STUDIES

in Philosophical Linguistics

SERIES ONE
AUXILIARIES AS MAIN VERBS

o. This paper has two major parts. In paragraph 1, I

present ten arguments that indicate that auxiliaries and verbs
are really both members of the same lexical category, verb. In
paragraph 2, I present two arguments which indicate that they
must be main verbs. That is, I will argue that in deep structure,
each of the five underlined words in (1 )

(1 ) Boris must have been being examined by the captain,

must be the main verb of some underlying S: all verbs (this term
covers what have traditionally been called adjectives, auxiliaries,
the copula, and true verbs) are directly dominated by VP in deep
structure.

1.
1.1, Under the present analysis of English, there are at
least three rules (Subject-Verb Inversion. Neg Placement and
VP Deletion) which mention the term shown in (2) as part of their
SD:

(2 ) Tns

For instance, VP Deletion, which produces sentences like (3),

built a house did

s building a house is
(3 )Mike ay build a house and Tom nay ^tc
must have been building a house aust have beer
has a house las
l!s sick Loes
1 * 1 1 Ls Jl
presumably mentions the term shown in (2). In the present theory,

this Is a double accident: firstly, (2 ) Is a very strange term

(It Is not even a constituent, and there Is no explanation for
why such a term should appear in widely separated rules, which
appear to have nothing to do with one another); and secondly, the
theory makes the claim that the Items mentioned in (2 ) have no
similarity which would predispose them to function together -
(2 ) is as natural a term, in this theory, as the one shown in ( M

I suggest that (2) should be replaced in all rules

which mention it by the entirely natural constituent shown in (5 ).

(5)

..
1 2 The so-called copula, be, should really be analyzed as
a true verb and should be assigned the feature
of evidence for this is the fact that in languages whose basic
order is SVO, the order in copular sentences is S be 0; in SOV
languages, the order is SO be. Another is the fact that be undergoes
Gapping just as real verbs do:

(6 ) S. I ate fish, and Bill (ate) steak.

* b. I am American, and Bill (is) Canadian

Furthermore, there is a rule, Q Hopping, which moves

quantifiers like all, both, each, etc. over be.
 fall ' (an \
] both Iboth |
(7 ) They < each are Hkndsorae -*• They are J each L handsome.
] etc.
r to-J

Q Hopping also moves quantifiers over auxiliaries

(under various conditions) - they have all gone, they must both
have left, etc. Since be is a verb, and jince this rule groups be
and the auxiliaries, it provides sonle evidence that the latter are
also verbs. I propose that both be treated as having the feature

analysis shown in (5 )*

1.3. For many speakers, there is a difference in acceptability

between (8a) and (8b),

(8 ) a. He^ forced me to be examined by Dr. Hito.

b»* He^ forced me to be examined by him^.

although other speakers do not make this distinction. This appears

to be due to a transformational constraint, of a very mysterious sort,
which may be able toextend indefinitely far down into embedded
sentences, if (9 ) is ungrammatical.

(9 ) ?? I want Mary to convince Tom to get Peter to

try to force Jack to be examined by me.

The constraint seems to be that no agent in a for-to or

Poss-Ing complement can be identical to the subject of a higher
sentence, as long asonly for-to orPoss-Ing sentences Intervene

between this agent and the subject.

Given this very rough statement of the constraint, the

ungrammaticality of (10 b) provides some evidence that the may of

permission is a true verb, which has a first person subject, when
used as a performative, as in (ICa), or an unspecified NP
subject, as In Bars may stay open until 5 In England, and which
undergoes the rule of Flip.

(10) a. You may gladly be examined by Dr. Hlto.

b. *You may gladly be examined by me.

(Notice also the expected switch of first to second person In

questions: *May Tom be examined by you?)
If the sentences In (10) are derived from roughly the
structures that underlie (1 1 ), the ungrammatlcallty of (1 0 b) can
be explained by the constraint mentioned above.

