Beyond 'Any' and 'Ever': New Explorations in Negative Polarity Sensitivity 9783110305234, 9783110303728

Citation preview

Eva Csipak, Regine Eckardt, Mingya Liu and Manfred Sailer (Eds.) Beyond ‘Any’ and ‘Ever’

Trends in Linguistics Studies and Monographs

Editor Volker Gast Editorial Board Walter Bisang Hans Henrich Hock Natalia Levshina Heiko Narrog Matthias Schlesewsky Niina Ning Zhang Editor responsible for this volume Heiko Narrog

Volume 262

Beyond ‘Any’ and ‘Ever’ New Explorations in Negative Polarity Sensitivity

Edited by Eva Csipak Regine Eckardt Mingya Liu Manfred Sailer

ISBN 978-3-11-030372-8 e-ISBN 978-3-11-030523-4 ISSN 1861-4302 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliografische Information der Deutschen Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutschen Nationalbibliografie; detailed bibliographic data are available in the internet http://dnb.dnb.de. © 2013 Walter de Gruyter GmbH, Berlin/Boston Typesetting: PTP-Berlin Protago-TEX-Production GmbH, Berlin Printing: Hubert & Co. GmbH & Co. KG, Göttingen ♾ Printed on acid-free paper Printed in Germany www.degruyter.com

Contents Acknowledgements 

 1

Regine Eckardt and Manfred Sailer  3 Beyond “any” and “ever”  Johan van der Auwera & Lauren Van Alsenoy  21 Mapping the West Germanic any’s  Jack Hoeksema Polarity items in Strawsonian contexts – A comparison 

 47

Volker Gast From æghwæðer to either: The distribution of a negative polarity item in historical perspective  Johan Brandtler Evaluability – An Alternative Approach to Polarity Sensitivity  Luka Crnič How to get even with desires and imperatives  I-Ta Chris Hsieh On NPI Licensing in Possibility Conditionals 

 103

 127

 155

Jon Gajewski An analogy between a connected exceptive phrase and polarity items  Elena Herburger and Simon Mauck  213 The chance of being an NPI  Robert Levine The modal need VP gap (non)anomaly 

 241

Regine Eckardt and Eva Csipak Minimizers – Towards pragmatic licensing 

 79

 267

 183

vi   

   Contents

Ai Matsui Revisiting the Licensing Problem through Understating NPIs: The Case of  299 Japanese Anmari ‘(not) very/much’  Edgar Onea and Manfred Sailer  323 Really all that clear?  Mingya Liu, Regine Eckardt, and Janina Radό  351 Polarity in Context  Subject index 

 369

Acknowledgements This volume resulted from a workshop “Beyond Any and Ever: Challenging Theories of NPI Licensing” which took place in Göttingen in 2011. The selection of papers reflects our experiences and discussions on the occasion: Most of us started with some preconception of what NPIs are, and what a good licensing theory might look like. When thinking about other people’s favourite data and examples, however, we all got a feeling that some eventual theory of NPI licensing would need to reconcile a larger number of incommensurable proposals, each equally lucid, insightful and well-motivated in its own realm of data. Finding the unifying general principle behind this mulitplicity of patterns was, and remains, the challenge in the field. The papers in this volume reflect this variety of strategies, principles and mechanisms that seem necessary, and will hopefully spur future research towards the optimal eventual theory. We would like to express our warm thanks to Ilaria Frana, Jon Gajewski, Remus Gergel, Anastasia Giannakidou, Yurie Hara, Jack Hoeksema, Gianina Iordachioaia, Agnes Jäger, Larry Horn, Michael Israel, Magda Kaufmann, Stefan Kaufmann, Utpal Lahiri, Robert D. Levine, Eric McCready, Ira Noveck, Doris Penka, Frank Richter, Roland Schäfer, Anna Szabolcsi and Thomas Weskott, who took part in the review process and offered critical and stimulating comments. We are also grateful to Birgit Sievert and Julie Miess from De Gruyter Mouton and to the series editor, Heiko Narrog, for their encouragement and support in the preparation of the final manuscript. The book emerged as part of a research project funded by the Deutsche Forschungsgemeinschaft, which we gratefully acknowledge.

Regine Eckardt and Manfred Sailer

Beyond “any” and “ever” 1 The problem Negative polarity items (NPIs) are restricted in use to contexts which can be superficially characterized as “negative”. Words like any and ever are acceptable in sentences like (1) but ungrammatical in (2). (1)

John didn’t buy any bagel (ever).

(2)

a. * John ever bought a bagel. b. * John was buying any bagel.

At least since Klima 1964 it is clear that while negation is the classical licensing context for NPIs, NPIs may also occur in other contexts which can be considered more or less “negative.” Some of these contexts are illustrated in (3) (the putative licensors are given in boldface). (3)

a. Few students have ever attended this class. b. Every student who has ever attended this class, will pass the test. c. Only Pat has any objection against the proposal. d. If Pat has any objection against the proposal, we should brainstorm again.

The exact characterization of the licensing contexts as well as the question of why an expression should show such a sensitivity to a particular context – or to a particular polarity – have been a continuing challenge for syntax, semantics, and pragmatics. In this introduction we will sketch the types of theories that have been proposed in formal linguistics since Klima 1964 and situate the contributions of this volume with respect to these theories. We highlight two underlying questions in NPI research: the problem of defining what an NPI is once we go beyond any and ever, as well as its dual problem: permissible contexts of use for any/ever which are not expected in mainstream theory. We outline below how authors in this volume address these questions.

4   

   Regine Eckardt and Manfred Sailer

2 Generations of theories Theories of NPI licensing since Klima 1964 not only differ in detail but also take different perspectives on the phenomenon. At the beginning, we see what could be called monolithic theories. They attempt to define some syntactic or semantic property of contexts which host NPIs without ever asking why an item needs the shelter of a syntactic licensor, or why an item has a liking for certain logical properties of contexts. The first monolithic approach is Klima 1964. Klima assumes a feature [affective+]. NPIs are required to stand in a particular syntactic constellation with an element that bears this feature, an early version of c-command in Klima’s case. Since the set of NPI licensing contexts looked too disparate to give an independent semantic characterization, a syntactic feature seemed to be the appropriate solution. Later syntactic approaches tried to reduce the feature [affective+] to a syntactic negation feature and assume variants of Klima’s c-command condition. Linebarger 1980 proposes that NPIs must be c-commanded by a syntactic NEG node at LF without there being any intervening operator. Progovac 1993 proposes a binding-theoretic account of NPIs: NPIs must be bound by a syntactic negation node. Depending on whether an NPI is more like a reflexive pronoun or a nonreflexive pronoun, there could be more or less material between the negation and the NPI. While the syntactic theories account successfully for various configurational aspects of the NPI licensing, they have notorious problems with licensers for which a NEG feature seems implausible. Going back to Klima’s original arbitrary feature [affective+] does not seem to be a welcome alternative, though. Ladusaw 1979 proposes a logico-semantic interpretation of Klima’s feature: He shows that the core contexts of NPI licensing can be characterized as being downward entailing/monotone decreasing. This means that they allow for inferences from supersets to subsets, a test which has become classic. Among our authors, Hoeksema, Herburger/Mauck, and Levine offer an introduction to Ladusaw’s account. While Ladusaw 1979 simply intended to give a semantic interpretation to Klima’s syntactic feature, later incarnations of the logico-semantic theories attempted to replace the syntactic definition of the licensing relation with the semantic notion of scope altogether. Zwarts 1995, 1998 provides a refinement of the logical characterization of the licening condition. In addition to downward monotonicity, Zwarts introduces the entailment notions of antimorphicity, antiadditivity, and non-veridicality as being relevant for NPI licensing. His typology of licensing criteria is introduced in detail in the contribution by Hoeksema, while van der Wouden 1997 and Giannakidou 1998 offer detailed empirical arguments in favour of the resulting classes of NPIs. In later work, Giannakidou argues that

Beyond “any” and “ever”   

   5

licensing should not only be semantically coded, but also semantically motivated and hence moves more towards what we will call “conspiracy theories”. Finally, the collocational approach presented in Richter and Soehn 2006 and Sailer 2009 is another variant of monolithic licensing. The proponents of this approach claim that NPI licensing is just a special case of collocational restrictions, coded by the presence of a particular semantic operator in the semantic representation of a sentence. The approach is monolithic because there is a separate module of the grammar that takes care of collocational requirements, even though it operates at the logical form of a sentence. The second group of theories consists of what we call conspiracy theories. They assume that NPI-hood is not a primitive property but follows from the interaction of pragmatic properties of an NPI and general properties of the licensing contexts. The most prominent representative of this approach is Krifka 1995 and related work by Fauconnier 1975, Lahiri 1995 and Eckardt 2005. These authors propose that licensing results from conspiring pragmatic factors. Theories assume that an NPI comes with a set of alternatives which lead to alternative salient propositions. NPIs are licensed iff the proposition asserted is stronger than any of these alternatives. This account is introduced and illustrated in the contributions by Crnič, Matsui and Eckardt/Csipak in the present volume. Conspiracy theories have attracted a lot of attention in recent years because of a general increase of work in alternative semantics and work in formal pragmatics (a.o. Kratzer/Shimoyama 2002, Chierchia 2004). Variants of this theory assume different alternatives and different kinds of assertion operators or other speech act operators, i.e., some NPIs are said to require an scalar assertion, others an emphatic assertion (Krifka 1995). These accounts are particularly suited for indefinites like any/ever, as well as minimizers and maximizer NPIs such as a drop, or budge. If conspiracy theories are on the right track, ideally the phenomenon of NPI licensing should supervene on more basic factors. On the other hand it seems that a certain degree of grammaticalization has taken place for (highly) frequent NPIs with a functional meaning. These may include any, ever, auxiliary need, as well as, possibly, particles such as yet and anymore. A contrast between such functional NPIs and “content” NPIs has been observed in Sedivy 1990: While the latter can occur in denials without an overt licenser, the former cannot: (11)

a. A: B:

I didn’t think Pat would lift a finger to help. But Pat DID lift a finger.

b. A: I didn’t think Pat would ever pass the exam. B: * But Pat DID ever pass the exam.

6   

   Regine Eckardt and Manfred Sailer

Finally, there are what we’ll call hybrid theories. These analyze NPI licensing in terms of operators in syntax and semantics which were proposed on independent grounds. Yet, the interaction of operators is not motivated by the syntax or pragmatics of the sentence but has to be stipulated as part of the analysis. The first such account was proposed by Chierchia (2004) where a syntactic feature mechanism checks features which are introduced by the NPI. Chierchia alludes to general pragmatic principles, yet the more “tricky” cases of NPI use require idiosyncratic assumptions in order to predict licensing correctly. Hence, the approach re-introduces a refined version of Klima’s original account. Guerzoni (2006) presents a hybrid theory that assumes a conspiracy-based account for minimizers and other non-functional NPIs, but a syntactic, featurebased account for any and ever. Looking at the historical development of NPIs and at closely related languages and dialects can be very insightful in this respect. We expect to find an element moving from a conspiracy-based distribution to a more grammaticalized distribution. At the same time, since the core pattern can be explained by both types of accounts, it will require very detailed empirical discussion to determine which type of theory is correct for a given item. Let us situate the contributions of the present volume in this landscape of theories.

3 The contributions 3.1 Pragmatic conspiracy accounts The contribution by Matsui discusses the Japanese understaters a(n)mari and sonnani which are similar to English all that in that they modify degree predicates and lead to downtoned statements. While the distribution of sonnani patterns with all that, a(n)mari differs from common NPIs in that it is acceptable in because-clauses, but bad in questions. Matsui takes this as reason to develop a new analysis for a(n)mari in terms of van Rooij’s (2003) notion of strength as relevance. Van Rooij’s view explains licensing of NPIs in questions, but Matsui opens up the possibility that the basic notions of his account can serve to analyze NPIs in more contexts. She proposes that a(n)mari shares the meaning of intensifying very but fulfills a different pragmatic function in language. A(n)mari is licensed if the resulting statement is less easily refuted, and hence more relevant in terms of van Rooij than the unmodified version of the sentence. She thus claims that not all NPIs lead to logical strengthening, but all NPIs relate to the same limited spectrum of pragmatic functions. The second Japanese downtoner, sonnani, is classed

Beyond “any” and “ever”   

   7

as another case of Krifka’s (1995) analysis of at all which the paper summarizes in a very clear and insightful manner. Onea and Sailer offer a study which complements Matsui’s paper in an interesting way. Like Matsui, they focus on English all that and approach the expression on a pragmatic basis. However, they argue that all that is not an NPI and requires an essentially different analysis. Firstly, introspective, corpus and questionnaire based data show that the distribution of all that does not fully match the set of contexts which allow (strong or weak) NPIs. The authors point out that at that seems to encode a reaction to another agent’s belief and propose two presuppositions which, put simply, state (i) that someone else believed that ‘x is P to a very high degree’ but (ii) the speaker does not believe that ‘x is P to a very high degree’. They demonstrate how these two will lead to inconsistency if all that is used in positive episodic statements. Moreover, the analysis predicts the peculiar distribution patterns of all that which are almost, but not exactly, the patterns of NPIs. Onea and Sailer refute existing analyses for all that, notably Krifka 1995 whereas Matsui, in her somewhat more detailed discussion, comes to a different conclusion for a(n)mari. One might speculate that the anaphoric/deictic property of all that is absent in Japanese a(n)mari, which may be a reason for the differences between seemingly similar items. Crnič draws attention to the fact that even, in the use which refers to low ends on a scale (“weak even”, corresponding to German auch nur) can be used in desire contexts and imperatives like Show me even one party that cares for the people. He demonstrates that an analysis which interprets even as part of the prejacent will not only compute wrong truth conditions but moreover predicts that such examples should be unacceptable. Taking up earlier proposals by von Fintel 1999, he argues that exhaustification is part of the sentences in question. To put it simply, he assumes that imperatives as the one above involve a tacit only-like component, roughly conveying the message that the speaker will be happy already when only a modest request is fulfilled, i.e. one party of a certain kind has been produced. This elegant idea not only captures the intended desires and requests correctly, Crnic also demonstrates how the ingredients can be put to work in his formal analysis of licensing “weak” even in desire and imperative contexts. Eckardt and Csipak start from the observation that existing pragmatic analyses for NPIs in the spirit of Krifka 1995 and Chierchia 2006 fail to predict the more limited range of contexts that we observe for strong NPIs, and specifically for minimizer NPIs like lift a finger. This is all the more surprising as the basic elements of pragmatic licensing (salient alternatives, logically stronger state-

8   

   Regine Eckardt and Manfred Sailer

ments) intuitively seem to play a role in our understanding of sentences like he didn’t lift a finger. They show why pragmatic licensing isn’t taylored to predict the narrow distribution of minimizer NPIs (echoing Crnič’s observation on the link between likelihood and logical strength). In search for criteria which limit the use of strong NPIs, they propose to take a look at side messages of scalar emphatic statements. Whereas strong contexts are nonveridical (the sentence denies that certain types of events took place), weak contexts suggest that certain types of events happened, or might happen in limited numbers or rare situations. Eckardt and Csipak argue that such side messages are nonsensical, and minimizers are excluded from contexts which give rise to nonsensical side messages.

