Draft March 2000; comments welcome
✞
Infinitival to
Yehuda N. Falk
Department of English
The Hebrew University of Jerusalem
✟
✒
✖
✖
✗
✘
✙
✙
✗
✠
✌
✡
✓
☛
✖
☞
✚
✌
✑
✍
✟
✎
✠
✟
✏
✠
✏
✏
✏
✑
✑
✒
✓
✒
✔
✓
✔
✕
✕
✑
✑
☞
☞
✏
✑
✏
✕
✑
✌
✕
✙
✌
✛
✟
✠
✡
☛
☞
✌
✍
✙
In the welcome rush to study linguistic universals, narrow parochial questions of the
syntax of English tend to get ignored. This paper is a study of a narrow parochial question of the
syntax of English: the categorization of the infinitive marker to. Naturally, the resolution of
questions of this nature will also have consequences for a theory of Universal Grammar.
For the most part, we will try to remain as theory-neutral as possible. This means that
theory-internal arguments will not be invoked. However, the conclusion that we reach will have
clear theoretical implications. We also must assume some theory of categories. We will assume
a fairly conventional theoretic view of categories (Chomsky 1986; Bresnan to appear). In the
verbal/clausal system, we assume the existence of three categories: V(erb), I(nfl), and
C(omplementizer). The category V is the lexical core of a clausal constituent. The category I
carries tense, aspectual, and agreement features: in English, modal auxiliaries belong to this
category and (in surface constituent structure) tensed forms of be and auxiliary have occupy the
I position as well. In other languages, all tensed verb forms may be in I, or I may always be a
(morphosyntactically) free form with no verbal elements.1 The category C is the familiar
category of complementizers. In English, the specifier position of IP is the (surface) position of
subjects and the specifier position of CP is the (surface) position of wh operators. We take this
to be a fairly middle-of-the-road position: unlike some lexicalist researchers, we accept the
category Infl, but unlike much research in the GB/Minimalist tradition, we do not recognize
other functional categories (AGR, ASP, TRANS, etc.) which are not lexically justifiable.
There seem to be three basic views in the theoretical literature concerning the
categorization of to: C (Postal and Pullum 1978, Falk 1984a, Sag and Wasow 1999), I (Chomsky
1981, Koster and May 1982), and V (Pullum 1982, Pollard and Sag 1994).2 Couched in this
superficial disagreement is the consensus that to is a head within the clausal system: it heads
either a CP, an IP, or a VP.
✜
This paper was inspired by Pullum (1982). I hereby acknowledge this debt. This paper has benefitted
from comments on earlier drafts by Joan Bresnan, Edit Doron, Tsipi Kuper-Blau, Anita Mittwoch, and Yael Ziv.
As in Pullum (1982), the sample sentences in this paper are inhabited by characters from
.
I would like to thank the producers of the various
series, and the late Gene Roddenberry, for
creating such a “fascinating” universe.
✁
✁
✂
✄
☎
✂
☎
✆
✂
✄
☎
✂
☎
✆
✝
✝
1
Whether Infl can include sublexical features is a theory-internal question: in derivational theories it
generally does, while in theories that have some version of the lexical integrity principle (such as LFG) it does
not.
2
In addition, Mittwoch (1990) argues that to is a modal auxiliary, without committing to a specific
category identity.
2
(1)
C
a.
C
I
b.
✢
VP
to
I
V
c.
✢
VP
to
V
✢
VP
to
We assume the head status of to in what follows. (For discussion and rejection of other
possibilities, see Pullum 1982.)
The position to be argued for here is that to is a member of category C. The argument
will be somewhat indirect. We will first show that the to infinitive must be a CP. This constitutes
an indirect argument for the C-hood of to, because it seems to head a CP. However, the CP-hood
of a to infinitive does not necessarily mean that to is C, because infinitives could also be
analyzed with to as I or V with a null complementizer heading the CP. We will then proceed to
show that arguments purporting to show that to is an I or V are faulty.
