Something Out of Nothing: the Semantics and Pragmatics of Implicit Quantification [1 ed.] 9789004432598, 9789004431485

Some sentences contain no overt quantifier, yet are interpreted quantificationally, e.g., Plumbers are available (entail

217 36 869KB

English Pages 183 Year 2020

Report DMCA / Copyright

DOWNLOAD PDF FILE

Recommend Papers

Something Out of Nothing: the Semantics and Pragmatics of Implicit Quantification [1 ed.]
 9789004432598, 9789004431485

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

Something out of Nothing

Current Research in the Semantics/Pragmatics Interface Series Editors Klaus von Heusinger (University of Cologne) Ken Turner (University of Brighton)

Editorial Board Nicholas Asher (Université Paul Sabatier) Johan van der Auwera (University of Antwerp) Betty Birner (Northern Illinois University) Ariel Cohen (Ben Gurion University) Paul Dekker (University of Amsterdam) Regine Eckardt (University of Constance) Markus Egg (Humbolt University Berlin) Donka Farkas (University of California, Santa Cruz) Brendan Gillon (McGill University) Jeroen Groenendijk (University of Amsterdam) Yueguo Gu (Chinese Academy of Social Sciences) Larry Horn (Yale University) Yan Huang (University of Auckland) Manfred Krifka (Humboldt University Berlin)

Chungmin Lee (Seoul National University) Claudia Maienborn (University of Tübingen) Alice ter Meulen (University of Geneva) Jaroslav Peregrin (Czech Academy of Sciences and University of Hradec Králové) Allan Ramsay (University of Manchester) Rob van der Sandt (Radboud University Nijmegen) Kjell Johan Sæbø (University of Oslo) Robert Stalnaker (Massachusetts Institute of Technology) Martin Stokhof (University of Amsterdam) Henk Zeevat (University of Amsterdam) Thomas Ede Zimmermann (University of Frankfurt)

volume 38

The titles published in this series are listed at brill.com/crispi

Something out of Nothing The Semantics and Pragmatics of Implicit Quantification

By

Ariel Cohen

LEIDEN | BOSTON

Library of Congress Cataloging-in-Publication Data Names: Cohen, Ariel, author. Title: Something out of nothing : the semantics and pragmatics of implicit quantification / by Ariel Cohen. Description: Leiden ; Boston : Brill, 2020. | Series: Current research in the semantics/pragmatics interface, 14727870 ; 38 | Includes bibliographical references and index. Identifiers: LCCN 2020020265 (print) | LCCN 2020020266 (ebook) | ISBN 9789004431485 (hardback) | ISBN 9789004432598 (ebook) Subjects: LCSH: Grammar, Comparative and general–Quantifiers. | Semantics (Philosophy) Classification: LCC P299.Q3 C55 2020 (print) | LCC P299.Q3 (ebook) | DDC 401/.4301–dc23 LC record available at https://lccn.loc.gov/2020020265 LC ebook record available at https://lccn.loc.gov/2020020266

Typeface for the Latin, Greek, and Cyrillic scripts: “Brill”. See and download: brill.com/brill‑typeface. ISSN 1472-7870 ISBN 978-90-04-43148-5 (hardback) ISBN 978-90-04-43259-8 (e-book) Copyright 2020 by Ariel Cohen. Published by Koninklijke Brill NV, Leiden, The Netherlands. Koninklijke Brill NV incorporates the imprints Brill, Brill Hes & De Graaf, Brill Nijhoff, Brill Rodopi, Brill Sense, Hotei Publishing, mentis Verlag, Verlag Ferdinand Schöningh and Wilhelm Fink Verlag. Koninklijke Brill NV reserves the right to protect this publication against unauthorized use. Requests for re-use and/or translations must be addressed to Koninklijke Brill NV via brill.com or copyright.com. This book is printed on acid-free paper and produced in a sustainable manner.

To Keren, the light of our lives



Where got I that truth? Out of a medium’s mouth. Out of nothing it came, Out of the forest loam, Out of dark night where lay The crowns of Nineveh. William Butler Yeats, Fragments



Contents 1 What You See Is Not Always What You Get

1

2 Reinterpretation Mechanisms 4 1 Reinterpretation 4 2 Type-Shifting 5 2.1 The Nature of Type-Shifting 5 2.2 Type-Shifting and Scope 10 2.3 Type-Shifting and Gradable Adjectives 12 2.4 Type-Shifting and Discourse Referents 15 2.5 Blocking 16 2.6 Type-Shifting or Sort-Shifting? 17 2.7 Summary 18 3 Predicate transfer 18 3.1 The Nature of Predicate Transfer 18 3.2 “Noteworthy” Properties 21 3.3 The Transfer Function 22 3.4 Intensional Properties 23 3.5 When Does Predicate Transfer Apply? 25 3.6 Scope 29 3.7 Summary 31 3 Bare Plurals 33 1 The Ambiguity of Bare Plurals 33 2 A Non-quantificational Theory? 33 2.1 Carlson (1977) 33 2.2 The Existential Reading 34 2.3 The Characterizing Generic Reading 36 3 Quantificational Accounts of English Bare Plurals 39 3.1 Points of Agreement 39 3.2 Characterizing Readings 42 3.3 Existential Readings and Blocking 43 4 Kinds or Properties? 46 4.1 An Argument for Properties—Unmotivated Type-Shifting 46 4.2 An Argument for Kinds—Bare Singulars 51 4.3 English Indefinite Singulars 52 4.4 An Argument for Kinds—Scopally Ambiguous Bare Plurals 56 5 A Synthesis 58 6 Deciding between the Theories 62

viii 4 Beyond English Bare Plurals 64 1 Italian Bare Plurals: Direct Kind Predication 64 1.1 The Data 64 1.2 The Theory 65 2 Definites and Type-Shifting 66 2.1 The Interpretation of Definiteness 66 2.2 Definites as Kind-Referring 71 2.3 Definite Plurals 75 2.4 Definites in Italian 76 3 Italian BPs: Existential Readings 77 4 Italian BPs: Characterizing Generics 78 4.1 Are Characterizing Generics Possible? 78 4.2 The Heaviness Requirement 79 4.3 A Comparison with English Indefinite Singulars 4.4 Generic Italian BPs Are Normative 81 4.5 Normative English BPs 83 5 An Account of Italian BPs 84 6 Hungarian Bare Plurals 85 7 Bare Singulars 86 7.1 Bare Singulars and Type-Shifting 86 7.2 Existential Readings 87 7.3 Reference to Kinds 92 7.4 Characterizing Readings 96 8 Conclusion 98 5 Generics and Habituals 100 1 Generics and Scope 100 1.1 Negation 100 1.2 Opaque Contexts 105 1.3 Transparent Contexts 106 2 Habituals and Scope 106 3 Types of Explanation 108 3.1 The Facts so far 108 3.2 Number 109 3.3 A Verb-Level Null Operator 110 3.4 A Syntactic Account? 110 4 Reinterpretation Mechanisms Revisited 112 4.1 The Introduction of Implicit Quantifiers 112 4.2 Null Categories in Syntax 113 4.3 Null Categories in Semantics 116

contents

80

ix

contents

5 6

7

The Generic Quantifier 118 Generics 120 6.1 Analysis 120 6.2 Formalization 124 Habituals 127 7.1 Analysis 127 7.2 The Imperfective 129 7.3 Formalization 130 7.4 Habituals and Bare Plurals 7.5 The Progressive 133

131

6 Iterativity 135 1 Durative Adverbials 135 2 The Origins of Iterativity 136 3 Iterativity and Scope 138 4 Alternative Explanations 139 4.1 Iterativity as an Inherently Narrow Scope Operator? 140 4.2 Iterativity as a Verb-Level Operator? 140 5 Iterativity as a Quantifier 142 7 The Nature of Implicit Quantification 145 1 Two Implicit Quantifiers 145 1.1 The Existential and the Generic Quantifiers 145 1.2 Are Generics Modal Universals? 145 2 A Preference for Inference 147 2.1 Unmarked Forms and Unmarked Interpretations 147 2.2 Bare Forms and Stereotypicality 148 2.3 Bare Forms and Non-stereotypicality 150 2.3.1 Generic Bare Plurals 150 2.3.2 Existential Bare Plurals 154 2.4 Inference and Preference 155 References 157 Name Index 168 Subject Index 170

chapter 1

What You See Is Not Always What You Get It has been said that “[n]o idea is older in the history of linguistics than the thought that there is, somehow hidden underneath the surface of sentences, a form or a structure which provides a semantic analysis and lays bare their logical structure” (Seuren 1973:528). The exact nature of this hidden form or structure has been hotly debated: it has been referred to by various names, ranging from the semantic logical form to the syntactic Deep Structure, but there seems to be quite a strong intuition that there has to be something there. There is more to language than just the sounds you hear or the squiggles on paper you see. Yet how do we find this elusive underlying structure? Chomsky (1973) had a remarkable idea: there are traces of it in the surface structure. So, surface structure is not as simple as its name may suggest; it contains more than we can hear. There are phonologically null elements in surface structure, which we cannot hear, yet are important for our understanding of the utterance. This proposal proved to be very fruitful in linguistic research, especially in syntax, but one can find the influence of trace theory in semantics as well. Of particular interest to this book are cases where a sentence contains no overt quantifier, yet its interpretation is quantificational. For example, consider the following two sentences: (1) a. Dogs are barking outside right now. b. Dogs are intelligent. These sentences contain no overt quantifier. However, their interpretation appears to imply quantification. Although (1a) contains no existential quantifier, it seems to be saying that some dogs are barking outside right now. The interpretation of (1b) is considerably less clear, but it seems to imply that a significant number of dogs are intelligent. In such cases, a common view is that the sentences contain a phonologically null quantifier; we can’t hear it, but it is there all the same, and it affects the interpretation. In this book I argue that this common view is, in fact, wrong. While the interpretation of sentences such as (1) is, indeed, quantificational, the sentences do not contain any quantificational element. Quantification results as a consequence of reinterpretation of the sentences. Moreover, the reinterpretation

© Ariel Cohen, 2020 | doi:10.1163/9789004432598_002

2

chapter 1

mechanism that is responsible for (1a) is fundamentally different from that responsible for (1b), resulting in different properties of these constructions. This book is structured as follows. In the next chapter I consider two types of reinterpretation mechanism that may introduce quantifiers: semantic typeshifting and pragmatic predicate transfer. I argue that these mechanisms have different properties, and have different testable implications, particularly with regard to scope: quantifiers introduced by type-shifting must take narrow scope, while those introduced by predicate transfer are generally scopally free. Chapter 3 is concerned with the interpretation of bare plurals in sentences such as those in (1). I consider three influential theories of bare plurals: one non-quantificational theory (Carlson 1977), and two quantificational ones (Chierchia 1998 and Krifka 2004). According to these three theories, the denotation of the bare plural in both (1a) and (1b) is the same. They disagree, however, on what this denotation is: Carlson and Chierchia propose that, in both cases, the bare plural denotes a kind, while Krifka argues that it denotes a property. Both Chierchia and Krifka agree that, when necessary, the denotation of the bare plural can shift from a kind to a property or vice versa, following an application of type-shifting operators. I argue that a deeper consideration of Chierchia’s and Krifka’s arguments shows that the arguments supporting one theory apply only to existential readings, and the arguments supporting the other—only to generic readings. I therefore suggest a synthesis of Chierchia’s and Krifka’s theories, according to which the denotations of bare plurals in existential and generic sentences are different: the subject of (1a) denotes a property, while the subject of (1b) refers to a kind. Type-shifting still plays an important role in the proposed synthesis, relating the two denotations of bare plurals. Chapter 4 extends the discussion of bare plurals to other languages, to definites, and to bare singulars. It is argued that the main claim, namely that generic readings require reference to kinds, while existential readings involve properties, is maintained throughout the constructions and languages discussed. Chapter 5 is concerned with the generation of the generic quantifier in generics and habituals. I demonstrate that, despite their conceptual similarities, generics and habituals behave rather differently, in particular with respect to scope: habituals take narrow scope only, whereas the scope of generics is generally free. I consequently argue that although in both cases the same quantifier is generated, the mechanisms that introduce it are different: type-shifting in the case of habituals, whereas predicate transfer is responsible for generics. The thesis discussed in this chapter gives rise to a prediction: implicit quantifiers that are introduced by type-shifting receive narrow scope only, whereas those that are introduced by predicate transfer will have generally free scope.

what you see is not always what you get

3

I test this prediction in chapter 6, which is concerned with durative adverbials. Durative adverbials may receive an iterative interpretation, and there are good reasons to suppose that this interpretation is generated by type-shifting. I demonstrate that, as predicted, iterativity can receive narrow scope only. These chapters are concerned with the behavior of two implicit quantifiers: existential and generic. In the conclusion of the book I consider the reason why only those two types of quantifier can be produced by reinterpretation mechanisms. I argue that the main criterion is the ability of these quantifiers to give rise to useful inferences. I demonstrate that both quantifiers have implications beyond their truth conditions, and that, as a consequence, inferences based on these quantifiers are particularly useful for the purposes of the speaker and hearer. Thus, implicit quantifiers are not merely phonologically null versions of explicit ones: they are generated by special means, and they have special uses.

chapter 2

Reinterpretation Mechanisms 1

Reinterpretation

Ideally, a semantic theory should translate a disambiguated natural language expression into an expression in some logical language, where it can be evaluated and assigned a denotation. However, it turns out that this ideal picture cannot be maintained. Even if an expression is disambiguated and translated into some logical expression, this is not, in general, sufficient to determine its meaning, and additional, semantic or pragmatic mechanisms are necessary. I will refer to such phenomena, following Egg (2005), as cases of reinterpretation. One note of clarification is in order: reinterpretation, as discussed in this book, relates to competence, not performance. That is to say, its goal is to capture various semantic and pragmatic properties of natural language, but no claim is made about the cognitive processes that are involved in the generation of utterances by speakers or their understanding by recipients. One well known example is (2), as said by one server to another in a restaurant. (2) The ham sandwich is getting restless (Nunberg 1977). The literal meaning of (2) is clearly inappropriate here; the sentence needs to be reinterpreted as saying, not that the ham sandwich is getting restless, but that the customer who ordered the ham sandwich is getting restless. Another case of reinterpretation involves sentences like (3). (3) Elizabeth sneezed for two months. Again, we do not interpret (3) as saying that Elizabeth produced a sneeze that lasted for two months; rather, we interpret it iteratively, meaning that Elizabeth sneezed many times throughout a period of two months. There are many more such cases of reinterpretation, and we will see some of them throughout this book. However, I should emphasize that I do not intend this book as an overview of reinterpetation phenomena in general, and the theories accounting for them. In fact, it must be noted that there is no widespread agreement on what phenomena, if any, ought to be considered cases of reinterpretation.

© Ariel Cohen, 2020 | doi:10.1163/9789004432598_003

reinterpretation mechanisms

5

How does reinterpretation work? How do we produce the intended meaning in cases like (2), (3), and others like them? On the face of it, it would be desirable to have a uniform account of all cases of reinterpretation—indeed, such a theory is proposed by Egg (2005). However, in this book I will argue that a unified view is misguided. In particular, I will examine several cases where reinterpretation results in the introduction of a quantifier: existential bare nouns, generics, habituals, iteratives. These phenomena have different properties, especially with respect to the scope of the quantifier, and I will show that these properties fall out of the mechanism used to introduce the quantifier. Specifically, I will discuss two reinterpretation mechanisms: type-shifting and predicate transfer. They have very different properties, and, consequently, the quantifiers they introduce behave differently. Let us consider each one in turn.

2

Type-Shifting

2.1 The Nature of Type-Shifting In recent years, type-shifting has become ubiquitous in semantic theories. Rare indeed is the theory that does not make some use, in one way or another, of some type-shifting operator. Before seeing how type-shifting works, let us consider first why it occurs at all. Why do we shift an expression from one type to another? The answer is that type-shifting applies when there is a type mismatch. In other words, we get a logical form that is semantically ill-formed: the elements of the expression are not of the right types to combine with each other. For example, some predicate requires an argument of type e, but its argument is actually of type ⟨e, t⟩. In such cases, the type of at least one of the offending elements is shifted to the appropriate type. Type-shifting was first carefully introduced into semantics by Partee and Rooth (1983), and they were led to this conclusion after studying the behavior of conjunction. Let me briefly illustrate their argument. Take an extensional transitive verb like buy, as it occurs in a simple sentence like (4a). Plausibly, the logical form of (4a) is (4b). (4) a. John bought War and Peace. b. buy(j, w)

6

chapter 2

What is the type of buy? Let us start from scratch, and consider first the type of its direct object. At this point, let us try to make things as simple as possible (if necessary, we will complicate them later); thus, we will ignore intensionality, and take the simplest possible type that does the job—the type of individuals, e. Once the verb combines with its object, the result is a VP that requires a subject, just like an intransitive verb. Hence, we can say that the type of buy is a function from the type of individuals to the type of intransitive verbs. If the type of intransitive verbs is simply ⟨e, t⟩, we get the following as the type of an extensional transitive verb: (5) ⟨e, ⟨e, t⟩⟩.

This works nicely for (4a), where the object is simply a name. What if it is a quantifier, as in (6)? (6) John bought a new coat In this case, quantifying in (or some other scoping device) will apply; the variable that is left in place of the quantifier will, again, be of type e, and we will get (7) as the logical form of (6). (7) ∃x(new-coat(x) ∧ buy(j, x))

So far so good; but this type of analysis will not work for intensional verbs like need. When the object is a quantifier, we can quantify in as before—this results in the de re reading. However, there is another option: to have the whole generalized quantifier be the argument of the verb, resulting in the de dicto reading. Thus, the logical form of (8a), under the de dicto reading, is (8b). (8) a. John needed a new coat b. need(j, λP.∃x(new-coat(x) ∧ P(x)))

If, again, we ignore intensionality,1 we can take the type of a generalized quantifier to be ⟨⟨e, t⟩, t⟩. Once need combines with this object, the result is of the type of an intransitive verb, as before. Hence, the type of need is:

1 It may seem perverse to discuss the type of intensional verbs while ignoring intensionality, but this is not really so: all that matters for our concern here is that extensional and

reinterpretation mechanisms

(9) ⟨⟨⟨e, t⟩, t⟩, ⟨e, t⟩⟩.

7

So, extensional and intensional verbs are of different types. However, we know that they can be conjoined, as in the following example: (10) John needed and bought a new coat. Normally, only elements of the same type can be conjoined. How, then, is (10) possible? Montague (1973) proposes a well known solution. According to him, individuals are of the type of generalized quantifiers: a set of properties. Thus, the term War and Peace does not denote the individual book w, but the set of properties that w has: λP.P(w). Hence, the logical form of (4a) is actually (11), rather than (4b). (11) buy(j, λP.P(w))2 Similarly, the logical form of (6) would actually be (12) rather than (7). (12) buy(j, λP.∃x(new-coat(x) ∧ P(x)))

Montague’s suggestion has the effect that extensional verbs have the same type as intensional verbs, hence solving elegantly the puzzle of their conjunction. The price we have to pay is the abandonment of the intuitively satisfying idea that names denote individuals. Most people would be willing to pay this price, provided that the solution works. But does it? Partee and Rooth claim that the answer is no. They argue that Montague’s proposal does not cover the scopal behavior of conjunctions. Consider, first, a conjunction of two intensional verbs, e.g. want and need. The logical form of the conjoined predicate is: (13) λP.λx.(want(x, P) ∧ need(x, P))

intensional verbs differ in their respective types, and this fact can be demonstrated even without intensionality. 2 Actually, the subject, too, ought to be λP.P(j) rather than j, but I am omitting the detailed treatment of the subject for the sake of clarity.

8

chapter 2

We can combine this conjoined predicate with an object and a subject, and get a sentence such as (14a), whose logical form is (14b). (14) a. John wanted and needed a new coat. b. want(j, λP.∃x(new-coat(x) ∧ P(x))) need(j, λP.∃x(new-coat(x) ∧ P(x)))



Note that the conjunction has wide scope over the existential. This means that (14a) ought to be equivalent to (15), as indeed it is. (15) John wanted a new coat and John needed a new coat. But what if we conjoin two extensional verbs, as in (16)? (16) John bought and put on a new coat. Sentence (16) is not equivalent to (17) John bought a new coat and John put on a new coat. While (16) requires that John put on the same coat that he bought, (17) allows John to put on a different coat, e.g. one that he had received as a gift. If, as Montague proposes, extensional verbs had the same type as intensional verbs, they would have the same logical forms, and we would predict again conjunction to take wide scope, and the equivalence of (16) and (17). Partee and Rooth conclude that Montague’s approach faces a serious difficulty here. Crucially, if we maintained the original, simple picture, where extensional verbs are just relations between two individuals, the empirical facts would be accounted for. Thus, the logical form of bought and put on would be: (18) λy.λx.(buy(x, y) ∧ put-on(x, y)) Now, the logical form of (16) would be:

(19) ∃y(new-coat(y) ∧ buy(j, y) ∧ put-on(j, y))

Clearly, this is not the logical form of (17), hence the non-equivalence of the sentences. We seem to be facing a dilemma. The acceptability of conjunctions of extensional and intensional verbs leads us to expect that they have the same type;

reinterpretation mechanisms

9

but conjunctions of extensional verbs with extensional verbs, and intensional verbs with intensional verbs, lead us to believe that their types are different. It is as if extensional verbs sometimes have one type, and sometimes another.3 In fact, this is precisely what Partee and Rooth propose. According to them, expressions have the simplest type that they can. Thus, the type of extensional verbs is really a relation between individuals, while intensional verbs denote relations between an individual and a generalized quantifier. When one type of verb is conjoined with another, the type of the extensional verb is shifted to that of the intensional verb. Specifically, instead of keeping the relation buy, we would shift it to (20) SHIFT(buy) =def λQ.λx.Q(λy.buy(x, y)).

Note that shifting changes the type of an expression, but does not change its meaning: SHIFT(buy) is the type of an intensional verb, ⟨⟨⟨e, t⟩, t⟩, ⟨e, t⟩⟩ but it still expresses the notion of buying. In particular, it would still provide the right truth conditions for a sentence like (6). If the logical form of buy is λy.λx.buy(x, y), then (21) SHIFT(buy) = λQ.λx.Q(λy.buy(x, y)). The logical form of a new coat is

(22) λP.∃y(new-coat(y) ∧ P(y)).

Hence, the logical form of the VP of (6) is an application of (21) to (22): (23) SHIFT(buy)(λP.∃y(new-coat(y) ∧ P(y))) = λQ.λx.Q(λy.buy(x, y))(λP.∃y(new-coat(y) ∧ P(y))) = λx.(λP.∃y(new-coat(y) ∧ P(y)))(λy.buy(x, y)) = λx.∃y(new-coat(y) ∧ buy(x, y)).

This is the same logical form we would get if we didn’t shift the verb, and kept its extensional type.

3 It should be emphasized that this is not an isolated, esotheric example; this phenomenon is widespread, and the problem can be, and has been, demonstrated through myriads of examples.

10

chapter 2

So, only when necessary, do extensional verbs have the type of intensional verbs. Indeed, in such cases, the conjunction has wide scope, just like the case of conjunction of intensional verbs. For example, (10) is indeed equivalent to: (24) John needed a new coat and John bought a new coat. Of course, (10) also has a de re reading, where there is a coat that John needed and bought. But in this case, need has the type of an extensional verb, so there is no need for type-shifting, and, as mentioned above, this reading is obtained by some scoping device such as quantifying in. Indeed, in this case (10) is not equivalent to (24): we don’t have a reading where John needed one coat but bought another coat. Type-shifting, then, occurs only as a last resort; if it is not triggered by type mismatch, it does not apply. For example, (16), which is a conjunction of two extensional verbs, does not cause type mismatch; hence, type-shifting does not have to apply, and it therefore mustn’t apply. 2.2 Type-Shifting and Scope An important question involves the stage at which type-shifting occurs. We have established that (10) requires type-shifting. But at which level of the derivation does it apply? There are, in principle, two options. We could shift buy immediately at the time when it is conjoined with need, which is of a different type. If SHIFT is our type-shifting operator, the logical form of the conjoined predicate will be: (25) λQ.λx.(need(x, Q) ∧ SHIFT(buy)(x, Q)). The first argument that (25) is applied to is: (26) λP.∃y(new-coat(y) ∧ P(y)).

This argument is of type ⟨⟨e, t⟩, t⟩, which is precisely the type required by (25). Hence, we can perform λ-reduction, to produce the desired logical form for (10): (27) need(j, (λP.∃y(new-coat(y) ∧ P(y)))) ∧ SHIFT(buy)(j, (λP.∃y(new-coat(y) ∧ P(y)))).

Alternatively, we could wait until buy is actually applied to its generalized quantifier argument. In this case, the logical form of the conjoined predicate will simply be:

reinterpretation mechanisms

(28) λQ.λx.(need(x, Q) ∧ buy(x, Q)).

11

Only when we apply this predicate to the representation of a new coat and perform λ-reduction do we discover that there is a type mismatch between buy and its object. At this point we apply type-shifting, and get the same logical form as before—(27). If type-shifting applies as a last resort, it would appear that we want to wait as much as possible until we apply it. It would therefore appear that we choose the second derivation. However, since the end result of both derivations is the same, a natural question arises: does it matter? The answer is, in fact, yes. Consider: (29) John didn’t buy a new coat. This sentence is scopally ambiguous; it could mean either that John didn’t buy any coat, or that there is one coat that he didn’t buy.4 But now contrast (29) with (30). (30) John needed but didn’t buy a new coat. This sentence only has the reading where negation takes scope over the existential, namely where John needed a new coat, yet didn’t buy any.5 This fact can be explained if we take the view that type-shifting applies as late as possible. When we conjoin the two predicates, we do not yet shift; hence, the logical form of the conjunction is: (31) λQ.λx.(need(x, Q) ∧ ¬buy(x, Q)).

We defer type-shifting until the last possible moment, namely when buy is applied to the wrong type of predicate. In a way similar to the derivation of (27), this results in the following logical form:

4 The second reading is more prominent in an exchange such as the following: (i) Q: Did John buy the entire stock of new coats? A: No, he didn’t buy a new coat, the one with the gold buttons. 5 Of course, it also has a de re reading, where the negation takes narrow scope (I thank Alessandro Zucchi for drawing my attention to this point). But this reading is generated by a scoping device such as quantifying in, and does not require type-shifting. The important point is that

12

(32) need(j, (λP.∃y(new-coat(y) ∧ P(y)))) ∧ ¬SHIFT(buy)(j, (λP.∃y(new-coat(y) ∧ P(y)))).

chapter 2

In this case, negation does, indeed, take wide scope, as desired. Now, consider what would happen if we didn’t wait, and shifted immediately. In this case, we would conjoin need with ¬buy; hence, we would need to shift the entire second predicate, including the negation. We would get the following conjoined predicate: (33) λQ.λx.(need(x, Q) ∧ SHIFT(λz.λx.¬buy(x, z))(x, Q)).

Now, if we applied it to the denotation of a new coat, the result would be: (34) need(j, (λP.∃y(new-coat(y) ∧ P(y)))) ∧ SHIFT(λz.λx.¬buy(x, z))(j, (λP.∃y(new-coat(y) ∧ P(y)))).

Here negation would take narrow scope, contrary to the empirical facts. Type-shifting, then, applies locally, as late as possible, resulting in narrow scope. One could speculate that this is because type-shifting is a costly operation, hence we prefer to shift as little as possible, as late as possible. 2.3 Type-Shifting and Gradable Adjectives Once we have established that operators introduced by type-shifting receive narrow scope only, we can use this fact to solve a number of semantic problems. As an illustration, consider the case of gradable adjectives. There is a widely (though by no means universally) accepted theory of gradable adjectives, first formalized in Seuren (1973) and Cresswell (1976). According to it, gradable adjectives in their positive form, e.g. tall, big, expensive, and the like, denote functions from individuals to abstract representations of measurements, which are called degrees. Degrees are assumed to be an irreducible type, d, so a gradable adjective is a measure function, whose type is ⟨e, d⟩. However, when a positive gradable adjective is used in a sentence, as in (35), it is predicated of an individual, and produces a truth value: hence, its type needs to be ⟨e, t⟩.

the sentence does not have a reading where need is interpreted de dicto, yet negation takes narrow scope.

reinterpretation mechanisms

13

(35) Mary is tall. How is this change in the type of the adjective brought about? One way is the addition of overt degree morphology, e.g., the comparative morpheme more (or -er). For example, the role of the comparative morpheme in (36) is to turn the measure function tall into a relation between two individuals. (36) Mary is taller than John. However, unlike the comparative, the positive form of the adjectives contains no overt degree morphology. What is it, then, that triggers the change in its type? Kennedy (2007) points out that there are two possible solutions: One solution to this problem would be … that the grammar includes a type-shifting rule that turns measure functions into properties of individuals … A second solution would be to assume … a null morpheme that has the same semantic function as overt degree morphology: it takes a gradable adjective denotation (a measure function) and returns a property of individuals. p. 7

How can we decide between the two options? I suggest that we should look for scope as a deciding factor. Kennedy himself, together with many other researchers, chooses the second option, namely a phonologically null morpheme, which he calls pos. If so, we would expect the positive form to exhibit the same scope ambiguities as the comparative. If, however, the change of type is generated by a type-shifting operator, we would expect the positive to be restricted to narrow scope only. It is well known that comparatives may exhibit scope ambiguities. Russell (1905) has already pointed out that (37) is ambiguous. (37) Jones thinks Smith’s yacht is larger than it is. Under one reading, paraphrased as (38a), Jones has a mistaken belief about the size of Smith’s yacht. The second reading, paraphrased in (38b), is that Jones believes a contradiction. (38) a. The size of Smith’s yacht is x, but Jones thinks that the size of Smith’s yacht is y, and y is greater than x.

14

chapter 2

b. Jones thinks that the size of Smith’s yacht is greater than the size of Smith’s yacht. Note that under the “sensible” reading, (38a), the size of the yacht in Jones’s belief worlds is compared to the size of the yacht in the actual world: hence, the comparative (or at least its second argument) takes wide scope with respect to the intensional verb. The opposite is true under the reading paraphrased as (38b), where Jones believes a contradiction: the size of the yacht in Jones’s belief worlds is compared to itself, i.e. to the size of the yacht in his belief worlds. Hence, the comparative has narrow scope with respect to the intensional verb. Let us now consider the positive form of the adjective: (39) Jones thinks that Smith’s yacht is large. If, indeed, large is the same as larger, except that instead of the overt -er it has a phonologically null morpheme, we ought again to have the same ambiguity. Under the wide scope reading, Jones thinks that the yacht is large with respect to the global standard of comparison, i.e. the standard of comparison considered by the speaker; under the narrow scope reading, Jones thinks that the yacht is large with respect to the standard of comparison in his belief worlds, i.e. whatever counts as large given Jones’s beliefs, desires, goals, and interests. Interestingly, we only get the narrow scope reading: it is impossible to interpret (39) in such a way that in Jones’s opinion the yacht is small, but according to the speaker’s standard it is large. Although such a state of affairs is perfectly conceivable, this interpretation is impossible. Suppose Jones thinks that the yacht is 30 feet long, and, according to his standards, this counts as a rather small yacht. Suppose further that, according to the speaker’s standards, a 30 footer is quite a large yacht. In such a state of affairs, (39) is unequivocally false. In contrast, the reverse is perfectly acceptable: if Jones thinks that a 30 footer is a large yacht, then (39) is true, even if the speaker considers it to be a small yacht. This can be demonstrated by the acceptability of (40). (40) Jones thinks Smith’s yacht is large, but in fact it is small. The sentence means that according the speaker’s standards the yacht is small, but according to Jones’s standards the yacht is large. The scopal behavior of the positive form would be surprising if we assumed a null degree morpheme, but is exactly as predicted if this form is generated by type-shifting. Thus, the fact that elements introduced by type-shifting receive narrow scope only can serve as a diagnostic for whether type-shifting

reinterpretation mechanisms

15

is involved. There are probably other examples of phenomena where this diagnostic can be used fruitfully, but I will not explore this issue further here. 2.4 Type-Shifting and Discourse Referents If type-shifting applies locally, as late as possible, we would expect it to apply at the last stage of semantic derivation. My reference to stages of derivation draws on Partee’s (1993) distinction between semantic structure as reflected in the compositional interpretations of linguistic expressions on the one hand and “computed out” semantic values that may result from carrying out logical or nonlogical inferences on the immediately linguistically determined interpretation. For example, the semantic structure of a sentence may include parts related as function and argument; computing out a semantic “value” may involve applying such functions to their arguments and preserving only the result … [Examples like (41)] illustrate pairs of expressions which differ in semantic structure but perhaps not in computed-out semantic value. p. 10

(41) a. Nine of the ten marbles are in the bag. #It’s under the sofa. b. One of the ten marbles is not in the bag. It’s under the sofa. The respective first sentences of (41a) and (41b) have the same “computed out” meaning: both presuppose that there are ten marbles, and say that nine of them are in the bag, while one is outside. However, (41b) allows a singular pronoun to refer to the marble that is outside the bag, while (41a) does not. This must have something to do with the different semantic structures of these sentences. Heim (1982) argues that the difference between these two semantic structures has to do with the different discourse referents that they introduce. Sentence (41b) introduces a singular discourse referent, hence it allows reference by a singular pronoun; however, (41a) introduces a different discourse referent, a plural one, which cannot serve as the antecedent of a singular pronoun. Hence, it is at the stage of semantic structure that discourse referents are determined (as well as some other non-truth conditional elements). If typeshifting applies as late as possible, it ought to apply after discourse referents are determined. This means that even if type-shifting introduces an existential quantifier, it would be unable to introduce a discourse referent by itself. This prediction will, in fact, be confirmed when we discuss bare plurals in chapter 3.

16

chapter 2

2.5 Blocking Another implication of the notion that type-shifting applies as a last resort is that if there are overt means to perform the same effect as a type-shifting operator, then type-shifting does not apply—it is blocked. Or, as Chierchia (1998) puts it: “Don’t do covertly what you can do overtly” (p. 360). For example, in English, a common noun can receive an indefinite interpretation with the determiner a/an, and a definite interpretation with the definite determiner the. In languages without determiners, such as Russian, such interpretations are the result of type-shifting.6 So, a bare singular noun may shift into either the definite or the indefinite interpretation (the choice between them depends on the context). Chierchia illustrates this point with the following Russian example: (42) V komnate byli malcik i devocka. Ja obratilsja k malciky. in room were boy and girl I turned to boy ‘In (the) room were (a) boy and (a) girl. I turned to (the) boy.’ Chierchia explains: The first occurrences of malcik ‘boy’ and devocka ‘girl’ in a presentational context are interpreted as indefinites. The second occurrence of malcik is anaphoric to the first and, hence, definite. p. 361

But note that in English, unlike Russian, bare singular nouns cannot be typeshifted to receive either the definite or indefinite meaning; hence, the corresponding English sentences (without determiners) are ungrammatical:7 (43) a. *Boy and girl were in room. b. *I turned to boy. Why is that? Chierchia’s explanation is based on the fact that English does have definite and indefinite determiners. Since they can do the job of the covert type-shifting operator, it is blocked from applying. He proposes the following principle:

6 We will discuss the nature of this type-shifting operator in chapter 3, section 3.1 below. 7 Bare plural nouns are a different story; we will deal with them extensively in chapter 3.

reinterpretation mechanisms

17

Definition 1 (Blocking Principle) For any type-shifting operation τ and any X: τ(X) is ruled out if there is a determiner D such that for any set X in its domain, D(X) = τ(X).

Type-shifting operators are not tied in with a specific language; they are assumed to be universal semantic principles, available in all languages. Hence, if a certain operator appears not to apply in a certain language, the reason must be that it is blocked; that is to say, the language must have overt means to achieve exactly the same meaning. 2.6 Type-Shifting or Sort-Shifting? Every type defines a set of expressions—the expressions that are of that type. But, of course, not every set of expressions is a type. For example, it makes sense to say that individuals and properties are different types, but not that fruits and vegetables are different types. Any remotely plausible semantic theory would have to assign to apple and cucumber the same type. Yet, sometimes there are linguistically significant distinctions within the same type. For example, the distinction between physical and abstract objects, or between animate and inanimate objects, may be important for certain linguistic constructions. In such cases, we talk about sorts, rather than types. The distinction between types and sorts may seem fairly straightforward when we only consider individuals. But when our domain also contains times and eventualities, the distinction can be somewhat blurred. For example, later in this book it will turn out to be important to treat points of time and intervals as distinct types, and events as types that are distinct from states or processes. This will be necessary, because we will see that the properties of some constructions can be explained if we assume type-shifting operators that convert, say, an event to a state or a process. While many researchers assume that these are, indeed, different types, some might claim that these are sorts (or ‘subtypes’) rather than distinct types. Such a dispute would not be easy to decide rigorously; but do we even need to? Suppose it turned out that events, states, and processes are of the same type— would this put a type-shifting theory of them in jeopardy? Fortunately, the answer is no. Partee (1987) points out that, in addition to type-shifting operations, there are also sort-shifting operations, i.e. conversions from one sort into another. Moreover, she sees no fundamental differences between them: one chooses whatever best fits one’s theory. For example, as part of a discussion of a preliminary version of Pelletier and Schubert (1989), and

18

chapter 2

some operators proposed there, Partee says: “Some of these operators perform type-shifting operations, others have what we might call ‘sort-shifting’ effects within a single type; … the distinction is a very theory-dependent one that we don’t wish to lay any emphasis on here” (p. 130). In other words, whether one wishes to refer to a particular operation as type-shifting or sort-shifting depends not on some deep, fundamental principles, but simply on the choice that works best for one’s favored theory. It should be emphasized that sort-shifting has the same properties as typeshifting: it applies only as a last resort, and would therefore apply as locally as possible and would lead to narrow scope readings, and would be blocked if the same result can be accomplished overtly. Hence, even if one would like to argue that some of what I call types are, in fact, sorts, an account in terms of type-shifting (or sort-shifting) can still apply to them. Therefore, in this book I will keep referring to type-shifting only, without loss of generality. 2.7 Summary To summarize this section, we should note the following properties of typeshifting: 1. It is semantically triggered, by type mismatch 2. Type-shifting changes the type, but not the meaning of an expression. 3. It occurs as locally, or as late, as possible. 4. Operators introduced by type-shifting receive narrow scope only 5. Type-shifting is blocked whenever it can be replaced by overt means.