3.2 Hybrid analyses in syntax and semantics The chapter by Gajewski offers an insightful illustration of the sophisticated and powerful analyses which arise in modular treatments of polarity licensing. Gajewski addresses exceptive but as in Nobody but Sally knew an answer. Its distribution derives from NPI and free choice items in that it can be combined with free choice any and NPI any. Gajewski takes set subtraction to be the basic meaning of but. However, “X but Y” statements entail not only facts about X\Y but also facts about Y. Our example entails both ‘Nobody who isn’t Sally knew an answer’ but also ‘Sally knew an answer’. In a series of analyses, Gajewski uses basic operators of NPI licensing (Krifka 1995, Chierchia 2006) to develop an analysis of but on basis of recursive exhaustification (hidden only) and a minimality condition on the exception set. Unlike its predecessors, this analysis is hosted in a general framework of NPI licensing, and Gajewski can demonstrate that this is not love’s labours lost: His integrated account can correctly predict the interaction of but with any in various constellations which were not treated correctly by previous analyses. We count Gajewski as a proponent of hybrid theory, as he suggests that alternative formation and exhaustification could be semantically universal operations like quantification and functional application. His view leads to ancillary questions: if such operators are available in general, what can be said about their distribution at LF, their scope taking behaviour, nesting and crossover, and other structural properties? Research in NPI licensing can tell us more about the atomic parts of semantic composition. Hsieh’s study illustrates the potential of combined semantic and pragmatic operators in the analysis of NPI licensing. Hsieh takes a closer look at examples like If we don’t pay any fees, we might get into trouble which contain polarity sensitive

Beyond “any” and “ever”   

   9

items in the restrictor of possibility conditionals. Such data are problematic for existing theories of NPI licensing, given that the standard Kratzer-Lewis semantics of conditionals assumes that they are existential quantifications over sets of possible worlds. Logical accounts of NPI licensing would hence predict that any in a sentence like the one above should not be licensed, given that existential quantification does not create a downward entailing context. In answer to this problem, Hsieh refers to observations about St’at’imcets modals where possibility modals have been shown to be universals over a limited domain of quantification, rather than existentials. Making use of proposals by Rullmann et al. 2008, Hsieh devises an analysis of possibility modals in English which predicts that they are acceptable hosts for NPIs. Herburger and Mauck advocate an account of syntactic NPI licensing which integrates insights of semantic theories. They start with a detailed discussion of problems for accounts of NPI licensing which rest on semantic or pragmatic terms. Semantic licensing in terms of downward entailment, they argue, cannot satisfactorily be extended to licensing contexts like hardly, barely, conditionals, regret/sorry and few, as well as the occasional licensor most. Pragmatic accounts of licensing are to date not explicit enough in relating the meaning of exhaustive operators to explicit even. The authors point out that similar lexical items in related languages underwent different developments as polarity sensitive items and that the diachronic developments seem driven by accident.1 In a second step, the authors answer the diagnosed problems by outlining the basics of a syntax-driven licensing mechanism which draws on semantic properties of parts of the sentence. They propose a mapping from semantics to syntax which turns the logical scope of operators into an LF feature: Downward entailing and other operators which can license NPIs in their argument are proposed to assign their arguments a minus feature at LF. With this feature mapping, LF structures represent whether a syntactic context is (locally or globally) negative and can host NPIs. Further restrictions could be used to fine-tune the range of contexts which are predicted to license NPI. What is original in the analysis is that it explicates a way to transport semantic features into syntax. The paper is hence one of the few which explore the mechanisms of grammaticalization on a formal basis.

1 The argument leaves aside the common observation that inseminating contexts of use are needed for linguistic changes (e.g. Eckardt 2006, Eckardt 2012 on NPIs in diachronic perspective).

10   

   Regine Eckardt and Manfred Sailer

3.3 (Neo-)monolithic approaches Levine treats one of the lesser researched NPIs, the modal need. While at first sight, need shares the distribution of weak NPIs, Levine draws attention to a peculiar connection between possible licensors and need in gapping constructions like He thought he would have to come, but he needn’t – . Whereas need can be licensed by Neg-raising here, more remote licensing is prohibited: *… but I don’t think he need – . He shows that the version without an adjacent VP is a superstrong NPI, an NPI that requires a clause-mate sentential negation. His observation is relevant for our understanding of NPIs, since so far no such superstrong NPIs have been identified for English. He also shows how this special property of modal need with no overt complement can be used as an empirical test to group various syntactic phenomena together. Levine’s paper shows how a detailed look at a single NPI makes an important contribution to our understanding of a wide class of syntactic phenomena. He argues that an integrated framework of linguistic analysis, like HPSG, can represent syntactic properties of items in the lexicon and connect them with other features like licensing requirements. Whereas Levine uses the logical terminology to characterize types of NPIs, the focus of his paper is to show the degree of arbitrariness that we have to expect when looking at NPIs: There is no obvious difference in the lexical meaning of the main verb need (to), the auxiliary need followed by a VP, and the auxiliary need followed by a gap. Nonetheless the syntactic differences come along with important differences in licensing conditions – ranging from a non-NPI, to a weak NPI, to a superstrong NPI. Brandtler proposes his very own and special mix of syntactic and semantic criteria in NPI licensing, based on his 2012 PhD thesis. He takes his starting point from Mainland Scandinavian which, according to Brandtler, shows two basically different constellations for NPIs. Firstly, he considers “non-evaluable utterances”. These include if-clauses, y/n-questions and optatives, and allow NPIs without further additional licensing. His second class, called “evaluable statements,” include all propositions about which an utterance makes a claim as to truth or falsity. These include assertions, negative assertions, presuppositions and entailments. Brandtler argues that these two classes also differ in syntax. Typical nonevaluable clauses in Swedish are V1 whereas typical evaluable clauses in Swedish show V2 syntax. Hence, the author diagnoses a syntactic correlation between (non-)evaluability and CP syntax of the sentence. Non-evaluable sentences are assumed to lack Spec-CP whereas evaluable sentences possess Spec-CP which can, or even has to contain overt material. The latter kind of sentence requires an appropriate licensor to host NPIs whereas the former is a “free host”, so to speak.

Beyond “any” and “ever”   

   11

The paper leaves it open whether the syntactic presence of a licensor is taken as necessary or whether an operator which creates the appropriate semantic context for NPI is indicative, as logical and pragmatic approaches to licensing would assume. While terminology may still need a bit of sharpening at some points, the paper puts into focus the fact that non-assertive speech acts (like questions) as well as their diachronic descendent constructions (such as antecedents of conditionals, optatives) share similarities both in NPI licensing and in assertiveness.

3.4 Use and history: assessing the data A final range of papers is concerned with the actual use of NPIs in corpora and speech. Van der Auwera and van Alsenoy trace the diachronic development of the cognates of any in English, Dutch and German, and supplement it with additional data from lesser-studied West Germanic languages such as Afrikaans, Pennsylvania Dutch and West Frisian. The comprehensive survey not only covers quantificational readings/uses – as those classically covered in the work of Haspelmath 1997 – but also uses as content words. For example, Dutch enig can be used in the sense of the English adjective wonderful, and German einig denotes a property of pluralities which can be glossed as ‘consensual(ly)’. The authors argue that such readings can link various quantificational or polarity sensitive uses, a hypothesis which is nicely illustrated by their diachronic record. What may be noteworthy is that intermediate content words like enig (‘wonderful’) and einig (‘consensual’) can be used in all contexts. Given that at least some of these content readings emerge from NPI readings, we can conclude that licensing requirements do not emerge by speakers simply imitating earlier usage. If that were so, then all readings/versions of any-words that go back to an NPI should be NPIs, as well. The study of van der Auwera and van Alsenoy offers indirect justification for the shared assumption of authors of the whole volume: NPI licensing is rooted in grammar and not simply a cultural practice, traded by imitation of older patterns. The paper by Gast is likewise situated in diachronic linguistics. He offers a precise and detailed study on the emergence of a single negative polarity item, English either. Gast’s study reminds us of the fact that the paths of language development and change are not always as cleanly delineated as writers of linguistic theory might hope. He reports that the uses of modern either derive from a morphologically transparent paradigm of dual quantifiers in Old English which distinguished ‘a and b’ (æghwæðer), ‘a or b’ (awðer) and ‘Not ( a or b )’ (nawðer). Gast starts from the puzzling observation that either turned from a dual universal

12   

   Regine Eckardt and Manfred Sailer

(both of two) into a dual existential (there’s one of two) which is, moreover, polarity sensitive. While first order logic tells us that (∀¬) and (¬∃) are equivalent, the true history of either, as Gast demonstrates, is not confined to one simple semantic reanalysis from ∀ to ∃ in negative contexts. After centuries of uses of æghwæðer (∀) along with awðer (∃) , the latter predominating in (certain) negative contexts, the phonological merge of the two words in late ME led to the emergence of NPI either in the Modern English sense. Crucial were utterances where hearers/readers faced a word that sounded like ægder and was used in a negative context. For contemporary hearers, this word could be analysed either as a (low scope) existential, or else as a universal which looked as if it were positioned syntactically too low for the intended meaning of the sentence. Interpretation of the sentence with a high scope universal (in the ∀¬ sense) would have required quantifier raising beyond clause boundaries. Current syntactic theories of movement between syntax and logical form (LF) claim this type of movement to be impossible (called island violation). It may be surprising that the linguistic behaviour of speakers before 1400 corroborates this rather abstract claim of modern linguistics. Obviously, the speaker community at the time chose to understand æghwæðer in a new, existential sense rather than standardly allowing for island violations in negative contexts. The corpus study by Hoeksema takes a closer look at Strawson downward-entailing contexts (superlatives, the only, the first, the last, only XP) in comparison to standard licensing contexts (negation, questions). While all these contexts are known to license the standard NPIs any and ever, Hoeksema’s research project does justice to the title of this volume in that he offers detailed distribution profiles for a much wider range of NPIs in English, Dutch and German. The study covers anymore, D meer, G mehr, E in weeks, months, …, E remotely, minimizers like care a pin, lift a finger, E yet, as yet and other expressions which have been noted to be restricted to negative contexts. For each of his items of interest, Hoeksema reports its occurrence in various Strawson DE contexts, but also gaps in the corpus record, i.e. combinations of item and context which one would expect but which never occur. In a next step, Hoeksema confirms the gaps by introspection. He demonstrates that the respective combinations in corpora are not simply unattested but are missing because they would actually be ungrammatical. To name just one example, the verb need is sharply ungrammatical in superlative constructions, in spite of its occurrence in other weak and Strawson DE contexts. Hoeksema demonstrates how theoretically motivated corpus studies can bring to the fore new facts and examples which might go unnoticed in purely descriptive corpus research. His findings pose a challenge for NPI analyses of all kinds, because the great variety of individual patters of distribution seems to defy

Beyond “any” and “ever”   

   13

the hope for simple and uniform factors of NPI licensing. Intriguingly, many of Hoeksema’s gaps look like structurally motivated ommissions. We therefore hope that the material in this chapter will spur future research in NPI licensing rather than offering proof that any and ever are the only existing NPIs. Liu, Eckardt and Radó, finally, investigate speakers’ intuitions about the acceptability of minimizer NPIs in various linguistic environments. They show that there is a clear pragmatic link between the use of scalar particles (auch nur) and minimizers, however in a surprising direction. While earlier authors anecdotically remark that NPI use seems to be improved stilistically by even/auch nur, their study shows that it is the minimizers which faclitate the accommodation of a scale and interpretation of auch nur. The effect even remained when sentences were put into discourse context. This corroborates the view that minimizer NPIs are lexically linked to a scale of alternatives, whereas ordinary expressions are not. It seems therefore theoretically justified to use scalar approaches of NPI licensing at least in these cases.