1. Infinitives as CP
Infinitives must be CPs because they have the same distribution as CPs. In environments
where it is possible to distinguish between CP, IP,3 and VP distribution, it is clear that infinitives
pattern with CPs.4 For example, CPs can be subjects; IPs and VPs cannot. Infinitives can also
be subjects.5
(2)
a.
b.
c.
d.
That Kirk defeated a Gorn surprised me.
*Kirk defeated a Gorn surprised me.
*Defeat a Gorn surprised / would surprise me.
To defeat a Gorn would surprise me.
The CP status of infinitives also emerges from a consideration of complements to various
categories. For example, CPs and infinitives can serve as arguments to nouns, while IPs and
VPs cannot.
(3)
a.
b.
Picard's intention that the Enterprise would fight the Borg
*Picard's intention the Enterprise would fight the Borg
3
I am assuming here that finite clauses with no overt evidence of a complementizer or material in [SPEC,
CP] are bare IPs, rather than CPs with null complementizers. In a theory in which this is impossible, IPs have no
distribution other than as complement of C. I am also assuming that the analysis of Raising-to-Subject infinitives
and ECM/Raising-to-Object infinitives as bare IPs is not a given. It seems to me that this is a position that needs
to be argued for rather than stipulated for theory-internal reasons. If the conclusion to be reached here that to is
not an I is correct, this analysis of Raising infinitives must be wrong.
4
Many of the distributional properties would also be consistent with the to infinitival being a PP, which
is its historical source. For arguments against a PP analysis, see Pullum (1982).
5
This argument is mentioned by Pollard and Sag (1994). However, since they accept Pullum's (1982)
analysis of to as a verb, they analyze infinitives of this kind as having an empty complementizer. The structure
(replacing their VP[comp] with CP) is:
(i)
[CP e [VP to defeat a Gorn]]
3
c.
d.
*Picard's intention fight(ing) the Borg
Picard's intention to fight the Borg
VPs cannot serve as complements to adjectives; CPs, IPs, and infinitives can.
(4)
a.
b.
c.
d.
proud that he serves in Starfleet
proud he serves in Starfleet
*proud serve/serving in Starfleet
proud to serve in Starfleet
On the other hand, clausal complements to prepositions can only be IP; CP, VP, and infinitives
are all ungrammatical.
(5)
a.
b.
c.
d.
*before that she served in Starfleet
before she served in Starfleet
*before serve in Starfleet
*before to serve in Starfleet
Another distributional argument for the CP status of to infinitivals is the fact that they
can be introduced by wh operators.
(6)
a.
b.
Janeway asked how to protect the Ocampa.
Odo didn't know who to charge with the crime.
On the generally accepted view that the structural position occupied by wh operators is the
specifier position of CP, this is further evidence that infinitival clauses are CP.
As noted earlier, the demonstration that infinitives are CP is not sufficient to argue that
to is C. It is possible that infinitival CPs have an empty C with an IP or VP complement. That
is to say, the possible structures for infinitives are now as follows (with a neutral position on the
structural presence of the unexpressed subject).
(7)
C
a.
C
C
b.
✣
VP
✣
C
IP/VP
✤
to
e
DP
PRO
I /V
✥
✣
I/V
✣
VP
to
However, we assume that the simpler structure in (7a) is a priori the preferable one. That is to
say, the burden of proof is on the proponent of the more complex structure in (7b) to show that
to is an I or V.
4
2. NonArguments for I- or V-hood
In this section we show that arguments that have been raised in the past that to is I or V
are invalid.
2.1. VP Ellipsis
VP Ellipsis has been used as an argument both for the analysis of to as I (Koster and May
1982) and for the analysis as V (Pullum 1982). Apparently, the argument is that English has a
rule of the form:
(8)
In the configurations [I VP] and [V VP], the VP can be optionally ellipted/omitted/
deleted. The I/V is stranded.
