a functional uncertainty (Kaplan and Zaenen, 1989): the functional annotation does not state exactly
what the path is, but still constrains what a possible resolution might be, so it can
resolve differently in different sentences. Definitions for both of these uncertainties
can be found in the appendix.
The remainder of the annotations on the ExtrP node of (63) are no less crucial.
First, the int value of the clause is constrained, but not defined, to be positive: this
value must be assigned from somewhere else. Second, the int value of the outermost
f-structure is equated with the int value of the focus; this ensures that the positive
int value of “est-ce” or “c’est” is passed down to quewh , which requires such a value,
and that the positive int value of other question-words is passed up to the sentence,
which also requires this value. These two equations also interact to constrain the
value of focus int to be positive, so that only question-words can be focussed via
the CP rule. The last annotation assigns a positive extr value to the f-structure of
focus; “quoi”, which has a negative extr value, cannot be used in this position,
29
because its negative value clashes with the positive one assigned by the ExtrP node.
But quewh , on the other hand, requires just such a positive extr value, and fulfills
this requirement by being fronted; the extraction of quewh can be forced if we assume
that positive extr values are not assigned to the f-structures of words in argument
positions.
In short, the CP rule in (63) interacts with the lexical entries of wh-pronouns to
create just the right kind of f-structure in just the right set of cases. But what ensures
that wh-NPs have the same distribution as wh-pronouns? Assume the following
schematic NP rule and lexical entries for a wh-determiner and common noun:
(64)
NP −→
(65)
quelpre
(66)
homme N
Det
N
↑=↓ ↑=↓
Det (↑ int)=+
(↑ pred)=‘man’
Thus the NP rule equates the f-structures of determiners and nouns. The effect is
that a common noun with a wh-determiner becomes an interrogative noun, which
fulfills the requirements of the focus function as defined by the CP rule.
4
A Closer Look at Anti-extraction
With this analysis of wh-interrogatives as a tool, it becomes possible to investigate
the anti-extraction phenomenon illustrated in (3), which I reintroduce here:
(67)
a. Qui est-ce que Marie voit?
who is-it that Marie sees
‘Who does Marie see?’
b. Quelle personne est-ce que Marie voit?
which person is-it that Marie sees
‘Which person does Marie see?’
c. Qui est-ce qui voit Paul?
who is-it that sees Paul
‘Who sees Paul?’
d. *Quelle personne est-ce qui voit Paul?
which person is-it that sees Paul
‘Which person sees Paul?’
30
The problem is that, even though both wh-pronouns and wh-NPs can be extracted
from object position (67a,b), wh-NPs cannot be felicitously extracted from subject
position: compare (67c) to (67d). There are two questions here: what makes subject
position special, and what makes wh-NPs special. I will tackle each of these questions
in turn, then propose an analysis of anti-extraction.
4.1
What Makes Subject Position Special
Extraction from subject position differs from other kinds of extraction in one significant way: when extraction takes place across a complementizer, quic is used when the
extraction site is in subject position, and quec appears otherwise. The complementizers are not interchangeable. This suggests that they have lexical properties that force
this behavior. In this section, I will investigate what these lexical properties could
be, and propose that the lexical properties of quic must shoulder at least part of the
burden for the ungrammaticality of questions like (67d). But I will go on to show that
quic is not entirely responsible for anti-extraction: interrogative cleft constructions
provide evidence that the specifier position of CP must also be involved.
The que/qui Alternation
A well-known problem in French syntax, to which QF is no exception, is the alternation between quec and quic . The example pairs in (67a,b) and (67c,d) illustrate this
phenomenon. It is readily observable that the occurrence of quec corresponds with
a gap in object position, and the occurrence of quic with a gap in subject position.
This distribution is complementary: the following examples show that quec cannot be
used with gaps in subject position, and that quic cannot be used with gaps in object
position.
(68)
a. *Qui est-ce que voit Paul?
who is-it that sees P.
‘Who sees Paul?’
b. *Qui est-ce qui Marie voit
who is-it that M. sees
‘Who does Marie see?’
?
The generalization made above must however be amended to cover the following data:
(69)
a. Pierre croit que Marie voit Paul.
P. believes that M. sees P.
‘Pierre believes that Marie sees Paul.’
31
b. *Pierre croit qui Marie voit Paul.
These examples show that quec can occur when no gap is present, but that quic
cannot. The correct generalization thus seems to be that quic occurs with a gap in
subject position, while quec occurs elsewhere, where elsewhere means either with a
gap in nonsubject position, or no gap at all.
Inside-out functional uncertainty (Kaplan and Zaenen, 1989) provides the means
for capturing this generalization. The functional uncertainty used in the CP rule (63)
works from the outside-in: the value of a higher attribute is equated with the value
of another attribute embedded arbitrarily far down into the f-structure. Inside-out
functional uncertainty is just the opposite: the value of an embedded attribute is
equated with the value of another attribute arbitrarily far up into the f-structure.
Suppose that the complementizers are associated with the following, preliminary lexical entries, in which gf stands for any grammatical function, and the superscrpit
asterisk is the the Kleen closure operator, indicating that the superscripted attribute
can be repeated any number of times, includng none:29
(70)
quecpre :
C ((gf∗ ↑) focus) 6= (↑ subj)
(71)
quicpre :
C ((gf∗ ↑) focus) = (↑ subj)
The equation in (70) is satisfied if and only if the f-structure corresponding to the C0
node, which contains quec , is such that it does not contain a subj attribute that has
a value token-identical to the value of some focus attribute. The equation in (71) is
satisfied if and only if the f-structure corresponding to the C0 node, which contains
quic , is such that it does have a subj attribute whose value is token-identical with
the value of some focus attribute. The following illustrates the targeted structure:


h
i
(72)
focus
h
i


...
subj
Here, the dots stand for an attribute path of any length, including zero (that is, it is
possible for focus and subj to be attributes of the same f-structure, or for the subj
to be embedded arbitrarily far down into the f-structure that focus is an attribute
of). The constraint in (70) is violated by an f-structure subsumed by (72), while the
constraint in (71) states that an f-structure subsumed by (72) must exist. It should
be clear that these constraints will generate the distributional patterns shown in (67),
regardless of the degree of embedding of the complementizers. This is indeed what is
observed:
29
The subscript pre indicates that the entries will eventually be revised. Note that a constraint
like that in (70) is also used by Falk to account for the “that”-trace effect.
32
(73)
a. Qui que/*qui Marie voit?
who that Marie sees
‘Who does Marie see?’
b. Qui qui/*que voit Paul?
who that sees Paul
‘Who sees Paul?’
c. Qui est-ce que Marie a dit que Paul sait que/*qui Pierre voit?
who is-it that Marie has said that Paul knows that Pierre sees
‘Who did Marie say that Paul knows that Pierre sees?’
d. Qui est-ce que Marie a dit que Paul sait qui/*que voit Pierre?
who is-it that Marie has said that Paul knows that sees Pierre
‘Who did Marie say that Paul knows sees Pierre?’
In (73a,b), the extractum is an argument of the only verb of the sentence, which it
seems to make sense to assume is not embedded, though a complementizer is present.
In (73c,d), the extractum is an argument of a verb embedded two clauses down.
The patterns noted above still hold, and the constraints in (70,71), which are lexical
properties of the complementizers, generate them.
The constraints in (70,71) must, however, be adjusted to cover cases in which the
focus is in fact embedded inside the extracted subj function. These are sentences
like the ones in (60), for which the FocusPath is not the empty string. I propose
the following, revised lexical entries for quec and quic :30
(74)
quec :
(75)
quicpre :
C ((gf∗ ↑) focus) 6= (↑ subj gf∗ )
C ((gf∗ ↑) focus) = (↑ subj gf∗ )
Each of these constraints is a concatenation of an inside-out functional uncertainty,
on the lefthand side of the equation, and an outside-in functional uncertainty, on the
righthand side of the equation. They state that a relationship of equality must hold
(75), or not hold (74), between the focus attribute of an f-structure arbitrarily far
above the f-structure of the complementizer, and an f-structure embedded arbitrarily
far down into the subj attribute of the f-structure of the complementizer.
The Culprit
Examining the sentences of (67) yields arguments in favor of not treating antiextraction as resulting from the presence of a CP barrier between extractum and
gap. This kind of explanation has been offered in the transformational literature (for
30
Notice that the entry for quic is still preliminary.
33
instance, Chomsky 1986) for phenomena as wide-ranging as the “that”-trace effect
(CP is a barrier to proper government of the trace), and the behavior of control verbs
(CP is a barrier to illicit government of PRO). Because only the sentence in (67d)
is ungrammatical, though all four of them have overt complementizers which necessarily head CPs, an explanation along these lines would be insufficient. Especially
damning to the prospect of such an approach to the problem of anti-extraction is
the evidence that the phenomenon takes place even when no CP bars the relation
between extractum and gap. In the following c-structure, the functional annotations
have been removed, and the gap has been made explicit by an e:
(76)
CP
C′
SpecCP
qui
C0
qui
TP
T′
NP
e
T0
VP
voit
V′
NP
P ierre
Here, “quip ” is in specifier position within the CP, which can then hardly be said to
intervene between the pronoun and its corresponding gap. Consider now the following
example, parallel to (76), but with an NP extractum:
(77)
*Quelle personne qui voit Paul?
which person that sees Paul
‘Which person sees Paul?’
Crucially, this sentence is ungrammatical. Anti-extraction crops up though no barrier
to government is present. The barrier explanation can therefore not account for this
phenomenon.
Another step towards determining whether anti-extraction is an effect of cstructure configuration is to show that the same sentential constituent structure does
not always cause this effect. This can be done by comparing (67c) and (67d), whose
structures are shown, packed, as (78):
34
(78)
CP
SpecCP
C′
qui/
*quelle personne
TP
T′
T0
VP
est-ce
V′
CP
C′
C0
qui
TP
T′
NP
e
T0
VP
voit
V′
NP
P ierre
Both of these sentences feature extraction from subject position. They are structurally
identical except for the extractum, which is a wh-pronoun in one case, and a wh-NP in
the other. When the extractum is a wh-pronoun, the sentence is grammatical; when
it is a wh-NP, the sentence is ungrammatical. It seems clear from these examples
that the constraints responsible for (67d), whatever they are, cannot be constraints
on sentential constituent structures.
The possibility that the structure of the extractum is responsible is open, of
course, but this alone could not explain what we are observing, since (67b) and (67d)
both have wh-NP extracta, but differ in grammaticality. The essential difference
between these two examples is the extraction site: subject vs. object position. This
difference corresponds to a difference in complementizers, and in grammaticality: quec
is used for object extraction, while quic is used for subject extraction. This points to
quic as the factor responsible for the ungrammaticality of (67d).
35
The smoking gun is the possibility, illustrated up in (22b) of not using the
complementizer in monoclausal interrogative constructions:
(79)
a. Qui voit Paul?
who sees Paul
‘Who sees Paul?’
b. Quelle personne voit Paul?
which person sees Paul
‘Which person sees Paul?’
Example (79b) differs from (77) in two respects: it lacks quic , and it is grammatical. This seems to indicate that the presence or absence of quic is crucial to the
phenomenon of anti-extraction of wh-NPs from subject position.
Cleft Constructions
Here, I consider the evidence that wh-NP can be clefted from subject position,
and that this clefting in fact involves extraction. The conclusion will be that antiextraction of wh-NPs results from extraction from subject position to SpecCP position.
The cleft construction can be used in French, like in English, to place discourse
emphasis on a particular constituent:
(80)
a. C’est [ lui ] qui est venu.
it-s him that is come
‘It is him who came?’
b. C’est [ pour les Canadiens ] que Pierre a joué.
it-is for the C. that P. has played
‘It is for the Canadiens that Pierre played.’
In example (80a), it is assumed that a particular person came, and asserted that this
person is in fact Pierre. Example (80b), can be understood in two different ways:
either there is an assumption that Pierre played for a particular hockey team, and
an assertion that this team is in fact the Canadiens; or there is an assumption that
Pierre has played hockey such that the Canadiens were involved, and an assertion that
Pierre played for them, not against them. In effect, the clefted constituent, or part
of it, is focalized, in much the same way that the question-word in wh-interrogatives
is. The essential difference here is that the focalized constituent in clefts need not be
interrogative, as indeed it is not in (80). But it can be:
36
(81)
a. C’est [ qui ] qui est venu?
it-is who that is come
‘Who is it that came?’
b. C’est [ pour qui ] que Pierre a joué?
it-is for whom that P. has played
‘Who is it that Pierre played for.’
A notable characteristic of interrogative clefts is that the focalized constituent must
be the question-word: (81b) is not ambiguous in the same way that (80b) is, and
only has the reading under which it is assumed that Pierre played for someone, and
a question is asked about exactly who that person is.
On cursory inspection, it seems as though the clefted constituent is modified by
a relative clause. There are several arguments that this is in fact not the case, and
that clefts and relative clauses must be analyzed differently. What these arguments