3

Predicate Transfer

3.1 The Nature of Predicate Transfer We have seen that when we shift the type of an expression, its meaning remains the same. So, SHIFT(buy) has a different type from buy, but both refer to an act of purchase. Sometimes, however, reinterpretation affects a change in meaning, while leaving the type unchanged. A classic example of such cases is (2), repeated below: (44) The ham sandwich is getting restless. The intended meaning of (44) is, of course, that the customer who ordered a ham sandwich, rather than the sandwich, is getting restless. But both sandwich and customer are of the same type: e.

reinterpretation mechanisms

19

Nunberg (1977) analyzes (44) as a case of reference transfer: the reference of the subject is transferred from the sandwich to the customer, while the predicate remains the same. Thus, the initial logical form of (44) involves reference to the ham sandwich; this reference is then transferred to the customer. In contrast, Sag (1981) argues that what is transferred is the meaning of the NP, rather than the entire DP.8 Thus, the denotation of ham sandwich is reinterpreted from the property of being a ham sandwich to the property of having ordered a ham sandwich. As supporting evidence, Sag presents the following examples: (45) a. There are five ham sandwiches sitting at table 9. b. Every ham sandwich at that table is a woman. Sag points out that both sentences in (45) are about customers who have ordered ham sandwiches, rather than the sandwiches themselves. This is even clearer for (46), another variation on this sentence, due to Egg (2005). (46) Most ham sandwiches are angry. Suppose there are five customers who have ordered a ham sandwich. Four of them ordered one ham sandwich each, and are angry; the fifth ordered ten sandwiches, and is happy and not at all angry. Clearly, in this situation (46) would be true. This is the result we would expect if (46) expresses quantification over customers, since four out of five of them are angry. In contrast, if we were quantifying over ham sandwiches, (46) would be false, since ten out of fourteen sandwiches are eaten by a non-angry customer. Since there is no reason to think that the words five or every behave any differently from the way they usually do, Sag concludes that it is the predicate that changes its sense: from a property of sandwiches to a property of people. Sag argues that the same thing happens with (44): the determiner the maintains its usual meaning, and it is only the predicate whose meaning is transferred. In later work, Nunberg (1995) comes to a similar conclusion. He points out that the definite determiner in (44) presupposes the existence of a uniquely identifiable ham sandwich orderer, not a uniquely identifiable ham sandwich. Thus, it is clear that the determiner combines with the nominal predicate only after is meaning has been transferred.

8 Of course, predating Abney (1987), Sag uses a different syntactic terminology.

20

chapter 2

Another argument for predicate transfer comes from cases where there is a clash in grammatical number between the literal and transferred meaning, e.g., when the food item is referred to by a plural expression, whereas a singular customer is intended, as in: (47) {

That is } French fries { } getting impatient. *Those *are

If (47) were a case of reference transfer, the demonstrative would have to agree with the plural predicate, and then the reference of the entire DP would be transferred. But this is clearly not the case: the predicate French fries is transferred from a (plural) property of foods to a (singular) property of customers before it is combined with the demonstrative, which is why the demonstrative has to be singular in number. The distinction between reference transfer and predicate transfer is even clearer when the predicate is verbal. Nunberg discusses a situation in which a customer hands his key to an attendant at a parking lot and says: (48) I am parked out back. Normally, (48) is interpreted as saying that the speaker’s car is parked out back, not that the speaker herself is parked out back. How does this transfer of meaning come about? One way this could conceivably happen would be if the argument changed meaning. So, in this example, I would change in meaning from the speaker to the speaker’s car. But Nunberg opts for the second option, namely that it is the predicate whose meaning changes: the sense of is parked out back is transferred from the property of being parked out back to the property of having a car that is parked out back. His arguments for this claim come from examples such as the following: (49) a. #I am parked out back and may not start. b. I am parked out back and would like to make a phone call. Suppose it were the argument that changed meaning, from a person to a car. Then (49a) would be predicted to be good: it predicates a conjunctive property—being parked out back and being unable to start—of a car. In contrast, (49b) would be bad, since the second conjunct clearly denotes a property of persons. However, the facts are exactly the opposite: (49a) is unacceptable, while (49b) is good. This is explained if, indeed, what changes meaning is the predicate: in (49a) we conjoin a property of persons (having a car that is

reinterpretation mechanisms

21

parked out back) with a property of cars, hence the unacceptability of the sentence. In contrast, (49b) conjoins a property of person with another property of person, hence its acceptability. Note that both the original predicate and the property it transfers to have the same type: they are both properties of individuals. Hence, predicate transfer is not a case of type-shifting (cf. Nunberg 1995, 130–131, n. 20). Therefore, it is not triggered by semantic mismatch. How is it triggered, then? The answer is that predicate transfer is triggered pragmatically: when the original meaning is well formed, but does not make sense. For example, sentence (48), on its literal interpretation, is acceptable syntactically and semantically; however, it is problematic pragmatically, since it normally does not make sense for a person to be parked. 3.2 “Noteworthy” Properties One may wonder why the predicate may not start in (49a) is not transferred to a property of persons, something like have a car that may not start. If it could, (49a) would be fine; but apparently such a transfer does not occur. Nunberg does not give a complete answer to this question, but he does talk briefly about the conditions under which predicate transfer is effected: predicate transfer is only possible when the property contributed by the new predicate is ‘noteworthy,’ which can be … exemplified in an utterance like ‘I’m parked out back,’ or ‘I’m in the phone book.’ The purposes of these utterances is to classify the members of one set of things (e.g., garage customers) relative to the immediate conversational purposes (of discharging the attendant’s responsibilities, say), in virtue of their relation to the identifying properties of some other group of things (the cars). In cases like these we will say that a property is noteworthy if it offers a useful way of classifying its bearer relative to the immediate conversational interests. From the point of view of a garage attendant, for example, a customer is usefully classified in terms of the properties he acquires from the location of his car (‘I’m parked out back’) or its overall appearance (‘I’m a red Mustang’) rather than in terms of the properties he acquires from its provenance or mechanical condition (‘I was bought from a friend,’ ‘I shimmy at high speed’)—though of course some of these properties might be more useful for other conversational purposes. p. 114

While Cruse (2004) identifies some problems with Nunberg’s specific definition of noteworthiness, the notion itself appears to be reasonably well estab-

22

chapter 2

lished. In later work, Nunberg (2004) argues that noteworthiness is not special to predicate transfer, but is instrumental in other linguistic phenomena, such as noun-noun compounds and zero-derivation (after Downing 1977 and Clark and Clark 1979, respectively). Slabakova et al. (2013) and Featherston et al. (2011) have found experimental evidence for noteworthiness. At any rate, in this book I am not concerned with noteworthiness in general, but only one specific aspect of it, to be discussed in section 3.4 below. 3.3 The Transfer Function Sag (1981) proposes formalizing predicate transfer using a function that maps properties to properties. For example, this function may map the property of being parked out back to the property of owning a car that is parked out back. Sag proposes adding a distinguished operator to our language, to indicate predicate transfer: for every predicate P, the denotation of 𝒯(P) will be the transfer of the denotation of P. Sag places no constraints on the denotation 𝒯(P), completely leaving it for pragmatics. But Dölling (1995) further specifies the nature of this operator. He proposes that transfer introduces an existential quantifier, so that, for every predicate P, Definition 2 𝒯(P) =def λx∃y(R(x, y) ∧ P(y)).

R is a free variable over relations; its value is what is left for pragmatics, according to Dölling.9 For example, in the case of (44), R will presumably be the relation order, so that: (50) 𝒯(ham-sandwich) = λx.∃y(order(x, y) ∧ ham-sandwich(y)). In the case of (48), R will probably be drive, so that: (51) 𝒯(P) =def λx.∃y(drive(x, y) ∧ P(y)).

Thus, the logical form of (48) will be something like (52a), which, after transfer, becomes (52b). Given (51), this is equivalent to (52c). 9 Actually, Dölling also has an additional operator, which is exactly the same, except that the order of the arguments of R is reversed. But, of course, if R is chosen carefully, only one of the two operators is needed.

reinterpretation mechanisms

23

(52) a. parked-out-back(I) b. (𝒯(parked-out-back))(I) c. ∃y(drive(I, y) ∧ parked-out-back(y)).

Dölling’s proposal is particularly interesting for the purposes of this book, because, according to him, predicate transfer always involves the introduction of a quantifier; specifically, an existential quantifier. As we will see later, this restriction only to existential quantification will need to be relaxed; but for now, his formalization provides a good starting point to investigate the properties of implicit quantification introduced by predicate transfer. 3.4 Intensional Properties One constraint on the noteworthiness of properties that is particularly important follows from Nunberg’s paper, though he seems to be unaware of it. Nunberg points out that predicate transfer is indifferent to how the bearer of this new or derived property is referred to—by an indexical or description or whatever. For example, in this situation the parking lot manager could say to the attendant: “The man with the cigar (Mr. McDowell, etc.) is parked out back”. pp. 111–112

Such indifference to the form of their argument is characteristic of extensional properties. For example, be a US citizen is an extensional property; given that, at the time of writing, Donald J. Trump and the President of the US are two different ways to refer to the same individual, (53a) and (53b) must have the same truth value: (53) a. The President of the US is an American citizen. b. Donald J. Trump is an American citizen. In contrast, a predicate like must be an American citizen is intensional; (54a) and (54b) have different truth values. While, by law, the President of the US must be an American citizen, there is no such obligation on the individual Donald J. Trump; he may renounce his citizenship if he so wishes (although in that case, of course, he would not be President anymore). (54) a. The President of the US must be an American citizen. b. Donald J. Trump must be an American citizen.

24

chapter 2

Since intensional predicates are, in fact, sensitive to the expressions by which their argument is referred to, it follows from Nunberg’s characterization that they cannot be subject to predicate transfer. Consider, for example, the following sentence: (55) Picasso is in Wikipedia. The plausible interpretation of (55) is that there is an article about Picasso in Wikipedia. Note that the property of being in Wikipedia is an extensional property: an article either is or is not in Wikipedia, regardless of how it is referred to. Hence, the property can be transferred (assuming it is “noteworthy”), and the desired interpretation is generated. Let us assume the following transfer function: (56) 𝒯(P) =def λx.∃y(about(y, x) ∧ P(y)).

Then, the initial logical form of (55) would be (57a), which is transferred to (57b), whose interpretation is (57c). (57) a. in(p, w) b. (𝒯(λx.in(x, w)))(p) c. ∃y(about(y, p) ∧ in(y, w)) Now consider the very similar (58).

(58) Picasso must be in Wikipedia. As before, the extensional property of being in Wikipedia can be transferred to the property of having an article in Wikipedia, and (58) receives the interpretation that there is an obligation to have an article about Picasso in Wikipedia: (59) a. □in(p, w) b. □(𝒯(λx.in(x, w)))(p) c. □∃y(about(y, p) ∧ in(y, w)).

Crucially, the intensional property of being obligated to be in Wikipedia, cannot be transferred. If it could, we would generate the following interpretation: (60) a. □in(p, w) b. (𝒯(λx.□in(x, w)))(p) c. ∃y(about(y, p) ∧ □in(y, w)).

reinterpretation mechanisms

25

The interpretation generated in (60c) means that there is an article about Picasso that must appear in Wikipedia. However, (58) has no such interpretation; it can only mean that some article about Picasso must appear on Wikipedia, not that there is a specific article that must appear there. Hence, the prediction is borne out: intensional properties cannot be transferred. 3.5 When Does Predicate Transfer Apply? Predicate transfer is a pragmatic process: it applies in order to make sense of the utterance of the speaker. Does this mean that it operates globally, only on the output of semantic processing? Or can it interact with the composition of semantic meaning, applying at more local levels? It appears that predicate transfer can apply at any level in semantic composition. Take, for example, the classic (61).10 (61) There is a lion in the courtyard. Sentence (61) can mean that there is a real lion in the courtyard, but, in most contexts, the plausible interpretation is that there is a representation of a lion— most probably, a statue—in the courtyard. This interpretation is provided by predicate transfer. Let 𝒯(P) be the property of being a representation of a P: (62) 𝒯(P) =def λx.∃y(representation-of(x, y) ∧ P(y))

Then, the logical form of (61) is (ignoring the courtyard): (63) ∃x(𝒯(lion))(x).

Recanati (2003) considers a slight variation of (61): (64) There is a stone lion in the courtyard. He points out that the most plausible interpretation of this sentence is that there is a statue of a lion, and this statue is made of stone. The logical form of (64), then, would be: (65) ∃x(stone(𝒯(lion)))(x). 10

After an example by L.J. Cohen (1985) who, in turn, bases it on examples in Plato’s Republic and Aristotle’s Categories.

26

chapter 2

Note that, in (65), predicate transfer operates locally: the predicate stone applies to the result of the application of 𝒯. It is, of course, also possible for predicate transfer to apply globally; in this case we would get (66) as the logical form of (64). (66) ∃x(𝒯(stone(lion)))(x)

In this formula, 𝒯 applies to the result of the semantic combination of the predicates stone and lion, and the interpretation is that there is (in the courtyard) a statue, and this statue depicts a lion that is made of stone (the statue itself may be made of, say, bronze). This is, of course, not the preferred reading of (64); but it would be the preferred reading of (67). (67) There is a roaring lion in the courtyard. The most plausible interpretation of (67) is that the statue depicts a roaring lion, not that the statue itself is roaring. Nunberg (1995), too, argues that predicate transfer can apply at any level in the semantic derivation of the meaning of a sentence, or, indeed, not at all—it is optional. As evidence, he presents the following example: (68) The article contains a description of a widely-studied Peruvian virus.

Let 𝒯(P) be the property of being a disease caused by something that is P. (69) 𝒯(P) =def λx.∃y(disease-caused-by(x, y) ∧ P(y))

If predicate transfer can apply at any level, the italicized material in (68) may have any of the following logical forms:11 (70) a. b. c. d.

widely-studied(Peruvian(virus)) widely-studied(Peruvian(𝒯(virus))) widely-studied(𝒯(Peruvian(virus))) 𝒯(widely-studied(Peruvian(virus))).

In (70a), there is no transfer, and we get the literal meaning: “a microorganism that is native to Peru and is widely studied.” If, for whatever reason, this 11

Assuming, following Montague (1970), that adjectives are predicate modifiers.

reinterpretation mechanisms

27

interpretation does not make sense (perhaps because it contrasts with our world knowledge), we can apply the function 𝒯; and we can apply it as locally or as globally as we need. If it applies locally, only to virus, we get the interpretation in (70b): the predicate “virus” is transferred to the predicate “disease caused by a virus”, and we get the meaning: “a virally-caused disease that is endemic to Peru and is widely-studied.” We can apply transfer more widely, to the combination Peruvian virus. Then we get the interpretation: “a disease that is caused by a Peruvian virus and is widely studied.” Or, we can apply it globally, to the whole phrase widely studied Peruvian virus, and then we get the interpretation: “a disease that is caused by a virus that is widely studied and is native to Peru”. It seems that all these interpretations are possible, though, of course, some may be preferred to others. Asher (2011) and Asher et al. (2015) argue against the free application of predicate transfer. They claim that this would cause overgenealization. They present the following example: (71) The ham sandwich that hasn’t been eaten is on the counter. They point out that (71) would be predicted have a reading where the eater of the ham sandwich that hasn’t been eaten is on the counter. It is not clear to me, however, why this is a problem. First, the predicate is not transferred to the properry of eating the sandwich, but rather to the person who ordered it, so if Asher and Asher et al. believe there is a contradiction here, this is not the case. And if they mean that there is no context where (71) is interpreted to be about the customer, then this, too, is not the case. One can easily imagine (71) uttered in a context where some unidentified person is dancing on the counter, and the waiters are asking each other who the dancer might be. More generally, we have seen that the application of predicate transfer is not as free as Asher and Asher et al. seem to believe, and is subject to the noteworthiness constraint. The phenomenon of predicate transfer has important implications for views on the semantics-pragmatics interface: it is clearly a pragmatic process, yet it takes part in the semantic operation of the construction of logical form. In this it may seem similar to a phenomenon that has attracted much attention recently, namely embedded implicature. As an illustration, consider (72a): it implicates that the stronger statement, (72b), is false. (72) a. Mary is seeing some of her students. b. Mary is seeing all of her students.

28

chapter 2

Now note what happens when (72a) is embedded inside the scope of another operator, say disjunction: Chierchia (2004) notes that (73a) does not implicate that the stronger (73b) is false—if it did, it would implicate that Mary is not working at her paper, which it most certainly does not. (73) a. Mary is either working on her paper or seeing some of her students. b. Mary is either working on her paper or seeing all of her students. Implicatures, then, appear to be calculated locally. One may be tempted to conclude that predicate transfer, too, is some kind of embedded implicature, and consequently applies locally. This, however, would not do: embedded implicatures and predicate transfer are rather different. To see why, consider the common view of embedded implicature12 (Landman 2000; Chierchia 2004): implicatures are generated by the computational system of the grammar, just like normal meanings, hence their locality. The only difference is that implicatures are defaults, and are therefore defeasible. Recanati (2003) points out two important differences between this view of implicature and predicate transfer. One is that the generation of (scalar) implicatures is bottom-up: it is directly triggered by linguistic material in the sentence, namely the scalar term. In contrast, predicate transfer is top-down: it is not triggered by a specific element in the sentence, but takes place in order to make sense of the communicative act performed by the speaker. Another difference is that, according to the view of implicature as default, it makes no use of the context, background knowledge, and the like.13 However, predicate transfer, of course, makes heavy use of contextual information. Therefore, it is unlikely that the mechanisms accounting for implicature may apply to predicate transfer.14 In fact, even a researcher who denies that Chierchia’s data constitute locally calculated implicatures (Geurts 2009) admits that predicate transfer is a local pragmatic process.

12 13

14

Or, at least, of its paradigmatic case of scalar implicature. Recanati emphasizes that this property is not entailed by the fact that implicature is bottom-up: for example, the process of contextually assigning a value to an indexical is triggered by the occurrence of that indexical in the sentence, yet extralinguistic information is clearly and crucially involved. There are, in fact, theories of embedded implicature which are top-down and context sensitive, such as Bach’s (1994) impliciture or Carston’s (2002) explicature; however, Recanati points out that even if they can account for some cases of implicature, they cannot account for all of them, e.g., the exclusive interpretation of disjunction.

reinterpretation mechanisms

29

3.6 Scope We have established, then, that predicate transfer interacts closely with semantics: it can apply at any stage of the semantic derivation. Depending on the stage of the derivation in which 𝒯 applies, the existential quantifier it introduces can have scope above or below other quantifiers. Consider the following sentence, for example: (74) Marvin has read Chomsky. The normal interpretation of (74) is through predicate transfer: since it makes no sense to read an actual person, the sentence is reinterpreted to mean that Marvin read some piece that Chomsky had written. Formally, let us define the following transfer function: (75) 𝒯(P) =def λx.∃y(write(x, y) ∧ P(y)).

The logical form of (74) (ignoring tense and aspect) is (76a). We can say that it predicates of the individual Chomsky the property of being read by Marvin, and express this, equivalently, as (76b). We can apply the transfer function to this property, to produce (76c), which, given (75) (and after lambda reduction), is equivalent to (76d). (76) a. b. c. d.

read(m, c) (λx.read(m, x))(c) (𝒯(λx.read(m, x)))(c) ∃y(write(c, y) ∧ read(m, y))

This means that there exists a piece that Chomksy wrote and that was read by Marvin. Now consider what happens if we introduce a quantifier: (77) Every linguist has read Chomsky. The initial logical form of (77) is:

(78) ∀z(linguist(z) → read(z, c)).

Now we can apply predicate transfer as in (79a), which, given (75), reduces to (79b).

30

(79) a. ∀z(linguist(z) → (𝒯(λx.read(z, x)))(c)) b. ∀z(linguist(z) → ∃y(write(c, y) ∧ read(z, y)))

chapter 2

This logical form means that for every linguist there is a piece written by Chomsky that he or she read. In this interpretation, the existential quantifier introduce by 𝒯 takes scope under the universal. There is, however, an alternative. We can apply transfer globally, to the property of being read by every linguist. This produces (80a), which reduces to (80b). (80) a. (𝒯(λx.∀z(linguist(z) → read(z, x)))(c)) b. ∃y(write(c, y) ∧ ∀z(linguist(z) → read(z, y)))

According to this reading, there is one piece written by Chomsky (say, Aspects) that every linguist read. Now, the existential introduced by 𝒯 takes scope above the universal. The wide scope reading of the existential is more prominent where an author is usually identified with one of his or her works. For example: (81) Dr. Johnson was used by every English writer of the 19th century. Normally, (81) is not interpreted as saying that every English writer of the 19th century used some piece or other that was written by Dr. Johnson; rather, it means that there is one piece, namely his famous dictionary, that was used by all. Thus, the manner of application of predicate transfer enables us to account for scope ambiguities; this will turn out to be important further in this book. However, we have to be careful: recall that not any property may be subject to predicate transfer—only those that are, in Nunberg’s words, “noteworthy.” In particular, we have seen that intensional predicates cannot be subject to predicate transfer. It therefore follows that a scope ambiguity will be attested only if in both options for the position of 𝒯, it does not apply to an intensional property. Consider the following sentence, for example: (82) The attendant believes that a car that I drive is blocking the entrance. This sentence is ambiguous. It may receive a de dicto reading, according to which the attendant has the belief that some car or other that I drive is blocking the entrance; alternatively, it may be read de re, namely that there is a specific

reinterpretation mechanisms

31

car that I drive, of which the attendant believes that it is blocking the entrance. The two readings are distinct: suppose, for example, that the attendant believes of my car that it is blocking the entrance, without realizing that it is my car. In this case, the de re reading will be true, whereas the de dicto reading—false. Now, consider what happens when the existential quantifier is generated by way of predicate transfer: (83) The attendant believes that I am blocking the entrance. Unlike (82), (83) is not ambiguous: it can only mean that in the attendant’s mind, the car that is blocking the entrance is mine. In the scenario described above, the sentence would be unambiguously false. Why is this? The logical form of (83) is something like the following: (84) Bιattendant block-entrance(I) We can apply predicate transfer locally, to the predicate block-entrance, as in (85a). This is equivalent to (85b), which correctly entails that the attendant believes that the offending car is mine. (85) a. Bιattendant (𝒯(block-entrance))(I) b. Bιattendant ∃y(drive(I, y) ∧ block-entrance(y)).

If we wanted to get the de re reading, i.e. the reading where the existential scopes above the belief operator, we would need to apply predicate transfer as in (86). (86) (𝒯(λx.Bιattendant block-entrance(x)))(I)

However, the property (λx.Bιattendant block-entrance(x)) is intensional, hence cannot be transferred. This is why the wide scope reading of the existential is not available, and (83) cannot be interpreted to entail the existence of the speaker’s car. Hence, operators introduced by predicate tranfer cannot scope out of contexts created by intensional predicates, i.e., they cannot scope out of opaque contexts. 3.7 Summary To summarize this section, we should note the following properties of predicate transfer:

32 1. 2. 3. 4.

chapter 2

It is pragmatically triggered. Predicate transfer changes the meaning, but not the type of an expression. It occurs at any point in the derivation. Not every predicate can undergo transfer: in particular, intensional predicates cannot. 5. Operators introduced by predicate transfer exhibit scope ambiguities (with the exception of opaque contexts created by intensional predicates).

chapter 3

Bare Plurals 1

The Ambiguity of Bare Plurals

Since explicit quantifiers are often determiners, an attractive place to look for implicit quantifiers is constructions that lack determiners: in English these are bare plurals (henceforth BPs). I will not deal here with the syntax of BPs, and, in particular, will have nothing to say on the question of whether they are NPs or DPs with a null determiner. Since I remain agnostic on this issue, I will henceforth in this book use the term Traditional Noun Phrase (TNP) to refer to both NPs and DPs, following Bošković’s (2008) terminology. It is well known that English BPs are ambiguous. They can get a reading of direct kind predication, as in (87a), which predicates a property of the kind as a whole. They can get a characterizing generic reading, as in (87b), which predicates a property of “many” members of the kind.1 Alternatively, they can get an existential reading, as in (87c), which predicates a property of some members of a kind. (87) a. Dogs have evolved from wolves. b. Dogs are intelligent. c. Dogs are barking outside right now. It is fairly uncontroversial that the reading exemplified by (87a) does not involve quantification. What about the readings exemplified by (87b) and (87c)? Are they predicational, or do they involve quantification?

2

A Non-quantificational Theory?

2.1 Carlson (1977) The best known theory that claims that such sentences are predicational is Carlson (1977). Since much of this book hinges on the assumption that BPs are quantificational, this theory needs to be confronted. 1 How many? This is a fascinating but a very difficult question, and I will not attempt to settle it here—though see the rest of this chapter and chapter 7 for some pertinent remarks.

© Ariel Cohen, 2020 | doi:10.1163/9789004432598_004

34

chapter 3

Carlson proposes that bare plurals unambiguously refer to kinds; it is the predicate that determines how the sentence is to be interpreted: as direct kind predication, a characterizing generic, or an existential. Carlson draws a distinction between kind-level, individual-level, and stage-level predicates. Kind-level predicates, such as be extinct or evolve from wolves hold of a kind directly. Individual-level predicates express a timeless property of an individual; a property that is perceived to be permanent, and not dependent on time, e.g. be altruistic, be lazy, be intelligent. In contrast, stage-level predicates, such as be available, be late, be busy, bark outside right now, express a temporary, transitory property of an individual.2 How does the existential reading induced by stage-level predicates, and how does the generic reading induced by individual-level predicates come about? For Carlson no quantifier is introduced, and what appears to be the quantificational force comes only from our interpretation of the predicate. Carlson argues for this claim by pointing out that BPs behave differently from overtly quantified indefinites, particularly with respect to scope. 2.2 The Existential Reading According to Carlson, a kind has a stage-level property just in case some spatiotemporal slice, or stage, of the kind has this property. For example, (87c) is true just in case some stage of the kind canis (the kind of dogs) barks: that is to say, there are some dogs that bark at the time of utterance. Crucially, this existential interpretation is only a consequence of the meaning of stage-level predicates, and in no way can the logical form of existential BPs be said to contain an existential quantifier. One argument that Carlson puts forward to substantiate his claim involves the fact that existential BPs are restricted to narrow scope only. Quantified expressions are expected to interact scopally with other operators and exhibit scope ambiguities. But while overt indefinites do display this behavior, BPs do not, and can only receive narrow scope.

2 It should be emphasized that this is just a rough characterization, and it is not meant as a definition; neither is the description above of stage-level predicates as transitory properties. A definition along these lines would be unlikely to work; for example, Carlson points out that although the adjective dead expresses a permanent property, it behaves like a stage-level predicate: (i) Firefighters are dead. The preferred reading of (i), just like that of the classic (ii), is existential. (ii)Firefighters are available. See Cohen (2003) for a story that exploits this idea.

bare plurals

35

For example, (88a) is ambiguous: it may mean either that for everyone there is a book on giraffes that he or she read, or that there is one specific book on giraffes that everyone read. Sentence (88b), in contrast, can only receive the first reading. (88) a. Everyone read a book on giraffes. b. Everyone read books on giraffes. Similarly, while (89a) may mean either that John saw no spots on the floor, or that there is one spot that he didn’t see, (89b) only has the narrow scope reading of spots. (89) a. John didn’t see a spot on the floor. b. John didn’t see spots on the floor. In fact, there are cases when the scope of BPs is narrower that the scope that overt indefinites can receive. This phenomenon, which Carlson calls differentiated scope, is demonstrated by the following pair: (90) a. #Carla kills a fly (for a living). b. Carla kills flies (for a living). Sentence (90a) only has a very bizarre reading, where Carla’s job is to kill the same fly again and again. The indefinite cannot scope under the habitual, to get a more acceptable (if still strange) reading, where Carla’s job is to kill a different fly each time. In contrast, this reading is easily obtainable for the BP in (90b); in fact, it is its only reading. So the BP can receive narrow scope even when the overt indefinite cannot.3 Yet another argument of Carlson’s has to do with opacity. While overt indefinites are ambiguous between de re and de dicto readings in intensional contexts, BPs can only get de dicto readings. Thus, while (91a) may mean either that any honest politician will satisfy John, or that there is some honest politician whom he is seeking, (91b) can only receive the first interpretation. (91) a. John is seeking an honest politician. b. John is seeking honest politicians. 3 Carlson’s better known example of differentiated scope is: (i) Max killed rabbits/a rabbit for three hours. However, is it not clear that this is really a matter of scope—see chapter 6, note 3.

36

chapter 3

On the basis of these aruments,4 Carlson proposes that BPs involve predication rather than quantification There is no question that Carlson’s arguments present a challenge to any quantificational account of existential BPs; however, they do not rule it out. They do not show that existential BPs do not involve a quantifier; but they do show that if there is such a quantifier, it behaves in a special way. Hence, any account of existential BPs will have to face the facts discussed by Carlson. For example, an account of BPs simply as existentially quantified variables (Wilkinson 1991; Diesing 1992; Gerstner and Krifka 1993; Kratzer 1995) will not do; but other accounts might, as we will see in the following sections. 2.3 The Characterizing Generic Reading Carlson also argues for a non-quantificational account of characterizing generic readings. According to Carlson, the truth of a sentence like (87b) is, in a sense, elementary; it is not logically dependent on properties of individual dogs. He concedes that some generic sentences may, indeed, depend on such specific facts for their truth, but such dependence is not required in general. In particular, Carlson claims that there is no general method to count instances and then determine that a sufficient number of them satisfy a property to make a given generic sentence true. In other words, there is no logical connection between (87b) and facts about the intelligence of Fido, Lassie, and other individual dogs. Properties of individual dogs may be part of the evidence for the truth or falsity of (87b), but this evidence should not be entailed by its logical form. According to Carlson, this is analogous to the fact that the evidence required to make (92) true need not be entailed by its logical form. (92) John is a spy.5 The force of this argument is not clear; there certainly are cases where the evidence needed to confirm or refute a sentence is entailed by its logical form. Quantificational sentences are a case in point. The truth value of such a sen-

4 Carlson also has a fourth argument, which has to do with the behavior of existential BPs with respect to anaphora; but this argument suffers from empirical and theoretical problems (cf. Delfitto 2006, footnote 14, p. 251), so I will omit it here. 5 This evidence may be entailed by the logical form in combination with some appropriate meaning postulates, but not by the logical form itself; contrast this with (i), which is entailed by the logical form of (92): (i) Somebody is a spy.

bare plurals

37

tence is dependent on the number of individuals in the domain satisfying some property, and this dependency is clearly expressed by its logical form: (93) Most fishermen have a Bachelor’s degree. When we want to evaluate whether (93) is true or false, we simply count the number of fishermen who have a Bachelor’s degree, and check whether it exceeds half the total number (in context) of fishermen. It is also not entirely clear that sentences such as (92) do not have logical forms that determine the kind of evidence they require, in terms of more basic properties; theories that make use of lexical decomposition6 can be seen as attempts in exactly this direction. Carlson’s predicational theory is conceptually problematic. By avoiding a quantificational treatment, Carlson’s account of the meaning of BPs gives up one of the strongest weapons in the semanticist’s arsenal. In his own words, Carlson’s approach amounts to dumping a problem into the hands of another realm of enquiry … Now we don’t have to deal with all that messiness in the semantics, and all us semanticists can breathe easier. Of course we would be fooling ourselves to think that this ‘solves’ the generic puzzle in any way. Carlson 1977:65

Carlson’s theory is even more problematic empirically, since there is evidence that the interpretation of BPs does involve quantification. Some care must be taken in collecting this evidence, however. The obvious way to test whether sentences like (87b) introduce a quantifier would be to see whether generic BPs exhibit scope ambiguities. But we should use this test with caution: as we will see throughout this book, implicit quantifiers turn out to behave in special ways with respect to scope, depending on the mechanism by which they are triggered. I will therefore not deal with the scope facts here, and defer their discussion to chapter 5, section 1. There are, however, other tests, one of which is quantificational variability. Consider, for example, Milsark’s (1974) well known sentence: (94) Typhoons arise in this part of the Pacific.

6 For example, it has been sugested that the logical form of kill is actually something like λy.λx.cause(x, die(y)); see Fodor (1977) for an overview.

38

chapter 3

Sentence (94) is ambiguous; according to one reading, it expresses a property that holds, in general, of typhoons, namely that they arise in this part of the Pacific. Under the second reading, in contrast, the sentence expresses a property of this part of the Pacific, namely that typhoons arise there. Given the second reading, then, typhoons is interpreted existentially. Carlson’s theory would be hard-pressed to account for this ambiguity.7 Arise is either an individual-level or a stage-level predicate. If the former, Carlson would predict that the subject is interpreted generically; if the latter—existentially. But in neither case would it predict both readings. It would seem that Carlson has the option of claiming that arise is lexically ambiguous: under one reading, it is an individual-level predicate, whereas under the other, it is stage-level. However, such a move would be implausible for two reasons. One is that there is no evidence whatsoever for such a lexical ambiguity, and its introduction would be nothing more than an artifact whose sole purpose would be to save Carlson’s theory. The second, and worse problem, is that for the stage-level sense of arise, Carlson’s approach would predict a simple existential reading for (94), i.e. that there exist some typhoons that arise at this part of the Pacific; this interpretation leaves out the generic character of (94), namely that it describes (under the second reading) some property that holds, in general, of this part of the Pacific. If the occurrence of typhoons in this part of the Pacific were rare, but typhoons did arise there on occasion, (94) would be false, whereas Carlson’s existential account would predict it to be true. Why, then, does Carlson propose such a theoretically and empirically problematic theory? His answer is simple: a quantificational theory requires a quantifier. But it is impossible to find a generic quantifier that will do justice to all the readings of generics. For example: (95) a. b. c. d. e. f. g. h. i.

Dogs are mammals. Birds fly. Mammals bear live young. Frenchmen eat horsemeat. Bulgarians are good weightlifters. Primary school teachers are female. People are over three years old. Members of this club help each other in emergencies. Supreme Court judges have a prime Social Security number.

7 In later work, Carlson (1989) acknowledges these difficulties.

bare plurals

39

j. A: Nobody in India eats beef. B: That’s not true! Indians do eat beef. Sentences (95a)–(95e) are all presumably true, but what is it that makes them true? Sentence (95a) seems to hold for all dogs, (95b) for most birds, (95c) for most female mammals (presumably less than half the total number of mammals), (95d) for rather few Frenchmen, and (95e) for very few Bulgarians. Things are even more complicated: the majority of primary school teachers are female, and the majority of people are over three years old, and yet (95f) and (95g) are not true. Even if no emergencies ever occured, (95h) may be true, and even if all Supreme Court judges happened to have a prime Social Security number, (95i) may be false. The truth of B’s answer in (95j) requires only that some Indians eat beef. The diversity of interpretations of generics, as exemplified by the sentences in (95), poses severe problems for any quantificational theory of genericity. But the same response to his arguments concerning the existential interpretation applies here: while Carlson has shown that, if there is a generic quantifier, it must have special properties, he has failed to rule out the existence of such a quantifier.8 There have, in fact, been many attempts at a quantificational theory of generics, with varying levels of success.9 I will not argue here in favor of one particular theory; all I claim is that a successful quantificational theory is possible in principle. Since such a theory is desirable, given the problems with a non-quantificational approach to generics, in what follows I will assume that there is a generic quantifier, without getting into its precise semantics.

3

Quantificational Accounts of English Bare Plurals

3.1 Points of Agreement Two influential, more recent theories of BPs (Chierchia 1998; Krifka 2004) assume type-shifting principles in their analyses of BPs. Several assumptions are common to both theories. Both agree that existential and characterizing generic BPs involve a phonologically null quantifier (existential and generic, respectively). However, they differ on the details of how this quantification is produced, as we will see below. 8 But see Liebesman (2011) for interesting arguments against the generic quantifier. 9 The interested reader is referred to my own proposals for an approach that, I have argued, successfully accounts for all readings of generics (Cohen 1996, 1999, 2001c, 2004a). For some recent competing approaches, see Leslie (2007), Sterken (2015), Nickel (2016), Nguyen (2019), and Tessler and Goodman (2019).

40

chapter 3

Chierchia and Krifka agree on a basic inventory of type-shifting operators. Note that, being type-shifting operators, they satisfy all the properties of such operators from chapter 2, e.g. they are subject to blocking, give rise to narrow scope readings, etc. A particularly important point to bear in mind is that these operators, like other type-shifting operators, are assumed to be available universally across languages. I will briefly discuss some of them, which will be relevant for our current purposes. One is an existential type-shifting operator, which turns a property into a generalized quantifier: Definition 3 ∃P =def λQ ∃x(P(x) ∧ Q(x)).

Another is ι, which corresponds to the definite determiner in English: Definition 4 ιP =def the maximal element in P (where defined).

This definition has the effect that if P is a singular property, say dog, ιP picks the unique element that satisfies it (if there is such an element), i.e. the unique dog in context. And if P is a plural property, e.g. dogs, ιP is the largest plurality of P, i.e. the plurality of all dogs. The two operators above go back to Partee (1987). Two additional operators are the ∩ and the ∪ operators, which turn a property into a kind and vice versa, respectively. Chierchia and Krifka attempt to reconcile the intuition that kinds are (special sorts of) individuals with the intuition that they are also related to properties: the kind canis is related to the property of being a dog. The solution they propose is to treat a kind as an individual concept, i.e. a function from possible worlds to individuals. In this case, the individual is a plural individual, the sum of all instances of the kind. In Chierchia and Krifka’s framework, this is simply a set. Chierchia is careful to note that this representation is not meant to carry any metaphysical baggage: Let me immediately qualify: I am not saying that kinds are individual concepts. Kinds are whatever your favorite worldview says that they are. However, in a compositional semantic system that computes truth conditions they can be represented as individual concepts of sort. p. 350

bare plurals

41

Chierchia and Krifka, then, treat the kind ∩ P as a function that produces, for each world (or situation), the maximal plurality of Ps in that world (or situation).10 For example, if the BP dogs denotes the property dogs (which is the property of pluralities of dogs), then the kind ∩ dogs is a function that provides for every world (or situation) the sum of all dogs in that world (or situation). Formally, the ∩ operator is defined as follows (where, for any x, Ps (x) is true iff x is in the extension of P in world or situation s): Definition 5 ∩P = def λs ιPs (where defined).

It should be noted that Chierchia claims that not all properties can be shifted into kinds. In particular, it follows from definition 5 that if P is a singular property, i.e. a property that applies only to singularities, ∩ P is undefined. Thus, for example, in contrast with ∩ dogs, the singular ∩ dog is undefined. The reason is that if there are worlds (or situations) with more (or fewer) than one dog, ιdog is not defined for such worlds (or situations). The operator that turns a kind into a property is the ∪ operator. It denotes a function that, for each world (or situation), returns the property of being a part of the plurality that comprises all the atomic elements of the kind. Formally: Definition 6 ∪k = def λs.λx.x ≤ ks .

Thus, for example, if d is the kind canis, then ∪ d is the property of being a part of the maximal plurality of dogs. Note that both singular and plural instances are considered parts of a plurality. Hence, ∪ d is the property of being a dog or a plurality of dogs. Let us now turn to Chierchia’s and Krifka’s theories of the interpretation of BPs. According to both, BPs start their “lives” as properties. That is to say, their initial denotation, before type-shifting, is that of a property. However, the ∩ operator can apply to shift this property to a kind. This can come about in a case such as (96a), whose initial logical form is (96b). Here, a predicate that selects for kinds, extinct, is applied to a BP, which

10

Chierchia is careful not to commit to either worlds or situations as the basis of the intensionality of kinds. As we will see in chapter 4, section 4.2, there are some reasons to think that situations are more appropriate.