(11) a. I gladly allow you to be examined by Dr. Hlto.

b. * 1 gladly allow you to be examined by me.

But the rule of Flip applies only to verbs - hence the

fact that It must apply to may argues that this modal Is also a verb,

1.^. The verb force requires a £-statlve^ verb as the main

verb of Its complement sentence, while the verb seem, with a for-to
complement, requires a C+stative} main verb.

(12) a. I forced Dr. Mensch to f*know J answer.

/*learn)
b. Dr. Mensch seems to \ know j the answer.
f
If the full range of auxiliaries which appear In the
complements of these verbs Is studied, the Interesting complementa­
rity shown In (1 3 ) comes to light;
) f learn the answer ^ /♦learn the answer
♦know the answer I I know the answer
♦be sleeping I I be sleeping
forced him to \ ♦have slept I _jje stems to \ have slept
I♦be allowed to leave( \ be allowed to leave
\*be bald J I be bald

If the auxiliaries be(lng) and have(en). the passive

auxiliary be(en) and the copula be are analyzed as being true
verbs, with the features £+Aux} and £+statlve} , then the facts In
(1 3 ) can all be subsumed under the generalizations expressed In (14)

(14) force (also coax, avoid, etc.) requires a f-statlveQ

verb In the next lowest sentence
seem (also be reported, turn out. happen, etc.), when used
with a for-to complement, require a [>statlve3 verb In
the next sentence down.

The problems raised by such sentences as (15) require

special treatment.

(15) a. Max forced me to be photographed.

b. I forced John to be reading when Judy left.
c. Dr. Mensch seems to learn the answers.

In order to state (14) In a maximally simple way. It Is

necessary to analyze the auxiliaries of (1 3 ) not only as verbs, but
as main verbs. This argument, therefore, belongs in part in £ 2 below.

1.5. The word so is a pro-S (cf. (16)).

(16) a. I hope that we will win in Vietnam, but no sane

man hopes so.
b. It may seem that we will win, to our glorious
president, but it doesn't seem so to me.
 Thus the fact that so can replace what follows
auxiliaries indicates that this constituent is a sentence.

(17)
likes ice cream
may be here
They said that Tom 4 is working hard ^and so he
had left
might have been singing)

r does
I may
< is
' /has^l
\hadj
might have been
jmight have
jy^might '}

The fact that s£ can replace either singing, or

been singing, or have been singing, in the last line of (1 7 )

constitutes a particularly telling criticism of the analysis of
auxiliaries in Chomsky's Aspects of the Theory of Syntax, for in
that analysis, neither of these last two phrases is even a
*
constituent, let alone being a sentence. In my analysis, however,
the derived structure of the embedded sentence in this last line
would be approximately that shown in (18).
singing
and since exactly the same structure would appear in the second
conjunct of the last line in (!?)„ the rule which substitutes so
for an identical sentence would be able to replace S^, Sg, or S-^,
thus yielding the three possible output sentences, after the so
has been permuted to the front of the second conjunct. The reason
that S^f S2 . and do not prune will be discussed in Chapter 3 of
my forthcoming monograph Variables in Syntax.

1.6. The NP nodes in (18) are motivated by the appearance of

which and that in (1 9 ).

(19) flikes ice pream, / wh^c!?uh! *oe? \

land that he doesf

may be here. / wh^c^ hf *ay \

^and that he mayjr
They said
that Tom ^ is working hard, J whlch he is ^
land / he is that")
^and \that he is)
had he { S j
(„and that he </^asl
\hadj
might have been sleeping,
„ land that he might (have(been
J
for in other sentences, which and that replace NP. Thus, this
argument and the one in paragraph 1.5 both show auxiliaries to mani~
fest synthetic phenomena characteristic of true verbs with sentential
complements.
1.7 The rule of S Deletion converts (20a) into (20b ).
(20) a. I know that our cause is lost, but no one else
knows that our cause is lost.
b. I know that our cause is lost, but no one else knows it<
If the deep structure of Max was chortling is that shown in (21),
where the verb be requires the participle lng to be added to the
 next verb down,

(2 1 )

VP

chortle

then the same rule of S Deletion can be used to produce (22).

