Partitives: Studies on the Syntax and Semantics of Partitive and Related Constructions [Reprint 2011 ed.] 9783110908985, 9783110147940

Citation preview

Partitives

Groningen-Amsterdam Studies in Semantics (GRASS) This series of books on the semantics of natural language contains collections of original research on selected topics as well as monographs in this area. Contributions from linguists, philosophers, logicians, computer-scientists and cognitive psychologists are brought together to promote interdisciplinary and international research. Editors Alice ter Meulen Martin Stokhof Editorial Board Renate Bartsch University of Amsterdam

Johan van Benthem University of Amsterdam Henk Verkuyl University of Utrecht Co Vet University of Groningen

Other books in this series: 1. Alice G.B. ter Meulen (ed.) Studies in Modeltheoretic Semantics 2. Jeroen Groenendijk, Theo M.V. Janssen and Martin Stokhof (eds.) Truth, Interpretation and Information 3. Fred Landman and Frank Veltman (eds.) Varieties of Formal Semantics 4. Johan van Benthem and Alice G.B. ter Meulen (eds.) Generalized Quantifiers in Natural Languages 5. Vincenzo Lo Cascio and Co Vet (eds.) Temporal Structure in Sentence and Discourse 6. Fred Landman Towards a Theory of Information 7. Jeroen Groenendijk, Dick de Jongh, Martin Stokhof (eds.) Foundations of Pragmatics and Lexical Semantics 8. Jeroen Groenendijk, Dick de Jongh, Martin Stokhof (eds.) Studies in Discourse Representations Theory 9. Michael Moortgat Categorial Investigations 10. Irena Bellert Feature System for Quantification Structures in Natural Languages 11. R. Bartsch, J. van Benthem and P. van Emde Boas (eds.) Semantics and Contextual Expression 12. D. Zaefferer (ed.) Semantic Universale and Universal Semantics 13. Renate Bartsch Situations, Tense, and Aspect: Dynamic Discourse Ontology and the Semantic Flexibility of Temporal System in German and English

Jacob Hoeksema (ed.)

Partitives Studies on the Syntax and Semantics of Partitive and Related Constructions

Mouton de Gruyter Berlin · New York 1996

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter & Co., Berlin.

The series Groningen-Amsterdam Studies in Semantics was formerly published by Foris Publications, Holland.

© Printed on acid-free paper which falls within the guidelines of the ANSI to ensure permanence and durability. Library of Congress Cataloging-in-Publication

Data

Partitives : studies on the syntax and semantics of partitive and related constructions / Jacob Hoeksema, editor. p. cm. - (Groningen-Amsterdam studies in semantics (GRASS) ; 14) Includes bibliographical references and indexes. ISBN 3-11-014794-7 1. Grammar, Comparative and general-Partitives. I. Hoeksema, Jacob, 1956. II. Series: GroningenAmsterdam studies in semantics ; 14. P299.P39P37 1996 415-dc20 96-663 CIP

Die Deutsche Bibliothek — Cataloging-in-Publication

Data

Partitives : studies on the syntax and semantics of partitive and related constructions / Jacob Hoeksema (ed.). — Berlin ; New York : Mouton de Gruyter, 1996 (Groningen-Amsterdam studies in semantics ; 14) ISBN 3-11-014794-7 NE: Hoeksema, Jacob; GT

© Copyright 1996 by Walter de Gruyter & Co., D-10785 Berlin. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording or any information storage and retrieval system, without permission in writing from the publisher. Printing: Arthur Collignon GmbH, Berlin. Binding: Dieter Mikolai, Berlin. Printed in Germany.

Table of Contents

Jacob Hoeksema Introduction

1

Barbara Abbott Doing without a Partitive Constraint

25

Jacob Hoeksema Floating Quantifiers, Partitives and Distributivity

57

Jaklin Komfilt Naked Partitive Phrases in Turkish

107

Ann Reed Partitives, existentials, and partitive determiners

143

Anne Vainikka and Joan Maling Is partitive case inherent or structural?

179

Karina Wilkinson Bare plurals, plural pronouns and the partitive constraint

209

Index of names

231

Index of subjects

235

Introduction Jacob Hoeksema

1. Linguistic relevance of the part-of relation The part-of relation plays an important role in cognition and language. For example, in the design of WordNet, the computational lexicon for cognition research developed by G.A. Miller and his associates at Princeton, one of the prime structural relations is meronymy, the relation between words denoting things and words denoting parts of these things (e.g. bird and wing, cf. Miller (1990)). The syntactic expression of the part-of relation takes many forms. Often, the relation is not formally expressed, but understood, as in a large class of English compounds, e.g. (1)

chicken-feet church-door mountain-top bike-steer committee-chair

It is the central importance and general salience of the part-of relation which renders formal expression unnecessary in these cases. Similarly, part-whole relations are crucial in understanding the use of mass nouns (cf. Ter Meulen 1980, Link (1983) and Moltmann (1992)), as well as the theory of aspect (see Dowty (1979), Hoeksema (1985), chapter 6, Bach (1986) or Verkuyl (1993)). The general importance of the relation is also illustrated by the fact that it supports definite descriptions, as in (2). (2)

I still have a jar of gooseberry jam, but I cannot get the lid o f f .

In this sentence, the use of the definite article the is licit because of a salient

2

Jacob Hoeksema

link between jars and their parts, such as lids. Given ajar, 'the lid' is uniquely identifiable as the lid of that jar. In other cases, the part-of relation is involved in certain ambiguities of indefinites, e.g. the difference between stressed and unstressed some (see Milsark (1977), De Hoop (1992) and Diesing (1992) for discussion of weak and strong readings of indefinites and the configurations where they arise): (3) (4)

Some senators were acting strange (There is a set of strange-acting senators) SOME senators were acting strange (Part of the senators is acting strange)

The stressed or partitive reading of indefinites is often formally expressed by a partitive construction or case. In English, the partitive construction takes the form of a frame detj (one) of det2 + common noun. In the following examples, the partitive expressions are indicated by roman type font: (5)

Some of the senators were acting strange None of us left before 11 p.m. One or two of the first 70 members had died by then Many of his friends are still at the funeral Not too many of these problems have been solved Every single one of a number of proposals was rejected

In these examples, the part-of relation is of a special kind, because it involves a group or collection (expressed by a plural noun phrase or a collective noun) and its members. When dett is a mass noun determiner, we may also express part-of relations between an individual and its parts, or a quantity of some substance and its subquantities. Partitive noun phrases of this kind will be called mass partitives here. (6)

Most of the city is off-limits to foreigners. Some of him had stayed behind in his native Rumania. Some of the water was murky. We did not get to see all of her new garden. Half of every donation goes to administrative costs. Rick is not much of a hero.

Introduction

3

2. A brief overview of partitive noun phrases The precise characterization of the partitive construction is one of the main topics of this collection. Some of the questions to be answered are: (a) (b) (c) (d) (e)

What is the internal structure of this construction and how does it fit in a general account of (noun) phrase structure? What are the constraints on detj (the "upstairs" determiner), and det2 (the "downstairs" determiner)? What is the distribution of partitive noun phrases (e.g. can they occur in existential constructions?) What is the quantificational force and what the anaphoric potential of partitives? What are the ways in which the various partitive structures of natural language are related?

In this section, I take a quick and somewhat eclectic look at what has been said about these matters and add a couple of my own observations. 2.1. Syntax and semantics of partitive noun phrases Important earlier studies of the English partitive construction were the Xbar theoretic studies of Jackendoff (1977) and Selkirk (1977). Jackendoff proposed the structure below, where an empty head element PRO serves as the head of the construction.1 N'

(7) Art'

N'

Art'

N'

Art'

Ν

Ν'

Art all

1.

PRO

of

the men

Similar empty-head proposals can be found in HPSG, cf. Nerbonne, lida and Ladusaw (1989, 1990), and within the currently popular DP-analysis of noun phrases, cf. Abney (1987).

4

Jacob Hoeksema

One thing that may appear odd about this structure, apart from the empty head, is the fact that the preposition of does not head a PP here. This is due to Jackendoff s assumption that of is not a regular preposition, but a case marker, inserted at a late stage of the derivation. If this unnecessary assumption is dropped, a more regular X-bar structure emerges, which treats all of the men as structurally parallel to a nonpartitive construction like all friends of the king: (8)

N'"

Art'" I Art" I Art'

I

N" ι N' I N'

I

1 PP

I

I

Art I

Ν

Ρ I

NP C=]

all

PRO

of

the men

The price to be paid for syntactic regularity is the assumption of an empty head. However, in defense of this assumption it must be added that the empty position is filled when the determiner requires the presence of a following nominal. Compare every (single) one of my friends, the only one of them that got away, neither one of the two women. Note that the presence of one versus PRO (or the empty noun) need not be stipulated just for the sake of the partitive construction. Determiners which require one in the partitive, also require it elsewhere, when other determiners may be used intransitively. Compare To each his own with To every *(one) his own. The main point to be made here is of course that the empty noun makes it unnecessary to assume a different syntactic structure of each of the men on the one hand and the synonymous every one of the men on the other. Another structure that has been suggested (cf. Keenan and Stavi (1986)) treats the string all of the in all of the men as a complex determiner, which takes men as its argument. The assumption is that of the men is not a constituent in this partitive. It may be noted that the string all of the can semantically be viewed as a determiner, because it has the logical property of conservativity,

Introduction

5

which Keenan and Stavi (1986) and Barwise and Cooper (1981) established to be the core feature of determiners. For a binary relation R between sets A and Β to be conservative, the following equivalence must hold: (9)

Conservativity R(A,B) = R(A,AnB)

That is to say, R relates A and Β just in case it relates A with the intersection of A and B. To illustrate the constraint, note that the validity of the following biconditional shows that all can be viewed as a conservative relation between two sets A and Β (where A and Β are expressed by the predicates philosophers and mortal, respectively). (10) All philosophers are mortal All philosophers are mortal and philosophers A similarly valid biconditional can be provided for all of the: (11) All of the dancers were blond

All of the dancers were blond dancers

The fact that the string in question may be interpreted as a conservative relation does not mean, however, that it must be a determiner. Conservativity is a necessary but not a sufficient condition on determiners. Note, for example, that some connectives, such as and, are also conservative: (12) happy and rich = happy and (rich and happy) The Keenan and Stavi analysis, then, has some semantic plausibility.2 Syntactically, the idea is that partitives are generated by rules in (13):

2.

An argument for the Keenan-Stavi analysis of partitives which has not, to my knowledge, been put forth so far can be based on the existence of idiomatic strings of the form detl of detr Consider the string much of a(n), as in He's not much of a marksman. In this use, the string has the idiosyncratic property of being a negative polarity item (cf. *He's much of a marksman) that appears to be restricted to predicate nominals. While it seems preferable to treat much of a(n) as a unit, it does not follow, of course, that all strings of the form dett of det2 have to be treated as such.

6

Jacob Hoeksema

(13) a. Det b. NP

Det of Det[+plur,+def] Det CN

As an argument in favor of this analysis, Keenan and Stavi note that it predicts correctly the ungrammaticality of NPs such as *each of John, since rule (13a) requires the presence of a second determiner after partitive of The same argument also excludes strings such as one of us or most of them however. The ungrammaticality of each ofJohn is more simply explained by the fact that John is singular. Mass determiners could be used here, however, as in much of John or most of France. The Keenan-Stavi account makes it hard to relate partitives to split partitives (as in e.g. Of the students, at least ten were drunk) where there clearly is a PP constituent. Moreover, in this type of account, it is a mystery why so many languages have partitives with PP-like substrings, usually with a preposition equivalent to of or from. Finally, the recursive rule (13a) leads to wild overgeneration of complex determiners. Using the rule a number of times, one may concoct such strings as *[[one of the] of my] (brothers). A different type of account was given in Barwise and Cooper (1981), who applied their theory of generalized quantifiers to the analysis of partitives. They attempted to reformulate JackendofFs (1977) Partitive Constraint, which stated (14) Partitive Constraint In an of-N"' construction interpreted as a partitive, the N'" must have a demonstrative or genitive specifier. Barwise and Cooper did away with the ugly disjunction in the Partitive Constraint by proposing that the NP following partitive of must be definite, in the sense of denoting a proper principal filter in every model. That is to say, the denotation of a definite NP must always be the set of all supersets of some subset A of the domain of quantification E. When A is neither the empty set nor Ε itself, the generalized quantifier {XcE | A c X } is a proper principal filter, a collection of sets which is closed under intersection and supersets. Examples of definite NPs, and their semantic definitions in a generalized-quantifier framework are:

Introduction

(15) a. the men b. my students

{XdE | [[man]] c X} if [[man]] * 0 , undefined otherwise {Xc:E | {x| χ G [[student]] & R(I,x)} c X} 3 if {x| Χ G [[student]] & R(I,x)} * 0 , undefined otherwise

c. John and Mary

{Xc;E | {j,m} G X}

7

By making the assignment of denotations a partial function, Barwise and Cooper incorporate a Strawsonian view of definite descriptions, according to which the men is not well-defined (or creates a presupposition-failure) in case there are no men, and similarly for my students in the special circumstance that I do not have any. In this respect, definite descriptions differ from true universal quantifiers, such as all women or every woman: the latter are defined even if there are no women. In this way, Barwise and Cooper are able to characterize with some precision the class of expressions that may combine with partitive of. These include definite descriptions, such as the men, those men, my men, but no universals, cf. *one of every student, *some of all books. The semantic function of partitive of is to take a principal filter and return a set, the generator of the filter. This is done by taking the intersection of the filter: (16)

M(Q)=nQ

The result is a common-noun denotation which can be the argument of a determiner. Partitive of is the inverse of a determiner: Whereas a determiner maps a common noun denotation (a subset of E) onto a generalized quantifier (a collection of subsets of E), of maps a quantifier onto a common noun denotation. And so we find that some of the boys denotes the same set as some boys in case there are (two or more) boys. Syntactically, Barwise and Cooper envisaged the following structure for partitive noun phrases:

3.

That is, the set of students who are related by some contextually specified relation R to me must be a subset of any X in the generalized quantifier family of sets.