42

chapter 3

denotes a property. We get a type mismatch, and the ∩ operator comes to the rescue, to shift the type of the BP to a kind, with (96c) as the resulting logical form. (96) a. Dogs are extinct. b. extinct(dogs) c. extinct(∩ dogs) In this way Chierchia and Krifka can account for direct kind predication. It is important to note that, according to both of them, BPs in cases of direct kind predication (after type-shifting) denote kinds. Chierchia and Krifka agree on one additional point: that in both characterizing and existential readings, the denotation of the BP is the same, though it may later be type-shifted to an existentially or generically quantified variable. They disagree, however, on what, in these cases, the denotation of the BP is. Chierchia argues that the denotation of the BP is a kind, just like it is in the case of direct kind predication. In contrast, Krifka proposes that, on both their characterizing generic and existential readings, BPs denote properties. More specifically, Krifka argues that these are not plural properties, but properties that are unspecified for number—i.e., they denote pluralities as well as individuals. Krifka presents some evidence supporting the ability of BPs to quantify over individuals. For example, (97a) can be answered by (97b) but not (97c). (97) a. Do you have dogs? b. Yes, one. c. No, only one. Also, (98a) is false if John has one dog, and if Mary has one dog, (98b) entails that John cannot stay with her. (98) a. John doesn’t have dogs b. If Mary has dogs, then John cannot stay with her, as he is allergic. This disagreement between Chierchia and Krifka leads to different accounts of characterizing and existential readings, as we will see below. 3.2 Characterizing Readings Krifka accounts for characterizing readings rather simply: the property denoted by the BP is mapped onto the restrictor of the generic quantifier, thus deter-

bare plurals

43

mining its domain. For example, the logical form of a characterizing generic like (99a) is simply (99b). Since for Chierchia BPs in characterizing generics denote kinds, in order to restrict the domain of the generic quantifier they must be type-shifted back to properties. This is brought about by an application of the ∪ operator, so that the logical form of (99a) is actually (99c). (99) a. Dogs are intelligent. b. genx [dogs(x)][intelligent(x)] c. genx [∪∩ dogs(x)][intelligent(x)]

Note that according to both theories, gen may quantify not only over singular individuals, but over plural individuals as well. For Krifka, this is because dogs is inherently a property that is unspecified for number; for Chierchia, it is because the ∪ operator yields such number-neutral properties. Whichever way one explains it, the option for quantifying over pluralities is necessary in order to account for characterizing generics with collective predicates, such as the following example, attributed by Krifka et al. (1995) to Gerstner (1979): (100) Lions gather near acacia trees when they are tired.11 3.3 Existential Readings and Blocking The case of existential readings is more complex; but despite the differences in their respective views, Chierchia and Krifka derive the existential reading in very similar ways: both derive the existential reading by an operator that is restricted to narrow scope only. For Krifka, this is simply the existential type-shifter ∃. It is triggered by type mismatch: a predicate that requires objects (non-kind individuals) as arguments is applied to a property (the denotation of the BP) instead. As we have seen in chapter 2, section 2.2 above, operators introduced by type-shifting receive narrow scope only, hence the scopal behavior of existential BPs is explained. For Chierchia, the existential reading is provided by an additional, special type-shifting operation, which he calls Derived Kind Predication (DKP), and defines as follows:

11

Things are, in fact, more complicated, as (100) quantifies only over pluralities, since individual lions cannot gather. In Cohen (1996) I propose using coordinates to solve this problem; for simplicity, I will omit coordinates in this book.

44

chapter 3

Definition 7 If P applies to objects and k denotes a kind, then P(k) = ∃x(∪ k(x) ∧ P(x)).

This operator, too, is used to fix the mismatch: this time, it is a mismatch between a predicate that applies to objects, and the kind denoted by the BP. Chierchia, too, believes that this operator applies locally and leads to narrow scope readings: “the mismatch fixing operation DKP will always apply inside the body of the abstract, for outside there is no mismatch to fix” (p. 369, original emphasis). Note that for both Chierchia and Krifka, existentially interpreted BPs can quantify over both individuals and pluralities, just like characterizing BPs: for Chierchia, it is because the DKP involves the ∪ operator, and for Krifka, it is because plural properties are unspecified for number. A problem that is common to both Chierchia and Krifka is this: why isn’t the existential type-shifting of BPs blocked by the plural existential determiner some? We have seen in chapter 2, section 2.5 above that bare singulars can receive an existential interpretation, via type-shifting, in a language like Russian; but not in English, which has overt means to generate an existential interpretation—the determiner a/an. But English also has overt means to provide existential interpretations of plurals, namely some. Why doesn’t this fact block the application of type-shifting to produce the same result? Chierchia and Krifka are aware of this difficulty, and they propose that some contributes more information than mere existential quantification: specifically, some introduces a choice function. Hence, the interpretation of some dogs and the existential interpretation of the BP dogs are not the same, and consequently the former does not block the latter. Some evidence that this is what some does, and that it does not act simply as a determiner (unlike the singular a/an), comes from the fact that some cannot introduce a variable to be bound by an adverbial quantifier. Thus, while (101a) means (roughly) that most dogs are intelligent, (101b) and (101c) cannot be so understood; they can only get the bizarre interpretation that there exists some dog (or dogs) whose intelligence keeps changing as time goes by, but is usually high. (101) a. A dog is usually intelligent. b. *Some dog is usually intelligent. c. *Some dogs are usually intelligent.

bare plurals

45

Thus, the singular determiner a/an blocks the existential interpretation of bare singulars in English, while the plural some does not block the existential interpretation of BPs. It is possible to generalize this explanation across languages. In Italian, for example, it is quite clear that no indefinite plural determiner exists; instead, Italian uses the bare partitive dei (Chierchia 1998). (102) Dei ragazzi sono arrivati in ritardo. of-the boys arrived late ‘(Some) boys arrived late.’ Being a partitive, the meaning of dei is not simply that of an existential, hence it does not block the existential reading of BPs, as we will see in chapter 4, section 3 below. In Spanish, too, existential readings of BPs are possible: (103) Marta compró naranjas. ‘Marta bought oranges.’ (McNally 2004) Unlike Italian, Spanish does appear to have a plural existential determiner, unos/unas. However, it behaves more like English some than like a true determiner. For example, unos/unas apparently cannot restrict the domain of an adverbial quantifier (Louise McNally, pc) and there are reasons to believe that, like some, it denotes a choice function (Palma 2007). Hence, the existence of unos/unas in Spanish does not block the existential interpretation of the BP. The situation in French is different: in this language, BPs are not allowed.12 Instead, French uses the indefinite determiner des. This fact can be explained nicely by the blocking principle: the existential reading of BPs is blocked by this determiner.13 One potential problem with this view is that binding a des NP by an adverb of quantification appears to be impossible. The following examples are from Dobrovie-Sorin (2004): (104) a. *Des enfants marchent rarement avant 10 mois. ‘Children rarely walk before the age of 10 months.’

12 13

Except for a few special circumstances, such as predicative structures, coordination, and enumerations. And, incidentally, the generic reading is blocked by the definite article, just like in Italian, as discussed in chapter 4, section 1 below.

46

chapter 3

b. *Des Indiens meurent en général jeunes. ‘Indians usually die young.’ Thus, it may appear that, just like some in English, des is not really a determiner, hence it should not block existential type-shifting. However, the impression created by the sentences in (104) may be misleading. Dobrovie-Sorin points out that while des NPs cannot form the restriction of quantification over singular individuals, they do allow quantification over plural individuals: (105) a. Des pays limitrophes ont souvent des rapports difficiles. ‘Neighboring countries frequently have difficult relations.’ b. Des amis intimes se critiquent toujours. ‘Intimate friends always criticize each other.’ Sentence (105a) does not quantify over individual countries, but over groups of neighboring countries; and (105b) only makes sense if quantifying over groups of friends. In such cases, des NPs can, in fact, restrict the domain of a quantificational adverb. Hence, des is a real determiner, and is, therefore, able to block the existential type-shift of BPs in French.

4

Kinds or Properties?

While, as we have seen, Chierchia and Krifka agree on many important issues, they still have a substantial disagreement: Chierchia proposes that, under both their existential and characterizing readings, BPs denote kinds, while Krifka argue that, in both cases, they denote properties. Let us consider a few of their respective arguments in more detail. 4.1 An Argument for Properties—Unmotivated Type-Shifting Krifka agrees that in cases of direct kind predication, the BP denotes a kind, after an application of the ∩ type-shifting operator. With regard to characterizing and existential sentences, however, he disagrees with Chierchia. As we have seen, for Krifka, the BPs in both these cases simply express properties, with no reference to kinds. Moreover, the predicate in the restrictor is a predicate of individuals, not kinds. Hence, Chierchia’s proposed type-shift from properties to kinds is not motivated by type mismatch, hence, by the last resort principle, it shouldn’t occur. Therefore, Krifka concludes, the BPs in such sentences remain property-denoting, and are not shifted to kinds.

bare plurals

47

I believe this argument succeeds with the existential reading: the typeshifting of BPs to kinds does indeed seem unmotivated. Even worse, this move requires the application of the special operator DKP, which is stipulated for this use only and for which there is no independent evidence. In contrast, as proposed by Krifka, a simple application of the ∃ type-shifting operator does the job and is fully motivated. Things are more complicated with the characterizing generic reading, however. One might argue that with a sentence likes (106), there is a type mismatch after all: the predicate requires an individual, but its argument is a property instead. (106) Dogs are intelligent. Krifka would probably argue that once the generic quantifier is introduced, it relates two properties, and the mismatch disappears. Since the generic quantifier must be introduced anyway, the type-shifting to kinds is still unmotivated. As both Chierchia and Krifka (and practically everybody else) agree, the generic quantifier is phonologically null. Some scholars (Leslie 2007b; Sterken 2015a; Nickel 2016), have even attempted to derive the meaning of the generic quantifier from this fact, each in a different way.14 The full implications of this fact, however, are rarely appreciated. If the generic quantifier is phonologically null, this means that the hearer cannot hear it, and must reconstruct it somehow. Therefore, the hearer must first generate an initial logical form, of a generic, in which the generic quantifier is not yet present. Hence, the initial logical form of a generic like (106) is predicative, not quantificational. How is a predicative sentence evaluated? Take a simple sentence like (107). (107) John loves Mary. Usually, we think of its logical form as (108a); but we could alternatively think of it as (108b) or (108c). (108) a. love(j, m) b. (λx.love(x, m))(j) c. (λy.love(j, y))(m) 14

In contrast, Liebesman (2011) draws the conclusion that there is no generic quantifier. But, as we have seen in section 2.3, this is quite problematic.

48

chapter 3

The logical form in (108a) treats the sentence as a relation between two individuals; (108b) treats it as predicating of John the property of loving Mary; and (108c) predicates of Mary the property of being loved by John. Of course, after lambda reduction, all three logical forms are logically equivalent. However, following Reinhart (1981), we can maintain that there is, after all, a difference between them. Reinhart argues that predication involves information about the topic of the sentence.15 More specifically, she follows Heim’s (1982) file card metaphor. According to Heim, discourse operates like a filing systems based on cards: the introduction of new individuals into the discourse opens new cards for them; subsequent reference to them provides information to be writtn on the cards. According to Reinhart, (108b) involves adding to the file card corresponding to John the information that he loves Mary, while (108c) adds to Mary’s file card the information that John loves her. Reinhart demonstrates that there are differences in interpretation between the two assignments of topic, especially with regard to presupposition.16 From this it follows that the identification of the topic is necessary for the evaluation of a predicative form. In other words, topicality is logically prior to predication: the predicative structure depends on the topic. Topics are therefore obligatory. Indeed, Krifka himself acknowledges this: “As a general rule of discourse coherence, every sentence must have a topic.”17 Heim (1982) identifies file cards with discourse referents: elements that can be referred to in discourse. Therefore, in order for topics to have an associated file card they must be referential. Indeed, Reinhart (1981) writes: “NP sentencetopics … will be referential entries under which we classify propositions in the context set” (p. 80). The view of topics as referential is widely accepted, even by scholars who do not follow Reinhart’s account. As Gundel (2012) puts it: It is generally agreed … that topics must be at least referential. There must be an individuated entity, or class/group of entities, for the utterance, sentence, or proposition to be about, and in order for truth value to be assessed in relation to that entity. p. 591

15 16

17

Henceforth I refer to topics in Reinhart’s sense, sometimes called aboutness topics. See Portner and Yabushita (1998) for a formalization of this idea in a dynamic setting. See also Nehaniv (2000, 2005) and Krifka (2008) who argue that the development of the notion of topic is what allowed human language to evolve beyond the simple capabilities of animal language by expressing predication. See also Erteschik-Shir (1997).

bare plurals

49

This is the view of topics I am going to assume in this book.18 Consequently, in order to generate the initial, predicative logical form of a sentence like (106), its topic must be established. Properties are not referential, hence they cannot serve as topics. In contrast, kinds are referential. Hence, the property dogs is type-shifted to the kind ∩ dogs, so that the initial logical form of (106) is: (109) intelligent(∩ dogs). Only at this stage may the generic quantifier be introduced, and the kind typeshifted back to a property.19 The resulting logical form is: (110) genx [∪∩ dogs(x)][intelligent(x)].

Crucially, the topic requirement had to be satisfied already at the predicative stage, hence the motivation for the type-shift. Thus, the application of the ∩ type-shifting operator to BPs in characterizing generics is, in fact, well motivated. In contrast, type-shifting of properties to kinds is not necessary in the case of existential readings. For example, in (111a), the BP is interpreted existentially: the sentence says that there are some dogs that John owns. The BP does not have to be a topic; in fact, it can’t be: if the BP is forced to be a topic, as in (111b), the existential reading disappears: (111) a. John owns dogs. b. #Let me tell you something about dogs: John owns them. The only way (111b) is interpretable at all is as a generic, where John owns the vast majority of dogs in the world. Note that the obligatoriness of topics is not violated, because another TNP, John in this case, is the topic. What about existentially interpreted BPs in subject position, as in (112)? (112) Dogs are barking outside right now. If the BP is not the topic, then something else must. But what is it? Note than in such cases, the predicate must be stage-level (Carlson 1977). This fact is significant, for the following reason. 18 19

See also Rochemont (2019). I will discuss the precise nature of the derivation of the generic quantifier in chapter 5.

50

chapter 3

Kratzer (1995) proposes that the different behaviors of stage-level and individual-level predicates stem from their different argument structures. Stagelevel predicates have an additional “Davidsonian” argument, indicating a spatio-temporal location; individual-level predicates lack such an argument.20 Evidence for this claim comes from minimal pairs such as the following: (113) a. When Mary speaks French, she usually speaks it well. b. *When Mary knows French, she usually knows it well. (114) a. Mary often speaks French. b. *Mary often knows French. Kratzer’s claim is that these sentences express quantification over a spatiotemporal variable. In (113a) and (114a), this variable is provided by the stagelevel verb speak, hence the sentences are fine. In contrast, (113b) and (114b) contain an individual-level verb, know, which does not provide a spatio-temporal variable. Hence, this is a case of vacuous quantification, which is prohibited by the grammar.21 Following Cohen and Erteschik-Shir (2002), we can use this fact to account for (112) as follows. Since the predicate is stage-level, it has a spatio-temporal argument, and it is this argument that is the topic. Therefore, the obligatoriness of topics is satisfied, and the subject BP need not be a topic, and is therefore not type-shifted to a kind. Since individual-level predicates, as in (106), do not have a spatio-temporal argument, the only possible topic is the BP, which consequently must be referential and denote a kind. Thus, the argument for properties succeeds for existential readings: there is, indeed, no need to shift the BP to a kind. However, regarding characterizing readings, the argument fails: the BP must be type-shifted to a kind in order to satisfy the obligatoriness and referentiality of topics.

20

21

Of course, this claim does not, in itself, imply that individual-level predicates express properties that hold throughout all time; analogously, the fact that heat may affect the length of objects does not require that the predicate long have a temperature argument. It should be noted that Kratzer’s claim has not been without its challengers. See de Swart (1991) and Chierchia (1995) for alternative accounts, and Cohen and Erteschik-Shir (2002) for counterarguments.

bare plurals

51

4.2 An Argument for Kinds—Bare Singulars Chierchia begins his paper by noting that, in English, bare singular count nouns are not allowed in either direct kind predication, characterizing, or existential sentences: (115) a. *Dog is extinct. b. *Dog is intelligent. c. *Dog is barking.22 According to Chierchia, all these cases require type-shifting to kinds. As we have seen above, Chierchia claims that such type-shifting is possible only with cumulative (plural and mass) properties, but not with singular properties. Hence, Chierchia uses the unacceptability of the sentences in (115) to argue that all the acceptable corresponding sentences with BPs involve reference to kinds: (116) a. Dogs are extinct. b. Dogs are intelligent. c. Dogs are barking. This argument does not apply equally well to all three interpretations of BPs. With respect to direct kind predication, as in (116a), the argument is quite powerful; as we have seen, there is wide agreement that the BP denotes a kind in such cases. However, with the existential reading there is an alternative, simpler explanation. As we have seen in chapter 2, section 2.5 above, Chierchia himself suggests that the existential reading of bare singulars in English is blocked by the indefinite existential determiner, a/an, and this is the reason why in Russian, which lacks determiners, the existential reading of bare singulars is available. This explanation accounts for the fact, noted by Chierchia, that there are no languages that allow bare singulars but not bare plurals. The reason is that there are no languages with a plural indefinite determiner but no singular indefinite determiner. What about the characterizing reading? On the face of it, it seems that a similar blocking account can be provided for the inability of bare singulars to 22

This is not, however, the whole story, since in languages without determiners, bare singulars can receive existential readings. It has been argued that, at least in some such languages, they can also refer to kinds, and/or receive characterizing interpretations. The issue is a complex one, and is therefore deferred until chapter 4, section 7.

52

chapter 3

receive a characterizing generic reading. It seems that (117) means the same as (116b), and therefore blocks (115b). (117) A dog is intelligent. But does (117) really receive a characterizing generic reading? To answer this question, we need to look a little more closely at English indefinite singulars (henceforth ISs). 4.3 English Indefinite Singulars English ISs sometimes appear to have characterizing generic readings: (118) A big watchdog is efficient/aggressive. However, the distribution of generic readings of ISs is more restricted than that of BPs. For example, while (119a) is fine, (119b) is odd (under its generic reading).23 (119) a. Big watchdogs are hairy/black. b. *A big watchdog is hairy/black. For another example, consider the following well known minimal pair (Lawler 1973): (120) a. Madrigals are popular. b. *A madrigal is popular. Additional examples are presented by Burton-Roberts (1977): (121) a. Kings are generous. b. *A king is generous. (122) a. Rooms are square. b. *A room is square.

23

One might object that, in the right context, this and the following IS generics which I have starred can actually be quite acceptable. This point is, in fact, absolutely correct; I will turn to the significance of context in a moment.

bare plurals

53

Note that an overt modal or quantificational adverb turns a bad IS generic into a perfectly acceptable sentence:

(123) a.

b.

c.

d.

⎧ may be ⎫ { } A big watchdog ⎨ is always ⎬ hairy/black. { is often } ⎩ ⎭ ⎧ may be ⎫ { } A madrigal ⎨ is always ⎬ popular. { is often } ⎩ ⎭ ⎧ may be ⎫ { } A king ⎨ is always ⎬ generous. { is often } ⎩ ⎭ ⎧ may be ⎫ { } A room ⎨ is always ⎬ square. { is often } ⎩ ⎭

Elsewhere (Cohen 2001b) I have proposed the following explanation of these facts. We have seen in section 4.1 above that the initial logical form of a characterizing generic is predicative, and that at this stage a topic must be identified. BPs may become topics by referring to kinds; however, indefinite singulars do not have this option, as they cannot denote kinds: (124) a. *A white-colored elephant has become extinct. b. *A white-colored elephant grows larger as one drives north c. *A white-colored elephant is so-called because of the pigmentation of its skin.24 Hence, indefinite singulars cannot have characterizing generic readings. It follows inevitably that, despite appearances, sentences like (125) are not characterizing generics, and do not involve the generic quantifier. (125) A big watchdog is efficient/aggressive.

24

All these examples become good if the IS is used taxonomically, i.e. as referring to a subspecies of while-colored elephants, e.g., the Indian white-colored elephant. I will return to taxonomic readings in chapter 4, section 2.2 below.

54

chapter 3

One piece of evidence for my proposal is the fact IS generics do not exhibit scope ambiguities. A BP generic is, in general, scopally ambiguous. For example, (126a) can mean either that “every” big watchdog has a (possibly different) special trainer, or that there is one special trainer common to “all” big watchdogs. In contrast, (126b) is not ambiguous in this way, and only receives the former (wide scope) reading. .

(126) a. Big watchdogs have a special trainer. b. A big watchdog has a special trainer. What, then, is the meaning of (125)? I have suggested that it predicates of some rule or regulation that it is in effect, namely accepted by the language community (though not necessarily adhered to). Note that such a sentence satisfies the topic constraint, since it has a topic: the topic is not the indefinite, which cannot be referential, but the rule, which is considered an individual in its own right. Thus, (125) predicates of a rule, associating the size of watchdogs with their efficiency/aggressiveness, that it is in effect. So, an IS generic refers to a rule or a regulation, and applies the predicate be in effect to it. Just like definite descriptions may fail to refer, if their referents are not in the common ground, IS generics may also fail to refer, if the rules they attempt to refer to are not “conceivable”. One type of conceivable rule (among others) is a linguistic rule—a definition.25 For example, being polyphonic is part of the definition of a madrigal: (127) A madrigal is polyphonic. An IS generic whose form is that of a definition, even though its meaning isn’t, may be acceptable: (128) a. A madrigal is a popular song. b. A king is a generous ruler. c. A room is a square enclosure. If the sentence is clearly in the form of a definition, even kind predicates may modify a generic IS. Consider (129a), which is the definition of dinosaur from the Cobuild English Language Dictionary (1987), and contrast it with (129b):

25

See Krifka (2009) for additional arguments that generic ISs express definitions.

bare plurals

55

(129) a. A dinosaur was a large reptile which lived in prehistoric times and which is now extinct. b. *A dinosaur is extinct. Another example of a conceivable type of rule, which is particularly relevant here, is an action trigger. Consider the following minimal pair, from BurtonRoberts (1977). (130) a. Gentlemen open doors for ladies. b. A gentleman opens doors for ladies. Burton-Roberts notes that while the most natural reading of (130a) is a generalization about gentlemen, (130b) is more likely to be said by Emile’s mother to get him to open doors for ladies, i.e. to trigger a certain kind of action from Emile. Note that if the context makes it clear that the sentence is used to trigger action, bad IS generics become good. Note how the bad (121b) and (122b) become perfectly good if used in the following respective contexts: (131) a. Sire, please don’t send her to the axe. Remember, a king is generous! b. How dare you build me such a room? Don’t you know a room is square? We can now explain why bad IS generics become good with an overt modal or quantificational adverb, as in (123). Unlike the generic quantifier, these operators are “there” in the sentence. Hence, the interpretation of the sentence does not pass through the initial stage of direct kind predication, which is then followed by the derivation of an implicit quantifier. Therefore, reference to kinds is not necessary, and an indefinite singular is perfectly good. The upshot of this discussion is that what looks like a characterizing generic reading of an indefinite singular, really isn’t; hence, the indefinite determiner a/an cannot block the characterizing reading of the bare singular in English. Hence, Chierchia’s argument succeeds with charactering readings: the reason must be that characterizing readings require kinds, which bare singulars are unable to denote. We can conclude that Chierchia’s argument from bare singulars succeeds with both types of generic readings—direct kind predication and characterizing, but fails with existential readings, where a blocking account seems quite capable of doing the job.

56

chapter 3

4.4 An Argument for Kinds—Scopally Ambiguous Bare Plurals As we have seen above, existential BPs are restricted to narrow scope only. But there are cases of BPs that do exhibit scope ambiguities. An example of such a BP is parts of that machine: (132a) is ambiguous, and can mean either (132b) or (132c). (132) a. John didn’t see parts of that machine. b. John didn’t see any parts of that machine. c. There are parts of that machine that John didn’t see. Chierchia notes a similar behavior with BPs like boys sitting here and people in the next room. Following Carlson (1977), he explains this phenomenon by claiming that such BPs cannot denote a kind. Therefore, DKP cannot apply; instead, we have an application of the type-shifting operator ∃. Crucially, Chierchia assumes that ∃, unlike DKP, is not restricted to narrow scope only, and allows scope ambiguities. Thus, Chierchia links the ambiguity of (132a) with the inability of the BP to denote a kind, and concludes that narrow scope existential BPs must denote kinds. One problem with Chierchia’s approach is that, as we have seen, there are good grounds to believe that quantifiers introduced by type-shifting receive narrow scope only. Moreover, in the specific case of BPs, Chierchia’s denial of the locality of type-shifting leads to the following problem: his theory entails that BPs that unquestionably may denote kinds, like birds, lions, etc., do not have to. They could also be subject to ∃, hence we would expect them to exhibit scope ambiguities, which they do not. In order to solve this problem, Chierchia assumes a hierarchy of type-shifting operators: ∩ outranks ∃. Thus, he postulates that if a BP may denote a kind, in fact it does have to. Another problematic aspect of Chierchia’s proposal is that it crucially relies on the assumption that not all BPs may denote kinds. Recall that Chierchia argues that singular properties cannot be shifted into kinds. But can ∩ apply to all cumulative (plural and mass) properties? Chierchia’s answer is no: “The property of being a broken old shoe that Leo left behind is unlikely to have a corresponding kind” (p. 351). He therefore postulates a set K of kinds, and claims that ∩ P is undefined if λs ιPs ∉ K. I am not, however, convinced by this claim, as a kind predicate can apply to a BP expressing this property. Sentence (133) seems to make perfect sense: (133) Broken old shoes that Leo left behind became bigger the older Leo got.

bare plurals

57

Note that the predicate of (133) must be a kind predicate: no individual shoe gets bigger as Leo grows older! The predicate expresses a property of the kind, namely that as Leo grows older, the instances of the kind at that time are bigger than in previous times (cf. the well known Wolves grow bigger as you drive north). Even if we grant that BPs like parts of that machine cannot denote kinds, it is not clear that this is what allows them to receive wide scope. There is an alternative explanation that does not require existential BPs to denote kinds, keeps the idea, defended in chapter 2, that type-shifting is local, and does not require the postulation of ranking between type-shifting operators. I suggest that the apparent scope ambiguity of the BPs in question is really an ambiguity between a specific and a non-specific interpretation. It is well known that indefinites may receive a specific, or referential, interpretation (Fodor and Sag 1982). For example, the most likely interpretation of (134) is that the speaker has some specific individual in mind, and is not merely making an existential statement. (134) A student that Betty used to know in Arkansas cheated on the exam. Note that this specific interpretation does not involve reference to kinds, as can be seen by the felicitous use of the indefinite singular. I propose that the wide scope reading of BPs in sentences such as (132a) is generated when they receive a specific interpretation; when they are not specific, the ∃ type-shifting operator applies normally, and the BP receives narrow scope. The problematic BPs mentioned by Chierchia are perfect examples of BPs that can be interpreted as referring to a specific group of individuals, restricted in time and space. It appears, then, that, in many cases, BPs that cannot denote kinds also are those BPs that may be interpreted specifically. How can we determine which of these two characteristics (the inability to denote kinds or the possibility of a specific interpretation) is responsible for the scopally ambiguous reading? In general, it is quite hard to distinguish the two. But consider this example from Geurts (2010): (135) Several students reported that they had been harassed by professors wearing false beards and pink gowns. This sentence certainly has a reading, probably the preferred reading, where the BP receives wide scope: the same professors harassed all the students. Clearly, the BP has a specific interpretation: it is possible to conceive of a spe-

58

chapter 3

cific group of professors wearing false beards and pink gowns. But does this BP fail to denote a kind? It appears that we can talk about the kind professors wearing false beards and pink gowns, and the BP may be the argument of kind predicates, as in: (136) Professors wearing false beards and pink gowns grow more numerous as one gets closer to the area where the alien virus was released. Additional evidence that wide scope BPs are interpreted specifically comes from Hindi. In this language, specificity is indicated by an overt particle, ko (Butt 1993; Bhatt and Anagnostopoulou 1996). Without this particle, a noun phrase cannot receive a specific interpretation. Interestingly, when (132a) is translated into Hindi, the BP is not marked with ko; as predicted by the current proposal, it can only receive the narrow scope interpretation, and the sentence is not ambiguous: (137) anu-ne is mashiin ke TukRe nahiiN dekhe Anu-ERG this machine of parts not see-PAST ‘Anu didn’t see parts of this machine.’ = Anu didn’t see any parts of this machine.26 I therefore conclude that when BPs receive wide scope, they are interpreted specifically, and hence the phenomenon of scopally ambiguous BPs fails to provide evidence that existential BPs refer to kinds.

5

A Synthesis

Let us take stock. The arguments for kinds appear to succeed only for characterizing readings (and, of course, direct kind predication), and the argument for properties—only for existential readings. This situation calls for a synthesis of the views of Chierchia and Krifka. In this section, which is based on Cohen (2007), I propose such a synthesis. I accept the type-shifting operators that Chierchia and Krifka propose, but reject their assumption that in both characterizing and existential readings, the denotation of the BP is the same. Following Cohen and Erteschik-Shir (2002; cf. Van 26

This example is from Dayal (2004b). Dayal actually has a different explanation for this fact: she claims that the BP receives a definite reading rather than an existential one. I will return to Dayal’s theory in chapter 4, section 7.2 below.

bare plurals

59

Geenhoven 1996; McNally 1998; Dobrovie-Sorin and Laca 1996; Doron 2003), I suggest that characterizing BPs, like Chierchia claims, denote kinds, but existential BPs, like Krifka argues, denote properties. Thus, the BP dogs in (138a) denotes the kind ∩ dogs = λs ιdogss , while in (138b) it denotes the property dogs. (138) a. Dogs are intelligent. b. John saw dogs. The initial logical form of (138a) is: (139) intelligent(dogs). This is a predicative logical form; hence, before it is generated, the topic must be identified. Therefore, the BP must be referential, hence it is type-shifted to the kind ∩ dogs, and we get: (140) intelligent(∩ dogs).

This logical form is semantically well formed, but pragmatically odd: it is not the kind ∩ dogs that is intelligent; only individual dogs can be intelligent. Hence, the generic quantifier is introduced,27 resulting in: (141) genx [∪∩ dogs(x)][intelligent(x)].

The existential reading is generated by the application of the ∃ operator. The initial logical form of (138b) would be (142), where e is the spatio-temporal location variable. (142) see(j, dogs, e) This logical form is predicative, and therefore, as we have seen in section 4.1, requires a topic. This requirement is, in fact, satisfied, because either John, or e, the spatio-temporal location, can serve as a topic. However, the logical form suffers from a type mismatch, because the second argument of see ought to be an individual, and not a property. The mismatch

27

See chapter 5 below for more on the motivation for the introduction of the generic quantifier, and the mechanism that brings it into effect.

60

chapter 3

is resolved by type-shifting: we apply ∃ to the property λx.see(j, x, e), and, as desired, we get: (143) ∃x(dogs(x) ∧ see(j, x, e)).28

If we assume that type-shifting occurs as locally as possible, the existential interpretation receives narrow scope only. For example: (144) John didn’t see dogs. The initial logical form of (144) is: (145) ¬see(j, dogs, e).

If type-shifting applies locally, we first resolve see(j, dogs, e), as before. Only then do we apply negation to the result, to get: (146) ¬∃x(dogs(x) ∧ see(j, x, e)).

Thus we account for Carlson’s (1977) observation that existential BPs receive narrow scope only.29 This proposal can also explain why BPs do not get de re readings in opaque contexts, and the explanation is compatible with both main theories of opacity. Montague (1973) claims that the de re/de dicto ambiguity is a matter of scope: wide scope TNPs are interpreted de re, narrow scope TNPs are interpreted de dicto. Since BPs receive narrow scope only, the unavailability of de re readings is thereby explained. Alternatively, we could follow Zimmermann (1993), who claims that the ambiguity hinges on different interpretations of the TNP: if the TNP is interpreted as a property, we get de dicto readings, whereas when the TNP introduces a discourse referent, we get de re readings. Since existential BPs denote properties, the opacity phenomena are explained. To strengthen this conclusion, let us investigate more closely the question whether BPs introduce discourse referents. In chapter 2, section 2.4 above it

28 29

In addition, the existential reading is augmented by an interesting implicature, as discussed in chapter 7, section 2.2. Recall that, according to Carlson, BPs sometimes receive scope that is narrower than the scope of overt indefinites; the explanation of this fact will be provided in section 5.2.

bare plurals

61

was established that type-shifting occurs after discourse referents are determined. Therefore, if BPs are introduced by type-shifting, they should not introduce discourse referents. This prediction is, in fact, borne out. There are a number of phenomena that demonstrate this fact. One such phenomenon is the observation that BPs support maximal anaphora only: (147a) and (147b) leave open the possibility that there were other men who came in but did not start a riot. This is to be expected if the subject introduces a discourse referent, which serves as the antecedent of the pronoun. In contrast, the truth of (147c) requires that all men who came in started a riot. (147) a. A man came into the bar. He started a riot. b. Some men came into the bar. They started a riot. c. Men came into the bar. They started a riot. Since men does not introduce a discourse referent, the only possible interpretation of they is as an E-type pronoun, paraphraseable as the men who came in; hence the maximal interpretation of the BP. Another relevant phenomenon is Szabolcsi’s (1997) “other” test.30 B’s answer in (148) is quite felicitous: (148) A: A student/some students misunderstood the question. B: Maybe you will find others, too. This is explained by the claim that the subject introduces a discourse referent; others is anaphoric to this referent, and refers to individuals, which are not included in it. B’s answer in (149), however, is bad: (149) A: Students misunderstood the question. B: *Maybe you will find others, too. This is because here there is no discourse referent that serves as the antecedent of others. Another test suggested by Szabolcsi (1997) involves specificity. She argues that discourse referents may be interpreted as referring to a specific individual or set. Thus, for example, (150a) has an interpretation where a specific teacher

30

Szabolcsi applies this and the following tests to various types of Hungarian TNP; however, she makes no claim about BPs.

62

chapter 3

checked that a specific boy was present; similarly, (150b) may be interpreted as a statement about a specific set of teachers and a specific set of boys. (150) a. Yesterday, a teacher checked that a boy was present. b. Yesterday, some teachers checked that some boys were present. However, no such reading is available for (151), where no discourse referents are introduced. (151) Yesterday, teachers checked that boys were present. Hence, BPs do not introduce discourse referents, and, therefore, the fact that they receive only de dicto readings is explained by Zimmermann’s (1993) theory.

6

Deciding between the Theories

We have to choose between three theories: Chierchia (1998), Krifka (2004), and the synthesis proposed here. How do we decide the matter empirically? How can we determine, for each of the two relevant interpretations of a BP (existential and characterizing) whether the BP in question denotes a kind or a property? Crucial pieces of evidence would be sentences where one occurrence of a BP is interpreted in more than one way. For example: (152) a. Dodos lived in Mauritius and (they) became extinct in the 18th century (after Heyer 1990). b. Dinosaurs were large reptiles that are now extinct. The VPs in these sentences consist of two conjoined predicates. The first predicate, lived in Mauritius or were large reptiles, forces the BP to receive a characterizing reading. The second predicate, became extinct in the 18th century or are now extinct, forces it to be interpreted as direct kind predication. The fact that the sentences are fine shows that in both cases, the BP denotes the same object. Hence, the BP must denote a kind in a characterizing generic as well as in sentences of direct kind predication. Another crucial example is the following: (153)

??Dogs are man’s best friend and (they) are barking outside right now (Schubert and Pelletier 1987).

bare plurals

63

Again, we have a conjunction of two predicates. The first forces a generic interpretation of the BP, while the second forces an existential reading. The fact that this sentence is bad shows that the BP denotes different objects in each of the readings: if it denotes a kind in generic sentences, it must denote a property in existential ones. If these judgments are granted, the synthesis presented here is confirmed empirically. However, it turns out that judgments are not so clear. As a matter of fact, Schubert and Pelletier (1987) present sentence (153) as an example of a good sentence, and use it to argue that existential BPs do denote kinds. We must, therefore, look elsewhere for data to resolve the matter. A good place to look would be a language where BPs cannot denote kinds. Then, if in this language BPs get a characterizing reading, we would know that reference to kinds is not necessary for such readings; otherwise, we will conclude that it is. And the same goes for the existential reading. Since I argue in this book that both the characterizing and existential readings of BPS are generated by reinterpretation mechanisms, and since reinterpretation mechanisms are universal, facts from such a language will provide evidence about the status of BPs in other languages, including English. In the next chapter I will argue that Italian is precisely such a language.

chapter 4

Beyond English Bare Plurals We ended the previous chapter noting that, in order to study the properties of English BPs, we need to look beyond English. It will turn out that we will also need to look beyond BPs, and also at bare singulars and definites. We will begin our journey with Italian BPs.

1

Italian Bare Plurals: Direct Kind Predication

Can an Italian BP be the argument in a sentence expressing direct kind predication? In this section I argue on both empirical and theoretical grounds that the answer is no; hence, Italian is a perfect testbed for comparison between the three theories of BPs discussed in the previous chapter. 1.1 The Data The issue of direct kind predication in Italian deserves some discussion, because judgments are, unfortunately, controversial. Chierchia (1998) claims that direct kind predication is possible, and presents the following examples: (154) a. Qui, ragazze in minigonna sono rare. ‘Here, girls in miniskirt are rare.’ b. Dopo il disastro nucleare, purtroppo, cani con difetti congeniti sono molto comuni. ‘After the nuclear disaster, unfortunately, dogs with birth defects are very common.’ However, these examples are not very convincing. Chierchia’s argument crucially depends on the assumption that the predicates be rare and be common are kind predicates. However, Krifka et al. (1995) argue persuasively against this assumption. One of their arguments is that these predicates, unlike real kind predicates, are acceptable with indefinite singulars in English. Note the contrast between (155a) which is unequivocally bad (except under its taxonomic reading), and (155b), which is, if not perfect, substantially better.

© Ariel Cohen, 2020 | doi:10.1163/9789004432598_005

beyond english bare plurals

65

(155) a. *A black rhino is extinct. b. A rhino with blue eyes is common. Krifka et al. analyze (155b) not as a predication of the property of being rare of the kind ∩ rhinos, but as a statement that the chance of encountering a blueeyed rhino is high.1 Note that with uncontroversially kind predicates, BPs are impossible (Longobardi 2001): (156) a. *Elefanti di colore bianco sono estinti.2 ‘White-colored elephants have become extinct.’ b. *Elefanti di colore bianco diventano sempre più grandi man mano che si va a nord ‘White-colored elephants grow larger as one drives north.’ c. *Elefanti di colore bianco sono così chiamati per la pigmentazione della loro pelle ‘White-colored elephants are so-called because of the pigmentation of their skin.’ I therefore conclude that, in Italian, direct kind predication is not possible with BPs.3 1.2 The Theory Assuming that, indeed, Italian BPs cannot be interpreted as direct kind predication, what is the explanation of this fact? It is known that Italian allows overt shifting of BPs to kinds, through the definite determiner. The following example is due to Longobardi (2001): (157) Gli elefanti di colore bianco sono estinti. ‘The white-colored elephants have become extinct.’