4 What counts as an NPI? In building theories for a range of items, it is essential to take a balanced decision on the data to be included. Traditionally, English (existential) any and ever offer the model case of negative polarity items which are standardly used in the literature to illustrate the phenomenon. Looking beyond any and ever, scholars have to decide which other words and expressions should count as negative polarity items. One simple answer could be that all and only expressions which can occur in exactly the same kinds of contexts as any and ever should qualify as negative polarity items. However, this strategy is unattractive. For one, it limits the range of items to a much smaller number than what is currently investigated in the field. More importantly, it excludes expressions which are limited to contexts of negative polarity for seemingly systematic reasons: minimizers like budge, red cent, lift a finger, but also maximizers like in years, in ages, in the world. These are more restricted in distribution than any and ever and hence classed as strong NPIs. The class is intriguing because it is potentially open and can be traced in many nonrelated languages, e.g. Slavic languages, Hungarian, Hindi, Mandarin Chinese (ta lian ge-pi dou mei fang ‘she didn’t give a fart/ say anything’), Japanese

14   

   Regine Eckardt and Manfred Sailer

(hitokoto ‘a/one word’ and itteki ‘a/one drop’) and more.2 Therefore, strong NPIs have always been part of NPI research. However, strong NPIs are not as homogenous as a first glance at examples might suggest. Along with NPIs of a common semantic “build” we find words like until in English which has been argued to be a strong NPI (de Swart 1996, Gajewski 2005), but which is not a minimizer. Features like the semantic “build” do not matter for monolithic theories of licensing, e.g. syntactic or logic licensing. Conspiracy accounts, however, rest on notions like alternatives to a given word. While it is pretty intuitive to decide what alternatives to “(not) lift a finger” or “any potatoe” might be, it is less obvious whether a word like until or NPI need (German brauchen) gives rise to alternatives in the same way. A radical definition of negative polarity item could run like this: An NPI is a word/expression which is restricted in use to those contexts which can host any and ever or to any subclass of these contexts.

This definition, simple and unaffected by theoretical considerations, covers hypothetical expressions with really odd restrictions in use. They could be limited to the scope of few, or only occur in the restrictor of universal quantifiers, or just with negation and in the scope of only and so on. Indeed, some interesting items suggest that this definition is needed. Hoeksema 2010b, Richter and Sailer 2009 discuss the Dutch minimizer kip (‘chicken’) which can only be used when a scene is described in which there are no humans (Ik heb geen kip gezien ‘I didn’t see anyone/any chicken’ vs. ??Geen kip sprak met de kinderen. ‘No one/no chicken has spoken to the children’). The German expression Federlesen ‘fuzz/ado’ not only is an NPI but it also requires a further quantification such as ohne langes /viel Federlesen ‘without long/much ado’. Likewise, Giannakidou’s extensive work on polarity sensitivity in Greek, resulting in a rich range of licensing criteria based on veridicality, demonstrates that not all NPIs fall neatly in the weakstrong-superstrong typology proposed by Zwarts (1995, 1997). Her notion of NPI likewise includes expressions with a wider distribution than any/ever in English. Obviously, the definition of “negative polarity item” eventually has repercussions on the success of various approaches of analyses. While some of our authors carefully restrict their study to any and ever, several papers highlight the problem of classing an expression as negative polarity sensitive.

2 We thank Stefan Kaufmann, Mingya Liu and Edgar Onea for valuable input.

Beyond “any” and “ever”   

   15

Levine investigates need as an NPI.3 It is inspiring to read his contribution in parallel to Hoeksema’s paper who shows that need does not pattern with either strong or weak NPIs in distribution. Hoeksema reports that while it is acceptable in the scope of only, the words need, hoeven, brauchen are inacceptable in the scope of superlatives, the first, the last and similar uniqueness contexts. Mysteriously, though, these NPIs are fine in the scope of only.4 Hence, licensing contexts for need/hoeven/brauchen range somewhere between downward entailing and much stronger contexts and it is unclear how criteria can be adjusted. Gajewski develops a sophisticated analysis of English exceptive but as a negative polarity sensitive item. What is striking for the German reader is the fact that the best German translation, außer, is not limited to negative contexts in the same way as the English exceptive marker is. The difference can be illustrated by the following pair of sentences. (2) * Two students but Maria speek French. √ Zwei Studenten außer Maria sprechen Französisch. two students but Maria speak French The German sentence is acceptable and conveys that Maria, plus two more, speak French. Van der Auwera and van Alsenoy might point out that both außer and but derive from a local preposition which means “outside”. Hence, it would be interesting to understand the branching point where the use of but became subject to licensing mechanisms which außer could evade. The expression all that in English and Japanese sonnani (‘all that’) were classed differently by different authors in this volume. Whereas Matsui adopts the analysis of Krifka (1995) and hence views Japanese sonnani as an NPI, Onea and Sailer argue that the distribution of all that, although superficially similar to other NPIs, should be analysed in terms of speaker belief. If we adopt the view that distribution is criterial to classify an expression as negative polarity item, then the architecture of the best analysis should not matter for the status of the item. Many linguists, however, would subscribe to the view that Onea and Sailer,

3 Or rather, as a pair of NPI homonyms – for simplicity’s sake, we refer to need as if it were a single item here. The argument can be recast in terms of addressing two homonyms of possibly different strength. 4 Hoeksema 2010b shows that the three NPIs are usually disallowed in the restrictor of a universal quantifier (*Every student that needs to worry will be informed.). The only exception is when there is an only/nothing else interpretation of the universal (All you need know is in the book.).

16   

   Regine Eckardt and Manfred Sailer

to the extent that their analysis is superior to the Krifka-type account and hence the “correct” one, have demonstrated that all that is not an NPI even though it may look like one at first sight. From a conspiracy perspective, however, we may still call it an NPI and simply state that its distribution follows without requiring a particular operator – which is exactly the direction in which conspiracy theories tend to develop. Gast as well as van der Auwera and van Alsenoy draw attention to the fact that the status of words as NPI can change over language history. Van der Auwera and van Alsenoy’s map of cognates of any suggests that simple properties (‘wonderful’, ‘consensual’) are less likely expressed by NPIs than existentials, free choice items, and other quantificational elements. Gast’s case of meaning change is, perhaps, more challenging in this respect. He traces a meaning shift from universal to existential quantifier which goes along with a change in status (from nonsensitive to sensitive). Taken together, these studies raise challenging questions for NPI theory: Given that NPI-hood is not just a matter of imitation of earlier use (as entailed by Auwera and Alsenoy), what other factors in communication led speakers of Early ModE to use either with a licensing requirement? Hoeksema, finally, poses the ultimate challenge to a uniform notion of negative polarity sensitive items. He effectively demonstrates that the items in his study (e.g. minimizers, yet, as yet, alive, need, hoeven) have distribution patterns which differ significantly. He shows that some items are just downright bad in contexts in which the majority of other NPIs is acceptable. So, which of the expressions on Hoeksema’s list should be called NPIs? And how do we decide?

5 Any and ever on forbidden grounds? In spite of decades of research in negative polarity sensitivity, even the distribution of any and ever has not as yet been fully understood and some uses of any and ever follow criteria beyond logic or syntax. Linebarger 1980 argued that imperatives may contain NPIs when used as threats, but not when used as promises, even though the syntactic form and, arguably, the logical form is identical. In some instances, even synonymous expressions behave differently in licensing. It has been observed that while expressions such as no students can license NPIs, semantically similar expressions such as zero students don’t (for example in Krifka 1995).

Beyond “any” and “ever”   

(3)

   17

a. No students have ever read this book. b. * Zero students have ever read this book.

Similarly overt as if as a form of grammaticalized irony can license ever/jemals, but not formally marked irony does not (see Hoeksema and Sailer 2012). (4)

a. Als ob Fred jemals den Abwasch macht! As if Fred ever does the dishes! b. * Klar, und Fred macht jemals den Abwasch! * Sure, and Fred ever does the dishes!

Several authors in the volume discuss the use of any/ever in contexts which are unforeseen in the light of licensing theories. This situation can be met in two different ways: Either the contexts are analyzed in a way that makes them compatible with current theorizing, or the data are taken as evidence that NPI licensing is a highly idiosyncratic phenomenon with at best an independently delineable core set of contexts. Both approaches are proposed in the current volume. Crnič considers any in imperatives and in the scope of desire predicates, in uses which cannot successfully be viewed as free choice any. Hsieh demonstrates that any/ever can appear in contexts that look like existential quantification. Both authors work in the realm of logic/semantic NPI licensing and argue that the standard semantic analysis of the examples in question should be replaced with one which reveals their downward entailingness. Their proposals not only predict that these contexts can host NPIs, they also can explain the special flavour added by NPIs in these cases, which corroborates the proposed analyses. Herburger and Mauck take observations about NPIs in uncommon contexts as their prime argument in favour of syntactic licensing. They take a historical perspective on NPI licensing and propose that the synchronic distribution of any and its mates consists of a systematic core and its non-systematic extension which emerged by grammaticalization. The patterns that we find today combine logical and structural licensing criteria (e.g. locality restrictions) and therefore, they argue, syntax is the correct level of representation at which structural features like dominance or c-command should be combined with logical characterizations of context. Their study is complemented by the framework of Brandtler who proposes his own original combination of syntactic and semantic licensing factors, the potential of which still remains to be explored. Eckardt and Csipak point out differences in acceptability of (minimizer) NPIs in threats and promises

18   

   Regine Eckardt and Manfred Sailer

which were first observed by König. Following van Rooij, they tentatively suggest an analysis in terms of utilities, according to which NPIs in promises are more costly for the speaker than threats.

6 Conclusion This introduction should not leave the reader with the impression that all is lost in the arena of NPI licensing theory. We see how the same basic licensing mechanisms can be coded in syntax, semantics, pragmatics or in overarching hybrids. While the boundaries between modules of language become permeable, we are drawn to crucial fundamental issues in linguistics. Should meaning and grammar be autonomous bodies of rules and principles, or does it make more sense to assume that rules in grammar can have homomorphic pendants on the side of meaning and use? Beyond a certain core type of NPIs and NPI-licensing contexts, there is attested and documented variation. In our view, this variation is no reason to abandon the quest for theory. An overarching theory of NPI licensing should integrate and moderate between good solutions for clusters of data, as those presented in the papers collected in this volume which illustrate the power of modules of grammar and their combinatorial machinery in explaining negative polarity sensitivity. The contributions of the current volume also show how the study of NPIs can advance the study of grammar in various modules. To name a few cases, Levine’s contribution leads to direct consequences for the theory of extraction. Gajewski’s study addresses the question of which abstract operators are part of the core semantic computational system. NPIs can also provide important arguments for a particular analysis of certain non-canonical structures as the imperatives explored by Crnič. Another example is Gast’s observation that island constraints in syntax and semantics can shape the diachronic reanalysis of functional elements. Negative polarity items remain a challenging field of research where pragmatics, convention, and the logic of language conspire to shape the grammar of languages. We hope that the present volume fuels future research.

Beyond “any” and “ever”   

   19

7 References Chierchia, Gennaro. 2004. Scalar implicatures, polarity phenomena, and the syntax/ pragmatics interface. In: A. Belletti (ed.) Structure and Beyond: The Cartography of Syntactic Structures. Oxford: Oxford University Press: Vol. 3, 39–103. Chierchia, Gennaro. 2006. Broaden your views. Implicatures of domain widening, and the “logicality“ of language. In: Linguistic Inquiry 37(4): 535–590. de Swart, Henriette. 1996. Meaning and use of ‘not … until’. In: Journal of Semantics 13: 221–263. Eckardt, Regine. 2005. Too poor to mention. Subminimal Eventualities and Negative Polarity Items. In Maienborn, C., A. Wöllstein-Leisten (eds.): Events in Syntax, Semantics and Discourse, 301–330. Tübingen: Niemeyer Verlag. Eckardt, Regine. 2006. Meaning Change in Grammaticalization. An Enquiry into Semantic Reanalysis. Oxford: Oxford University Press. Eckardt, Regine. 2012. The many careers of negative polarity items. In B. Cornillie, H. Cuyckens, K. Davidse, T. Mortelmans und T. Leuschner (eds.): Proceedings of New Reflections on Grammaticalization 4. Amsterdam: John Benjamins. Fauconnier, Giles. 1975. Polarity and the Scale Principle. In: Papers from the Eleventh Regional Meeting of the Chicago Linguistic Society. 188–199. von Fintel, Kai. 1999. NPI-licensing, Strawson-entailment, and context-dependency. Journal of Semantics 16: 97–148. Gajewski, Jon. 2005. Neg-Raising: Presupposition and polarity. PhD Thesis, MIT. Giannakidou, Anastasia. 1998. Polarity Sensitivity as (Non)veridical Dependency. Amsterdam: John Benjamins. Guerzoni, Elena. 2006. Intervention effects on NPIs and feature movement. In: Natural Language Semantics 14(4): 359–398. Haspelmath, Martin. 1997. Indefinite pronouns. Oxford: Oxford University Press. Hoeksema, Jack. 2010. Negative and positive polarity items: An investigation of the interplay of lexical meaning and global conditions on expression. In: L. R. Horn (ed.): The Expression of Negation, 187–224. Berlin: Mouton de Gruyter. Klima, Edward. 1964. Negation in English. In J.A. Fodor and J.J. Katz (eds.): The Structure of Language. Englewood Cliffs, NJ: Prentice Hall. Kratzer, Angelika and J. Shimoyama. 2002. Indeterminate pronouns. The view from Japanese. In: Y. Otsu (ed.): Proceedings of the Third Tokyo Conference on Psycholinguistics, 1–25. Tokyo: Hituzi Syobo. Longer version electronically available at Semantics Archive, URochester. Krifka, Manfred. 1995. The semantics and pragmatics of polarity items. Linguistic Analysis 25(3/4): 1–49. Ladusaw, William. 1979. Polarity sensitivity as inherent scope relations. Ph.D. dissertation, The University of Texas at Austin. Lahiri, Utpal. 1998. Focus and negative polarity in Hindi. Natural Language Semantics 6(1): 57–123. Linebarger, Marcia C. 1980. The Grammar of Negative Polarity. Ph.D. dissertation, Massachusetts Institute of Technology. Progovac, Lijliana. 1993. Negative Polarity: entailment and binding. Linguistics and Philosophy 16: 149–180.