8

Jacob Hoeksema

(17)

Ν'"

DET

CN

NP

all

of

my

friends

Barwise and Cooper's account was only partially successful. It provided an explicit semantics for the partitive construction, which in a clear and intuitively appealing way linked up with the distributional peculiarities of the partitive construction. A number of problems with their account were solved when proper attention was paid to the semantics of plurality (see the discussion of the difference between one of the two and *one of both in Ladusaw (1982) and Hoeksema (1984)). The syntactic part of the BarwiseCooper account is not overly attractive. It fails to predict the ungrammaticality of partitive expressions such as *every of the students, *no of the women, *the of my friends, and at the same it does not predict the acceptability of none of her friends, every one of these dissertations. The reason for this failure is easy to pinpoint. By treating partitive of as the inverse of a determiner, the Barwise and Cooper account predicts that any determiner can function as the upstairs determiner of a partitive, and that expressions which do not combine with common nouns, such as none (cf. *none students) cannot. As the examples show, these predictions are incorrect, as must be the assumption that partitive of maps NPs into common nouns. Semantically, the Barwise and Cooper proposal relied too much on the correctness of the definiteness constraint on the embedded NP. Just as with that other famous definiteness effect, also treated in Barwise and Cooper (1981) from the perspective of generalized quantifier theory, the definiteness effect in existential sentences, there turned out to be several types of exceptions, which together suggest something is wrong with the account given. Some of these were pointed out in Stockwell, Schachter and Partee

Introduction

9

(1973), Ladusaw (1982), Hoeksema (1984), de Jong andVerkuyl (1985) and Roberts (1987), while others are discussed in the contributions to this volume by Abbott, Reed and Wilkinson. One problem concerns the acceptability of universals after partitive of. While universally quantified NPs are usually excluded from that position, they do not always yield ungrammatical partitives, cf.: (18) the best of all possible worlds or its Dutch translation (19) de beste van alle mogelijke werelden In such cases, there is a dependency between the upstairs and the downstairs determiner: the presence of all downstairs is compatible only with an upstairs determiner that is superlative (as in 18-19) or a fractional expression (such as 24%, three quarters or half).* In a large electronic corpus, partitive strings containing the substring of all were typically like the examples in (20). (20) a. USA Today recently stated that such allegations now come up in 25% of all divorces, and the Denver Post stated the number was 30%. b. In the Bundesliga about 70 percent of all penalty-kicks would be repeated by the referee due to premature movement of the keeper. c. Half of all inmates are drug addicts. It is not entirely clear what superlatives and fraction indicators have in common (to the exclusion of all other determiners), but it should be noted they also feature in another set of exceptions to the constraint involving partitives with bare plurals after of cf. (21) a. the most eloquent of men 5 4.

One might perhaps add a third category here, partitives with wh-pronouns, as the following example from my corpus suggests: (i)

5.

Who of all people would Jennifer believe but Vem.

A peculiar property of this kind of partitive appears to be that it does not permit numerals, cf. the regular partitive in (ia) with the bare plural construction in (ib).

10

Jacob Hoeksema

b. the best of friends c. the most tenacious of enemies and (22) a. 90% of gay service men6 b. 74% of whites and more than half of Asians c. half or more of high school students and 90% of college

students

It might be noted here that the examples in (21) are somewhat stilted and that the Dutch counterparts to (21) and (22) are not grammatical: (23) a. *de meest welsprekende van mannen b. *de beste van vrienden c. *de volhardendste van vijanden (24) a. *90% van homosexuele Soldaten b. *74% van blanken en meer dan de helft van Aziaten c. *de helft of meer van scholieren en 90% van Studenten

(i)

a. The best seven/seven best of the contenders will proceed to the next round, b. They were the best of friends /*the best seven/seven best of friends

Also, the use of most is more restricted in bare plural partitive: (ii) a. most of us / most of the men / most of all b. *most of friends 6.

I do not have a ready explanation for these observations. This type of partitive appears to be typical of journalistic prose. Some real-world examples (taken from an electronic corpus) are given below. (i)

a. His studies show that more the 80% of school children don't need this type of teaching and should be taught a fact based drug prevention program. b. Approximately 10% of pregnant females have high blood pressure. c. About half of cases begin before age 4, and it is frequently not recognized before the child enters school. d. She loves him like no other man, and their marriage was very special, but statistics say half of marriages end in divorce (..)

Introduction

11

It was suggested by an anonymous reviewer that if superlatives and fraction indicators themselves induce a partition of the domain, perhaps of in these examples is not partitive of but simply the semantically vacuous dummy element often postulated in nominalizations such as the destruction of the city. If that were to be the case, we would not expect the partitive constraint to hold. I do not think this suggestion is on the right track, however. Note that it does not explain why in Dutch the partitive constraint remains operative, even though Dutch van acts in all others ways just like its English counterpart. In addition, if the partitioning effect of the determiner were at stake, one might also expect the acceptability of of + bare plural after, say, partitive readings of indefinites, e.g. *SOME (but not all) of men or its Dutch translation *sommige van mannen. Therefore I prefer to think of the superlative and fractional cases as still unexplained problems for the Partitive Constraint. Other types of examples that are not easily reconciled with the Partitive Constraint include mass partitives {more than half of a pizza, 30% of a nuclear site in Nevada, a quarter of an hour) and partitives with specific indefinites (e.g. one of several proposals emanating from the Massachusetts Institute of Technology). On the other side, it is not easily explained within a theory such as that of Barwise and Cooper (1981), why conjunctions of singular NPs may not occur after partitive of. (25) a. *one of Jack and Jill (cf. one of the two children) b. *each (one) of Dr. Jekyll and Mr. Hyde cf. each one of his two personalities) c. *one of my proposal, your proposal and her proposal For an account of this observation, see Reed (1988), who makes a distinction, at the level of discourse representation, between conjoined phrases and definite plurals. Another type of explanation could be based on the lack of definite determiners as main operators in conjunctions (as suggested both by the original formulation of the Partitive Constraint in Jackendoff (1977) and by Keenan and Stavi (1986)). Note however, that these alternative accounts fail to predict that mass partitives are acceptable with conjoined singulars, and that count partitives are acceptable with conjoined plurals: (26) a. Only about half of Jack and Jill was visible for the sniper. b. None of the plumbers and the carpenters were fired.

12

Jacob Hoeksema

2.2. The distribution of partitives 2.2.1. Regular partitives A good theory of partitives should not only explain the various restrictions that hold within the confines of the partitive construction itself, but also shed light on the distribution of partitives within the larger context of the sentence. The distribution of partitives has been discussed, among other things, in connection with the definiteness effects found in existential sentences.7 Partitives often combine an indefinite upstairs determiner with a definite downstairs determiner. This gives them a somewhat dual character in terms of definiteness and the question that arises is whether the indefiniteness of the one or the definiteness of the other determiner will be crucial in either allowing or barring partitive subjects of existential sentences. Keenan (1987: 296) suggests that partitives are not grammatical in existential sentences:8 This analysis [= the one advanced in Barwise and Cooper (1981), JH] then would predict that sentences such as There are at least two of the ten students in the garden would be natural. My best judgment here, admittedly somewhat shaky, is that such sentences are not natural (..). However, if further and more systematic judgments are elicited and accord such sentences a natural status, I would abandon this analysis in favor of the one suggested by Barwise and Cooper. Hoeksema (1989), on the other hand, maintains that there are two types of there-sentences to be distinguished here. Purely existential sentences, that is to say, sentences which make claims about existence only, do not permit partitives, whereas presentational fAere-sentences may contain them. The following examples illustrate this claim: 7.

See also De Hoop (1992) and Diesing (1992) for a discussion of (in)definiteness effects in scrambling. The so-called scrambling order of objects in Dutch and German is the one where they precede sentential adverbs.

8.

Diesing (1992: 72) does not hesitate to star existential sentences with partitives in them. Diesing's assessment of the facts might reflect a conservative dialect of English, but fails to reveal the important difference in acceptability between purely existential and presentational sentences pointed out in the main text, and ignores the fact that actual usage attestations of partitives in existential sentences of the presentational kind can easily be found.

Introduction

13

(27) a. #There is/exists one of the two boys, (purely existential) b. There were several of us in the room, (presentational) The presentational sentences are characterized by an XP (sometimes called the coda) following the subject. The basic idea is that in the purely existential sentence (27a), the existence of two boys is already given ("presupposed" if you will), and hence the existence of one of them does not have to be asserted. In the presentational sentence (27b) on the other hand, the assertion is not superfluous: Not existence is asserted, but presence in the room. The following two example sentences illustrate the difference between the two types of there-sentences for Dutch: (28) a. #Er is een van de honden. there is one of the dogs b. Er blafte een van de honden. there barked one of the dogs A corpus study of partitives in existential sentences reveals that the actual situation is somewhat more complicated than the brief discussion in Hoeksema (1989) suggests. Purely existential sentences without a coda are in fact not at all uncommon in English, as the following sample of sentences from my corpus may illustrate: (29) a. There were three of us - Chris, Martin, and moi. b. There are two of you and but one of me! c. There were plenty of them. These sentences are all cardinality statements. They state the size of a certain set. Determiners such as some do not appear to be equally felicitous in existential sentences: (30) ??There are some of us. It is interesting to note that sentences such as the ones in (29) are not acceptable in Dutch: (30) a. ??Er waren drie van ons. there were three of us

14

Jacob Hoeksema

b. ??Er waren twee van jullie. there were two of you Instead of (30), one uses the idiomatic construction in (31):9 (31) a. We waren met z'n drieen. we were with their threes b. Jullie waren met z'n tweeen. you were with their twos These observations strongly suggest that there are three separate uses of ί/iere-sentences in English: (1) existence statements, (2) presentational statements and (3) cardinality statements. Partitives are barred only from existence statements. In other types of environment, partitives pattern with definite expressions.10 For example, in certain types of measure-adverbials, defmites and partitives are excluded alike: (32) a. many years ago/*the years ago/*those years ago/*several of the years ago b. weigh a few pounds/*a few of the pounds/*those pounds c. It took several years/many decades/*those years/*many of the years Similarly, partitives may not serve as inalienably-possessed objects of the verb have, just like defmites: 9.

Related to this observation is the following. In English, but not in Dutch, one may have cardinal partitives after the: (i)

a. The three of

us had

a

good time,

b. *De drie van ons hadden een goede tijd. (ii)

a. We divided

the work between the three of us.

b. *We verdeelden het werk over

de drie van ons.

Again, a different idiom must be used: (i)

a. Wij drieen hadden een goede tijd. we

threes had

a

good time

b. We verdeelden het werk over ons drieen. we 10.

divided

the work over us threes

Cf. Szabolcsi (1986) and De Jong (1987) for an overview and discussion of various contexts where definiteness effects hold.

Introduction

15

(33) a. I have seven cousins/*the cousins/*seven of the cousins b. Fred has many warts/*the warts/*many of the wartsu We may conclude then, that English partitives pattern sometimes with definites, and sometimes with indefinites, just as one might expect given their dual nature.

2.2.2. Bare partitives Other interesting questions of distribution come up when we extend the purview of our investigation to what one might call bare partitives', noun phrases in the partitive case, or preceded by a partitive preposition. Bare partitives do not have the dualism of the upstairs and downstairs determiners, and as a consequence, they may have rather different distributional properties. In English, bare partitives show up in a rather rare verb alternation12, illustrated by the following sentences: (34) a. Again Tarzan came down into the village and renewed his supply of arrows and ate of the offering of food which the blacks had made to appease his wrath. (From: E. Rice Burroughs, Tarzan of the Apes)

11.

An apparent exception to these definiteness effects is posed by sentences such as (i) (i) Who has the most warts? (ii) Who weighs the most kilograms? These examples involve so-called comparative superlatives (Szabolcsi (1985)) and are semantically indefinite. Note, for instance, that (i) can be paraphrased as (iii): (iii) Who has more warts than anyone else? A different type of exception is to be found in statements such as (iv) and (v) below: (iv) He's got the money, she's got the looks. (v) She has neither the brains nor the ambition to become another Madonna.

12.

Although the looks in (iv) and neither the brains nor the ambition in (v) are clearly inalienably-possessed objects, they are acceptable in these examples. This alternation is not listed in Levin's (1993) meticulous overview of English verb classes and verb alternations, presumably due to its rather marginal status in English.

16

Jacob Hoeksema

b. In the breast of his blouse he carried some coarse dark bread; he ate of this between whiles, and sat munching and drinking near Madame Defarge's counter. (From: Ch. Dickens, A Tale of Two Cities) Here, the regular object NP is replaced by an of??, to indicate that the object does not wholly but only partly undergo the action of the verb. The number of verbs which allow this alternation is rather limited, it appears. For instance, while Tarzan may eat of the offering, he cannot be said to 'read of the newspaper' if all he did was look at part of it. Perhaps only the most frequent verbs of bodily ingestion undergo the alternation. In Dutch, the same alternation can be found: (35) De discipelen aten van het brood en de vissen The disciples ate of the bread and the fish In addition, Dutch has an interesting construction involving bare partitives of the form van die Ν "of those N", reminiscent of the English indefinite use of this/these, as in I have this brother who doesn't like cream cheese and She's got these really large eyes (cf. De Hoop, Vanden Wyngaert and Zwart (1990) for discussion of this construction). Other restrictions on the distribution of bare partitives are found in French. Kayne (1975: 27ff.) notes that when they lack an article bare partitives are sensitive to the presence of negation. Thus there is a difference in acceptability between the elements of the following pairs of negative and affirmative counterparts in (36) and (37). (36) a. Elle n'a pas mange de carottes. she has not eaten of carrots "She didn't eat any carrots" b. *Elle a mange de carottes. she has eaten of carrots (37) a. II n'a pas pu gagner d'argent. he has not been-able win of money "He was unable to earn any money" b. *// a pu gagner d'argent he has been-able win of money

Introduction

17

Besides negation, certain other adverbs also permit the presence of bare partitives of this type13, but the above-mentioned sensitivity to negation is interesting to keep in mind, as other languages may exhibit similar restrictions on the use of partitives (the genitive of negation in languages such as Russian14, Polish and Lithuanian comes to mind, as well as the role of negation in licensing partitive case in Finnish and certain partitives in Basque (cf. Horn 1978)). Other types of bare partitives in French, of the form des Ν or du/de la Ν have a wider distribution, and function as indefinites. More discussion of bare partitives can be found in Kornfilt's and Vainikka and Maling's contributions to this volume.

3. Overview of this volume The papers in this volume can be divided into a few natural classes. There is a set of papers which deal with the Partitive Constraint. The papers by Abbott, Reed and Wilkinson all fall within this group. Hoeksema's paper is concerned with the relationship between partitives and floating quantifiers and deals with a close relative of the Partitive Constraint which was proposed for quantifier float. The two papers by Vainikka and Maling and by Kornfilt are concerned with partitive case (with special reference to Finnish and Turkish, respectively), and have a more syntactic orientation than the other papers. I will give a brief summary of each paper. Abbott sets out to show that the Partitive Constraint is really a figment of the linguist's imagination. She presents evidence showing that not only the constraint itself, but also various weakened versions of it, cannot be maintained in light of a fuller set of data. The oddness of many partitives, Abbott argues, is due to pragmatic effects, rather than a semantic constraint against the use of indefinites in partitive constructions. Given a proper context, most of the "ungrammatical" examples cited in the literature become

13. 14.

For instance, trop "too, too much",peu "little", assez "enough, sufficiently", tellement "so". For discussion of the Russian genitive of negation, see Babby (1980) and Pesetsky (1982).