1 See Cohen (1999) for a probabilistic account of such sentences. 2 Chierchia judges the similar (i) to be good, but even he admits that “[s]ome informants find [(i)] more marked than [the sentences in (154)]” (p. 385). (i) Insegnanti davvero dediti nella scuola di oggi sono quasi estinti. ‘Really devoted teachers in today’s school are nearly extinct.’ 3 For additional arguments supporting this conclusion, see Dayal (2004a) and Robinson (2005).

66

chapter 4

Since an overt determiner does the job, the blocking principle dictates that such type-shifting cannot apply covertly. Hence, Italian BPs cannot denote kinds. This explanation relies on the fact that Italian allows definite plural generics, but English does not. Before we continue, we need to figure out why that is so.

2

Definites and Type-Shifting

2.1 The Interpretation of Definiteness Let us first consider what the definite determiner does. Chierchia (1998), following Sharvy (1980) and many others, assumes that it is a maximizing operator: ιP is the largest member of P if there is one; otherwise, it is undefined. Thus, the logical form of the dog is ιdog; its meaning is the maximal element in the denotation of dog. Since the denotation of dog is a set of individual dogs, it does not have a maximal element unless it contains exactly one dog—hence the uniqueness presupposition of definite singulars. Things are different with the plural the dogs, whose logical form is ιdogs. Since dogs denotes a set of pluralities of dogs, there is always a maximal element, which is the plurality of all dogs. For Chierchia, pluralities are sets, hence ιdogs denotes the set of all dogs. This interpretation only pertains to the specific, non-generic use of the definite determiner. But definites can also receive a generic interpretation. Sentence (158a) is interpreted as direct kind predication, and (158b) is a characterizing generic. Note that unlike BPs, a definite cannot be interpreted existentially: (158c) can only mean that a uniquely identifiable tiger is chasing a monkey, not that some tiger or other is doing so. (158) a. The tiger is in danger of extinction. b. The tiger has stripes. c. The tiger is chasing a monkey. These facts follow naturally under the proposal made in this book, if we assume that definite singulars may, in addition to their specific interpretation, refer to kinds; thus, they can be interpreted as direct kind predication or characterizing generics, just like BPs. And, since in our system there is no operation corresponding to DKP, there is no way for a definite singular to receive an existential reading. This type of account has, in fact, been challenged. Krifka (2004) and Dayal (2004b) deny that definites can be interpreted as characterizing generics, and

beyond english bare plurals

67

Chierchia (1998) goes so far as to deny that definite singular generics involve reference to kinds in the first place. Let us consider each objection in turn. In fact, my proposal that definite singulars refer to kinds is in agreement with Krifka (2004). However, as we have seen above, for Krifka characterizing generics involve properties, and, consequently, characterizing readings with definite singulars, as in (158b) above, pose a problem for him. Krifka is therefore led to the conclusion that such sentences, despite appearances, are not characterizing after all, but rather express direct kind predication. What are the truth conditions of such predications? Krifka suggests that the truth conditions of a definite singular generic are defined to be those of the corresponding BP sentence—where the BP, according to Krifka, denotes a property. Thus, the truth conditions of (158b) are simply stipulated to be the same as the truth conditions of (159). (159) Tigers have stripes. Of course, if, as in the current proposal, characterizing generics involve reference to kinds rather than properties, such a move becomes unnecessary, and the equivalence of (158b) and (159) follows naturally, and does not need to be stipulated. Dayal (2004b) also claims that sentences like (158b) are not characterizing generics. She claims that, unlike BPs, definite generics are restricted to properties that are, in some sense, “necessary” rather than “accidental”. As evidence, Dayal produces the following minimal pair, which she ascribes to Edwin Williams: (160) a. Rutgers Professors seem to be born on weekdays. b. #The Rutgers Professor seems to be born on a weekday. Dayal claims that the property of being born on a weekday is accidental rather than necessary, and uses this point to argue that definite generics in English are only good with essential or necessary properties. It should be pointed out that such claims are usually not made about the definite determiner, but rather the indefinite determiner. For example, we have seen that Burton-Roberts (1977) considers the following minimal pair: (161) a. Gentlemen open doors for ladies. b. A gentleman opens doors for ladies. He notes that (161b), but not (161a), expresses what he calls “moral necessity.”

68

chapter 4

However, it is far from clear that the oddness of (160b) is due to the accidental nature of the generalization. Indeed, Krifka (2004) points out that, unlike (161b), (162) is, in fact, bad. (162) #The Gentleman opens doors for ladies. Note that (163) is much better than (160b) as a generic, yet it predicates exactly the same property of a definite singular: (163) The successful Rutgers Professor seems to be born on a weekday. If indeed definite singular generics were bad with accidental properties, the acceptability of (163) would not be expected. The factors affecting the distribution of the definite generic are, at present, poorly understood, and appear idiosyncratic and language dependent. Contrast (164a), which is fine, with (164b), which is odd (under the generic reading). (164) a. The tiger lives in the jungle. b. #The dog barks. This contrast seems to be an idiosyncratic property of English; indeed, there are languages where the equivalent of (164b) is perfectly acceptable, e.g. German: (165) Der Hund bellt.4 Therefore, the greater productivity of BPs, as opposed to definite singulars, provides no evidence for the claim that their interpretations are different. On the contrary: whenever the definite singular generic is acceptable, it can receive a characterizing interpretation, just like the corresponding BP.5

4 This example is taken from Heyer (1990). 5 Dayal also presents the following minimal pair: (i) a. Yesterday between 3 and 4 whenever thieves entered the house, the police caught them. b. #Yesterday between 3 and 4 whenever the thief entered the house, the police caught him. It is not clear, however, what this comparison can tell us about characterizing generics, since the BP in (i.a) is clearly interpreted existentially rather than generically—indeed, the BP can be replaced by the overtly existential some thieves with no significant change in meaning.

beyond english bare plurals

69

Chierchia (1998), of course, would face no problem accounting for the characterizing readings of definites, if he proposed that they, like BPs, refer to kinds: type-shifting would apply, just like it does with BPs, in order to produce the characterizing interpretation. If he did so, however, he would face a problem. As we have seen above, existential readings are not available for definite singulars: (158c) may mean that a uniquely identifiable tiger is chasing a monkey, but not that some tiger or other is doing so. If the definite singular may refer to a kind, DKP ought to apply to it to generate an existential reading. Chierchia is therefore forced to conclude that definite singulars do not refer to kinds after all. Instead, he suggests that they refer to groups. A group is a plurality that is viewed as a “unit”; groups are in the extension of nouns like committee or team. Hence, the tiger denotes the group of all tigers. The group interpretation is generated as follows. Chierchia assumes an additional type-shifting operator, MASS. When MASS applies to a singular property such as dog, it returns the set of individual dogs as well as sets of dogs. When the definite determiner applies to MASS(dog) it returns, as it always does, the maximal element: in this case, just like with plural definites, it is the maximal set of dogs. Chierchia then assumes yet another type-shifting operator, g, which turns this set into a group. Thus, the logical form of the dog, under its generic interpretation, is actually (166) g(ιMASS(dog)). Note that (166) has three nested operators: the MASS type-shifter applies to the predicate, producing the set of individual dogs and sets of dogs. The definite determiner ι applies to the result, producing the maximal set of dogs. The g type-shifter turns this set into a group. Groups are similar to kinds, yet are not the same. The kind ∩ tigers, for Chierchia, is a function that, for every situation s, provides the set of tigers in s. In contrast, the definite generic the tiger is not a function: it denotes the group (rather than the set) of tigers, and in the actual world (rather than a function from worlds to sets). A group is sufficiently different from a kind so as not to be the right sort of argument for a kind predicate, like extinct. Chierchia suggests that, in this case, the group is type-shifted by Montague’s intensionality operator, ˆ. This makes it into a function from situations to groups of individuals, which is very simlar to a kind. Still, it is not quite the same, since a kind is a function from situations to sets of individuals. Nonetheless, Chierchia assumes that kind predicates can

70

chapter 4

apply to such intensionalized groups as well, with exactly the same truth conditions. In this way he generates an interpretation of (167a) that is equivalent to (167b). (167) a. The dinosaur is extinct. b. Dinosaurs are extinct. How ia a characterizing generic produced? Chierchia cannot use the ∪ operator because he “cannot use ‘≤’, as that is defined only for pluralities, not for groups” (p. 381n27). He must therefore introduce yet another operator, which uses the primitive member-of instead of ‘≤’. Again, Chierchia has to assume that the similarity is strong enough to allow for the mechanism of generic quantification to apply. In this case, however, this assumption, in addition to being stipulative, has an unwelcome consequence. Since there is no type mismatch, the intensionality operator does not apply. The result is that a characterizing generic with a definite singular, according to Chierchia’s theory, is predicted to be extensional, surely an undesirable result. While groups are sufficiently similar to kinds to allow definites to be interpreted as direct kind predication and characterizing generics, they are postulated to be sufficiently dissimilar so as to fail to trigger DKP, thus preventing existential readings with episodic predicates. But this distinction between cases where kinds and groups are treated on a par, on the one hand, and cases where they are treated differently, on the other, appears completely ad hoc (cf. Krifka 2004). In fact, an unappealing feature of Chierchia’s approach is that it postulates two types that kind predicates may apply to: kinds and groups. There are no truth conditional or pragmatic differences between them: the sole purpose of making the distinction is to make the system work.6 In my opinion, this added complication, of a group forming operator, which behaves exactly like kinds with respect to some operators, but is different with respect to others, is an aritificial construct. If the definite generic behaves exactly as if it denotes kinds, we must assume that it does.7

6 Of course, it may be that groups themselves have an independent motivation, to explain the semantics of plurals; but their stipulated ability to serve as arguments of kind predicates is not supported by anything other than Chierchia’s theory-internal considerations. 7 Another problem with Chierchia’s approach is the fact the MASS type-shifter would undermine his treatment of bare singulars, as discussed in section 7.3 below.

beyond english bare plurals

71

2.2 Definites as Kind-Referring Why can definite DPs denote kinds? In her explanation, Dayal (2004b) takes as a starting point sentences like (168), which involve quantification over subkinds of lion. (168) a. Every/a/one (kind of) lion is extinct. b. Two/three/most (kinds of) lions are extinct. Dayal writes: Now there are two ways of deriving the taxonomic interpretation, either by building an ambiguity into the denotation of the common noun or into the determiner. Common nouns, under the first approach, would systematically denote properties of ordinary individuals or properties of subkinds, and standard determiners would combine compositionally with the latter to yield taxonomic readings. Alternatively, determiners could be ambiguous between the familiar functions built on properties of individuals and functions that would take properties of individuals but map them onto the taxonomic domain before building up the relevant meanings. For convenience, I will adopt the first approach, though the proposal I want to make regarding definite generics does not hinge on this choice. p. 424

I will follow Dayal’s choice for now, though this decision will be reconsidered in section 7.3 below. According to Dayal, then, when we predicate dog of some individual x, this can mean either “x is a dog” or “x is a kind of dog” (e.g. terrier). Dayal points out that if dog is ambiguous, so is ιdog. If we take dog to denote a property of individuals, ιdog is the unique dog in its extension, which is only defined if there is a unique dog. But if we take dog to mean a kind of dog, then ιdog means the unique kind of dog. What is this element? Dayal proposes that the predicate dog, under its taxonomic reading, is underspecified with respect to the level at which it applies: its extension may include subkinds of dog, but it may also include only the kind ∩ dogs itself. In the latter case, there is only one element in the denotation of the predicate, and the use of the definite is acceptable. For example, (169a) has (169b) as its logical form. (169) a. The lion is likely to become extinct. b. become-extinct(ιlion)

72

chapter 4

If, in context, the denotation of lion (under its taxonomic interpretation) contains only the kind ∩ lions, (169) means that the kind leo is becoming extinct; this is the desired interpretation. Some evidence that a taxonomic reading of a kind can include the kind itself comes from sentences like the following, from Geurts (2001): (170) Fred invented a pumpkin crusher. Since indefinite singulars cannot denote kinds, they can only serve as arguments of kind predicates (like invent) under the taxonomic reading.8 Thus, (171) is normally interpreted as saying that Fred invented some sub-type of computer, not that he is the father of computing. (171) Fred invented a computer. However, sentence (170) would be taken to mean that Fred invented a whole new kind of machine, not merely one sub-type of pumpkin crushers. This is explained by Dayal’s suggestion that the taxonomic reading may refer to the kind itself, not only to its subkinds. Dayal does not discuss the conditions under which context restricts the subkinds of a kind to include only the kind itself. It should be noted that such an interpretation is, in general, dispreferred: we cannot interpret (171) as saying that Fred invented the computer. However, in the case of (170) we have to do this, since the sentence refers to a novel kind, so no subkinds of it are recognized. It is therefore to be expected that, if and when pumpkin crushers become common appliances, (170) would no longer mean that Fred is the father of these machines. Since the interpretation of a kind as containing only itself among its subkinds is dispreferred, the generic interpretation of the definite singular should also be dispreferred. This is, in fact, the case. For example, (172a), under its generic reading, is not a very good sentence—compare it with (172b), which is better, and (172c), which is better still. (172) a. ##The dog is intelligent. b. #The terrier is intelligent. c. The Boston terrier is intelligent.

8 But see Müller-Reichau (2006) for the view that sentences like (170) demonstrate that indefinite singulars can denote kinds.

beyond english bare plurals

73

It appears, then, that the more specific the predicate, the better the definite singular generic. Why does the specificity of the predicate affect the acceptability of the definite generic? The reason, I suggest, is that, although possible, it is not easy to interpret a predicate like dog as excluding all subkinds of dog from its extension. Most of us, even those who don’t have Macbeth’s extensive knowledge that “hounds and greyhounds, mongrels, spaniels, curs, shoughs, water-rugs, and demi-wolves, are clept all by the name of dogs”, are simply too familiar with various subkinds of dog. But if we keep subkinds of dog under consideration, we lose uniqueness, and the definite is undefined. This is why (172a) is somewhat odd. The more specific the predicate, the lower the level of the kind it corresponds to; hence it is easier to ignore its subkinds, since normally we are not interested in subkinds that are of too low a level. This is why (172b) is better, and (172c)—perfect. Compared with (172a), (164a), which is repeated below as (173), is better, because we are less familiar with subkinds of tiger, hence it is easier to exclude them from the extension of the predicate. (173) The tiger lives in the jungle. I have no explanation for the different distribution of definite singular generics across languages, exemplified by the higher acceptability of the German (165), repeated below. (174) Der Hund bellt. ‘The dog barks.’ I can, however, offer a tentative speculation. The availability of the definite singular is inversely proportional to the salience of taxonomic level of subkinds. One need not buy the stories about Eskimo words for snow (Pullum 1989) in order to accept the assumption that if a language has more terms referring to the subkinds of a certain kind, this would affect the salience of those subkinds in that language. Perhaps the difference between the availabilities of the generic interpretation of der Hund vs. the dog has to do with a difference between the two languages in the richness of the vocabulary used to describe breeds of dogs.9 9 Compare Blok and Eberle (1999). They consider the following German sentence and its English translation:

74

chapter 4

The proposal that specificity is what facilitates the generic interpretation of definite singulars is in contrast with a commonly held view (see, e.g., Krifka et al. 1995) that definite singular generics are only possible with predicates that denote “well-established kinds.” A classic argument for this claim, originating with Barbara Partee, involves the following minimal pair: (175) a. The Coke bottle has a narrow neck. b. #The green bottle has a narrow neck (fine non-generically). It is claimed that Coke bottle is a well-established kind, whereas green bottle is not; and that this is the reason why (175b) cannot be interpreted generically. I suggest, instead, that the difference is one of specificity: Coke bottle is a very specific kind of bottle, and it is hard to envision subkinds of it; in contrast, it is quite natural to think of subkinds of green bottle, e.g. green beer bottle, green wine bottle, etc. Indeed, in a context that clearly places green bottle at the bottom of the taxonomy, (175b) is perfectly fine. Consider the following scenario, from Dayal (1992): You are on a tour of a plant which makes bottles and the tour guide says, “we manufacture three types of bottles at this plant, green, blue and clear. The green bottle is our particular specialty. It has a long neck.” I think there is no problem now in a generic interpretation for the singular term p. 56

The differences between the two theories (the requirement for “well-established kinds” or the specificity requirement proposed here) can be illustrated clearly with the following example, from Cohen (1996). Contrast the oddness of (176a) (under the generic reading) with the acceptability of (176b). (176) a. #The politician never misses a photo opportunity. b. The successful politician never misses a photo opportunity.

(i) a. John trinkt nur pils. b. John only drinks lager. Blok and Eberle state: “All our German and English informants confirmed that … the alternatives of which the native speaker is aware in [(i.a)], are those kinds of beer that have a name in German, and correspondingly for [(i.b)]” (p. 111).

beyond english bare plurals

75

Clearly, “successful politician” is more specific than “politician”, which explains the difference in acceptability between the members of this pair of sentences. Yet one would be hard-pressed to argue that successful politician is a well-established kind, whereas politician is not. 2.3 Definite Plurals Having discussed definite singulars, let us now turn our attention to plurals. I assume a completely straightforward view of the plural morpheme: it turns a property of objects of type τ to a property of pluralities of type τ—I remain agnostic here on what pluralities actually are. Just like the predicate dog is ambiguous between properties of individual dogs and properties of kinds of dog, dogs is also ambiguous: it can have in its extension pluralities of dogs, or pluralities of kinds of dog. When dogs is combined with the definite determiner, the result is correspondingly ambiguous. If dogs is interpreted to be about pluralities of dogs, the result is the maximal plurality of dogs.10 If we interpret dogs to be about pluralities of kinds of dog, the result is the maximal plurality of kinds of dog, i.e. a plurality consisting of the kinds ∩ terriers, ∩ bulldogs, etc., as well as the kind ∩ dogs itself. That is to say, the dogs denotes all the dogs or all the kinds of dog, depending on whether dogs is read taxonomically or not. Because of its plural meaning, what the dogs normally does not denote is the singleton consisting only of the kind: (177a), in contrast with the BP sentence (177b), can only mean that all the subspecies of dogs are in no danger of extinction. (177) a. The dogs are in no danger of extinction. b. Dogs are in no danger of extinction. Suppose, for example, that terriers, bulldogs, and rottweilers were, in fact, in danger of extinction, but no other type of dog was. Hence, dogs, as a species, were quite safe. Then (177a) would be false, although (177b) would be true. The definite plural the dogs, then, cannot normally refer to the kind ∩ dogs in English. And, in general, definite plurals in English cannot usually denote kinds: even in a context in which no subkinds of ∩ dogs are considered, i.e. where the denotation of dogs (under the taxonomic reading) is the singleton whose sole member is ∩ dogs, it is not possible to refer to a singleton with a plural noun.

10

Note that this is not the same as the kind ∩ dogs—see below.

76

chapter 4

In English, of course, there is a way to refer to kinds with plural nouns: to use BPs. In this case, the plural predicate dogs is type-shifted to the kind ∩ dogs, and there is no plural reference to a singleton.11 2.4 Definites in Italian We now turn to Romance, exemplified by Italian. Dayal proposes that, in Italian, the definite determiner is ambiguous. In English, the meaning of the P is ιP, and this is also one reading of the definite in Italian. But there is another reading, which, in a sense, is an intensional version of the first one: λs.ιPs (where s is a world or situation variable). Now recall that this is precisely the definition of the ∩ type-shifting operator—definition 5. This means that, in Italian, the type-shifting operation ∩ can be performed overtly by the definite determiner. When the definite applies to a plural noun, the result can be a kind (the other possibility is, of course, simply a definite plural). Consequently, BPs in Romance cannot refer to kinds, because the type-shifting is blocked: its job is taken over by an overt operator. Why do the meanings of the definite in English and Romance differ in this way? Recall that this difference entails that, in English, BPs can refer to kinds but definite plurals cannot, while in Romance, definite plurals can refer to kinds but BPs cannot. An interesting way to characterize this distinction has been proposed by Farkas and de Swart (2007), using the notions of Optimality Theory. A kind, as we have seen, refers to the maximal element in the denotation of its defining property. On the one hand, this calls for kinds to be referred to by the definite determiner, since, normally, it serves as the expressions of maximality. On the other hand, the definite determiner is also normally used to refer to individuals that have been introduced previously, but usually a kind does not have an antecedent in previous discourse; hence, we would not expect the definite determiner. Some languages, such as Romance, prefer to express maximality over novelty, hence they use the definite determiner as a function from 11

In contexts in which the kind is already introduced in prior discourse using a plural form—a BP—definite singular generics are, in fact, possible. Consider the following example, which Farkas and Sugioka (1983) found in the Encyclopedia Britannica: (i) Saurischian Bipeds—The saurischians were the first of the two great groups to assume prominence … From certain of these forms, the saurischians were certainly derived. The heading of the article (in bold) is a BP that introduces the kind; later references to the kind can be made using a definite plural. Presumably, in this context the saurischians is interpreted as something like a quotation, i.e. “the kind that was previously introducied as saurischians”.

beyond english bare plurals

77

indices to maximal sets, and allow it to produce a kind when combined with a plural noun. Other languages, such as English, prefer to express novelty over maximality, hence they do not allow the definite determiner to be interpreted in this manner.12

3

Italian BPs: Existential Readings

Italian, then, is a language where BPs may not refer to kinds. It is therefore important to find whether it allows existential readings of BPs. If it does not, this fact would indicate that kind reference is indeed, as Chierhcia argues, essential for an existential interpretation. If it does, however, it would follow that existential readings are produced by other means, such as the ∃ typeshifting operator. All too rarely, an important question can be answered clearly and uncontroversially. But this is the case here: there is a consensus that existential readings of BPs are readily available. Moreover, just like in English, these readings exhibit narrow scope only. This can be demonstrated by the following examples (Chierchia 1998: ex. 83, p. 388): (178) a. Non ho conosciuto belle ragazze. ‘(I) didn’t meet beautiful girls.’ b. Ho ucciso conigli ripetutamente. ‘(I) killed rabbits repeatedly.’ c. Voglio conoscere belle ragazze. ‘(I) want to meet beautiful girls.’ Sentence (178a) can only mean that it is false that there exist beautiful girls whom I met. It cannot get the interpretation where there exist beautiful girls whom I didn’t meet. Sentence (178b) exemplifies differentiated scope: it can only mean that each killing involved different rabbits. And (178c) demonstrates that BPs can only be read de dicto in opaque contexts: the sentence cannot mean that there are some beautiful girls that the speaker wants to meet.

12

Of course, the definite can still pick up a kind if it is the unique individual in the denotation of the noun it combines with, as is the case with definite singular generics.

78

chapter 4

This fact poses a problem for Chierchia, since existential readings are supposed to be derived from kind reference by DKP: otherwise, he would predict incorrectly that they exhibit normal scope ambiguities, just like parts of that machine (cf. chapter 3, section 4.4). As we have seen, Chierchia is therefore forced to argue that reference to kinds is possible in Italian, but the evidence he uses to support this claim is quite weak. In contrast, this fact is quite consistent with both Krifka (2004) and the current proposal, since for both, the existential reading of a BP does not require the BP to denote a kind, but rather a property.

4

Italian BPs: Characterizing Generics

4.1 Are Characterizing Generics Possible? If characterizing generics constitute reference to properties (Krifka), such readings should be freely available in Italian. If, on the other hand, characterizing generics constitute kind reference (Chierchia and the current proposal), such readings should not be available at all. At first sight, it looks like the second prediction is clearly borne out: indeed, many characterizing generic BPs are bad in Italian: (179) *Linguisti sono bravi. ‘Linguists are clever.’ (Chierchia 1998) (180)

??Cani da guardia di grosse dimensioni sono più pelosi/neri. ‘Watchdogs of large size are hairier/blacker.’ (Longobardi 2001)

However, it turns out that, at least in some dialects, characterizing generics do seem to be possible, though their distribution is severely restricted. For example, Longobardi contrasts the unacceptable (180) with the fine (181). (181) Cani da guardia di grosse dimensioni sono più efficienti/aggressivi. ‘Watchdogs of large size are more efficient/aggressive.’ There are, admittedly, dialects of Italian where (181) is bad; but the fact is that in other dialects, it is good. We therefore cannot take the easy way out and simply ignore it and sentences like it. The facts, then, appear to be unexpected under any of the proposed theories (Chierchia, Krifka, and the current proposal). Characterizing generic readings of Italian BPs are neither freely available nor ruled out altogether.

beyond english bare plurals

79

How can we explain this situation? Before we do, let us first get one red herring out of the way. 4.2 The Heaviness Requirement The subject of (181) is rather “heavy”; could this be the reason why the sentence is good? The answer is no. The heaviness requirements for subjects is not related to the possibility of a characterizing interpretation. First, we have already seen an example of a sentence with a heavy subject BP that, nonetheless, cannot get a generic interpretation: (180). Second, the heaviness requirement of Italian subject BPs applies even when they are interpreted existentially. For example, Chierchia points out that the sentences in (182) are bad; but when the subject becomes heavy, as in (183), the sentences become considerably better. (182) a. *Studenti hanno telefonato. students have phoned b. *Hanno telefonato studenti. have phoned students ‘Students have phoned.’ (183) a. Studenti e colleghi hanno telefonato. students and colleagues have phoned ‘Students and colleagues have phoned’ b. Hanno telefonato studenti che volevano apere la data have phoned students that wants to find out the date dell’esame. of the exam ‘Students who want to find out the date of the exam have phoned.’ Whatever the reason for the heaviness requirement is, it is clear therefore that this requirement is orthogonal to the possibility of a characterizing generic interpretation of the BP. An account of the heaviness requirement is, of course, important in its own right, and deserves a short detour. Dayal (2004a) makes an attractive proposal. Following Enç (1986), she proposes that every noun phrase must have, in addition to a referential index, a temporal situation index. Dayal assumes that bare plurals lack such a situation variable. When inside the VP, the BP may

80

chapter 4

inherit the temporal specification of the c-commanding verb. However, this is impossible when the BP is in subject position, which is why Italian BPs are generally bad in subject position. But a heavily modified BP in subject position manages to escape this restriction, and is acceptable. Why? Based on evidence involving the behavior of subtrigged any, Dayal proposes that modification, in itself, provides a temporal situation variable; thus, a heavily modified BP acquires the necessary temporal index, and is therefore acceptable. Why doesn’t the heaviness requirement apply to English? Dayal points out that English BPs have an option that Italian BPs lack, namely the option to refer to kinds. Recall that the kind ∩ dogs is a function that provides for every situation the sum of all dogs in that situation: λs ιdogs. It is this situation variable that satisfies the requirement for the BP to have a temporal index. Crucially, for this explanation to work, the situation variable must be temporal, rather than, say, an atemporal world variable. In other words, kinds are intensional, but their intensionality is intimately related to the time index, rather than (or, perhaps, in addition to) the world index (cf. Cohen 1999). In this way, the acceptability of non-heavy generic BPs in English is accounted for. What about existential BPs? Since Dayal believes that existential BPs denote kinds, she applies the same solution to them too. This way, however, is not to open to us, if we accept the proposal made in this book that existential BPs in English denote properties. There is, however, an alternative explanation. Based on evidence from Dutch, Van Geenhoven (1996) argues that existential BPs are not really subjects in English, and the real subject is a null pleonastic. If so, the BP is inside the VP, and receives its temporal index from the verb in the normal way. 4.3 A Comparison with English Indefinite Singulars So the heaviness requirement is no help, and we are left with the problem of the characterizing generic reading of BPs. Why is it acceptable at all, and why only in restricted cases? The situation is reminiscent of the case of indefinites singulars in English. There are quite a few similarities between Italian BPs and English ISs. For instance, as we have seen in chapter 3, section 4.3 above, English ISs also cannot denote kinds (unless, of course, they are used taxonomically).13 13

Based on this fact (and a number of others), Longobardi (2001) suggests that Italian BPs, on both their characterizing and existential readings, are free variables, like English ISs. However, this proposal cannot explain the distribution and scopal behavior of Italian BPs described above.

beyond english bare plurals

81

Another relevant similarity is the fact that generic readings of Italian BPs are severely restricted. For example, as we have seen, (180), repeated below as (184a), is odd; compare it to the unacceptable English IS sentence (119b), repeated below as (184b). (184) a. ??Cani da guardia di grosse dimensioni sono più pelosi/neri. ‘Watchdogs of large size are hairier/blacker.’ (Longobardi 2001) b. *A big watchdog is hairy/black. Note further that just like with English ISs, an overt modal or quantificational adverb can turn a bad BP generic to an acceptable one (Longobardi 2001). Thus, while (185a) is odd (except under an existential reading), (185b) and (185c) are fine. (185) a. *Elefanti di colore bianco hanno creato in passato grande curiosità. ‘White-colored elephants raised a lot of curiosity in the past.’ b. Elefanti di colore bianco possono creare grande curiosità. ‘White-colored elephants may raise a lot of curiosity.’ c. Elefanti di colore bianco hanno creato sempre/spesso in passato grande curiosità. ‘White-colored elephants always/often raised a lot of curiosity in the past.’ 4.4 Generic Italian BPs Are Normative I therefore propose that acceptable BP generics in Italian, just like English ISs, are normative: they predicate of a rule that it is in effect. Consider some minimal pairs, of acceptable (at least in some dialects) vs. unacceptable BPs (Longobardi 2001): (186) a. Cani da guardia di grosse dimensioni sono più efficienti/aggressivi. ‘Watchdogs of large size are more efficient/aggressive.’ b. ??Cani da guardia di grosse dimensioni sono più pelosi/neri. ‘Watchdogs of large size are hairier/blacker.’ (187) a. Stati di grandi dimensioni sono pericolosi. ‘States of large size are dangerous.’

82

chapter 4

b. ??Stati di grandi dimensioni sono prosperi. ‘States of large size are prosperous.’ (188) a. Uccelli di zone paludose sono ghiotti di insetti. ‘Birds from marshy areas are greedy for insects.’ b. ??Uccelli di zone paludose sono scuri/intelligenti. ‘Birds from marshy areas are dark/intelligent.’ Longobardi notes that the predicates of good BP generics are “somewhat more eventive” than those of bad BP generics. We can explain this observation as follows. We have seen that one of the ways English ISs can receive a quasi-generic reading is if they call for action—if they are action triggers. For example (130b), repeated below, can be interpreted as calling on the hearer to open doors for ladies. (189) A gentleman opens doors for ladies. The good BP generics are “more eventive” because they, too, constitute calls for action—they are action triggers, hence conceivable rules. Bad ones, in contrast, are not. For example, (186a) can quite naturally be seen as a call for some action, and can naturally be followed by something like (190); but it is hard to think of an action that would be triggered by (186b). (190) … perquesto evitare Fido! ‘… therefore avoid Fido!’ Similarly, the continuations in (191a) and (191b) make perfect sense; but what sort of actions can reasonably be triggered by (187b) or (188b)? (191) a. Stati di grandi dimensioni sono pericolosi, perquesto attaccateli. ‘States of large size are dangerous, therefore attack them!’ b. Uccelli di zone paludose sono ghiotti di insetti, perquesto date a Tweety degli insetti! ‘Birds from marshy areas are greedy for insects, therefore feed Tweety some insects!’ And yet, there are two obvious differences between Italian BPs and English ISs: the former are bare plurals, whereas English ISs are neither bare nor plural. But do English BPs also have a normative reading?

beyond english bare plurals

83

4.5 Normative English BPs We have seen that English ISs are normative: they state that some norm is in effect. While the most natural interpretation of BPs is descriptive, they, too, can be read normatively. Consider the following sentence, for example: (192) Police officers do not behave like that in an emergency situation. The sentence can be read as making an observation about the way police officers behave (uttered by, say, a reporter who is preparing a piece about the police force). But (192) has an additional reading: it can, for example, be uttered by the Police Chief, chastising some officers under his command. Under this reading, the sentence tells the police officers what (not) to do, rather than says what they actually do do. Note that just like the normative reading of ISs, normative BPs are restricted to wide scope only: the BP in (192) takes scope over the negation operator. Hence, (192) can only mean that there is a rule that prohibits police officers from behaving like that, not that there is no rule that forces them to behave like that. For example, if the behavior in question is panicking, (192) means that the norm of police behavior is that they do not panic in an emergency situation. But the sentence cannot get the reading where negation takes wide scope, i.e. it cannot mean that there is no norm of police behavior according to which they must panic in an emergency situation. A natural question to ask at this point is whether this ambiguity is somehow related to the generic/existential ambiguity of BPs. I have argued that the generic reading requires reference to kinds, whereas, under the existential reading, BPs denote properties. Since the normative reading has a generic “flavor”, most researchers have implicitly assumed that it is really nothing but the generic reading. If we accept the view of ISs proposed here, this position would mean that there is a fundamental difference between the normative readings of BPs and ISs: the former express quantification with the generic quantifier, whereas the latter do not involve quantification, but rather predicate of a rule or a regulation that it is in effect. I would like to propose an alternative, though. While the kind-denoting, generic reading of BPs is unambiguously descriptive, the property-denoting reading is ambiguous: as discussed above, the BP could be subject to the ∃ type-shifter, to receive a narrow scope existential reading; but it can also be interpreted normatively, as predication over a rule or regulation.14

14

In this case, since there is no type-shifting, the BP receives wide rather than narrow scope.

84

chapter 4

Some evidence for this position comes from French. In this language, BPs are almost never allowed. Generic readings, either direct kind predication or characterizing generics, use the definite determiner les; existential readings use the indefinite determiner des. We can say that les indicates reference to kinds (resulting in one of the types of generic reading), while des indicates reference to properties (resulting in the existential reading). Interestingly, when a sentence is interpreted normatively, this is indicated by the indefinite determiner, not the definite (generic) determiner (Carlier 1989, as reported by de Swart 1996). Thus, a reporter writing a piece about the police will say (193a), whereas a superior chastising a subordinate will use (193b). (193) a. Les agents de police ne se comportent pas ainsi dans une situation d’alarme. ‘DEF-PL police officers do not behave like that in an emergency situation.’ b. Des agents de police ne se comportent pas ainsi dans une situation d’alarme. ‘INDEF-PL police officers do not behave like that in an emergency situation.’ Plausibly, in English too, the normative reading of BPs is not generated by their kind-denoting, generic reading, but by the property-denoting interpretation. We can extend this conclusion to Italian BPs too: Italian BPs, as we have seen, denote properties and cannot denote kinds. These properties can get a narrow scope existential interpretation, but they can also, like property-denoting English BPs, receive normative readings.

5

An Account of Italian BPs

Let us put together our findings. Italian BPs do not denote kinds; rather, they denote properties. Consequently, they cannot participate in sentences expressing direct kind predication. They can get an existential reading, where the property is type-shifted to an existential quantifier. Since type-shifting is local, we get a narrow scope reading. What looks like a generic sentence involving an Italian BP is not really generic, and does not involve the generic quantifier. Rather, it says that some

beyond english bare plurals

85

rule (in the cases discussed by Longobardi, it is an action trigger) is in effect. Hence, there is no need to assume that the BP denotes a kind in such a sentence; it still denotes a property, just like it does under the existential reading. Hence, the proposal made regarding English BPs applies equally well to Italian: existential BPs denote properties, or may receive a normative reading; generic BPs denote kinds. Since Italian BPs denote only properties, and not kinds, they can get existential and normative readings, but not generic ones.

6

Hungarian Bare Plurals

It should be emphasized that this account of Italian BPs is not dependent on idiosyncratic properties of Italian, or even Romance languages in general. Other, typologically unrelated languages, behave similarly: if BPs cannot denote kinds, they cannot receive characterizing generic readings, and only existential (and normative) readings are possible. Consider Hungarian, for example. Just like Italian BPs, Hungarian BPs cannot refer to kinds, and the definite determiner is required: (194) *(A) dinoszauruszok kihaltak. (the) dinosaurs died-out ‘*(The) dinosaurs died out.’ As predicted, characterizing generic readings are impossible (Farkas and de Swart 2007): (195) *Kutyák veszélyesek mikor éhesek. Dogs dangerous when hungry ‘*Dogs are dangerous when hungry.’ And yet, when the interpretation is that of a rule or a regulation, BPs are acceptable, as pointed out to me by Beata Gyuris (pc): (196) Rendőrtisztek nem viselkednek így szükséghelyzetben. Police-officers not behave-so in-emergency-situation ‘Police officers do not behave like that in an emergency situation.’ Interestingly, a definition is not sufficient to permit BPs in Hungarian:

86

chapter 4

(197) *Madrigálok polifonikusok. Madrigals polyphonic ‘*Madrigals are polyphonic.’ The explanation for this fact, however, may be syntactic. É. Kiss (2002) argues that BPs in Hungarian are not subjects; therefore, in order to satisfy the EPP, something else must be the subject.15 In sentence (196), the predicate is stagelevel, so it introduces a Davidsonian argument (Kratzer 1995; cf. the discussion in chapter 3, section 4.1 above), which may serve as the subject; the EPP is satisfied, and the sentence is fine. In contrast, in (197) the predicate is individual-level, and there is no Davidsonian argument. Hence, there is no subject, and the sentence is ruled out as a violation of the EPP.

7

Bare Singulars

7.1 Bare Singulars and Type-Shifting English does not generally allow bare singulars in argument positions, but some languages do. Bare singulars are usually available in languages that do not have determiners, like Hindi and Russian. We have seen in chapter 2, section 2.5 above, that bare nouns in such languages can receive both existential and definite interpretations, as illustrated by Chierchia’s (1998) Russian example, repeated below: (198) V komnate byli malcik i devocka. Ja obratilsja k malciky. in room were boy and girl I turned to boy ‘In (the) room were (a) boy and (a) girl. I turned to (the) boy.’ Chierchia argues that these interpretations are generated by type-shifting: the definite interpretation is the result of type-shifting by ι, and the existential interpretation is brought about by type-shifting with ∃. This view immediately raises a number of questions. If the existential interpretation is the result of type-shifting, does it receive narrow scope only? Can the definite interpretation be used to refer to a kind, like it does with English definite singulars? Can bare singulars refer to kinds through application of ∩ ? 15

As we have seen in section 4.2 above, Van Geenhoven (1996) argues that English BPs are also not subjects, and that the real subject is a null pleonastic. This option, however, does not seem to be open for Hungarian.

beyond english bare plurals

87

Do they have characterizing generic readings? And what about languages like Hebrew, which allows bare singulars even though it has a definite determiner, or like Brazilian Portuguese, which has both definite and indefinite determiners, yet also allows bare singulars? I this section we will examine these questions, and consider whether they can receive answers that are compatible with the theory proposed here. 7.2 Existential Readings As we have seen, Chierchia argues that the existential reading of bare singulars in Russian and Hindi is generated through type-shifting. Hebrew has a definite determiner, but lacks an indefinite one; hence, the existential reading in Hebrew is plausibly generated in the same way. However, there is at least one language that presents us with a puzzle: Brazilian Portuguese has both definite and indefinite determiners, yet allows bare singulars in argument positions. If the existential reading is generated by typeshifting, why isn’t it blocked by the indefinite determiner? Recall that Chierchia (1998) and Krifka (2004) argue that, in English, the existential reading of BPs is not blocked by some, because this determiner provides more information than simple existential quantification. Specifically, it introduces a choice function. Evidence for this claim was presented in chapter 3, section 3.3, where it was pointed out that some cannot introduce a variable to be bound by an adverbial quantifier: (199) a. A dog is usually intelligent. b. *Some dog is usually intelligent. c. *Some dogs are usually intelligent. Krifka hypothesizes that the same account can be made of the indefinite determiner in Brazilian Portuguese, and that it introduces a choice function. This, however, is doubtful, given that the indefinite determiner can, in fact, be bound by an adverbial quantifier, as the following example, from Müller (2001), indicates: (200) Uma lagartixa sempre perde seu rabo. A gecko always loses its tail ‘A gecko always loses its tail.’ There is, however, an alternative explanation. Bare singulars in Brazilian Portuguese, although morphologicaly singular, are semantically number neutral (Müller 2001; Munn and Schmitt 1999, 2005).