20   

   Regine Eckardt and Manfred Sailer

Richter, Frank and Philip Soehn. 2006. Braucht niemanden zu 
scheren: A survey of NPI licensing in German
. In: Stefan Müller (ed): Proceedings of the 13th International 
Conference on HPSG, 421–440. Stanford: CSLI 
Publications. Rullmann, Hotze, L. Matthewson and Henry Davis. 2008. Modals as distributive indefinites. Natural Language Semantics 16(4): 317–357. van Rooij, Robert. 2003. Negative Polarity Items in questions: Strength as relevance. Journal of Semantics 20: 239–273. Sailer, Manfred. 2009. A representational theory of negative polarity item licensing. Habilitation thesis, University of Göttingen. Sedivy, J. 1990. Against a unified analysis of negative polarity items. In: Cahiers Linguistiques d’Ottawa, 18. Ottawa, University of Ottawa: 72–90. van der Wouden, Ton. 1997. Negative Contexts. Collocation, Polarity and Multiple Negation, Routledge: London and New York. Zwarts, Frans. 1995. Nonveridical contexts. Linguistic Analysis 25(3/4): 286–312. Zwarts, Frans. 1998. Three types of Polarity. In: F. Hamm, E. Hinrichs (eds.): Plurality and Quantification, 177–228. Dordrecht: Kluwer.

Johan van der Auwera & Lauren Van Alsenoy

Mapping the West Germanic any’s* 1 Introduction Hoeksema (2010: 188) claims that any is “without any doubt the world’s bestknown, and most intensively studied polarity item”. If Hoeksema says so, it must be true. Any’s West Germanic cousins, however, i.e., etymological counterparts such as German einig or Dutch enig, have not been that fortunate. For German there are some isolated remarks in Fobbe (2004) and Jäger (2008) – as well as in older (historical) work. For Dutch, the situation is a bit better, in part thanks to Hoeksema (Hoeksema & Klein 1995; Hoeksema 2005, 2007/2010a; see also Giannakidou 2010, Willis 2012, van der Auwera & Van Alsenoy 2012). For West Frisian there may just be one study (Hoekstra & Siebinga 2008), for the other West Germanic languages, vernaculars and dialects there is next to nothing. This paper begins to remedy this situation. At least, this paper will put Afrikaans, Pennsylvania Dutch and Yiddish on the agenda, with some observations, and for German and Dutch we offer an integrated account, which further relates them to English and to the older stages of these languages. Another issue of which this paper attempts to advance our understanding is that all West Germanic any words had or have adjectival uses. Apart from Jäger’s (2008: 194–196) discussion of Old High German the research literature mentions this fact only in a side remark or footnote (Hoeksema & Klein 1985: footnote 21; Giannakidou 2010: footnote 1). Our approach uses semantic maps. In section 2 we will critically discuss the semantic map account offered in Haspelmath (1997). In section 3 we treat some West Germanic any words with a map that differs from that of Haspelmath’s. We focus on English, Dutch and German and briefly also discuss West Frisian, Afrikaans, Pennsylvania German, and Yiddish. Section 4 is the conclusion.

* Thanks are due to the Belgian Federal Government (Federal Science Program, project on ‘Grammaticalisation and Intersubjectification’) as well as the Research Foundation Flanders (projects on ‘Jespersen cycles’ and ‘Indefiniteness’). We are very grateful also to Mark Louden (Madison, Wisconsin) for informing us on Pennsylvania German, to Han Nijdam (Frisian Academy, Ljouwert/Leeuwarden) for help on West Frisian, and to Jef Van Loon for help on the etymology of the any words. Finally, we also thank the two anonymous reviewers and the editors.

22   

   Johan van der Auwera & Lauren Van Alsenoy

2 Haspelmath (1997) The mapping that we will undertake is not geographical, but semantic. A semantic map, at least a classical one1, arranges uses in conceptual space such that semantically related uses are connected. The further claim is that whenever a construction, whether it is a morpheme, a word or a phrase, is used for more than two uses, then, everything else being equal, these uses must be connected – this is the so-called ‘contiguity’ or ‘adjacency’ requirement.2 It so happens that arguably the best book length illustration of the semantic map approach is one about indefinite pronouns and determiners, i.e. Haspelmath’s Indefinite pronouns (1997). As is to be expected, Haspelmath provides an account of English any. Haspelmath’s indefiniteness map has nine uses. These are illustrated in (1) to (9) with English somebody, anybody, and nobody. Six of these uses are available for anybody. These are the ones shaded in Figure 1. (1)

Somebody called you while you were away: guess who? [specific known]

(2)

I heard somebody, but I couldn’t tell you who. [specific unknown]

(3)

Please ask somebody else. [irrealis non-specific]

(4)

a. If you see somebody, tell me immediately. b. If you see anybody, tell me immediately. [conditional]

(5)

a. Did somebody tell you about it? b. Did anybody tell you about it? [question]

1 For the difference between classical and non-classical, “multidimensional scaling” maps, see Croft and Poole (2008) vs. van der Auwera (2008, in print). 2 The ‘everything else being equal’ phrase refers to some types of exceptions – see van der Auwera & Temürcü (2006).

Mapping the West Germanic any’s   

(6)

He is better than anybody else in the group. [comparative]

(7)

I don’t think that anybody knows the answer, [indirect negation]

(8)

Anybody can solve this problem. [free choice]

(9)

a. I haven’t seen anybody.

   23

b. I have seen nobody. [direct negation]

Figure 1: English any on Haspelmath’s 1997 indefiniteness map

Much of what the literature says about any is ‘translated’ onto the map. Thus conditional, question, comparative, indirect and direct negation are what have been called ‘negative polarity’ contexts (see e.g. Ladusaw 1979). What is nowadays called ‘free choice’ is there as well, and interestingly, the map itself does not force the analyst to take up a stand as to whether there is really only one any. The map, we have stated in the above, is a map of uses (or ‘functions’, as Haspelmath 1997 calls them). It is non-committal as to the number of any meanings it involves. One may advance the map with a monosemy hypothesis implying that all uses instantiate one meaning. Or one may equally well accept two meanings (e.g. one for free choice and another one for everything else) or even six (for the map after all distinguishes six any uses). Note also that we do not claim that the map says all there is to say about any, and neither does Haspelmath for that matter. For example, Haspelmath (1997: 190–192) is aware of what he calls the ‘anti-depreciative’ use, illustrated in (10). (10)

I don’t read just any old linguist, I only read Antoine Meillet.

24   

   Johan van der Auwera & Lauren Van Alsenoy

But for some reason, this use does not make it into the map, although it would seem to be easy to do this. The ‘anti-depreciative’ use, we think, is simply the negation of free choice. Haspelmath (1997: 248) also provides an analysis of Dutch enig. It is shown in Figure 2.

Figure 2: Dutch enig on Haspelmath’s 1997 indefiniteness map

The four uses of Dutch enig ‘any’ distinguished by Haspelmath are all negatively polar. They are illustrated in (11) to (14).3 (11)

Had hij toen enige vrienden? had he then any friends ‘Did he have any friends then?’

(12)

Hij deed het zonder enige hulp. he did it without any help ‘He did it without any help.’

(13)

Als hij enig succes heeft, dan komt het door if he any success has then comes it through zijn goede contacten. his good contacts ‘If he has any success, then it is due to his good connections.’

(14)

Het is nu kouder dan in enige vorige winter. it is now colder than in any previous winter ‘It is now cooler than in any previous winter.’

3 (12) is supposed to illustrate indirect negation. The latter is a somewhat vague category, covering both superordinate negation, illustrated in (7), and ‘without phrases’, as illustrated in (12). See also Haspelmath (1997: 80–81).

Mapping the West Germanic any’s   

   25

These uses are typical for singular count nouns. One further use, not mentioned by Haspelmath, is direct negation, as illustrated in (15).4 (15)

Niemand had enig boek gelezen. nobody had any book read ‘Nobody had read even one book.’

So the enig ‘any’ sphere extends a little bit more to the right than what see on Figure 2. There is, however, another enig or, if one prefers, another use of enig, most strongly brought to the foreground by Hoeksema (2007/2010a). This other enig corresponds to some. It occurs with quite a few of the Haspelmath functions, but it very nearly but not quite (Hoeksema & Klein 1995: 167; van der Auwera & Van Alsenoy 2012: 339–340) only combines with plural count nouns or singular mass nouns. (16) to (18) illustrate the specific known, specific unknown, and the irrealis non-specific uses for plural nouns. (19) is a mass noun example, which could be taken as specific known or unknown. (20) illustrates the unacceptability of a singular count use. (16)

Ik heb enige boeken meegebracht, Lucky Jim, Requiem en L’Étranger. I have some books brought.along Lucky Jim, Requiem en L’Étranger ‘I have brought along some books, Lucky Jim, Requiem en L’Étranger.’

(17)

Ik heb enige boeken meegebracht, maar ik weet niet welke. I have some books brought.along but I know not which ‘I have brought along some books, but I don’t know which ones.’

(18)

Geef me enige boeken. give me some books ‘Give me some books.’

(19)

Dat duurt enige tijd. that lasts some time ‘It lasts some time.’

4 Dutch enig does not, however, combine with niet ‘not’. This is perhaps a reason for considering (15) as a case of indirect negation. The latter notion, we have pointed out in note 3, is a vague notion, which needs further study.

26   

   Johan van der Auwera & Lauren Van Alsenoy

(20) * Geef me enig boek. give me some book ‘Give me some book.’ Enig ‘some’ also occurs in three contexts that allow enig ‘any’, viz. question, conditional, and indirect negation. (21)

Heeft hij enige boeken meegebracht? has he some books brought.along ‘Has he brought along some books?’

(22)

Als hij enige boeken meebrengt, dan is het goed. if he some books brings.along then is it good ‘If he brings along some books, then it is good.’

(23)

Ik geloof niet dat hij enige boeken heeft meegebracht. I believe not that he some books has brought.along ‘I don’t think that he has brought along some books.’

Interestingly, these three contexts are also the ones in which English allows both some and any.5 Since enig can correspond to both some and any, it is not surprising that the polysemy shows up in these very same contexts. This enig, corresponding to some, can of course be argued to be a different enig. And this is in fact Haspelmath’s position. Haspelmath (1997: 11–12) would consider it a quantifier, more precisely a ‘mid-scalar quantifier’, instead of an indefinite determiner or pronoun. He knows that the two uses are related and also that other linguists would not consider the uses to be that different. Perhaps even most linguists would study the ‘some’ and ‘any’ uses together, for in the standard approach all ‘some’ and ‘any’ uses are treated in quantificational terms.6 Furthermore, from Haspelmath’s own perspective, the inclusion of enig ‘some’ would seem easy. Just like the English map has a place for both free choice any and negatively polar any, if one cares to claim that English has two any words, the Dutch

5 Haspelmath (1997) accepts the question and conditional uses of some, but not the indirect negation use as in (a). See Szabolsci (2004: 415) and van der Auwera & Van Alsenoy (2011: 11). (a) I don’t think that somebody will be there. 6 The term ‘mid-scalar’ does not strike us as felicitous either. The fact that enig can be paraphrased as ‘a few’ suggests a positioning on the lower end of the scale rather than in the middle. Just how this element of low scalarity accrues to the ‘some’ use of enig is an interesting further issue.

Mapping the West Germanic any’s   

   27

map could have a place for both enig meaning ‘any’ and enig meaning ‘some’. The uses in questions, conditionals and indirect negation require no further action, for enig ‘any’ is also registered there. So all we have to do, in the Haspelmath perspective, is to extend the map to the right, so that enig also covers direct negation (the use illustrated in (15)). This is shown in Figure 3.

Figure 3: Dutch enig on Haspelmath’s 1997 indefiniteness map – second version

The map does not of course show that when enig occurs in the question, conditional or indirect negation cells, it can correspond to either any or some. This is a problem of the Haspelmath map that has been noted for other phenomena as well and which Haspelmath himself is aware of (see van der Auwera & Van Alsenoy 2001: 6–7). It is also one of the reasons why we will offer another map (section 3).7 But the story of enig is not finished yet. There are two more uses, illustrated in (24) and (25). (24)

Dit vind ik enig. this find I wonderful ‘I find this wonderful.’

(25)

De enige manier om hem te laten zwijgen is met bier. the only way in.order.to him to let be.quiet is with beer ‘The only way to make him quiet is with beer.’

In (24) and (25) we see an adjective. Here the pressure to consider these uses as uses of a different enig is even stronger, for the other enig uses are determiner uses (and also pronoun uses, but we haven’t illustrated those). Yet the senses of the adjective and determiner seem to related, at least diachronically. The deter-

7 We will then also argue that English some is inherently specific and any inherently nonspecific but that question, condition and indirect negation contexts make some contextually non-specific and hence very close to any, a closeness that is further reflected by the fact that in these contexts Dutch enig can be translated by both some and any.

28   

   Johan van der Auwera & Lauren Van Alsenoy

miner enig and any can often be paraphrased as ‘even one’, as in (15), and the uniqueness in (25) and the derived sense of excellence in (24) (‘excellent’ because ‘unique’) is a dimension of ‘oneness’ too. Of course, the adjective senses do not find a place on the Haspelmath map. So we may need to extend it or revise it in a more drastic way, which is what we do in the next section. Let us now look at German einig. The first thing to note is that Haspelmath (1997) does not discuss einig at all, for the same reason that he does not discuss Dutch enig ‘some’: both would classify as what he calls ‘mid-scalar quantifiers’. Thus German has an einig ‘some’ use, similar to the Dutch enig ‘some’ use, as illustrated in (26) to (28), which are the translations of (16) to (18). (26)

Ich habe einige Bücher mitgebracht, Lucky Jim, Requiem I have some books brought.along Lucky Jim, Requiem und L’Étranger. en L’Étranger ‘I have brought along some books, Lucky Jim, Requiem en L’Étranger.’

(27)

Ich habe einige Bücher mitgebracht, aber ich weiß nicht, welche. I have some books brought.along but I know not which ‘I have brought along some books, but I don’t know which ones.’

(28)

Gib mir einige Bücher. give me some books ‘Give me some books.’