18

Jacob Hoeksema

acceptable. The only truly unacceptable cases are partitives with embedded plurals, such as *some of women (with the still unexplained exception noted above of superlative and fractional partitives, e.g. the most unlikely of coincidences). A semantic treatment is sketched which makes use of Kamp's (1981) Discourse Representation Theory. The contribution by Reed similarly addresses a host of problems that beset the Partitive Constraint, and relates the distribution of definite and indefinite noun phrases in existential and partitive constructions to their contribution to, and demands on, the discourse setting. Reed concludes that the distribution of noun phrases in existentials and partitives supports a three-way classification into those which evoke discourse entities, those which access them, and those which evoke subgroups of discourse entities (partitives belong to this last kind). (An interesting puzzle for this proposal that has not received any attention in the literature concerns partitives such as all/some/most of the time, where the definite is neither "old" or "familiar" information, nor uniquely described. They appear to be semi-idiomatic.) Reed's paper also contains a discussion of the constraints on upstairs and downstairs determiners. Both Reed's and Abbott's paper are interesting not only for what they have to say about partitives, but also their discussion of the problem of definites in existential sentences. Wilkinson's paper is primarily concerned with the question of why pronouns embedded in a partitive construction may have bare plural antecedents, although bare plurals themselves may not themselves appear inside a partitive. This paper is of interest both for its discussion of pronominal relations and the semantics of bare plurals. Two distinct types of anaphora are argued to be involved: Etype pronouns in the sense of Evans (1977) and common-noun anaphora (cf. for instance Van Eijck (1983)). The paper by Hoeksema on floating quantifiers relates to the study of partitives in a number of ways. First of all, floating quantifiers are often partitive phrases, or have been argued in the literature to derive from partitive-like constructions. Second, floating quantifiers have been argued to obey a definiteness condition similar to the Partitive Constraint (Dowty and Brodie (1984)). Hoeksema shows that this definiteness condition, just like the Partitive Constraint, is spurious. He also argues at length and on the basis of data from a number of languages, that syntactically, floating quantifiers cannot be directly related to underlying partitives by means of a movement transformation, as suggested by Sportiche (1988). Instead, the

Introduction

19

adverbial theory of floating quantifiers, promoted among others by Dowty and Brodie, is argued to be correct, and the semantic properties of floating quantifiers are studied in some detail. Kornfilt's paper argues that the Turkish so-called 'Ablative Partitive' functions not as an oblique, but as a structural case. The Turkish equivalent to of the cake in John had been eating of the cake behaves in a number of respects like a regular direct object. For instance, ablative partitive object are treated the same as accusative objects in the causative construction. A comparison of bare and regular (or "headed", as Kornfilt calls them) partitives is made and the syntactic structure of the latter is determined on the basis of a number of tests for constituency. The properties of the bare partitives are then captured in an analysis embedded within the Government-Binding framework, interesting in particular for students of the relationship between abstract and morphological case. The framework used is that of classical GB, in which the notion 'government' plays a crucial role. It would be interesting to see if and how this analysis can be translated into the currentlypopular minimalist program, in which agreement between specifiers and heads is the crucial mechanism. A similar emphasis on the relation between overt and abstract case can be found in Vainikka and Maling's study of the Finnish partitive. These authors argue that Belletti's (1988) account of partitive case as an abstract inherent case associated with indefiniteness leads to problems for the analysis of the Finnish partitive. Just like Kornfilt did for the Turkish partitive ablative, they argue against Belletti that partitive case is structural, rather than inherent. In this respect, they follow Lasnik's (1992) amendation of Belletti's theory. This paper is also of interest for its discussion of current treatments of indefiniteness in general in a Government-Binding type theory of case. Whereas the traditional view would have it that case and definiteness distinctions are orthogonal, these treatments postulate a direct link between the two. There is considerable evidence for such a direct link, not just in Finnish, but also in Russian, where the genitive of negation is associated with indefiniteness (and has the properties of a structural case, according to Pesetsky's (1982) analysis), and Turkish (cf. e.g. Εης (1991)). There is still plenty of work to do in this area, however, especially as regards the integration of syntactic phenomena such as case marking and word order on the one hand, and the pragmatics and semantics of indefiniteness on the other hand.

20

Jacob Hoeksema

4. Conclusion The part-of relation of central importance in human cognition. The study of partitives pursues the ways in which this relation is expressed in the noun phrase system of natural languages and the ways in which it relates to issues of definiteness and case. Two types of partitives can be distinguished: full, or headed, partitives and determinerless, or bare, partitives. These types do not only differ in their internal structure, but also in their distributional properties. Many questions touched upon in this collection could stand further scrutiny. Much needs to be done to gain a better crosslinguistic perspective on partitives. To mention one case, the status of the sentences in (21) and (22) above is still largely mysterious. However, a number of conclusions emerge from the present collection. First of all, the Partitive Constraint (as well as a similar constraint on the hosts of floating quantifiers) is beset with many problems and best thought of in pragmatic terms, rather than as a syntactic or a semantic condition (as conceived of earlier). In this way, it follows the fate of the definiteness effect in existential sentences (cf. e.g. Abbott (1992), (1993) and Ward and Birner (1993) for recent discussion). Second, the distribution of bare partitives in some languages appears to require that partitive case is treated as a structural case (and partitive arguments therefore as directly related to direct objects or subjects). However, we have also seen that languages may have several types of bare partitives (e.g. Dutch and French). A fuller account of bare partitives should shed light on the variation to be found in this area and the ways in which bare partitives are related to headed partitives. More work needs to be done as well on questions such as the relationship of partitive case to genitive and ablative case, and the typology of languages that mark partitives. Acknowledgements I wish to thank the following people for their assistence in the completion of this volume: Emmy Jacobs, Helen de Hoop, Peter Lasersohn, Alice ter Meulen, Ton van der Wouden and Frans Zwarts and the anonymous reviewer.

Introduction

21

References Abney, Steven P. 1987

The English Noun Phrase in its Sentential Aspect. Ph.D. dissertation, MIT.

Abbott, Barbara 1992

"Definiteness, existentials, and the 'list' interpretation". In: Chris Barker- David Dowty, eds., Proceedings of Semantics and Linguistic Theory II, Columbus: The Ohio State University, 1-16.

1993

" A pragmatic account of the definiteness effect in existential sentences", in: Journal of Pragmatics 19, 39-55.

Babby, Leonard H. 1980

Existential sentences and negation in Russian. Ann Arbor: Karoma.

Bach, Emmon 1986

"The Algebra of Events," in: Linguistics and Philosophy 9, 5-16.

Barwise, Jon-Robin Cooper 1981

"Generalized Quantifiers in Natural Language", in: Linguistics and Philosophy 4, 159-220.

Belletti, Adriana 1988

'The Case of Unaccusatives', Linguistic Inquiry 19-1, 1-34.

Coppen, Peter-Arno 1991

Specifying the Noun Phrase. Amsterdam: Thesis Publishers.

Diesing, Molly 1992

Indefinites. MIT-Press, Cambridge.

Dowty, David R. 1979

Word Meaning and Montague Grammar. Reidel, Dordrecht.

Dowty, David-Belinda Brodie 1984

" A Semantic Analysis of'Floated' Quantifiers in Transformationless Grammar", in: Mark Cobler, Susannah Mackaye and Michael Wescoat, eds., Proceedings of the West Coast Conference on Formal Linguistics, vol. 3, 75-90.

Eijck, Jan van 1983

"Discourse Representation Theory and Plurality", in: Alice ter Meulen, ed., Studies in model-theoretic semantics, Dordrecht: Foris, 85-106.

Εης, Mürvet 1991

"The Semantics of Specificity", Linguistic Inquiry 22, 1-25.

22

Jacob Hoeksema

Evans, Gareth 1977

"Pronouns, Quantifiers, and Relative Clauses I", Canadian Journal of Philosophy 7-3.

Hoeksema, Jacob 1984

"Partitives", unpublished paper, University of Groningen.

1985

Categorial morphology, Garland Press, New York.

1989

review of Eric Reuland and Alice ter Meulen (eds.) 'The Representation of (In)definiteness', in Language 65-1, 115-125.

Hoop, Helen de 1992

Case Configuration and Noun Phrase Interpretation. versity of Groningen.

Ph.D. dissertation, Uni-

Hoop, Helen de-Guido Vanden Wyngaerd-C. Jan-Wouter Zwart 1990

"Syntaxis en semantiek van de van i//e-constructie", in: Gramma 14, 81-106.

Horn, Laurence R. 1978

"Remarks on Neg-Raising." In: Peter Cole, ed., Syntax and Semantics, vol. 9: Pragmatics. New York, etc.: Academic Press, 129-220.

Jackendoff, Ray S. 1977

X'-syntax. Cambridge, Massachusetts: MIT-Press.

Jong, Franciska de 1987

"The Compositional Nature of (In)definiteness", in: Eric Reuland-Alice ter Meulen, eds., 270-285.

Jong, Franciska de-Henk J. Verkuyl 1985

"Generalized Quantifiers: The Propemess of Their Strength", in: Johan van Benthem-Alice ter Meulen, eds., Generalized Quantifiers in Natural Language, Foris, Dordrecht.

Kamp, Hans 1981

Ά theory of truth and semantic representation', in: Jeroen Groenendijk-Theo Janssen-Martin Stokhof, eds., Formal Methods in the Study of Language, Vol. 1. Reprinted in: Jeroen Groenendijk-Theo Janssen-Martin Stokhof, eds., 1984, Truth, interpretation and information, Dordrecht: Foris, 1-41.

Kayne, Richard S. 1975

French Syntax. The Transformational

Cycle. Cambridge: MIT-Press.

Keenan, Edward L. 1987

"A Semantic Definition of'Indefinite NP'", in: Eric Reuland-Alice ter Meulen, eds., 286-317.

Introduction

23

Keenan, Edward L.-Yonathan Stavi 1986

"A Semantic Characterization of Natural Language Determiners", in: Linguistics and Philosophy 9, 253-326.

Klein, Maarten 1981

"De interne structuur van partitieve constructies", in: Spektator

10, 295-309.

Ladusaw, William A. 1982

"Semantic constraints on the English partitive construction", in: Daniel Flickinger-Marlys Macken-Nancy Wiegand, eds., Proceedings of the First West Coast Conference on Formal Linguistics, Stanford: Linguistics Department, Stanford University, 231-242.

Lasnik, Howard 1992

"Case and Expletives: Notes toward a Parametric Account", Linguistic 23-3, 381-405.

Inquiry

Levin, Beth 1993

English Verb Classes and Alternations. and London.

University of Chicago Press, Chicago

Link, Godehard 1983

"The Logical Analysis of Plural and Mass Terms: a lattice-theoretical approach," in: Rainer Bäuerle-Christoph Schwarze-Amim von Stechow, eds., Meaning, Use and Interpretation of Language, Berlin: De Gruyter, 302-323.

Meulen, Alice ter 1980

Substances, Quantities and Individuals. A study in the formal semantics of mass terms. Dissertation, Stanford University, reproduced by the Indiana University Linguistics Club, Bloomington, Indiana.

Miller, George A. (ed.) 1990

"WordNet: An On-line Lexical Database", in: International cography 3, 235-312.

Journal of Lexi-

Milsark, Gary 1977

Towards an explanation of certain peculiarities of the existential construction in English', Linguistic Analysis 3, 1-30.

Moltmann, Friederike 1992

Parts and wholes in semantics. Manuscript, to appear with Oxford University Press.

Nerbonne, John, Masayo lida, and William Ladusaw 1989

'Running On Empty: Null Heads in Head-Driven Grammar', in: Katherine Hunt-Jane Fee, eds., Proceedings of the Eighth West Coast Conference on Formal Semantics, Stanford Linguistics Association, Stanford, 276-288.

24

Jacob Hoeksema 1990

'Semantics of Common Noun Phrase Anaphora', in: Aaron Halpem, ed., Proceedings of the Ninth Annual West Coast Conference on Formal Linguistics, 379-394.

Pesetsky, David 1982

Paths and Categories. Doctoral dissertation, MIT.

Reed, Ann 1988

"Semantic Groups and Discourse Groups", in: Joyce Powers-Kenneth de Jong, eds., Proceedings of the fifth Eastern States Conference on Linguistics, Columbus: The Ohio State University, 416-27.

Reuland, Eric J.-Alice G.B. ter Meulen, eds. 1987

The Representation

of (In)definiteness,

Cambridge: MIT-Press.

Roberts, Craige 1987

Modal subordination, anaphora, and distributivity. PhD dissertation, University of Massachusetts, Amherst.

Selkirk, Elisabeth O. 1977

"Some Remarks on Noun Phrase Structure", in: Adrian Akmajian-Peter Culicover-Thomas Wasow (eds.), Formal Syntax, New York: Academic Press, 285-316.

Sportiche, Dominique 1988

"A Theory of Floating Quantifiers and Its Corollaries for Constituent Structure", Linguistic inquiry 19, 425-449.

Stockwell, Robert P.-Paul Schachter-Barbara Hall Partee 1973

The major syntactic structures of English. New York etc.: Holt, Rinehart and Winston.

Szabolcsi, Anna 1985

"Comparative Superlatives," in: MIT Working Papers in Linguistics,

245-265.

1986

"From the Definiteness Effect to Lexical Integrity", in: W. Abraham and S. de Mey, eds., Topic, Focus and Configurationality, Amsterdam: Benjamins, 321348.

Verkuyl, Henk J. 1993

A Theory of Aspectuality. Cambridge etc.: Cambridge University Press,

Ward, Gregory-Betty Birner 1993

'TAere-sentences and information status", paper presented at the LS A Annual Meeting in Los Angeles, January 1993.

Doing without a Partitive Constraint* Barbara

Abbott

1. Introduction The Partitive Constraint seems to be an elegant and satisfying dividend of recent work elaborating the generalized quantifier approach to noun phrases, especially given that the constraint was already acknowledged in the existing syntactic literature on the partitive construction, having outlived its earlier incarnation as a sometimes obligatory "of dropping" transformation (Jackendoff 1968). Stated syntactically, the constraint requires an NP embedded in a partitive to be definite, more specifically to be determined by the, a demonstrative, or a possessive (Selkirk 1977, Jackendoff 1977). Thus examples like those in (la) would be allowed, while those in (lb) would be ruled out. (1)

a. many of these books, three of your apples, some of the pencils b. many of all books, three of no apples, some of many pencils

But, as noted by Barwise and Cooper (1981), the permitted NPs are exactly the ones that, under the generalized quantifier approach, are guaranteed to denote a proper principal filter, and thus to provide a CN-type denotation

I would like to express my appreciation to the Department of Linguistics and Germanic, Slavic, Asian, and African Languages, and the College of Arts and Letters, of Michigan State University for a research leave during which this paper was written. Some early work on the topic was presented at the University of Illinois, Wayne State University, and the Cognitive Discussion group at MSU and I would like to thank those audiences for their comments. I would also like to thank Larry Hauser, Jacob Hoeksema, Ellen Prince, and an anonymous reviewer for reading and commenting on an earlier version. Their comments have resulted in improvements; responsibility for remaining faults, of course, rests with me.