(22) Max was chortling when I got up yesterday morning

and he was still at it when I went to bed that night.

1.8. The rule of S Deletion appears in the grammar of German,

and if the deep structure of Ottokar muss singen. 'Ottokar must
sing1, is that shown in (2 3 ).
 (23)

Ottokar V1 NP

+V / \
+Modal
I / \
es NP VP
I
muss V
Ottokar

+V
-Modal

sing

sentence (2*4-) can be produced with already available rules.

du musst es auch
(2*0 Ottokar muss singen, und
das musst du auch
'Ottokar must sing, and you must (it) too.'

There are sentences which are identical to (2k) except

that forms of thq verb müssen 'must' have been replaced by forms of
können 'can', wollen 'want to', sollen 'ought to', or dürfen 'be
allowed to', which indicates that these elements which have been
analyzed as modal auxiliaries in the past, should really be analyzed
as being trfie verbs, differing from 'real' verbs like versuchen 'try'
*
only in haying the feature [+Modall , where the latter has the
feature ^-ModalJ .
The fact that (25) is ungrammatical
 Idu musst es auch
(25) »Ottokar muss Krebs haben, und l
Idas musst du auch
(Ottokar must have cancer and you must (it) too.)
is related to the fact that (2 6 ) is grammatical:
(26) Es muss sein, dass Ottokar Krebs hat.
•It must be that Ottokar has cancer.'
I would claim that both come from roughly the same deep structure,
a structure containing an intransitive verb müssen, such as the one
shown in (2 7 ), which underlies the first clause of (2 5 ).

(27)

hat Krebs

1.9» There are such transitive-intransitive verb pairs in

English, too; the syntax of these will be investigated in detail
in David M. Perlmutter's forthcoming dissertation (M.I.T. 19 6 7 ).
In particular, the verb may appears both transitively and intransi­
tively in deep structures. Sentence (28) is ambiguous: one structure
which underlies it is shown in (2 9 ), the other in (3 0 ).
 (28) Windows may be broken by rioters.

(29

break windows

(30)

Sentence (28) can be disambiguated by the adverbs gladly

and possiblyi (3 1 a) can only derive from a structure like (2 9 ), and

(3 1 b) from one like (3 0 ).

(3 1 ) a. Windows may gladly be broken by rioters.
b. Windows may possibly be broken by rioters.
 The fact that (31b), but not (31a.), has a version which
begins with the expletive there

(32) a, *There may gladly be windows broken by rioters,

b. There may possibly be windows broken by rioters.

seems to parallel the fact that sentences with intransitive verbs

like happen, turn out. etc. can also start with this expletive.

(33) a. There happened to'be a commissar present.

b. There turned out to be a catfish in the drain.

This could be accounted for by making the rule which

inserts there cyclic. It would apply on S2 in (30) after the
passive had been formed. Then, when It^ Replacement applies, on S^,
the derived subject there would end up as the superficial subject
of the intransitive verb may. It is not clear to me at present how
(32a) is to be excluded, for if the rules of Passive and
There Insertion apply to in (2 9 ), and this derived subject there
is then made a constituent of the higher sentence by It Replacement
on S^, the rule of Flip should be able to apply, resulting in the
ungrammatical (32a). I include a discussion of this case here not
because I have an analysis which can account for the ungrammaticality
of this sentence, but only because I feel that it is significant
that the sense of (28) for which There Insertion can apply is the one
related to sentence (3 *0 » which contains may as an intransitive.

(3^) It may be that windows will be broken.

 But many problems remain, and at present I cannot
account for them satisfactorily with my analysis.

1.10. The sentences in (35) are felt to be variants of one

another.

(35) a» Ella doesn't need to go.

b. Ella need not go.