Einig ‘some’ also occurs in questions, conditionals, and in indirect negation – examples simplified from web pages. (29)

Hast du auch einige schöne Zitate oder have you also some beautiful quotations or ein Lebensmotto? a motto.for.life ‘Do you also have some beautiful quotations or a motto for life?’ (cp. http://de.toluna.com/opinions/897448/Hast-auch-einige-schoneZitate-oder-Lebensmotto.htm, June 28 2011)

Mapping the West Germanic any’s   

   29

(30)

Wäre schön, wenn du einige Eigenschaften wie Treue were nice if you some characteristics such.as faithfulness und Ehrlichkeit hast. and honesty have ‘It would be nice if you have some characteristics such as faithfulness and honesty.’ (cp. http://www.kalaydo.de/anzeigen/kleinanzeigen/gemeinschaft/ersucht-sie/hallo-waere-schoen-wenn-du-einige-eigenschaften-wie-treueehrlichkeit-nachsichtigkeit-hast-u.-natue./anzeige/?kalaydonr=17496123, June 28 2011)

(31)

Aber wir wollen natürlich nicht, dass einige unserer but we want naturally not that some of.our alten Lieblinge völlig aus dem Sichtfeld verschwinden. old loved.ones fully out the view disappear ‘But we do not want of course that some of our old favorites disappear completely.’ (cp. http://eu.leagueoflegends.com/de/news/einige-klassischechampions-wurden-eben-noch-erschwinglicher, June 28 2011)

German does not, however, have an einig that corresponds to the English any. In questions, for example, einig only corresponds to some and not any. (32) * Hat er einigen Erfolg gehabt mit all seinen Bemühungen? has he any success had with all his efforts ‘Has he had any success with all his attempts?’ Figure 4 maps einig ‘some’ on the Haspelmath map.

Figure 4: German einig on Haspelmath’s 1997 indefiniteness map

Note that the German einig map in Figure 4 is similar to the Dutch enig map in Figure 3. The only difference would seem to be that Dutch enig reaches into direct

30   

   Johan van der Auwera & Lauren Van Alsenoy

negation and comparison and that German einig does not. However, the similarity is deceptive, for German differs from Dutch in not allowing an ‘any’ reading. But this is invisible on the map. The question, conditional and indirect negation cells that allow an ‘any’ use in Dutch also allow a ‘some’ use and both in Dutch and German. Put differently, the question, conditional, and indirect negation cells are shaded on both the Dutch and the German maps; this is because of the ‘some uses’, and the representational technique cannot show that the shading for Dutch has a double reason: they house both enig ‘some’ and enig ‘any’. Then German einig has an adjectival use. (33)

Experten sind einig, dass die Zinsen in der experts are in.agreement that the interests in the Zukunft steigen werden future raise will ‘Experts agree that interest will go up in the future.’

This adjectival einig sense is different from those of Dutch, but it is also related to ‘oneness’. To be in agreement means that to be of one opinion. To conclude, even for just the determiner (and pronominal) uses the actual Haspelmath maps are arguably insufficient, most clearly so for Dutch and German. The technique has an important drawback too: when a function is accessible to a strategy and there are different senses (e.g. Dutch questions allow both enig ‘some’ and enig ‘any’), this double motivation or polysemy cannot be shown on the map. Finally, both Dutch and German have adjectival einig uses, which seem to be related to the determiner uses.

3 Alternative mapping For various reasons, including ones that relate to the treatment of any, linguists have proposed two types of changes to the Haspelmath map. One approach keeps the geometry proposed by Haspelmath intact but adds uses. The other approach does not keep the Haspelmath geometry intact, at least not entirely. This distinction depends on the purpose of the study. Aguilar-Guevara et al (in print) exemplifies the first approach and their new map is enriched with a cell that is dedicated to the ‘anti-depreciative’ use illustrated in (10). Van der Auwera & Van Alsenoy (2011) exemplify the second approach: their point is to take a skeleton version of the Haspelmath map and to show how it relates to the Square of Oppositions, also known as the ‘Aristotelian’ Square. In this paper the alternative map will be

Mapping the West Germanic any’s   

   31

a mixture. On the one hand, it will use a skeleton version of the Haspelmath map. On the other hand, the interpretation of the cells is a bit different and the skeleton is also extended to cover the adjectival uses. The part of the map that resembles Haspelmath’s has four cells. ‘Specific’ covers both specific known and unknown but also those uses that allow English some, Dutch enig ‘some’ and German einig ‘some’. Consider (29) again. (29)

Hast du auch einige schöne Zitate oder ein Lebensmotto? have you also some beautiful quotations or a motto.for.life ‘Do you also have some beautiful quotations or a motto for life?’

We claim that German einig and English some are by themselves specific. (29) asks whether the hearer has some specific quotations. Of course, the question context has a ‘despecifying’ effect, which explains why English Do you also have some beautiful quotations? is very close to Do you also have any beautiful quotations? – and why Dutch enig can be translated by both some and any.8 For the determiner or pronominal English any and Dutch enig ‘any’, we allow three cells, all of them called ‘non-specific’: free choice (including its negation, illustrated in (10)), direction negation, and everything else, called ‘non-specific negative polarity’. We do not deny that the finer distinctions embodied in the Haspelmath map are important, only that the scope of the study does not allow us to go into that level of granularity. Then, from another point of view our alternative map shows more details than the Haspelmath map, for we cover the adjectival uses. The new map is shown in Figure 5.

Figure 5: An alternative map

8 The some and any questions are very close, but not identical. See Haspelmath (1997: 82–86), also for further references.

32   

   Johan van der Auwera & Lauren Van Alsenoy

We will use this map for the description of the synchrony and diachrony of any and its Dutch and German counterparts and it will be vindicated by showing how every any word covers or covered a contiguous stretch and develops incrementally, without hopping over a stage.9 The facts of present-day English any are simple. In terms of the categories of the new map any has three functions, all non-specific: (i) negative polarity, as in (4b), (5b), (6), and (7), direct negation as in (9a), and free choice as in (8).

Figure 6: English any

Each of the three uses was also available to the Old English ancestor of any¸ viz. ænig, as illustrated in (34) to (36). (34)

þæt ic me ænigne under swegles begong gesacan ne tealde. that I me any under sky region enemy not counted ‘so that I didn’t count any enemy under heaven’s expanse’ (Beowulf 1772–1773)

(35)

Nu ic suna minum syllan wolde now I son mine give would guðgewædu, þær me gifeðe swa war garments if me granted so ænig yrfeweard æfter wurde any heir after became ‘I would give my son the war garments, if any heir was granted to me then’ (Beowulf 2729–2731)

9 This gives the semantic map a diachronic interpretation, which is highly desirable if indeed ‘the best synchronic semantic map is a diachronic one’ (van der Auwera 2008: 43).

Mapping the West Germanic any’s   

(36)

   33

forþy bið ænig þing full þe sum bið wana & therefore is any thing full that some is lacking & forðy bið ænig þing wana ðe sum bið full; therefore is any thing empty which some is full ‘Therefore anything is full which is empty to some and therefore anything is empty which is full to some’ (HC: second subcorpus (OE2) Alfred’s Boethius 83)

In a corpus study of the Helsinki corpus Van Alsenoy (2011) shows that the negative polarity uses are the most frequent ones, which is not surprising, for the label covers various uses (question, conditional, etc.). The direct negation use, illustrated in (34), must have been well entrenched too, for it developed the univerbation nænig ‘not any’. (37)

Nænig heora þohte þæt he þanon scolde not.any of.them thought that he thence should eft eardlufan æfre gesecean again home.love ever seek ‘None of them thought that they would ever go to their beloved home again.’ (Beowulf 691–692)

The free choice use, as exemplified in (36), is rare.10 Besides the determiner and pronoun ænig there was also an adjective anga.11 Bosworth & Toller (1898, lemma anga) mention two meanings: (i) one and no

10 In the Helsinki corpus, free choice any is only attested twice, as compared to 293 attestations of the non-specific negation use and 108 attestations of the non-specific negative polarity uses. Hofer (2005:27) notes that “ænig, when occurring in assertive contexts, seems very close to present-day free choice any and its meaning”; however, of the four examples mentioned, only one is a convincing case. The rarity of Old English free choice ænig contrasts with its current high frequency. In a random selection of 200 written attestations in the written part of the British National Corpus, roughly one third are free choice uses. 11 They are both formed on the numeral ‘one’ but the suffixes are different. One is a very general derivational suffix and the other is a negatively polar indefiniteness marker. Surprisingly, it is the determiner/pronoun that uses the general suffix and it is the adjective that uses the negatively polar indefiniteness marker (EWN, lemma enig; cp. also Lehmann 1986: 194 – and Gonda 1975: 32 for more distant parallels). We will argue that the determiner/pronoun and the adjective got confused in the histories of English, Dutch, and German. It seems that there was confusion even at their common Proto stage.

34   

   Johan van der Auwera & Lauren Van Alsenoy

more, only, sole, singular and (ii) any, every one, all. The first use is of course similar to the adjectival use of Dutch enig, illustrated in (25). (38)

ond ða Iofore forgeaf angan dohtor and then Iofor gave only daughter ‘and then gave his only daughter to Eofor’ (Beowulf 2997)

The second use may have been rare. At least there is no attestation in the Helsinki corpus. (39) is the Bosworth & Toller example, and it only illustrates the ‘any’ gloss. (39)

hwæðer whether ðara of.them ‘whether them?’

ðu æfre gehyrdest, ðæt wisdom angum you ever heard that wisdom single eallunga þurhwunode altogether remained you have ever heard that wisdom always remained in any of

The any translation is that of Bosworth & Toller’s, but we have purposely kept ‘single’ as a gloss. We have argued before, on account of Dutch enig, that the meaning of singularity and that of any are close. We think that we see this reflected in Bosworth & Toller’s any use of anga too. If we gloss (39) as ‘whether you have ever heard that wisdom remained in even just a single one of them’ we arrive at the any reading. This further suggests, we propose, that the adjective and the determiner sense were confused – we will argue the same and with more evidence, for Dutch and German – and that the adjective could have been encroaching on the terrain of the determiner. The Old English situation, with its two players, ænig and anga, is shown in the maps in Figure 7. For anga, the only non-specific use for which we have an attestation is the negatively polar one, illustrated in (39), but there could, of course, have been a wider usage.

ænig Figure 7: Old English ænig and anga

anga

Mapping the West Germanic any’s   

   35

What happened later is that the anga word disappeared completely and left its sense to only (Old English ænlic, used from Old English onwards in the sense of ‘only’, ‘single’, and also ‘wonderful’), unique (ca. 1600 from French unique) and single (first attestation in 1303, from Old French sengle). Modern Dutch enig is mapped in Figure 8. For Old Dutch the map is identical except for the ‘wonderful’ use, which does not appear before the 20th century (EWN lemma enig).

Figure 8: Dutch enig

Compared to English, the semantic range of enig is very wide. To start from the top, enig has the adjectival use ‘only’ (and ‘wonderful’). In this connection, dictionary makers point out that Dutch enig goes back to two different words, (i) the ‘unique’ word, the Old English version of which is anga, and (ii) the ‘any’ word, which appears as ænig in Old English (EWN, lemma enig). In Dutch the two words got confounded even before the earliest written sources. The forms fell together and the one remaining form kept both the ‘unique’ and the ‘any’ sense. This confusion is similar to the one we assumed for English, but also a little different. First, for English we hypothesized that the confusion happened only at the semantic level, with the ‘unique’ sense allowing an ‘any’ reading. And second, in English the one surviving form, viz. any, only continued the ‘any’ sense, not both the ‘any’ and the ‘unique’, as with Dutch enig. The second aspect of the heavy semantic load of Dutch enig is the specific use, as in (16) to (19). The details of the genesis of this remain to be investigated, but it is striking that it seems to be around already in an example dated to 1100 (ONW, lemma enigII).

36   

(40)

   Johan van der Auwera & Lauren Van Alsenoy

Sin eineghe in thero ucatione gentium, thie are some in the calling heathen that there êr thes waran clari ingenio … there earlier this were clear mind ‘There are some in the calling of the heathens that earlier had a clear understanding of this …’

(41) is a later example (dated 1285), with a singular count noun (EWN, lemma enighII). (41)

Die vrouwe dat suaer verdroech. Ende vloecte de dieue that woman that heavy tolerated and swore the thieves dicke ghenoech. Omdat so gherne adde verdient. Dat often enough because so readily had deserved that enech viant ofte vrient. Haer dat lijf adde ghenomen. some fiend or friend her that body had taken. ‘The lady didn’t tolerate it and cursed the thieves often enough because she would have readily deserved that some friend or foe had taken that body.’

The enig ‘some’ examples are very rare, which makes sense if one assumes this to be a later development. This is a sensible assumption, given that the specific use is not attested in Old English or in Old German. The very idea of a negative polarity item shedding its negative polarity to become neutral or positive is uncontroversial: we need it for e.g. English need (van der Auwera & Taeymans 2009) and for German jemand (Jäger 2007, Willis 2012). Also, as we will see, German einig developed a specific use, too, but later than Dutch. The story of Dutch enig is not quite complete yet. At least from the 18th century (WNT, lemma enig) onwards, enig in the sense of ‘unique’ began to allow a superlative ending, yielding what would correspond to ‘onliest’. It is a use frowned upon by prescriptive grammars or tolerated only in informal spoken language (http://taaladvies.net/taal/advies/vraag/262/, 1 July 2011). (42)

Het was de enigste mogelijkheid. it was the onliest possibility ‘It was the only possibility.’

The form is robust though, and not isolated. There is also the even stronger form allerenigst ‘onliest of all’, the latter also in the ‘wonderful sense’.