26

Barbara Abbott

(the generator set of the principal filter) to combine with the first, or 'upstairs', determiner in the partitive construction.1 This happy result does require a couple of auxiliary assumptions. First, it must be assumed that these definite NPs are undefined if their CN sets are empty - the situation in which they would denote an improper filter if they were defined. Second, it must be assumed that NPs determined by universal quantifiers {every, each) are always defined - so that when their CN sets are empty they are allowed to denote an improper filter. This is so that universally quantified NPs will be ruled out of the embedded position in a partitive. One small problematic detail noted by Barwise and Cooper led to further advances. Although both is definite, one does not say one of both apples. Ladusaw (1982) and Hoeksema (1984)2 each solved this problem and in so doing provided more general accounts of the semantics of partitives. The basic insight was that the embedded NP in a partitive must denote a group, while both is inherently distributive.3 Ladusaw proposed expanding the universe of discourse to contain group level entities - singleton sets containing nonempty nonsingletons - which generate principal filters just like ordinary entities do. The embedded NP in (2a) below denotes such a group level individual while that in (2b) does not. (2)

a. one of the two students b. *one of both students

Ladusaw's interpretation rule then interprets ofNP in a partitive as the value of a downstepping 'consists of' function which maps the generator of a group level individual onto the set it contains - in the case of (2a), the set of students (which is presupposed to consist of just two). Hoeksema's 1.

I'm generalizing somewhat here on Barwise and Cooper, who did not explicitly provide a semantics for demonstrative and possessive determiners or for plural definites without numerals.

2.

Hoeksema's paper was not specifically motivated by the both problem, but proposed an alternative to Barwise and Cooper's syntactic treatment of the partitive construction, while incorporating and extending some of their semantic insights.

3.

Keenan and Stavi (1986: 289) give a solution which is similarly motivated but syntactic. They propose to derive both by means of a "spelling" rule from each of the two. One problem for this approach is number agreement, given that both is plural and each is singular.

D o i n g without a Partitive Constraint

27

solution is similar, but it does not require the downstepping function. Instead, his universe of discourse has a part-whole relation defined on it directly. Hence the two students would denote (the filter generated by) a complex entity having each of the two students as a part. Hoeksema's interpretation rule gathers the parts of the generator into a set. Again, as with Ladusaw's account, a generator is required - only principal filters need apply. This insight concerning groups solves another, more subtle, problem with the original Barwise and Cooper analysis (as Roberts notes (1987: 204)). Universally quantified NPs always denote principal filters, and - unless their CN sets are empty - proper ones as well. The requirement that the embedded NP in a partitive always denote a proper principal filter rules these out, at the expense of much of the explanatory power of the analysis. (That is, if the reason definites are allowed in partitives is because they provide a generator set, then universally quantified NPs ought to be allowed too.) The requirement that the embedded NP in addition denote a group level individual rules out the universal (distributive) cases naturally and automatically. Hoeksema, however, noting examples like those in (3) (3)

a. the best of all possible worlds b. the most beautiful of all

provides an additional group interpretation for all.4 Both Ladusaw and Hoeksema recognize the existence of what have been called "mass" partitives - examples such as those in (4): (4)

a. most of the gold b. some of the book

4.

I agree with Hoeksema that all needs a group as well as a distributive interpretation. However I am not sure that the examples in (3) are genuine partitives. Note that with an upstairs superlative w e can also get embedded bare plurals, which (as discussed at length below) are not allowed otherwise in partitives: (i)

a. the happiest of men b. the best o f friends

(ii)

a. *all of men b. *some o f friends

I provide a more clearly partitive example, showing all with a group interpretation, below in section 4.

28

Barbara Abbott

Ladusaw suggests a continuity between his downstepping function and a function which would yield, for any individual, the stuff of which it consists. A unified analysis would further support the downstepping function and its explanatory role in the Partitive Constraint. Hoeksema (following Link 1983) explicitly provides for a things-to-stuff function, which plays a role in his interpretation of examples like those in (4). In short, bolstered by the group/distributive distinction, the Partitive Constraint looks close to becoming "a theorem of the semantics" (Ladusaw 1982: 239). Indeed, it has been asserted or assumed by many others in addition to Barwise and Cooper, Ladusaw, and Hoeksema; e.g. Westerstähl ("Concerning the position DET2 [in a 'DETl of DET2...N' construction], we find that (a) the possessives (b) the demonstratives...(c) the definite article will do. In fact, ...precisely these DETs fit here" (1985: 63-64)), Keenan and Stavi ("[In a det of the form DET of DET] the second DET position must be definite (and syntactically plural)" (1986: 297)), and Link ("It follows that the core NP [i.e. "the NP following of"] has to be definite (cf. *some of all men, *all offew books)', this is the Partitive Constraint" (1984: 15; cf. also Link 1987)). What is the matter with this picture? The main problem is a very large group of counter-examples. In view of these, I will argue below that there is no semantic definiteness constraint on the embedded NP in a partitive. Instead, the types of examples which have been cited in the literature as evidence in favor of the constraint are, with one exception, anomalous only for very general pragmatic reasons. The exception - partitives with embedded bare NPs (e.g. *some of apples) - are genuinely ill-formed and ruled out by independently motivated principles. In addition, this analysis supports an alternative, independently motivated view of NP interpretation which would not, in any case, yield the Partitive Constraint as a theorem. The remainder of this paper is organized as follows: in section 2 I review some data which are problematic for the Partitive Constraint. This will raise the question of the relationship between ordinary partitives and mass partitives, such as those above in (4), and I will argue in favor of considering them to belong to a single construction. In section 3 I will examine two revisions of the Partitive Constraint which have been suggested in the literature, and I will argue that they too are not supported by the evidence. In section 4 I will propose an alternative explanation for the strangeness of examples which have led people to believe in the Partitive

Doing without a Partitive Constraint

29

Constraint, and distinguish the pragmatically odd cases from the genuinely ill-formed ones. In section 5 we will turn more explicitly to the semantics of partitives, and I will show that interpretations can be provided for all the examples except those which are genuinely ill-formed.

2. Counterexamples to the Partitive Constraint 2.1. Ordinary partitives with embedded indefinites It has been acknowledged by many of those who have written about partitives that there are some pesky counterexamples to the assumption that the embedded NP is always definite, i.e. determined by the, a demonstrative, or a possessive. The examples in (5)-(9) are culled from the literature and the ones in (10) are my own. Note that (6) is a "mass" (i.e. nonplural) example. More on this below. (5)

a. One of some boys who were playing in the alley got arrested. b. He ate three of some apples he found on the ground. (Stockwell, Schachter, and Partee 1973: 144)

(6)

I heard too much of one speech and not enough of the other. (Selkirk 1977: 315, η 7)

(7)

a. That book could belong to one of three people. b. This is one of a number of counterexamples to the PC. c. John was one of several students who arrived late. (Ladusaw 1982: 240)

(8)

They called the police because seven of some professor's were missing.5 (Keenan and Stavi 1986: 297)

5.

Note that although the embedded determiner in this example is a possessive, it is not definite according to Keenan and Stavi's semantics, as they point out.

manuscripts

30

Barbara Abbott

(9)

a. Only one of many people who saw the accident would testify. b. Only one of many applicants passed the test. (Reed 1989: 421.)

(10) a. Ants had gotten into most of some jars ofjam Bill had stored in the basement. b. Three quarters of half the population will be mothers at some point in their lives. c. We put two strawberries on each of three pies (and kiwi slices on the rest). d. Any one of several options are open to us at this point. e. All of three people (out of the 50 I wrote to) had the politeness to respond to my invitation. f. The five worst of any problems that you find that can't be fixed should be listed at the beginning. Examples such as these suggest that much of the semantic motivation for the Partitive Constraint has already been lost. We saw above in the Introduction that the fact that an NP always denotes a proper principal filter under a generalized quantifier analysis was not sufficient for its being embedded in a construction where a CN type of denotation must be derived These examples show that it is not necessary. More seriously, if Barwise and Cooper's generalization were an accurate one, examples such as those in (5) and (7)-(10) should be uninterpretable, but they are not. Fortunately we have an alternative at hand. Roberts (1987) has argued on independent grounds for extending the variable/quantifier distinction of Kamp (1981) and Heim (1982) to a division of NPs generally speaking into two categories which she calls "individual denoting" and "quantificational". 6 The individual denoting NPs are regarded as denoting individuals, where "individual" here includes groups, while the quantificational NPs have an inherently distributive interpretation. If (as we might expect) it should turn out that the NPs embedded in group partitives were all in Roberts' individual denoting 6.

As Roberts notes, this is just one of many proposals for dividing NPs semantically into two categories, some of which relate more closely to Roberts' distinction than others. Cf. Milsark's 1977 "weak" (cardinality words) vs. "strong" (quantifiers), Scha's 1981 "distributive" vs. "collective" (discussed in Roberts 1987), or Reuland and ter Meulen's "nonquantificational" vs. "quantificational" uses of NPs (1987: 13f¥). See also the references in Reinhart (1987: 143).

Doing without a Partitive Constraint

31

category, in particular in the group type subset of that category, then we would have an appropriate kind of interpretation for them automatically. This is, in fact, half of the analysis I am proposing, but we need to turn to mass partitives to get the other half of the picture.

2.2. Mass partitives which violate the PC There is a difference of opinion in the literature as to the relation between ordinary partitive examples such as those in (la) and (2a), and the mass examples like those in (4) (and Selkirk's (6)). Is this the same construction, or a different one? Selkirk, Ladusaw, and Hoeksema (either tacitly or explicitly) assume it is the same. Keenan and Stavi (1986: 290) suggest otherwise, and provide a distinct type of structure for the mass construction. 7 Example (11) (= their (86)) has each as the embedded determiner, which would not be allowed in an ordinary partitive. (11) Most of each of the houses was destroyed. Stockwell et al. (1973: 144ff.) argue that the example in (12a), with a fraction, belongs to a different construction; their argument is based (in part) on (12b) which violates the Partitive Constraint (in their terms, a constraint against generics in partitives). (Another of their arguments is looked at below.) (12) a. One-half of the broom is red. b. One-half of a broom is not very useful. Reed describes such examples (hers are given in (13)) as having a "discourse use and distribution" {op. cit.: 425, η. 1) distinct from ordinary partitives, but does not amplify on this point. (13) a. half of a cake b. some of each book 7.

It should be noted that Keenan and Stavi's analysis of the ordinary partitive construction differs in interesting ways from that of Barwise and Cooper. Since they agree on the points relevant to this paper, however, I won't go into those differences.

32

Barbara Abbott

Putting aside for a moment the question of the relation between these examples and the ordinary partitive examples usually cited, we can observe many additional examples which contain embedded NPs that would be ruled out in ordinary partitives by the Partitive Constraint. (14) a. I'll be back in three quarters of an hour. b. Why settle for half of a loaf? c. There was most of a birthday cake and all of a large vegetarian pizza sitting on the buffet. d. That sounds like too much of a good thing. (15) a. The Smithsonian donated most of both rare book exhibits. b. One third of every book Chomsky writes is footnotes. c. Those who ate some of any entree prepared with mayonnaise should report to the local health facility. d. At least a quarter of most fruits consists of rind and seeds. e. I could follow a little of some of those arguments. f. Eating too much of any kind of food is unhealthy. It should be noted that the examples in (15) also violate the group denotation requirement. In fact the embedded NP in each case has a distributive interpretation, which is why we have a mass reading in these examples - the interpretation of the upstairs determiner applies individually to entities picked out or described by the embedded NP. An example such as (16) shares this distributive characteristic without having a mass interpretation. (16) I had one of each kind of cookie. For the purposes of this paper I will continue to refer to partitives like those in (4), (6), and (11)-(16) as mass partitives, even though this terminology is a little strange for (16). (In any case part of my purpose here will be to argue that the distinction is artificial and unnecessary.) Mass partitives, then, are partitives in which parts of entities are referred to, where "entity" must be taken so as to include kinds while excluding groups referred to by plural NPs. Note that groups referred to with singular nouns such as crowd, group, class, etc., do behave like singular (atomic) entities in this context, and such nouns are permitted in mass partitives, e.g.:

Doing without a Partitive Constraint

33

(17) a. Some of the crowd was stranded behind the barricades. b. Most of Bill's family smokes. (See Roberts 1987: 187ff, for another argument that such nouns must be regarded as referring to atomic entities.) I will call the more traditional examples, as well as those in (5) and (7)—(10), group partitives. Group partitives, then, are partitives which refer to parts of groups. It was suggested above that, under the approach of Roberts (1987), group partitives could be provided with an interpretation whether or not they contained embedded definites, as long as the embedded NP was a grouptype individual-denoting NR If the NPs embedded in mass partitives are either singular or quantificational, we have the possibility of a unified analysis for partitives of both types and with no need for a separate Partitive Constraint. This is essentially what I will propose, although there are a few complications. We will look at the details of the semantics in section 5 below. At this point I would like to look briefly at the question of whether group and mass partitives should be considered to be part of a single construction. Anyone who assumes that they should be so considered can skip the following section without loss of continuity.

2.3. Mass partitives are partitives The main arguments for considering mass partitives to be partitives are that they look like partitives, and that the simplest syntactic analysis would probably generate both kinds. Furthermore as we will see below in section 5, the differences in their semantic interpretation will follow naturally from the interpretation of the embedded NP, without any need for special stipulation. Thus the burden of proof seems to be on anyone who wants to maintain that they are in fact a different construction. One argument against considering them to be partitives which has appeared more than once in the literature but which is obviously not going to carry a lot of weight in the present context is the fact that they seem even more freely than group partitives to violate the Partitive Constraint. (Nevertheless this is a fact that needs an explanation, and I will try to provide one in section 4 below.) Selkirk (1977) gave a number of arguments for distinguishing (group) partitives from what she called "pseudopartitives". These are phrases like a

34

Barbara Abbott

bunch of bananas, a lot of turkey, a cup offlour. One might try to argue that mass partitives are a species of pseudopartitive, but there are a number of arguments against such a view. The strongest argument against considering mass partitives to be pseudopartitives is that pseudopartitives are distinctive for embedding bare plurals and mass nouns, which, as noted above, are prohibited in mass partitives (as well as group partitives): cf. *most of brooms, *some of milk. Indeed this characteristic is essential for Selkirk's structure for pseudopartitives, according to which what is embedded in them is an N' rather than an NP. All of the examples of mass partitives given above have determiners, and thus could hardly be assigned Selkirk's pseudopartitive structure. Furthermore insofar as the syntactic differences between partitives and pseudopartitives pointed out by Selkirk are related specifically to this difference in structure, we would expect mass partitives to behave like group partitives rather than pseudopartitives, and would have no explanation if they failed to. There is one further difference pointed out by Selkirk, and that is the deletability of (or option not to insert) of in pseudopartitives as compared with true partitives: one cup (of) flour. It should be noted that this characteristic is not completely reliable cf. the regular partitives all (of) the apples, half (of) the students. However it does group the mass partitives with the group partitives, since the only o / ' s that are deletable in examples (11)-(16) are those which follow half. Another possible argument for distinguishing mass partitives from the group examples comes from Stockwell et al. (1973: 144) in connection with the examples in (12). They note the inability in these cases, in contrast with more typical partitives, to prepose the o/-phrase. It is not clear how conclusive this kind of evidence is. For one thing, there is a problem preposing the ofphrase in a number of the group partitive examples, cf. (18) a. ?Of some boys who were playing in the alley, one got arrested. b. ?Of three people, that book could belong to one. One might suppose that the indefiniteness of the embedded NP is having an effect here; the preposing construction might be thought to be connected in some way with discourse topics, which should be denoted by definite NPs. Note that although Stockwell, et al. star the preposed version of (12a), the essentially parallel (19a) seems fine, while (19b) with an embedded indefinite does not.