88

chapter 4

For example, Müller notes that seu, the anaphoric third person possessive, gets different interpretations relative to the type of its antecedent. If the antecedent is an overt IS, as in (200), the nominal complement of seu is interpreted as atomic, and the sentence means that every gecko loses one tail. However, if the antecedent is a bare singular, as in (201), the nominal complement is unmarked for number, and the sentence allows that each gecko loses an unspecified number of tails. (201) Lagartixa sempre perde seu rabo. gecko always loses its tail ‘A gecko always loses its tail.’ For another demonstration of the number neutrality of bare singulars in Brazilian Portuguese, note that Munn and Schmitt (2005) point out that either (202b) or (202c) can felicitously follow (202a). (202) a. Eu vi criança na sala. I saw child in-the room ‘I saw child in the room.’ b. E ela estava ouvindo. And she was listening ‘And she was listening.’ c. E elam estavam ouvindo. And they were listening ‘And they were listening.’ Thus, the meanings of the overt IS and the bare singular are not the same: while the former is singular, the latter is unspecified for number. Hence, the indefinite determiner does not block the existential reading of the bare singular. In all languages under consideration, then, the existential reading of bare singulars is generated by type-shifting. We therefore predict that existential bare singulars receive narrow scope only. Is this prediction borne out? The facts about Brazilian Portuguese appear to fulfill this prediction, as the following examples, from Schmitt and Munn (1999), demonstrate: (203) a. Pedro quer encontrar um policial. ‘Pedro wants to meet a policeman.’ (ambiguous: transparent or opaque)

beyond english bare plurals

89

b. Pedro quer encontrar policial. ‘Pedro wants to meet policeman.’ (only opaque) (204) a. João não viu uma mancha no chão. ‘João didn’t see a spot on the floor.’ (scopally ambiguous) b. João não viu mancha no chão. ‘João didn’t see spot on the floor.’ (only ¬ > ∃)

(205) a. Todo mundo leu um livro sobre girafas. ‘Everyone read a book on giraffes.’ (scopally ambiguous) b. Todo mundo leu livro sobre girafas. ‘Everyone read book on giraffes.’ (only ∀ > ∃)

The same phenomenon obtains in Hebrew, as the following examples, from Danon (2008), indicate: (206) a. dan mexapes iša. Dan seek woman ‘Dan is seeking a woman/wife.’ b. dan xošev še- kelev našax oti. Dan thinks that dog bit me ‘Dan thinks that a dog bit me.’ Sentence (206a) can only be read de dicto: it cannot mean that there is a particular woman that Dan is seeking. Similarly, (206b) can also be read only de dicto. Suppose I was bitten by a dog, and Dan saw the event, but he thinks the animal in question was actually a wolf. Then, although there is a dog such that Dan thinks of this dog that it bit me, (206b) cannot get this de re reading and is, therefore, false. The following attested example makes this point even clearer. (207) kshe-natan ha-kadosh barux hu ‘et ha-tora when-gave the-holy blessed he ACC the-Torah tsipor lo tsavax bird not screamed

90

chapter 4

‘of lo parax fowl not flew shor lo ga‘a ox not lowed ela ha-‘olam shotek ‘u-maxrish but the-world silent and-speechless ‘When God gave the Torah, no bird twittered, no fowl flew, no ox lowed, but the whole world was hushed into breathless silence.’ (Exodus Rabbah, 29:9) Since the last clause states that the entire world was silent, it is clear that negation takes wide scope over the bare singular: the sentence says that no bird, fowl, or ox made noise, not that some of them didn’t. In Hindi and Russian, too, the existential reading of bare singulars appears to receive narrow scope only. The following examples are from Dayal (2004b): (208) a. mujhi lagtaa hai ki kamre meN cuuha ghuum rahaa hai. to-me seems that room in mouse moving around is ‘It seems to me that a mouse is moving around in the room.’ (Hindi) b. mne kazhetsja chto v komnate mysh. me seems that in room mouse ‘It seems to me that there is a mouse in the room.’ (Russian) c. kot ne sidit na stule. cat not sits on stool ‘There isn’t any cat sitting on the stool.’ (Russian) Dayal points out that (208a) and (208b) can only be interpreted with seem taking scope over the bare singular; and (208c) can only mean that no cat is sitting on the stool, not that there is a cat that doesn’t (and perhaps others that do). There is, however, a fly in the ointment. Dayal presents other examples, in which the bare singular appears to get wide scope; moreover, the wide scope is obligatory in such cases. The following examples are from Hindi: (209) a. #caroN taraf baccaa khel rahaa thaa. four ways child was-playing ‘The same child was playing everywhere.’

beyond english bare plurals

91

b. #kamre meN bacca khel rahaa thaa aur bacca room in child was-playing and child so (bhii) rahaa thaa. was-sleeping (also) ‘The same child was playing and sleeping in the room.’ Sentence (209a) can only mean that the same child was playing everywhere, and (209b) can only mean that the same child was both playing and sleeping, hence their oddity. Dayal presents similar examples from Russian, which are odd for the same reason: (210) a. #Sobaka byla vesde. dog was everywhere ‘The dog/A particular dog (the same one) was everywhere.’ b. #V etoj kletke, tigr jest i tigr spit. In this cage, tiger is eating and tiger is sleeping ‘In this cage, the same tiger is eating and sleeping.’ How can this be? According to Dayal, in these cases the bare singular actually receives a definite reading, via the ι type-shifting operator. This is surprising, since these examples are quite acceptabe in contexts where the referent of the bare singular is not familiar. Dayal is therefore led to postuate that in such contexts the definite requires only uniqueness, not familiarity. Moreover, to avoid a narrow scope existential interpretation, she stipulates that ι outranks ∃, and blocks its application. There is, however, a simpler solution, which maintains the normal meaning of definites, and does not require the stipulation of a hierarchy between type-shifting operators. Following Krifka (2004), we can propose that existential bare singulars presuppose uniqueness. The bare singulars in (209) and (210) do receive narrow scope only; however, they presuppose uniqueness, hence their oddity. Uniqueness is also presupposed in the examples in (208); the difference is that in these cases, uniqueness makes sense, and does not result in oddity: the existence of a unique mouse in context does not require that the speaker identify the mouse; and the existence of a unique cat in context is quite compatible with the fact that no cat is sitting on the stool. Some evidence for the uniqueness presupposition of bare singulars actually comes from a language like English, which usually uses overt indefinite singulars, but also, in restricted cases, bare singulars. In those cases where bare singulars are used, they appear to presuppose uniqueness.

92

chapter 4

One context where bare singulars may be allowed is coordination. Heycock and Zamparelli (2003) note that, in this context, the bare singular requires uniqueness. For example, since there is only one president and one vice president, (211a) is fine; but since there are typically many employees and inspectors, (211b) is odd. (211) a.

At the company meeting, president and vice president gave an optimistic speech. b. ??At the company meeting, employee and inspector talked about their colleagues’ motivation.

Another context where bare singulars are permitted is illustrated in (212), from Munn and Schmitt (2005). (212) The President made Powell chairman of the Joint Chiefs of Staff. Munn and Schmitt point out that here, again, the bare singular presupposes uniqueness. Sentence (212) is fine, since there is only one chairman of the Joint Chiefs of Staff. However, (213) is bad, since there are typically several 5 star generals. (213) *The President made Powell 5 star general. 7.3 Reference to Kinds Chierchia (1998) points out that while the BP is fine in the Russian example (214a), the bare singular in (214b) is bad. (214) a. Dinosavry vymerly. Dinosaurs died-out ‘Dinosaurs are extinct.’ b. *Dinosavr vymer. Dinosaur died-out As discussed in chapter 3, section 3.1 above, he argues that the inability of bare singulars to denote kinds follows from his definition of the ∩ operator, which is repeated below: (215)

∩P

=def λs ιPs (where defined).

beyond english bare plurals

93

According to this definition, if P is a singular property, and there are worlds where there is more than one individual satisfying P, ∩ P will be undefined. The claim that bare singulars cannot denote kinds has not gone unchallenged. It appears that in Brazilian Portuguese, bare singulars may, in fact, denote kinds, as the following examples from Dobrovie-Sorin and Pires de Oliveira (2004), indicate:16 (216) a. Baleia está em extinção. Whale is in extinction ‘Whales are on the verge of extinction.’ b. Rato foi introduzido na Austrália em 1770. Rat was introduced in-the Australia in 1770 ‘Rats were introduced in Australia in 1770.’ Does the fact that Brazilian Portuguese bare singulars can denote kinds constitute a counterexample to Chierchia’s definition? The answer is no. Recall that bare singulars in this language are number neutral, i.e. they have in their denotation both individuals and pluralities. Dobrovie-Sorin and Pires de Oliveira point out that, in this case, the ∩ operator is well defined; for example, ∩ baleia, just like the English ∩ whales, is a function that, for each situation, returns the maximal set of whales in this situation. As an aside, recall, from section 2.1 above, that Chierchia’s theory leads him to postulate a MASS type-shifting operator, which turns a singular property into a number-neutral property. But if such an operator existed, it would be able to apply to all bare singulars, in all languages, hence bare singulars would be able to denote kinds, undermining Chierchia’s own generalization. A different challenge to Chierchia’s generalization comes from Hindi, where bare singulars are not number neutral (Dayal 2004b: 407n11). Dayal claims that, in Hindi, bare singulars can denote kinds, and presents the following example: (217) kutta aam janvar hai. dog common animal is ‘Dog is a common animal.’ However, the significance of this example is not clear. As discussed in the case of Italian, it is doubtful that the predicate be common is really a kind predicate.

16

But see Müller (2003) for different judgments.

94

chapter 4

Moreover, as pointed out to me by Ashwini Deo (pc), the predicate in this example is aam janvar ‘common animal’, rather than simply aam ‘common’, and without such a definitional form, the bare singular cannot denote a kind. Thus, (218) is quite marked. (218) ??Is ilake-me kutta aam hai. this area-in dog common is ‘Dog is common in this area.’ When we use clear cases of kind predicates, the bare singular is not acceptable: (219) *Jaise jangal ghanaa hotaa jaata hai, lomdi badi hotii jaa-tii hai. ‘As the jungle becomes denser, wolf becomes bigger.’ The presence of overt taxonomic information, however, may enable a kind reading in cases where the sentence may be read as a definition, just like it does with with English indefinite singulars, as demostrated by (129a), repeated below: (220) A dinosaur was a large reptile which lived in prehistoric times and which is now extinct. Indeed, if (217) and (218) are translated into English, with an indefinite singular replacing the bare singular, similar acceptability judgments obtain: (221) a. A dog is a common animal. b. ??A dog is common in this area. A similar account can be made of Hebrew. This language has a definite determiner, but not an indefinite one. Hence, it allows bare singular nouns in argument positions, and they are interpreted existentially. No definite interpretation is possible, and no kind reference; both are available with the overt definite determiner: (222) a. kelev noveax. dog barks ‘A dog is barking (not: The dog is barking).’ b. *(ha) namer nimtsa be-sakanat hakxada. *(the) leopard is in-danger(of) extinction

beyond english bare plurals

95

‘*(The) leopard is in danger of being extinct.’ Interestingly, Doron (2003) points out that there are cases when Hebrew bare singulars can refer to kinds, as in: (223) namer hu min be-sakanat hakxada. leopard he kind in-danger(of) extinction ‘The leopard is a kind in danger of being extinct.’ I suggest that, once again, the definitional form of a sentence is the factor that makes kind predication acceptable. It might be thought that there is an alternative way for bare nouns to denote kinds. If, in languages like Hindi and Russian, they can be interpreted as definite, we might expect them, just like the overt (singular) definite in English, to be able to denote kinds. Just like in English, this would be brought about when the predicate is interpreted taxonomically. Why isn’t such a reading available? Recall that, in her discussion of taxonomic readings, Dayal considers two options: either the common nouns themselves are ambiguous, or the determiners that apply to them are. She chooses the first option, though she emphasizes that her proposal does not hinge on this choice. There is, however, some evidence that the second choice is preferable: the ambiguity between taxonomic and non-taxonomic readings lies with the determiner rather than with the common noun itself. Krifka et al. (1995) point out that Chinese uses different classifiers to distinguish between object-related readings and taxonomic readings: (224) a. yi zhî xíong one CL bear ‘an individual bear’ b. yi zhōng xíong one CL bear ‘a kind of bear, a bear species’ Since classifiers function as determiners (Cheng and Sybesma 1999), this provides evidence for the position that determiners are responsible for taxonomic readings. Overt determiners, then, are ambiguous between object-related and taxonomic readings. But there is no reason to assume that operations introduced

96

chapter 4

by type-shifting are ambiguous in the same way. Indeed, the type-shifter ∃ does not appear to induce taxonomic readings. For example, bare singulars in Hebrew receive their existential reading via ∃: while the overt numeral in (225a) allows a taxonomic interpretation, the bare singular in (225b) does not. (225) a. livyatan exad (ha-livyatan ha-kaxol) nimtsa be-sakanat whale one the-whale the-blue is in-danger-of hakxada. extinction ‘One whale (the blue whale) is in danger of extinction.’ b. *livyatan nimtsa be-sakanat hakxada. whale is in-danger-of extinction ‘*Whale is in danger of extinction.’ Therefore, it is reasonable to propose that the type-shifter ι is also unable to generate the taxonomic interpretation. Consequently, bare singulars cannot refer to kinds via ι. 7.4 Characterizing Readings In Brazilian Portuguese, as we have seen above, bare singulars refer to kinds. We therefore expect characterizing generic readings of bare singulars to be freely available. This is, indeed, the case, as demonstrated by the following examples, from Schmitt and Munn (1999): (226) a. Criança lê revistinha. Child read-3sg comic book ‘Children read comic books.’ b. Coelho vê o mundo de outra maneira por que é muito Rabbit see-3sg the world of other manner because is very pequeno. small ‘Rabbits see the world in a different way because they are very small.’ In contrast, in languages where bare singulars cannot denote kinds, we expect that they cannot get characterizing generic readings, except for normative statements. However, we are in for a surprise: characterizing generics do appear to be freely available for bare singulars in languages that permit them to occur in argument positions.

beyond english bare plurals

97

Take Hindi, for example: the following is perfectly acceptable (Ashwini Deo, pc). (227) Kamraa caukor hotaa hai. room square pronominal copula is ‘Room is square.’ Note that (227) cannot be explained away as a normative statement, since, as we have seen, the corresponding English IS sentence (122b), repeated below, is clearly unacceptable (in the null context): (228) *A room is square. Bare singulars can get characterizing readings in Hebrew too, as the following example, from Greenberg (2002), demonstrates: (229) ‘orev hu Saxor Raven pronominal copula black ‘Raven is black’ How can we account for these facts? In order to answer this question, let us look more closely at Chierchia’s explanation for the fact that bare singulars cannot denote kinds: ∩ dog denotes a function that is undefined for worlds (or situations) that have more than one dog. But it is not necessary for ∩ dog to be defined for all worlds (or situations): as is usual in modal logic, the domain of the function defined by the ∩ operator is not all worlds (or situations), but only all accessible worlds. But what, in this case, counts as an accessible world? I follow Stalnaker’s (1974; 1998; 2002) suggestion that presupposition is a propositional attitude, and should be represented as an accessibility relation. That is to say, in all accessible worlds (or situations), all the presuppositions of the interlocutors are true. Recall that in the previous section we have reached the conclusion that bare singulars presuppose uniqueness; this means that in all accessible worlds (or situations), there exists exactly one dog. But this would mean that ∩ dog is not undefined after all: it denotes a function that would pick out, in every situation (or world) that has exactly one dog, that dog.17

17

Krifka (2004) reaches the same conclusion, though for different reasons.

98

chapter 4

Note that the fact that ∩ dog is defined does not mean that it denotes a kind. The domain of this function consists of situations with exactly one dog, i.e. minimal situations. But kinds ought to be defined for all situations, not merely minimal ones, otherwise almost all kinds would be undefined in the actual world. Thus, Chierchia’s explanation for the inability of bare singulars to denote kinds remains intact. However, minimal situations are sufficient to express the characterizing generic reading. As Heim (1990), von Fintel (1994), and others have demonstrated, quantificational adverbs can be represented as quantifiers over minimal situations; the same holds for the generic quantifier. Yet one puzzle remains: if ∩ dog does not denote a kind, or, in fact, any entity, it cannot serve as a topic. But recall the discussion from chapter 3, section 4.1 above: the initial logical form of characterizing generics is predicative, at which point the sentence must have a topic. But if the bare singular cannot serve as the topic, is the topic constraint satisfied? To answer this question, look again at our examples (227) and (229), and note that they contain a pronominal copula, or Pron (Doron 1983). In fact, its presence in such cases is obligatory (Greenberg 1998; Doron 2003). I suggest that it is the pronominal copula that may serve as the topic (Junghans 1997; Błaszczak and Geist 2000; Geist 2007; Cover 2006; Markman 2008). Therefore, the sentence does contain a topic, which is not the bare singular, the topic constraint is satisfied, and the sentence is consequently acceptable.

8

Conclusion

In this and the previous chapter I have proposed that, in both English and Italian, BPs denote properties. There are two ways in which these BPs can be used. They can either be type-shifted, to receive narrow scope existential readings; or they can form part of a rule or a regulation that is predicated to be in effect. In English, there is a third option: BPs can be type-shifted to kinds, to receive generic readings. In Italian, however, this is prohibited by the blocking principle—reference to kinds can be accomplished by an overt determiner. In both languages, then, BPs are properties subject to type-shifting, and constrained by the blocking principle. This proposal uniformly accounts for their rather different behaviors in two languages, and perhaps universally: we have seen that the same principles can also apply to Hungarian, and to various languages that allow bare singulars, such as Hindi, Russian, Brazilian Portuguese, and Hebrew.

beyond english bare plurals

99

One piece of the puzzle is still missing, though: how do we get a characterizing generic from the kind-denoting reading of BPs? We will explore this question in the next chapter.

chapter 5

Generics and Habituals We have argued above that generics express quantification, and the same is usually said about habituals. This quantifier is, as all agree, phonologically null. This means that it does not occur in the sentence; it needs to be derived somehow. What is the process by which this quantifier is produced? We have in our arsenal two processes by which quantifiers are introduced: type-shifting and predicate transfer. These processes have different properties, as we have seen in chapter 2 above. In particular, they behave differently with respect to scope: quantifiers introduced by type-shifting receive narrow scope only, whereas those introduced by predicate transfer engage in free scope ambiguities in transparent contexts. The first step in an account of the origins of the generic and habitual quantifiers will, therefore, be to look at their scopal properties. We will consider generics first, and then move on to habituals.

1

Generics and Scope

What is the scopal behavior of generics? Do they display normal scope ambiguities, no scope ambiguities, or are they somewhere in between? This question is surprisingly hard to answer. It is not immediately clear whether generics can interact scopally even with a simple operator, such as negation. 1.1 Negation Carlson (1977) argues that generics do not exhibit any scope ambiguities. As an example, he considers the effect of negation. Carlson notes that (230a) is ambiguous between readings that can be paraphrased as (230b) and (230c). (230) a. Bill doesn’t like all wombats b. For all x, if x is a wombat then it is false that Bill likes x. c. It is false that, for all x, if x is a wombat then Bill likes x. In contrast, (231) is not ambiguous; it can only get a reading that can be approximated by (230b), not the reading approximated by (230c).1 1 I say “approximated”, since, of course, the quantifier in (231) is the generic quantifier, not the universal one, and it allows exceptions. © Ariel Cohen, 2020 | doi:10.1163/9789004432598_006

generics and habituals

101

(231) Bill doesn’t like wombats. Hence, Carlson concludes that generics are not ambiguous with respect to negation. This is not the end of the story, however. Krifka et al. (1995) reach the opposite conclusion, and argue that generics are ambiguous with respect to negation. They discuss (232), which, they claim “can mean either that cows do not have the habit of eating nettles or that they have the habit of not eating nettles” (p. 123). (232) Cows do not eat nettles. Krifka et al. do not explain exactly what the truth-conditional difference between these two readings is (we will return to this issue in a moment), but it is fairly clear that there is, indeed, a real ambiguity here. Hence, there is evidence that generics are not ambiguous with respect to negation, and there is evidence that they are. How can we resolve this conflict? One possible solution might be to deny that (231) has any quantificational reading. We could say that (231) simply expresses predication of the kind wombat, expressing the claim that it does not have the property of being liked by Bill. According to this interpretation, (231) is not about individual wombats at all. Bill does not even have to know any wombats for the sentence to be true; he just has to have a general dislike toward the species. Since there is no quantifier involved, there is no puzzle why there is no scope ambiguity. In contrast, when a quantifier does show up, as in (232), the ambiguity is there. It cannot be denied that (231) does, indeed, have such a reading. Moreover, this is probably the preferred interpretation. So, perhaps it really is the only reading, and (231) does not involve quantification. But this proposed solution would only push the problem one step further: why does (231) have only a kind reading? We know that like can, in fact, take individual objects: ⎧ Mary ⎫ { } (233) Bill likes ⎨ every politician ⎬. { most colleagues } ⎩ ⎭

There is no way to interpret (233) as kind predication, but only as a statement about Bill’s feelings toward individuals. Moreover, a quantificational reading of (231) is possible, if somewhat unnatural, if we add an overt adverb of quantification, as in:

102

chapter 5

(234) John usually likes wombats (but he hates my pet wombat). Many more examples are provided by Diesing (1992, 113–114):2 (235) a. b. c. d. e. f. g. h. i.

I usually like pictures of manatees. I usually love sonatas by Dittersdorf. I usually appreciate good jokes about violists. I generally hate articles about carpenter ants. I generally detest operas by Wagner. I usually dislike movies about vampires. I generally abhor books about Brussels sprouts. I usually despise paintings of Chester A. Arthur. I generally loathe stories about stockbrokers.

Incidentally, there are two options to negate (234) (and the sentences in (235)), depending on whether negation is placed before or after the quantificational adverb (in surface sturcture), resulting in two different readings: (236) a. John doesn’t usually like wombats. b. John usually doesn’t like wombats. Sentence (236a), but not (236b), is compatible with a situation in which John is indifferent towards wombats. Even if it turns out that (231) does not, in fact, have a quantificational reading, there are a great many other cases with a generically interpreted object, where it is clear that we do have a characterizing reading: (237) a. b. c. d. e.

Neil doesn’t fly Apollo missions. This security guard doesn’t identify employees. This program doesn’t support women. This court doesn’t hear appeals. The big gun doesn’t mark Royal events.

Note that these cases allow exceptions: Neil may fly an Apollo mission on a rare occasion, the security guard may identify employees from time to time, etc. The question, then, is this: do we have a scope ambiguity here?

2 Diesing’s examples actually involve indefinite singular objects, rather than BPs, but the judgments are the same.

103

generics and habituals

This question is not so easy to answer as it stands. But note what happens if we stress the negation: (238) a. Neil {

DOESn’t } fly Apollo missions. does NOT

b. This security guard { c. This program { d. This court {

DOESn’t } identify employees. does NOT

DOESn’t } support women. does NOT

DOESn’t } hear appeals. does NOT

e. The big gun {

DOESn’t } mark Royal events. does NOT

Now we get a much stronger reading, which allows no exceptions: Neil doesn’t fly any Apollo missions, the security guard doesn’t recognize any employees, etc. I think the same effect is also present in (231), but it is hard to detect, because of the dominance of the kind reading of this sentence. It is, however, clearly present in (232): (239) Cows do NOT eat nettles. While (232) tolerates exceptions, the truth of (239) permits no cow to eat nettles. What is the nature of this effect of focus on the interpretation of negated generics? Specifically, and most relevantly for our purposes, is it a scope ambiguity? Following Cohen (2001a), I suggest an answer to the first question that will entail an affirmative answer to the second one. Consider (237b), for example. Let us first assume that negation takes narrow scope. In this case, we get the reading where, in general, if x is an employee, the security guard doesn’t identify x. Of course, the domain of quantification is restricted only to relevant employees, say employees that tried to pass through the door guarded by the security guard. It is well established that focus can further restrict the domain of quantification. For example: (240) a. This security guard doesn’t identify the [signatures]F of employees. b. This security guard doesn’t [identify]F the signatures of employees.

104

chapter 5

Here, the role of focus is to introduce alternatives; the disjunction of the alternatives restricts the domain of quantification (Rooth 1985). Thus, we interpret the sentence as follows: in general, if x is a security guard, and x satisfies the disjunction of the alternatives, then the security guard doesn’t identify x. In (240a) we consider alternative means of identification; and the sentence says that employees who are identified by the security guard are not identified on the basis of their signatures. In contrast, (240b) is about employees who present their signatures for inspection, and says that, in general, they are not identified. Let us now consider what happens when we focus the negation itself: (241) The security guard [doesn’t]F identify employees. In this case, I suggest that negation takes wide scope. Note that there is focus on the auxiliary, and therefore no focus inside the scope of the generic. Hence, no alternatives are introduced, and the set of alternatives is simply the set consisting of a single property: the property of not being identified by the security guard.3 What happens when the set of alternatives is a singleton? Sentence (241) is then interpreted as follows (with wide scope negation): it is not the case that, in general, if x is an employee and the security guard identifies x, then the security guard identifies x. This will of course be trivially false, so long as the security guard identifies at least one employee. Hence, we get the interpretation that the truth of (241) requires the security guard not to identify any employee. For another example, consider the following exchange: (242) A: I just saw a documentary about the platypus. What a funny mammal—it lays eggs! B: Don’t be silly. Mammals [don’t]F lay eggs. B’s response means that it is not the case that there exists x s.t. x is a mammal and x lays eggs. Is it therefore, of course, false. Crucially, B cannot be understood as making the (true) statement that it is not the case that, in general, if x is a mammal then x lays eggs. This, then, is the nature of the ambiguity of (232) above. Under one reading, negation takes narrow scope, and the sentence says that, in general, cows do

3 Compare Krifka (1999, p. 269): “the alternatives that are introduced by focus are used by … focus-sensitive operators … No alternatives are projected beyond that”.

generics and habituals

105

not eat nettles. But when negation is focused and takes wide scope, the sentence expresses the negation of the existential statement that some cows eat nettles, i.e. the claim that no cows eat nettles. Hence, contrary to Carlson’s claim, we can conclude that generics do interact scopally with negation. 1.2 Opaque Contexts Carlson (1977) notes that (243), which employs an overt quantifier, is ambiguous. (243) John believes that all professors are insane. Under one reading, the quantifier has narrow scope with respect to the belief operator, and the sentence means that John has the following belief: for all x, if x is a professor then x is insane. Under the other reading, the quantifier has wide scope, and the sentence means that for all x, if x is a professor, then John has the following belief: x is insane. In contrast, the generic (244) only has the first type of reading. (244) John believes that professors are insane. Sentence (244) can only mean that John believes the following: in general, if x is a professor then x is insane. It cannot mean that, in general, if x is a professor, then John believes that x is insane. Judgments are rather hard, as Carlson himself admits. Let us try to find an example with clearer judgments. (245) The King believes that every enemy spy is loyal to him. Under one reading, the King has a rather irrational belief: “for all x, if x is an enemy spy, then x is loyal to me.” Under the other, more plausible reading, the King is rational, but has simply failed to expose the spies in his camp. Thus, for every x, if x is an enemy spy, the King has the following belief: “x is loyal to me.” If we change the quantifier from the determiner every to the adverb always, we can see the two readings in surface structure, depending on whether the adverbial is inside or outside the intensional context: (246) a. The King believes that an enemy spy is always loyal to him. b. The King always believes that an enemy spy is loyal to him.

106

chapter 5

Sentence (246a) describes an irrational king, whereas (246b) is perfectly compatible with a rational king (who, presumably, has an incompetent counterintelligence service). What about the generic? (247) The King believes that enemy spies are loyal to him. Indeed, this only has the irrational reading. It can only mean that the King does not understand what it means to be a spy, not that he simply failed to expose the spies. Hence, Carlson is right: generics do not scope out of opaque contexts. 1.3 Transparent Contexts In contrast, in transparent contexts, generics exhibit the usual scope ambiguities (Schubert and Pelletier 1987): (248) a. b. c. d. e.

Canadian academics are supported by a single granting agency. Storks have a favorite nesting area. Dogs have a tail. Sheep are black or white. Whales are mammals or fish.

Of course, some readings are implausible (e.g., the reading of (248c) where the indefinite takes wide scope) but they are all, in principle, available. The conclusion of this section is that generics do exhibit scope ambiguities, but they are more restricted than these of overt quantifiers; specifically, generics cannot scope out of opaque contexts.

2

Habituals and Scope

It is commonplace to treat generics and habituals on a par, and many authors do not bother to distinguish between them. Intuitively, the two phenomena do have a great deal in common: both involve no overt quantifier, and both seem to imply some sort of lawlike, permanent regularities, yet both tolerate exceptions. There are, however, important differences between the two constructions. Of particular relevance to our discussion are the different extents to which generics and habituals exhibit scope ambiguities. Krifka et al. (1995) note that the habitual quantifier takes narrow scope with respect to an indefinite. Thus,

generics and habituals

107

(249) is odd, because its only interpretation is that there is a cigarette that John habitually smokes. (249) #John smokes a cigarette. Since normally cigarettes are smoked only once, the sentence is pragmatically odd. Crucially, (249) cannot get the pragmatically plausible reading, where John habitually smokes a cigarette (different cigarettes on different occasions). It should be pointed out that some informants do manage to get this reading, but I believe that they do this by accommodating an appropriate context (see below), in order to avoid the oddity. Indeed, (250) is not odd, hence does not get accommodated. And all informants agree the only reading it can get is that there is one brand of French cigarette that John habitually smokes, i.e. the habitual takes narrow scope. (250) John smokes a brand of French cigarette. Krifka et al. make a similar point. They note that, in contrast with cigarettes, pipes are smoked multiple times, and consequently (251) is fine. (251) John smokes a pipe. Again, the habitual takes narrow scope: (251) means that there is a pipe that John habitually smokes,4 but this is no longer a pragmatic anomaly. In this, the habitual differs from an overt adverbial quantifier. For example, the sentences in (252) are ambiguous. (252) a. John never smokes a cigarette. b. John smokes a cigarette every day. Sentence (252a) can have either the implausible reading where there is one cigarette that John never smokes, or the much more reasonable reading where there are no occasions of John smoking. Similarly, (252b) can mean either that there is (implausibly) one cigarette that is smoked every day, or (much more plausibly) that on every day there is a cigarette that John smokes.

4 Of course, this does not exclude the possibility that there are additional pipes that John habitually smokes.

108

chapter 5

A similar phenomenon is observed when the habitual is restricted overtly, or even by the context. In such cases, habituals are ambiguous between wide and narrow scope readings. Consider: (253) a. John smokes a cigarette when he is nervous. b. Q: What does John do when he is nervous? A: He smokes a cigarette. c. John bakes a cake for dinner (Hans Kamp, pc). In (253a), a when-clause introduces the restrictor of the quantifier. This, apparently, is sufficient to allow the habitual to receive wide as well as narrow scope, and the sentence is ambiguous. Wide scope may even be enabled by contextual restriction. In (253b), the question sets up a restriction of the habitual quantifier, so that it ranges over situations in which John is nervous. Consequently, the answer is ambiguous, and can easily receive the plausible wide scope reading. Similarly, (253c) provides a restriction of the quantifier over dinner situations, and, consequently, a wide scope interpretation is easily available. Things are even more complicated, in fact. As we have seen in chapter 3, section 2.2 above, if the indefinite is a bare plural, the habitual can have only wide scope, and it is the narrow scope reading that is unavailable. Thus, (254) means that John is habitually in a situation where there are some cigarettes that he smokes. It does not, indeed cannot get the interpretation corresponding to (249), namely that there are some cigarettes that John habitually smokes. (254) John smokes cigarettes.

3

Types of Explanation

3.1 The Facts so far Let us recap the scope facts we have observed so far. Generics exhibit scope ambiguities, with one significant limitation: they cannot scope out of opaque contexts. In contrast, habituals receive narrow scope only, except when they are overtly or contextually restricted, in which case they may take wide scope, and with bare plurals, where they have to take wide scope. Surprisingly few accounts of these puzzling differences in scopal behavior between generics and habituals have been offered in the literature. Let us take a look at two of them.

generics and habituals

109

3.2 Number Krifka et al. (1995) assume that generics have normal scope ambiguities, and focus on the difference between sentences such as (249) and (254). They propose to account for the difference on the basis of the number of the indefinite: singulars allow only narrow scope to the habitual, whereas plurals allow both readings. Krifka et al. assume that unrestricted habituals express quantification over a situation variable; this variable may be a simple situation, but it may also be a sum situation, i.e. a situation composed of a plurality of simple situations. There are two ways in which quantification over sum situations can be acceptable. One way is to associate the sum situation with a sum individual, as in (254). In this case, every simple part of the sum situation is associated with a simple part of the sum individual. This is how we get the interpretation where different cigarettes are smoked on different occasions. An alternative is to associate one simple individual with the sum situation as a whole, as in (251). This is how we get the narrow scope interpretation of the habitual, namely that a single pipe is associated with multiple smoking events. However, quantification over sum individuals is not compatible with associating a different simple individual with each one of the simple situations in the sum. This is the interpretation that would make sense of (249), and the fact that it is unavailable explains the unacceptability of the sentence. Krifka et al. assume that explicitly restricted habituals, as in (253a), can only quantify over simple situations. Presumably, the same would hold for overt adverbial quantifiers, as in (252), and cases where the restriction is provided by the context, as in (253b) and (253c). There are, however, problems with Krifka et al.’s account. First, the difference between unrestricted habituals on the one hand, and restricted habituals (and overt Q-adverbs and certain contexts) on the other hand, is unmotivated. Why is is that the former, but not the latter, may quantify over sum situations? Second, while Krifka et al. provide an explanation for why the habitual in (254) may have wide scope, they fail to explain why it must. Since habituals, by this account, may quantify over simple situations as well as sum situations, why can’t the bare plural scope over the habitual, associating each of the simple individuals in its sum with a simple situation? Third, consider the following sentence: (255) #John smokes some cigarettes. The direct object is plural, just like the direct object of (254). If, as Krifka et al. 1995 claim, all that matters were the number of the TNP, it should behave sco-

110

chapter 5

pally just like it does in (254), namely take narrow scope with respect to the habitual. But the facts are exactly the reverse: (255) can only have the implausible reading that there are some cigarettes that John habitually smokes,5 and does not have the reading where John is habitually in a situation where there are some cigarettes that he smokes. 3.3 A Verb-Level Null Operator A different account has been proposed by Van Geenhoven (2004). She does not deal with generics, but only habituals, and assumes a phonologically null operator, whose role is to express habituality.6 This operator applies directly at the level of the verb—hence, any operator that scope over the verb, also scopes over this operator. This results in habituals having narrow scope only. This account is, however, problematic, on both theoretical and empirical grounds. From a theoretical point of view, this proposal is not well motivated. Why would the habituality operator apply at the verb level, and not scope freely like most other operators? Moreover, the explanation has empirical difficulties: Van Geenhoven’s proposal fails to account for the examples in (253), where a habitual does receive wide scope.7 3.4 A Syntactic Account? Perhaps the explanation of these facts is syntactic, rather than semantic, in nature. No such idea has been explored, as far as I am aware, but suppose that implicit quantification (generic or habitual) applies to a phrase structure tree at the lowest level possible. Habituals quantify over the event argument, which is inside the VP. Hence, these operators apply at the level of the VP, so that everything that has scope over the VP also scopes over them—this is why habituals receive narrow scope only. In contrast, generics quantify over the subject. Hence, they apply at the IP level, and are therefore scopally ambiguous. The relevant fact about opaque contexts is not their opacity, but the fact that they are embedded; and the generic cannot scope over the embedding verb. Would such an account be syntactically well motivated? I am not sure: many questions would remain to be worked out. For example, why do implicit oper-

5 Under the taxonomic interpretation, where some cigarettes is interpreted as brands of cigarette, the sentence actually makes perfect sense. 6 In fact for Van Geenhoven habituality is reduced to iterativity; we will return to iterativity and to Van Geenhoven’s theory in chapter 6. 7 Nickel (2016) also proposes that the generic operator applies at the level of the predicate, but not out of considerations of scope: he argues that gen is a distributivity operator. I will address his account in chapter 7, section 1.2.

generics and habituals

111

ators have to operate at the lowest level possible, while explicit operators are not so restricted? What, precisely, is the syntactic location of these operators, and why can’t they QR?8 However, we needn’t worry about the theoretical justification for such a theory, because it is empirically incorrect. The crucial test cases are sentences where the object is interpreted generically, and takes wide scope. This should not be possible according to the syntactic account, since the object is inside the VP, hence the generic operator should also apply at this level, and receive narrow scope only. However, it is easy to construct this type of sentence: (256) a. {

A well trained team flew } Apollo missions. Three astronauts flew b. A personal code identifies admissible users. c. Two bodyguards protect Ministers, but a single bodyguard protects Deputy Ministers.

Under their most plausible readings, sentence (256a) means that, in general, for each Apollo mission there was a team that flew it; (256b) refers to a different code for different users; and in (256c), different bodyguards are assigned to different politicians. Hence, in all these sentences the object is interpreted generically, and receives wide scope. The sentences above are manufactured examples. But one can also find naturally occurring examples on the Web: (257) a. Family planning helps women or girls to avoid pregnancies and continue with their education.9 b. Three judges sit together to hear appeals from decisions of the Trial Division.10 c. At sexual maturity a modified anal fin distinguishes the male Guppy.11 d. Twenty one gun salutes were fired from the large guns to mark Royal events.12 8

9 10 11 12

Rimell (2004) assumes that the habituality quantifier applies at the level of AspP; she further assumes that no element can scope below AspP. She does not, however, provide an argument for either of these claims. https://repository.unam.edu.na/bitstream/handle/11070/421/nandjila2008.pdf?sequence =1&isAllowed=y http://www.isn.net/cliapei/pub/78b.htm https://www.practicalfishkeeping.co.uk/features/the‑guppy‑first‑loves‑never‑die/ http://www.caithness.org/caithnessfieldclub/bulletins/1996/artillery_batteries_castletow n_and_mey.htm

112

chapter 5

The dominant reading of (257a) is that, if x is a woman or a girl, family planning helps x, not that family planning helps women, or family planning helps girls; (257b) allows different judges for different appeals; (257c) is about different fins for different guppies; and (257d) is about different gun salutes for different occasions of royal events. We can conclude, then, that all three suggestions made above do not seem to explain the scopal facts about generics and habituals successfully—can we do better?