Mapping the West Germanic any’s   

   37

(43)

Van de 6 miljard mensen op deze planeet zijn jullie 2 of the six billion people on this planet are you two wel de allerenigste die er zo over denken. really the all.onliest who it like.that about think. ‘Of the 6 billion of people on this planet, you two are the only ones who think about it like that.’ (http://www.veeteeltforum.nl/viewtopic.php?f=13&t=2804&start=45, 1 July 2011)

(44)

Absoluut, ik vind kinderen allerenigst en zij absolutely, I find kids all.onliest and they zijn ook dol op mij. are also crazy on me. ‘Absolutely, I find children wonderful and they are also crazy about me.’ http://marleeninbloomington.weblog.nl/marleeninbloomington/2006/ 03/t_schrijven_van.html, July 1 2011)

Yet another form is allerenig ‘only of all’, attested in the ‘only’ sense. (45)

Het allerenige verschil tussen Zeebrugge en the all.only difference between Zeebrugge and Pyeongtaek is het feit dat deze laatste haven een Pyeongtaek is the fact that this last port a belangrijk centrum is van zware industrie important centre is of heavy industry. ‘The only difference between Zeebrugge and Pyeongtaek is the fact that this last port is an important industrial centre.’ (http://www.apzi.be/news/detail_nl.phtml?id=401, July 1 2011)

The Maps in Figure 9 show the use of enigst, allerenigst and allerenigst.

enigst/allerenig Figure 9: Dutch enigst, allerenig, and allerenigst

allerenigst

38   

   Johan van der Auwera & Lauren Van Alsenoy

The emergence of all these forms can to some extent be interpreted as the Dutch attempt to get rid of the adjective vs. determiner polysemy. In English the attempt was completely successful: the determiner and the adjective now have different forms, any vs. only (and other words). Dutch kept enig both as an adjective and as a determiner: Dutch did develop the unambiguously adjectival forms, enigst, allerenig, and allerenigst, but they didn’t replace the enig adjective. Let us now turn to German. Present-day German is shown in Figure 10 on the left and we add Old High German on the right.

 

Modern German einig

Old High German einîg

Figure 10: Modern German and Old High German einîg

In Old High German, einîg was used in direct negation, indirect negation, questions and conditionals. (46)

Uns nist erloubit zi slahanne einingan us not.is permitted to slaughter anyone ‘We are not allowed to slaughter anyone.’ (Fobbe 2004: 143)

(47)

dei er nalles ano that before not without forahtun piheialt, anoo einikeru arbeiti fearing fulfills without any work ‘he will do what he didn’t do without fear before, without any effort’ (Kali Korpus online, http://www.kali.uni-hannover.de/)

(48)

Enoni ening fon then heriston giloubta in PARTICLE any of the lords believed in Inan odo fon pharisaeis him or of Pharisees ‘and did any of the lords or the Pharisees believe in him?’ (Jäger 2008: 196)

Mapping the West Germanic any’s   

(49)

   39

Ibu einic er dem ist arhaban fora if any of them is proud for kiuuizidu deru sineru listi … knowledge of his art… ‘But if anyone of them should grow proud by reason of his art, …’ (Kali Korpus online, http://www.kali.uni-hannover.de/)

The same Old High German einîg was also used as an adjective meaning ‘single’, ‘only’, at least in some texts. Jäger (2008) studied three relatively long texts, all of them 9th century (see Jäger 2008: 5–6). According to Jäger (2008: 196) einîg had both ‘any’ and ‘single’ uses in the Tatian text, pace Fobbe (2004: 141). The Otfrid text, however, used einîg only with the ‘single’ sense, and the Isidor text only with the ‘any’ sense (Jäger 2008: 195). This suggests that the ‘single’ and ‘any’ uses got confused (compare Jäger 2008: 194), which is also what happened in English and Dutch. What happened in German afterwards is that the ‘any’ use disappeared, with rare attestations in the 18th century (Fobbe 2004: 255–256, Jäger 2008: 259) and even a 19th century one, from the pen of Goethe. (50)

… dasz sie weder an leibe noch vermögen einige kränkung that they neither ar body nor property any loss erdulden sollen. suffer shall. ‘… that they shall not suffer any loss concerning their body or property.’ (Jäger 2008: 260, based on Grimm & Grimm 1862: 210)

At some point, perhaps as late as the beginning of the New High German period, einig developed a ‘some’ sense to refer to an indefinite but specific number of referents. According to Götze (ed) (1940: 155) this use was well-established in the 17th century. (51)

Es sind aber einige unter euch, welche nicht glauben. there are but some among them, who not believe. ‘There are, however, some among them who don’t believe.’ (Fobbe 2004: 217)

We assume that it developed out of the ‘any’ sense, just like Dutch enig. It would appear (Jäger 2008: 259–260) that the ‘any’ sense had nearly disappeared before ‘some’ took off. But presumably the continuity is real, a presumption more plausible than the assumption that ‘any’ had totally disappeared before the start of

40   

   Johan van der Auwera & Lauren Van Alsenoy

‘some’ – a view that is explicit in the Benecke et al (1854) lemma of the einic use of the term ‘return’: the ‘any’ use would ‘return’ as a ‘some’ use. Meanwhile the adjectival sense of ‘only’ continued and we assume that it gave rise to a new adjectival sense, viz. ‘in agreement’. (52)

Darüber sind sie about that are they ‘They agree about that.’

sich einig. themselves in.agreement

According to Grimm & Grimm (1862: 207) this use is first attested in early New High German (see also Jäger 2008: 194). Just like in Dutch, the ‘only’ sense had competition, initially not from a superlative, which we see in Dutch, but from a form einzig, which had been around since Middle High German (Kluge 1989: 171) and which replaced einig ‘only’ in the 18th century (Behaghel 1923: 421). Interestingly, in the 17th century example in (54), we see that even the new ‘only’ word was liable to be interpreted as ‘any’.12 (53)

Das ist das einzige was That is the only what ‘That is the only thing I need.’

ich brauche. I need

12 A whole set of issues for further study relate to the observation that whereas in positive contexts German einzig corresponds to Dutch enig or enigst, in negative contexts einzig corresponds to yet another ‘one’ based word in Dutch, viz. enkel – see Hoeksema (2005) for a study on the negative enkel construction. (a) German Dutch

kein einziges Buch geen enkel boek no single book ‘not a single book’

With the numeral ‘one’ Dutch enkel also has a positive use, and then German comes up with another ‘one’ word, viz. einzelner. (b) Dutch German

één enkel boek ein einzelnes Buch a single book ‘one single book’

And in the plural, enkele means ‘some’. (c)

Dutch German

enkele boeken einige Bücher some books

Mapping the West Germanic any’s   

(54)

   41

Ich darf mich gar nicht kehren an einzig Ungemach. I may me at.all not turn to single inconvenience. ‘I may not turn to a single inconvenience.’ (Paul Gerhardt 1607–1676 cited in Behaghel 1923: 421)

And meanwhile einzig also spawned the prescriptively rejected allereinzig and allereinzigst, both meaning ‘only’. Once again, we see a language getting rid of the adjective vs. determiner polyfunctionality, but the outcome is a little different from what we have seen in English and Dutch. Like in English, there is no one any word anymore that has both the adjectival ‘single’ use and the determiner ‘any’ use. Both German and English only kept a determiner use, albeit a different one, ‘any’ in English and ‘some’ in German.13 Like Dutch, German developed an adjectival sense different from ‘single’, i.e., ‘wonderful’ in Dutch and ‘in agreement’ for German; in both languages, these senses are still around. Interestingly, in modern German the ‘some’ and ‘in agreement’ senses are not adjacent. Normally, the uses shown on a semantic map have to be adjacent. ‘Normally’, however, is a hedge allowing for exceptions, and one exception is that the ancestor(s) adjacent to one or more non-adjacent uses got lost (see van der Auwera & Temürcü 2006), and this is what happened in German. It is also interesting that both the specific determiner and the adjective use are polarity neutral. In Dutch and German we find both uses and in English none. One wonders whether the developments of the two polarity neutral senses could have reinforced one another. On the remaining West Germanic languages, we will be brief. Old West Frisian, first of all, had an ‘any’ determiner, spelled in various ways, aeng, eng, enig, and enich (Hoekstra & Siebinga 2008: 153). According to Hoekstra & Siebinga (2008) only negative polarity uses are attested. It is intriguing that there are no attestations for direct negation, but that may be due to the small size of the corpus.14 The enich form also had the adjectival ‘single’ use (Hoffmann et al. 2008, lemma ēnich). For Modern West Frisian we have the Wurdboek fan de Fryske Taal (1984: 206–207) listing ienich with the adjectival uses ‘only, unique’, the German style

13 Dutch is in between English and German, in that it kept the determiner both in the ‘some’ and in the ‘any’ sense. This may further be put into the wider context in which Dutch grammar is in a non-trivial sense ‘in between’ those of English and German (see Van Haeringen 1956; Hüning et al (eds.) 2007, Vismans et al (eds.) 2010, van der Auwera & Noël 2011). 14 Hoekstra and Siebinga (2008) found comparative uses and consider them as negative polarity, just like Haspelmath (1997). But there is reason to doubt this and consider comparative uses a subtype of free choice (van der Auwera & Van Alsenoy 2011: 5–6). If that is correct, then Old Frisian has to be credited with a free choice use.

42   

   Johan van der Auwera & Lauren Van Alsenoy

‘in agreement’ use and also a use ‘lonely’, possibly unique to West Frisian. For the determiner use both ‘some’ and ‘any’ uses are mentioned. For the adjectival ‘unique’ use West Frisian also has the superlative ienichst. The resulting picture is that Modern West Frisian is very much like Dutch and one wonders whether contact influence has not played a role here. Contact influence is definitely a factor in the study of the any words in Afrikaans and Pennsylvania Dutch. Both have an any word that functions exactly like English any. So, different from their closest sister languages, Dutch and German, respectively, Afrikaans and Pennsylvania Dutch have a free choice any word (Afrikaans enig and Pennsylvania German ennich). For Afrikaans enig the suspicion that this use is due to contact from English has been on record since at least Mansvelt (1884: 41) and is supported by Donaldson (1989: 232) and Hoeksema (2010a: 843). The Pennsylvania Dutch ennich, Mark Louden (p.c.) informs us, is borrowed from English (and adapted to Pennsylvania Dutch phonology). Like English also, neither Afrikaans nor Pennsylvania German have a positive ‘some’ use, and like all West Germanic languages considered in this paper, Afrikaans and Pennsylvania Dutch have an unambiguously adjectival form. Afrikaans uses the superlative enigst ‘only’. Pennsylvania Dutch has both the einzig-like form eensich, but more commonly (Mark Louden p.c.), this form appears in a superlative and comes out as eensichscht. Yiddish, finally, does not have any determiner cognate with German einig, but it does have einikh ‘in agreement’ and eintsikh ‘only’. Whether these result from indigenous developments or are taken from German is unclear. Interestingly, Yiddish uses eyner ‘one’ together with an indefinite article as a specific determiner, and it has borrowed free choice determiners and pronouns from Slavic (see van der Auwera & Gybels in print for details and references). (55)

Eyner an alter shadkhn hot gehat a talmid. some a old marriage.broker had had a pupil ‘Some old marriage broker once had a pupil.’

(56)

Haynt tsu tog starbt nisht abi ver. now to day dies not no.matter who ‘These days not just anybody dies.’

Mapping the West Germanic any’s   

   43

4 Conclusion In this paper we put the study of present-day English any in a wider perspective, both historically and cross-linguistically. We sketched (i) how in English, German and Dutch a non-specific any determiner and pronoun got confounded, formally and/or semantically, with an adjective meaning ‘single, only’, (ii) how each language ‘solved this problem’, either completely (English) or partially (Dutch and German), each time with spawning new adjectives, and (iii) how in Dutch and German the determiner and pronoun developed a specific use. These changes are all well-behaved, as projected on a semantic map similar to, yet also different from the one proposed by Haspelmath (1997). The descriptions remain sketchy and need to be supported by detailed diachronic corpus work. Even more sketchy and in need of further study are the comparative notes on West Frisian, Afrikaans, Pennsylvania Dutch, and Yiddish.

5 References Aguilar Guevara, Ana, Maria Aloni, Angelika Port, Radek Simik, Machteld de Vos and Hedde Zeijlstra. In print. Semantics and pragmatics of indefinites: methodology for a synchronic and diachronic corpus study. In: Beyond Semantics – Corpus-based investigations of pragmatic and discourse phenomena, http://www.linguistics.ruhr-uni-bochum.de/bla/ Behaghel, Otto. 1923. Deutsche Syntax. Eine geschichtliche Darstellung. Band I. Die Wortklassen und Wortformen. Heidelberg: Carl Winter Verlag. Benecke, G.F. et.al. 1854. Mittelhochdeutsches Wörterbuch. Leipzig: Hirzel. Bosworth, Joseph and T. Northcote Toller. 1898. An Anglo-Saxon Dictionary. Oxford: Clarendon Press. Croft, William and Keith T Poole. 2008. Inferring universals from grammatical variation: multidimensional scaling for typological analysis. Theoretical Linguistics 34: 1–37. Donaldson, Bruce C. 1988. The influence of English on Afrikaans: a case study of linguistic change in a language contact situation. Pretoria: Serva. Fobbe, Eilika. 2004. Die Indefinitpronomina des Deutschen. Aspekte ihrer Verwendung und ihrer historischen Entwicklung. Heidelberg: Winter. Giannakidou, Anastasia. 2010. The dynamics of change in Dutch enig: from nonveridicality to strong negative polarity. Natural Language and Linguistic Theory 28: 861–875. Götze, Alfred (ed.). 1940. Trübners Deutsches Wörterbuch. Berlin: Walter de Gruyter. Gonda, Jan. 1975. The use of the particle ca. In: Jan Gonda, Selected Studies. Volume 1. Indo-European linguistics, 320–392. Leiden: Brill. Grimm, Jacob and Wilhelm Grimm. 1862. Deutsches Wörterbuch. Leipzig: Verlag von S. Hirzel. Haspelmath, Martin. 1997. Indefinite pronouns. Oxford: Oxford University Press. HC = Rissanen, Matti, Ossi Ihalainen and Merja Kytö (compilers). 1991. Helsinki Corpus of English Texts (diachronic part). Norwegian Computing Centre for the Humanities and the Oxford Text Archive. Helsinki: University of Helsinki.