Doing without a Partitive Constraint

35

(19) a. Of my introductory syntax class, one half got 4.0s (and the other halfflunked). b. ?Of an introductory syntax class, I usually expect one half to flunk. However intuitions here are anything but robust, and Quirk, Greenbaum, Leech and Svartvik (1972) give the following examples and judgments, concluding that "the initially placed o/-phrase is functionally different, although semantically similar to the postmodifying phrase". (20) a. b. c. d.

Of ten reviewers, only a few praised his play. ?*Only a few of ten reviewers praised his play. Of fourteen women, ten were highly critical of the proposal. ?*Ten of fourteen women were single.

The evidence here is inconclusive, and must await further research on the preposed construction. To summarize the results of this section, there does not seem to be any strong evidence in favor of regarding mass partitives as belonging to a different type of construction from group partitives. Given the advantages noted above in grouping them together, I will henceforth make that assumption.

3. Two possible revisions of the Partitive Constraint As the citations above make clear, the existence of counterexamples to the Partitive Constraint has been known for some time. There are two main suggestions for revision existing in the literature. The most common of these - made by Selkirk, Ladusaw, and Hoeksema - is that the embedded NP must be, if not definite, at least specific in some sense. Ladusaw remarked, in connection with his examples in (7), repeated here: (7)

a. That book could belong to one of three people. b. This is one of a number of counterexamples to the PC. c. John was one of several students who arrived late.

that they "are appropriately used only when the user has a particular group of

36

Barbara Abbott

individuals in mind. ... It seems that the pragmatic notion of an introduced discourse entity is relevant here" (p. 240); and Hoeksema suggested, with respect to the same examples, that "perhaps an analysis of specificity along the lines of Fodor and Sag (1982) is needed" (p. 40, n. 10). Let us first look at this idea.

3.1. Fodor and Sag's 1982 referential indefinites Fodor and Sag presented convincing evidence that at least some indefinite NPs are actually ambiguous, having a "referential" reading in addition to a "quantifier" one. The semantic analysis they propose for the referential interpretation would treat these NPs on a par with proper names, demonstratives, and definite descriptions as denoting proper principal filters. Although most of their examples contained NPs of the form a CNP, they did note in their introductory remarks: we believe that a comparable semantic ambiguity can be demonstrated for numerical determiners such as two, three, seventeen, and so on, and also for some, several and many (though every, all, each, most, few, no and the null plural determiner appear to have only a quantifier interpretation). (Fodor and Sag (1982: 355)) The relevant group of NPs would thus include many of those which occur in the problematic examples given above, including all but one of the group partitives. Assuming the correctness of Fodor and Sag's analysis, if it could be shown that the embedded NP in a group partitive always had a referential interpretation then the Partitive Constraint would again follow from the semantics, at least for these cases. However this does not seem to be the case. Fodor and Sag characterize the referential reading of indefinites as one which is "used for the purpose of making an assertion about an individual" (p. 380) and remark that even though not enough information is supplied for the hearer to identify the referent, "the speaker could specify which individual his assertion was about" (p. 382). They assume also that "the referent... is fixed by the speaker's intentions" (p. 383). Thus it seems clear that to use an indefinite referentially a speaker must have some particular individual or individuals in mind - exactly what Ladusaw was suggesting. The examples below, however, could clearly be used by a speaker who has no particular

Doing without a Partitive Constraint

37

individuals in mind as referents of the underlined indefinites. (21) a. John was apparently one of several students who arrived late - / have no idea how many, or who the others were. b. Mary thinks that this is only one of a number o f counterexamples to the PC. c. Probably every speaker will only tell one of many possible jokes related to their topic. d. Kim asked them to tell the caterer to put two strawberries on each of three pies, and kiwi slices on the remainder. In addition, Fodor and Sag note some crucial characteristics of referential indefinites, which indefinites embedded in partitives need not have. These primarily involve having the widest possible scope over any other operators in the sentence8, but in the examples in (21b-d) above the partitive may have narrow scope with respect to dominating propositional attitude verbs, as well as the future tense, sentence adverb, and quantified subject in (21c). Further examples of partitives with indefinites having narrow scope with respect to other quantifiers, and having 'island-bound' interpretations, are given in (22). (22) a. Each of us has any one of several options open to us, if only we would stop to think. b. Every year only one of many applicants is admitted to the program. c. If three or more of some professor's manuscripts get published, the University benefits. d. Anybody who breaks more than one of any dishes they're given won't get any more. Examples such as those in (21) and (22) show quite clearly that, not only does the embedded NP in a partitive (even a group partitive) not have to be definite, it does not even have to be specific in any semantic sense yet explicated in the literature. Thus we seem to be thrown back on pragmatics. This brings us to the other existing suggestion for modifying the Partitive Constraint. 8

More properly, entailing a widest possible scope reading. Fodor and Sag argue that referential indefinites simply don't participate in scope relations.

38

Barbara Abbott

3.2. Reed 1989: Discourse groups9 Reed (1989) offers a pragmatic account which would seem to entail the Partitive Constraint. Specifically, she claims that the function of partitives is to introduce subgroups of existing discourse groups, where a discourse group is distinct from a semantic group and is defined as a "plural discourse entity" (p. 420). She argues that this account better explains the badness of (23a), since it also explains (23b, c). (23) a. *one of both students b. *one of the crowd c. *one of John and Mary Group nouns like crowd, army, class and so forth are used to refer to singular discourse entities, and hence are unacceptable embedded in a (group) partitive. Conjoined NPs, and NPs of the form both CNP introduce - or evoke, in the sense of Webber (1983) - discourse groups rather than accessing existing groups, according to Reed, and that is why they are unacceptable. Although there is much appeal in this analysis, there are some problems with it as it stands. One is giving a good account of the occurrence of embedded indefinites in partitives. Reed mentions such examples but, besides noting the need for the embedded NP to be related in some way to the discourse (which must surely be true of any NP, on general Gricean grounds), can only say that "the embedded indefinite NP is somehow taken to access rather than evoke a discourse group" (p. 421).10 However this does not seem to be an accurate description of such sentences. If we consider the examples above in (21), for instance, it seems clear that the entities referred to by the underlined NPs need not already exist in the discourse. Furthermore even if this description were accurate it would not explain what appears to amount to unusual behavior on this account, much less show why both and conjoined NPs cannot display the same unusual behavior. Another problem 9.

In some respects Reed's analysis bears a resemblance to that in Westerstahl 1984, which crucially invokes the notion of context sets. Westerstähl's analysis is semantic, however.

10.

Reed (1988) contains more extensive discussion of indefinites, but does not describe them as accessing discourse groups. It is unclear to me what exactly is intended in her discussion here, which is summarized as follows: "...an indefinite will be acceptable in a partitive when its reference becomes fixed in some way through context" (p. 19).

Doing without a Partitive Constraint

39

is the failure of this account to explain why both cannot typically appear naturally in existential sentences, one of whose functions, according to Reed, is to introduce discourse groups (p. 422). But the most fundamental problem with this analysis is that its basis must simply be stipulated. Why should partitives be confined to introducing subgroups of existing discourse groups? Could we have a construction that could introduce a subgroup of a new group? If not why not? If so, why should partitives be unable to do so? Such questions remain unanswered.

4. Doing without the constraint It does not appear that the embedded NP in a partitive must have a specific or referential reading (in the sense of Fodor and Sag) nor that it must access an existing discourse group. Let us take a fresh look at the situation. The simplest, and therefore most desirable, way for things to turn out would be for any informed or anomalous partitive examples to follow from the semantics of the construction plus any obvious or independently motivated pragmatic principles. Let us take a closer look at examples which have led to the postulation of the Partitive Constraint.

4.1. Examples from the literature The following examples are a representative sample: (24) a. three of some men b. many of all women c. several of no books d. two of too many acquaintances e. several of twenty of his roses that were sick f. nine of many of the lathe operators who were from Sicily g. three of nine planets of the solar system h. not much of little of Jane's wine that remained i. few of many questions j. any of many answers (Selkirk 1977: 294f.)

40

Barbara Abbott

(25) all of many men (Barwise and Cooper 1981: 184.) (26) a. some of all books b. many of no books c. each of few books (Ladusaw 1982: 281.) (27) a. each of no students b. at least two of exactly six students (Keenan and Stavi 1986: 289.) There are two main things to notice about these examples. One is that they are all given without any context - not even a surrounding sentence; and the other is that we know what they mean - we can interpret them without any trouble at all. I claim that each of these examples is both syntactically and semantically well-formed, and that the reason they have been cited as ungrammatical is purely pragmatic.11 The principle involved is a very general one that prohibits mentioning entities unless there is some reason for mentioning them. Note that these are all group partitive examples, and in most cases what is being described is an indefinite subgroup of some indefinitely specified group. Outside of any context at all, one's reaction is one of being left holding an empty bag. One wants to know why the 'outer' containing 11.

De Hoop (1992: 222-224) has argued for a semantic analysis. Noting the difference between Ladusaw's (7c), and (i), (7) c. John was one of several students who arrived late. (i)

?* John was one of several students, [evaluation is De Hoop's]

De Hoop argues that weak determiners embedded in a partitive must take two sets as arguments. This analysis is problematic however. One problem is the ad hoc assignment of a novel type to these weak determiners precisely for this situation. It may be that De Hoop did not realize that a new type was required, because of some confusion about the difference between a determiner expressing a relation between two sets, and a determiner expressing a function which takes two sets as arguments. In fact it would seem that under her analysis weak determiners embedded in partitives must actually be regarded as functions from pairs of sets to sets - that is of type

most

sitting o/t buffet(y)

The representation for an example like (15b), with an embedded quantificational NP, is given in (38). (15) b. One third of every book Chomsky writes is footnotes. 16.

As Jacob Hoeksema has pointed out to me, I am ignoring complications that arise in the semantics for most as it combines with mass nouns. Given my assumption of a nonatomic lattice for the mass domain, it will be infinite and it is not always clear what counts as most under those circumstances. Perhaps the approach sketched by Roberts (1987: 208ff), in which homogeneous parts of mass entities are introduced, should be adopted here.

48

Barbara Abbott

(38) χ book (x) written by Chomsky (x)

all

y y * h(x> 3 χ μ ( ν ) = μ(1ι(χ)) footnotes (y)

As noted previously we might expect that NPs embedded in a mass partitive would have to be either singular, or one of Roberts' quantificational NPs. However this does not turn out to be the case. Some, for example, can occur embedded in a mass partitive and with a distributive interpretation, as in (15e). (15) e. I could follow a little of some of those arguments. This suggests that some must be regarded as ambiguous, being able to have either a group or a quantificational interpretation. Providing an interpretation for an example like (15e) is not a problem, as shown in (39) χ = those arguments y y

some =>

ζ ζ > h(y) I could follow (z) a little (μ (ζ))

(Possible consequences of such examples for Roberts' overall approach will be touched on in the concluding section.) Finally consider an example like (16). (16) I had one of each kind of cookie. Let us suppose that words like kind, sort, etc., denote sets of suprema of the sublattices denoted by plural common noun phrases. Then a representation for (16) would look like (40).

Doing without a Partitive Constraint

49

(40) X kind of cookie (x)

all

y y These kids are all potential criminals. This kind of analysis does not seem to have any recent endorsements, however, because of a number of basic problems that it leads to. First of all, the rule would be an instance of a lowering rule, which is capable of moving a quantifier element from subject position into the VP. Such lowering rules are often considered to be ruled out by general principles. 5 Second, sometimes there are simply too many floating quantifiers per host: 6 (18) These kids are all potential criminals and have each received several warnings already. Here, it is inconceivable that both all and each have floated off the same subject NP. A third problem arises with plural agreement (as noted e.g. in Partee 1971): (19) a. Each of these men shaves himself. b. These men each shave themselves. The putative source for (19b), (19a), has singular number agreement, whereas (19b) itself has plural agreement. Unless we allow transformations to globally change syntactic features, this kind of effect is very difficult to deal with.

5.

In GB, this follows if all movement creates traces and traces have to be bound by ccommanding elements (which are structurally superior). But see Stowell (1981) for a proposal which allows lowering. In GPSG, lowering is ruled out by the slash mechanism, which calls for elimination of the slash feature at some higher level.

6.

Cf. Dowty (1986). Van der Does (1992) calls sentences such as (18) Dowty-sentences.

Floating Quantifiers, Partitives and Distributivity

65

A fourth problem that arises has to do with restrictions on partitive noun phrases. As discussed in Hoeksema (1984), and Reed (1988, this volume), partitives with a conjunction following of are not good: *each of Tom, Dick and Harry, *none of Fred, Ed, and Ned. However, conjunctions can be hosts/triggers of floating quantifiers: (20) a. Tom, Dick and Harry have each had a BLT sandwich. b. Fred, Ed and Ned have all been to France.

