Springer is collaborating with JSTOR to digitize, preserve and extend access to Foundations of Language. http://www.jstor.org Quantifiers in English Author(s): Ray S. Jackendoff Source: Foundations of Language, Vol. 4, No. 4 (Nov., 1968), pp. 422-442 Published by: Springer Stable URL: http://www.jstor.org/stable/25000347 Accessed: 15-07-2015 09:54 UTC Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/ info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTC All use subject to JSTOR Terms and Conditions
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Springer is collaborating with JSTOR to digitize, preserve and extend access to Foundations of Language.
http://www.jstor.org
Quantifiers in English Author(s): Ray S. Jackendoff Source: Foundations of Language, Vol. 4, No. 4 (Nov., 1968), pp. 422-442Published by: SpringerStable URL: http://www.jstor.org/stable/25000347Accessed: 15-07-2015 09:54 UTC
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/ info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
This paper will examine the relations between several noun phrase con structions in English. Evidence will be presented that they are all instances of the same deep structure configuration, and transformations will be pro posed to produce the observed surface structures.
I. THE THREE CONSTRUCTIONS
The first construction involves what will be called Group I words, of which the following are examples:
Group I: a group, a herd, a wagonload, a score, a pound, a gallon, a
number, a lot, a pair
All of these words can occur independently as noun phrases, and they can also have complements with an of prepositional phrase:
(1) a group (2) a group of men
(3) a group of the men
The words in Group I are fairly clearly nouns: they can pluralize, and (in
most cases) be counted and modified by adjectives.
(4) groups of men
(5) two gallons of water
(6) an incredibly large herd of cattle
The second construction involves the so-called quantifiers.
Group II: some, each, few, which, all, both
Traditional grammarians have said that these words have two uses. The first
is pronominal: the words can occur alone or with a following of prepositional
phrase.
(7) some (of the men)
* This work was supported in part by U.S. Air Force Contract 19(628)-5524. It has
appeared in an earlier form in Stephen R. Anderson, Ray S. Jackendoff, and Samuel Jay Keyser, Studies in Transformational Grammar and Related Topics (AFCRL Report No.
68-0032, March 1968). I wish to thank Arthur Schwartz for many useful discussions of
genitive constructions and quantifiers which helped lead to this paper.
422
Foundations of Language 4 (1968) 422-442. All rights reserved.
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
The other use is called adjectival, because the words directly precede a noun:
(8) some men
However, the paradigm is not that simple. The prepositional phrase following one of these words cannot be indefinite:
(9) *some of men
Furthermore, the adjectival use is not truly adjectival, since no article may precede such a noun phrase:
(10) *the some men
The paradigm may be summed up as (11).
(11) some ((of the) men)
In addition, we have the forms in (12) for all and both, which do not occur with the other Group II words:
(12) *some both the men
all
The third construction shares characteristics of Groups I and II:
Group III: a few , many, one, three
Group III has the same paradigm of following nouns and prepositional phrases as Group II:
(13) three ((of the) men) (14) *three of men
(15) *three the men
However, Group III words do permit articles in front of them, like Group I:
(16) the three men
II. BEHAVIOR COMMON TO THE THREE GROUPS
Let us examine the restrictions on complements in all three groups. First of
1 Few (Group II) and a few (Group III) are treated as separate lexical items here. This is justified by the fact that they behave differently syntactically. Few is inherently negative, and conditions the change of some to any: Few of the men had seen any rats. A few is not negative, and the change from some to any does not take place: *A few of the men had seen any rats.
423
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
In Groups I and III, which permit the use of articles, there is a very interesting restriction holding between the article of the main noun phrase and the article of the complement: both cannot be definite at the same time
unless there is a relative clause present. Thus we have
(33) a group of the men
(34) three of the men
as well as
(35) the group of men
(36) the three men
but not
(37) *the group of the men
(38) *the three of the men
However, the addition of a relative clause makes (37) and (38) acceptable:
(39) the group of the men that had already left
(40) the three of the men that you met yesterday
This relative clause is associated not with the complement noun phrase, but with the Group I or Group III word itself. This is shown syntactically by two facts. First, the complement noun phrase can be a pronoun, which does not permit attachment of relative clauses:
(40.1) the three of them that you met yesterday (40.2) *them that you met yesterday
Second, when the of-phrase preposes from (41) and (42), the relative clause does not move with it to form (43) and (44), but rather remains behind as
in (45) and (46).