amount to is that “qui” and “que” in (80,81) are in fact the complementizers, not
relative pronouns, and that clefting involves extraction across those complementizers.
This matters, because wh-NPs can in fact be clefted:
(82)
C’est quelle personne qui est venue?
If clefting involves extraction, but not anti-extraction, then the latter phenomenon is
more than just a restriction against extraction across quic .
The first piece of evidence that clefts warrant a different analysis than relative
clauses is that it is not generally possible to predicate a relative clause of pronouns
or proper names, except under contrastive focus for the latter:
(83)
a. ?J’ ai donné un livre à lui [ que j’ ai vu ].
I have given a book to him who I have seen
‘I gave a book to him that I saw?’
b. ?J’ ai parlé à lui [ qui fume ], pas à lui [ qui boit ].
I have spoken to him who smokes not to him who drinks
‘I spoke to him who smokes, not him who drinks.’
(84)
a. ?J’ ai donné un livre à Pierre [ que j’ ai vu ].
I have given a book to P. who I have seen
‘I gave a book to Pierre that I saw.’
b. J’ ai parlé à Pierre [ qui fume ], pas Pierre [ qui boit ].
I have spoken to P. who smokes not P. who drinks
‘I spoke to Pierre who smokes, not Pierre who drinks.’
37
The sentence in (84b) is felicitous, as long as there are two men called Pierre in the
context, one of which happen to be a smoker, and the other of which happens to be
a drinker; the relative clauses in this case are used to distinguish the two. All of the
other examples are anomalous. In contract, both of the cleft examples in (80) are
perfectly grammatical, though the focalized constituents are a pronoun and a proper
name, respectively.
A second piece of evidence is that it is not possible to predicate two relative
clauses of the same noun (85). But the focalized constituent in a cleft constituent
can be a noun modified by a relative clause (86):
(85)
*Pierre a parlé à l’ homme [ que j’ ai frappé ] [ que j’ ai vu ].
P. has spoken to the man that I have hit that I have seen
‘Pierre spoke to the man that I hit that I saw.’
(86)
C’est l’ homme [ que j’ ai frappé ] [ que j’ ai vu].
it-is the man that I have hit that I have seen
‘It is the man that I hit that I saw.’
This would be unexplained if the second “que”-clause was a relative clause in both
instances; but if the second “que”-clause in (86) is the complement of a cleft construction, and not a relative clause, then an explanation can be found.
The third argument for treating relatives and clefts as entirely different constructions comes from this set of constrasts in the formation of clefts relative clauses
(88) and clefts (89) from simple declarative sentences (87):
(87)
a. Marie à préparé une embuscade pour l’ homme.
M. has prepared an ambush for the man
‘Marie prepared an ambush for the man.’
b. Le frère de l’ homme est venu.
the brother of the man is come
‘The man’s brother came.’
c. J’ ai donné un livre à l’ homme
I have given a book to the man
‘I gave a book to the man.’
(88)
a. J’ ai vu l’ homme [ pour qui Marie a préparé une embuscade ].
I saw the man for whom M. has prepared an ambush
‘I saw the man for whom Marie prepared an ambush.’
b. J’ ai vu l’ homme [ dont le frère est venu ].
I saw the man of-whom the brother is come
‘I saw the man whose brother came.’
38
c. J’ ai vu l’ homme [ à qui j’ ai donné un livre ].
I have seen the man to whom I have given a book
‘I saw the man to whom I gave a book.’
(89)
a. C’est pour l’ homme [ que Marie a préparé une embuscade ].
it-is for the man that M. has prepared an ambush
‘It is for the man that Marie prepared an ambush.’
b. C’est [ le frère de l’ homme ] [ qui est venu ].
it-is the brother of the man that is come
‘It is the man’s brother who came.’
c. C’est à l’ homme [ que j’ ai donné un livre ].
it-is to the man that I have given a book
‘It is to the man that I gave a book.’
The examples in (88) show that “l’homme” can be relativized from the object position of a preposition, leaving this preposition in the relative clause modifier with
a pronominal object. But (89) shows that “l’homme” must take the preposition of
which it is an object along when it is clefted. The following sentences are grammatical,
but not with a cleft interpretation:
(90)
a. C’est l’homme pour qui Marie a préparé une embuscade. (*cleft /
√
relative)
b. C’est l’homme dont le frère est venu. (*cleft /
√
relative)
√
c. C’est l’homme à qui j’ai donné un livre. (*cleft / relative)
The discourse function of these examples is to individuate a particular man among a
contextually salient set of men, while the discourse function of the examples in (89)
is to indicate that the entity clefted, and no other, participates in the event denoted
by the complement in the relevant way.
It is apparent from (88-90) that it is the relative pronoun, not the relativized
noun, that pied-pipes prepositions. In clefts, it is the clefted noun itself that does the
pied-piping. This suggests that relativized nouns are not extracted, but that relative
pronouns and clefted nouns are.
For these reasons, the analysis of clefts must differ from the analysis of relative clauses. I assume a standard analysis of relative clauses: the relativized NP is
coindexed with a pronoun which occupies the topic function of the relative clause,
which pronoun is itself structure-shared with a grammatical function of that clause.
A clefted constituent, in contrast, will itself be structure shard between some grammatical function in the matrix clause, and the grammatical function it occupies in
the complement.
39
Formally, the analysis of clefts must then be similar to that of wh-interrogatives,
though not identical. A notable differences is that clefts are not always interrogative.
This may seem to motivate a change in the CP rule of (63), which would allow it to
be used in declarative contexts. But if the clefted constituent is generated under CP,
even if this rule is altered to fit non-interrogatives, then a focus inside the comp
f-structure is predicted, whereas the focus of clefts should be an attribute of the
matrix clause. This problem, and the fact that clefts only ever occur with the verb
“être”, suggests that clefts should perhaps not be handled through c-structure rules.
Such an analysis would predict a wider range of cleft verbs that just this one: cstructure rules should not apply to individual words to the exclusion of all others.
Consequently, my analysis of clefts will involve a special, cleft version of the verb
“être”, which will fit smoothly in with the machinery already constructed in this
paper. I propose the following lexical entry for the cleft version of “être”:31
(91)
êtreclef t
V
(↑
(↑
(↑
(↑
pred)=‘be < cf,comp > subj ’
cf)=(↑ comp CleftExtrPath)
focus)=(↑ cf CleftFocusPath)
int)=(↑ focus int)
CleftExtrPath identifies the grammatical function the clefted constituent occupies in the complement, and equates it with the value of cf, a meta-function required
by the verb. CleftFocusPath identifies an f-structure in that grammatical function
to put in the focus of the outermost f-structure. CleftExtrPath, CleftFocusPath and cf are defined in the appendix. Note also that the cleft’s matrix f-structure
takes on the int value of its focus: in this manner, clefts can be interrogative when
the clefted constituent is an interrogative word, and non-interrogative otherwise.32
The analysis can be illustrated with the following c- and f-structure pair:
31
This is really more like a lexical entry associated with the entire paradigm of this verb. I specify
category information for the infinitive, but only the properties specific to clefts are given in the entry
itself.
32
The analysis of clefts I present here is slightly idealized. It allows any grammatical function
inside the clefted constituent to be the focus, which is correct for a non-interrogative clefts, like the
one in (80b), but incorrect for interrogative clefts, like the one in (81b). In these, only the question
word in the clefted constituent can be the focus. I assume that the analysis can be adjusted to
predict this.
40
(92)
C’est Pierre que j’ai vu
‘It is Pierre that I have seen.’
TP

