Carlos de Cuba

advertisement
Carlos de Cuba
SUNY Stony Brook
Draft 5/8/02
A Phase-derivational Analysis of Limited Embedded V2 in Mainland Scandinavian
1. Introduction
It has been widely noted in the literature that many languages show an asymmetry in word order
in root vs. embedded contexts. Certain constructions that occur in main clauses are not possible in
embedded clauses. One such asymmetry is the Verb second (V2) effect in many Germanic
languages. Den Besten (1983) provides the classic analysis of the V2 effect: the finite verb moves
to C and a topic fronts to Spec-CP. V2 is then blocked in subordinate contexts because of the
presence of a complementizer in the embedded C. Den Besten’s analysis accounts for much of the
Germanic data, but it does not explain the pattern of limited embedded V2 in Mainland
Scandinavian (MSc) (examples in Swedish)1.
(1) a. Rickard ångrade att han inte var hemma
R.
regretted that he not was home
b. *Rickard ångrade att han var inte hemma
R
regretted that he was not home
‘Rickard regretted that he was not home.’
(2) a. Rickard sa att han inte var hemma
R. said that he not was home
b. Rickard sa att han var inte hemma
R. said that he was not home
‘Rickard said that he was not home.’
The (a) examples above show the standard word order for Swedish embedded complement
clauses, where the verb var ‘was’ is below sentential negation, indicating that V2 movement of V
to C has not taken place. The complementizer att ‘that’ is in the head of C: [CP [C att] [IP han inte
var hemma]]2. (1b) shows that embedded V2 (EV2) is not generally permitted, while (2b) is an
example of limited embedded V2, the restriction being that it is only possible under matrix
1
For more examples and discussion see Heycock (2000), Holmberg & Platzack (1995), Vikner (1995), Iatradou &
Kroch (1992), Holmberg (1983), Andersson (1975).
2
‘bridge verbs’. The grammaticality of (2b) presents a problem for Den Besten’s analysis, because
it predicts that the V2 effect should only be possible in the absence of a complementizer in C. In
fact, an EV2 sentence without the complementizer is ungrammatical in Swedish.
The focus of this paper is to account for the cluster of facts surrounding EV2 in MSc,
using a derivational style analysis in the spirit of Chomsky (2000, 2001). I propose that there is
extra syntactic structure selected in the sentential complements of bridge verbs, which allows for
the extra movement in (2b). This extra structure is not available under non-bridge verbs, so (1b) is
ruled out in the standard manner. I will also attempt to sharpen the notion of what a bridge verb
is, following in the footsteps of Kiparsky & Kiparsky (1970), Hooper & Thompson (1973),
Cattell (1978), and Hegarty (1992), and use this notion to motivate the extra syntactic structure.
My hope is to improve upon the CP-Recursion analysis, which has been widely argued for
(Vikner 1995, Holmberg & Platzack 1995, Watanabe 1992, Iatradou & Kroch 1992, Heycock
2000). In section 2 I present my proposal for an extra syntactic projection between V and CP. In
section 3 I implement the proposal to account for limited EV2 and other word order facts in
Swedish. Section 4 presents more evidence for the proposed projection from Hungarian and
Japanese. In section 5 the structure is used to explain factive island effects. Section 6 presents a
classification of bridge and non-bridge verbs. Section 7 is a summary.
2. The Proposal
To account for the syntactic difference between bridge verb and non-bridge verb sentential
complement clause constructions, I propose that there is extra structure between V and CP in
bridge constructions. This extra projection is a CP with special selectional properties that
differentiate it from a standard CP. I will henceforth refer to this projection as CP VS, with the VS
standing for ‘Volunteer-stance’ (VS). I take this terminology from work on why-extraction by
Cattell (1978), who divided verbs that take sentential complements into three classes, Volunteerstance, Response-stance and Non-stance. Only VS verbs allow why-extraction from their
2
For arguments against V to I movement without V to I to C in Swedish, see Holmberg & Platzack (1995).
3
complements. In this paper I define bridge verbs as being VS verbs. I take the VS class to be a
subset of what are traditionally referred to as non-factives, and I take non-bridge verbs to be
factives. Following Hegarty (1992), I make the link between Cattell’s VS verbs and Vikner’s
(1995) list of Danish verbs that allow EV2 complements, and I call these bridge verbs. For a more
detailed discussion, see section 6.
A bridge verb selects for CPVS as opposed to a standard CP. CVS then selects for CP. Nonbridge verbs do not select for CPVS, and therefore can never license this recursive CP
construction. CP-recursion is strictly limited by selectional properties, as C can not select for CP
or CPVS, and CVS can not select for CPVS. Thus CP-recursion is limited to bridge (VS) verb
complement clauses. The structures are given below, (3) for non-bridge verbs and (4) for bridge
verbs.
(3)
VP
ru V’
ru CP
Non-bridge V
ruTP
5
(4)
VP
ru V’
ru CPVS
Bridge V
ru CP
ruTP
5
M.Sc. EV2 facts provide syntactic evidence that an extra position exists in the CP field. In (2b)
we saw the presence of the overt complementizer att ‘that’ in conjunction with EV2. The classic
analysis of V2 is of verb and topic movement to CP, and complementizers presumably occur in
C, so it is natural to assume a recursive CP structure for (2b). In (3), the complementizer att
(‘that’ in Swedish) is in the head of CP, blocking V2 movement. In (4), att is in the head of CPVS,
and the standard V2 movement takes place in CP. Various versions of the CP-recursion analysis
4
of EV2 have been proposed, including Vikner (1995), Holmberg & Platzack (1995), Iatridou &
Kroch (1992), Watanabe (1992) and Hegarty (1992). The present account differs in that it is the
selectional properties of the verb that license CP-recursion, not the semantic content of the lower
CP. The semantic interpretation of the CP is the same in (3) and (4), but the truth-value of a CP is
put into question when embedded under a CPVS. Looking at data from English applied to (3) and
(4), the lower CPs in the sentences in (5) are interpreted semantically in the same way, but the
upper CPVS effects the truth conditions of the bridge verb complement3.
(5) a. John forgot [CP that he went to the store] (he went to the store = true)
b. John thinks [CPVS [CP that he went to the store] (he went to the store = true or not)
The presence of CPVS changes the interpretation of the lower CP from factive to non-factive. In
(5a) the truth of the entire sentence depends on the truth of the lower clause. In (5b) the sentence
can be true whether or not the lower clause is true. I take this semantic difference to be caused by
a Volunteer-stance operator [VSOP] that heads CPVS. The meaning that this operator provides is
that the information (sentence) embedded under CPVS is Volunteer-stance. All of the semantic
theories that I am familiar with share the need to add extra structure in order to get the correct
truth conditions for sentences with bridge verb sentential complements4. My proposal fits this
need easily by providing an extra syntactic position that houses an operator that provides the
semantic notion of non-factivity of a complement clause.
The operator contained in CPVS can affect movement possibilities. The presence of matrix
negative verbs, matrix irrealis verbs and matrix negation with bridge verbs all block EV2 in
M.Sc.. In my proposal, what stops movement in these cases is the negative and/or irrealis
licensing of CVS, making [VSOP] negative [NVSOP] or irrealis [IVSOP]. The negative or irrealis
3
EV2 is not available in restricted or non-restricted relative clauses in Swedish, in spite of the fact that it has been
argued that RRCs are presupposed while NRRCs aren’t. However, the present analysis gives the correct prediction as I
argue that EV2 is a result of verb selection, I have nothing further to say about the semantic differences between RRCs
and NRRCs.
4 Lambda operators, quotational theories, world theories etc. For a summary of some of these theories see Larson &
Segal (1995: ch. 11).