4

Reinterpretation Mechanisms Revisited

4.1 The Introduction of Implicit Quantifiers Explicit quantifiers, as in (252), are produced by the speaker, and appear in the input to the hearer. And, as we have seen, they exhibit the normal scope ambiguities. In contrast, generics and habituals involve implicit quantification: the quantifier is phonologically null. But this means that, in a very real sense, the quantifier is not “there”; it is not produced by the speaker, nor does it appear in the input to the hearer. How, then does it appear in the interpretation? Where does it come from? Sometimes, although the quantifier itself is implicit, its presence is indicated by explicit means, such as the occurrence of an overt restrictor, as in (253a). Even context, as in (253b) and (253c), may indicate the presence of the quantifier. In such cases, the quantifier is, in a sense, really “there”, as if it were pronounced overtly. Hence, again, scope ambiguities are normal. However, when there are no such overt means, the presence of the quantifier involves a change of meaning from the original input, where there is no quantifier, to the semantic representation, where a quantifier does occur. However, in many accounts of generics and habituals this simple fact is ignored. The implicit assumption appears to be that, in some sense, the quantifier is already “there”, and no special mechanisms are needed to introduce it. In other words, the only difference between overt quantifiers and the generic/ habitual quantifier is phonological: the former have phonological content, while the latter are “null”—we can’t hear them. But what, exactly, does it mean for a quantifier to be phonologically null?13 13

As mentioned above, some researchers (Leslie 2007b; Sterken 2015a) attempt to derive the meaning of the generic quantifier from the fact that it is phonologically null; however, they still assume that the quantifier is essentially “there”, and do not discus the mechanisms by which it is derived. A notable exception is Nickel (2016); I will discuss his theory in chapter 7, section 1.2 below.

generics and habituals

113

4.2 Null Categories in Syntax The idea seems to come from syntax, where there is a long history of theories assuming empty categories of various kinds: traces (or unpronounced copies), PRO, pro, etc. Syntacticians propose that semantic interpretation makes use of these traces, so they must be available to semantics. How is this possible? The common answer is that such empty categories are simply “there”: they have no phonological content, but are syntactically and semantically significant. Consider, for example, the remarkable (if controversial) phenomenon of wanna contraction. In colloquial English, (258a) can quite acceptably be paraphrased as (258b). (258) a. I want to come to the party. b. I wanna come to the party. Strikingly, however, (259b) is not an acceptable paraphrase of (259a), and is, in fact, ungrammatical. (259) a. Who do you want to come to the party? b. *Who do you wanna come to the party? The common argument is that want to cannot be turned into wanna if there is an empty element between want and to; and in (259a) there is such an unpronounced element, a trace (or an unpronounced copy), so its surface structure is actually (260). (260) Who do you want t to come to the party? Such examples have been taken to indicate that traces are very real: you can’t hear them, but they are there all the same.14 One almost feels that, like the Emperor’s new clothes, if you can’t see them, you must be a fool. Yet we mustn’t lose sight of the fact that the Emperor is, in fact, naked: when we hear a sentence, what we are presented with is an expression without empty categories. In fact, Chomsky (1995) argues that, at PF, empty categories are deleted, in accordance with his proposed Full Interpretation principle: “if a symbol in a representation has no sensorimotor interpretations, the representation does not qualify as a PF representation” (p. 27). Since what we hear is fed by PF,

14

This account of wanna contraction is, in fact, quite controversial, but I will not join the debate here.

114

chapter 5

and empty categories are deleted at PF, it follows that empty categories are not present in the input we hear, not even in some obscure “abstract” sense. What is often not appreciated is the fact that if hearers do not hear these elements in the input, they need to reconstruct them somehow.15 One might think that it does not really matter how we decide to treat empty categories. We may decide to treat empty categories as really “there”, though inaudible; alternatively, we may choose to accept that they are not “there”, and are reconstructed by the hearer. But is there an empirical fact that can distinguish between the two views? Such an empirical fact would be provided by syntactic phenomena that are only sensitive to material that is really “there”. If there are syntactic, nonphonological rules that ignore unpronounced material and only apply to material that can actually be heard, such rules would provide evidence that the empty categories are not really “there”, not only phonologically but also syntactically. It appears that there are, in fact, such rules. Holmberg (1999) notes that in Scandinavian languages, object shift is impossible if there is material inside the VP preceding the object. For example, in Swedish, the object cannot shift across an indirect object, from its position in (261a) to its position in (261b). (261) a. Jag gav inte Elsa den. I gave not Elsa it b. *Jag gav deni. I gave it Similarly, while (262a) is fine, (262b), where the object has shifted across a verb particle, is bad. (262) a. Dom kastade inte ut mej. they threw not out me b. *Dom kastade meji inte ut ti. they threw me not out

15

Syntacticians who are interested in ellipsis often do, in fact, discuss the need to reconstruct the elided material (see, for example, Wasow 1972; Williams 1977; Haïk 1987; Kitagawa 1991; Fiengo and May 1994).

generics and habituals

115

However, if the intervening material is phonologically null, object shift is perfectly acceptable; thus, the following sentences are good: (263) a. Vemj gav du deni inte tj ti? who gave you it not ‘Who didn’t you give it to?’ b. UTj kastade dom meji inte tj ti (bara ned för trappan). out threw they me not only down the stairs ‘They didn’t throw me OUT (only down the stairs).’ Thus, Holmberg’s Generalization appears to provide evidence that empty categories are not really “there”; they need to be reconstructed by the hearer.16 Reconstructing empty categories is not a trivial task: indeed, according to the Trace Deletion Hypothesis (Grodzinsky 1986, 1995), agrammatic aphasics are unable to do it.17 So why do it at all? Why is it that, when we hear (264a), we reconstruct the phonologically null element PRO as the subject of leave, and derive (264b)? (264) a. John wants to leave. b. Johni wants PROi to leave. For another example, consider (265a): why do we reconstruct a phonologically null trace as the object of love, and generate (265b) as the result? (265) a. Who does Mary love? b. Whoi does Mary love ti ? The answer is that, otherwise, syntactic constraints would be violated, and the sentences would be ill-formed. Thus, (264a) is ill-formed, because the inner clause lacks a subject, hence PRO is introduced to “save” the situation. Similarly, (265a) is bad, because the transitive verb lacks an object; hence, the trace is introduced, and the resulting structure, (265b), is fine.

16 17

There are alternatives to Holmberg’s theory (e.g., Bobaljik 2002; Erteschik-Shir 2005; Fox and Pesetsky 2005). His empirical generalization, however, appears to be largely accepted. Ironically, the name of the theory implies that traces are “there”, and agrammatic aphasics somehow delete them. Of course, this is not the actual content of the theory: rather, agrammatics cannot recover traces that are not present in the sentences they hear.

116

chapter 5

It should be emphasized that, of course, introducing phonologically null elements is only one of several possible ways to save the situation. There are competing syntactic theories in which no such elements are posited. But the point of these examples is not to argue for a particular syntactic theory, but rather to show the rationale behind the postulation of empty categories: if they exist, the reason for their introduction is to save a sentence that, otherwise, would be bad. 4.3 Null Categories in Semantics There are a few cases where null categories have been proposed on semantic, rather than syntactic grounds. But the reason for their introduction is similar: to save a sentence that, otherwise, would be unacceptable. A classic case is that of quantificational domain restriction. It is widely agreed that when a quantifier is used in natural language, its domain is assumed to be restricted by the context. More specifically, the context provides a set, which is not included in the surface structure of the sentence, and the domain of the quantifier is understood to be intersected with this set. For example, Neale (1990) writes: Suppose I had a dinner party last night. In response to a question as to how it went, I say to you: (266) (Neale’s example #2) Everyone was sick. Clearly I do not mean to be asserting that everyone in existence was sick, just that everyone at the dinner party I had last night was. pp. 95–96

Sentence (266) does not provide any explicit restirction on the domain of the quantifier. Why, then, is the sentence not understood to express a statement about everyone in existence? Presumably, because such a statement would be blatantly false—it’s a statement that anyone would immediately recognize as being false, and nobody in their right mind would have any reason to utter. In Gricean terms, such a statement would not merely violate the maxim of Quality—it would flout it. Therefore, we look to the context for a domain, which provides us, in this case, with the set of people at the dinner party, and that serves as the domain of the quantifier. Similarly, Recanati (1996) notes that “when we say ‘The burglar took everything’, we have the feeling that ‘everything’ ranges over the domain of valuable objects in the house—not everything in the world” (p. 445). Again, saying that

generics and habituals

117

the burglar took everything in the world would be a nonsensical statement, hence the context provides us with an acceptable domain for the quantifier to range over. Contextual restriction operates even when the sentence itself does provide some restriction. For example, Lewis (1996) writes: “If I say that every glass is empty, so it’s time for another round, doubtless I and my audience are ignoring most of all the glasses there are in the whole wide world throughout all of time” (p. 553). Even though the sentence restricts the domain of quantification to glasses only, a statement about all the glasses in the world would be nonsensical; hence, context provides further restriction, and the quantifier is understood to range over the intersection of the set of glasses with this set, resulting, presumably, in the set of glasses held by the speaker and his friends on that occasion. Stanley and Szábo (2000) discuss this issue at length. They write: What is the problem of quantifier domain restriction? Consider the sentence: (267) (Stanley and Szábo’s example #1) Every bottle is empty. Suppose someone utters [(267)] in a conversation. It is unlikely that what she intends to convey is that every bottle in the universe is empty; she most likely intends to convey that every one of a restricted class of bottles (say, the bottles in the room where she is, the bottles purchased recently, etc.) is empty. p. 219

Once again, it is clear that a statement about every bottle in the universe would make little sense, and the quantifier needs to be further restricted by the context. Stanley and Szábo’s focus is on the question of where the restriction provided by the context is introduced: whether it is part of syntactic structure (which is elided), a variable at the level of logical form (their preferred solution), or introduced only after pragmatic processing. Regardless of the level at which this restriction is introduced, it is clear that the motivation for postulating it in the first place is that, without it, the sentence would be unacceptable, as patently false.

118 5

chapter 5

The Generic Quantifier

Seen in this light, the generic quantifier joins syntactic traces and contextual restrictions as another empty category. Generics and habituals trigger the introduction of this empty category: gen. Chierchia (1995) identifies many similarities between the generic quantifier and (overt) adverbs of quantification. Following him, and many others, I assume that the generic quantifier is a kind of adverb of quantification. Then, we would expect that, just like other adverbs of quantification, it has temporal as well as atemporal interpretations. When gen is used atemporally, we get a generic; and when it is used temporally, the result is a habitual. Thus, I assume that generics and habituals are similar in that they share the same quantifier (Cohen 1996); the differences between them will be discussed anon.18 Thus, (268a) turns into (268b), and the representation of (269a) is (269b), where gen is the generic quantifier. (268) a. Birds fly. b. Birds gen fly. (269) a. John smokes. b. John gen smokes. Of course, there is a clear difference between gen and overt quantificational adverbs: it is phonologically null. But is this difference significant? It is often assumed in the literature that the answer is no: that gen is just like an overt quantificational adverb, and that the fact that it is phonologically null is of no significance.19 However, I argue that this view is deeply problematic: the generic quantifier is not the same as overt quantifiers, and the fact that it is phonologically null has quite significant implications. One such implication of this view is its cross-linguistic predictions. If gen is just another quantifier that happens to be phonologically null, we would expect to find languages where it is not null, but overt. However, there is apparently no language that has an overt element that expresses genericity (Dahl 1995). A related problem is that if gen is just a phonologically null version of an overt quantifier, we would expect languages without quantifiers to lack gener18 19

But see Boneh and Doron (2008), who claim that there are, in fact, two habitual quantifiers, only one of which appears to correspond to gen. Though we have seen some exceptions to this view (Leslie 2007b; Sterken 2015a; Nickel 2016).

generics and habituals

119

ics as well. But Everett (2005) argues that Pirahã is a language without phonologically overt quantifiers, which, nonetheless, contains generics: (270) kaoáíbogi hi sabí ’ágahá. evil spirit he mean is (permanent) ‘Evil spirits are mean’ If Everett’s claim is granted, gen cannot be a mere phonologically null version of an overt quantifier. Even if we only look at English, it turns out that there are empirical differences between overt adverbial quantifiers and generics. One difference we have already seen in chapter 3, section 4.3 above—overt quantifiers are always good with indefinite singulars, but generics are often bad (Cohen 2001b): (271) a. *A madrigal is popular. b. A madrigal is usually popular. We have also seen in section 1.1 that generics can give rise to existential readings: (272) A: I just saw a documentary about the platypus. What a funny mammal—it lays eggs! B: Don’t be silly. Mammals [don’t]F lay eggs. However, such a reading is not possible with adverbs of quantification: (273) Mammals [don’t]F usually lay eggs. While (272) denies that any mammal lays eggs, and is therefore false, (273) allows that there are some mammals that lay eggs, hence it’s true (Cohen 2001a).20 It turns out that it is actually beneficial to realize that gen is not really “there”, and to pay close attention to the mechanisms by which it is introduced. In fact, the differences in the scopal behavior between generics and habituals are due to the fact they they use different reinterpretation mechanisms. Specifically,

20

Incidentally, these differences may account for Sterken’s (2015b) observation that generics vary their interpretations more readily across different contexts, but I will not discuss this issue firther here.

120

chapter 5

habituals involve type-shifting, whereas generics involve predicate transfer. Let us consider each one of these constructions in turn.

6

Generics

6.1 Analysis Recall that some generics are fine without the generic quantifier: these are cases of direct kind predication. Thus, for example, it is generally assumed that (274a) expresses no quantification, but rather the predication of the property of being an endangered species of the kind ∩ pandas. Hence, its logical form ought to be something like (274b). (274) a. Pandas are endangered. b. endangered(∩ pandas) In contrast, other generics cannot receive this interpretation. For example, (275) does not express predication of the property of liking bamboo shoots of the kind ∩ pandas. (275) Pandas like bamboo shoots. At an intuitive level, I think this is quite clear: kinds are not the sort of things that can eat; only individual animals can eat. But there is a stronger argument than just intuition. Consider, for example, (276a), from Carlson (1977) (his example #35, p. 165). Carlson points out that if (276a) were interpreted as predicating the property like oneself directly of the kind ∩ cats, it would have (276b) as its logical form. (276) a. Cats like themselves. b. like(∩ cats, ∩ cats) But this logical form says that cats like cats, and we would lose the more plausible reading where, in general, a cat likes itself, but not necessarily other cats. Consequently, in cases where direct kind predication will not do, Carlson proposes that a phonologically null generic operator is necessary. I accept this view, though for me, unlike Carlson, and like most scholars, the generic operator is a quantifier (see Cohen 1996 and chapter 3, section 2.3 for the arguments for this move). Sterken (2016) makes this argument (which she calls the binder

generics and habituals

121

argument, after Stanley and Szábo’s 2000 binding argument) explicitly. She considers another one of Carlson’s examples (his #25, p. 163): (277) Goldfish like everyone who likes them. Sterken points out that the object of likes is a variable, which must be bound by a quantifier. This quantifier is not everyone, which binds the subject of the verb, hence it must be the generic operator, which consequently must be a quantifier. Thus, the logical form of (275) must include the generic quantifier; it cannot be something like (278). (278) like-bamboo(∩ pandas) Of course, these considerations do not preclude the proposal, made in this book, that (278) is the initial logical form of (275); and then the generic quantifier is introduced to produce its final logical form. The question, then, is this: why must the generic quantifier be introduced? What is wrong with (278)? One possible answer is that (278) contains a type mismatch: like-bamboo is a predicate that applies to objects, i.e. individual pandas, and not to the kind ∩ pandas. This is, essentially, Carlson’s (1977) view; he also draws a corresponding distinction between object-level and kind-level predicates.21 If, indeed, type mismatch is the cause, then the operation introducing the generic quantifier ought to be type-shifting. There is, however, little evidence for a type difference between objects and kinds. Granted, there are predicates that only apply to kinds, such as be extinct, be invented, etc. However, this shows nothing more than that these predicates have certain selectional restrictions. Almost every predicate carries selectional restrictions, some of which are extremely specific. McCawley (1976) points out that for example, the verb diagonalize requires as its object a noun phrase denoting a matrix (in the mathematical sense), the adjective benign in the sense ‘noncancerous’ requires a subject denoting a tumor, and the verb devein as used in cookery requires an object denoting a shrimp or a prawn. p. 67

21

More precisely, Carlson treats objects and kinds as two subtypes of individuals.

122

chapter 5

Of course, we wouldn’t want to say on the basis of these selectional restrictions that matrices, tumors, and shrimp are distinct logical types! There are, in fact, many predicates that apply to both kinds and objects: (279) a. {

Moby Dick } is a mammal. The whale

b. There is a a story in the newspaper about { c. John loves { d. { e. {

Mary }. tigers

Mary }. tigers

James Bond } will not be forgotten. The lion Mary } lives in Africa. The lion

If we maintained that kinds differ from objects in their type, we would be forced to conclude that all these predicates are ambiguous. Hence, and for reasons of parsimony, a more attractive alternative is that both kinds and objects are of the same logical type, namely individuals; correspondingly, there is no distinction between predicates that apply to objects and predicates that apply to kinds (Cohen 1996). In the grammar, any predicate may apply to any individual. However, some of these applications will be rejected22 due to violations of certain selectional restrictions. Of course, selectional restrictions (or some similar devices) are needed on independent grounds, e.g. to account for the unacceptability of sentences like the following: (280) #The rock is sad. Sentence (281), just like (280), will be generated by the grammar, but both will be rejected because they are semantically anomalous: the property of being sad requires its argument to be animate, which a rock is not; and the property of being an endangered species requires its argument to be a kind, which an individual panda is not: (281) #Panny, my panda, is an endangered species. 22

Or, perhaps, considered necessarily false.

generics and habituals

123

In other words, the fact that kinds induce special selectional restrictions does not entail that they belong to a special logical type; world knowledge, rather than the grammar, would determine which individuals are considered to be kinds and which are not. Let us consider (275) again. Because of the topic constraint, the BP pandas is first type-shifted into the kind ∩ pandas, resulting in the logical form (278). However, as it stands, this logical form is bad; not because of semantic mismatch, but because it violates selectional restrictions—it doesn’t make pragmatic sense. We need to “save” the sentence, and the way to do it is to derive from (278), which expresses predication of a kind, a logical form that expresses quantification over instances of the kind, using the ∪ operator. Thus, (282) is the representation of generic quantification as a standard tripartite structure.23 (282) genx [∪∩ pandas(x)][like-bamboo-shoots(x)]

In general, if predication of property P over kind k does not make pragmatic sense, we derive instead: (283) genx [∪ k(x)][P(x)].

What is the process by which gen is introduced? If it is derived because of type mismatch, it ought to be type-shifting; but if, as I argue here, there is no type mismatch, this process ought to be predicate transfer. Which is it? One piece of evidence that the process in question is indeed predicate transfer is the fact that introducing the generic quantifier is sometimes optional. For example, (284) is ambiguous. (284) Animals are listed in the encyclopedia. Under one reading, the kind ∩ animals is listed in the encyclopedia. But there is another reading, where, in general, if x is a subkind of animal, then x is listed in the encyclopedia. Note that both readings involve predicating a property— being listed in the encyclopedia—of kinds: either the kind ∩ animals, or its subkinds. However, only one of the readings involves the generic quantifier. Hence, the introduction of this quantifier is optional, rather than mandatory. This is a strong indication that the relevant phenomenon is not type-shifting, which is mandatory, since without it the sentence would be ill-formed. In contrast, as we have seen in chapter 2, section 3, predicate transfer is, in fact, optional.24 23 24

Alternatively, we could represent this logical form using minimal situations. Sterken (2016) acknowledges that if the derivation of the generic quantifier is optional,

124

chapter 5

Indeed, we can apply Nunberg’s (1995) test. Recall example (49b), repeated below: (285) I am parked out back and would like to make a phone call. Since the argument of both conjoined predicates is the same, and since it is clearly the speaker (not the speaker’s car) for the second conjunct, it follows that the argument of the predicate in the first conjunct is also the speaker. Therefore, Nunberg takes the acceptability of (285) as evidence that the predicate, but not the argument, changes meaning. We can apply the same test to generics: (286) Pandas like very restricted foods and are therefore an endangered species. The fact that (286) is fine indicates that only the predicate changes meaning, not the argument: the subject of the sentence still denotes a kind, and is acceptable as an argument of a kind predicate, even though it is used in a characterizing, quantificational generic.25 The picture that emerges is as follows. A generic sentence is initially interpreted as predication, without the generic quantifier (cf. Cohen 2001b). If this interpretation makes sense, as in (274), nothing else needs to be done. However, if it results in a pragmatic anomaly, the generic quantifier is introduced by way of predicate transfer. 6.2 Formalization In order to formalize this operation, let us generalize Dölling’s (1995) definition from chapter 2, section 3.1, repeated below: (287) 𝒯(P) =def λx∃y(R(x, y) ∧ P(y)).

25

this constitutes an argument in favor of the proposal that it is derived by predicate transfer. Curiously, however, Sterken uses this point as an argument against the derivation of gen by predicate transfer: the reason is that she has not been able to find cases that demonstrate the optionality of the generic quantifier. But surely (284) is precisely such an example. The proposal that chacracterizing generics are generated by predicate transfer may remind some readers of Liebesman’s (2011) claim that kinds can inherit properties from their instances. However, as Liebesman and Magidor (2017) point out, Liebesman’s process differs from predicate transfer. More importantly, unlike the proposal made here, Liebesman denies that the result of this process in an occurrence of a quantifier in the logical form of the generic.

generics and habituals

125

Recall that 𝒯 is a predicate transfer operation, which introduces an existential quantifier, and R is a free variable over relations, whose value is left for pragmatics. We will now define a similar operator, 𝒯𝒢 , which generates the generic quantifier, and, instead of Dölling’s pragmatically determined relation R(x, y), the specific relation ∪ x(y), meaning that y is an instance (or a plurality of instances) of the kind x: Definition 8 𝒯𝒢 (P) =def λx.geny [∪ x(y)][P(y)].

For example, the logical form of (288a) is (288b), to which 𝒯𝒢 applies, as in (288c), resulting in (288d). (288) a. b. c. d.

Pandas like bamboo shoots. like-bamboo(∩ pandas) (𝒯𝒢 (like-bamboo))(∩ pandas) genx [∪∩ pandas(x)][like-bamboo-shoots(x)]

The generation of the generic quantifier via predicate transfer can explain the scopal behavior of generics. Consider again one of Schubert and Pelletier’s (1987) examples: (289) Storks have a favorite nesting area. Initially, this sentence is interpreted as predication, so its logical form is: (290) ∃x(favorite-nesting-area(x) ∧ have(∩ storks, x)).

However, this sentences doesn’t make sense, since only individual storks, and not the kind, can have a nesting area. Hence, predicate transfer applies. Recall that, in chapter 2, section 3 above, we have established that predicate transfer can apply at any stage of the derivation, generating scope ambiguities. One possibility is to apply it locally, as in (291a). This will result in (291b), where the generic taks narrow scope. (291) a. ∃x(favorite-nesting-area(x) ∧ (𝒯𝒢 (λy.have(y, x)))(∩ storks)) b. ∃x(favorite-nesting-area(x) ∧ geny [∪∩ storks(y)] [have(y, x)])

126

chapter 5

Alternatively, we could apply predicate transfer globally, i.e. to the property of having a favorite nesting area, as in (292a). This results in (292b), where the generic takes wide scope. (292) a. (𝒯𝒢 (λy.∃x(favorite-nesting-area(x) ∧ have(y, x))))(∩ storks) b. geny [∪∩ storks(y)] [∃x(favorite-nesting-area(x) ∧ have(y, x))]

Hence, the scope ambiguities of generics in transparent contexts are accounted for by the assumption that the generic quantifier is introduced by predicate transfer. Note that if the responsible process were type-shifting rather than predicate transfer, generics would be predicted to have narrow scope only, which is not the case. Recall our conclusion from chapter 2, section 3.6 that predicate transfer cannot apply to intensional properties. This accounts for the fact that generics cannot scope out of opaque contexts. For example, the generic in (247), repeated below, can only get narrow scope with respect to the belief operator. (293) The King believes that enemy spies are loyal to him. The initial logical form of (293), where Ba ϕ means that a believes ϕ, is something like: (294) Bιking loyal(∩ enemy-spies). We can apply predicate transfer locally, as in (295a), which results in (295b), the interpretation where the generic has narrow scope. (295) a. Bιking (𝒯𝒢 (loyal))(∩ enemy-spies) b. Bιking genx [∪∩ enemy-spies(x)][loyal(x)]

If, however, we wanted to get the wide scope interpretation of the generic, we would need to apply predicate transfer globally: (296) (𝒯𝒢 (λx.Bιking loyal(x)))(∩ enemy-spies).

However, the property of being believed by the King to be loyal is an intensional property, hence cannot be subject to predicate transfer. This is why the wide scope reading of the generic is not available.

generics and habituals

7

127

Habituals

7.1 Analysis What is the origin of habituals? Consider a sentence in the present simple tense, such as: (297) John smokes. How is it interpreted? Following Taylor (1977) and Dowty (1979), I assume that eventive verbs are evaluated at time intervals that are longer than moments. As an illustration, Dowty asks us to consider a segment of a motion picture film showing a ball rolling down an inclined plane. A single frame of this film does not in itself offer us the evidence to say that the ball is really in motion … [This] would require access to information about the physical state of the world at at least two moments in time. p. 168

In contrast, sentences with stative verbs can be true at moments. The meaning of the present tense is a moment, the speech time (cf. von Stechow 2002). Hence, (297) involves a type mismatch: the eventive verb smoke requires an interval that is longer than a moment, but it is in the present tense, which is a moment.26 The way to resolve this mismatch is to turn the eventive predicate into a stative one. One way to turn an eventive verb into a stative one is to make it a habitual: applying standard tests, it is clear that habituals are stative (Leech 1971; Mufwene 1984; Partee 1984; Chung and Timberlake 1985; Rot 1987; Smith 1991; Carlson 1995). For example, stative, but not eventive verbs in the present perfect entail that the situation they refer to persists into the present: (298) a. She has lived here since early childhood. b. The child has coughed all night.

26

Perhaps the distinction between moments and intervals is of a sort rather than a type; this, however, should not affect the arguments made here—see the discussion in chapter 2, section 2.6 above.

128

chapter 5

While (298a), whose main verb is stative, entails that she still lives here today, (298b) does not entail that the child is still coughing now. Habituals pattern like statives in this respect: (299) He has played tennis for many years. Under its habitual interpretation, (299) entails that he still plays tennis.27 Vlach (1981) points out that, when modified by a when-clause, statives, but not eventives, imply that the state has already obtained before the event described by the when-clause. Thus, (300a) entails that Max was here before I arrived, but (300b) is quite compatible with Max starting to run the moment I arrived. (300) a. Max was here when I arrived. b. Max ran when I arrived. Again, habituals pattern with statives: (301) entails that Max’s habit of running in marathons began before the invitation arrived. (301) Max ran in marathons when the invitation to the Olympic team arrived.28 I suggest that turning the verb into a habitual involves type-shifting, which introduces the generic quantifier. Note that this proposal captures both similarities and differences between generics and habituals. They are similar in that they both involve the same quantifier—gen. However, they differ with respect to the mechanism that generates this quantifier. With generics, gen is generated by pragmatic predicate transfer; whereas in the case of habituals, semantic type-shifting is responsible.29

27 28 29

These examples are taken from Brinton (1987), though, in fact, she objects to these conclusions. Vlach actually makes this point about the progressive, but it can just as easily apply to habituals. An interesting question is why two very different mechanisms result in the generation of the same quantifier; I will return to this issue in chapter 7.

generics and habituals

129

7.2 The Imperfective Taylor (1977) and Dowty (1979) argue that while stative verbs are evaluated at moments, the truth of a sentence with such a verb at a given moment t is dependent not on what happens at t alone, but on a longer interval surrounding it. Dowty says: John knows French is made true not by John’s doing anything at that moment, but by past occasions of John-stages having stage-properties of speaking French, and John loves Mary is somehow or other made true by past (and presumably future) instances of John-stages bearing certain relations to Mary-stages. To the extent that an interval of time could be said to be “the” interval of their truth, it would seem to be (in most cases) only a large and vaguely defined interval including a vague number of past instances of the truth of certain stage-predicates, and presumably including a vague number of future instances of certain stage-predicates. p. 179

Let us, therefore, assume the following (context dependent) function: for every moment t, int(t) is the relevant interval in context.30 It is important to emphasize that this is not the interval with respect to which an eventive verb is evaluated—let us indicate the latter by e. Thus, a sentence with a stative verb will have a truth value at a moment t, by virtue of int(t). Therefore, in order for the mismatch to be resolved, and the habitual reading to be obtained, the existence of int(t) must be presupposed. This presupposition can be put in Reichenbach’s (1947) terms: the reference time is included in the relevant event time. Interestingly, this is precisely what the imperfective morpheme is usually taken to mean, in languages that have imperfective morphology (Klein 1992, 1994). The role of the imperfective, then, is to indicate that this presupposition is satisfied. What happens in languages without imperfective morphology? In such languages, the presupposition is simply accommodated. Since accommodation is a rather common, presumably cheap operation, it is not surprising that many languages manage without an imperfective morpheme.31

30 31

See Cohen (1999) for some thoughts on how this interval is determined. Boneh and Doron (2008) claim that there is another interpretation of habituals that is not dependent on imperfectivity. I have nothing to say about this reading here.

130

chapter 5

7.3 Formalization For simplicity, I will assume henceforth that the arguments of eventive verbs are intervals, rather than, as is more plausible, spatio-temporal events whose running times are intervals (Link 1987). I propose that if t is the time of evaluation of a habitual, it expresses quantification over intervals that are contained in int(t). In order to formalize this idea, let us define the following type-shifting operator: Definition 9 γP =def λt.gene [e ≤ int(t)][P(e)].

This operator turns an eventive predicate (i.e. a property of intervals) into a property of moments of time, binding the interval (or event) variable by the generic quantifier. The notation ≤ indicates interval containment. The (intransitive) tenseless verb smoke is a relation between an individual and an interval: (302) λx.λe.smoke(x, e). When the verb applies to an individual, as in (297), we get the following logical form: (303) λe.smoke(j, e). Let t0 designate the speech time. Thus, if the verb is in the present tense, it is predicated of t0 : (304) (λe.smoke(j, e))(t0 ).

As discussed above, this causes a type mismatch: the eventive verb requires as an argument an interval that is longer than a moment, but is applied to a moment. We need to turn the sentence into a stative, by making it habitual. This is performed by an application of the type-shifting operator γ: (305) γλe.smoke(j, e) = λt.gene [e ≤ int(t)][smoke(j, e)].

Finally, when we apply (305) to t0 , we get (306), which is the intended interpretation. (306) gene [e ≤ int(t0 )][smoke(j, e)]

generics and habituals

131

Note that, since γ is a type-shifting operator, it operates as locally as possible, resulting only in interpretations where the habitual quantifier receives narrow scope. Consider, again, (249), repeated below: (307) #John smokes a cigarette. The transitive verb smoke is a relation between two individuals and an interval: (308) λy.λx.λe.smoke(x, y, e). Therfore, the initial logical form of (307) is (309a). This causes a type mismatch, because the third argument of the verb is a moment rather than an interval. Consequently, we apply γ, as (309b), to turn an eventive into a stative. This application results in (309c), where the existential takes scope over the habitual, hence the infelicity of the sentence. (309) a. ∃x(cigarette(x) ∧ smoke(j, x, t0 )) b. ∃x(cigarette(x) ∧ (γλe.smoke(j, x, e))(t0 )) c. ∃x(cigarette(x) ∧ gene [e ≤ int(t0 )][smoke(j, x, e)])

If it were possible to apply γ globally, as in (310a), we wold get the logical form in (310b), where the habitual takes wide scope. However, this reading, although eminently plausible, is not available. (310) a. (γ∃x(cigarette(x) ∧ smoke(j, x, e)))(t0 ) b. gene [e ≤ int(t0 )][∃x(cigarette(x) ∧ smoke(j, x, e))]

Thus, because habituals generate the generic quantifier by type-shifting, it takes narrow scope with respect to other operators. In contrast, when the quantifier is an overt quantificational adverb, as in (252), or when the existence of the quantifier is indicated by an overt restrictor, as in (253a), or by context, as in (253b), there is no need for type-shifting, and scope ambiguities are normal. 7.4 Habituals and Bare Plurals We have, consequently, an explanation why habituals receive narrow scope. But what about BPs? Why do they get even narrower scope? The answer, I propose, hinges on the fact that BPs do not introduce discourse referents, as we have seen in chapter 3, section 5. In contrast, the vague interval int(t0 ) does introduce a discourse referent.

132

chapter 5

Consider the following discourse: (311) John smokes. It started two months ago. Note that the pronoun it indeed refers to the interval of John’s smoking, and not to individual smoking events. Note further that (311) leaves open the possibility that John also smoked at earlier stages of his life. Hence, habituals, unlike BPs, do introduce discourse referents. Using DRT we can, therefore, represent the meaning of (297) as the following DRS: (312)

h t0 j

e≤h

h = int(t0 ) gen e

smoke(j, e)

It is this difference between habituals and BPs that accounts for the scope facts, namely that BPs scope even below the habitual quantifier. Consider (254) again, repeated below: (313) John smokes cigarettes. Recall that we have concluded that existential BPs denote properties, and that they receive their existential reading through type-shifting. Hence, we need to perform two type-shifting operations on this sentence: one, introducing the generic quantifier, results in the habitual interpretation of the verb; and another, introducing the existential quantifier, results in the existential interpretation of the BP. Now consider the DRS corresponding to the attested reading, where the habitual scopes over the BP: (314)

h t0 j

e≤h

gen e

h = int(t0 )

∃x(cigarettes(x) ∧ smoke(j, x, e))

133

generics and habituals

Note that the habitual introduces a discourse referent, h, but the BP does not—its existential quantifier is static.32 Note further that the discourse referent is introduced globally, in the highest DRs. This is permitted by the rules of DRT, which “introduce discourse referents either into the universe of the very DRS to which they are applied or into the universe of a superordinate DRS” (Kamp and Reyle 1993:168). In contrast, Szabolcsi (1997) notes: “No mechanism with a comparable effect is available to [TNPs] that do not introduce a referent” (p. 136). Therefore, the existential quantifier corresponding to the BP does not introduce a discourse referent, and, consequently, cannot escape the scope of the duplex condition. If it could, the DRS corresponding to the interpretation where the BP scopes over the habitual would be generated: (315)

h t0 j

∃x(cigarettes(x)∧ e ≤ h

h = int(t0 ) gen e

cigarettes(x) ∧ smoke(j, x, e) )

But note that, in this DRS, the existential quantifier is outside the duplex condition, which, as we have just seen, is not permitted by the rules of DRT. It is for this reason that the reading where the BP scopes over the habitual is not available. 7.5 The Progressive In this section I have proposed that habitual interpretations result because of type-mismatch: an eventive verb requires a temporal argument that is longer than a moment, but is evaluated at a moment. It should be noted that another way in which an eventive verb can hold at a moment is by putting it in the progressive (Bennett and Partee 1978): (316) is fine, and does not require a habitual interpretaton. (316) John is smoking. In many languages, a verb in the simple present can receive a progressive interpretation. For example, the Dutch sentence (317) can mean that the speaker is dancing at the time of speech. 32

Note also that because, following Krifka (2004), BPs are unspecified for number, (254) does not allow John to smoke more than one cigarette on any relevant occasion.

134

chapter 5

(317) Ik dans. I dance Why is such an interpretation impossible in English? In my explanation I follow Landman (2008).33 In English, the progressive is produced with the auxiliary be. However, in many other languages, including Dutch, no such particle marks the progressive.34 Assuming a suitable type-shifting operator, the verb can be type-shifted to the progressive in Dutch.35 In contrast, in English this is impossible, because the use of the overt auxiliary blocks the option of type-shifting. Hence, the only available option is a shift to the habitual. 33 34 35

Though he uses the notion of accommodation rather than type-shifting. Though, as Landman points out, Dutch does have several progressive constructions. Specifying the nature of this operator requires an investigation of the semantics of the progressive, an issue that lies outside the scope of this book.

chapter 6

Iterativity 1

Durative Adverbials

I have argued above that when an operator is introduced by type-shifting, it can get narrow scope only; but when it is introduced by predicate transfer, its scopal behavior is mostly free (with the exception of opaque contexts). We can test the predictions of this theory by looking at yet another implicit quantifier: the one introduced by iterative readings of durative adverbials (also called foradverbials). The following contrast is well known: (318) a. Mary swatted flies for half an hour. b. #Mary swatted a fly for half an hour. While (318a) makes perfect sense, (318b) is distinctly odd. Various explanations of this fact have been offered (e.g., Dowty 1979; Krifka 1989; Moltmann 1991; Zucchi and White 2001). While these accounts differ on many substantial points, there is a general idea that is, in essence, common to all. Accounts of durative adverbials assume a domain of eventualities. Eventualities are divided into (at least) two types (or, perhaps, sorts): homogeneous eventualities (which are often further divided into states and processes) and non-homogeneous eventualities (which are often called points, or simply events). The difference between the types of eventuality is best explained via an example. An eventuality of swatting flies is homogeneous. This means that it can have a proper part that is also an eventuality of swatting flies. Thus, if Mary swatted flies for half an hour, there were subintervals of this period of time that also consitute an eventuality of swatting flies. Correspondingly, the predicate swat flies is homogeneous (sometimes called cumulative) in that it denotes a set of homogeneous events. In contrast, an eventuality of swatting a fly is not homogeneous, because it does not have proper parts that are also eventualities of swatting a fly. Correspondingly, the predicate swat a fly is not homogeneous (such predicates are often called quantized). Now, the claim is that for-adverbials are restricted in such a way that they do not allow quantized predicates in their scope; this is the reason why (318a) is good but (318b) is bad. © Ariel Cohen, 2020 | doi:10.1163/9789004432598_007

136

chapter 6

There is, however, a (non-swatted) fly in the ointment. It is well known that there are cases when a quantized predicate is actually fine inside the scope of a durative adverbial. For example: (319) John read the letter for a week. The predicate read the letter is quantized: there is no proper subpart of an eventuality of reading the letter that is also an eventuality of reading the letter (though, of course, there may be proper parts that are eventualities of reading parts of the letter). It would seem, then, that (319) ought to be bad, but it is, in fact, perfectly acceptable. Intuitively, the reason is quite clear: (319) is interpreted iteratively. That is to say, we take (319) to mean that John read the letter again and again in the course of one week. In essence, what iterativity does is turn a non-homogeneous eventuality into a homogeneous one. Note that an eventuality of reading the letter again and again is homogeneous: it has proper parts that are also eventualities of reading the letter again and again. Thus, the restriction on durative adverbials, namely that they require homogeneous predicates, is satisfied, and the acceptability of (319) is explained. The object of the verb in (319) is definite; but iterative readings are also possible with indefinites: (320) Mary dialed a phone number for half an hour (Van Geenhoven 2004). The predicate dial a phone number, just like read the letter, is quantized: there is no subpart of an eventuality of dialing a phone number (in its entirety) that is also an eventuality of dialing a phone number. Hence, (320) ought to be bad. However, under an iterative interpretation, the predicate becomes cumulative, and the sentence is acceptable: it means that Mary dialed a phone number again and again, and this happened during the course of one hour.