3.2. Sportiche's (1988) account Some of the problems raised above for the traditional movement account of floating quantifiers are countered in the alternative transformational theory of Sportiche (1988). According to Sportiche, the so-called floating quantifier is syntactically inert. What moves is its host. This entails that the subject originates inside the VP, and not just for passive and ergative predicates, as previously thought. Sportiche's proposal immediately solves the first problem. The movement operation in question is no longer a lowering rule. Similarly, the second problem disappears. Cases such as (18) above are now seen as involving extraction-across-the-board, a well-attested phenomenon whereby multiple gaps are satisfied by a single filler (cf. Ross 1967). The third problem, involving the pair of sentences in (19), may or may not be solved, depending on whether one treats the agreement after or before movement. If agreement is marked at surface structure, the particular problem of (19) might go away, although non-local versions of this problem remain. Consider the following pair of sentences: (21) a. Each of the men left after he had been insulted. b. *The men each left after he had been insulted. The only way to account for the ill-formedness of (21b) on Sportiche's theory would be to show that neither the NP [each t] can be an antecedent of he nor the NP the men. The latter part is easy: mismatch in number features prevents any direct link between the men and he. However, the first part of the task is more difficult. One might suppose that [each t] is too low to c-command the adverbial clause headed by after, but this is incorrect,

66

Jacob Hoeksema

since it is possible for quantifiers even when they occur in object position to bind pronouns in q/ier-clauses: (22) Fred kissed each girl after she kissed him. Another alternative might be to suggest that expressions containing traces cannot bind pronouns. But the examples in (23) show that expressions containing traces may bind pronouns. (23) a. Which students did you obtain [the phone number of tj. after it. had become obsolete? b. What city could you remember [the name of t]. only after it. had been changed by the revolutionaries? I conclude that the problem involving the binding of singular pronouns raised by Partee (1971) still poses a major obstacle for modern transformational accounts. One might note that Sportiche's paper concentrates on French tous and English all, both of which are plural floating quantifiers and do not create the kind of problem that each creates here. The fourth problem for the traditional account, which involves the examples in (24), also still stands, although Sportiche (1988: 440) dismisses it as no more than the sort of complication that any theory of floating quantifiers must face. But rather than dismiss this problem, one could also say that it is no more than one case of a very general problem for any movement account, which is that the moved element and the stranded element do not really fit together in their putative source. For example, there is the problem of partitive floating quantifiers which have a pronoun instead of a gap: (24) a. b. c. d.

These guys were neither of them very smart. We were all of us really delighted to come. They were all of them trained linguists. You are none of you in very good shape.

A simple movement account would have to postulate the following ungrammatical strings as the underlying forms for these sentences:

Floating Quantifiers, Partitives and Distributivity

(24') a. b. c. d.

[e [e [e [e

BE BE BE BE

67

[neither of them these guys very smart]] [all of us we really delighted to come]] [all of them they trained linguists]] [none of you you in very good shape]]

Sportiche (1988: 445) supposes that the pronouns in (24) are spelled-out gaps, resumptive pronouns in other words. Thus, (24a) could be derived as follows: (24") [e BE [these [these [these

[neither of these guys very smart]] => movement guys BE [neither t very smart]] => trace spelling guys BE [neither them very smart]] of-insertion guys BE [neither of them very smart]]

This suggestion lacks plausibility, however. There is no great similarity between the pronouns in the floating partitive quantifiers and the types of resumptive pronouns that are otherwise found in English. In particular, there is no evidence for resumptive pronouns in so-called Α-dependencies (to use the basic concepts and terminology of Sportiche's framework) such as passive or raising constructions in English: (25) a. b. c. d.

This problem was talked about on Thursday. *This problem was talked about it on Thursday. This problem seems to be unsolvable. *This problem seems it to be unsolvable.

And even if one can find a way around this problem, there is the additional problem of partitive quantifiers in Dutch. These do not contain personal pronouns and thus cannot be given a treatment in terms of resumptive pronouns. Yet they, too, indicate that floating quantifiers and their 'hosts' do not in fact need to be able to form a well-formed NP constituent: (26) a. De vrouwen waren geen van The women were none of 'The women were none of b. *Geen van allen de vrouwen

allen verlegen. all shy them shy' waren/was verlegen.

68

Jacob Hoeksema

(27) a. De decanen waren geen van drieen optimistisch. The deans were none of threes optimistic 'None of the three deans were optimistic' b. *Geen van drieen de decanen waren optimistisch. (28) a. Wij waren elk van beiden koppig. We were each of both stubborn 'We were both of us stubborn' b. *Elk van beiden wij/ons waren/was koppig. Each of both we/us were/was stubborn The type in (27), with a plural form of the numeral, is especially interesting. Normally, numerals cannot occur after partitive 'of unless they are either accompanied by a definite determiner, usually de 'the', or else specific indefinites (cf. Abbott, 1995). Here however, they must appear without a determiner: (27') *De decanen waren elk van de drieen koppig. Similar problems with partitive floating quantifiers arise in French (cf. Kayne 1975): (29) Ces hommes avaient tous les trois connu Garbo. Those men had all the three known Garbo 'Those men had all three of them known Garbo' Again, there is no possible source NP for a movement account, since tous les trois ces hommes is not grammatical. In German, the evidence is hardly better for a movement account. Giusti (1990a,b) extends Sportiche's theory to German, and argues for example that the neuter pronoun alles "all, everything" can be stranded by a moved NP, as in (30) Wer is heute abend alles da? Who is tonight all there 'Who will all be there tonight?'

Floating Quantifiers, Partitives and Distributivity

69

And indeed, wer alles "who all" can form a nominal constituent, as shown by example (31): (31) Wer alles war da? Who all was there Nevertheless, it is easy to find cases where floating quantifier and "host" cannot be pieced together like two parts of the same jigsaw puzzle. For example: (32) Es war alles gelogen. It was all lied 'It was all a lie' (33) *Es alles war gelogen. (34) *Alles es war gelogen. Indeed, there are floating quantifiers that never function as determiners or predeterminers in the way required by the movement theory. Such a quantifier is Dutch allemaal "all".7 8 This expression must float (as in sentence a below) or else sink (as in sentences b-e).

7.

There is actually a determiner allemaal, but this is not semantically related to the floating quantifier allemaal. The determiner does not mean "all" or "every", like the floating quantifier, but something like "a lot", and unlike universal quantifiers, it is weak in the sense of Barwise and Cooper (1981), which entails that it can be used in existential sentences: (i)

Er waren allemaal problemen. There were alot-of problems

Perhaps it is best compared to English phrases like all kinds of, which also have the property of being weak. The distribution of the determiner allemaal is peculiar. For instance, it does not appear in measure-objects, just as English all kinds of doesn't appear there: (ii) *Het *the (iii) *Het *the 8.

feest duurde allemaal uren (OK: vele, verscheidene) party lasted all-kinds-of hours (OK: many, several) boek kostte allemaal guldens (OK: vele, verscheidene) book cost all-kinds-of guilders (OK: many, several)

In this respect Dutch allemaal differs somewhat from its Afrikaans cognate and counterpart almal which can head a partitive construction (Oosthuizen 1989: 2):

70

Jacob Hoeksema

(35) a. De boeren hadden allemaal hooikoorts. The farmers had all hay-fever 'The farmers all had hay-fever1 b. *De boeren allemaal hadden hooikoorts,9 c. *Allemaal de boeren hadden hooikoorts. d. *Allemaal van de boeren hadden hooikoorts. e. *Allemaal hadden hooikoorts.10 (i)

a. Sy haat almal She hates all b. Sy haat hulle She hates them

van hulle. of them almal. (with floated almal) all

It can also occur as a pronominal argument (Oosthuizen 1989: 156) and so it has the distribution of English all rather than that of Dutch allemaal: (ii) Almal blyk ongelukkig te wees. All seem unhappy to be 9.

Note however that allemaal is grammatical as a postmodifier of a pronoun, as in: (i)

Wij allemaal hebben gefaald. we all have failed

(ii) Een geschenk van ons allemaal a present of us all However, unlike the floating quantifier allemaal, the postmodifier does not cooccur with a weak (unstressed or clitic) pronoun, cf. the following contrast (we is the weak variant of wij): (iii) We hebben allemaal hard gewerkt we have all hard worked 'We have all worked hard' (iv) *We allemaal hebben hard gewerkt. This makes it unlikely that (iii) is to be derived from the same source as (iv), unless an independent explanation can be found for the grammaticality of (iii). 10.

There is one use of allemaal as a pronominal element of which I am aware, and this use appears to be idiosyncratic, given that the expression otherwise lacks the distributional properties of pronouns. This is the use of allemaal in superlative constructions: (i)

Freddie is het mooist van allemaal. Freddie is the prettiest of all

cf.: (ii) *Freddie is zat van allemaal Freddie is tired of all

Floating Quantifiers, Partitives and Distributivity

71

Parallel cases can be found in the Groningen dialect of Dutch. This dialect makes a morphological distinction between pronouns and determiners. Pronouns are frequently derived from the determiners by adding the suffix -ent, cf. e.g. baaide "both" (determiner) and baaident "both" (pronoun). The floating quantifier use turns out to involve the pronominal form, not the determiner form, cf. Baaide/ *baaident kiender gingen mit "Both children came along" versus Dij kiender gingen baaident/*baaide mit "Those children both came along". Yet other classes of examples in which floating quantifiers cannot form a constituent with their 'host' involve full quantificational NPs. In earlier stages of English, an NP such as every man could serve as a floating quantifier. Instances of this use can be found for instance in the King James (or Authorized) translation of the Bible. Some examples are given in (36) below, together with a similar one from the Book of Mormon: 1 ' (36) a. (..) then shall they give every man a ransom for his soul unto the Lord (Exodus 30:12) b. For they cast down every man his rod, and they became serpents: but Aaron's rod swallowed up their rods. (Exodus 7:12) c. Then they speedily took down every man his sack to the ground, and opened every man his sack. (Genesis 44: 11) d. Then they rent their clothes, and laded every man his ass, and returned to the city. (Genesis 44: 13)12 e. they shall eat every man the flesh of his own arm (2 Nephi 19:20) One may also note here cases with every one as a floating quantifier: f. all the Dutch fleet, man-of-war and merchant East India ships, are got every one in from Bergen (S. Pepys, Diary entry, 9 Sept 1665)

11.

It is striking that all examples have a singular pronoun which is bound by every man. If this is indeed significant, and not accidental, a different analysis might be called for which exploits the presence of the pronoun.

12.

Note the stereotyped character of these examples, all of which involve pronominal subjects and the fixed expression every man.

72

Jacob Hoeksema

There is no reason from either distribution or meaning not to call such quantificational NPs floating quantifiers, and the fact that they are complete, and not to be analyzed as remnants stranded by some fronting operation, makes it highly doubtful that Sportiche's theory can be extended to account for them. If floating-quantifier constructions derive from partitives through movement, then we would expect to find more floating quantifiers than we actually do. For instance, just as we derive The boys all left from All (of) the boys left, we can derive *The boys three left from Three of the boys left. Indeed, there are no universal laws barring numerals from functioning as floating quantifiers, as languages such as Japanese shows (cf. Miyagawa 1989, Fukushima 1991): (37) Otoko-ga go-nin ki-ta man-nom five come-past 'Five men came' Hence we must conclude that Sportiche's approach sometimes overgenerates (in the case of floating numerals in English) and sometimes undergenerates, in the case of floating quantifiers which do not have a source as a determiner or predeterminer (as we saw in the case of Dutch allemaal). A final problem which needs to be mentioned here involves the use of floating quantifiers in absolute constructions. These constructions, which have the subject-predicate structure of regular sentences but lack verbs and inflection, may also contain floating quantifiers: (38) a. With these enemies both on the same planet, it was too dangerous to beam Kirk down. b. Only with his enemies each on a different planet would he stand a chance, Kirk realized. c. With his crew members not all ready to beam down, Kirk had to play for time. The point about these examples is that they lack a VP out of which the subject of the absolute construction could have moved. To the best of my knowledge, there is no evidence whatsoever for movement in these constructions. While movement could always be postulated here, it would only be an ad-hoc move to save the theory.13

Floating Quantifiers, Partitives and Distributivity

73

To sum up: The evidence for a local movement account of floating quantifiers along the lines of Sportiche (1988), Giusti (1990a,b) and others is weak. The stranding evidence crucial to such an analysis is available only for a small number of floating quantifiers.14 In many other cases a movement analysis is forced to postulate impossible phrases as underlying sources. There are additional problems with Sportiche's theory that have been discussed in Doetjes (1992). 15 Doetjes' account of the French data assumes 13.

A somewhat similar problem arises in connection with the following example from Lewis Carroll: (i) Another Rule of Battle, that Alice had not noticed, seemed to be that they always fell on their heads, and the battle ended with their both falling off in this way, side by side: when they got up again, they shook hands, and then the Red Knight mounted and galloped off. Here the element serving as the antecedent to the floating quantifier both is the possessive pronoun their. To accomodate this example in Sportiche's theory, it is necessary to assume movement of the pronoun they into the specifier slot of the NP, where it is assigned genitive case. But then it is unclear how to treat the corresponding Acc-ing case: (ii) the battle ended with them both falling off in this way

14.

Not even all cases of French tous are accounted for by Sportiche's movement analysis, such as cases where tous precedes the complementizer que (examples from Kayne 1975: 63): (i)

II faut toutes qu'elles s'en aillent. it must all that they go away 'It is necessary that they all go away'

(ii) It faut tous qu'on se tire it must all that one refl beat-it 'It is necessary we all beat it' However, I will assume with Kayne that these cases present a special case (there is variation in acceptability among speakers concerning these cases, and there are curious restrictions on the morphological forms which may appear before the complementizer, as well as on the embedding verb). 15.

One such problem is the lack of evidence for stranding from object position in English, something that Sportiche's theory would predict to occur. As a matter of fact, sentence (i) below (taken from Doetjes 1992: 328) is out but predicted to be OK on a stranding account, while (ii), is rather better, but predicted to be out, since the position of the floating quantifier is not the deep structure position of the moved object. (i) *The books, which I will have to read all, are interesting. (ii) ?The books, which I will all have to read, are interesting.

74

Jacob Hoeksema

that floating quantifiers are adverbial in nature and is therefore compatible with my analysis as sketched below.

4. Floating quantifiers as predicate modifiers The last type of account I will discuss in this paper is the adverbial or modifier account of floating quantifiers. According to this type of analysis, which can be found in one form or another in Dowty and Brodie 1984, Roberts (1987), Fukushima (1991), and Van der Does (1992), floating quantifiers are essentially adverbial elements which serve as operators on the verb phrase or parts thereof. The adverbial theory of floating quantifiers immediately explains why floating quantifiers in English may show up in sentence-medial position (see the examples in (3) above), a position where otherwise only adverbials and parentheticals occur.16 Other positional aspects of floating quantifiers are a

16.