(41) The group of the men that had already left were disgusted with the proceedings.
(42) The three of the men that you met yesterday have not left yet. (43) *Of the men that have already left, the group were disgusted
with the proceedings. (44) *Of the men you met yesterday, the three have not left yet. (45) Of the men, the group that had already left were disgusted
with the proceedings. (46) Of the men, the three you met yesterday have not left yet.
A relative clause is however possible on the complement noun phrase; it will
prepose along with the complement.
425
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
(47) The three of the men I spoke of that you met yesterday have not left yet.
(48) Of the men I spoke of, the three you met yesterday have not left yet.
To further point up the similarity between Groups I and III, we note the existence of doublets, words which alternate between Groups I and III. First there are the two words couple and dozen, which, depending on dialect, can belong to either group:
(49) a couple of (the) theorems (I)/a couple (of the) theorems (III) (50) a dozen of (the) eggs3 (I)/a dozen (of the) eggs (III)
Even more curious, hundred and thousand are in Group III in their singular but Group I in their plural:
,hundred hundreds t (51) a thund d (of the) cats (III) / thundds of (the) cats (I) thousand thousands
These variations would seem to indicate that the difference between Groups I and III is very trivial, perhaps the difference of one feature. Furthermore, the similar restrictions on complements and articles suggest that all three
groups are examples of the same deep structure complement construction.
III. THE STRUCTURE OF GROUP I. II, AND III NOUN PHRASES
Let us first look at the structure of Group I NPs. As we noted earlier, the
members of Group I are clearly nouns, with the possibility of a fully filled-out
determiner. Thus about the only possible structure is something like (52).
(52) a group of men: NP
Det N PP
Art group of NP
a Det N
Art men
indef
The members of Group III can also take articles and relative clauses,
although there does not seem to be much possibility of their taking adjectives
3 The phrase a dozen of eggs can be heard in many small Boston grocery stores.
426
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
(except perhaps without the complement, as in the doughty three). However, this is true with some members of Group I as well, particularly those that are semantically numeral-like, such as score and the doublets mentioned above. Thus it is plausible to consider Group III members also as nouns, giving structures like (53).
(53) three of the men: NP
Det N PP
Art three of NP
indef Det N
Art men
the
Since Group II members do not take articles, it is not clear whether they are nouns or determiners or neither or both. However, the nature of the restrictions indicates that they bear the same grammatical relation to their
complements as do Groups I and III. For the present, we will put off giving any further detail for the structure of Group II than (54).
(54) some of the men: NP
DetN PP some
of NP
Det N
l I Art men
I the
Of course, there is one problem with the structures (53) and (54). If the noun phrase in the complement is indefinite instead of definite, we will get the incorrect
(55) *three of men
(56) *some of men
We will remedy this by adding to the grammar a transformation which
427
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
obligatorily deletes of with indefinite complements of Groups II and III
giving us the correct forms
(57) three men
(58) some men
The following examples show that such an of-deletion is indeed the correct transformation to handle this situation.4
(58.1) Guess what we don't have any of: insect repellent. (58.2) We don't have any insect repellent. (58.3) *Guess what we don't have any: insect repellent. (58.4) *We don't have any of insect repellent.
The postulation of an of-deletion transformation permits us to explain these
examples easily: in (58.1), the preposing of what prevents the of from
dropping as it does in (58.2). Before stating the transformation, let us make note of a couple more
things it should do. First, there are the forms (12) with all and both, which could be derived by optionally extending the transformation to definite
complements of these two words. Second, there is the problem of number
agreement. Certain members of Groups II and III, in particular which (in the singular sense), each, and one, take singular verbs when they are subjects of sentences; also, singular pronouns and reflexives are used for core
ferentiality with these NPs.
/was true to himself? (59) which of the mentor himself? (different meanings)
were true to themselves?
! was true to himself (60) Each of the men (60) Each of the men
*were true to themselves
was true to himself (61) One of the men
true to msel
*were true to themselves
Furthermore, these same words take singular rather than plural nouns after them in the 'adjectival' use:
(62) which each man
one
This can be accounted for by assigning an underlying number to the Group II and III words, which determines the number of following verbs and pronouns.