5
status of the operator blocks a merge that would have allowed for EV2. This has similarities to
earlier proposals (negative complementizers: Laka (1990); operators: Watanabe (1992), Hegarty
(1992); transparent vs. contentful CPs: Iatradou & Kroch (1992)), but differs from all of them in
significant respects. This is discussed further below.
3. Limited Embedded V2 in Swedish
In this section I apply the proposal in section 2 to data from Swedish. The distribution V2 in
Swedish is roughly as follows:
A.
B.
C.
D.
V2 is obligatory in matrix clauses.
EV2 is prohibited under non-bridge verbs.
EV2 is optional under bridge verbs.
EV2 is prohibited under negated bridge verbs, negative verbs, and irrealis verbs.
I also examine the optionality of the complementizer att and propose a syntactic explanation for
its distribution. Complementizers are only optional under bridge verbs. In addition, I will offer a
derivation for a construction in Swedish that has to the best of my knowledge escaped previous
analysis.
3.1. The Standard Cases: Matrix V2 and Non-Bridge Verbs (No EV2 allowed)
In (6) we see cases in Swedish where no CPVS is generated5. (6a) is a standard matrix V2
construction with the finite verb var moving from V to T to C, and the subject han ‘he’ moving
locally from SpecVP to SpecTP to SpecCP. The movement of the verb is needed to check the
Finite feature [+Fin] in C. Han moves to SpecTP to check Nominative case features [+NOM],
then up to SpecCP to check the [+EPP] feature6. The phrase that moves to SpecCP does not need
to be the subject, as other XPs (objects, adverbials, negation, non-finite verbs) can also move and
check [+EPP]. However, subject SpecVP to SpecTP movement is always obligatory for case.
(6) a. [CP Hani [C vark [TP ti tk [NegPinte] [VP ti tk hemma]]]]
He
was
not
home
5
This analysis is based mostly on Holmberg & Platzack (1995), but the [+EPP] feature in CP driving the movement of
an XP to SpecCP is my addition.
6 For me, the [+EPP] is checked by a lexical item that bears syntactic features and is able to participate in a checking
relationship. Note that while the verb checks [+Fin] in CP, it does not check [+EPP]. The fact that verbs don’t check
[+EPP] is evidenced by obligatory subjects in TP in Swedish.
6
b. Rickard ångrade [CP [Catt] [TPhani [NegPinte] [VP ti var hemma]]]
R
regretted
that
he
not
was home
The same clause is shown in (6b) embedded under a non-bridge verb. Han moves up to SpecTP
for case, but the verb var remains in VP. This is evidenced by the position of sentential negation
inte above the verb7. The [+Fin] feature in C is checked by the overt complementizer att.
Crucially, I assume that the complementizer has the ability to check both [+Fin] and [+EPP] 8.
Therefore, there is no motivation for any XP to move to SpecCP for checking purposes. This
explains the ungrammaticality topicalization over the complementizer, as illustrated in (7).
(7) *Rickard ångrade [CP hani [Catt] [TP ti [NegPinte] [VP ti var hemma]]]
R.
regretted he that
not
was home
Semantically, the truth-values are straightforward. In order for (6a) to be true, ‘he was not home’
has to be true. For (6b) to be true: 1. Rickard has to regret that he was not home 2. He must not
have been home. It is impossible to regret an event that did not occur (as opposed of course to
regretting that an event did not occur, which is entirely plausible). Since there is no [VSOP]
intervening between the matrix clause and the CP in (6b), the semantic module will have no
trigger to cast the truth of the CP in doubt. In other words, there is nothing in the syntax to tell the
semantics that the complement should not be interpreted as true. Hence, the CP gets a factive
reading. This can be taken as the default interpretation of CPs.
3.2. Bridge Verbs With Embedded V2
7
The placement of sentential negation and adverbs is constant in Swedish, directly above VP. I will not address the
question of negation being above or adjoined to VP as it is not the focus of this paper.
7
I now turn to cases in which we find EV2 in Swedish. In (8) there is a CP embedded under the
M.Sc. bridge verb tror ‘think’. EV2 is evidenced by the position of the finite verb läste ‘read’
above the negation inte ‘not’ in the lower clause. The matrix clause above VP has been omitted9.
(8) Dan tror att Rickard läste inte boken i dag.
D. thinks that R. read not book-the today
VP
ru V’
ruCPVS
V
ttror
CVS
tDan
ru CVS’
ruCP
ru C’
att[+EPP]
DP
[VSOP] 5 C
Rickardt lästet tRickard
[+EPP]
[+Fin]
ru TP
ru T’
ru NegP
T
ru vP
tläste
ru v’
inte
ru VP
tRickard
ru V’
v
tläste
DP
ruAP
5
V
5
boken
tläste
i dag
The crucial differences between (8) and (6b) are the presence of CPVS, and the fact that in (6b),
att merges in C, while in (8), att merges in CVS. I assume that att and [VSOP] are compatible in
the same head, and that the presence of [VSOP] is not enough to check [+EPP]. Because att
merges later in (8), V2 occurs in the lower CP, with the verb läste moving to the head of C to
check [+Fin] and Rickard moving to SpecCP checking [+EPP] 10. At first glance, one may notice
8
In this paper I argue that [+EPP] in CP and CP VS can be checked by either X (att) or XP (a phrase moved to the
specifier). However, it is crucial for me that verb movement to C can not check the EPP. For discussion of different
methods of EPP checking see Alexiadou & Anagnostopoulou (1998).
9 And Dan actually merges in SpecvP. This convention is used for the remainder of the paper.
10 In (8), i dag or boken could move to SpecCP and check the [+EPP] feature instead of Rickard. I have nothing to say
about the local movement of non-subjects to the topicalized position as the focus of this paper is the CP-phase level. I
assume a Chomsky (2000, 2001) style derivation can account for movement out of the v-phase.
8
that this derivation seems to violate the principle that merge is cheaper than move (Chomsky
2000, 2001). In order to check the [+Fin] and [+EPP] features in CP, the merge of att should be
cheaper than moving läste and Rickard. However, if we think of derivations proceeding in
phases, there is a simple explanation. I assume that both CP and CPVS are a strong phases, and
therefore have different numerations. If att is merged in CPVS, then att must be in the numeration
for CPVS, not in the numeration for CP. Thus, att is not available for merge in C, so V2 movement
must take place in order to check the features [+Fin] and [+EPP] in CP. Att then merges in CVS in
the next phase11. In other words, the CP in (8) is built up just as if it were a matrix CP. It can then
be embedded under a bridge verb because of the presence of CPVS directly above CP, which
providing a position where ‘nominalizing’ att can merge. Without CPVS (as in non-bridge
contexts) there is no way to mark a V2 CP in order to embed it.
The presence of CPVS containing [VSOP] above CP insures that the semantic component
will interpret the CP not as true, but as something along the lines of ‘possibly true’, meaning that
it may or may not be the case. [VSOP] serves to separate the truth-value of the embedded CP
from the truth-value of the matrix clause (or to somehow modalize it). In other words, just
because you ‘think’ something doesn’t make it so. The CPVS provided by the syntax is then used
by semantics to get the correct interpretation.
3.3. Bridge Verbs Without EV2, With and Without a Complementizer
As in English, an overt complementizer is optional under bridge verbs. In Swedish, the
complementizer is only optional in non-EV2 contexts. The complementizer is obligatory in
sentences like (8). In (9), the complementizer is optional12.
11
I take the lexical item att to serve as a kind of nominalizer, marking the contents of CP as a kind of nominal to the
syntax, as opposed to a verbal-like matrix clause. This eliminates the possibility of there being two occurrences of att
(in both C and CVS). More than one nominalizing marker for CP would be redundant, as there is no need to nominalize
a nominal. Following this line of thinking, CP is not projected by att, but by its syntactic features. Att only serves to
nominalize a CP. The presence of CP VS allows for a ‘second chance’ for CP to be nominalized by att, even after V2
movement in CP. This ‘spilling over’ of att into the next phase is therefore only possible under bridge verbs.