This rule accounts for the grammaticality of the following cases of VP Ellipsis.
(9)
a.
b.
c.
d.
Uhura canI [fix the transmitter].
In a fit of anger, Q didI [send the Enterprise into Borg space].
Zephram Cochrane might haveV [settled in the Alpha Centauri system].
Trelane stoppedV [chasing Kirk].
The requirement that an I or V be stranded is supposed to account for the ungrammaticality of
the following.
(10)
a.
b.
*In a fit of anger, Q [sent the Enterprise into Borg space].
*On its maiden voyage, Kirk [saved the Enterprise-B].
Given this analysis of ellipsis, there is a contrast between the ability of uncontroversial
complementizers and to to be stranded by ellipsis.
(11)
a.
b.
c.
*McCoy thinks that [Spock is too unemotional].
*The Horta attacked as soon as [her eggs were stolen].
*Guinan asked if [anything was wrong].
(12) *Kira hoped for [Bareil to be elected Kai].
(13)
a.
b.
Data wants to [be human].
Odo expects Quark to [smuggle contraband onto the station].
Since complementizers cannot be stranded, so the argument goes, the fact that to can be stranded
shows that it is not a complementizer.
This argument suffers from a fatal flaw. The only category in English that freely ellipts
is VP. In the sentences in (11) (and, under some assumptions, (12)), what is ellipted is IP, not
VP. Therefore, there is no reason to expect these ellipses to be grammatical. There is no reason
on the basis of these data to consider to a V or I.
If to is a complementizer, the rule of VP Ellipsis would have to be slightly restated.
5
(14)
In the configuration [C/I/V VP], the VP can be optionally ellipted/omitted/deleted. The
head is stranded.
On the assumption that the categories C, I, and V form a natural class, this may well be a simpler
rule than one that specifies only I and V. We leave open the question of whether this rule needs
to be stated in terms of what can be stranded, which seems dubious in any case. The point is that
even if it does, VP Ellipsis does not provide an argument that to is not a complementizer.
2.2. Bare VP complements
Another argument that to is either V or I is provided by Pullum (1982) and Mittwoch
(1990). They observe that the complement of to is a bare VP with an uninflected verbal head.
Bare uninflected VP complements are also attested in cases such as the following.
(15)
a.
b.
c.
Trelane made Spock [recite a poem].
The wormhole lets people [travel to the Gamma Quadrant].
The Federation helps the Bajorans [rebuild their planet].
(16)
a.
b.
c.
The Romulans might [attack Vulcan].
The Borg will [assimilate everyone].
Uhura can [open hailing frequencies].
In the theory of categories that Pullum assumes, the italicized heads in both (15) and (16) are
verbs. Thus, Pullum concludes that only verbs can take bare VP complements, and to must
therefore be a verb. In the categorization assumed by Mittwoch, the heads in (15) are verbs and
those in (16) are modals. Mittwoch’s conclusion is that to is a modal. We categorize modals as
Infl (Falk 1984b).
The basic generalization is that V and I can take bare VP complements; C, as the other
verbal category, should be expected to have this capability as well. None of the uncontroversial
complementizers can take bare VP complements. If our proposal is correct, to fills this otherwise
open gap. The bare VP complement thus provides no evidence for categorizing to as V (or I),
and can in fact be used as a suggestive argument for categorizing it as C.
2.3. Need for a structural subject position
Koster and May (1982) argue that, since infinitival clauses demonstrably have syntactic
(although unexpressed) subjects, they must have a structural subject position. Under current
versions of theory, such a position would be either [SPEC, IP] or [SPEC, VP]. Plausibly, then,
to is either I or V.
This argument rests on the assumption that subjects must be represented in a structurally
uniform manner. This is not uncontroversial. In theoretical frameworks like LFG and HPSG, for
example, subjects need not be represented structurally at all. Even within the GB framework
other structural positions are available for subjects, such as small-clause subject or [SPEC, VP]
in the complement of to. This is therefore not an empirical argument, but a theoretical claim to
be confirmed empirically.