pred

focus



subj


obj






comp


T′
T0
VP
c’est
V′
NP
CP
P ierre
C′
C0
TP
que
T′
T0
VP
j’ai
V′
D
E
‘be obj,comp subj ’
h
i
pred ‘Pierre’
h
i
pred ‘pro’
h
i













D
E 
pred ‘see subj,obj ’ 


h
i


subj

pred
‘I’


h
i



obj
V0
vu
But this formalization of cleft constructions is not required for the problem they represent to emerge. If clefting involves extraction, and the evidence strongly suggests
that it does, then it is possible, as shown in (82) to extract wh-NPs from subject
position when quic is present. Anti-extraction cannot merely be a restriction against
extraction from subject position when quic is present, but must instead be a complex
interaction between the presence of quic , which entails extraction from subject position, and the instantiation of the interrogative focus position in SpecCP. In other
words, wh-NPs in SpecCP are anti-extracted from subject position conditional to the
presence of quic .
4.2
What Makes Wh-NPs Special
Assuming that the conclusion reached at the end of the previous section is correct,
then it is not an inherent property of wh-phrases that makes them extractable from
41
subject position, or not. Rather, wh-phrases must have an inherent property to which
quic is sensitive, such that certain wh-phrases, but not others, are assigned by quic a
feature that makes them incompatible with SpecCP. A glance back at the CP rule
in (63) reveals that extr is a good candidate for the feature assigned by quic : if the
complementizer assigns a negative extr value to just the set of wh-phrases that are
anti-extracted, then it will not be possible for those wh-phrases to be in the focus
attribute defined by SpecCP, since the value of focus extr would then be both
positive and negative, a clear violation of Functional Uniqueness. The question I
address in this section is what exactly the feature of wh-phrases is that causes quic
to assign a negative extr value to their f-structure. First, I return to the suggestion
I made at the very beginning of this paper that D-linking, in the sense of Pesetsky
(1987), could be causing the mischief. As it turns out, D-linking is not a sufficient
explanation, so an exploration of alternatives will be necessary.
D-mystifying D-linking
I mentioned at the very beginning of this paper the distinction that has been drawn
by Pesetsky (1987) between wh-NPs and wh-pronouns. Wh-NPs restrict the range
of felicitous answers to a question to members of a set of contextually salient entities
picked out by the NP. Wh-pronouns impose no such restriction. This can be seen in
the following mini-discourses:
(93)
a. Some men entered the room. Which ones did Mary talk to?
b. Some men entered the room. Who did Mary talk to?
A natural answer to the question in (93a) involves a subset of the men introduced in
the previous sentence, but this is not necessarily so for a natural answer to (93b). In
Pesetsky’s terms, wh-NPs are D-linked, while wh-pronouns are not.
Being D-linked generally correlates with greater freedom in extraction possibilities for the D-linked phrase. This is true of English, Japanese and Polish (Pesetsky,
1987), Italian (Cinque, 1990) and Chamorro (Chung, 1994), at the very least. The
anti-extraction phenomenon of Québécois French is interesting because the opposite
conclusion seems to be true: wh-pronouns can be freely extracted from subject position, but wh-NPs are bound by a number of constraints. But is D-linking really
the correct way of thinking about the distinction between those wh-phrases that are
anti-extracted and those that are not?
One place to start looking for an answer is a proper characterization of the set
of anti-extracted wh-phrases. Up to now, the discussion has centered on wh-NPs as
the entities affected by anti-extraction. However, the taxonomy of entities subject to
it is somewhat broader that I have let on:
42
(94)
a. Quel homme (*qui) voit Paul?
which man that sees Paul
‘Which man sees Paul?’
b. Lequel (*qui) voit Paul?
which-one that sees P.
‘Which one sees Paul?’
c. Combien d’hommes (*qui) sont morts dans les tranchées?
how-many of-men that are died in the trenches
‘How many men died in the trenches?’
d. Combien (*qui) sont morts dans les tranchées?
how-many that are died in the trenches
‘How many died in the trenches?’
Sentences like the one in (94a) I have used throughout to illustrate anti-extraction:
an NP with the wh-determiner “quel”. The sentence in (94b) shows that the pronoun “lequel” also triggers anti-extraction.33 The next example (94c) shows that
anti-extraction also occurs with “combien de”, which could be treated as a complex
wh-determiner, or simply as a pronoun taking a partitive oblique. The difference is inconsequential from the perspective of the analysis I give here: according to the NP rule
in (64), determiners and their nouns have the same f-structure, and whichever functional information contributed to this f-structure by a “combien de” determiner would
also be contributed to it by a “combien” pronoun. Even if it turns out that “combien
de” is a complex determiner, “combien” also has an unambiguously pronominal use,
shown in (94d), which is subject to anti-extraction. Let me call wh-quantifiers the
full set of anti-extractable wh-phrases.
The D-linking approach begins to unravel when wh-quantifiers, quewh and quiwh
are subjected to a mini-discourse test. The test is whether a natural answer to the
questions in these examples necessarily involves members of the appropriate set of
entities mentioned in the context sentence of each example, mainly the set of men
who got on the bus, went to war, or entered the room in (93a,b,c) respectively, and
the set of pianos in (93d).34
(95)
a. Trois hommes sont montés dans l’autobus. Lequel est sorti en premier?
‘Three men got on the bus. Which one got off first?’
33
Other members of these paradigms are also anti-extracted: “quelle”, “quels”, “quelles” and
“laquelle”, “lesquels”, “lesquelles”. However, “auquel”, “auxquels”, “auxquelles”, and “duquel”,
“desquels”, “desquelles”, which incorporate the case-marking prepositions “à” and “de”, cannot be
used as subjects, and cannot be shown to be anti-extracted.
34
I omit here the wh-quantifiers “quel homme” and “combien d’hommes”, because there is always
the possibility that they could pick out some contextually salient set of men other than the one
introduced in the immediately preceding context sentence.
43
b. Mille hommes sont partis à la guerre. Combien sont morts dans les
tranchées?
‘A thousand men went to war. How many died in the trenches?
c. Cinq hommes sont entrés dans la salle. Qui a parlé a Paul?
‘Five men entered the room. Who talked to Paul?’
d. Trois pianos sont tombés du deuxième étage. Qu’est-ce qui a écrasé
Fido?
‘Three pianos fell from the second story. What crushed Fido?’
The results are telling: though “lequel” does have to pick out one of the three men
who got on the bus in (93a), “combien” does not have to pick out a set of men among
the thousand who went to war: any set of entities that died in the trenches could
constitute an answer to the question in (93b).35 “Quec ” and quic behave as expected:
any person could be the one that talked to Paul in (93c), and any heavy object could
have been the thing that crushed poor Fido in (93d). It seems as though “combien”
shares with quewh and quiwh the property of being non-D-linked, even though it does
not share their property of being exempt from anti-extraction.36
But the link between anti-extraction and D-linking is not as tenuous as the
previous examples would have one believe. In English, is is possible to force the Dlinking of wh-pronouns like “who” and “what”. In those sentences, the wh-pronouns
acquire the greater freedom of extraction that inherently D-linked wh-phrases like
wh-NPs enjoy. The following example is from Pesetsky (1987:109):
(96)
I know that we need to install transistor A, transistor B and transistor C,
and I know that these three holes are for transistors, but I’ll be damned if
I can figure out from the instructions wherei what goes ei !
If one assumes that movement is the only scoping mechanism, the “what” in (96)
would outscope “where” at LF, so that the dependencies between the two items and
their traces would cross. This is the kind of example that would be ungrammatical, or
at least anomalous, if the wh-pronouns were non-D-linked,37 and which leads Pesetsky
35
Though these entities would have to be masculine, given the inflection of the verb in the interrogative.
36
According to the characterization of D-linking given by Pesetsky, a wh-NP like “how many
angels” is not D-linked because when it is used in a question, because “there is no presumption
that either speaker or hearer has a particular set of angels in mind” (1987:108). Assuming that
this transfers over to French, and my intuition suggests that it does, then a wh-NP like “combien
d’anges” is non-D-linked, in spite of the difficulty of establishing this through a test like the one in
(95). Yet example (94c) shows that “combien d’hommes” is anti-extracted. This makes the thesis
that D-linking correlates with anti-extraction hard to maintain.
37
That is, “Wherei does what go ei ?” should be anomalous out of context. It doesn’t sound too
bad to me, but that’s the argument.
44
to suggest that there is more to scoping than merely moving; since the example is fine,
the conclusion is that “what” is D-linked, and does not scope through movement.
Though a movement analysis of scoping is not at stake here, the possiblity
of D-linking wh-pronouns should make sentences in which quewh and quiwh are Dlinked constructible. It could then be verified whether D-linked wh-pronouns are
anti-extracted:
(97)
a. Je sais que quelqu’un a vu Paul, mais qui (?est-ce qui) l’a vu?
‘I know that someone saw Paul, but who saw him?’
b. ?Je sais que quelque chose a mordu Paul, mais qu’est-ce qui l’a mordu?
‘I know that something bit Paul, but what bit him?’
Both of these examples are in fact slightly anomalous when quic is present.38 So
D-linking of the extractum does seem to correlate with anti-extraction, but only to a
certain degree: there must be a property of “combien” and “combien de” that makes