12 This optionality remains under matrix negation of bridge verbs and under irrealis verbs.
9
(9) Dan tror (att) Rickard läste boken i dag.
D. thinks (that) R. read book-the today
In this section I present two derivations of (9), one with att and one without.
3.3.1.
Bridge Verbs Without EV2, With a Complementizer
The structure of (9) when the complementizer is present is given in (10). There are two major
differences between (8) and (10). First, att merges in C in (10), checking both the [+Fin] feature
and the [+EPP] feature. For this reason, no XP needs to move to SpecCP to check [+EPP].
Second, the [+EPP] in CPVS is checked in (10) by movement of CP to SpecCPVS13. These
differences result because att is in the CP numeration in (10), but is in the CPVS numeration in (8).
(10) Dan tror att Rickard läste boken i dag.
D. believes that R. read book-the today
VP
ruV’
tDan
ruCPVS
V
ru CVS’
ttror
CP[+EPP]
to
ruC’ CVS[VSOP]
tCP
ru TP
C
ru T’
att
DP
ru vP
[+Fin]
5
T
ru v’
[+EPP] Rickardt
tRickard
ru VP
v
ru V’
lästei
V
ruAP
5
DP
5
boken
tläste
i dag
Because att is in the CP numeration, it must merge in C to check [+Fin], as opposed to the more
costly move of finite verb to C. Att also checks [+EPP] in CP, so XP movement to SpecCP is
ruled out. CP then merges with CVS. Since att has merged in C, CPVS still needs its [+EPP] feature
13
Note an important difference between CP and CPVS. CPVS bears a [+EPP] feature, but does not bear a [+Fin] feature,
while CP bears both [+Fin] and [+EPP].
10
checked. In order to check this feature, CP moves to SpecCPVS, providing CPVS with lexical
content14,15. Again, the semantic module is given [VSOP], rendering a non-factive interpretation
of CP.
3.3.2.
Bridge Verbs Without EV2, Without a Complementizer
In this section I argue that it is possible to give a syntactic explanation for complementizer
optionality. Instead of proposing a null complementizer or some form of complementizer
deletion, I show that the selectional properties of CVS can account for this phenomenon. Suppose
that CVS, in addition to CP, could alternatively select a TP directly. This derivation is given in
(11).
(11) Dan tror Rickard läste boken i dag.
D. believes R. read book-the today
VP
ruV’
tDan
ruCPVS
V
ru TP’
ttror
CVS[VSOP]
DP
5
Rickardt
TP[+EPP]
ru CVS’
tp
tTP
ru vP
ru v’
tRickard
lästei
ru VP
ru V’
DP
5
ruAP
V
5
boken
tläste
i dag
The TP is constructed in exactly the same way in (10) and (11), but then there is no CP and no
complementizer in the numeration. CVS selects TP directly. This is not surprising, as it is standard
for CPs to select TP, and CPVS is a type of CP. There is also no [+Fin] feature to be checked
14
Alternatively, att head movement from C to CVS could also check the EPP. However, I argue that the optionality of
the complementizer can be explained by selection. In order to keep the proposal consistent (see 3.3.2.) I will stay with
the CP/TP to Spec CPVS analysis. There are also benefits to this analysis that I exploit in section 5 to analyze factive
islands.
15 At this point, one might ask why the subject of the lower clause, the DP Rickard, does not move up to check [+EPP]
in SpecCPVS. The decision is made by the principle of economy, with CP being closer, and therefore a shorter move.
11
because of the lack of CP16. All that remains is [+EPP] in CPVS, and this is checked by TP moving
to SpecCPVS in the same manner as the CP moves in (10)17. The semantics of (11) work in the
same way as in the previous derivations.
3.4. Extraction
Another fact about EV2 in MSc is that EV2 clauses are islands for extraction. This fact is
accounted for in the derivation in (8). As the tree is building up, the topicalized V2 constituent
Rickard (which moves to check [+EPP]) in SpecCP blocks the avenue for movement. The A-bar
position is thus unavailable as a landing site for wh-movement, so there is no escape. The Spec of
CPVS is unavailable for movement for two reasons: First, movement to SpecCPVS would disobey
locality, skipping the A-bar position SpecCP. Second, even if locality could be violated, a whphrase would already be spelled out in its lower position by that point in a phase-based derivation
(Chomsky, 2001). I am assuming that when the next highest phase is completed, only the head
and Spec positions are still available for movement: the complement is spelled out by the
phonology. I take CP and CPVS to be strong phases. Therefore, once CPVS is complete, the only
two constituents available for movement in (8) would be Rickard and läste, in the Spec and head
position respectively. Any wh-phrase below the head position would be spelled out and hence
unavailable to the syntax.
Extraction is possible out of non-EV2 embedded clauses such as (10). Examining the
structure in (10), we see that SpecCP is an available A-bar position for movement. A wh-phrase
can escape through this position obeying locality conditions and phase construction rules. Once in
SpecCP, a wh-phrase gets a free ride to SpecCPVS when the CP moves up to check [+EPP]. It can
then move on from there to a higher position in the structure. Even if we consider the movement
16
As discussed in footnote 13, there is no [+Fin] feature associated with CP VS, only [+EPP].
A question that arises now - how come there is no att necessary for nominalization in (11)? The answer comes
through selection. I stipulate that V can select for CP VS or a nominalized CP, and CVS can select for a nominalized CP
or a TP. In other words, the only context in which there is no complementizer necessary (and in fact where no
complementizer is allowed) is when a V selects a CP VS in which CVS has selected a TP.
17
12
of CP to SpecCPVS as a completion of the next highest phase (CPVS), the Spec and the head of CP
are still available to the syntax (only the complement spells out).
Extraction in (11) works in a similar fashion to (10), with some minor differences due to
the lack of a CP in (11). Since there is no CP, TP moves into CPVS to check [+EPP]. If we assume
that this movement completes CPVS, then the complement of the next phase down spells out. In
this case the next lowest strong phase in vP. Therefore a wh-phrase is still able to move out from
under TP18. Since the derivations in the next section (3.5) are essentially the same as those in
(10) and (11), extraction will be assumed to work in the same manner.
3.5. Negated Bridge Verbs, Irrealis Verbs, and Negative Verbs (No EV2 allowed)
In this section I examine contexts in which EV2 is prohibited under bridge verbs. I argue that
CPVS is still present in these cases, as it is necessary for semantic interpretation. However,
movement possibilities are restricted because of a change in the nature of [VSOP] under certain
syntactic conditions. In her dissertation, Laka (1990) presented arguments for the existence of
negative complementizers that were licensed by negative verbs and negation, and showed how
they in turn licensed Negative Polarity Items (NPIs) in complement clauses. I propose that in
Swedish, the complementizer is not negative, but that instead the [VSOP] is. When licensed by
negation, negative verbs and/or irrealis verbs, [VSOP] takes on the character of the licenser,
much like Laka’s negative complementizer. This change in status of the operator is responsible
for different syntactic effects, including ruling out EV2.
3.5.1.
Negated Bridge Verbs, With and Without a Complementizer (no EV2 allowed)
As with non-negated bridge verbs, negated bridge verbs can appear with or without a
complementizer. This is shown in (12). Derivations are provided for each case below.
18
As mentioned in footnote 15, I am leaving aside the question of movement out of the vP-phase.
13
(12) Jag tror inte (att) Rickard läste boken i dag.