4 I am grateful to Heidi Roehrig for losing the insect repellent, making the discovery of these examples possible.
428
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
This same number will be used in the of-dropping transformation to induce number agreement on the reduced complement.
We will use a feature [+ Q] to determine the applicability of the transfor mation. Members of Groups II and III will be marked [+ Q] and members of Group I will be [-Q]. Thus the variation among the doublets will be marked by the presence or absence of this single feature. The transformation can then be stated as given in (63).
(63) of-dropping:
NP1 NP1
Det N PP et NNP +Q +Q [P of NP2 et p[ Det N2
/ \ [ocplJ
Det2 N2
Obligatory if Det2 is C-definite3
Optional if Det2 is ?+ definite] and Q is all or both
The transformation as stated here makes the minimal possible reduction in structure. It will produce the correct surface strings, since the indefinite article for mass and plural count noun phrases is null and will not show up in the phonological output. However, it is not clear whether the derived structure is correct. There may also be deletion of the nodes NP2 and Det2
(except with all and both, which would then have to be handled separately), but I have no evidence for or against such further reduction.
By using the underlying structures (53) and (54) and the transformation (63) for quantifiers, then, we can explain three things automatically: first, the
similarity of restrictions on quantifier complements to restrictions of Group I
complements; second, the paradigms (11) and (12) for quantifier comple ments; and third, number agreement with verbs and pronouns, as in (59)-(61). Let us examine some other proposals for the underlying structure of quan tifiers and see how well they do.
IV. OTHER PROPOSALS FOR QUANTIFIERS
The earliest transformational approach to quantifiers was that developed by Chomsky and elaborated by Barbara Hall5. This approach puts the quan
5 Barbara Hall: 1962, 'All About Predeterminers', M.I.T., unpublished.
429
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
tifiers in the determiner, in a constituent called prearticle or predeterminer. Thus the structures look as shown in (64).
(64) some of the men: NP
Det
Pre Art Art N
some of the men
The problems with this approach are many. First, it requires that of the men not be a constituent, thus immediately losing the generalization that it
is a plain ordinary prepositional phrase. Second, it cannot bring out the
similarity to Group I unless Group I is also put in the Prearticle. But Group I
cannot be put in the prearticle, since the members of Group I are clearly nouns. If we admit the possibility of noun phrases in the prearticle, we might as well call the quantifiers nouns too, and then the structure (64) can be
replaced by (65). But this structure still does not treat of the men as a prepo sitional phrase.
(65) some of the men: NP
Det N
Pre Art Art men
NP of the
some
Presumably, this approach could account for the paradigms (11) and (12) by a variant of the transformation (63). However, it cannot easily account
for the facts of number agreement. To account for (59)-(61), both the number
agreement rule for verbs and the rules that determine the number of a
pronoun will have to refer to the prearticle when there is one, and to the
number of the head noun when there is no prearticle. In contrast, in the
analysis proposed in this paper, the number agreement rules can refer to the
number of the head noun in cases of quantification as well as in ordinary sentences.
Another popular analysis for quantifiers is the one proposed by Janet
Dean6. In this analysis, the structure as given in (66) is proposed. Then, by
6 Janet Dean: 1966, 'Determiners and Relative Clauses', M.I.T., unpublished, ditto. I believe Postal has suggested a similar analysis.
430
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
a transformation, N1 is deleted, provided it is identical with N2; otherwise, the derivation blocks. The advantage of this analysis over Hall's is that it
does recognize that of the men is a prepositional phrase and that both the
quantifier and the complement can take relative clauses. It also makes number agreement easy to explain.
However, this analysis has two faults. First, it makes explanation of the
paradigms (11) and (12) a great deal less natural. If Det2 is generated as
indefinite, we get the ungrammatical
(67) *some men of men,
which must be disposed of with a second transformation to become some
men; this transformation is also obligatory and requires identity of N1 and N2. Thus we need two transformations to account for a paradigm consisting of two examples, and neither of the surface structures are identical with
either of the deep structures. In the analysis proposed in the present paper, only the of-dropping transformation is needed, and the definite case remains as generated.