them anti-extractable even when they are not D-linked, and there must be a property
of wh-quantifiers in general that makes their anti-extraction much worse than the
anti-extraction of D-linked wh-pronouns. Obviously, it would be preferable if these
two properties were in fact one and the same. Determining which of these perspectives
is the correct one will involve delving into the properties of wh-quantifiers and whpronouns and examining how they differ. This is a task I undertake immediately.
Beyond D-linking
A person asking a question with the D-linked wh-quantifiers, “quel N” and “lequel”,
expects to receive as an answer a nominal whose referent is a single, masculine entity
in a contextually salient set of masculine entities. The feminine and plural versions
of those wh-quantifiers reveal different expectations: the expected answer can be a
single nominal with a referent of a specific gender, picked from a contextually salient
a set of entities of that gender, or a collection of nominals with referents of either
gender from a contextually salient set of entities. The contextual salience of the set
of which the expected referent must be part is what makes these wh-quantifiers Dlinked. The requirements for the expected referent to be of a particular gender and
a particular number are other things entirely, and are reflected in the gender and
number features of the wh-quantifiers. Are either of these requirements shared by the
other wh-quantifiers, but not by the wh-pronouns? The answer seems to be yes. Sort
of.
38
Recall that quic must be present when “est-ce” is, and that “est-ce” must be present when quewh
is used. This explains the distribution of parentheses and anomalousness marks in (97): “est-ce qui”
is optional only as a whole in (a), but not optional at all in (b).
45
The “yes” part is that a person asking a question with “combien de N” and
“combien” is expecting to receive as an answer a number or quantity, such that this
number or quantity of entities are indeed in the relation denoted by the question. In
other words, the answer expected to a question like “Combien d’hommes R?” is the
number of men that are in relation R. There is no restriction imposed on this number:
“a single one” or “a hundred” are equally acceptable answers to such a question. The
answer expected to a question like “combien d’eau R” is a quantity of water that is in
relation R, with no restriction imposed on the magnitude of the quantity. Whether the
request is for a number or quantity largely depends on the grain of the wh-quantifier’s
semantic restriction: a count noun will require an answer in terms of number, and a
mass noun will require an answer in terms of quantity.39 An interesting observation
here is that requests for quantity can also be achieved with D-linked wh-quantifiers.
For instance, “quelle quantité d’eau” contains a request for information about the
magnitude of a quantity of water, or about a number of quantities of water, as the
context dictates. The count/mass distinction seems to be having an effect on the
nature of the information asked.
So all wh-quantifiers, whether D-linked or not, seem to be requests for information that involves number or quantity of entities. The difference is that the D-linked
wh-quantifiers “quel N” and “lequel” are used in requests for information about the
entities of which a particular number is known to be in the relation denoted by the
question: there is a particular number of entities in some relation, and the asker wants
to know what those entities are. On the other hand, the non-D-linked wh-quantifiers
“combien de” and “combien” are used in requests about the number or quantity of
entities known to be in a particular relation: there is a particular number of entities
in some relation, and the asker wants to know what that number is.
Someone asking a question with a wh-pronoun, expects an answer that consists
of a nominal or a set of nominals with referent of a particular degree of humanness, of
which no particular number is expected to be in the relation denoted by the question.
So quiwh and quewh are used in requests for information about entities known to be in
a particular relation: some entities are known to be in some relation, and the asker
wants to know what those entities are.
The difference between wh-quantifiers and wh-pronouns seems to be that the
former, but nor the latter, involve an implicit assumption by the asker about a number
of entities, and an expectation that the answer to a question will contain information
about a number of entities. But if there is a property that can be used to distinguish
those wh-phrases that are subject to anti-extraction from those that aren’t, why did
I express above a reservation about there being such a difference? The answer is that
the property I just identified cannot be syntactic. Syntactically, quewh and quiwh are
39
So French collapses the distinction that English draws between “how many” and “how much”.
46
unambiguously singular, as shown by their agreement properties:
(98)
a. Qui est venu?
b. *Qui sont venus?
(99)
a. Qu’est-ce qui est tombé?
b. *Qu’est-ce qui sont tombés?
Even in context, say a context in which someone just asserted that many people
came or many people fell and the asker didn’t quite catch who or what did so, the (b)
examples are bad. So a syntactic distinction cannot be drawn between the number
of the wh-pronouns and the number a singular wh-quantifier like “lequel”.
However, the assumptions a speaker makes when asking a question (or when
speaking in general) are hardly syntactic. They can be reflected in the syntax, but
there is no guarantee that the syntactic properties of an utterance will be iconic to its
discourse properties. So I propose that the analysis of anti-extraction must involve a
level on non-syntactic representation: a d(iscourse)-structure.
4.3
Anti-Extraction and Discourse-Structure
The framework of LFG provides a formal mechanism for the flow of information
between levels of stucture. As suggested in Kaplan (1987), different structures are
related by correspondence functions, that take a given structure within their domain
as input, and return another structure with their range as output. For instance, the
c-structure is related to the f-structure by the correspondence function φ. Because
φ is a function from the c-structure to the f-structure, it is said that the c-structure
projects the f-structure, and conversely that the f-structure is a projection of the cstructure. This projection architecture is motivated by the idea that different types
of linguistic information may be best represented as different types of formal objects;
correspondence functions are used to relate these objects in a principled way. Various
levels of representation beyond the c- and f-structure have been introduced. For
instance, Butt et al. (1996) assume a morphosyntactic-structure projected by the
c-structure, and current work on semantics in LFG (Dalrymple, 1999) assumes a
semantic-structure projected from the f-structure. The argument-structure (Bresnan
and Kanerva, 1989) is instrumental to the analysis of variation is argument realization,
like dative shift and the passive.
My proposition is thus that wh-quantifiers contribute to a discourse-structure a
presupposition40 about the number of entities about which a question is being asked,
while wh-pronouns do not. These two classes of entities update the d-structure in
40
This may not be the best term for the kind of property I am suggesting here, which is really
47
different ways. My interest here is not in the actual representation of the d-structure,41
but merely in the contributions made to it by wh-NPs and wh-quantifiers. In effect, I
want to define a correspondence function from the f-structure to the d-structure that
can be referenced formally to capture the phenomenon of anti-extraction.42 Let me
define this correspondence function:
(100)
If f is an f-structure, then fδ is the d-structure corresponding to f .
This allows for the following representation for wh-determiners and wh-quantifiers,
and wh-pronouns:43
(101)
quel
Det (↑ int)=+
(↑δ presup)=number
(102)
laquellef in
(103)
combien N (↑ int)=+
(↑δ presup)=number
(104)
quep,f in
Det
Det
(↑
(↑
(↑
(↑
(↑ pred)=‘pro’
(↑ pers)=3
(↑ num)=sg
(↑ gend)=fem
(↑ int)=+
(↑δ presup)=number
pred)=‘pro’
type)=nonhuman
extr)=c +
int)=c +
The attribute-value pair [presup number] in (101,102,103) represents the presupposition they contribute to the d-structure.44
an assumption made by the asker that a certain number of entities are in a particular relation, and
an expectation that the answer to the question asked will contain information about that certain
number of entities, though this number may not be what the question is directly about. But I do
not know of a better one.
41
Though I do think this is a very interesting topic; see Crouch and Van Genabith (1999) for
discussion.
42
To follow the tradition of English-to-Greek alliterative nomenclature for correspondence functions, I will call the the function from f-structure to d-structure δ.
43
The lexical entry for “quel” is reprised from (65), and that for quewh from (55), though only the
former is modified.
44
I make no claim as to whether this is indeed the way the presupposition should be represented.
This is merely a device I use to represent the distinction between the entities affected by antiextraction and the entities that aren’t.
48
But recall that the presence of this presupposition in the d-structure cannot be
what actually causes anti-extraction: wh-quantifiers can be extracted to SpecCP when
quic is not present, and can be clefted. Anti-extraction is actually an incompatibility
between wh-quantifiers and SpecCP that is conditional to the presence of quic . This
means that quic must be sensitive to the property that differentiates wh-quantifiers
from wh-pronouns.
The problem is that this property, the presupposition about number is projected
from the f-structure that will wind up in the focus function, an f-structure to which
quic only has access via functional uncertainty. Imagine that the f-structure in (105)
is the value of subj:
(105)
le frère de quelle personne
‘the brother of which person’