I believe not (that) R. read book-the today
3.5.1.1. Negated Bridge Verbs, With a Complementizer (no EV2 allowed)
(13) Jag tror inte att Rickard läste boken i dag.
I believe not that R. read book-the today
NegP
ruVP
inte
ruV’
tjag
ruCPVS
wo CVS’
CP[+EPP]
ru C’
CVS[NVSOP]
tCP
ru TP
C
ru T’
att
DP
[+Fin]
5
T
ru v’
[+EPP] Rickardt
tRickard
ru VP
V
ttror
ru vP
ru V’
v
lästei
5
tp
V
ruAP
DP
5
boken
tläste
i dag
The derivation in (13) is essentially the same as in (10). However, the operator in C VS has been
negativized. This negative operator [NVSOP] is licensed by the matrix negation inte. The
negative nature of this operator in CVS eliminates the possibility of the att merging there. We can
think of this prohibition as arising from the fact that the complementizer can not be negative.
Unlike Basque, as discussed in Laka (1990), Swedish (as does English) lacks separate lexical
negative and non-negative complementizers. Att appears in both contexts. I therefore take the
operator and the complementizer to be separate entities, and also see them as being noncompatible in the same head19. Since higher att merge is ruled out in the presence of [NVSOP],
EV2 as in (8) is also ruled out. As in (10), CP moves up to SpecCP VS to check [+EPP]. This
19
We can think of this incompatibility as a mismatch of features, with [NVSOP] having a negative feature.
14
movement puts CP above the operator. However, before CP moves, as the structure is being built,
the operator is in a C-command relationship with the CP and can therefore license any NPIs.
Semantically, we can think of the operator [NVSOP] as adding more content to CP VS for
interpretation. Namely, that the CP is expressly NOT the volunteered-stance of the speaker (or
subject of the matrix clause). In other words, the possibility of the embedded CP being true is
NOT what is being volunteered, the possibility of it NOT being true is. This information is thus
contained in CPVS.
3.5.1.2. Negated Bridge Verbs, Without a Complementizer (no EV2 allowed)
(14) Jag tror inte Rickard läste boken i dag
I believe not R. read book-the today
NegP
ruVP
inte
ruV’
tjag
ruCPVS
ttror
ru T’
DP
5
Rickardt
T
tRickard
V
TP[+EPP]
CVS[NVSOP]
ru vP
ru v’
ru VP
wp CVS’
tp
tTP
v
ru V’
lästei
DP
ruAP
5
V
5
boken
tläste
i dag
The derivation in (14) works much the same as (11), and (11) and (14) are parallel to (10) and
(13). The numeration for (14) lacks att and CP with its [+Fin] and [+EPP] features. TP moves to
SpecCPVS to check the[+EPP] there. The semantics works as in previous derivations.
3.5.2. Negation of Non-bridge Verbs (no EV2 allowed)
The analysis of negated non-bridge verbs is straightforward. There is no operator under these
verbs to begin with, so negation has no effect on any movement (and there is no EV2 movement
15
in this context anyway). NPIs are licensed in the embedded clause in the standard manner. (15)
shows the non-bridge context, and (16) the bridge.
(15)
VP
ru V’
ru CP
ruC’
ruTP
V
Non-bridge
C
5
att
(16)
VP
ru V’
V
bridge
ru CPVS
ruCVS’
ruCP
CVS
[NVSOP]
ruC’
ruTP
att
5
In (15), att merges in C. In (16), att also merges in C, and it is never able to merge higher because
it is incompatible with [NVSOP].
The presence of the operator in (15) versus (16) affects the semantic interpretation of the
lower clause. As discussed in section 3.5.1.1, [NVSOP] makes a different semantic contribution
than [VSOP]; namely, that the not only is the lower CP Volunteer-stance, but also that the lower
CP is NOT the stance of the speaker (or subject). Consider the English sentences in (17).
(17) a.
b.
c.
d.
John thinks [ (that) he went to the store]
John doesn’t think [ (that) he went to the store]
John regrets [ that he went to the store]
John doesn’t regret [ that he went to the store]
In (17a) it is John’s stance that ‘he went to the store’ is true. In (17b) it is John’s stance that ‘he
went to the store’ is NOT true. In both cases these are volunteer stances. The operator in (17a)
makes the CP Volunteer-stance, while the operator in (17b) gives the lower CP a negative
16
Volunteer-stance. (17d) on the other hand does not have an operator, so there is no effect on the
interpretation of the CP. Therefore, the truth-value of ‘that he went to the store’ is interpreted in
exactly the same way in (17c) and (17d), regardless of the presence of matrix negation. In both
sentences the contents of CP are interpreted as true (factive).
3.5.3. Irrealis Verbs and Inherently Negative Verbs
The data for Irrealis complement in Swedish is mixed. On one hand, the verbs önska ‘wish’ and
hoppas ‘hope’ seem to work in the exact same manner as negated bridge verbs. I analyze these
verbs as licensing an irrealis operator [IVSOP] in the head of CPVS just like the [NVSOP]
operator is licensed by negation. Att is optional for these complements, just as with negated
bridge verbs. The derivations proceed in the same way as in (13) and (14), and the semantic
contribution of [IVSOP] is that the embedded CP represents an ‘unreal’ situation or event.
On the other hand, there are irrealis predicates such as föredra ‘prefer’ and insistera på
‘insist’ (on) that require att to be present. This is a problem for me, because my proposal predicts
that these verbs, which are non-factive, should have the option to omit the complementizer. At
present I do not have a good explanation for these facts. I leave aside this class of irrealis
predicates (insist, prefer, demand etc.) for future research.
A similar problem arises for inherently negative verbs like betvivla/tvivla på ‘doubt’ and
förneka ‘deny’. These predicates are predicted to work in the same way as negated bridge verbs,
but they also take obligatory att. The English translations of these verbs work as expected, with
that being optional. I will also leave these verbs aside now, with the hope that future research will
bring a better understanding of what special properties they have in Swedish.
Finally, modalized bridge verbs (skulle + V ‘would’, borde + V ‘should’, etc.) behave in
the expected manner, like önska ‘wish’ and hoppas ‘hope’. EV2 is not allowed in their
complements, and att is optional. The derivations for these constructions work in the same way as
(13) and (14).
17
3.6. A Possible Solution to Some Puzzling Data From Swedish
Holmberg (1986: 144) presents some interesting Swedish data, repeated here as (18). These
sentences all have subordinate clause word order (non-V2) despite being main clauses.
(18) a. Kanske att hon kommer senare.
Maybe that she comes later
b. Aldrig att jag skulle gå dit.
Never that I would go there
c. Möjligen att Peter vet var hon är.
Possibly that P. knows where she is
In this fairly common construction, it seems like a bare sentence adverb is able to take a CP
argument. This construction is limited to certain adverbs, and Holmberg (1986:144) provides only
three examples; maybe, never, and possibly. Holmberg notes that these constructions seem to fall,
“outside of Swedish ‘core grammar’ ”, even though they are, “an at least partly productive
construction, and hence should be provided for by UG.” The present analysis offers a way to
account for these constructions. Kanske ‘maybe’ and möjligen ‘possibly’ are both epistemic,
giving a non-factive reading to a sentence. This is not unlike the semantic effect of [VSOP] on a
CP. Likewise, aldrig ‘never’ gives a similar semantic effect to the negative operator [NVSOP].
It is noteworthy that alltid ‘always’ does not license this construction.
(19) *Alltid att jag skulle gå dit.
Always that I would go there
The derivation given in (20) shows that the constructions in (18) are easily accommodated in the
present proposal. Negation is included to show that only non-V2 is grammatical in these
constructions.
18
(20) Möjligen att Peter inte kommer.