Yet in the Aspects analysis, these sentences come from totally

different deep structures: (3 6 ) for (3 5 a), and (3 7 ) for (3 5 b)

(36)

it NP PdP
 Ella Aux VP

s°

In this analysis, then, the only similarity between the

sentences in (3 5 ) is phonological - it is a phonetic accident that

there is a modal which is a homonym of a true verb, an accident
which requires no more comment than the fact that the modal may
is also homonymous with the name of a month.
In my analysis, however, both sentences in (35)
would be derived from (3 8 ).

(38)

V
 I am not sure of the deep structure source of

negation, and so I have temporized by analyzing It as a feature

of the highest £+Auxl • carries the tense. To this structure,

the optional rule In (3 9 ) can apply.
OPT r* -|
(dare 1
(39) n need j — ► l+Modall / negative contexts

This rule Is of course only a mnemonic for the

correct rule - In particular, I have not tried to reconstruct the
phrase "negative contexts". But (39) should apply to produce the
sentences In (*1-0), but not those In (*4-1).

(*1-0) a. Willy need not leave.

b. How long need he fear your wrath?
c. Need he be so cruel?
d. He was crueler than he need havebeen.
e. I don't thinkhe need have eaten so much.
f. I dare eat only what my doctor allows me to eat.

(41) a. *Who need telephone her?

b. *If she need leave, she's to call me.
c.??It's too bad that he need be so heartless.
d. * 1 don't think he dare eat so much.

Clearly much detail must be filled In to convert (39)

into an adequate rule, but the outlines, at least, seem to be
reasonably clear.
If (39) does not apply to (38), the Infinitival
complementizer, to, will not be deleted; and, after C+Neg} has been

spelled out as not, to the right of do, the verb on which It is

marked, the rule which substitutes the next lowest verb for do, when

this verb immediately follows do, will not apply. But if (39) does

apply, a rule substituting modals for the tense verb will apply,

and the not will again get spelled out to the right of the verb
on which it is marked, which is now the new modal need. The rules
then, must be ordered as shown in (42),

(42 ) /dare] __. j+Aux 1

I needj L+ModalJ

Modal Substitution for Tense Verb

Negative Spelling Out
Subject Verb Inversion
Verb Substitution for Tense Verb
Q Hopping

One final rule must apply to delete the infinitival

complementizer to in (35b), It is possible that this rule can be

made to be the same as the independently necessary rule which
deletes to after the active forms of make. see, and hear
(contrast (43a) with (43b)),
(made j
(4 3 ) a. The FBI i saw y Peter curse,
] heart J

^made
Peter was < seen to curse by the FBI,
heard
Although many details remain to be filled in, the
broad outlines of this analysis should be clear. In the Aspects
analysis, however, not only would no rules directly relate (35®)
and (35b) (it would only be possible to exhibit their relatedness
by claiming that the lexical entry for the verb need and the one
the modal need are somehow similar), but it would be difficult to

exclude sentences like (44),

\
dare
(44) *James r read this book.
yjnee&j
Presumably some new kind of deep structure constraint would
have to be devised to exclude (44). This is possible only if the
term “negative context" turns out to be definable in terms of deep
structure properties - not derived structure ones. Whether this is
possible is not known at present.
2.
2.1. The first piece of evidence for the stronger hypothesis,
that not only are auxiliaries [+vj , but they are each the head V
of some VP, comes from German. In my forthcoming paper "Gapping
and the Order of Constituents", I argue that the direction in which
the rule of Gapping operates depends on the phrase-structure
configurations which are the input to the rule. In languages like
English, verbs gap forwards, because the V is on the left branch
of a VP. Thus (45a) is grammatical, but not (45b).

(45) a. I ordered peaches, and Tom cream, and Bill Sterno.

b,*I peaches, and Tom cream, and Bill ordered Sterno.