✦
2.4. Finiteness features
Koster and May (1982) observe that to carries a feature [ finite]. Since the feature
[±finite] is associated with the category I, it follows that to must be a I. As with the previous
argument, this is a theoretical claim rather than an empirical argument.
✧
6
The assumption made by Koster and May is in fact only one of three possible
alternatives. It is equally possible that only the feature [ finite] is associated with the category
I, or that there is no one-to-one correlation between categories and features. Under either of
these alternatives, Koster and May's conclusion would not follow.
Ultimately, the correctness of one of the possible views about the relationship between
the feature [finite] and the category I is an empirical question. The categorization of to is a
prerequisite for answering this question.
★
2.5. Entailment of complement
Mittwoch (1990) argues that the states of affairs expressed by bare infinitive
complements are entailed by the larger sentence while the states of affairs expressed by to
infinitives are not.
(17)
a.
b.
Kirk saw Edith cross the street. [entails that Edith crossed the street]
Kirk expected Edith to cross the street. [does not entail that Edith crossed the
street]
She attributes this nonentailment to to, and suggests that this is a modal-like property.
It is unclear that a syntactic property is involved here at all. But even if the lack of
entailment is to be traced back to syntax, it is not clear that it is the presence of a modal that is
involved. The same lack of entailment is possible with a complement that has no modal in it.
(18)
Kirk believes that Edith crossed the street. [does not entail that Edith crossed the street]
It is possible that the presence of a complementizer is crucial here. This does not seem to be an
argument for categorizing to as an Infl.
2.6. Adverbials and floated quantifiers
A syntactic distributional argument offered by Mittwoch (1990) is based on the
positioning of words like all and really.
(19)
a.
b.
The Klingons can all board the ship.
Kirk might really enjoy being captain.
(20)
a.
b.
The Klingons want to all board the ship.
Kirk seems to really enjoy being captain.
However, the position of all and really is based on the following VP, not a preceding modal.
(21)
a.
b.
Worf made the Klingons all board the ship.
Good luck helped Kirk really enjoy being captain.
This has no bearing on the categorization of to.
2.7. Other arguments
Pullum (1982) provides several additional arguments ostensibly supporting the verb
analysis of to. We will not deal with them extensively, but we will show briefly that a couple
7
of them do not distinguish between V and C, although they may provide arguments against the
Infl analysis.
One of the points Pullum makes, based on Zwicky and Levin (1980), is that to is resistant
to stress.6
(22)
a.
b.
*I want the Enterprise TO boldly go where no one has gone before.
*The Yorktown didn’t go there, but I expect the Enterprise TO.
Pullum claims that in this way it is like infinitive forms of auxiliaries. This property of to
contrasts with the behavior of Infls, which can be freely stressed.
(23)
a.
b.
The Enterprise WILL boldly go where no one has gone before.
The Yorktown didn’t go there, but the Enterprise WILL.
However, complementizers are also resistant to stress.
(24) *I think THAT the Enterprise should boldly go where no one has gone before.
Neither to nor complementizers are completely unstressable; some complementizers (like
conditional if) can be stressed,7 and Mittwoch shows that to can receive contrastive stress.
(25)
a.
b.
I’ll beam aboard the Enterprise IF O’Brien operates the transporter.
I wouldn’t be upset not to be chosen for the away team, but on the other hand,
I wouldn’t be upset TO be chosen either.
But they are both more resistant to stress than Infl. This resistance to stress is actually expected
of functional elements, including complementizers and auxiliaries, and it is the free stressability
of Infl that is surprising. In any case, while the stress facts can serve as an argument against the
Infl analysis of to, they cannot be used to argue against the analysis of to as complementizer.
Another observation made by Pullum is that not can precede to.
(26)
Q wants the Enterprise not to boldly go where no one has gone before.
Since not follows Infls, this is another argument against the Infl analysis.