Possibly that Peter not come
CPVS
ru CVS’
Möjligen
ruCP
[+EPP]
CVS
[IVSOP]
ru C’
ru TP
ru T’
C
att
DP
ru NegP
[+Fin]
[+EPP]
Petert
5
T
inte
tPeter
ru vP
6
kommer
The CP is constructed in the usual manner and merged with CVS. Möjligen merges in SpecCPVS,
licensing [IVSOP] and checking [+EPP]. Since CPVS has the properties of a CP, it is able to head
the whole clause.
3.7. Summary
In this section I implemented the extra structure proposed in section 2 in a number of syntactic
derivations of Swedish sentences. I showed that extra structure (CPVS selected by bridge verbs)
and the operators contained therein can be exploited to explain a number of phenomena,
including limited EV2, the optionality of the complementizer att, and main clause att
constructions introduced by certain lexical adverbs. Also presented was a principled way to
strictly limit CP-recursion, and syntactic and semantic motivation for the extra structure and
operators. In the next section I give more evidence for CPVS from Hungarian and Japanese.
4. Hungarian and Japanese
In this section I give more evidence for the proposed extra structure from two non-Germanic
languages. These languages exhibit extra morphology in bridge verb contexts that I take as more
evidence for the existence of CPVS. This extra morphology, which appears to be generated in the
CP field, is associated with a non-factive interpretation of CP. This morphology, if analyzed as
being generated in a separate syntactic position from the complementizer, seems to go against the
19
Kiparskian (1970) prediction that extra structure (‘the fact’) should be associated with a factive
interpretation. What follows is a brief discussion of two languages that display extra morphology
in non-factive contexts. I begin with data from Hungarian and then present data from Japanese.
4.1. Hungarian Azt
As in Swedish, Hungarian embedded clauses (constructions with verbs taking a finite sentential
complement) exhibit two different patterns, one for bridge verbs and another for non-bridge
verbs. The difference is shown in (21)20.
(21) a. Azt
hiszem
that-ACC
I-think
'I think that Mary is smart.'
b. (*Azt)
sajnálom
that-ACC
I-regret
'I’m sorry that Mary is smart.'
hogy
Comp
Mari
Mary
okos.
smart-is
hogy
Comp
Mari
Mary
okos.
smart-is
In the examples above, the pronominal element azt comes from the lower clause, since it
represents the object of the matrix verb, which is the lower CP itself (For an analysis along these
lines, see Kiss, 1998). This pronoun is only present in cases where the matrix predicate is nonfactive (bridge). The fact that azt bears accusative case gives evidence that it comes from beneath
the verb. Therefore, azt can be argued to originate from a position between the matrix verb and
the overt complementizer hogy ‘that’. This is exactly the position where I argue that CPVS is
projected. (21b) shows the ungrammaticality if azt appears with a factive verb.
More evidence that the proposed extra structure leads to a non-factive reading comes
from the semantic interpretations of the sentences in (22). The interpretation of certain Volunteerstance predicates can move closer to a factive reading by leaving out azt.
20
Thanks to Barbara Ürögdi for the Hungarian data and discussion about them.
20
(22) a. Azt
hiszem
hogy
that-ACC I-think
Comp
'I think that Mary is smart.'
b. Hiszem
hogy
Mari
I-think
Comp
Mary
'I have no doubt that Mary is smart.'
Mari
Mary
okos.
smart-is
okos.
smart-is
More such examples can be constructed with other bridge verbs as well. The fact that the
omission of azt (which I assume originates in CPVS) is associated with the loss of a non-factive
interpretation is not surprising if one analyzes that the extra non-factive morphology (azt)
originates in CPVS. If certain bridge verbs are allowed to choose whether to take CPVS or CP
directly, the facts in (22) fall out. (22a) gets a Volunteer-stance reading because of the presence of
[VSOP] (evidenced by the presence of azt), while (22b) gets a factive reading because it lacks
[VSOP] (evidenced by the absence of azt).
The facts in Hungarian, along with the Japanese data to follow, give evidence that the semantic
information about Volunteer-stance vs. factive interpretation is located in the head of CVS (in the operator)
as opposed to the matrix verb. In these languages, complements to semifactives like ‘know’ differ in
interpretation depending on which structure the verb selects, CP or CP VS. This variation in interpretation of
complements under individual verbs leads me to believe that the determination of whether a complement is
factive or non-factive is decided by the presence or absence of the operator, not by the semantic/pragmatic
class of the verb. I therefore assume that the VS-operator provides the semantics with a Volunteer-stance
interpretation, and this interpretation is unavailable under factive predicates.
4.2. Japanese
Satoko Suzuki (1997) gives an interesting account of the role of factivity in complementizer
choice in Japanese21. The general pattern given for Japanese complementizers is in (23) and
(24)22.
21
For alternative analyses of Japanese, see Fukui (1998), Kurafuji (1996), and Yanagida (1996). Yanagida argues for a
factive operator analysis, a la Melvold (1986).
22 For more Japanese data, the reader is referred to Suzuki (1997).
21
(23) Predicates which take sentential subjects
a. Factives take no, koto, to yuu no, and to yuu koto.
b. Non-factives take to yuu no and to yuu koto.
(24) Predicates which take sentential objects
a. Factives take no and koto.
b. Non-factives take to
However, Suzuki notes that there are exceptions to these generalizations. Much like in the
Hungarian examples in (22), extra structure (or the lack of extra structure) seems to affect the
interpretation of embedded clauses with regard to factivity. I will first discuss predicates that take
sentential subjects. As shown in (25), factive predicates have more complementizer options than
non-factives23. Data from Kuno (1973: 218-9).
(25) a. John ga kekkon tyokugo sindesimatta {koto/no/to yuu koto/to yuu no} wa higeki da.
J. Sub. marriage right-after died { complementizer
} Top. tragedy is.
‘It is a tragedy that John died right after he got married.’
b. John ga Mary o nagutta {*koto/*no/to yuu koto/to yuu no} wa uso da.
J Sub. M. DO. hit:past {
complementizer
} Top. lie is.
‘It is false that John hit Mary.’
Only the factives can have bare no or koto (25a), while non-factives obligatorily have to yuu as
well (25b). As for the difference in meaning in factives with or without to yuu, Suzuki writes the
following: (Suzuki 1997: 294)
Since to yuu functions to indicate lack of strong conviction on the part of the speaker about the truth of the
proposition expressed in the complement clause, its occurrence is obligatory when the predicate of a
sentence is a noncommittal non-factive or a counterfactual non-factive. When it occurs with factives or
assertive non-factives, it indicates weakening of conviction.
It seems that by adding structure in Japanese sentential subjects, one can get a non-factive, or less
factive reading24. This mirrors the Hungarian data discussed in section 4.1., where the presence of
extra morphology (and in my view, extra structure) is associated with a non-factive interpretation.
23
I assume that the whole complex (to yuu koto for example) is not simply one long complementizer, but is a
combination of elements housed in both CP VS and CP.
22
One could imagine to yuu being housed in CPVS along with [VSOP], and thus the non-factive
interpretation would be provided for the semantics25.
In the domain of sentential objects, the complementizer choice is koto or no for factives
and to for non-factives26. As with sentential subjects, the situation with sentential objects is more
complicated than the generalization. In some cases, factive predicates can occur with the nonfactive complementizer to. Data from Suzuki (1997:296)27
(26) Sorekara firipin to yuu kuni
ga firipain to hatsuonsareru koto mo
and Phillipines Quot. say country Sub. Filipain as pronounce:Pass. thing also
aru to
shitta
toki mo bikkurishita
exist Quot. find:out:Past. Time also be:surprised:Past.