In languages like Japanese, where the underlying order

is SOV,the reverse is true: Gapping operates backwards because
the Vis on a right branch of VP. Thus what corresponds to (46b)
is grammatical, but (46a) is not.

(46) a, *1 peaches ordered, and Tom cream, and Bill Sterno.

b. I peaches, and Tom cream, and Bill Sterno ordered.

German exhibits SVO word order in main clauses, and

only sentences like (45a) can be the result of Gapping.In dependent
clauses, however, both sentences like (46b), in whichGapping
operates backward, and sentences like (46a), in which
It operates forward, are possible. Since Gapping can operate
forward even in dependent clauses, I argue that basic order in
German is not SOV, as has previously been held, but SVO; and that
there is a rule. Verb Final, which obligatorily moves verbs in
dependent clauses to the end of their VP. This rule is stated in

(^7)
(47) Verb Final

[V x]
VP OBLIG

=>
0 2 + 1
Condition: this rule works only in dependent clauses.
In languages with "free" word order, such as Latin,
Russian, etc., where both SOV andSVO are possible output strings,
sentences corresponding to (45a), (46a), and (46b) exist. No
language exhibits sentences like (45b). These facts can be accounted
for if it is assumed that Gappingis an "anywhere" rule, and can
operate at any stage of a derivation, and that the underlying o
of all free word-order languages is SVO. Thus the effective order
of the rules of Gapping and Scrambling. the rule which permutes
major elements within clauses, would be that shown in (48).

(^8 ) a. Gapping (OPT)

b. Scrambling (OPT)
c. Gapping (OPT)

Sentences like (45a) would be derived by merely applying

Gapping to the underlying SVO order. Those like (46b)
would be derived by first applying Scrambling, which could
give conjuncts of the form SOV, and then gapping backward. And
sentences like (^6 a) would be derived by first gapping forward,
yielding a string like (^5a),and then applying Scrambling to the
first conjunct, converting SVO to SOV order. It is impossible to
derive (^5 b) with these rules.
It is easy to see that a parallel explanation ofthe
German facts can be given if the rules in German are orderedas
in (^9) and the underlying order of constituents in SVO.

(4-9) a. Gapping (O P T )
b. Verb Final (CBLIG)
c. Gapping (OPT)

In order to derive sentences like (50)» which contain

many "Hilfsverben" (=auxlliary verbs), the structure of (50) at
the time Verb Final applies must be roughly that shown in (51).

(50) Gwendolyn muss von Kasimir gesehen worden sein

Gwendolyn must by Casimir seen been be (=have)
'Gwendolyn must have been seen by Casimir.'
L+Modall
VP,
I
muss

f
r+v
L+InfiI VP

I
sein

1+V I I
|+Past I
I Particlplel

I
worden
 Verb Final will move Vg to the end of VP2 , V3 to the
end of VP^, and V^ to the end of VP^, thus reversing the order of
the bottom three verbs.(Note that the order before the application
of this rule corresponds exactly to the order of the corresponding
verbs in English.) If (5 1 ) were itself in a dependent clause,
Verb Final would also have to move Vj to the end of VP1# as has
happened in (5 2 ).

(52) weil Gwendolyn von Kasimir gesehen worden sein muss,

•because Gwendolyn must have been seen by Casimir•

It is absolutely necessary to postulate four verb

phrases in (51). because there are other main clause-dependent
blause pairs which show that the rule of Verb Final must produce
order alternations with the main verb sehen ‘see' (ef.(5 3 )), with
the passive "auxiliary" werden »become' (cf. (5 4 )), with the past
tense verb sein 'be' (cf. (5 5 )), and with the "modal auxiliary"
müssen 'must'. (Compare (5 0 ) and (5 2 )).

(53) a. Kasimir sieht Gwendolyn.