(27)
a.
b.
*The Enterprise not will boldly go where no one has gone before.
The Enterprise will not boldly go where no one has gone before.
However, as noted by Hornstein (1977), not can also precede constituents of categories other
than VP.
6
Zwicky and Levin limit the unstressability of to to situations when it is stranded by VP Ellipsis, and
Pullum follows them on this. It seems to me, though, that it is (at least relatively) resistant to stress in other
contexts as well. By limiting discussion to VP Ellipsis constructions, Pullum automatically excludes consideration
of complementizers. The particular resistance to stress before a VP Ellipsis site is interesting and in need of
expanation, as Zwicky and Levin point out, but it is not clear to me that it is germane.
7
I thank Tsipi Kuper-Blau (personal communication) for this observation.
8
(28)
a.
b.
c.
[DP Not one starship] was able to defeat the Enterprise.
Spock seemed [AP not very logical] during pon-farr.
[CP Not that it matters], but Picard was assimilated by the Borg.
It is thus not clear that the positioning of not before to can be used to argue that to is a verb.
2.8. Conclusion
The arguments that have been made for categorizing to as V or I are not very compelling.
We conclude, then, that since the infinitival clause is a CP, to is a C.
It is striking that the argument for the complementizer status of to is indirect. In this
respect, we echo the observation made by Pullum (1982: 205) that
[n]one of the above arguments is compelling enough to sway a hardened conservative… An
empiricist who believes that syntactic analyses should spring out of the raw facts will not be
convinced, for the arguments I have put forward rest on subtle simplifications of the grammar at a
fairly abstract level.
However, as Pullum thought was true for his analysis of to as a verb, the analysis of to as a
complementizer argued for here results in a simpler grammar than either of the alternatives.
If the analysis proposed here is correct, it has interesting implications for syntactic
theory. It means, for example, that contrary to the assumption in the transformational tradition,
the presence of an understood subject cannot be used as evidence for an IP node. It also means
that the assumption that the feature [±finite] is associated invariably with the category I is also
incorrect. Conversely, since infinitives are CPs, it also means that Brame's (1976) proposal that
infinitives constitute a distinct category to account for subcategorization in obligatory equi verbs
is also incorrect.
3. On For Infinitivals
One interesting problem created by our analysis of to infinitives is the structure of the
bracketed element in (23).
(29)
[For Picard to beam down to the planet] would be dangerous.
It is usually assumed that the structure of the for clause is something like:
CP
(30)
C
for
IP/S
DP
I /VP
✩
Picard
to beam down to the planet
9
Under this analysis, for is a complementizer and the DP after for is analyzed as occupying the
normal structural position for subjects ([SPEC, IP] or NP-under-S). Our analysis of the to-VP
construction as a CP renders such an analysis impossible.
While the analysis of for as complementizer is traditional in generative work, it is not the
only possibility. Jespersen (1940: 308) characterizes for as a “mere grammatical sign of the
subject (S) of the infinitival nexus,” i.e. as a preposition Case-marking the subject of the
infinitive. Such an analysis draws a connection between English and languages that mark the
subjects oif infinitives dative. We can adopt this idea in phrase structure terms as follows.8
CP
(31)
PP/KP
C
P/K
DP
C
for
Picard
to
✪
VP
beam down to the planet
This analysis has the advantage of capturing directly the mutual dependence between for and the
subject, which is captured indirectly in other analyses. There also seems to be some evidence for
treating for and the DP as a constituent: for + DP can be coordinated, while complementizer–
subject strings are not generally coordinatable.
(32)
a.
b.
Trelane wishes [[for Kirk and for Spock] to beam down].
*Trelane expects [[that Kirk and that Spock] will beam down].
For this to work, subjects would have to be allowed to occupy [SPEC, CP] position. A structure
in which for and the subject form a constituent was first proposed (as a derived structure) by
Emonds (1976). Emonds observes that adverbials cannot be positioned between for and the
nominal, even though with other complementizers adverbials can occur in that position. He also
observes that such an analysis accounts for the accusative marking on the nominal.