‘And when I found out that the country called the Philippines can be pronounced
filipain, I was also surprised.’ (Hotta 1957:101)
In (26), to is used despite the fact that shitta ‘find out’ is factive. Suzuki argues that to is used
instead of koto/no because the information in the complement clause is not really ‘knowledge’
that is presupposed, but a ‘sudden realization’. In Cattell’s terms we could say that this
information is Volunteer-stance. The examples Suzuki gives are all with predicates referred to by
Karttunen (1971) and Hooper (1975) as 'semifactives' (realize, discover, notice, know, etc.). They
can pattern with both factives and non-factives. The choice of complementizer in Japanese signals
which interpretation the clause leans towards.
In addition to factive predicates occurring with to, there are also non-factive predicates
that appear with koto, a factive complementizer. As would be expected, koto gives the non-factive
predicate a more factive reading. An example of this is given in (27) (Suzuki 1997:300).
24
Suzuki argues that factivity is not the deciding factor for complementizer choice, but speaker conviction is. Her
arguments are compatible with a Cattell/Hegarty classification scheme (stance or familiarity) as discussed below in
section 6.
25 I abstract away from the actual workings of Japanese syntax, as it is not the focus of this paper. However, I assume
that the details could be worked out, and I leave this to future research.
26 I take bare to to be in CP
VS like to yuu, and to be related to to yuu because of the morphological similarity and the
semantic effects they both signal.
23
(27) Demo anata no hoo koso konna katachi de wakare o tsugeru jinsei
but you Link. side indeed this:kind form in farewell DO. say
life
nado
zettai ni erabitakunakatta
koto o omoi, watashi
something:like never Adv. Want:to:choose:Neg.:Past.
DO. think:Inf. I
wa umarete hajimete chooji
nado
o yomu no desu
Top. born:and first:time memorial:address something:like DO. read Nom. Is
‘But, thinking that you never wanted to choose a life that ends like this, I will read a
memorial address for the first time in my life.’ (Bungei Shunjuu 1986:245)
(27) was uttered by a speaker at the funeral of her best friend, who was killed in a bomb
explosion. This to me would be defined as a Non-stance context, with the speaker assuming that
the complement clause is presupposed (factive reading)28. As with the Hungarian examples in
(22), I see these Japanese cases as a function of selectional properties of certain bridge verbs
(semifactives seem to be able to select either CPVS or CP), with the semantic consequences
entailed.
As in sentential subjects, to yuu can be added to koto (and infrequently to no) in
sentential objects29. Suzuki explains that to yuu koto as compared to bare koto:
...should be regarded not merely as showing lack or weakening of the speaker's conviction but as
indicating a lesser degree of internalization. (Suzuki 1997: 305)
I see this 'weakening' and 'lesser degree of internalization' as a Volunteer-stance reading, caused
by the presence of [VSOP] (signaled by to yuu, which I take to be in CPVS).
4.3. Summary
In this section I have presented morphological and semantic evidence from Japanese and
Hungarian that a non-factive reading is associated with more structure in the CP field, not less
(contra Kiparsky & Kiparsky, 1970). The absence of this structure leads to a factive
interpretation. While a detailed analysis of the syntactic workings of the extra structure in these
27
The reader will notice that Suzuki glosses to as a quotation marker. For me what is important is its place in the CP
field, not its labeling. In addition, I have nothing to say about the other occurrences of to and/or yuu in (28).
28 For more detail on stance verbs, see section 6.
29 For the relevant data, see Suzuki (1997).
24
two languages is beyond the scope of this paper, the present proposal is a promising direction for
future research.
In the next section I discuss another area where the CPVS analysis looks
promising, factive islands.
5. More on Extraction: Factive Islands
It is widely recognized that there is an extraction asymmetry from the clausal complements of
factive vs. non-factive predicates. Factive CPs are weak islands for adjunct extraction in both
English and M.Sc.. This is evidenced in (28) (non-factive) and (29) (factive).
(28) a. How do you think that you behaved?
b. Hur tycker du att du uppträdde?
c. Hvordan tror du at du oppførte deg?
d. Hvordan synes du at du opførte dig?
(Eng.)
(Swe.)
(Nor.)
(Dan.)
(29) a. *How do you regret that you behaved?
b. *Hur ångrar du att du uppträdde?
c. *Hvordan angrer du at du oppførte deg?
d. *Hvordan fortryder du at du opførte dig?
(Eng.)
(Swe.)
(Nor.)
(Dan.)
Further evidence of factive islands is given in (30) and (31). Why-extraction gives two possible
interpretations of the sentences with ‘think’ in (30), while the factive verb ‘forget’ allows only
one interpretation, questioning the matrix verb.
(30) a. Why do you think that John went to the store?
b. Varför tror du att John gick till affären?
c. Hvorfor tror du at John dro til butikken?
d. Hvorfor tror du at John gik i butikken?
(Eng.)
(Swe.)
(Nor.)
(Dan.)
Two possible answers:
e. Because he told me he was going.
f. Because he needed milk.
(31) a. Why did you forget that John went to the store?
b. Varför glömde du att John gick till affären?
c. Hvorfor glemte du at John dro til butikken?
d. Hvorfor glemte du at John gik i butikken?
One possible answer:
e. Because I was drunk.
f. *Because he needed milk.
(Eng.)
(Swe.)
(Nor.)
(Dan.)
25
There also seem to be differences in movement possibilities out of Japanese complements under
factive vs. non-factive verbs, as data from Yanagida (1996:292) illustrates30.
(32) a. Taroo-wa wine-wa
katta
to
omotta.
Taroo-Top. wine-Top./Foc. bought Comp. thought
‘Taroo thought that as for wine, he bought it.’
‘Taroo thought that he bought wine (but not beer).’
b. Taroo-wa wine-wa
katta koto-o
kookai-sita.
Taroo-Top. wine-Top./*Foc. bought Comp-Acc. regretted
‘Taroo regretted that as for wine, he bought it.’
*‘Taroo regretted that he bought wine (but not beer).’
Yanagida analyses focus as being achieved by syntactic movement, and this movement is
unavailable under factives, ruling out contrastive reading of wa in (32b)31. Another asymmetry in
Japanese is in LF extraction of naze ‘why’. LF adjunct extraction is available under bridge verbs
like ‘say’, but not under non-bridge verbs like ‘regret’32.
(33) a. Bill-wa John-ga naze kubi-ni natta tte itta no?
Bill-Top. John-Sub. why was
fired Comp. said Q.
‘Why did Bill say that John was fired t ?’
b. ?? Bill-wa John-ga naze kubi-ni natta koto-o oshinda no?
Bill-Top. John-Sub. why was
fired Comp. regretted Q.
‘Why did Bill regret that John was fired t ?’
The facts in (33) mirror the English and M.Sc. facts in (30) and (31).
As in English and M.Sc., extraction of arguments in Japanese is allowed (at LF) under
both factives and non-factives. The analysis argued for in this paper gives a way to account for
the facts. The structures for non-bridge vs. bridge sentences given in (3) and (4) are repeated here,
and (34) represents non-EV2 bridge verb embedded clauses like (10) and (11).
30
Thanks to Miyuki Sawada for Japanese references and discussion.
Yanagida’s analysis of what causes a factive island (factive operator) differs from the analysis that I am exploring.
32 The examples in (35) are adapted from Fukui (1988:508), but in (35b) I have replaced the manner of speaking verb
‘whisper’ with the factive verb ‘regret’ and changed the complementizer accordingly. Fukui observes that manner of
speaking verbs in Japanese, as in English, block adjunct extraction. While manner of speaking verbs pattern with nonfactives in extraction, they seem to pattern with Volunteered-stance predicates in interpretation. I leave the patterning
of manner of speaking verbs to future research.