Link (1974) notes that the set of positions for floating quantifiers in German is a subset of the set of positions for parentheticals. One position where parentheticals and some adverbs (usually adverbs with a parenthetical intonation) may occur is that between the first constituent of a main clause and the finite verb, which normally occupies the second position in German main clauses. Here floating quantifiers and most adverbial expressions may not occur. Similar observations can be made for Dutch, cp. the following sentences: (i)

De mannen zijn gisteren allebei gearresteerd. the men are yesterday both arrested "the men were both arrested yesterday"

(ii) De mannen zijn allebei gisteren gearresteerd. the men are both yesterday arrested "the men were both arrested yesterday" (iii) De mannen echter/*allebei zijn gisteren gearresteerd the men however/both are yesterday arrested "the men however/both were arrested yesterday" (iv) Allebei zijn de mannen gisteren gearresteerd both are the men yesterday arrested "Both men were arrested yesterday"

Floating Quantifiers, Partitives and Distributivity

75

bit harder to account for, perhaps, and need a more detailed investigation than I can offer here.17 If floating quantifiers are really adverbials, one would also expect them to coordinate with other adverbials. While floating quantifiers do not tend to conjoin much, it is possible to come up with a few conjunctions that do not sound too bad: (39) a. De hinderen hebben allemaal en binnen 14 dagen het diploma gehaald. the children have all and within 14 days the diplom earned b. We komen of allemaal of helemaal niet we come either all or totally not 1 'We will either all come, or not at all

17.

Johnson (1992) notes that floating quantifiers may occur inside the VP in nonfinal positions: (i)

a. I put the bottles all on the table. b. I looked the numbers all up. c. I read the numbers all quickly. d. John believes the men all liars. e. John gave the men all a letter.

In this respect, floating quantifiers appear to differ from VP-adverbs such as quickly, which must occupy the sentence-final position. Johnson also notes that it is not enough if some material follows the floating quantifier: expressions which are arguable not VP-constituents do not accept an immediately preceding floating quantifier: (ii) a. *I met the boys all surely. b. *So many men met the boys all that they grew weary. c. *I met the boys all nude. (* on the subject predication reading) One may add to these observations that in Dutch, floating quantifiers may occur sentence-finally, but not following the VP. Sentence-final occurrence is shown in (iii): (iii) Ad kende de Studenten allemaal. Ad knew the students all while the impossibility of post-VP occurrence is shown by (iv): (iv) *Ad wilde de Studenten kennen allemaal Ad wanted the students know all (Sentence iv is OK when there is a heavy intonational break before allemaal and this expression is interpreted as an afterthought.)

76

Jacob Hoeksema

4.1. The principal-filter condition 4.1.1. Restricting quantifier floating by semantic filtering It would seem that it is harder on the adverbial theory to capture the local relation between the floating quantifier and the noun phrase from which, in the traditional transformational analysis, it has floated away. As Dowty and Brodie (1984) have argued, however, a proper formulation of the semantics of floating quantifiers might suffice to express all relevant properties of this relationship. In their analysis, a floating quantifier restricts the domain of a functional expression. A verb phrase, for instance, can be viewed as a mapping from generalized quantifiers to truth-values (cf. Keenan and Faltz (1985) for a detailed proposal). Adding a floating quantifier may restrict this mapping to a subset of the original class of quantifiers. They suggest that this subset is the class of principal filters. Principal filters are defined as follows: (40) Definition Let Q be a collection of subsets from the domain of discussion E. Then if there is a subset A of E, such that for all X in Q, A c= X, we say that Q is a principal filter, more precisely, the principal filter generated by A.18 In set notation: the principal filter generated by A is {X c Ε | A c= X}. In Barwise and Cooper (1981), principal filters are denoted by definite NPs and universally quantified NPs, such as the students, those of us who knew

18.

Dowty and Brodie (1984) use a slightly different definition: a principal filter is a collection of sets with a nonempty intersection. This rules out so-called non-proper principal filters which correspond to the powerset of E, the universe (because the supersets of the empty set include every subset of E). For universal quantifiers, which are commonly assumed to have no existential presuppositions (but see Verkuyl and De Jong (1985) for an opposing view), this means that they can only cooccur with floating quantifiers when their noun denotation is nonempty. In other words, the floating quantifier will add an existential presupposition, according to Dowty and Brodie's proposal. When that presupposition is violated, ungrammatically ensues. This may well be incorrect, but I won't pursue this matter here.

Floating Quantifiers, Partitives and Distributivity

77

Freud, we the people, all nations, every corner, each moment. The definition in (39) would shed light on the difference in acceptability between the examples in (41) below and those in (41') (all examples taken from Dowty and Brodie 1984)). (41) a. b. c. d.

John, Mary and Susan all left. John and Mary both left. The students all left. ?All students in my class must all turn in their exams on Friday.

(41') a. b. c. d. e.

*John, Mary or Susan all left. *John or Mary both left. *Few students all left. *No students all left. *At least five students all left.

The fact that the subject no students and the floating quantifier all are mutually exclusive now follows from the semantic properties of all as a verb-phrase modifier and not from the illformedness of a putative underlying string all (of) no students. An immediate advantage of this is that the grammaticality of sentences such as (41a) and (41b) is no longer a problem. Semantically, a conjunction such as John, Mary and Susan corresponds to a principal filter. The fact that there is no corresponding partitive *all of John, Mary and Susan is irrelevant. The local nature of the relation between floating quantifier and its host is also explained. By affecting the domain of a verbal predicate, the floating quantifier can only have effects on the direct arguments of that predicate. It cannot have similar effects on noun phrases that are not clause-mates. In the same fashion, the floating quantifier can have no effects on any modifiers of the verbal predicate, which explains observations in the literature (e.g. Seiter 1979) that modifiers are not suitable hosts for floating quantifiers. 19

19.

Some accounts, such as Sportiche's movement theory, would restrict things even further and exclude all cases where the host is not a noun phrase argument. On the whole, this is correct, although I am aware of one case, where the host is a prepositional phrase, thereby providing an additional refutation of Sportiche's theory. This concerns the floating quantifier allemaal in Dutch:

78

Jacob Hoeksema

In a similar way, superiority effects are captured. If a floating quantifier modifies a transitive verb, it may have effects on the choice of the direct object; if it modifies a verb phrase, it may affect the choice of the subject. However, it cannot combine with a verb phrase and still have effects on the selection of the direct object. This immediately accounts for the Dutch data below. (42) a. voordat ze allemaal ons vernederen before they all us humiliate 'before they all humiliate us' b. voordat ze ons allemaal vernederen before they us all humiliate 'before they humiliate us all' c. *voordai hij allemaal ons vernedert before he all us humiliates 'before he all humiliates us'20 If we view reflexives and reciprocals semantically as relation-reducers, that is, operators on predicates which affect their argument structure, the similarities in distribution between floating quantifiers and bound anaphors such as reflexives and reciprocals are easy to account for. Reciprocals are especially close to floating quantifiers, because they likewise restrict the domain of a function to a certain semantic class, in this case the class of plural quantifiers. The differences in distribution, which were discussed in section 2, are in fact harder to account for. Perhaps it is the fact that fronted

(i)

Waarover heeft hij allemaal zitten praten? where-about has he all sit talk 'What all has he been talking about?'

Note that it is also possible to have allemaal joined to the wh-word: (ii) 20.

Waar allemaal heeft hij over zitten praten?

For English, we must assume, in order to exclude (i), that the adverb occurs in a rightbranching structure: (i)

*He all humiliates us.

Otherwise, a bracketing as in (ii) would predict that (i) is possible. (ii)

He ((all humiliates) us)

Floating Quantifiers, Partitives and Distributivity

79

anaphors can also be analyzed either as arguments or as contained in arguments, while floating quantifiers are strictly modifiers, which accounts for the greater ease with which the anaphors can be fronted or extracted out of a subordinate clause. However, before such a claim can be accepted, it will be necessary to gain a deeper understanding of the factors influencing the fronting of adverbials. This problem is beyond the scope of the present paper, but see Szabolcsi and Zwarts (1990), among others, and the literature cited there, for discussion. 4.1.2. The empirical evidence for the principal-filter condition The principal-filter restriction proposed by Dowty and Brodie for hosts of floating quantifiers is reminiscent of a similar constraint proposed in Barwise and Cooper (1981) for partitive noun phrases. The noun phrase following partitive of is required to be a definite noun phrase. Definite noun phrases are defined by Barwise and Cooper as denoting proper principal filters. Despite similarities, the two classes of expressions are by no means identical. I have already mentioned the fact that conjunctions of singular terms may serve as the host to floating quantifiers, yet do not appear after partitive of.21 Before considering Dowty and Brodie's proposal in more depth, let us make a brief detour to look at some cases which I believe are only superficially a problem for the Dowty/Brodie approach, because they actually represent a different construction than the one we have been concerned with so far. Consider the examples below (examples a,b), where the initial constituent and apparent host of the floating quantifier is a bare noun and cannot denote a principal filter. These examples represent a construction which is archaic in English, but somewhat more alive in Dutch and fairly productive in German (cf. examples c,d below): (43) a. Friends have I none. b. But answer made it none (Hamlet, Act 1, Scene 2, line 215) 21.

This includes coordinations with as well as, judging from the following example: (i) his own restless bisexual bohemianism as well as the bitter tears of his films were both of them products of his chronically disturbed psyche (from: John Ardagh, 'Germany and the Germans', Penguin, p. 294)

80

Jacob Hoeksema

c. Bücher habe ich books have I d. Freunde hat sie friends has she

keine. none sehr viele. very many

Here the host of the 'floating quantifier' has to be indefinite (in most dialects of German, also bare, that is, without an overt determiner), compare (45). (44) *Die Bücher hatte ich keine. For these cases, I believe that an analysis in terms of movement (or some nonderivational equivalent thereof) is more nearly correct. This is not the place to go into the many difficult problems that arise, but it is enough for the purposes of this paper if it can be argued that the construction involved here is a different one from the regular floating quantifier construction. Of course, this is not hard to argue, since the fact that one construction became obsolete in English whereas the other one is alive and well suggests very strongly that they should not be viewed as the same. To mention one other difference, note that the host of the floating quantifier is typically topicalized in the construction exemplified in (43-44), whereas hosts in subject position are not usually acceptable, as the following German examples indicate: (45) *weil Bücher keine dem Herrn gewidmet waren because books none the Lord dedicated were Compare this with (46) weil diese Bücher alle dem Herrn gewidmet sind because these books all the Lord dedicated are It is not clear to me how movement theories of quantifier floating such as Sportiche's can make a proper distinction between regular quantifier floating as in (46) and the kind of quantifier split found in (43). What would prevent the element Bücher in (45) to move into subject position, while allowing it to move into the slot for topicalized elements? Standard devices such as case marking or theta-assigment can't be relied on since they would not distinguish between (45) and (46).

Floating Quantifiers, Partitives and Distributivity

81

Leaving aside now cases of quantifier split, let us take another look at the principal-filter restriction. It appears that this restriction is a good first approximation of the data, since it is compatible with the vast majority of cases. As for indefinite subjects, Dowty and Brodie allow them just in case they are specific indefinites, following earlier suggestions by Ladusaw (1982) regarding indefinites in partitive constructions. Thus for Dowty and Brodie, specific indefinites are semantically (and not just pragmatically) distinct from nonspecific indefinites. They give an example with a nonrestrictive relative clause, to force a specific interpretation: (47) Five contestants, who were selected as finalists by the judge yesterday, will all perform again tomorrow. To test this claim, I have done a little empirical investigation, checking through all occurrences of all in a 6 million word corpus collected mainly from postings on the various Internet bulletin boards. I found the following cases of floated^// with indefinite subjects (I discarded all cases where an indefinite subject clearly had a generic reading, but some doubtful cases remain): (48) a. Buildings, docks, vessels, and details of the Artie landscape are all clearly visible. b. The operation of the kiln during the trial burn was controlled and monitored far more carefully than under routine daily operation, when variations in waste type and quantity, human error, equipment malfunction, and combustion upsets all lead to increased emissions. c. Nations like Liechtenstein (24,000 people), Turks and Caicos Islands (7,000 people), and Tuvalu (6,000 people) all have representatives in the United Nations General Assembly. d. And a great many voices all said together ('like the chorus of a song,' thought Alice), 'Don't keep him waiting, child!' (Lewis Carroll) e. Unfortunately 4 starters, Brantley, Robinson, Henderson, and Minnesota recruit Ryan Wolf all fouled out in a span of two minutes (the Russians fouled out three players also but they had much more depth) and the Russians took over at the end. f. Horses, riders, people, were all blown about like ships at sea. g. Plaintiff demonstrated a probability of 1 in 4,000, by computing the chance that 12 consecutive mistakes would all fall against him.

82

Jacob Hoeksema

h. Unusual noises on the phone, intensifying whenever UFOs are mentioned, and voices breaking in on conversations, have all led many people to suspect that their phones are being tapped. i. Israeli historian Yehuda Bauer and Jewish-American historians like Raul Hilberg and Deborah Lipstadt all state that this anti-German hate story is untrue. j. Names like Ngozi (Blessingj, Obianuju (One who comes at the time of plenty), Nwa-amaka ("There is nothing as sweet as a child") are all popular girls names. k. Student demonstrations, guerrilla theatre, and strongfaculty support were all used to counteract the last attempt by the administration to use threats of eviction to modify the relationship between the coops and the university. 1. The reason is obvious: factionalism, policy problems, leadership problems have all rendered the ANC unready. Some of these cases are best viewed as involving specific readings. This probably includes indefinites with an added like (quite a few of the above examples are of this type). Certainly an expression such as nations like Liechtenstein, Turks and Caicos Islands, and Tuvalu appears more "specific" than an otherwise contextually equivalent expression such as miniature nations. In the case of (48e), it is even more obvious that we are dealing with a specific indefinite. In (48b) we might be dealing with a generic indefinite subject. Other cases, however, are genuine counterexamples. For example, (48g) can't be interpreted as being about 12 specific consecutive mistakes. The same is true for the subjects of (48d,f,h,k,l). In the same corpus, I found one example with each floating off an indefinite (and nonspecific) subject and two with both: (49) a. A mathematician, scientist, and engineer are each asked: "Suppose we define a horse's tail to be a leg." b. Early research results and practical experience both suggest that clarithromycin is much more promising than any of the standard treatments. c. How could it be, I wondered, that two seemingly upstanding, highly regarded people could both be speaking of such diametrically opposed scenarios?

Floating Quantifiers, Partitives and Distributivity

83

On the other hand, cases of all or each floating off regular quantifiers, such as no student or every administrator, were not found,22 and it is here that we find the part of Dowty and Brodie's condition which seems robustly confirmed by empirical evidence. I note, however, that even here, some cases appear to be better than others. In particular, I'd like to suggest that when no is used to quantify over groups, rather than individuals, the result seems better than expected, cp.: (50) a. b. c. d.

No two consecutive numbers are both divisible by two. No three consecutive numbers are each prime. *No students have each baked a pie. *No administrators have each/all made a difference.

Brame (1979: 134) offers the following sentence which is likewise problematic for the Principal Filter-condition: (51) Of the five boys, only John and Bill both shouted at each other simultaneously. Somewhat borderline are cases with two all's. Dowty and Brodie give them a question mark, and in the corpus, such examples are much rarer than one would expect if they were fully acceptable. On the other hand, I did find some examples of this type, which suggests they are not ungrammatical. (52) a. All the beautiful women you find in Bond movies were all drooling over these fat old guys with pot bellies and seventeen underchins! b. All the nice looking guest-hotels are all full c. all the DRUMS and SPACE segments I witnessed all had interesting moments It is probably the redundancy of the floating quantifier, together with a stylistic resistance to the double use of the same word in a single clause 22.