44   

   Johan van der Auwera & Lauren Van Alsenoy

Hoeksema, Jack. 2005. Geen enkel. Tabu 34: 171–186. Hoeksema, Jack. 2007. Dutch ENIG: From nonveridicality to downward entailment. In: Zeijlstra & Soehn, 8–15. [Revised version as 2010a, Natural Language and Linguistic Theory 28: 837–859.] Hoeksema, Jack. 2010. Negative and positive polarity items: an investigation of the interplay of lexical meaning and global conditions of expression. In: Laurence R Horn (ed.), The expression of negation, 187–224. Berlin: Mouton. Hoeksema, Jack and Henny Klein. 1995. Negative predicates and their arguments, Linguistic Analysis 25: 146–180. Hoekstra, Eric and M. Sjoerd Siebinga. 2008. Exploring the distribution of aeng, the old Frisian counterpart of Modern English any, Modern Dutch enig. In: Bremmer, Rolf H. Jr., Stephen Laker and Oebele Vries (eds.). Advances in Old Frisian Philology, 153–162. Amsterdam: Rodopi. Hofer, Maria. 2005. The development of any with a focus on non-assertive any and its prehistory. Vienna: Unpublished MA-Thesis. Hofmann, Dietrich, Anne Tjerk Popkema and Gisela Hofmann. 2008. Altfriesisches Handwörterbuch. Heidelberg: Winter. Hüning, Matthias, Ulrike Vogl, Ton van der Wouden, and Arie Verhagen (eds.). 2006. Nederlands tussen Duits en Engels: Handelingen van de workshop op 30 september en 1 oktober aan de Freie Universität Berlin. Leiden: Stichting Neerlandistiek Leiden. Jäger, Agnes. 2007. On the diachrony of polarity types of indefinites In: Zeijlstra and Soehn 79–84. [Revised version as 2010. Anything is nothing is something, Natural Language and Linguistic Theory 28: 787–822.] Jäger, Agnes. 2008. History of German Negation. Amsterdam: Benjamins. Kluge, Friedrich. 1989. Etymologisches Wörterbuch. 22. Auflage. Berlin, New York: Walter de Gruyter. Ladusaw, William A. 1979. Polarity sensitivity as inherent scope relations. Doctoral dissertation, University of Texas at Austin. [Also 1980. New York: Garland Press] Larrivée, Pierre and Richard Ingham (eds.). 2012. The evolution of negation: Beyond the Jespersen cycle. Berlin: de Gruyter Mouton. Lehmann, Winfred P. 1996. A Gothic etymological dictionary. Leiden: Brill. Mansvelt, N. 1884. Proeve van een Kaapsch-Hollandsch idioticon met toelichtingen over land, volk en taal. Cape Town. http://www.dbnl.org/tekst/merw002vroe01_01/ merw002vroe01_01_0008.php Szabolsci, Anna. 2004. Positive polarity – negative polarity. Natural Language and Linguistic Theory 22: 409–452. Van Alsenoy, Lauren. 2011. The origin of any: a study of Old English any using the Helsinki Corpus of English Texts. Manuscript. van der Auwera, Johan. 2008. In defense of classical semantic maps. Theoretical Linguistics 34: 39–46. van der Auwera, Johan. In print. Semantic maps: for synchrony and diachrony. In: Anna Giacalone Ramat, Caterina Mauri and Pierra Molinelli (eds.), Synchrony and diachrony: a dynamic interface. Amsterdam: Benjamins. van der Auwera, Johan and Paul Gybels. In print. On negation, indefinites and negative indefinites in Yiddish. In Bjoern Hansen and Marion Aptrot (eds.), Yiddish Language Structures, 193–249. Berlin: De Gruyter Mouton.

Mapping the West Germanic any’s   

   45

van der Auwera, Johan and Dirk Noël. 2011. Raising: Dutch between English and German. Journal of Germanic Linguistics 23: 1–36. van der Auwera, Johan and Martine Taeymans. 2009. The need modals and their polarity. In: Rhonwen Bowen, Mats Mobärg and Sven Ohlander (eds.), Corpora and discourse – and stuff. Papers in honour of Karin Aijmer, 317–326. Gothenburg: University of Gothenburg. van der Auwera, Johan and Ceyhan Temürcü. 2006. Semantic maps in Typology, Encyclopedia of Language and Linguistics. (2nd ed.) ed. by Keith Brown. Oxford: Elsevier, 11: 131–134. van der Auwera, Johan and Lauren Van Alsenoy. 2011. Mapping indefiniteness: towards a Neo-Aristotelian approach. In: Eliza. Kitis, Nikolas Lavidas, Nina Topintzi and Tasos Tsangalidis (eds.), 1–14. Thessaloniki: Monochromia. van der Auwera, Johan and Lauren Van Alsenoy. 2012. Indefinite pronouns: synchrony and diachrony – comments on Willis. In: Larrivée and Ingham, 325–345. Van Haeringen, C[oenraad] B. 1956. Nederlands tussen Duits en Engels. Den Haag: Servire. Vismans, Roel, Matthias Hüning and Fred Weerman (eds.). 2010. Dutch between English and German. Special issue Journal of Germanic Linguistics 22, 4. Willis, David. 2012. Negative polarity and the quantifier cycle: Comparative diachronic perspectives from European languages. In: Larrivée and Ingham, 287–323. Wurdboek fan de Fryske Taal. 1984. Ljouwert : Fedde Dykstra. Zeijlstra, Hedde and Jan-Philipp. Soehn (eds.). 2007. Proceedings of the Workshop on Negation and Polarity. http://www.sfb441.uni-tuebingen.de/negpol/negpol07.pdf BNC – British National Corpus http://www.natcorp.ox.ac.uk/ EWN – Etymologisch Woordenboek van het Nederlands http://www.inl.nl/nl/woordenboeken/etymologisch-woordenboek-nederlands-(ewn) Kali corpus http://www.kali.uni-hannover.de/ ONC – Oudnederlands woordenboek http://www.inl.nl/nl/woordenboeken/oudnederlands-woordenboek-(onw) Taaladvies.net, question on enigste/enige, http://taaladvies.net/taal/advies/vraag/262/ July 1 2011. VMNW – Vroegmiddelnederlands woordenboek http://www.inl.nl/nl/woordenboeken/vroegmiddelnederlands-woordenboek-(ivmnw) WNT – Woordenboek der Nederlandse Taal http://www.inl.nl/nl/woordenboeken/woordenboek-der-nederlandsche-taal-(iwnt)

Jack Hoeksema

Polarity items in Strawsonian contexts – A comparison 1 Introduction1 The theory of negative polarity licensing put forward in Ladusaw (1979) is admirable for a number of reasons. It is a unified analysis, treating all polarity items alike, and reducing all licensing contexts to a single factor, namely downward entailment, a factor which, moreover, is independently motivated. In this respect, Ladusaw’s theory is unlike earlier proposals, such as the syntactic feature account of Klima (1964), which offers a unified account, but it is not based on anything that is motivated independently of the licensing of polarity items. It also differs from the theory offered by Linebarger (1980), which is independently motivated, but not unified.2 For this reason, the field of polarity studies has primarily tried to build on Ladusaw’s work in trying to make further progress, in an attempt to retain as much as possible the attractive features of that account. Among the works building on Ladusaw’s original theory, several deserve to be mentioned. In this introduction, I will limit myself to three strands of research, associated with the names of Frans Zwarts, Anastasia Giannakidou and Kai von Fintel.3 Zwarts, in a series of publications (Zwarts 1981, 1986, 1998, among others, see also van der Wouden 1997), has refined Ladusaw’s account by proposing a hierarchy of polarity items: weak items licensed in any downward entailing context, strong items, licensed in anti-additive contexts and superstrong items, licensed only by regular negation (so-called antimorphic contexts). This account is exactly like Ladusaw’s for the weak items, and makes use of semantically-defined, and independently motivated notions for the strong and superstrong items as well.

1 I would like to thank the audience at my presentation in Göttingen, and two anonymous readers, for their astute remarks and helpful comments. 2 In particular, Linebarger assumes syntactic licensing by negation (at LF) as well as pragmatic licensing by negative implicatures. 3 The work of Kadmon and Landman (1993) on English any and that of Chierchia (2004, 2006) has also been very influential. Chierchia’s proposals make heavy use of a new theory of scalar implicatures, and build on the earlier work of Kadmon and Landman. However, their scope extends far beyond the problem of polarity sensitivity, and has been extensively debated elsewhere (e.g. Homer 2011).

48   

   Jack Hoeksema

However, it is no longer unified in the sense of treating all polarity items alike, but it offers an attractive typology of polarity items, based on various logical properties of negation. The idea of there being (at least) a binary distinction between weak and strong polarity items has been widely followed, although there are various empirical problems with the classification proposed (cf. e.g. Hoeksema 2010b, Gajewski 2011, Giannakidou 2011 for recent discussion). Giannakidou’s best-known contribution to the study of polarity items consists in a proposal to generalize the Ladusaw account to environments that are not downward-entailing, but share with negation the important property of being nonveridical, added with a couple of additional licensing and antilicensing conditions for special cases (cf. also Zwarts 1995 for an early discussion of nonveridicality). Nonveridical contexts are those contexts in which the truth of a proposition cannot be guaranteed if the proposition appears within that context. For instance, the truth of the proposition it is raining is not guaranteed in the context Jones hopes ___. The same is clearly true for sentential negation, which is equivalent to the following context: It is not the case that ___. Put differently, if a statement of the form XYZ such that Y denotes a proposition does not entail Y, then the context X__Z is said to be nonveridical.4 Indefinite noun phrases and pronouns are nonreferential in such contexts. A number of examples have been identified of expressions that appear in all and only nonveridical contexts, such as Modern Greek kanenas ‘anybody/nobody’ (when it is unstressed, cf. Giannakidou 1997, 1998, 2011) and the early-modern Dutch indefinite determiner enig ‘any’ (at least in combinations with singular count nouns, cf. Hoeksema 2010a). Addi-

4 Instead of using the context of a polarity item as licensing its occurrence, one also finds proposals in the literature in terms of operators. In that case a nonveridical polarity item is licit in the scope of a nonveridical operator. See Homer (2012) for some discussion of empirical differences between the two approaches in cases of multiple operators (see also note 8 below). There are other cases where environments make better predictions than operators, such as: (i) His chances of ever winning the cup are slim/*excellent. It is neither the presence of chances nor that of slim that licenses the polarity item ever, but rather their combination, which turns the context into a downward entailing one. Note that (ii), for instance, entails (iii): (ii) His chances of winning the cup or coming in second are slim. (iii) His chances of winning the cup are slim. This entailment is not valid if we change slim into excellent. To see this, suppose that excellent chances are chances of 80% or higher, and that the person under discussion has a 40% chance of winning the cup and another 40% chance of coming in second. Given that the two options are mutually exclusive, there is a 80% chance of either winning the cup or coming in second (which is excellent), but only a 40% chance of winning the cup, which, according to our criterion, is not excellent.

Polarity items in Strawsonian contexts – A comparison   

   49

tional licensing (or anti-licensing) conditions are invoked to deal with expressions that have a narrower distribution than kanenas, such as emphatic kanenas or English any. Giannakidou’s work has been influential, but will not be discussed here at length. However, it is obviously both possible and of interest to study the various nonveridical environments, such as disjunction, the scope of modals, the complement of verbs like wish and hope, imperatives, and so on, in the manner adopted here, with a view of determining if and to what extent different polarity items distinguish among these environments. Slowly but surely, more items are being discovered that show the general general distributional characteristics of weak kanenas (e.g. embedded who the hell-clauses (Den Dikken & Giannakidou 2002) and the German idiom von Dummsdorf kommen ‘come from Stupidville’ which was recently discovered to be of this type, cf. Hoeksema and Sailer (2012)), making such a comparison worthwhile. For the time being, such an investigation of nonveridical contexts will have to be deferred to future work. Von Fintel’s work, like that of Zwarts and Giannakidou, uses Ladusaw’s theory as its main point of departure. Unlike the proposals by Ladusaw, Zwarts, and Giannakidou, which are all couched in terms of entailment relations, Von Fintel (1999) is primarily concerned with presuppositional contexts. Ladusaw’s core notion is that of downward entailment (a.k.a. implication reversal, cf. Fauconnier 1978). Consider for instance the following entailment: (1)

Jones is a sheep farmer → Jones is a farmer

(2)

Nobody here is a farmer → Nobody here is a sheep farmer

While normally having the more specific property of being a sheep farmer would entail having the more general property of being a farmer, as in (1) above, negative contexts reverse the direction of implication, as we see in (2). If we see the entailment in (1) in terms of going upwards, from a small set (the set of sheep farmers) to a larger set (that of farmers), then the entailment in (2) will have to be seen as involving going downwards, hence the name downward entailment. In the definition used by von Fintel (1999): (3)

A function f of type is downward entailing (DE) iff for all x, y of type σ such that x ⇒ y: f(y) ⇒ f(x).

In this definition, the arrow ⇒ indicates a generalized entailment relation between expressions of any boolean type (cf. Keenan and Faltz 1985 for an early statement of this notion).

50   

   Jack Hoeksema

As a number of people have noted, not all triggers of polarity items are clearly downward entailing. One core case involves restrictive adverbs such as only (cf. Linebarger 1987, Atlas 1993, 1996, Horn 1969, 1996 for discussion): (4)

Only Jones is a farmer -/-> Only Jones is a sheep farmer

As noted by von Fintel, the inference from left to right is just fine, provided we assume that Jones is a sheep farmer. Otherwise, we have presupposition failure. If we add the presupposition of the consequent as an additional premise, we get an unobjectionable type of inference: (5)

Only Jones is a farmer. Jones is a sheep farmer. ∴Only Jones is a sheep farmer.