The second problem with this analysis is the example (68), pointed out to me by Chomsky:
(68) John and Bill, each of whom....
The only possible deep structure for this in Dean's analysis is the highly implausible (69).
(69) John and Bill, each John and Bill of wh-John and Bill....
The motivation behind both Dean's and Hall's approach is based on the
need to meet selectional restrictions. It is obvious that the acceptability of
(72)-(73), which contain quantifiers, should be based on the same factor as the acceptability of (70)-(71), which do not.
( the theorem (70) I proved the boy
(71) I socked the theo the boy
431
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
(72) I proved ( some of the theorems *some of the boys
(73) Isocked *some of the theorems ( some of the boys
In the theory of selection given in Chomsky's Aspects of the Theory of Syntax, it is assumed that the process of selection which determines the acceptability of (70)-(73) is based on the feature of abstractness on the head noun of the
object of the sentence. Given this assumption, it follows that theorems and
boys must be the head nouns of the objects in the deep structure, since some
clearly cannot be marked for abstractness (or the multitude of more finely divided semantic features). Therefore both Hall and Dean adopt structures which are unnatural for other reasons, just in order to get theorems and boys into the head position in the noun phrase. Unfortunately, the same problem arises with Group I constructions:
(74) I proved i a bunch of theorems (74) proved *a bunch of boys
*a bunch of theorems (75) I socked
' a bunch of boys
The prearticle analysis would of course handle this by putting the Group I noun in the prearticle. I have no idea how Dean's solution could take care of this.
Recent work7 has shown, however, that selectional restrictions are not
applied to only the head noun of a noun phrase. Instead, the selection is
performed on the basis of the semantic reading of the entire noun phrase. Since presumably the readings of some of the boys and a bunch of boys contain the fact that we are talking about boys and not theorems, this
mechanism of selection eliminates the need to somehow work boys and theorems into head position in the objects of (72)-(75). The selection appa ratus will have available to it the information that it needs to make the
proper distinctions in these sentences, even if the analysis is as in (52) and
(54), in which group and some are the syntactic heads of their noun phrases. Therefore, if this view of selection is correct, it destroys the chief reason for
preferring Hall's or Dean's analysis to the one presented here, namely the need to meet selection restrictions on head nouns. The greater syntactic
7 Ray Jackendoff: 1966, 'A Note on Selectional Restrictions', M.I.T., unpublished; James
McCawley: 1966, 'Concerning the Base Component of a Transformational Grammar', University of Chicago, unpublished; James McCawley: 1967, 'How to Find Semantic Universals if There Are Any', in Bach and Harms (eds.) Proceedings of the Texas Confer ence on Language Universals.
432
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
generality of the present analysis then makes it a clear choice over these other two approaches.
A third proposal for quantifiers, given in various forms by George Lakoff and Guy Carden8, involves putting the quantifier in a predicate of an addi tional underlying sentence. For Lakoff, this additional sentence can be either above or below the quantified noun in the tree; for Carden, the additional sentence must be a higher one. Thus the underlying structure as shown in (76) is typical.
(76) some of the men Left: S
NP VP
NP S (were)some
the men NP VP
the men left
A transformation then deletes a lot of nodes and moves the quantifier into the lower NP.
The motivations for this approach are fairly complex. One of Lakoff's
arguments is based on the existence of archaisms like
(77) Many were the men that perished.
This sentence, of course, can be generated by the mechanisms proposed in this paper as well: since many is an NP, (77) is an instance of the predicate nominal construction, just like the men are doctors. Lakoff's other arguments are based on the interpretation of questions containing quantifiers.
Carden's arguments are based on contrasts such as the following:
These contrasts cannot be accounted for with the conventional machinery of referential indices, as can, for example, John saw John and John saw
himself, since everyone always denotes the same set. Carden argues that the difference is due to the presence of different numbers of higher sentences,
8 George Lakoff: 1965, On the Nature of Syntactic Irregularity, Appendix F, Report NSF-16, Harvard Computation Laboratory; Guy Garden: 1967, English Quantifiers. Harvard MA dissertation.
433
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
so that, for example, (78) and (79) have the underlying forms as given in (82) and (83) respectively.