2

The Origins of Iterativity

Where does iterativity come from? Krifka (1989) contrasts iterative readings with “normal, that is, non-iterative interpretations” (p. 75). By implication, iterative readings are therefore somehow abnormal. Recall that these readings arise when, otherwise, the predicate would be quantized and the sentence would be bad. In other words: iterative readings arise when there is a type mismatch between a quantized predicate and the

137

iterativity

requirement of a durative adverbial. Hence, they constitute a case of typeshifting.1 Various authors (e.g., Pulman 1997; de Swart 1998; Egg 2005) have proposed type-shifting operators that derive the iterative interpretation; for concreteness, I choose de Swart’s formalization. The reason is that de Swart’s theory is explicitly couched in terms of Discourse Representation Theory (Kamp and Reyle 1993), and thus makes it easy to represent the scopal relations between the iterative operator and other operators. De Swart proposes that iteration results from the application of an aspectual coercion operator, Ceh , which turns an event (a non-homogenous eventuality) into a state (a homogeneous eventuality).2 Thus, for example, the plausible interpretation of (321) is that John played a single sonata again and again for eight hours. (321) John played a sonata for eight hours. De Swart representes this reading as the following Discourse Representation Structure (discarding elements that do not pertain to our purpose): (322)

h xj sonata(x) dur(h) ≥ eight-hours time(h) < now h:

e′

Ceh

e′ : play(j, x)

1 Perhaps there is only one type of eventualities, and the distinction between homogeneous and non-homogeneous eventualities is a sort, rather than a type distinction. In this case, we would have sort-shifting rather than type-shifting; but this should not affect the discussion in this chapter, as noted in chapter 2, section 2.6 above. 2 It is not clear to me why the resulting eventuality is a state rather than a process, as indeed

138

chapter 6

This DRS is interpreted as follows: there is a sonata such that, in the past, there was a homogeneous eventuality whose duration was at least eight hours, and which consisted of repeated events of playing this sonata by John.

3

Iterativity and Scope

Note that in DRS (322), the iterativity operator takes narrow scope with respect to the indefinite a sonata: this DRS means that the same sonata was played again and again. If, indeed, the iterative operator is generated via type-shifting, we predict that this ought to be the only scopal relation between the iterativity operator and the indefinite. More formally, we predict that while (322) is fine, (323), in which the indefinite is inside the scope of the iterative operator, ought to be bad: (323)

h j dur(h) ≥ eight-hours time(h) < now e′ x

*

h:

Ceh

sonata(x)

e′ : play(j, x)

Indeed, (323) corresponds to a reading where John played possibly different sonatas for eight hours, a reading which (321) does not have. The sentence can only be interpreted as saying that John repeated the same sonata again and again for eight hours—the reading captured by (322). proposed by Pulman (1997) and Egg (2005); but the exact type of the resulting eventuality is not crucial for our purposes here.

iterativity

139

This restriction of the iterative operator to receive narrow scope appears to be shared by other sentences we have considered. For example, (320) above can only mean that Mary dialed the same number again and again, not that she tried different numbers. Additional examples are presented by MacDonald (2006): (324) a. #The farmer dragged a log into the barn for an hour. b. #The kid pushed a stereo into the garage for an hour. c. #The girl carried a bag into the store for an hour. MacDonald notes that “the object in motion that arrives at the goal must be the same object undergoing the motion in each of the iterated events” (p. 29). The obligatoriness of narrow scope can solve a puzzle that has always been a thorn in the side of theories of durative adverbials. We have said that (318b) is bad because the for-adverbial requires a cumulative predicate, but swat a fly is quantized. But we have also said that a bad durative can become good if it is interpreted iteratively. Why, then, can’t iterativity “save” (318b) and make it acceptable? Why doesn’t this sentence have an iterative interpretation? The answer is that (318b) does, in fact, have an iterative reading. But it is a very odd reading, where the same poor fly is swatted again and again. The reason is that the iterative operator is restricted to narrow scope only. If it could scope over the indefinite, (318b) would be read as saying that Mary was engaged in a series of swatting events, but she swatted different flies every time. This would be a perfectly plausible scenario; such a reading, however, is unavailable.3

4

Alternative Explanations

I have proposed above that the scopal restrictions on iterative interpretations are due to the generation of the iterativity operator by type-shifting. This theory competes with two alternative explanations that have been made in the literature. Let us consider them in turn.

3 In contrast, (318a) does not have an iterative interpretation; the reading where Mary swatted different flies is generated naturally from the cumulativity of the BP. Carlson (1977) seems to think that both (318a) and (318b) have iterative readings, and characterizes the difference between them as another case of differentiated scope (see chapter 3, section 2.2); but there is no need for the iterativity operator in (318a), since there is no type mismatch.

140

chapter 6

4.1 Iterativity as an Inherently Narrow Scope Operator? One may, perhaps, suggest that the iterativity operator is restricted to narrow scope not because it is introduced by type-shifting, but because of something inherent in the nature of iterativity. But this is simply not the case. If iterativity is indicated overtly, and not as a result of type-shifting, wide scope is possible. For example, Zucchi and White (2001) note that while (325a) can only mean that John pushed the same cart again and again, (325b) allows him to push different carts on different days. (325) a. John pushed a cart for a year. b. John pushed a cart every day for a year. Similarly, while (326a) only has the odd reading where John rediscovers the same flea again and again, (326b) has the perfectly reasonable reading where John discovers different fleas every day. (326) a. #John found a flea on his dog for a year. b. John found a flea on his dog every day for a year. Note that even when iterativity is merely implied, this is sufficient to introduce the iterative operator, and no type-shifting is necessary. Therefore, the iterative operator is allowed to take wide scope, as can be seen by the following examples: (327) a. #John baked a cake for six weeks. b. John baked a cake for dinner for six weeks. Sentence (327a) can only mean that John baked and rebaked the same cake for six weeks. In contrast, it is sufficient fot the phrase for dinner in (327b) to induce iteration over dinners. Since this iteration is not produced by type-shifting, it is not restricted to narow scope only, and (327b) gets the plausible reading where John baked different cakes for different dinners. 4.2 Iterativity as a Verb-Level Operator? A different explanation is proposed by Van Geenhoven (2004). She considers West Greenlandic, where iterativity is indicated by the morpheme -tar-. Van Geenhoven presents the following examples:

iterativity

141

(328) a. #Nuka ullaa-p tunga-a tama-at N.ABS morning-ERG direction-3SG.SG.ABS3 all-3SG saniuqqut-puq. go.by-IND.[-tr].3SG ‘Nuka took the entire morning to pass by (moving very slowly).’ b. Nuka ullaa-p tunga-a tama-at N.ABS morning-ERG direction-3SG.SG.ABS3 all-3SG saniuqqut-tar-puq. go.by-repeatedly-IND.[-tr].3SG ‘Nuka went by repeatedly for the whole morning.’ Sentence (328a) only has the implausible reading where Nuka walked extremely slowly, taking the whole morning to pass by. In contrast, when the morpheme -tar- is combined with the verb, the sentence becomes quite acceptable: (328b) receives a natural interpretation, namely that Nuka came by several times throughout the morning. Van Geenhoven argues that -tar- is a verb-level operator. She claims that iterativity in English is expressed by a phonologically null morpheme, which is like -tar- in being restricted to the level of the verb; hence its narrow scope. In contrast, overt Q-adverbs can operate at levels that are higher than the verb, hence their scopal ambiguity.4 This proposal is problematic on several counts. Even looking only at West Greenlandic, it is far from clear that -tar- is, indeed, a verb-level operator. Indeed, Bittner (1987, 1995) has demonstrated that -tar- can operate at higher levels. Van Geenhoven acknowledges this fact but says that she wants “to focus on the ability of -tar- to operate on the verb-level” (p. 157n14). However, in order to explain the obligatoriness of narrow scope, she needs to demonstrate not that -tar- can operate at the level of the verb, but that it must. But, in fact, Van Geenhoven admits that -tar-, unlike iteratives in English, is not restricted to narrow scope only, and does exhibit scope ambiguities. She presents the following example, from Fortescue (1984): (329) marlu-riar-lu-ni quirsur-tar-puq. two-do.times-INF-3SG.PROX cough-repeatedly-IND.[-tr].3SG

4 Van Geenhoven also attempts to account for examples like (318a) using iterativity; I will not get into the details of her proposal here.

142

chapter 6

Sentence (329) is ambiguous: it could mean that he coughed twice, in which case -tar- takes scope only over the verb. But it could also mean that he repeatedly coughed, each time doing it twice; in this case -tar- takes wide scope. Therefore, it is far from clear how the comparison with West Greenlandic helps Van Geenhoven’s claims about English. Even granting Van Geenhoven’s claims concerning West Greenlandic, there is absolutely no evidence that, in English, the null iterativity operator applies to the verb. Van Geenhoven’s only argument for this is that if “silent operators … were allowed to operate on projections of [lexical or functional] categories, we would end up with scope ambiguities in natural language which are too numerous to be mastered” (p. 173). But even if this claim were true, it would amount simply to a statement of the problem, and not an explanation of it. But the fact is that the claim is not true. As we have seen with the examples in (327) above, in the appropriate context the iterative operator can receive wide scope, and, in particular, can scope over the verb. It is not clear how this observation can be explained under Van Geenhoven’s proposal. We can conclude, then, that the reason why iterativity is restricted to narrow scope only, does not come from something special about the meaning of iterativity. Rather, it is because iterativity is generated by type-shifting.

5

Iterativity as a Quantifier

The iterativity operator, then, is an operator that is generated by type-shifting, and is restricted to narrow scope only. But what sort of operator? Naturally enough, most authors formalize iterativity as some sort of quantifier. But what is its quantificational force? Dowty (1979) suggests that iteratives are quantifiers that do not have a fixed quantificational force. He considers (330). (330) John found his son’s tricycle in the driveway for six weeks (p. 82). Clearly, (330) can only be read iteratively: there were repeated events of John’s finding his son’s tricycle within an interval of six weeks. How is this best formalized? Dowty suggests that iterativity is understood as elliptical for some adverbial quantifier. The value of this quantifier could be, for example, every day, once a week, frequently, and so on. At a given utterance of (330), the precise value of the quantifier will presumably be recovered from the context.

iterativity

143

However, it is not clear how the force of the iterative quantifier is determined by the context. It might be argued that the explicit mention of weeks in (330) induces the interpretation once a week, but this is implausible. Suppose John found his son’s tricycle only on five of the six weeks in question; would this make (330) false? The answer appears to be no. In fact, if there is any plurality of events of John’s finding his son’s tricycle during the six weeks, (330) appears to be true.5 Therefore, it makes more sense to regard iterativity as an existential over pluralities: (330) says that there is a plurality of events of John’s finding his son’s tricycle in the driveway. We can formalize this interpretation of iterativity as explicit quantification over pluralities (Pulman 1997; de Swart 1998; Van Geenhoven 2004; Egg 2005). However, I choose to follow Krifka (1989), who treats iterativity as a double existential over singular events, thus obviating the need for pluralities. This approach is preferred for the purposes of this book, because it makes it easier to observe the relative scope of the operator, without intereference from the possible distributitive interpretation of pluralities. Specifically, Krifka (1989) proposes that iterativity is a relation between an event e, an individual x, and a thematic role θ, and it means that x is the θ of at least two different parts of e. For example, Krifka represents the predicate read a letter as follows (ignoring the subject argument): (331) λe.read(e) ∧ ∃x(letter(x) ∧ patient(e, x)).

This predicate is quantized; to combine it with a for-adverbial, we must make it iterative. The result is a predicate that says, essentially, that there is a letter that is read more than once: (332) λe.read(e) ∧ ∃x(letter(x) ∧ patient(e, x)∧ ∃e′ ∃e′′ (e′ ≠ e′′ ∧e′ ⊆ e ∧ patient(e′ , x)∧ e′′ ⊆ e ∧ patient(e′′ , x))).

Note that, instead, we could have placed quantification over letters inside the scope of quantification over sub-events. The resulting logical form would be:

5 If the plurality is too small, the sentence might be perceived to be misleading, so there might be an implicature associated with iterativity to the effect that the plurality is of “sufficient” size.

144

chapter 6

(333) λe.read(e)∧ ∃e′ ∃e′′ (e′ ≠ e′′ ∧e′ ⊆ e ∧ ∃x(letter(x) ∧ patient(e′ , x))∧ e′′ ⊆ e ∧ ∃x(letter(x) ∧ patient(e′′ , x))).

This logical form would mean that, more than once, a (possibly different) letter was read. Such a reading, however, is unavailable: read a letter just cannot mean that. So, iteration must be under the scope of the existential quantifier introduced by the indefinite. We can conclude that the prediction of the theory proposed in this book is borne out: iterativity is a quantifier invoked by type-shifting, and, therefore, it receives narrow scope only.

chapter 7

The Nature of Implicit Quantification 1

Two Implicit Quantifiers

1.1 The Existential and the Generic Quantifiers This book argues that there are two processes by which implicit quantifiers are introduced: type-shifting and predicate transfer. These processes are very different in their properties and in the conditions that invoke them. Typeshifting is semantic in nature, it is mandatory and is invoked when there is a type mismatch, and it allows the quantifier it introduces to receive narrow scope only; while predicate transfer is pragmatic in nature, it optional and is invoked when there is a pragmatic anomaly, and allows almost free scopal interactions. Interestingly, both of these very different processes result in the introduction of the same implicit quantifiers: existential and generic. Type-shifting introduces the existential quantifier (existential BPs and iterative interpretations) and generic quantifier (habituals); and predicate transfer, too, introduces the existential quantifier (“ham sandwich” sentences) and the generic quantifier (generics). The question is: why do such different mechanisms result in the introduction of exactly these two quantifiers, and no others? There must be something special about the existential and generic quantifiers, which causes them to be invokved as implicit quantifiers. But what is it? The introduction of the existential quantifier is, perhaps, less of a surprise. It is, in some sense, the simplest quantifier, and it is often assumed to apply with no phonological correlation, as an operation of existential closure (Heim 1982). But the fact that the second implicit quantifier is the generic is more surprising. If anything, we might expect the universal quantifier—the natural dual of the existential quantifier. 1.2 Are Generics Modal Universals? In fact, such a proposal is made by Nickel (2016). He discusses the phenomenon of distributivity: a sentence such as (334a) is normally interpreted as (334b). (334) a. The kids sneezed. b. Every one of the kids sneezed.

© Ariel Cohen, 2020 | doi:10.1163/9789004432598_008

146

chapter 7

This is usually accounted for by postulating a phonologically null distributivity operator with essentially the semantics of the universal quantifier (Link 1983 and many others). Nickel (2016) uses this fact to “argue that the generic operator just is a distributive operator” (p. 125). Hence, both the generic and the distributive operators are universal quantifiers, except that the former is modal, whereas “the latter applies to groups without modal import” (p. 125). In that, Nickel follows a long list of scholars (Delgrande 1987; Morreau 1992; Asher and Morreau 1995; Krifka 1995; Pelletier and Asher 1997; to name just a few) who argue that the generic quantifier is universal, albeit a modal universal. These are really two claims: the first is that generics are modal, and the second—that they are universal. We have already noted, in chapter 4, section 4.2 above, that there are reasons to think that the intensionality of kinds is related to time, rather than to possible worlds; if generics are modal, their modality is of rather a special sort (see Cohen 2012 for my take on the modality of generics). Even if the generic quantifier is modal, it is not a modal universal. There are several acknowledged tests for universal (and negative universal) quantifiers, and, as we will see, the generic quantifier fails all of them. One test involves the words almost and absolutely, which can only modify universals: (335) a. { b. *{

Almost } everyone/nobody was there. Absolutely

Almost } some/few/three/many/most people were there. Absolutely

The same results obtain when we apply the test to adverbial quantifiers: (336) a. Spencer { b. *Spencer {

almost } always/never comes to office parties. absolutely

almost } sometimes/seldom/twice/often/usually absolutely comes to office parties.

If we apply this test to generics, we see that they do not behave like universals, in that they cannot combine with almost or absolutely: (337) *{

Almost } birds fly. Absolutely

the nature of implicit quantification

147

A similar test involves even, which also combines only with universal quantifiers (Zuber 2004): (338) a. Every bird, even Tweety, is sad. b. No bird, not even Tweety, is sad. c. *Some/Few/Three/Many/Most birds, even Tweety, are sad. The test gives the same results with adverbial quantifiers: (339) a. Peter always, even on his birthday, comes to the office. b. Peter never, not even on his birthday, stays home. c. *Peter sometimes/seldom/twice/often/usually, even on his birthday, stays home. And, once again, generics pattern with the non-universal quantifiers, and cannot combine with even: (340) *Birds, even penguins, have wings.1 We can conclude, then, that the generic quantifier, whatever its exact interpretation, is not a universal. Why, then, is it the one, together with the existential, that is introduced by mechanisms of implicit quantification?

2

A Preference for Inference

2.1 Unmarked Forms and Unmarked Interpretations Let us begin with a trivial point. Sentences with implicit quantifiers, like all sentences, have both form and interpretation. Until the quantifier, whether existential or generic, is introduced, the form is bare—it contains no quantifier. And a bare form is obviously shorter than a form that is exactly the same except that it also contains an overt quantifier. There is an idea that has been formulated by Atlas and Levinson (1981) and Horn (1984), but goes back considerable time before them, and it can be expressed succinctly by the following slogan: unmarked forms receive unmarked interpretations. The underlying notion is simple: both speaker and 1 There are additional tests for universal quantification, which generics also fail (Moltmann 1995; Zuber 1998; Krifka 2001); however, these tests seem inapplicable to adverbial quantifiers, so I did not include them here.

148

chapter 7

hearer want to minimize their effort. Unmarked forms are easier for the speaker to produce; unmarked meanings are easier for the hearer to understand. Hence, unmarked forms are preferred to marked forms, and unmarked interpretations are preferred to marked interpretations. Since shorter forms are easier to produce for the speaker, they are considered unmarked; thus, it is often assumed that shorter forms receive an unmarked interpretation. For an illustration, consider the following minimal pair, from McCawley (1978): (341) a. Black Bart killed the sheriff. b. Black Bart caused the sheriff to die. McCawley notes that while (341a) implies that Black Bart killed the sheriff in the expected way (i.e. shot him), (341b) implies some indirect, perhaps unusual way of killing, e.g. sabotaging the sheriff’s own gun so that it backfires. The notion that unmarked forms receive unmarked interpretations has been formalized in Bidirectional Optimality Theory (e.g. Blutner 1998, 2000) and also received a Game Theoretical account (e.g. Dekker and van Rooy 2000; van Rooy 2004). Since bare forms are unmarked, we expect their meaning to be unmarked too. What is an unmarked meaning? The prevailing view is that the unmarked interpretation is the stereotypical one: it is easier for the hearer to reach a stereotypical interpretation, because it is more salient. This is why (341a) describes a stereotypical killing, while (341b) implies a more unusual event. 2.2 Bare Forms and Stereotypicality Let us now try to apply this idea to BPs. Concerning generic BPs, the idea that generics express streotypicality has at least an initial appeal. Consider a simple generic sentence: (342) Ravens are black. Clearly, this is not a universal, since the existence of the odd albino raven does not render (342) false. But what, then, does it mean? The meaning of a generic is a hotly debated topic, and I am not going to address it here.2 At a pretheoretic level, however, a reasonable approximation of the meaning of the sentence is

2 See Cohen (1996, 1999) for my take on this question.

the nature of implicit quantification

149

that it makes a statement not about all ravens, but about stereotypical ravens. Since an albino raven is not stereotypical, it does not count an an exception, and does not falsify (342). Note that I am not advocating here that the truth conditions of (342) are captured by an appeal to stereotypicality—indeed, in a moment I will argue against such a claim.3 All I am saying is that when (342) is interpreted, there is a “feeling” that stereotypicality is involved. Nothing more than this rather weak and, I believe, non-controversial claim is necessary for the purposes of this book, as we shall see. In contrast with generics, existential bare plurals are usually supposed to express nothing more than a simple existential claim, and people rarely associate them with notions of stereotypicality. And yet, stereotypicality does appear to play a role in their interpretation (Cohen 2005), as the following examples demonstrate: (343) a. This tractor has wheels. b. This tractor has some wheels. Suppose the tractor in question has only two wheels. Then (343a) would be odd, but (343b) would be fine. Sentence (343a), but not (343b), suggests that the tractor has four wheels, suitably arranged: two large ones in rear, two smaller ones in front. In other words, (343a) implies that the tractor has the stereotypical arrangement of wheels. For another example, consider the following pair, suggested by Tova Rapoport (pc): (344) a. John has playing cards. b. John has Victorian playing cards. Sentence (344a) suggests that John has the stereotypical set of cards, i.e. a full deck. Sentence (344b), in contrast, may be felicitously uttered even if John has only a few Victorian cards; in the context of this sentence, John is most probably a collector of Victorian cards, and there is no specific set of cards that would be considered stereotypical for collecting purposes. Even the “classic” example of an existential reading of a bare plural, namely the subject of available, may give rise to stereotypicality. Suppose we wish to send a spaceship to the moon. We contact NASA, and get the following response: 3 But see Geurts (1985) and Declerk (1986) who argue for this claim.

150

chapter 7

(345) Astronauts are available. Sentence (345) says more than simply that there exist some available astronauts. Rather, it implies that there is a set of available astronauts that is stereotypical, in terms of its size, the training of its members, etc., in the context of our mission: there are, say, three astronauts, who have the respective roles of Command Module Pilot, LEM Pilot, and Mission Commander. Note that this feeling of stereotypicality is perceived by the hearer, who may felicitously respond to it as if it were explicitly made: (346) a. A: This tractor has wheels. B: So where do you want to go with it? b. A: John has playing cards. B: Great, let’s start a game. c. A: Astronauts are available. B: But the mission cannot go ahead, because the rocket is still malfunctioning. B’s response in (346a) relates to the implied claim that the tractor has a set of wheels suitable for the purpose of riding it; in (346b), B responds to A’s implication that John has a set of playing cards suitable for playing; and in (346c), A’s purpose is to indicate that the mission is ready (and the set of available astronauts is the appropriate stereotypical set), and B responds to this. 2.3 Bare Forms and Non-stereotypicality It appears, then, that bare plurals, both generic and existential, are associated with stereotypicality. However, in Cohen (2006) I argue that the association of unmarked forms with sterotypical meanings does not hold in general, and suggest an alternative. 2.3.1 Generic Bare Plurals Let us first look at generics: in fact, it is quite easy to show that generics do not, in general, express stereotypicality. Take (347), for example. (347) #Mammals are placental mammals. The stereotypical mammal is certainly a placental mammal. Hence, if generics expressed statements about stereotypes, (347) ought to be unproblematically true; but the fact is that it is quite bad. It might be argued that the problem with (347) is simply the fact the same word, mammals, occurs in both subject and predicate—perhaps this is what

the nature of implicit quantification

151

makes the sentence awkward. To see that this is not the case, consider the following examples: (348) a. #Primary school teachers are female. b. #People are over three years old (Henk Zeevat, pc). Although the stereotypical primary school teacher is female, (348a) is bad, and although the stereotypical person is an adult, (348b) is very odd. In order to account for such facts, I argue elsewhere (Cohen 1996, 1999, 2004b) that generics carry a homogeneity presupposition.4 The generic gen(ψ, ϕ) presupposes that its domain, ψ, is homogeneous, in the following sense: for any psychologically salient criterion by which ψ may be partitioned into subsets, the conditional probability of ϕ is roughly the same given every such subset of ψ. That is to say, the domain of a generic may not have “chunks” where there are significantly more ϕs or significantly fewer ϕs than there are in the rest of ψ. Homogeneity corresponds rather well to the pre-theoretical notion of what a generic sentence means. For example, suppose a friend is coming to Israel for a visit, and is worried about whether she will be able to manage, speaking only English. We reassure her by saying (349) Israelis speak English. Observe that (349) means more than simply that if you meet an Israeli, he or she is likely to speak English; in addition, the sentence requires that, wherever you go in Israel, whichever group of Israeli society you associate with, a member of this community will be likely to speak English. Indeed, suppose the friend spent all her visit in a town where nobody spoke English, or with members of some group of Israeli society where English was rarely spoken. In such a case, she would be justified in accusing us of misleading her. Homogeneity can explain the oddness of examples (347), (348a), and (348b) above. If we partition the domain of mammals according to biological groups, homogeneity will be violated: one subset (the placental mammals) satisfies the property of being a placental mammal, while another (marsupials) does not. Hence, assuming the criterion of biological group is salient, (347) is bad. Sentence (348a) is odd because partition according to sex violates homogeneity: one subset (the females) satisfies the property, another (the males) does

4 Not to be confused with the homogeneity of eventualities, discussed in chapter 6.

152

chapter 7

not. And (348b) is bad because partition according to age violates homogeneity: some subsets (adults) satisfy the predicated property, others (babies and toddlers) do not. Of course, this theory crucially relies on what determines whether a partition is salient or not. There is some psychological evidence for such criteria, which are rather complex, and depend on one’s world knowledge, the linguistic and extralinguistic context, and even on the way in which the properties are expressed.5 See Cohen (2004b) for a thorough discussion. Why do generics have the requirement of homogeneity? In Cohen (1996) I suggest that homogeneity is useful for inference. If the domain is homogeneous, we are safe from local minima. That is to say, we will not find ourselves in a situation where, because of bad luck, we happen to be dealing with a subset of the domain where all our inferences are wrong. For example, recall that (342) is acceptable and true, despite the existence of albino ravens. Why? The reason is that albino ravens are homogeneously distributed throughout the raven community—there are no colonies of albino ravens. Hence if, whenever we encounter a raven, we infer that it is black, this inference will be justified. It might not always prove correct—sometimes we will encounter the odd albino raven—but it will not be repeatedly wrong. Do people, indeed, require that a domain be homogeneous when they draw inferences about it? Suppose we wanted to find out if a domain ψ is homogeneous with respect to property ϕ, by observing properties of its instances. If we pick individuals that are similar to one another, and find out that they have property ϕ, this would not be good enough; the specimens might all be subsumed by one subclass, which has properties different from other subclasses of the domain, which may not satisfy ϕ. If, on the other hand, the instances we examine are dissimilar, they are likely to belong to different subclasses, and the domain is more likely to be homogeneous. Therefore, if, indeed, when people draw inferences about a domain, they require it to be homogeneous, the following ought to be the case: when one notes that some subordinates, say s1 and s2 , of a domain ψ, satisfy property ϕ, and uses this fact to argue that ϕ applies to ψ as a whole, the strength of this argument will be judged to be proportional to the perceived dissimilarity of s1 and s2 . Exactly this behavior has, indeed, been observed by Osherson et al. (1990). They consider inductive arguments, for example:

5 This aspect of the theory is sometimes misunderstood, e.g., Nickel (2012) seems to think that only the last factor is argued to be important.

the nature of implicit quantification

153

(350) Sparrows have sesamoid bones. Eagles have sesamoid bones. All birds have sesamoid bones. In this argument, information about subclasses (sparrows and eagles) of a class (birds) is used to derive a rule about all the members of the class.6 Note that here, as in the rest of the arguments used in Osherson et al.’s experiments, the sentences use detailed biological information, which laypersons are not expected to know beforehand. Thus, their judgments are expected to rely solely on the perceived strength of the formal argument. Not surprisingly, Osherson et al. have found that if information about more subclasses is presented, the argument is judged to be stronger. For example, the argument in (351) is stronger than that in (350). (351) Hawks have sesamoid bones. Sparrows have sesamoid bones. Eagles have sesamoid bones. All birds have sesamoid bones. This is because in (351) we have evidence coming from an additional subclass, hawks, which was absent in (350). Knowing about more members of a class that satisfy the property (with none that violate it) increases the probability that all members satisfy it, hence strengthens the argument. It turns out that not all supporting evidence strengthens an argument in the same way. Osherson et al. have tested the following argument: (352) Hippopotamuses have a higher sodium concentration in their blood than humans. Hamsters have a higher sodium concentration in their blood than humans. All mammals have a higher sodium concentration in their blood than humans.

6 The conclusion here is not a generic, but a universal, so the rule does not allow exceptions; but this does not affect the point I am making here.

154

chapter 7

Information about hamsters adds to information about hippopotamuses to provide an argument about all mammals. This argument is judged by subjects to be quite strong. Interestingly, if, instead of hamsters, we add information about rhinoceroses, the argument is judged to be considerably weaker: (353) Hippopotamuses have a higher sodium concentration in their blood than humans. Rhinoceroses have a higher sodium concentration in their blood than humans. All mammals have a higher sodium concentration in their blood than humans. Now note that hippopotamuses and rhinoceroses are rather similar; they are both large herbivorous African mammals. But hippopotamuses and hamsters are rather different, in their size, habitat, behavior, etc. Thus, knowing that hippopotamuses and rhinoceroses share some property leaves open the possibility that this property is only characteristic of some subclass of mammals, say that of the large herbivorous African mammals; if this is the case, the domain of mammals would not be homogeneous with respect to this property, and hence the argument is weak. In contrast, it is hard to think of a natural subclass of mammals that includes both hippopotamuses and hamsters but leaves other mammals out. Therefore, if we find that both hippopotamuses and hamsters share a property, the domain is more likely to be homogeneous, hence the argument is judged stronger. These results, then, indicate that, given an inductive argument, the more evidence it provides that the domain is homogeneous, the stronger it is judged. Thus, a generic implies that inferences about its domain are generally reliable. Of course, one of the ways to aid inference is stereotypicality; indeed, this is arguably why we have stereotypes. But it is not the only way: another one is homogeneity. 2.3.2 Existential Bare Plurals We have seen above that existential bare plurals often imply stereotypicality. Indeed, stereotypicality can aid inference. For example, the hearer of (343a) can draw inferences about the tractor (e.g. that it can run). However, existential bare plurals are not always interpreted stereotypically. Consider the following sentence: (354) In this forest, trees are dying.

the nature of implicit quantification

155

The stereotypical, indeed the common situation where trees are dying is where the trees are concentrated in some area, the location that was first infected. However, in such a case, where only a single grove out of the forest is dying, (354) would be quite odd. Rather, (354) seems to indicate that the dying trees are homogeneously distributed throughout the forest.7 As in the case of generics, so in the case of existential bare plurals, homogeneity serves as an aid to inference. The hearer of (354) can draw inferences about the forest (e.g. that it is sick), which would not be possible if only a single grove were infected. For further examples of existential bare plurals that express homogeneity, consider the following sentences, after Greenberg (1994): (355) a. (Although it is winter now, in our country) lakes are dry. b. (Wellington’s army has won a great victory today, but) soldiers are tired. c. (It is New Year’s Eve, and) restaurants are full. Note that if only lakes located in a specific region of the country, or only saltwater lakes were dry, (355a) would be odd; rather, the sentence implies that throughout the country, lakes are dry. Similarly, (355b) would be bad if only the young soldiers, or only the soldiers in a specific platoon were tired; its acceptability requires that soldiers from a variety of ages, units, etc. be tired. As for (355c), if only Italian restaurants, or only restaurants on a specific street were full, it would be odd; (355c) implies that various types of restaurant, throughout town, are full. 2.4 Inference and Preference Hence, in Cohen (2006) I suggest that the crucial notion which prefers an interpretation of a bare nominal is not whether the interpretation is stereotypical as such, but whether it provides information that is useful for inference. It should be emphasized that homogeneity and stereotypicality are quite distinct concepts. While it might be argued that stereotypicality is the unmarked interpretation, no such argument can be made for homogeneity. Indeed, there is no reason to assume a-priori that the domain of inference is homogeneous—in most cases, this will not be so. Hence, the implication of homogeneity is actually quite informative, and it is therefore not simply the unmarked reading. 7 Or, at least, that the disease is spreading, and that the dying tress will be homogeneously distributed in the near future.

156

chapter 7

The crucial notion, then, is not unmarkedness, but aid to inference. We prefer an interpretation not on the basis of how easy it is to assume it, but on the basis of its potential to support inference. Of course, in any given situation there might be a different quantifier that best supports inference in that particular situation. But we are looking for interpretations that best support inference across a variety of circumstances. This means that an interpretation is preferred if it aids inference, by either: 1. being more stereotypical, or 2. by implying that the domain of inference is homogeneous. Since sentences with implicit quantifiers have an unmarked form, and consequently receive an unmarked interpretation, we can therefore conclude that the quantifiers induced by reinterpretation mechanisms are those that provide interpretations that are best associated with their unmarked forms, i.e. best support inferences. From this it follows that the universal quantifier could not have been generated by reinterpretation, since, although it does give us a fair amount of inference power, it is very rarely that we know anything with such assurance, so that we can use the universal for our inferences. Hence, the universal quantifier does not really aid inference very much in practice: we are bound to face exceptions.8 Since we can rarely have the complete assurance of the universal quantifier, we want the next best thing: an assurance that, if we make mistakes, we will not make them repeatedly, all the time. This is precisely what the generic qantifier provides us with, by way of homogeneity. And since the existential also has the property of homogeneity, and aids inference in this way, it, too, may be invoked. Hence, the two quantifiers that aid practical inference, the existential and the generic, are those that are best associated with unmarked forms, and are therefore those that may be introduced by reinterpretation mechanisms. 8 It should be noted that a modal universal would be even less help.

References Abney, S. 1987. The English Noun Phrase in Its Sentential Aspect. MIT dissertation. Asher, N. 2011. Lexical Meaning in Context: A Web of Words. Cambridge: Cambridge University Press. Asher, N., and Morreau, M. 1995. What some generic sentences mean. In Carlson and Pelletier (1995) 300–338. Asher, N., van de Cruys, T., and Abrusán, M. 2015. Lexical semantics today. In The Routledge Handbook of Linguistics, ed. by K. Allan 169–186. New York: Routledge. Atlas, J., and Levinson, S.C. 1981. It-clefts, informativeness, and logical form: Radical pragmatics (revised standard version). In Radical Pragmatics, ed. by P. Cole 1–61. New York: Academic Press. Bach, K. 1994. Conversational impliciture. Mind and Language 9.124–162. Bennett, M., and Partee, B.H. 1978. Toward the Logic of Tense and Aspect in English. Bloomington: Indiana University Linguistics Club. Bhatt, R., and Anagnostopoulou, E. 1996. Object shift and specificity: Evidence from kophrases in Hindi. In Papers from the 32nd Regional Meeting of the Chicago Linguistic Society, ed. by L.M. Dobrin, K. Singer, and L. McNair 205–239. Bittner, M. 1987. On the semantics of the Greenlandic antipassive and related constructions. International Journal of American Linguistics 53.194–231. Bittner, M. 1995. Quantification in Eskimo: A challenge for compositional semantics. In Quantification in Natural Languages, ed. by E. Bach, E. Jelinek, A. Kratzer, and B.H. Partee 59–80. Dordrecht: Kluwer. Błaszczak, J., and Geist, L. 2000. Kopulasätze mit den pronominalen Elementen to/ėto im Polnischen und Russischen. In Copular and AUX-Constructions, ed. by E. Lang 115–139. Berlin: ZAS. ZAS Papers in Linguistics 16. Blok, P.I., and Eberle, K. 1999. What is the alternative? the computation of focus alternatives from lexical and sortal information. In Focus: Linguistic, Cognitive, and Computational Perspectives, ed. by P. Bosch and R. van der Sandt 105–119. Cambridge: Cambridge University Press. Blutner, R. 1998. Lexical pragmatics. Journal of Semantics 15.115–162. Blutner, R. 2000. Some aspects of optimality in natural language interpretation. Journal of Semantics 17.189–216. Bobaljik, J.D. 2002. A-chains at the PF-interface: Copies and ‘covert’ movement. Natural Language and Linguistic Theory. Boneh, N., and Doron, E. 2008. Modal and temporal aspects of habituality. In Syntax, Lexical Semantics, and Event Structure, ed. by M. Rappaport-Hovav, E. Doron, and I. Sichel 338–362. Oxford: Oxford University Press. Bošković, Ž. 2008. What will you have, DP or NP? In Proceedings of the 37th Annual

158

references

Meeting of the North Eastern Linguistic Society, ed. by E. Elfner and M. Walkow 101– 114. Brinton, L.J. 1987. The aspectual nature of states and habits. Folia Linguistica 21.195–214. Burton-Roberts, N. 1977. Generic sentences and analyticity. Studies in Language 1.155– 196. Butt, M. 1993. Object specificity and agreement in Hindi/Urdu. In Papers from the 29th Regional Meeting of the Chicago Linguistic Society 89–103. Carlier, A. 1989. Généricité du syntagme nominal sujet et modalités. Travaux de Linguistique 19.33–56. Carlson, G. 1977. Reference to Kinds in English. University of Massachusetts at Amherst dissertation. Also published 1980, New York: Garland, to which edition page numbers refer. Carlson, G. 1989. On the semantic composition of English generic sentences. In Chierchia et al. (1989) 167–192. Carlson, G. 1995. Truth-conditions of generic sentences: Two contrasting views. In Carlson and Pelletier (1995) 224–237. Carlson, G., and Pelletier, F.J. (eds.) 1995. The Generic Book. Chicago: University of Chicago Press. Carston, R. 2002. Thoughts and Utterances!: The Pragmatics of Explicit Communication. Oxford: Blackwell. Chand, V., Kelleher, A., Rodriguez, A., and Schmeiser, B. (eds.) 2004. Proceedings of WCCFL 23. Somerville, MA: Cascadilla Press. Cheng, L., and Sybesma, R. 1999. Bare and not-so-bare nouns and the structure of NP. Linguistic Inquiry 30.509–542. Chierchia, G. 1995. Individual-level predicates as inherent generics. In Carlson and Pelletier 176–223. Chierchia, G. 1998. Reference to kinds across languages. Natural Language Semantics 6.339–405. Chierchia, G. 2004. Scalar implicatures, polarity phenomena and the syntax/pragmatics interface. In Structures and beyond, ed. by A. Belletti 39–103. Oxford: Oxford University Press. Chierchia, G., Partee, B.H., and Turner, R. (eds.) 1989. Properties, Types and Meaning. Dordrecht: Kluwer. Chomsky, N. 1973. Conditions on transformations. In A Festschrift for Morris Halle, ed. by S.R. Anderson and P. Kiparsky 232–286. New York: Holt, Rinehart and Winston. Chomsky, N. 1995. The Minimalist Program. Cambridge, Massachusetts: MIT Press. Chung, S., and Timberlake, A.H. 1985. Tense, aspect and mood. In Language Typology and Syntactic Description: Grammatical Categories and the Lexicon, ed. by T. Shopen 202–258. Cambridge: Cambridge University Press. Clark, E.V., and Clark, H.H. 1979. When nouns surface as verbs. Language 55.767–811.

references

159

Cohen, A. 1996. Think Generic: The Meaning and Use of Generic Sentences. Carnegie Mellon University dissertation. Published 1999, Stanford: CSLI. Cohen, A. 1999. Generics, frequency adverbs and probability. Linguistics and Philosophy 22.221–253. Cohen, A. 2001a. Existential generics. In Proceedings of the 17th Annual Conference of the Israeli Association for Theoretical Linguistics. Cohen, A. 2001b. On the generic use of indefinite singulars. Journal of Semantics 18.183– 209. Cohen, A. 2001c. Relative readings of many, often, and generics. Natural Language Semantics 9.41–67. Cohen, A. 2003. Bare plurals. Glot International 7.184–186. Cohen, A. 2004a. Existential generics. Linguistic and Philosophy 27.137–168. Cohen, A. 2004b. Generics and mental representations. Linguistic and Philosophy 27.529–556. Cohen, A. 2005. More than bare existence: An implicature of existential bare plurals. Journal of Semantics 22.389–400. Cohen, A. 2006. Bare nominals and optimal inference. In Proceedings of Sinn und Bedeutung 10, ed. by C. Ebert and C. Endriss 71–84. Cohen, A. 2007. Between kinds and properties: Bare plurals across languages. In Proceedings of the 17th Conference on Semantics and Linguistic Theory, ed. by T. Friedman and M. Gibson Ithaca, NY. Cornell University. Cohen, A. 2012. Generics as modals. Recherches linguistiques de Vincennes 41.63–82. Cohen, A., and Erteschik-Shir, N. 2002. Topic, focus, and the interpretation of bare plurals. Natural Language Semantics 10.125–165. Cohen, L.J. 1985. A problem about ambiguity in truth-theoretical semantics. Analysis 45.129–134. Cover, R. 2006. Focus on ko: the syntax, semantics, and pragmatics of identificational focus in Pulaar. ms., University of California, Berkeley. Cresswell, M.J. 1976. The semantics of degree. In Montague Grammar, ed. by B.H. Partee 261–292. New York: Academic Press. Cruse, D.A. 2004. Lexical facets and metonymy. Ilha do Desterro 47.73–96. Dahl, Ö. 1995. The marking of the episodic/generic distinction in tense-aspect systems. In Carlson and Pelletier (1995) 412–425. Danon, G. 2008. Definiteness agreement with PP modifiers. In Current Issues in Generative Hebrew Linguistics, ed. by S. Armon-Lotem, G. Danon, and S. Rothstein. Amsterdam: John Benjamins. Dayal, V. 1992. The singular-plural distinction in Hindi generics. In Proceedings of the Second Conference on Semantics and Linguistic Theory, ed. by C. Barker and D. Dowty number 40 in Working Papers in Linguistics 39–57. Ohio State University. Dayal, V. 2004a. Licensing by modification. ms., Rutgers.