pred
‘brother’



pred ‘of’

"


oblposs 
pred
obj

f:
int



# 

‘person’ 


+
h
fδ : presup
i
number
The presupposition is a feature of fδ , the d-structure that corresponds to the fstructure f of the embedded obj. But if the subject were simply “lequel”, the presupposition would be directly in the f-structure of subj.
So the path from the f-structure of quic , which is the same as the f-structure that
contains the subj, down to the f-structure that projects the crucial d-structure information, is indeterminate. A means whereby this f-structure can be reliably assigned
the anti-extraction feature must be provided. This cannot be accomplished by assigning the anti-extraction feature via functional uncertainty: there is no guarantee that
this second uncertainty will resolve to the same path as the first one, and therefore
no guarantee that the feature will be assigned to the correct f-structure. The solution to this problem is the device of local name, introduced by Kaplan and Maxwell
(1996): the f-structure picked out by a functional uncertainty can be assigned a label
(prefixed with a % sign so that the label’s nature is transparent), which can then be
used to place further constraints on that f-structure. The lexical entry of quic can be
revised to:
(106)
quic :
C ((gf∗ ↑) focus) = (↑ subj gf∗ ) = %qword
(%qwordδ presup)=number 7→ (%qword extr)=−
Recall that the functional uncertainties in this lexical entry state that an equality
relationship must hold between the focus attribute of an f-structure arbitrarily far
49
above that of the complementizer, and an f-structure embedded arbitrarily far down
into the subj attribute of the complementizer. Here, both of the f-structures of
which this equality is constrained to hold are assigned the arbitrary label %qword,
preceded by a percent sign to mark its status as a local name. The second part of
this lexical entry is an implication, stating that if the d-structure corresponding to
%qword has a presup attribute with the value “number”, then the f-structure of
%qword has a negative value for the feature extr.
When wh-phrases containing words like (101-103) are embedded somewhere in
the subj function of an f-structure of which the constraints introduced by quic must
hold, they will be assigned a negative value for the feature extr. These wh-phrases
are then compatible with the focus defined by the cleft verb “être” (91), which
has no requirements for the extr value of its focus, but are incompatible with the
focus defined by the CP rule (63), which is assigned a positive extr value. In other
words, wh-quantifiers are incompatible with extraction to SpecCP, conditional to
being extracted from subject position, but only when quic is present. This is exactly
the prediction that the data I have presented throughout impose on an analysis of
the anti-extraction phenomenon.
5
Conclusion
This paper started out with the idea that a precise analysis of wh-interrogatives
in Québécois French could help shed some light on the nature of anti-extraction, a
phenomenon whereby some question-words cannot be extracted from subject position.
First, I provided an analysis of wh-interrogatives that differs significantly from
the one in Kaplan and Bresnan (1982), and is arguably truer to the data, though
an investigation of other relevant phenomena such as binding and crossover might
disprove this claim. But this analysis does make for a cleaner distinction between the
surface structure of interrogatives, and their deeper grammatical relations: the constituent dislocated in the c-structure is not necessarily congruent to the constituent
dislocated in the f-structure. In reaching this analysis, I uncovered several interesting facts about “est-ce que”. For instance, contrary to what has been claimed by
some (e.g. Kayne, 1972), this orthographic sequence in fact consists of two discrete
syntactic units. Furthermore, “est-ce” is a complement-taking verb, though it does
not inflect, and does not take an overt subject; rather, it is inherently interrogative,
and defines its own expletive subject, via a grammaticalized subject suffix. The verb
retains properties of this former suffix, like a restriction against being embedded.
Second, I showed that anti-extraction is due partly to the lexical properties
of quic , which constrains extraction of the subject of its clause. This constraint is
a lexical property of the complementizer, and also controls the distribution of quic
50
relative to quec : the latter can be used only when its clause’s subject is not extracted.
The presence of quic is crucial to anti-extraction, but so is the specifier position
of CP: cleft constructions, which may feature extraction from subject position and
obligatorily have a complementizer, do not give rise to the phenomenon.
I showed that the wh-phrases that undergo anti-extraction do not do so as a
result of being D-linked, in the sense of Pesetsky (1987): non-D-linked wh-phrases
are also anti-extracted. Instead, an assumption made by the asker of a question about
the number of entities denoted by the wh-phrase, and the consequent expectation that
and answer to the question will provide information about said number, is at fault.
“Quic ” is sensitive to this property when it occurs in the d-structure corresponding to
a wh-phrase embedded in its own f-structure, and assigns a feature to this wh-phrase
that makes it incompatible with the focus function defined by SpecCP.
Together, the analyses of wh-interrogatives and anti-extraction I give here make
the prediction that anti-extraction shoul occur regardless of the degree of embedding
the the offending wh-phrase in the extractum. These predictions are in fact borne
out:
(107)
a. *Le cousin de quelle personne est-ce qui voit Paul?
the cousin of which person is-it that sees Paul
‘Which person’s cousin sees Paul?’
b. *La voiture que Marie a empruntée à quelle personne est-ce qui a
été volée?
the car that Marie has borrowed to which person is-it that has been
stolen
‘The car Marie borrowed from which person has been stolen?’
The wh-NPs in these examples are embedded inside the extractum, but the constraints defined in this paper predict that they will be assigned a negative extr
feature, because the extraction site is in subject position and “quic ” is present. Since
the extractum is in SpecCP, this negative extr value conflicts with the positive one
assigned by that node. The analysis predicts that these examples should be ungrammatical, and they are.
Nevertheless, further research must be undertaken to determine whether the
analysis presented herein extends to a broader range of phenomena in this particular
dialect, other dialects of French, or even cross-linguistically.
Appendix
Here I provide definitions and justifications for the various metacategories, metafunctions and meta-attributes I introduced throughout the paper.
51
Defining ExtrP
The metacategory ExtrP is used in the CP rule in (63) as a cover term for the range
of categories extracta can have. The following example illustrate:
(108)
a. NP: Qui vient?
who comes
‘Who is coming?’
b. PP: Pour qui Sakic joue?
for who S. plays
‘Who does Sakic play for?’
c. AdvP: Quand Marie revient?
when M. comes-back
‘When does Marie come back?’
d. AP: Comment grand Denis est?
how tall D. is?
‘How tall is Denis?’
Accordingly, the following disjunctive definition can be given for ExtrP:
(109)
ExtrP ≡ { NP | PP | AdvP | AP }
ExtrP is thus equivalent to the disjunction of the phrasal categories in within the
curly brackets.
Defining ExtrPath
ExtrPath is the functional uncertainty that retrieves a grammatical function and
assigns it to the constituent in specifier position of CP. In the following examples, the
path down to the function of the extractum is noted before each sentence:
(110)
a. subj: Qui
voit Pierre?