31
26
(3)
VP
ru V’
ru CP
Non-bridge V
ruTP
5
(4)
VP
ru V’
ru CPVS
ru CP
Bridge V
ruTP
5
(34)
VP
ru V’
ru CPVS
ru CVS’
Bridge V
CPt
4
ru
CVS
tCP
TP
As noted in this section, adjunct extraction is not allowed out of the CP in (3), and as noted in
section 4.4., all extraction from EV2 complements, which have the structure in (4), is disallowed.
The only structure that allows adjunct extraction is (34). (34) is also the only structure in which
CP has moved into SpecCPVS.
If we think of derivations taking place in phases (as in Chomksy, 2001), this move could
be significant for the timing of phonological spell-out of lower clauses containing adjuncts.
Movement of the CP up may allow for a later spell-out of material lower in the CP, as the
complement of C should not be spelled out until completion of the next highest strong phase.
CPVS, the next highest strong phase, is not completed until after CP has moved into SpecCPVS. At
that point in the derivation, the whole CP is no longer a complement of CPVS, so it is plausible
that it still would not be spelled out, leaving its contents (like adjuncts) available for further
movement. A related possibility is that a late spell-out allows for a late insertion of adjuncts,
which are then able to move up. Without this late spell-out, it may be impossible to insert
27
adjuncts. What the exact details are concerning factive islands will be left to further research33.
However, the present proposal seems to have promising possibilities for accounting for these
extraction facts.
6. What Verbs License EV2?
Having accounted for the different syntactic effects associated with bridge verb vs. non-bridge
verb contexts, I will now define what constitutes a bridge verb. In this section I review various
proposals for categorizing verbs and select what I find to be the most accurate classification.
6.1. Factivity and Assertion
In their classic paper, "Fact", Kiparsky & Kiparsky (1970) (K&K) explore the syntax-semantics
interface in the English complement system. They note that there are two classes of predicates,
those that assert the truth of their sentential complements (non-factives) and those that presuppose
the truth of their sentential complements (factives).
(35) a. I believe [that it is raining].
b. I regret [that it is raining].
In (35a),- the truth of the sentential complement is not presupposed, while in (35b) it is. This
remains the case if the matrix clause is negated, as in (36).
(36) a. I don’t believe [that it is raining].
b. I don’t regret [that it is raining].
K&K also note that this semantic classification is reflected in the different syntactic behavior of
complementation for the two types of predicate. Among these differences were (K&K 1970:
347):
(37) Only factive predicates can have as their objects the noun fact with a gerund or thatclause:
33
While this explanation looks promising for English and M.Sc., the possibility for a spell-out type solution for the
Japanese data is less clear. I leave this for future research.
28
a. Factive:
I want to make clear the fact that I don't intend to participate.
b. Non-factive: *I assert the fact that I don't intend to participate.
(38) Gerunds can be objects of factive predicates, but not freely of non-factive predicates:
a. Factive:
I regret having agreed to the proposal.
b. Non-factive: *I believe having agreed to the proposal.
(39) Only non-factive predicates allow the accusative and infinitive construction (ECM):
a. Non-factive: I believe Mary to have been the one who did it.
b. Factive:
*I resent Mary to have been the one who did it.
To account for the semantic and syntactic differences between factive and non-factive predicates,
K&K hypothesize that presupposition of complements is reflected in their syntactic deep
structure. According to their framework, the structure for sentences like those in (35) is (40).
(40) a.
NP
fact
b.
S
Factive
NP
S
Non-factive
Two optional transformations provide the surface structure in (37-39); deletion of the head noun
fact, and formation of gerunds from that-clauses in position after nouns. The facts in (37-39) fall
out from the structures in (40). (37a) is fine because the fact is present in the deep structure and
not deleted, but (37b) is out because the fact it is not originally there and there is no place to add
it in the structure. (38a) derives easily from the proposed deep structure, but (38b) is not possible
because the gerund formation transformation only occurs after nouns. (39b) is ruled out as a
Complex NP Constraint violation (Ross, 1967), assuming an operation of raising to object. While
I share the intuition that a semantic difference is responsible for a syntactic difference, the present
proposal differs significantly from K&K. For K&K, extra structure is associated with factives,
while I have presented arguments that extra structure goes with non-factives.
29
Hooper & Thompson (1973) (H&T) examine contexts in which root phenomena occur in
embedded clauses in English. They conclude that these phenomena can only occur in clauses that
are asserted. They divide factive and non-factive verbs into five distinct groups according to
whether or not their complements can be asserted. Andersson (1975) offers a modification of
these five groups, and provides a large amount of data from Swedish. Included in Andersson’s
data are verbs that license EV2, and these are the same as H&T’s verbs that allow embedded root
phenomena in English. Neither H&T nor Andersson give a detailed syntactic analysis using their
classifications. They rely more on a semantic/functional description of the phenomena in
question. Problems with their analyses are summarized in Heycock (2000:15-17).
6.2. Stance Verbs and Familiar Complements
Cattell (1978)34, in an investigation of Why-extraction in English, further modifies H&T’s
categories of verbs which take sentential complements. He divides these verbs into three classes:
Volunteer-stance, Response-stance, and Non-stance.Hegarty (1992: f.12) summarizes Cattell’s
classes as follows:
(41) a. Non-stance: Presuppose, as a matter of conventional meaning, that the content of
the complement is part of the body of background beliefs. This class includes most
standard factive verbs.
b. Response-stance: The content of the complement is under consideration for
inclusion in the body of background beliefs. This class includes the nonfactives
confirm and verify, and the negative verbs deny and doubt.
c. Volunteer-stance: The content of the complement is being offered by the speaker
for inclusion in the body of background beliefs.
As Hegarty notes, the classification in (41) represents a departure from the traditional factive vs.
nonfactive distinction. Cattell (1978: 69) notes that only Volunteer-stance (VS) verbs can occur
with ambiguous why, whereas Non-stance (NS) and Response-stance (RS) verbs do not show this
ambiguity:
(42) a. Why do they think (that) Sue killed Harry? (VS – ambiguous)
34
Thanks to Anna Szabolcsi for suggesting this reference.
30
b. Why do they accept35 that Sue killed Harry? (RS – not ambiguous)
c. Why did Richard comment that Sue killed Harry? (NS - not ambiguous)
While why in (42a) can be interpreted in the matrix or embedded clause, in (42b) and (42c) only
the matrix reading is available. Why-extraction thus seems to be available only from complements
of VS verbs as opposed to NS or RS verbs. Stance verb classes represent a more accurate
classification than a factive/non-factive distinction.
Hegarty (1992) classifies verbs in a similar manner to Cattell. His ‘Class A’ predicates
are identical to Cattell’s VS class, and his ‘Class B’ predicates are a union of RS and NS verbs
with negative verbs such as deny, regret and doubt removed. However, Hegarty also argues that
Class B predicates are syntactically marked if they meet various background knowledge criteria
in the discourse, or the linguistic or pragmatic context. A syntactic feature Familiar [F] is added
to C in ‘familiar’ complements. [F] marked complements become islands for extraction of why,
among other syntactic effects.
Hegarty also noticed that his verb classification closely fits Vikner’s (1995) list of Danish
verbs that allow EV2 and those that do not, repeated here as (43) and (44)36.
(43) Verbs which allow sentential complements with and without EV2:
(Dan.) antyde
(Eng.) hint
angive
indicate
svare
answer
påstå
claim
berette
report
betone
emphasize
beslutte
decide
erfare
learn
huske
remember
slå fast
ascertain
synes
think
tro
believe
håbe
hope
mene
mean
sige
say
se
see
føle
feel
formode
assume
vide
know
I have changed the verb from Cattell’s original ‘deny’ to another RS verb ‘accept’ because I would (as does Hegarty)
classify ‘deny’ as an inherently negative verb, and it therefore patterns differently. I will return to this point below.