160

references

Dayal, V. 2004b. Number marking and (in)definiteness in kind terms. Linguistics and Philosophy 27.393–450. Declerk, R. 1986. The manifold interpretations of generic sentences. Lingua 68.149–188. Dekker, P., and van Rooy, R. 2000. Bi-directional optimality theory: An application of game theory. Journal of Semantics 17.217–242. Delfitto, D. 2006. Bare plurals. In The Blackwell Companion to Syntax, ed. by M. Everaert and H.V. Riemsdijk volume 1, 214–259. Massachusetts: Blackwell Publishers. Delgrande, J.P. 1987. A first-order conditional logic for prototypical properties. Artificial Intelligence 33.105–130. Diesing, M. 1992. Indefinites. Cambridge, Massachusetts: MIT Press. Dobrovie-Sorin, C. 2004. Generic plural indefinites and (in)direct binding. In A Handbook of French Semantics, ed. by F. Corblin and H. de Swart 55–71. Stanford: CSLI. Dobrovie-Sorin, C., and Laca, B. 1996. Generic bare NPs. Unpublished manuscript. Dobrovie-Sorin, C., and Pires de Oliveira, R. 2004. Two ways of referring to kinds in Brazilian Portuguese. Unpublished manuscript. Dölling, J. 1995. Ontological domains, semantic sorts and systematic ambiguity. International Journal of Human-Computer Studies 43.785–807. Doron, E. 1983. Verbless Predicates in Hebrew. University of Texas, Austin dissertation. Doron, E. 2003. Bare singular reference to kinds. In Proceedings of SALT XIII. Cornell Working Papers in Linguistics. Downing, P. 1977. On the creation and use of English compound nouns. Language 53.810–842. Dowty, D. 1979. Word Meaning and Montague Grammar. Dordrecht: Kluwer. É. Kiss, K. 2002. The EPP in a topic-prominent language. In Subjects, Expletives, and the EPP, ed. by P. Svenonius 107–124. Oxford: Oxford University Press. Egg, M. 2005. Flexible Semantics for Reinterpretation Phenomena. Stanford: CSLI. Enç, M. 1986. Tense Without Scope: An Analysis of Nouns as Indexicals. University of Southern California dissertation. Erteschik-Shir, N. 1997. The Dynamics of Focus Structure. Cambridge: Cambridge University Press. Erteschik-Shir, N. 2005. Sound patterns of syntax: Object shift. Theoretical Linguistics 31.47–93. Everett, D.L. 2005. Cultural constraints on grammar and cognition in Pirahã. Current Anthropology 46. Farkas, D., and Sugioka, Y. 1983. Restrictive if/when clauses. Linguistics and Philosophy 6.225–258. Farkas, D., and de Swart, H. 2007. Article choice in plural generics. Lingua 117.1657–1676. Featherston, S., von Heusinger, K., and Weiland, H. 2011. I’m leaking oil and looking for a garage: Testing conditions on meaning transfer. In Syntax and Semantics 37: Experiments at the Interfaces, ed. by J.T. Runner 31–65. Emerald: Bingley.

references

161

Fiengo, R., and May, R. 1994. Indices and Identity. Cambridge, Massachusetts: MIT Press. von Fintel, K. 1994. Restrictions on Quantifier Domains. Amherst: University of Massachusetts dissertation. Fodor, J.D. 1977. Semantics: Theories of Meaning in Generative Grammar. New York: Crowell. Fodor, J.D., and Sag, I.A. 1982. Referential and quantificational indefinites. Linguistics and Philosophy 15.355–398. Fortescue, M. 1984. West Greenlandic. London: Croom Helm. Fox, D., and Pesetsky, D. 2005. Cyclic linearization of syntactic structure. Theoretical Linguistics 31.1–45. Geist, L. 2007. Predication and equation in copular sentences: Russian vs. English. In Existence: Semantics and Syntax, ed. by I. Comorovsky and K. von Heusinger 79–105. Dordrecht: Springer. Gerstner, C. 1979. Über generizität. Master’s thesis University of Munich. Gerstner, C., and Krifka, M. 1993. Genericity. In Handbuch der Syntax, ed. by J. Jacobs, A. von Stechow, W. Sternfeld, and Th. Vennemann 966–978. Berlin: de Gruyter. Geurts, B. 1985. Generics. Journal of Semantics 4. Geurts, B. 2001. Genericity, anaphora, and scope. Paper presented at the Workshop on Genericity, University of Cologne. Geurts, B. 2009. Scalar implicature and local pragmatics. Mind and Language 24.51–79. Geurts, B. 2010. Specific indefinites, presupposition, and scope. In Presuppositions and Discourse: Essays offered to Hans Kamp, ed. by R. Bäuerle, U. Reyle, and T.E. Zimmermann 125–158. Oxford: Elsevier. Greenberg, Y. 1994. Hebrew nominal sentences and the stage/individual level distinction. Master’s thesis Bar-Ilan University. Greenberg, Y. 1998. An overt syntactic marker for genericity in Hebrew. In Rothstein 125–143. Greenberg, Y. 2002. The manifestation of genericity in the tense aspect system of Hebrew nominal sentences. In Themes in Arabic and Hebrew Syntax, ed. by J. Ouhalla and U. Shlonsky 267–298. Dordrecht: Kluwer. Grodzinsky, Y. 1986. Language deficits and the theory of syntax. Brain and Language 27.135–159. Grodzinsky, Y. 1995. A restrictive theory of agrammatic comprehension. Brain and Language 50.27–51. Gundel, J.K. 2012. Pragmatics and information structure. In The Cambridge Handbook of Pragmatics, ed. by K. Allen and M.M. Jaszczolt 585–598. Cambridge: Cambridge University Press. Haïk, I. 1987. Bound VPs that need to be. Linguistics and Philosophy 10.503–530. Heim, I. 1982. The Semantics of Definite and Indefinite NPs. University of Massachusetts at Amherst dissertation.

162

references

Heim, I. 1990. E-type pronouns and donkey anaphora. Linguistics and Philosophy 13.137– 177. Heycock, C., and Zamparelli, R. 2003. Coordinated bare definites. Linguistic Inquiry 34.443–469. Heyer, G. 1990. Semantics and knowledge representation in the analysis of generic descriptions. Journal of Semantics 7.93–110. Holmberg, A. 1999. Remarks on Holmberg’s Generalization. Studia Linguistica 53.1– 39. Horn, L.R. 1984. Toward a new taxonomy for pragmatic inference: Q-based and R-based implicature. In Meaning, Form, and Use in Context: Linguistic Applications, ed. by D. Schiffrin 11–42. Washington: Georgetown University Press. Junghans, U. 1997. On the so-called eto-cleft construction. In Annual Workshop on Slavic Linguistics, ed. by M. Lindseth and S. Franks 166–190. Ann Arbor: Michigan Slavic Publications. Kamp, H., and Reyle, U. 1993. From Discourse to Logic. Dordrecht: Kluwer Academic Publishers. Kennedy, C. 2007. Vagueness and grammar: The semantics of relative and absolute gradable adjectives. Linguistics and Philosophy 30.1–45. Kitagawa, Y. 1991. Copying identity. Natural Language and Linguistic Theory 9.497–536. Klein, W. 1992. The present perfect puzzle. Language 68.525–552. Klein, W. 1994. Time in Language. London: Routledge. Kratzer, A. 1995. Stage-level and individual-level predicates. In Carlson and Pelletier 125–175. Krifka, M. 1989. Nominal reference, temporal constitution, and quantification in event semantics. In Semantics and Contextual Expression, ed. by R. Bartsch, J. van Benthem, and P. van Emde Boas 75–115. Dordrecht: Foris. Krifka, M. 1995. Focus and the interpretation of generic sentences. In Carlson and Pelletier (1995) 238–264. Krifka, M. 1999. At least some determiners aren’t determiners. In The Semantics/Pragmatics Interface from Different Points of View, ed. by K. Turner 257–291. Amsterdam: Elsevier. Krifka, M. 2001. Quantification into question acts. Natural Language Semantics 9.1– 40. Krifka, M. 2004. Bare NPs: Kind-referring, indefinites, both, or neither? In Proceedings of the 13th Conference on Semantics and Linguistic Theory, ed. by R.B̃ . Young and Y. Zhou. Krifka, M. 2008. Functional similarities between bimanual coordination and topic/ comment structure. In Variation, selection, development: Probing the evolutionary model of language change, ed. by R. Eckardt, G. Jäger, and T. Veenstra 307–336. Berlin: Mouton de Gruyter.

references

163

Krifka, M. 2009. Definitional generics as second-order predications. Paper presented at the workshop Genericity: Interpretation and Uses, Paris. Krifka, M., Pelletier, F.J., Carlson, G., ter Meulen, A., Link, G., and Chierchia, G. 1995. Genericity: an introduction. In Carlson and Pelletier 1–124. Landman, F. 2000. Events and Plurality. Dordrecht: Kluwer. Landman, F. 2008. 1066: On the differences between the tense-perspective-aspect systems of English and Dutch. In Theoretical and Crosslinguistic Approaches to the Semantics of Aspect, ed. by S. Rothstein. Amsterdam: John Benjamins. Lawler, J. 1973. Studies in English generics. University of Michigan Papers in Linguistics 1:1. Leech, G. 1971. Meaning and the English Verb. London: Longman. Leslie, S-J. 2007a. Generics and the structure of the mind. Philosophical Perspectives 21.375–403. Leslie, S.J. 2007b. Generics, Cognition and Comprehension. Princeton University dissertation. Lewis, D. 1996. Elusive knowledge. Australasian Journal of Philosophy 74: Papers in Metaphysics and Epistemology. 549–567. Liebesman, D. 2011. Simple generics. Noûs 45.409–442. Liebesman, D., and Magidor, O. 2017. Copredication and property inheritance. Philosophical Issues 27.131–166. Link, G. 1983. The logical analysis of plural and mass terms: A lattice-theoretical approach. In Meaning, Use and Interpretation of Language, ed. by R. Bäuerle, C. Schwartze, and A. von Stechow 302–323. Berlin: de Gruyter. Link, G. 1987. Algebraic semantics for event structures. In Proceedings of the Sixth Amsterdam Colloquium, ed. by J. Gronendijk et al. 243–262 Amsterdam. ITLI. Longobardi, G. 2001. How comparative is semantics? A unified parametric theory of bare nouns and proper names. Natural Language Semantics 9.335–369. MacDonald, J.E. 2006. The Syntax of Inner Aspect. Stony Brook University dissertation. Markman, V.G. 2008. Pronominal copula constructions are what? reduced specificational pseudo-clefts! In Proceedings of the 26th West Coast Conference on Formal Linguistics—WCCFL XXVI, ed. by C.B. Chang and H.J. Haynie 366–374 Somerville, MA. Cascadilla Proceedings Project. McCawley, J.D. 1976. Grammar and Meaning. New York: Academic Press. McCawley, J.D. 1978. Conversational implicature and the lexicon. In Syntax and Semantics 9: Pragmatics, ed. by P. Cole 245–259. New York: Academic Press. McNally, L. 1998. Stativity and theticity. In Rothstein 293–307. McNally, L. 2004. Bare plurals in Spanish are interpreted as properties. Catalan Journal of Linguistics 3.115–133. Milsark, G. 1974. Existential Sentences in English. Massachusetts Institute of Technology dissertation.

164

references

Moltmann, F. 1991. Measure adverbials. Linguistics and Philosophy 14.629–660. Moltmann, F. 1995. Exception sentences and polyadic quantification. Linguistics and Philosophy 18.223–280. Montague, R. 1970. English as a formal language. In Linguaggi nella Società e nella Tecnica, ed. by B. Visentini et al 189–224. Milan: Edizioni di Comunità. Reprinted in Thomason (1974), 188–221. Montague, R. 1973. The proper treatment of quantification in ordinary English. In Approaches to Natural Language, ed. by J. Hintikka, J. Moravcsik, and P. Suppes. Dordrecht: Reidel. Morreau, M.P. 1992. Conditionals in Philosophy and Artificial Intelligence. University of Amsterdam dissertation. Mufwene, S.S. 1984. Stativity and the Progressive. Bloomington: Indiana University Linguistics Club. Müller, A. 2001. Genericity and the denotation of common nouns in Brazilian Portuguese. In Proceedings of SULA (Semantics of Under-Represented Languages in the Americas), ed. by A. Weerle and J.Y. Im 72–80 Amherst. GLSA, University of Massachusetts. Müller, A. 2003. Generic sentences with indefinite and bare subjects in Brazilian Portuguese. In Proceedings of SULA 2, ed. by J. Anderssen, P. Menéndez-Benito, and A. Werle 71–86 Amherst. GLSA, University of Massachusetts. Müller-Reichau, O. 2006. Sorting the World: On the Relevance of the Kind-Level/ObjectLevel distinction to Referential Semantics. University of Leipzig dissertation. Munn, A., and Schmitt, C. 2005. Number and indefinites. Lingua 115.821–855. Neale S. 1990. Descriptions. Cambridge, MA: MIT Press. Nehaniv, C.L. 2000. The making of meaning in societies: Semiotic & informationtheoretic background to the evolution of communication. In Starting from Society 73–84. Society for the Study of Artificial Intelligence and Adaptive Behaviour. Nehaniv, C.L. 2005. Open problems in the emergence of evolution of linguistic communication: A road map for research. In Proceedings of the Second International Symposium on the Emergence and Evolution of Linguistic Communication 86–93. Nguyen, A. 2019. The radical account of bare plural generics. Philosophical Studies. Nickel, B. 2012. Dutchmen are good sailors: Generics and gradability. In Genericity, ed. by A. Mari, C. Beyssade, and F.D. Prete 390–405. Oxford: Oxford University Press. Nickel, B. 2016. Between Logic and the World: An Interpreted Theory of Generics. Oxford: Oxford University Press. Nunberg, G. 1977. The Pragmatics of Reference. CUNY, New York dissertation. Reprinted by Indiana University Linguistics Club, 1978. Nunberg, G. 1995. Transfers of meaning. Journal of Semantics 12.109–132. Nunberg, G. 2004. The pragmatics of deferred interpretation. In The Handbook of Pragmatics, ed. by L. Horn and G. Ward 344–364. Oxford: Blackwell.

references

165

Osherson, D.N., Smith, E.E., Wilkie, O., López, A., and Shafir, E. 1990. Category-based induction. Psychological Review 97.185–200. Palma, H.L. 2007. Plural indefinite descriptions with unos and the interpretation of number. Probus 19.235–266. Partee, B.H. 1984. Nominal and temporal anaphora. Linguistics and Philosophy 7.243– 286. Partee, B.H. 1987. Noun phrase interpretation and type-shifting principles. In Studies in Discourse Representation Theory and the Theory of Generalized Quantifiers, ed. by J. Groenendijk, D. de Jongh, and M. Stokhof 115–143. Dordrecht: Foris. Partee, B.H. 1993. Semantic structures and semantic properties. In Knowledge and Language, ed. by E. Reuland and W. Abraham volume II: Lexical and Conceptual Structure 7–29. Dordrecht: Kluwer. Partee, B.H., and Rooth, M. 1983. Generalized conjunction and type ambiguity. In Meaning, Use, and interpretation of Language, ed. by R. Bäuerle, C. Schwarze, and A. von Stechow 361–383. Berlin: de Gruyter. Pelletier, F.J., and Asher, N. 1997. Generics and defaults. In Handbook of Logic and Language, ed. by J. van Benthem and A. ter Meulen 1125–1177. Amsterdam: Elsevier. Pelletier, F.J., and Schubert, L.K. 1989. Mass expressions. In Handbook of Philosophical Logic, ed. by D. Gabbay and F. Günthner 327–407. Dordrecht: Reidel. Portner, P., and Yabushita, K. 1998. The semantics and pragmatics of topic phrases. Linguistics and Philosophy 21.117–157. Pullum, G.K. 1989. The great Eskimo vocabulary hoax. Natural Language and Linguistic Theory 7.275–281. Pulman, S.G. 1997. Aspectual shift as type coercion. Transactions of the Philological Society 95.279–317. Recanati, F. 1996. Domains of discourse. Linguistics and Philosophy 19.445–475. Recanati, F. 2003. Embedded implicatures. Philosophical Perspectives 7.299–332. Reichenbach, H. 1947. Elements of Symbolic Logic. Berkeley: University of California Press. Reinhart, T. 1981. Pragmatics and linguistics: An analysis of sentence topics. Philosophica 27.53–94. Rimell, L. 2004. Habitual sentences and generic quantification. In Chand et al. Robinson, H. 2005. Unexpected (In)definiteness: Romance Plural Determiners in Generic Contexts. Rutgers University dissertation. Rochemont, M. 2019. Topics and givenness. In Architecture of Topic, ed. by V. Molnár, V. Egerland, and S. Winkler 47–66. Berlin: de Gruyter. Rooth, M.E. 1985. Association with Focus. University of Massachusetts at Amherst dissertation. van Rooy, R. 2004. Signalling games select Horn strategies. Linguistics and Philosophy 27.493–527.

166

references

Rot, S. 1987. On aspect-tense-temporal reference in present-day English. Acta Linguistica 37.243–286. Rothstein, S. (ed.) 1998. Events and Grammar. Dordrecht: Kluwer. Russell, B. 1905. On denoting. Mind 14.479–493. Sag, I.A. 1981. Formal semantics and extralinguistic context. In Radical Pragmatics, ed. by P. Cole 273–294. New York: Academic Press. Schmitt, C., and Munn, A. 1999. Against the nominal mapping parameter: Bare nouns in Brazilian Portuguese. In Proceedings of NELS 29, ed. by P. Tamanji, M. Hirotani, and N. Hall 339–353. Schubert, L.K., and Pelletier, F.J. 1987. Problems in the representation of the logical form of generics, plurals, and mass nouns. In New Directions in Semantics, ed. by E. LePore 385–451. London: Academic Press. Seuren, P.A.M. 1973. The comparative. In Generative Grammar in Europe, ed. by F. Kiefer and N. Ruwet 528–564. Dordrecht: Kluwer. Sharvy, R. 1980. A more general theory of definite descriptions. The Philosophical Review 89.607–624. Slabakova, R., Amaro, J.C., and Kang, S.K. 2013. Regular and novel metonymy in native Korean, Spanish, and English: Experimental evidence for various acceptability. Metaphor and Symbol 28.275–293. Smith, C. 1991. The Parameter of Aspect. Dordrecht: Kluwer. Stalnaker, R.C. 1974. Pragmatic presupposition. In Semantics and Philosophy, ed. by M.K. Munitz and D.K. Unger 197–213. New York: New York University Press. Stalnaker, R.C. 1998. Belief attribution and context. In Contents of Thought, ed. by R. Grimm and D. Merrill 140–156. Tuscon: University of Arizona Press. Stalnaker, R.C. 2002. Common ground. Linguistics and Philosophy 25.701–721. Stanley, J., and Szábo, Z. 2000. On quantifier domain restriction. Mind & Language 15.219–251. von Stechow, A. 2002. German seit ‘since’ and the ambiguity of the German perfect. In More than Words: A Festschrift for Dieter Wunderlich, ed. by B. Stiebels and I. Kaufmann 393–432. Berlin: Akademie Verlag. Sterken, R. 2015a. Generics in context. Philosopher’s Imprint 15.1–30. Sterken, R. 2015b. Leslie on generics. Philosophical Studies 172.2493–2512. Sterken, R. 2016. Generics, covert structure and logical form. Mind & Language 31.503– 529. de Swart, H. 1991. Adverbs of Quantification: A Generalized Quantifier Approach. Groningen University dissertation. Also published 1993, New York: Garland. de Swart, H. 1996. (In)definites and genericity. In Quantifiers, Deduction and Context, ed. by M. Kanazawa, C. Piñon, and H. de Swart. Stanford: Center for the Study of Language and Information. de Swart, H. 1998. Aspect shift and coercion. Natural Language and Linguistic Theory 16.347–385.

references

167

Szabolcsi, A. 1997. Strategies for scope taking. In Ways of Scope Taking, ed. by A. Szabolcsi 109–154. Dordrecht: Kluwer. Taylor, B. 1977. Tense and continuity. Linguistics and Philosophy 1.199–220. Tessler, M., and Goodman, N.D. 2019. The language of generalization. Psychological Review 126.395–436. Thomason, R. (ed.) 1974. Formal Philosophy: Selected Papers of Richard Montague. New Haven: Yale University Press. Van Geenhoven, V. 1996. Semantic Incorporation and Indefinite Descriptions. University of Tübingen dissertation. Published by CSLI, 1998. Van Geenhoven, V. 2004. For-adverbials, frequentative aspect, and pluractionality. Natural Language Semantics 12.135–190. Vlach, F. 1981. semantics of the progressive. In Syntax and Semantics 14: Tense and Aspect, ed. by P. Tedeschi and A. Zaenen 271–292. New York: Academic Press. Wasow, T. 1972. Anaphoric Relations in English. Cambridge, MA: MIT dissertation. Wilkinson, K. 1991. Studies in the Semantics of Generic Noun Phrases. University of Massachusetts at Amherst dissertation. Williams, E. 1977. Discourse and logical form. Linguistic Inquiry 8.101–139. Zimmermann, T.E. 1993. On the proper treatment of opacity in certain verbs. Natural Language Semantics 1.149–179. Zuber, R. 1998. On the semantics of exclusion and inclusion phrases. In Proceedings of the 8th Conference on Semantics and Linguistic Theory, ed. by A. Lawson 267–283 Ithaca, NY. Cornell University Press. Zuber, R. 2004. Boolean semantics and categorial polyvalency. In Chand et al. Zucchi, S., and White, M. 2001. Twigs, sequences and the temporal constitution of predicates. Linguistics and Philosophy 24.223–270.

Name index Abney, S. 19n Anagnostopoulou, E. 58 Aristotle 25n Asher, N. 27, 146 Atlas, J. 147 Bach, K. 28n14 Bennett, M. 133 Bhatt, R. 58 Bittner, M. 141 Błaszczak, J. 98 Blok, P.I. 73n Blutner, R. 148 Bobaljik, J.D. 115n16 Boneh, N., 118n18, 129n31 Bošković, Ž. 33 Brinton, L.J. 128n27 Burton-Roberts, N. 52, 55, 67 Butt, M. 58 Carlier, A. 84 Carlson, G. 2, 33–39, 49, 56, 60, 100–101, 105–106, 120–121, 127, 139n Carston, R. 28n14 Cheng, L. 95 Chierchia, G. 2, 16, 28, 39–47, 50n21, 51, 55– 59, 62, 64–70, 77–79, 86–87, 92–93, 97–98, 118 Chomsky, N. 1, 113 Chung, S., 127 Clark, E.V. 22 Clark, H.H. 22 Cohen, A. 34n, 39n9, 43n, 50, 53, 58, 65n1, 74, 80, 103, 118–119, 120–124, 129n30, 146, 148–152, 155 Cohen, L.J. 25n Cover, R. 98 Cresswell, M.J. 12 Cruse, D.A. 21 Dahl, Ö. 118 Danon, G. 89 Dayal, V. 58n, 65n3, 66–68, 71–72, 74, 76, 79–80, 90–91, 93, 95 Declerk, R. 149n Deo, A, 94, 97

Dekker, P., 148 Delfitto, D. 36n4 Delgrande, J.P. 146 Diesing, M. 36, 102 Dobrovie-Sorin, C. 45–46, 59, 93 Dölling, J. 22–23, 124–125 Doron, E. 59, 95, 98, 118n18, 129n31 Downing, P. 22 Dowty, D. 127, 129, 135, 142 É. Kiss, K. 86 Eberle, K. 73n Egg, M. 4–5, 19, 137, 143 Enç, M. 79 Erteschik-Shir, N. 48n17, 50, 58, 115n16 Everett, D.L. 119 Farkas, D., 76, 85 Featherston, S. 22 Fiengo, R., 114n von Fintel, K. 98 Fodor, J.D. 37n, 57 Fortescue, M. 141 Fox, D., 115n16 Geist, L. 98 Gerstner, C. 36, 43 Geurts, B. 28, 57, 72, 149n Goodman, N.D. 39n9 Greenberg, Y. 97–98, 155 Grodzinsky, Y. 115 Gundel, J.K. 48 Gyuris, B. 85 Haïk, I. 114n Heim, I. 15, 48, 98, 145 Heycock, C. 92 Heyer, G. 62, 68n4 Holmberg, A. 114–115 Horn, L.R. 147 Junghans, U. 98 Kamp, H. 108, 133, 137 Krifka, M. 2, 36, 39–48, 54n, 58–59, 62, 64– 68, 70, 74, 78, 87, 91, 95, 97n, 101, 104n,

169

name index 106–107, 109, 133n, 135–136, 143, 146, 147n Laca, B. 59 Landman, F. 28, 134 Lawler, J. 52 Leech, G. 127 Leslie, S.-J. 39n9, 47, 112n, 118n19 Levinson, S.C. 147 Lewis, D. 117 Liebesman, D. 39n8, 47n, 124n25 Link, G. 130, 146 Longobardi, G. 65, 78, 80–82, 85 MacDonald, J.E. 139 Markman, V.G. 98 May, R. 114n McCawley, J.D. 121, 148 McNally, L. 45, 59 Milsark, G. 37 Moltmann, F. 135, 147n Montague, R. 7–8, 26n, 60, 69 Morreau, M.P. 146 Mufwene, S.S. 127 Müller, A. 87–88, 93n Müller-Reichau, O. 72n Munn, A., 87–89, 92, 96 Neale, S., 116 Nehaniv, C.L. 48n16 Nguyen, A. 39n9 Nickel, B. 39n9, 47, 110n7, 112n, 118n19, 145– 146, 152n5 Nunberg, G. 4, 19–27, 30, 124

Reinhart, T. 48 Reyle, U. 133, 137 Rimell, L. 111n8 Robinson, H. 65n3 Rochemont, M. 49n18 Rooth, M.E. 5–9, 104 van Rooy, R. 148 Rot, S. 127 Russell, B. 13 Sag, I.A. 19, 22, 57 Schmitt, C. 87–89, 92, 96 Schubert, L.K. 17, 62–63, 106, 125 Seuren, P.A.M. 1, 12 Sharvy, R. 66 Slabakova, R. 22 Smith, C. 127 Stalnaker, R.C. 97 Stanley, J. 117, 121 von Stechow, A. 127 Sterken, R. 39n9, 47, 112n, 118n19, 119n, 120– 121, 123n24 Sugioka, Y. 76n de Swart, H. 50n21, 76, 84–85, 137, 143 Sybesma, R. 95 Szábo, Z. 117, 121 Szabolcsi, A. 61, 133 Taylor, B. 127, 129 Tessler, M., 39n9 Timberlake, A.H. 127 Van Geenhoven, V. 59, 80, 86n, 110, 136, 140–143 Vlach, F. 128

Osherson, D.N. 152–153 Palma, H.L. 45 Partee, B.H. 5–9, 15, 17–18, 40, 74, 127, 133 Pelletier, F.J., 17, 62–63, 106, 125, 146 Pesetsky, D. 115n16 Portner, P. 48n16 Pires de Oliveira, R. 93 Pullum, G.K. 73 Pulman, S.G. 137–138, 143 Rapoport, T. 149 Recanati, F. 25, 28, 116 Reichenbach, H. 129

Wasow, T. 114n White, M. 135, 140 Wilkinson, K. 36 Williams, E. 67, 114n Yabushita, K. 48n16 Yeats, W.B. vi Zamparelli, R. 92 Zeevat, H. 151 Zimmermann, T.E. 60, 62 Zuber, R. 147 Zucchi, S. 11n5, 135, 140

Subject Index a/an 16, 44–45, 51, 55 absolutely 146 accessible worlds 97 accommodation 107, 129, 134n33 action triggers 55, 82, 85 adjectives gradable 12–15 adverbs of quantification, see quantification, adverbial adverbials, see durative adverbials almost 146 anaphora 36n4 maximal 61 bare plurals 2, 16n7, 33–63, 66–69, 75–76, 87, 98–99, 102n, 123, 133n, 139n, 148–155 and habituals 131–133 as both kinds and properties 58–61, 98– 99 as cumulative 139 as kinds 33–39, 42, 43, 51–52, 56–58 as properties 41–44, 46–50 characterizing reading of 33, 36–39, 42– 43, 47–49, 51–52, 58–59, 62–63, 98–99, 148–154 definite reading of 58n, 91 direct kind predication 33, 42, 51, 101, 120 existential reading of 34–36, 43–46, 49– 50, 51, 56–58, 58–60, 62–63, 98–99, 149–150, 154–155 generic reading of, see bare plurals, characterizing reading of heaviness requirement 79–80 in Hungarian 85–86 in Italian 64–66, 76–85, 98 in Russian 92 non-quantificational theories 33– 39 normative reading of 83–84, 98 quantificational theories 39–63 specific interpretation of 57–58 bare singulars 2, 16, 51–52, 86–98 and type-shifting 86–87 characterizing reading of 96–98 existential reading of 87–92

reference to kinds 92–96 uniqueness presupposition of 91–92, 97 Bidirectional Optimality Theory 148 blocking 16–18, 44–46, 51 BPs, see bare plurals Brazilian Portuguese 87–89, 93, 96, 98 choice functions 44–45, 87 coordinates 43n de dicto 6, 11n5, 30–31, 35, 60–62, 77, 89 de re 6, 10, 11n5, 30–31, 35, 60, 89 definites 2, 65–77 as maximizing operators 66 as kind-referring 71–75 generic interpretation of 66–70 in Italian 65–66, 76–77 plural 65–66, 75–76, 86 singular 66–75, 76n, 77n uniqueness presupposition of 66, 91 definitions 54–55 dei 45 Derived Kind Predication 43–44, 47, 56, 66, 69–70, 78 des 45–46, 84 differentiated scope 35, 77, 108, 139n discourse referents 15, 48, 60–62, 131–133 Discourse Representation Theory 132–133, 137–138 distributivity 110n7, 143, 145–146 DKP, see Derived Kind Predication DRT, see Discourse Representation Theory durative adverbials 3, 135–144 Dutch 80, 133–134 E-type pronouns 61 ellipsis 114n, 142 embedded implicature 27–28 empty categories, see null categories, in syntax EPP 86 even 147 eventive verbs 127–131, 133 eventualities homogeneous 135–138, 151n non-homogeneous 135–137

171

subject index existential quantifier, see quantification, existential experimental evidence 22, 152–154 explicature 28n14 focus 103–105 for-adverbials, see durative adverbials French 45–46, 84 Full Interpretation principle 113 Game Theory 148 gen, see quantification, generic generic quantifier, see quantification, generic generics 2, 120–126, 128, 145–147, 148 see also quantification, generic and scope 100–106, 125–126 homogeneity presupposition of, see homogeneity German 68, 73 habituals 2, 118, 127–134 and bare plurals 131–133 and scope 106–108 as stative 127–128 Hebrew 87, 89–90, 94–97, 98 Hindi 58, 86, 87, 90–91, 93–94, 95, 97, 98 Holmberg’s Generalization 114–115 homogeneity 151–156 Hungarian 61n, 85–86, 98 imperfective 129 implicit quantification, see quantification, implicit impliciture 28n14 indefinite singulars 52–55, 64–65, 67, 80– 82, 83, 88, 97, 119 individual concepts 70, 71 individual-level predicates, see predicates, individual-level inference 3, 15, 147–156 int 129–133 intensionality and kinds 41n, 80, 146 and predicate transfer 23–25, 30–31 and type-shifting 6–10, 14 operator 69–70, 76 intervals 127, 129–132, 135, 142 ISs, see indefinite singulars

Italian bare partitives in 45 bare plurals in 64–66, 76–85, 98 definites in 65–66, 76–77 iterativity 4, 135–144 and scope 138–140 and type-shifting 137 as a quantifier 110n6, 142–144 kind-level predicates, see predicates, kindlevel kinds 56–57, 120 as individual concepts 40 as referential 49 vs. groups 69–70 ko 58 les 84 lexical decomposition 37 meaning postulates 36n5 modality 97, 145–146, 156n modals 53, 55, 81 moments 127–131, 133 negation 11–12, 60, 83, 89–90, 100–105 noteworthiness 21–24, 27, 30 null categories in semantics 116–117 in syntax 113–116 null quantifier, see quantification, implicit number 20, 109–110, 133n neutrality 42–44, 87–88, 93, 133n object shift 114–115 opacity 30–31, 35, 60–62, 77, 88–89, 105– 106, 108, 110, 126 Optimality Theory 76–77 bidirectional, see Bidirectional Optimality Theory PF 113–114 Pirahã 119 predicate transfer 2, 18–32, 100, 120, 123– 128, 145 and intensionality 23–25 and scope 29–32 predicates collective 43

172 cumulative 135–136, 139 homogeneous, see predicates, cumulative individual-level 34, 38, 50, 86 kind-level 34, 54–58, 64–65, 69–70, 72, 93–94, 121–122, 124 non-homogeneous, see predicates, quantized object-level 121–122 quantized 135–136, 139, 143 stage-level 34, 38, 49–50, 86 probability 65n1, 151–153 progressive 128n28, 133–134 Q-adverbs, see quantification, adverbial quantification adverbial 44–46, 53, 55, 81, 87, 98, 101– 102, 105–106, 107, 109, 118–119, 131, 141–143, 146–147 domain of 36–37, 42–43, 45–46, 103– 104, 116–117 existential 15, 22–23, 145 explicit 33–35, 36–37, 44, 53, 55, 60n29, 81, 87–88, 91, 101–102, 105–107, 109, 112, 118–119, 131, 141, 147 generic 38–39, 42–43, 47, 49, 53, 55, 59, 112, 118–121, 123–125, 128, 145, 156 see also generics implicit 1–3, 23, 33, 37, 39, 47, 55, 100, 110, 112, 118–120, 145–156 null, see quantification, implicit overt, see quantification, explicit universal 145–147, 156 vacuous 50 quantificational variability 37–38 quantifying in 6, 10–11 regulations, see rules and regulations reinterpretation 1–32, 112–117, 119, 156 as a universal principle 63 rules and regulations 54–55, 83–85, 98 Russian 16, 44, 51, 86–87, 90–91, 92, 95, 98 scope and bare singulars 86, 88–91 and conjunction 7–8 and existential bare plurals 34–35, 43– 44, 56–58, 60, 77–78, 84, 131–133

subject index and generic bare plurals 37 and generics 100–106, 108–112, 119–120, 125–126 and gradable adjectives 13–15 and habituals 106–112, 119–120, 131– 133 and indefinite singulars 54 and iterativity 138–144 and normative bare plurals 83 and predicate transfer 29–31 and sort-shifting 18 and type-shifting 10–12, 40 differentiated, see differentiated scope selectional restrictions 121–123 semantics-pragmatics interface 27 seu 87–88 situations and worlds 41n, 80 minimal 98, 123n23 some 44–45 sort-shifting 17–18, 137n1 Spanish 45 stage-level predicates, see predicates, stagelevel stative verbs 127–131 stereotypicality 148–151, 154–156 sums of individuals 40–41, 80 of situations 109 Swedish 114–115 𝒯 22–27, 29–31, 124–125 𝒯𝒢 125–126 -tar- 140–142 taxonomic interpretations 53n, 64–65, 71–75, 80, 95–96, 110n5, 123 TNP, see Traditional Noun Phrase topics 48–50, 53–54, 59, 98, 123 Trace Deletion Hypothesis 115 trace theory 1, 113–115 Traditional Noun Phrase 33, 49, 60, 61n, 109, 133 transfer function, see 𝒯, 𝒯𝒢 type-shifting 2, 5–18, 21, 39–51, 56–61, 66– 77, 83, 84, 86–88, 91–93, 95–98, 100, 120–121, 123, 126, 128, 130–132, 134, 135, 137–140, 142, 144, 145 and scope 10–12, 18, 40 as a last resort 10–11, 16, 18, 46

173

subject index as a universal principle 17, 40 Ceh 137–138 g 69 MASS 69–70, 93 ∃ 40, 43, 47, 56–57, 59–60, 77, 83, 86, 91, 96 ι 40, 86, 91, 96 ∩ 40–42, 46, 49, 56, 59, 76, 86, 92–93, 97–98, 123



γ

40–41, 43–44, 49, 70, 123, 125 130–131

unmarkedness 147–148, 150, 155–156 unos/unas 45 wanna contraction 113 West Greenlandic 140–142 when-clause 108, 112, 128