A less radical alternative analysis would be as follows:
CP
(33)
C
DP
for
Picard
CP
to beam down to the planet
This analysis is basically a “Raising-to-Object” analysis of for clauses. The nominal following
for is the (nonthematic) object of for, and is understood as the subject of the complement CP.
8
I would like to thank Joan Bresnan (personal communication) for suggesting an analysis along these
lines to me.
10
Within a theoretical framework that recognizes Raising-to-Object, such as LFG or HPSG, an
analysis along these lines should be unproblematic.9
One way or another, for infinitives do not seem to present an insurmountable problem
for the analysis proposed here. What is required is a rethinking of the structure of for clauses,
and probably a rejection of the analysis of for as complementizer.
4. Conclusions
We have argued in this paper that to infinitivals are CPs, and that to is a complementizer.
This conclusion has important consequences for general syntactic theory. A theory of syntax has
to be based on careful empirical study, not just abstract theoretical principles. For example, the
theoretical desire to make finite and nonfinite clauses have the same constituent structure, which
motivates much work on infinitives in the GB and Minimalist traditions, has been shown to be
empirically inadequate.
The categorization of to as a complementizer is not in principle in conflict with any
current theory of syntax, although it is in conflict with specific analyses that have been proposed
in the contexts of various theories. For this reason, this paper has tried, as much as possible, to
be theory-neutral in its argumentation.
An interesting next step would be to examine infinitives in other languages in the light
of the arguments presented in this paper. It may turn out that infinitives are I-less CPs generally,
or it may turn out that the peculiarities of English, such as the existence of the infinitive particle
to, produce a structure that is significantly different from those found in other languages. Only
careful empirical argumentation will be able to resolve such questions.
References
Brame, Michael K. (1976) Conjectures and Refutations in Syntax and Semantics. New York:
Elsevier North-Holland.
Bresnan, Joan (in preparation) Lexical-Functional Syntax. ms., Stanford University.
Chomsky, Noam (1981) Lectures on Government and Binding. Dordrecht: Foris.
Chomsky, Noam (1986) Barriers. Cambridge, Mass.: MIT Press.
Emonds, Joseph E. (1976) A Transformational Approach to English Syntax: Root, Structure
Preserving, and Local Transformations. New York: Academic Press.
Falk, Yehuda N. (1984a) Grammatical Configurations and Grammatical Relations. Doctoral
dissertation, MIT
Falk, Yehuda N. (1984b) “The English Auxiliary System: A Lexical-Functional Analysis.”
Language 60: 483–509.
Hornstein, Norbert (1977) “S and X Convention.” Linguistic Analysis 3: 137–176.
Jespersen (1940) A Modern English Grammar on Historical Principles, Part V: Syntax, Fourth
Volume. London: George Allen and Unwin.
Koster, Jan, and Robert May (1982) “On the Constituency of Infinitives.” Language 58:
116–143.
✫
9
Within a framework that does not recognize Raising-to-Object, this analysis could perhaps be formalized
with a small-clause approach. The subject would be adjoined to the CP headed by to. I will not work out the
details here.
11
Mittwoch, Anita (1990) “On the Distribution of Bare Infinitive Complements in English.”
Journal of Linguistics 26: 103–131.
Pollard, Carl, and Ivan A. Sag (1994) Head-driven Phrase Structure Grammar. Stanford: CSLI
Publications.
Postal, Paul M., and Geoffrey K. Pullum (1978) “Traces and the Description of English
Complementizer Contraction.” Linguistic Inquiry 9: 1–29.
Pullum, Geoffrey K. (1982) “Syncategorematicity and English Infinitival to.” Glossa 16:
181–215.
Zwicky, Arnold M., and Nancy S. Levin (1980) “You Don’t Have Tó.” Linguistic Inquiry 11:
631–636.