As a matter of fact, one such case was found, but it seems to me a likely error, because the intended reading appears to be one where the subject of the VP will both be fine is understood to be they rather than neither of them. (i)

He says neither of them want Kim to be in jail any longer than is necessary and will both be fine when Kim is home.

84

Jacob Hoeksema

which is responsible for the low numbers of such cases and the intuition that these sentences are less than perfect. Combined, the data suggest the following explanation: When the subject is a quantified NP, with a distributive quantifier, there is no need for a distributive quantifier in the VP. In the case of 'semi-distributive'23 determiners such as all, the result is only a mild deviation from the norm. When the determiner is strictly distributive, such as each, the addition of a floating quantifier makes the sentence more clearly unacceptable. (53) a. Each student had a proposal. b. *Each student each/all/both had a proposal. Even here, it seems that there are some subtle differences. It seems that floating each is somewhat better in (53b) than floating all·, while the first strikes one as overly redundant, the use of all gives the feeling of a clash between the individuals quantified over by each student and the groups required for the proper use of all.24 23.

24.

See Dowty (1986) for discussion of the semantic properties of all. This determiner distributes certain properties over each member of a group, but at the same time it allows for group predicates, unlike true distributive quantifiers. Thus, All boys gathered in the park is grammatical, whereas Each boy gathered in the park is semantically deviant. These examples should get better when instead of student a collective noun is used such as jury or group. With definite determiners, I have found some cases of floated all in my corpus: (i)

a. The jury all wrote down on their slates, 'SHE doesn't believe there's an atom of meaning in it,' but none of them attempted to explain the paper. (Lewis Carroll, Alice in Wonderland)

b. The jury all brightened up again. c. The jury all looked puzzled. d. The flora is all continuously changing as we watch and the land continues to move much as rolling waves on the ocean do. e. Do staff all have CPR/First Aid training? f. Today's managerial and scientific elite of the Biosphere 2 project can all be traced directly back to John Allen's so-called "Theatre of All Possibilities": Along the lines of example f, say, we could construct similar sentences with quantifiers in subject position, which are more acceptable than they would have been with noncollective nouns. (ii) a. Neither elite can be all traced back to the Theatre of All Possibilities. b. Can any/each elite be all traced back to the Theatre of All Possibilities?

Floating Quantifiers, Partitives and Distributivity

85

To conclude, it is evident that Dowty and Brodie's filtering approach to the distribution of floating quantifiers is too restrictive. It rules out fairly infrequent but acceptable use of floating quantifiers with nonspecific indefinites and the interactions of quantified subjects with quantifier floating which it correctly rules out can also be explained by means other than the filtering approach. Apart from these empirical concerns, there is also a theoretical reason to favor a slightly different approach. 4.1.3. Scope and higher-order types The higher-order types (VPs take their denotation in the type ,t>,t> of functions from generalized quantifiers to truth-values) which Dowty and Brodie (1984) postulate for verb-phrase denotations can also be found in Keenan and Faltz (1985) as well as Montague's UG (Montague 1970). On a descriptive level, where one is concerned primarily with the correct statement of the truth-conditions of the sentence, there is no harm in using higher-order types instead of lower-order ones. However, from a more theoretical point of view, they raise the question if these types are ever crucially needed. In the case of noun phrases, some could be interpreted in the simple type of entities, whereas others crucially take their interpretation in the type of generalized quantifiers. Is there similar evidence that some VPs crucially have to be interpreted as sets of generalized quantifiers, one may ask. The answer to this question seems to be "No", at least if one disregards for the time being predicates with floating quantifiers. For example, if we could find a predicate which only applies truthfully to generalized quantifiers that are closed under intersection we would have an argument that some predicates must crucially take their interpretation in type ,t>,t>. However, no such predicate appears to exist. Keenan and Faltz (1985) impose a number of conditions on VPdenotations which serve to constrain the possible VP-denotations to only those which can be derived from denotations in type by type-lifting. As Van Benthem (1987) has stressed, it then becomes more attractive to view as the basic type of VPs, because under that assumption it follows automatically, without stipulating any meaning postulates or semantic universale, why there are no essential ,t>,t> denotations for VPs. And indeed Keenan's more recent work has dropped the higher-order types for predicates (cf. Keenan 1987).

86

Jacob Hoeksema

For the treatment of monotonicity properties, it is necessary to view subject and object as functions over one-place and two-place predicate denotations, respectively. To show this, I need a few definitions. First, and elementary, let entailment be a relation ' Nobody ate (59) John likes or respects nobody —> John likes nobody In each case we may replace a disjunction by one of its members. This is possible only in m o n l environments. In spite of common usage, which calls the subject and the direct object the arguments of the verb, we have to consider them functors, which take verbal elements as their arguments. More precisely, they are functors which send η-place predicates to n-1 place predicates. To return now to our main topic, we see that the higher-order treatment of Dowty and Brodie (1984) implies a view of function-argument structure which goes against our current understanding of monotonicity phenomena. Note also that negative floating quantifiers do not trigger polarity items in subject position, something which one would expect on the Dowty/Brodie analysis, cf.: (60) a. *The parents of any students were none of them very pleased. b. None of them knew the parents of any students. This concludes our argument that for purposes of scope, it is best not to give the VP a higher-level interpretation which makes it a functor with scope over the subject. For further discussion and additional problems which arise if NPs are not treated as functors over η-place predicates, I refer the reader to Keenan (1987) and Hoeksema (1989).

Floating Quantifiers, Partitives and Distributivity

89

4.2. The semantics of floating quantifiers 4.2.1. The operators of Van der Does (1992) Van der Does (1992), partly following earlier proposals by Roberts (1987), Link (e.g. 1991) and others, introduces a number of operators which model the meaning of various natural language floating quantifiers. These operators are of type « < e , t > , t > , « e , t > , t » , the type of functions which send collections of sets of individuals to collections of sets of individuals. The type itself, which is usually reserved for the generalized quantifiers which serve as NP-denotations, is used by Van der Does to provide interpretations for plural VPs, in order to capture the intuition that a plural VP is interpreted not a property of individuals but as a property of sets.27 The operators (Van der Does calls them "modifiers", a term which I will reserve for their natural-language counterparts) are called α, τ, δ, and π, and are defined as follows: (61) Definition of four operators. a := m Y l ( Y ) & |Y| = 1

(pure) atomic

τ := m Y I ( Y ) & |Y| > 1

collective

6 := m Y . A T ( Y ) c X

distributive

π := λΑΧΥΥ c u X

partaking in

The α-operator "selects the pure atoms or individuals from the denotation of a VP" (Van der Does 1992: 61). According to Van der Does, there does not appear to be an exact English counterpart to a , although it might seem at first blush that expressions such as alone, on his/her own, all by him/ herself would have the same force. Van der Does warns against any such presumption, explaining that when Perdeck buys a book all by himself, we 27.

Hoeksema (1983) on the other hand, treats VPs as denoting functions in , but has a different conception of the basic type : besides individuals, also groups (that is, sets of individuals, or sets of (sets of)* individuals) are taken to be of type .

90

Jacob Hoeksema

not only understand this to entail that the singleton {perdeck} is in the set of entities which buy a book, but also, that he is not a member of a larger set which also buys that book. On this matter, it seems to me, Van der Does is wrong. The fact that Perdeck cannot be part of a larger group which buys a book if he buys the book all by himself, while hard to deny, must not be ascribed to the semantics of the adverbial expression all by himself, but rather to the lexical semantics of the verb buy. Like other verbs of its kind, it resists multiple agents (at least per buying event): if a buys a book, then b cannot also buy it, unless we have a sequence of buying events. If we consider a different type of predicate, such as weigh more than 300 pounds, the situation changes. Clearly, Perdeck could weigh more than 300 pounds on his own, while belonging to a club of weight-watchers whose collective weight is also more than 300 pounds. However, there is a more compelling reason why alone or on his own is not quite synonymous with a : while a is purely quantiideational, alone, like its counterpart together, also has an important spatio-temporal meaning component (for a proposal on the proper treatment of together, see Lasersohn (1990)). Thus when we hear that Brad sleeps alone, we not only understand that {Brad} is in |sleep|, but also that his bedpartner Janet is spatio-temporally removed from him. If we ignore this aspect, we can view alone etc. as the English counterparts to a. 28 The τ-operator marks predicates for collectivity. Again, there is no precise counterpart to τ in English, although together could be used if one ignores its spatio-temporal aspects or in contexts where space/time plays only a minor role. Thus two writers can write a novel together without having to overlap in either space or time.

28.

Perhaps a closer analogue, in English, to α than alone or all by himself is the use of the singular. Van der Does (1992: 61) states: "One should be cautious, though, in using α to capture the syntactic number of the VP. In (5d) the VP is singular, but the predication is collective." Example (5d) is: (5) d. The quartet makes music. One could quibble here and treat the quartet as an individual similar in ontological status to Perdeck (cf. Landman 1989 for discussion). If one does not like this, one could move to those British dialects which would assign a plural to the verb in (5d) and say that singular number in those dialects corresponds to a . However, one should note that the use of number in these dialects is somewhat haphazard and influenced by matters of animacy.

Floating Quantifiers, Partitives and Distributivity

91

The δ-operator, borrowed from Link and defined here in terms of AT, the set of atoms (within , AT is the set of singletons), can be expressed in English by means of the floating quantifier expression each. Often, natural language leaves 6 unexpressed, because the meaning of the verb or verb phrase is such that a collective reading is ruled out or because the subject adds distributive force (cf. the discussion of the principal-filter constraint above). Thus in Van der Does' example below, the use of each appears somewhat redundant: (62) The Mitarios each admire the Montagues. The effect of each here is to force each singleton {x} such that χ is a Mitario to be included in |admire the Montagues|. The π-operator expresses the notion of partaking in. Van der Does illustrates this operator with the following example: (63) The Mormons spread the Word. Besides a distributive and a collective reading, this example also has a mixed reading, such that various groups (perhaps partly overlapping) of Latter Day Saints are involved in missionary activities. It is this reading that the operator π is intended to capture. Van der Does does not suggest a natural-language counterpart for π. The operators defined by Van der Does have a great many noteworthy formal properties. For example, they are all idempotent, which is to say that for any of these operators 0(0(X)) = O(X). (Hence O(X) is a fixed point for O.) This property might explain why iteration of the same floating quantifier is not found in natural language: If iteration does not affect the meaning of the predicate, then it is more economical to apply the operator only once. Combinations of operators are often equivalent to just one of the operators. Thus Van der Does remarks that the composition of δ and π equals π (i.e. that function h such that h(X) = δ(π(Χ)) equals π) while the composition of π and δ equals δ (hence the innermost operator "wins" in these combinations). Perhaps more interesting for us is the interaction of δ and τ. Here we note the following: (64) a. δ(τ(Χ)) = λΖ[ΑΤ(Ζ) < λΥ[Χ(Υ) Λ |Y| > 1]] = {0} b. τ(δ(Α9) = τ(Χ)

92

Jacob Hoeksema

From (64a) we see, that applying δ to the result of applying τ to an arbitrary set X leaves us with a trivial result. On the other hand, if we apply both operators in reverse order, the result is equivalent to just applying τ. This goes a long way toward explaining why the distributive operator each and the collective operator together normally do not occur together: (65) a. *The students each lifted the sofa together. b. *The students together each lifted the sofa. 4.2.2. Tinkering with the types Sometimes, we want to distribute a predicate over groups which are members of a collection, without distributing it over the members of those groups themselves. Consider for instance the following example: (66) The Beatles and the Stones both have recorded this song. Intuitively, it is appealing to interpret the subject of this example as a pair {b,s}, where b and s are themselves groups of individuals. What both does in this example is to distribute the predicate have recorded this song over these two groups, without entailing that each individual Beatle and each individual member of the Rolling Stones also have recorded the song. Lest it be thought I make the example up, let me present a collection of examples which I culled from my corpus. In each case, the total number of individuals or events involved is more than two, but there is a two-way partition in the subject denotation corresponding to the conjuncts. (67) a. In the church, men and women are both called to minister to the saints, but God only permits men to teach and exercise authority over other men. b. Fox executives and "Home Alone"producer John Hughes have both indicated they want to do a sequel. c. Hitler's successes and later downfall were both dependent on making illogical (and therefore unexpected) military moves. d. The mighty and the humble are both counted as one.

Floating Quantifiers, Partitives and Distributivity

93

Sentences of this kind are not so easy to capture in an approach such as the one offered by Van der Does. If the VP is of type , then it should take as its argument an element of type , which is to say, a set of individuals. Taking, for the moment, the groups b and s to constitute sets, not individuals, then one is forced to interprete the conjunction The Beatles and the Stones as the union of b and s (see Schwarzschild 1990 for a spirited defense of such a treatment). However, after taking the union of b and s, there is no obvious way in which the floating quantifier both could distribute the predicate over two elements: instead of the two elements b and s, we have the eight members of b υ s. One might suppose at this point that a contextually-provided partition of the subject denotation could provide us with the means to interpret cases like the above (cf. e.g. Gillon 1987, Schwarzschild 1990). However, there is no evidence that such an approach is correct. Quite on the contrary. If the context were to provide a partition, one would expect discourses such as the following to be acceptable: (68) The animals were separated by sex. #The pigs, the cows and the sheep were both sold to out-of-towners. Given that the first sentence sets up a partition of the animals in two groups, male and female animals, the second sentence should be interpretable as saying that both the male and the female pigs, cows and sheep were sold to out-of-towners, but the phrasing is definitely odd. The three-way partition of the animals suggested by the syntactic form of the subject clearly overrides the effects of context Taking this as a hint, I propose to adopt my earlier treatment in Hoeksema (1983b, 1988), in which the domain of quantification, where the expressions of type find their denotation, is enlarged to contain not just individuals, but also groups of individuals, groups of groups of individuals and so on.29 Predicates can then retain their familiar type 29.

More precisely, we can derive the domain of quantification Ε from some set of individuals I by iterated finite group formation in the following manner, due to Johan van Benthem (cf. Hoeksema 1983, note 1): (i)

LetE0 = I

(ii)

E n+| = E n uPOW a2 (E n ) (where POW a2 (X) denotes the set of all subsets of X with cardinality 32)

(iii) Ε = Εη Εη v '

94

Jacob Hoeksema

, rather than take a denotation in . The latter type is now reserved once again for non-referring or quantificational NPs (cf. Partee and Rooth 1983, Hoeksema 1988). Floating quantifiers have the type of adverbs, «e,t>,