(82) S
NP VP
NP S every
one/ NP VP
NP S every
one /
one. Loves one.
(83) S
NP VP
NP S every
one/
onei loves one,
Carden neglects to notice that the same contrasts exist for indefinite
plurals under similar circumstances:
(84) Senators from New England admire Senators from New England.
(85) Senators from New England admire themselves.
(86) Men who earn $ 50,000 a year expect men who earn $ 50,000 a
year to be treated with respect. (87) Men who earn $ 50,000 a year expect to be treated with respect.
Notice that in (85), even though the plural themselves is used, each individual is understood to admire only himself. The fact that these phenomena appear in ordinary plurals shows that the problem has nothing to do with quantifiers, but is in fact a more general problem of reference of sets. To claim that
plurals, like quantifiers, are higher sentences merely begs the question, and such a solution borders on the absurd. What is needed rather is a more
comprehensive study of coreferentiality in all kinds of noun phrases that denote more than a single individual.
What is worst, however, about the Lakoff-Carden analysis is that it
434
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
totally ignores the more obvious aspects of the syntax of quantifiers, namely the phenomena investigated here. The result is that all syntactic generality is lost. There is no way to explain the similarity to Group I, unless Group I are verbs too. But this is impossible, since Group I can be pluralized and counted. The fact that quantifiers can occur alone as pronouns is difficult to explain unless we have two dummy NPs in a structure like (76). The facts of number agreement are also very difficult to state: if each is a verb, how can it carry singular number so as to affect the number of verbs and pronouns in other sentences, since only nouns and determiners can be marked lexically for number? Furthermore, if one is a verb, how can we account for its simi
larity in behavior to the well-known pronoun one? All these considerations make the adoption of a Lakoff-Carden analysis look very implausible.
V. THE PROBLEM OF RECURSION
One serious difficulty faced by any analysis of quantifiers is the problem of recursion. There are very limited possibilities open for multiple quantifiers in a single noun phrase; but on the other hand, rather long expressions can be generated, such as (88).
(88) No three of any four given men can together lift this stone.
We have already noted one restriction on all quantifiers, namely that some, all, and few cannot occur in the of-phrase (see (27)-(32)). The distribution of acceptability in recursive quantification does not appear to relate to the structure of the noun phrase, but rather it seems to derive from idiosyncratic characteristics of the specific quantifiers involved; for example, we have (89) but not (90) and (91) but not (92).
(89) any one man which one man
(90) *every one man *which one man
(91) every three men all seven men
(92) *both two men *each two men
*any many men
Likewise, corresponding to (88), we have the unacceptable (93).
(93) *Some three of all four given men can together lift this stone.
There is no apparent structural difference among these contrasting examples, yet the difference in acceptability is perfectly clear.
435
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
The analysis of quantifiers proposed in this paper has no difficulty generating indefinitely deeply nested levels of quantifiers. All that must be done is to further expand the NP in the of-phrase into another NP with a
quantifier. Thus the subject of (88) might have an underlying structure roughly as shown in (94) (it will be simplified somewhat in the next section).
(94) NP1
/DetN\ PP no
^
of NP2
Det N PP
indef of NP3 three
Det N PP any
of NP4
Det N PP
I I / indef of NP5
four
given men
The problem, of course, is preventing incorrect recursions, such as some and all in NP1 and NP3 respectively. The analysis by Chomsky and Hall, the only attempt I know of to describe restriction on quantifier recursion, took care of this by further breaking up PreArt into various categories and
using context-sensitive phrase structure rules. Thus the differences were accounted for structurally, resulting in a very complex and artificial-looking system. However, this analysis was developed before the concepts of the lexicon and lexical features were introduced into the theory of grammar; thus a structural difference was the only way available to make these dis
tinctions. Hopefully, a more insightful account could be made, making use of the more powerful tools provided by lexical features. Validation of my
analysis of quantifiers will depend crucially on being able to use lexical
features to account for the restrictions; otherwise the rules generate far too
much. Of course, the same is true of any theory of quantifiers, such as Dean's
or Carden's, which permits free recursion in the quantifier system.