b. obj: Qui Pierre voit
?
c. oblθ : À qui Pierre a donné un livre
d. adjunct: Quand Pierre a vu Marie
e. comp subj: Qui Pierre dit qui
?
?
a vu Marie?
f. comp obj: Qui Pierre dit que Marie a vu
?
g. comp oblθ : À qui Pierre dit que Marie a donné un livre
52
?
h. comp adj: Quand Pierre dit qu’Étienne a vu Marie
i. (xcomp) subj: Qui
espère
gagner le gros-lot?
j. xcomp obj: Qui Pierre espère voir
?
k. xcomp oblθ : À qui Pierre espère donner un livre
l. xcomp adj: Quand Pierre espère gagner le gros-lot
m. subj oblθ : *De qui [ le frère
n. subj adj: *Où [ l’homme
?
?
?
] est venu?
] vend des aspirateurs?
o. obj oblθ : *De qui Pierre a vu [ le frère
p. obj adj: *Où Pierre a vu [ l’homme
]?
] se peigner?
q. oblθ adj: *Où Pierre a mis le livre [ dans la boı̂te
]?
r. adj adj: *Où Pierre a mis le livre [dans la boı̂te] [sur la table
]?
All of the grammaticality judgements are relative to the bracketings shown; some
sentences, especially those with adjuncts, are grammatical when a different structure
is evaluated. Note that all of these examples can be dummy embedded under the
comp of “est-ce” and maintain the grammaticality indicated.
The conclusion these examples seem to force is that the extractum can be a
subj, obj, oblθ or adj embedded under any number of xcomps or comps, but not
under even a single subj, obj, oblθ or adj. This gives us the following definition
for ExtrPath:
(111)
ExtrPath ≡ { xcomp | comp }∗ { subj | obj | oblθ | adj }
The superscript star in (111) is the Kleene operator, indicating the xcomp/comp
disjunction can be repeated any number of times, including zero. The star is outside
the brackets, so the path denoted by the first disjoint set, should it be of length
greater than zero, can be made up of only xcomps, only comps, or any combination
of the two. The ExtrPath must terminate at exactly one of the gfs in the second
disjoint set.
Defining FocusPath
The following examples will help determine the nature of FocusPath, the path
down through the f-structure of the extracted constituent in SpecCP that leads to
the function of the question-word. The grammatical functions of the path precedes
each sentence.
the “e” in front of (112a) denotes the empty string:45
45
Examples with a FocusPath starting at subj are impossible to construct: they would require
that a clause be extracted, which ExtrPath shows to not be possible.
53
(112)
a. e: Qui est-ce qui est venu?
b. obj: Pour qui est-ce que Pierre a pleuré?
c. adj: *Le livre où est-ce que Pierre a mis dans la boı̂te?
d. obj adj: *À l’homme où est-ce que Marie a donné un livre?
e. obj oblθ obj: Au cousin de qui est-ce que Marie a donné un livre?
f. oblθ obj: Le frére de qui est-ce qui est venu?
g. oblθ adj: *Le cousin de l’homme où est-ce qui est venu?
h. adj obj: Le livre sur quoi est-ce que Pierre a mis dand la boı̂te?
i. adj subj: L’homme que qui a vu est-ce qui est venu?
j. adj oblθ : L’homme qui a mis le livre où est-ce qui est venu?
k. adj adj: L’homme qui est tombé où est-ce qui est venu?
l. adj comp subj: L’homme qui a dit que qui parlait est-ce qui est
venu?
m. adj comp obj: L’homme qui a dit que Pierre voit qui est-ce qui est
venu?
n. adj comp oblθ : L’homme qui a dit que Pierre a mis un livre où est-ce
qui est venu?
o. adj comp adj: L’homme qui a dit que Pierre est tombé où est-ce qui
est venu?
p. adj (xcomp) subj: L’homme que qui espère
venu?
voir est-ce qui est
q. adj xcomp obj: L’homme qui espère voir quoi est-ce qui est venu?
r. adj xcomp oblθ : L’homme qui espére mettre un livre où est-ce qui
est venu?
s. adj xcomp adj: L’homme qui espère partir quand est-ce qui est venu?
These examples show several things. First, FocusPath consisting of only adj are
ungrammatical, but paths of length greater than one can terminate at adj. Second,
after it has gone through adj, a FocusPath can terminate at any gf, though paths
that terminate comp or xcomp cannot be constructed because these functions cannot
be wh-pronominalized. Third, non-null paths that terminate at oblθ cannot be
constructed unless they first go through adj; otherwise oblθ must be followed by
obj. The explanation for this is that the oblθ function is restricted to PPs, whose
preposition head requires an obj. This leaves the following definition for FocusPath:
54
(113)
FocusPath ≡ { (oblθ ) obj | adj gf+ }∗
The superscript “+” is the Kleene plus operator, indicating here that gf can be
repeated any number of times, but no less than one. Note that the star operator has
scope over the entire disjunction, making it possible for FocusPath to resolve to the
empty string. In such a case, as the CP rule in (63) makes clear, the ↓ metavariable
is equated to both (↑ focus) and (↑ ExtrPath): this is the case of wh-pronoun or
wh-NP extracta, which are congruent with the focussed element.
This definition is formulated in such a way as to allow the FocusPath to consist
at least partly of consecutive obj functions, a possbility not shown in (112). This is
not a problem: either this kind of embedding cannot be accomplished in the language,
in which case there can be no attribute paths with consecutive objs, or it can, in
which case the possibility is accounted for. The example I am thinking of is that of
complex prepositions: an object-taking preposition takes an object-taking preposition
as it object. As far as I know, apparent complex preposition are attested in French
(i.e. ”près de”, ”quant à”), but whether they are indeed best analyzed as complex
prepositions as opposed to lexicalized sequences is not certain. I remain neutral about
this particular issue, but wish to point out that question-words can indeed occur as
the object of these prepositions. Whether they are complex or not, this possibility is
accounted for by the analysis.
Defining CleftExtrPath
CleftExtrPath is the functional uncertainty that leads down from the comp function of the cleft verb “être” and retrieves a grammatical function to equate with that
verb’s cf (defined below):
(114)
a. subj: C’est Pierre qui
est venu.
b. obj: C’est Pierre que Marie voit
.
c. oblθ : C’est à Pierre que Marie à donné un livre
d. adj: C’est sur la tête que Pierre est tombé
.
.
e. oblθ obj: *C’est Pierre que Marie a donné un livre à
.
f. adj subj: *C’est Marie que Luc frappait Pierre pour que/qui
satisfaite.
g. adj obj: *C’est Pierre que Marie courait pour satisfaire
soit
.
h. adj oblθ : *C’est Pierre que Marie courait pour faire un cadeau à
.
i. adj adj: C’est dans l’entrée que Pierre est tombé sur [ la voiture
].
j. comp subj: C’est Pierre que Marie dit qui
55
est venu.
k. comp obj: C’est Pierre que Marie dit que Paul a vu
.
l. comp oblθ : C’est où que Pierre dit que Marie a mis le livre
m. comp adj: C’est quand que Pierre dit que Marie est partie
n. xcomp obj: C’est Pierre que Marie espère voir
.
?
.
o. xcomp oblθ : C’est où que Marie espère mettre le livre
?
p. xcomp adj: C’est sur la tête que Marie espère voir Pierre tomber
.
These examples reveal that be subjs, objs, oblθ s or adjs, which can be embedded
under comp or xcomp, but not other grammatical functions. This is exactly the
range of possibilities observed in (110) for the grammatical functions that extracta in
wh-interrogatives can occupy. Thus:
(115)
CleftExtrPath ≡ ExtrPath
Defining CleftFocusPath
CleftFocusPath is he functional uncertainty that reaches into the cf function
of the cleft verb and identifies a grammatical function to equate with that verb’s
focus. Here, interrogative clefts are used to illustrate the possiblities, to avoid the
degree of ambiguity that would be caused by using non-interrogative clefts; recall from
Section 4.1 that any non-interrogative cleft whose clefted constituent is pied piped is
ambiguous between a reading under which the whole clefted constituent is focusses,
and a reading under which only part of that constituent is. Should the following