'Casimir sees Gwendolyn.'
b. weil Kasimir Gwendolyn sieht
because Casimir Gwendolyn sees
'because Casimir sees Gwendolyn'

(54) a. Gwendolyn wurde von Kasimir gesehen

Gwendolyn became by Casimir seen
'Gwendolyn was seen by Casimir'
b,. well Gwendolyn von Kasimir gesehen wurde
because Gwendolyn by Casimir seen became
'because Gwendolyn was seen by Casimir'
(55) a • Gwendolyn 1st von Kasimir gesehen worden*
Gwendolyn Is by Caslmlr seen become.
'Gwendolyn has been seen by Caslmlr'.
b. well Gwendolyn von Kasim’r gesehen worden 1st.
because Gwendolyn by Caslmlr s^en become Is.
'because Gwendolyn has been seen by Caslmlr'.
These facts provide evidence of the strongest kind that
there Is no category difference between German auxiliaries and
other verbs, and that each auxiliary must be Immediately dominated
by VP. In passing It should perhaps be noted that the copula sein
•be' behaves Just like all other verbs with respect to the rule of
Verb Final; (56a) must be converted by this rule into (5 6 b).
Im Knäst
(56) a. Hans 1st • ekelhaft
„ein Scheisskop^
Ink '1
f In the clink
'Hans Is < repulsive

}
V,a bounder : ;
im Knast
b. well Hans ekelhaft 1 st
ein Schelsskopf,
(in the clink'
'because Hans ls^repulsive'
(a bounder' „
. .
2 2 The second argument that auxiliaries are main verbs comes

from Greenberg (cf. his "Some unlversals of grammar" in his

Unlversals of Language. M.I.T. Press), who notes that in languages
whose basic order Is SOV, if there is an auxiliary, it follows the
verb, while In languages whose basic order is SVO, if there is an
auxiliary, It precedes the verb. (Guarani, a language of South
America, provides the only counterexample to this latter claim -
according to Greenberg, It exhibits the
order SV Aux 0 .) These facts, which Greenberg merely notes, can
be explained under the hypothesis that auxiliaries are main verbs:
stating that the auxiliary was precedes writing a letter in (5 7 ).

(57) Bill was writing a letter.

is equivalent to stating that verbs precede their objects in English,

I propose to derive (57) from (58),

(58)

Bill

write a letter

so the fact that was precedes its object, writing a letter, is the
same as the fact that writing precedes its object, a letter, and
these facts need only be stated once.

Similarly, in Japanese, where the basic order is SOV, we

find the auxiliary lta ’was 1 following the verb kaite 'writing' in
(59). which corresponds to (5 7 ).
(59) Biru ga tegami o kalte lta.
Bill (particle) letter (particle) writing was
'Bill was writing a letter.*
This sentence would derive from (60):

(60) S

NP-"

I
Blru

tegami ksk

Once again, the fact that the phrase tegami o kalte

precedes lta is the same as the fact that tegami o precedes kalte:
all objects precede their verbs In Japanese.
The fact that auxiliaries which derive from intransitive
verbs, like müssen 'must' In (2 7 ), behave like the transitive aux­
iliaries Just discussed is accounted for by the fact that the rule
of Xt Replacement, which converts (27) into (25), is formulated in
such a way that it always adjoins the infinitive phrase to the same
side of the VP as the side on which the other objects appear. In
other words, It_ Replacement, Just like Gapping, operates in different
directions in different languages, the direction in particular
languages being dependent on the input phrase-structure configuration.
 3a This concludes this brief survey of the reasons I now

know for claiming that auxiliaries belong to the same major

category as verbs, and are Introduced into deep structures the

same way other verbs are. There are many problems which 1 have
only discussed cursorily in this preliminary version, and some 1
have not touched at all, such as, for instance, the problem of
insuring that the verbs in (1 ) are embedded into one another in
the right order, Still, it seems to me that I have given enough
evidence in support of my original hypothesis to justify the hope
that the problems which are still outstanding will be able to be
accounted for without it being necessary to abandon the basic
points of the analysis I have proposed,

John Robert Ross

Massachusetts Institute of Technology