436
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
So far we have only assigned Group II to some position under the Det or N node, as in (54). As we will see, further attempts to determine the con
stituency of Group II become rather confusing. The characteristic which distinguishes Group II from Group III is the fact
that Group II cannot occur with a preceding article, whereas Group III can. This suggests three possibilities: first, and most trivial, Group II members are nouns which, like proper nouns, cannot take determiners; second, the structure is exactly as in (54), with the Group II word subsuming both the Det and N nodes; third, and most interesting, Group II words are themselves articles. As far as I can see, the first two possibilities are essentially equiva lent, so the problem boils down to whether Group II members are nouns or articles.
Relevant to this question is the problem of the following paradigms, which seem to be very closely related to the Group II paradigm (11):
(95) every man/every one of the men
( man? (96) no mn/none of the men n men
In (97), none is apparently a portmanteau of no one(s). What makes these
paradigms particularly close to (11) is the fact that singular Group II members can optionally appear in this paradigm:
(97) each either neither
man
anyeihr (one) of the men
any which
The plural Group II members all, some, both, and plural any and which cannot take ones followed by a definite complement (although which ones
by itself is all right):
(98) all some
both (*ones) of the men
any which
In an account of Group II, it would clearly be desirable to handle the varying
participation of these quantifiers in the paradigm with one by means of some
simple feature, governing either a semantic selection or 9 syntactic transfor
mation.
437
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
Assume first that all Group II members are nouns, i.e. that (54) is the correct structure for some of the men. This requires that the deep structure of every one of the men be (99), produced from recursion of the base rule that generates the PP.
(99) every one of the men:
NP'
/Det N \ PP
every]
-pL of NP2
Det N PP
I I /\ indef ones of NP3
Det N
l I the men
By of-dropping in NP', we get the correct surface string for (99). This analysis is undesirable for a number of reasons. First, the structure (99)
for every one of the men is slightly horrendous, requiring as it does two levels of embedding. However, it is not only horrendous, but troublesome as well. How can the restriction of every to the paradigm with one be stated? If the
requirement is that ones be in the immediately lower NP, we also get (99) for the structure of every man, with only the definiteness of NP3 changed. But for the simple every man, a doubly embedded structure is unthinkable.
Furthermore, we need a transformation to delete one after of-dropping has
applied twice, in order to turn every one man into every man.
Suppose then that the requirement on every is that it have an indefinite NP below it. Then the structure of every man is (100), and we don't need a
transformation to delete one.
(100) every man: NP
DetN \ PP every
of NP
Det N
I I indef men
Every one of the men is still represented as (99), the justification being that if every requires an indefinite below it, and the class to be quantified is
438
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
definite, one must use the intervening minimally specified ones to meet the
requirement. How would this restriction be extended to the rest of Group II? For the
singular members that can optionally occur with one, we need put no re
quirement at all on their complements: each of the men will have a structure
like (54), and each one of the men like (99). For the plurals, we note simply that ones of the men never occurs in any context, so presumably no special explanation would be needed for (98). Thus the proposal that Group II
members are nouns can be made to work, provided we are willing to put
up with some rather ungainly base structures and some lopsidedness in the
underlying structures of the paradigms. Next suppose instead that Group II members are articles, i.e. that every
one of the men is represented as (101).
(101) every one of the men: NP1
Det N1 PP
Art one of NP2
I I every the men
Presumably the choice of N' is free, so we can generate every man simply as given in (102).
(102) every man: NP
Det N
I I Art man
every
Given this analysis of every, the restrictions on the occurrence of one
show up in an entirely different light than if every is a noun. Before discussing these restrictions, let us propose a transformation, ones-absorption, to oper ate as shown in (103).
(103) Ones-absorption:
NP = NP
Det N X Det N X
II Art
Art one(s) + subst
439
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
The feature [+ subst] placed on the article in many cases dictates a change in its phonological shape. The application of the rule is demonstrated in the
following examples. (104), with an adjective between the article and the noun, is acceptable. (105), with no such adjective, is out. Instead, ones
absorption produces the surface forms (106).