examples be replaces by non-interrogatives, and the focus kept on the underlined
constituent, the same grammaticality judgements as those given here for interrogative
foci would obtain.
(116)
a. e: C’est qui qui est venu?
b. obj: C’est pour qui que Pierre a pleuré?
c. adj: *C’est le livre oùoù que Pierre a mis dans la boı̂te?
d. obj adj: C’est *à l’homme où que Marie a donné un livre?
e. obj oblθ obj: C’est au cousin de qui que Marie a donné un livre?
f. oblθ obj: C’est le frére de qui qui est venu?
g. oblθ adj: *C’est le cousin de l’homme où qui est venu?
h. adj obj: C’est le livre sur quoi que Pierre a mis dand la boı̂te?
i. adj subj: C’est l’homme que qui a vu qui est venu?
56
j. adj oblθ : C’est l’homme qui a mis le livre où qui est venu?
k. adj adj: C’est l’homme qui est tombé où qui est venu?
l. adj comp subj: C’est l’homme qui a dit que qui parlait qui est venu?
m. adj comp obj: C’est l’homme qui a dit que Pierre voit qui qui est
venu?
n. adj comp oblθ : C’est l’homme qui a dit que Pierre a mis un livre où
qui est venu?
o. adj comp adj: C’est l’homme qui a dit que Pierre est tombé où qui
est venu?
p. adj (xcomp) subj: C’est l’homme que qui espère
voir qui est venu?
q. adj xcomp obj: C’est l’homme qui espère voir quoi qui est venu?
r. adj xcomp oblθ : C’est l’homme qui espére mettre un livre où qui est
venu?
s. adj xcomp adj: C’est l’homme qui espère partir quand qui est venu?
The interrogative clefts given here correspond exactly to the wh-interrogatives given
in (112) to define FocusPath. The grammaticality patterms observed there are
exactly replicated. Thu we have:
(117)
CleftFocusPath ≡ FocusPath
Defining cf
The metafunction cf is used in the lexcial entry for the cleft verb “être” in (91),
as a cover term for the geammatical functions that can be assumed by the clefted
constituent. These are the phrasal categories that project the f-structure at which
CleftExtrPath can terminate:
(118)
cf ≡ { obj | oblθ | adj }
The disjunctive definition is required (as opposed to a flat-out imposition of a
single specific function) because the clefted constituent will be either an NP, a PP,
and AP or an AdvP, and I assume that the VP rule will only generate NPs as objs
and PPs as oblθ , and AdvPs and APs as adjs.
57
References
Anderson, S. R. (1982). The analysis of French shwa: or, how to get something for
nothing. Language, 58(3):534–573.
Auger, J. (1993). More evidence for verbal agreement-marking in colloquial French.
In Ashby, W., Mithun, M., Perissinotto, G., and Reposo, E., editors, Linguistic
prespectives on the romance languages. Amsterdam: John Benjamins.
Auger, J. (1995). Les clitiques pronominaux en français parlé informel.
québécoise de linguistique, 24:21–60.
Revue
Baker, C. (1970). Notes on the description of English questions: the role of an abstract
question morpheme. Foundations of Language, 6:197–219.
Barnes, B. (1985). The pragmatics of left detachment in spoken standard French.
Amsterdam: John Benjamins.
Bouchard, D. (1988). French voici/voilà and the analysis of pro-drop. Language,
64(1):89–100.
Bresnan, J. and Mchombo, S. A. (1987). Topic, pronoun, and agreement in Chicheŵa.
Language, 63(4):181–254.
Bresnan, J. W., editor (1982). The mental representation of grammatical relations.
Cambridge, MA: MIT Press.
Bresnan, J. W. and Kanerva, J. (1989). Locative inversion in Chicheŵa: a case study
of factorization in grammar. Linguistic Inquiry, 20. 1-50.
Butt, M. and King, T. H. (2000). Null elements in discourse structure. In Subbarao,
K., editor, Papers from the NULLS Seminar. Motilal Banarasidas, Delhi.
Butt, M., Niño, M.-E., and Segond, F. (1996). Multilinguial processing of auxiliaries
in LFG. In Gibbon, D., editor, Natural language processing and speech technology: results of the 3rd KONVENS conference, pages 111–122. Mouton De Gruyter:
Universität Bielefeld.
Casagrande, J. and Saciuk, B. e. (1972). Generative studies in Romance languages.
Rowley, MA: Newbury House.
Chung, S. (1994). Wh-agreement and “referentiality” in Chamorro. Linguistic Inquiry,
25(1):1–44.
58
Cinque, G. (1990). Types of Ā-dependencies. Cambridge, MA: MIT Press.
Crouch, R. and Van Genabith, J. (1999). Context change, underspecification and the
structure of glue language derivations. In Dalrymple (1999), pages 117–189.
Dalrymple, M. (1999). Semantics and syntax in lexical functional grammar. Cambridge, MA: MIT Press.
Dalrymple, M., Kaplan, R. M., Maxwell, III, J. T., and Zaenen, A., editors (1995).
Formal issues in lexical-functional grammar. Stanford, CA: CSLI Publications.
Falk, Y.
Pivots and the theory of grammatical relations.
Proceedings of the LFG200 conference.
Available
online
http://cslipublications.stanford.edu/LFG/5/lfg00-toc.html.
In
at
Jones, M. A. (1996). Foundations of French syntax. Cambridge University Press.
Kaplan, R. M. (1987). Three seductions of computational psycholinguistics. In Whitelock, P., Wood, M. M., Somers, H. L. andJohnson, R., and Bennett, P., editors,
Linguistic theory and computer applications, pages 149–188. New York: Academic
Press. Also published in Dalrymple et al. (1995).
Kaplan, R. M. and Bresnan, J. (1982). Lexical-functional grammar: a formal system
for grammatical representation. In Bresnan (1982), pages 173–281.
Kaplan, R. M. and Maxwell, III, J. T. (1996). LFG grammar writer’s workbench. Technical report, Xerox Palo Alto Research Center, available online at
ftp://ftp.parc.xerox.com/pub/lfg/lfgmanual.ps.
Kaplan, R. M. and Zaenen, A. (1989). Long-distance dependencies, constituent structure, and functional uncertainty. In Baltin, M. and Kroch, A., editors, Alternative
conceptions of phrase structure. University of Chicago Press.
Kayne, R. S. (1972). Subject inversion in French interrogatives. In Casagrande and
Saciuk (1972), pages 70–126.
King, T. H. (1995). Configuring topic and focus in Russian. Stanford, CA: CSLI
Publications.
Kroeger, P. (1993). Phrase structure and grammatical relations in Tagalog. Stanford,
CA: CSLI Publications.
59
Labelle, M. and Hirschbuhler, P. (2000). Les “clitiques” arguments en serbo-croate
et dans l’histoire du français. In Muller, C., de Carralho, P., Labrune, L., Lambert,
F., Ngalasso, M. M., and Ploog, K., editors, Clitiques et cliticisation: actes du
colloque de Bordeaux. Paris: Honoré Champion.
Langacker, R. W. (1972). French interrogatives revisited. In Casagrande and Saciuk
(1972), pages 36–69.
Miller, P. (1991). Clitics and constituents in phrase structure grammar. PhD thesis,
Rijksuniversiteit te Utrecht.
Morin, Y. C. (1985). On the two French subjectless verbs voici and voilà. Language,
61(4):777–820.
Pesetsky, D. (1987). Wh-in-situ: movement and unselective binding. In Reuland,
E. J. and ter Meulen, A. G., editors, The representation of (in)definiteness, pages
98–129. Cambridge, MA: MIT Press.
Picard, M. (1991). Clitics, affixes, and the evolution of the question marker tu in
Canadian French. French Language Studies, 1:179–187.
Picard, M. (1992). Aspects synchroniques et diachroniques du tu interrogatif en
québécois. Revue québécoise de linguistique, 21:65–75.
Pollock, J.-Y. (1989). Verb movement, Universal Grammar, and the structure of IP.
Linguistic Inquiry, 20:365–424.
Rizzi, L. (1986). On the status of subject clitics in Romance. In Jaeggli, O. and SilvaCorvalán, C., editors, Studies in Romance linguistics, pages 391–419. Dordrecht:
Foris.
Rizzi, L. and Roberts, I. (1989). Complex inversion in French. Probus, 1:1–30.
Ross, J. R. (1967). Constraints on variables in syntax. PhD thesis, MIT.
Schane, S. (1968). French phonology and morphology. Cambridge, MA: MIT Press.
Selkirk, E. (1972). The phrasal phonology of English and French. Bloomington:
Indiana University Linguistics Club.
Zwicky, A. M. and Pullum, G. K. (1983). Cliticization vs. inflection: English n’t.
Language, 59:502–513.
60
Download