36 All the predicates listed that allow EV2 take or can take what Hegarty calls ‘novel’ complements (except for betone,
‘emphasize’), while all the non-EV2 predicates listed are either intrinsically negative or take ‘familiar’ complements,
which Hegarty argues have negative [N] or familiar [F] features blocking CP-recursion. This is a favorable result for
Hegarty’s analysis. Also note that Hegarty’s classes also match the Swedish data given in Andersson (1975). The
points of difference between Hegarty and Cattell have to do with some of H&T’s ‘semi-factive’ predicates like know,
realize and point out.
35
31
(44)
Verbs which allow sentential complements, but not EV2:
(Dan.) beklage
(Eng.) be sorry
bekræfte
confirm
fortryde
regret
bevise
prove
tvivle på
doubt
bede om
ask for
tænke på
think of
tillade
permit
holde hemmeligt
keep secret
være glad for
be happy
hade
hate
overse
overlook
overbevise om
convince
tillgive
forgive
forlange
demand
fortie
conceal
vise
show
indrømme
admit
This ties the facts together neatly, showing that bridge verbs are the class of verbs that allow EV2
and why-extraction. As far as I know, Hegarty is the first to have noticed this generalization. I
have therefore concluded that the best characterization of these groups is given by Cattell and
Hegarty37. I have adopted Cattell’s terminology of Stance verbs, with the caveat that I, like
Hegarty, separate inherently negative verbs (deny etc.) into their own category. I have not
adopted Hegarty’s notion of familiar complements, but I believe his account could be
incorporated into the present work. I leave a further sharpening of these verb classes to future
research. What is important for me is that the Cattell/Hegarty classes are to be preferred to a
factive/nonfactive distinction. Cattell’s slightly modified verb classes will suffice for the present
work.
7. Summary:
In this paper I have proposed an extra layer of CP structure to account for Limited Embedded
Verb Second facts in Mainland Scandinavian. This extra structure is selected by bridge verbs,
which I define as Volunteer-stance predicates as opposed to there being a factive/non-factive
distinction. Semantic interpretation is affected by the extra layer, and freer syntactic movement is
While I agree with Hegarty’s verb classification, I differ from his syntactic analysis of the facts. Hegarty argues that
extra structure is associated with ‘factive’ type verbs, blocking movement. I argue however, that extra structure is
associated with ‘nonfactive’ type (VS) verbs, and this extra structure facilitates movement.
37
32
facilitated. The proposal also accounts for complementizer optionality in English and Swedish,
and offers a way to account for factive island effects.
References
Alexiadou, Artemis, and Elena Anagnostopoulou. 1998. Paramaterizing AGR: Word order,
verb-movement and EPP checking. NLLT 16:491-539.
Andersson, Lars-Gunnar. 1975. Form and function of subordinate clauses. Doctoral
Dissertation. Göteborg Universitet: Göteborg, Sweden.
Besten, Hans den. 1983. On the interaction of root transformations and lexical deletive rules.
ed. John Benjamins On the formal syntax of the Westgermania. Amsterdam.
Cattell, Ray. 1978. On the source of interrogative adverbs. Language 54.1:61-77.
Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press.
Chomsky, Noam. 2000. Minimalist inquiries: The framework. In Step by step: Essays on
minimalism in honor of Howard Lasnik, ed. Roger Martin, David Michaels, and Juan
Uriagereka, 89-155. Cambridge, Mass.: MIT Press.
Chomsky, Noam. To appear. Derivation by phase. In Ken Hale: A life in language, ed. Michael
Kenstowicz. Cambridge, Mass.: MIT Press.
Cinque, Guglielmo. 1990. Types of A'-Dependencies. MIT Press: Cambridge, Mass.
de Cuba, Carlos, and Barbara Ürögdi. 2001. The Role of factivity in the syntax and semantics of
embedded clauses. Paper presented at SUNY/CUNY/NYU Mini-conference: SUNY
Stony Brook, Spring 2001.
Fukui, Naoki. 1988. LF Extraction of Naze: Some theoretical implications. NLLT: 6. 503-526.
Green, Georgia. 1976. Main clause phenomena in subordinate clauses. Language 52: 382-397.
Hegarty, Michael. 1992. Familiar complements and their complementizers: On some
determinants of A’- locality. Unpublished manuscript: University of Pennsylvania.
Heycock, Caroline. 2000. Embedded root phenomena. To appear in An electronic encyclopedia
of case studies. Henk van Riemsdijk and Martin Everaert, Editors. The LingComp
Foundation.
Presently
available
on
the
worldwide
web
at:
http://www.ling.ed.ac.uk/~heycock/papers/index.html
Holmes, Philip, and Ian Hinchliffe. 1994. Swedish: A Comprehensive Grammar. Routledge
Grammars.
Holmberg, Anders. 1983. The Finite sentence in Swedish and English. WP in Scandinavian
Syntax. Linguistics Department: University of Trondheim.
Holmberg, Anders. 1986. Word order and syntactic features in the Scandinavian languages and
English. Doctoral Dissertation, Department of General Linguistics: University of
Stockholm.
Holmberg, Anders, and Christer Platzack. 1995. The Role of Inflection in Scandinavian Syntax.
Oxford University Press.
Hooper, Joan. 1975. On assertive predicates. ed. Kimball, John P. In: Syntax and Semantics 4.
91-124. Academic Press: New York.
Hooper, Joan, and Sandra Thompson,. 1973. On the applicability of root transformations. LI
Vol. IV, number 4: 465-497.
Iatridou, Sabine, and Anthony Kroch. 1992. The Licensing of CP-recursion and its relevance to
the Germanic verb-second phenomenon. Working Papers in Scandinavian Linguistics.
50: 1-25.
Karttunen, Lauri. 1971. Some observations on factivity. Papers in Linguistics 4: 55-69.
Kiparsky, Paul, and Carol Kiparsky. 1970. Fact. ed. M. Bierwisch and K. Heidolph. In:
33
Progress in Linguistics. The Hague: Mouton.
Kiss, Katalin É. 1998. Hungarian syntax. in: Új magyar nyelvtan (New Hungarian Grammar).
Hungarian Academy of Sciences: Budapest 7:169-182
Koster, Jan. 1978. Why subject sentences don't exist. ed. Keyser. In: Recent Transformational
Studies in European Languages:53-64.
Kurafuji, Takeo. 1996. Unambiguous checking. MIT Working Papers in Linguistics 29: 81-96.
Laka, Itziar. 1990. Negation in syntax: On the nature of functional categories and projections.
Doctoral Dissertation, MIT, Cambridge, Massachusetts.
Larson, Richard K., and Gabriel Segal. 1995. Knowledge of Meaning. MIT Press: Cambridge,
Mass.
Melvold, Janis. 1986. Factivity and definiteness. Draft, Dept. of Linguistics and Philosophy
MIT.
Platzack, Christer. 1998. Svenskans inre grammatik - det minimalistiska programmet.
Studentlitteratur: Lund, Sweden.
Platzack, Christer. 1986. Comp, infl, and Germanic word order. ed. L. Hellan and K. Koch
Christensen. Topics in Scandinavian Syntax:185-234.
Ross, John Robert. 1967. Constraints on variables in syntax. Doctoral Dissertation. MIT,
Cambridge, Massachusetts.
Suzuki, Satoko. 1997. The relevance of factivity to complementizer choice in Japanese.
Studies in Language 21:2:287-311.
van Riemsdijk, Henk. 1982. A Case study in syntactic markedness. Foris Publications.
Vikner, Sten. 1995 Verb Movement and Expletive Subjects in the Germanic Languages. Oxford
University Press.
Yanagida, Yuko. 1996. Deriving surface word order in discourse configurational languages.
MIT Working Papers in Linguistics 29: 283-302.
Download