(104) no my9
John's red one(s) a the
(105) *no
*my *John's one(s) ka
*the
(106) none mine
John's one
those/it/they10
The use of ones-absorption gives us an exceedingly simple method to account for the use of one with Group II. This method has two notational
variants. The first one is to say that the phonological shape of + every is
every one, that the shape of + ubt
is either each or each one, and that + substj
the shape of + subst] is some. The other notational variant is to say that + substj
every cannot undergo ones-absorption, that each optionally does, becoming each if it undergoes the transformation and leaving each one if it does not, and that all always undergoes the transformation. As far as I can see, there
are no different claims made by these two variants. Both attribute the vari
ation in the use of one to an option in the operation of a trivial rule. This
is probably preferable to the analysis of Group II as nouns, where the
9 A fuller discussion of how ones-absorption interacts with the possessive system appears in my paper 'Possessives in English', contained in AFCRL Report No. 68-0032 (see note on page 422). 10 The possibility of generating it and they in this manner is reminiscent of Paul Postal's analysis in "On So-Called 'Pronouns' in English", Georgetown Monograph Series in
Language and Linguistics 19, 1966.
440
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
variation is attributed to an optional requirement on the complement, a rather deep semantic fact; the variation does not seem to be of that nature.
By the account of Group II as articles, the similarity of Group II to
Groups I and III is due to the presence of an underlying ones, a Group III
member, in Group II expressions. Group II does not have to be marked [+ Q], and of-dropping need only examine the head noun for this feature. Furthermore, this analysis offers a considerable reduction in the size of
underlying structures, compared to the analysis of Group II as nouns:
compare (99) to (101), (100) to (102), and (94), a more highly embedded
example, to (107).
(107) NP
Det N PP
Art three of NP
no Det N PP
Art four of NP
any given men
This approach to Group II has one difficulty, which is actually a more
general problem and must be dealt with regardless of our analysis of
Group II. The problem is this: what is the underlying form of Group II and the substantive possessives such as mine in the case of mass interpre tation? (106) can indicate mass pronominalization as well as count, yet there is nothing to correspond to (104) with mass noun phrases:
(108) no
the
There is no pronoun in English for indefinite mass nouns, except perhaps stuff. If the use of (106) for count nouns derives from the application of
ones-absorption, presumably its use for mass nouns should derive similarly. But there is no appropriate pronominal form to correspond to ones in the case (108) where ones-absorption does not apply. The only way out, still
preserving this analysis, is to postulate the existence of an indefinite mass
pronoun which has no phonological shape and which must be absorbed by an article. But this kind of totally abstract lexical item should be very costly
441
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions
in terms of an evaluation measure for theories; it should only be used as a last resort in trying to capture generalizations.
Another possible way out of this problem might be to abandon ones absorption and replace it with a lexical rule systematically relating articles and pronouns, much as other lexical rules relate verbs and their nominali zations. Then the use of quantifiers with one and the examples (104) would represent the article form, while the quantifiers without one and the ex amples (106) would represent the pronominal form. Then the non-existence of (108) could be explained by the non-existence of the appropriate mass pronoun. Unfortunately this approach leaves open the question of how to
get rid of (105), the article form with ones. Either a convention of obligatory suppletion or the rule of ones-absorption will have to be added back into
the grammar, so it is not clear what gain in economy has been made by
moving the rule into the lexicon. However, this tactic does seem like a
conceptual gain, even if the present framework of grammar cannot capture its advantage. I suspect that the ultimate solution to the constituency of these formatives lies somewhere between the two approaches suggested here; but I cannot suggest where, as the available theory of grammar provides no
such compromise solution.
VII. SUMMARY
The similarity between the three groups of words presented here leads us to
analyze Group III words as nouns exactly parallel to Group I nouns in
underlying complement structure. The postulation of the of-dropping trans formation (63) accounts for the surface structure paradigm of these words.
Group II words are analyzed as articles combined with the Group III
pronoun ones by ones-absorption or perhaps as nominalized articles. The
validity of these underlying structures rests crucially on two suppositions: first, that selectional restrictions are stated between verbs and readings of
entire noun phrases, not between verbs and head nouns; second, that an
account of quantifier recursion can be given in terms of features on the
quantifiers rather than in terms of structural differences.
Massachusetts Institute of Technology
442
This content downloaded from 14.139.86.99 on Wed, 15 Jul 2015 09:54:01 UTCAll use subject to JSTOR Terms and Conditions