This leads von Fintel to propose the following generalization of downward entailment, which he named after the philosopher Peter Strawson, who first put presuppositions on the map: (6)

Strawson Downward-Entailingness A function f of type is Strawson-DE iff for all x, y of type σ such that x ⇒ y and f (x) is defined : f(y) ⇒ f(x).

Superlatives are another prime example of a trigger that is not a downward-entailing operator (Hoeksema 1986a). However, they conform to the weaker requirements of Strawson downward-entailment: (7)

Jones is the richest farmer -/-> Jones is the richest sheep farmer

(8)

Jones is the richest farmer Jones is a sheep farmer [presupposition of conclusion] ∴Jones is the richest sheep farmer

A third context worth pointing out is that of factive verbs. These may serve as triggers for polarity items:

Polarity items in Strawsonian contexts – A comparison   

(9)

   51

a. Jones regrets ever having set foot on this miserable island. b. Frankly, I am amazed you would ever say such a thing!

Regular downward entailment is not valid: (10)

Jones regrets being a farmer -/-> Jones regrets being a sheep farmer

However, Strawson downward entailment holds:5 (11)

Jones regrets being a farmer Jones is a sheep farmer ∴Jones regrets being a sheep farmer

I will not discuss here the full list of contexts discussed in Von Fintel (1999). In particular, the discussion of conditionals in that paper will be left untouched, in light of the additional complexities involved in their analysis. I should also note that Strawson Downward Entailment does not deal with all contexts in which polarity items are licensed. In particular, questions remain a problem for this type of approach.6 Von Fintel’s approach to the treatment of polarity items in presuppositional contexts has been influential (see e.g. Rullmann 2003, Gajewski 2005, 2011, Sharvit and Herdan 2006, Guerzoni and Sharvit 2007, among others). However, an evaluation of the theory based on a broader set of polarity items than the

5 The facts are slightly more complicated than in the case of superlatives, however. See von Fintel (1999) for discussion. In particular, it is possible to say: Jones regrets being a farmer, but given that he is a farmer, he does not regret being a sheep farmer. This indicates that verbs such as regret (and prefer, etc.) are highly sensitive to context. The validity of (11) therefore depends on the context remaining fixed. 6 Note that the nonveridicality theory of Giannakidou (1998) has no problems with the appearance of polarity items in questions, since questions are by their very nature nonveridical. For a pragmatic approach, building on insights van Kadmon and Landman (1993) and Krifka (1995), see Van Rooy (2003). There are many complicating factors involved, including differences between rhetorical and nonrhetorical questions, between direct and indirect questions, and between yes/no and wh-questions. Note that nonveridicality in and of itself does not shed light on these finer-grained distinctions, and will have to be supplemented by additional requirements.

52   

   Jack Hoeksema

usual suspects any and ever, has not been carried out yet.7 Having sketched the approach to presuppositional contexts adopted by von Fintel, I will now proceed to a corpus-based comparison of these contexts. Using data from English, Dutch, and German, I will show that the various contexts discussed by von Fintel, which I refer to here as ‘Strawsonian’, do not necessarily line up neatly in their behavior as hosts of negative polarity items. My main concern is to show that polarity items are not a uniform class of expressions, and to pinpoint a number of interesting subclasses of NPIs. An additional goal of this paper is to look more closely at some of the environments with superlative-like characteristics, such as the first, the last, and the only. Although I claimed in Hoeksema (1986a) that they pattern with superlatives (see also Bhatt 2002, Herdan 2005 for similar ideas), I am now in a position to show that they differ from superlatives, and even from one another, in their ability to license some polarity items. In a similar vein, I will note differences in licensing between superlatives and the quantifier most, arguing against a recent proposal made in Gajewski (2010). I intend to show in this way that corpus data are not just useful for the classification of polarity items, but also for that of their host environments.

2 Any and ever A vast amount of the literature on polarity items is based on the distributional patterns exhibited by the most common English NPIs, any and ever. I will therefore start with a discussion of these two expressions. English any can be used both in a variety of polarity contexts, and as a free choice item in (mostly) modal and generic contexts. I will follow Linebarger (1980), Ladusaw (1979), Carlson (1980, 1981) and Dayal (1998), among others, in assuming that there are two uses of any: free choice any and polarity-sensitive any.8 My main reason for assuming two any’s is that indefinite NPIs do not automatically have this double interpretation (cf. Hoeksema 2010a for discussion of Dutch enig, a cognate of any, which is a polarity item, but lacks a free choice interpretation). Note also the ambiguity of questions such as the following:

7 The set of items considered in Gajewski (2011) is somewhat larger, but the binary classification proposed in that paper does not suffice to deal with the range of variation to be discussed in the present paper. 8 There is actually even a third any, so-called adverbial any, as in The weather is not getting any better. Adverbial any behaves like a regular polarity item, and does not have a free choice use.

Polarity items in Strawsonian contexts – A comparison   

(12)

   53

Would you kiss any girl?

Under the NPI reading, the adressee of this question is assumed to be shy about girls, while under the free choice reading, the opposite seems to be true. A variety of linguists have taken issue with this purported lexical ambiguity (Kadmon and Landman 1993, Lee and Horn 1994, Zwarts 1995, Giannakidou 2001, Horn 2005, inter alii). Lee and Horn (1994) for instance point out that any and no (which they see as the neg-incorporated counterpart of any, a la Klima 1964) can be modified by whatsoever: (13)

a. I don’t have any money whatsoever. b. Pick any card whatsoever.

What is interesting, according to Lee and Horn, is that both free choice and NPI any support the use of this modifier, and moreover, no other determiners do (apart from no): (14)

a. * I don’t have money whatsoever. b. * Pick a card whatsoever.

However, this particular argument is weakened by the possibility of whatsoever with a number of other universal and negative quantifiers (contra Lee and Horn 1994). The following examples were lifted from the Internet: (15)

a. The record is simply beautiful and inspiring in every way whatsoever. b. Machu Picchu is fascinating from every angle whatsoever. c. Design/Builder shall indemnify and hold Owner harmless from any and all claims, liability, damages, loss, cost and expense of every type whatsoever […]. d. Therefore all things whatsoever ye would that men should do to you, do ye even so to them: for this is the law and the prophets. (Matthew 7:12 KJV)

54   

(16)

   Jack Hoeksema

a. “There’s very little reason whatsoever for a New Yorker to go outside of New York State,” said Martin Mason, a broker with Joseph Gunnar & Company. b. It’s amazing how shitheads show themselves … They go on the attack for little reason whatsoever because you look like riffraff.

Ever also may be used in several ways: either as an NPI, or as a universal temporal adverb: (17)

a. I don’t think I’ll ever do that again. b. The rain fell ever faster.

The latter use is heavily restricted, showing up mainly in collations with comparatives, present participles (ever growing) and a few fixed expressions (e.g. for ever, as ever), cf. Israel (1998). Given these complications, it will be necessary to sift corpus data, pulling out the NPI uses of these items from among the other uses. Using material from books, newspapers and the Internet, I have hand-collected a set of occurrences of any and ever, and classified them according to their licensing environment. In cases where more than one licensing expression was available, I assumed that the active licensor was the one closest to the polarity item. For example, in the following sentence: (18)

If you don’t do anything, I will.

we are dealing with an occurrence of anything in a conditional sentence, in the scope of negation. Both conditional clauses and the scope of negation are fine contexts for polarity items, but negation is closer to anything than the conditional operator if. Consequently it will be classified as a case of triggering by negation.9 If you find this a dubious decision, bear in mind that most data points are unam-

9 Cases of double negation and other types of double licensing are of interest in their own right. They are well-known environments for positive polarity items as well (cf. Baker 1970, Szabolcsi 2004). In many cases, they still host negative polarity items, even though at the level of the entire sentence, the two negations cancel out and do not support downward entailing inferences (cf. Hoeksema 1986b). A forthcoming paper, Homer (2012), has identified cases where double negation contexts do not license polarity items, e.g. in cases where both licensing elements are in the same higher clause, and the polarity item in a subordinate clause, such as *I am not surprised he would ever do a thing like that, cf. I am surprised he would ever do a

Polarity items in Strawsonian contexts – A comparison   

   55

biguously classifiable, so a few potentially ambiguous cases won’t make a huge difference one way or another for the distributional patterns we will be considering below. In Table 1, I present the data I have collected for NPI any (including indefinite pronouns such as anything, anyone, anybody) and ever. Examples of adverbial any (see section 3 below) were not included, nor were cases of free choice any. In interpreting this table, and the ones that follow it, please bear in mind that some important baseline information is not available, such as the odds of something appearing in the scope of negation, or in the restriction of a superlative. Clearly, having such information would be extremely welcome, but for the time being, we’ll have to go without such data.10 It is very likely, that such odds

context

ever

%

any

%

superlative the only11 the first the last only XP negation question other

857 146 162 41 57 1337 1018 1842

16 2.6 3 0.8 1 24 19 34

58 49 33 19 28 3936 1248 2907

0.7 0.6 0.4 0.2 0.3 48 15 35

total

5460

100

8278

100

Table 1: Ever versus Any

thing like that. Using carefully defined licensing domains, Homer is able to predict both cases where two negations do not license polarity items, and cases where they do. 10 One of the reviewers suggests using Stefanowitsch’ (2008) notion of negative entrenchment. Using statistical methods, Stefanowitsch distinguishes between systematically absent and accidentally absent data. His study concerns a number of ditransitive verbs which may or may not appear in the double-object construction. For the relation between polarity items and their contexts, one may do something similar, although it should be noted that not all differences to be found are due to licensing behavior alone. Unlike ditransitive verbs, negative polarity items vary along a great many dimension, including syntactic category, semantic type, rhetorical value, etc. For this reason, I restrict statistical testing in the present article to cases where items are very similar. 11 Included in the category the only are a few instances of items in the restriction of the equivalent expression the one. Compare: the only man who ever loved her = the one man who ever loved her.

56   

   Jack Hoeksema

are highly text-type dependent, and so would have to be carefully calibrated for the mix of text types used for my data collection. However, there is one thing we can do, which is to compare various polarity items, and the result turns out to be interesting. What we may observe here is a striking difference among these two oftendiscussed items, a difference that is not absolute, which may be the reason it is not usually noted, but quite significant nonetheless. While ever is particularly common in superlative contexts, any is not. (Positive occurrences of ever as in he was ever the gentleman, were excluded from the tabulated data, as were occurrences of free choice any.) There is also a drop in frequency for the superlativelike categories the only, the first, the last, albeit a somewhat less severe drop.

3 Any and anymore While ever and any have by and large a rather similar distribution, showing differences primarily among superlative and related contexts, we start to see much bigger differences when we compare any with an altogether different type of polarity item, anymore. Although anymore is derived from any + more, it does not come from the determiner any, but from the related adverbial modifier, which is commonly found as a modifier of comparatives, as in the following pair of sentences:12 (19)

Could you be any more insulting?

(20)

I can’t sing any higher.

Neither anymore nor adverbial any shows up in the context of superlatives, cf.: (21) * Fred is the best teacher we have anymore.13 (22) * Fred is the best teacher who can dance any faster.

12 It may also be found in combination with too (He wasn’t any too friendly), some adjectives (Is it any good?) and a few verbs (That won’t hurt them any). 13 For those North-American or Irish speakers of English who use so-called positive anymore, this sentence should be acceptable.

   57

Polarity items in Strawsonian contexts – A comparison   

Corpus data confirm this assessment (the any data are the same as in Table 1):

context

anymore

%

any

%

adv any

%

superlative the only the first the last only XP negation question other

– 15 – – – 1029 25 58

– 1.3 – – – 91 2 5

58 49 33 19 28 3936 1248 2907

0.7 0.6 0.4 0.2 0.3 48 15 35

– 1 – – – 157 44 30

– 0.4 – – – 68 19 13

total

1127

100

8278

100

232

100

Table 2: Anymore versus (determiner/pronoun) any and adverbial any 14

Note that noun phrases headed by superlatives, the first or the last do not host anymore and adverbial any. Interestingly, though, we have to make an exception for the only, which is relatively well-attested. With one exception, all occurrences of anymore in this context are from the US, suggesting an Atlantic divide in usage. For this reason, I have divided the anymore data into two groups: US occurrences with the spelling anymore and British occurrences with the spelling any more.15 (Of course, cases where any more had a different meaning, as in (19) above, were not included in the data set.) In Table 3, these two groups are compared to German and Dutch data:

14 Adverbial any in the fixed expression any longer is not included here, because the high frequency of any longer would create a slanted picture. However, with regard to occurrences in the restriction of superlatives, the only, the first and the last, any longer patterns with adverbial any: among 128 occurrences, not a single one appeared in one of these contexts. A difference between any longer and adverbial any that can be noted is that any longer is far less common in questions. 15 The spelling difference anymore versus any more is not a very reliable test for distinguishing British from American data, and so all material was checked by hand as well as much as possible. In the case of newspapers and novels, attributing usage is easy, in the case of anonymous Internet postings, it is harder, and here I had to use spelling as the decisive criterion. However, I note that relatively little of the material comes from Internet sources.

58   

   Jack Hoeksema

context

anymore GB

meer

mehr

US

N

%

N

%

N

%

N

%

superlative the only the first the last 16 only XP negation question other

– 1 – – – 388 10 21

0.2 – – – 92 2 5

– 14 – – – 641 15 37

– 2 – – – 93 2 4

– – – – 15 2395 – 103

– – – – 0.6 95 – 4

– – – – – 713 – 7

– – – – – 99 – 1

total

420

100

707

100

2513

100

720

100

Table 3: Anymore (GB versus US), meer (Dutch), mehr (German)

The difference between the GB and the US data regarding the context the only is significant (a two-tailed Fisher’s exact test yielded p < 0.05). The comparison with Dutch meer and German mehr shows that British English is similar to the European Germanic languages, and that the US usage has diverged. It is likely that the presence of positive anymore dialects in the US, but not in Britain, has something to do with this Atlantic divide, but it should be noted that whereas positive anymore remains somewhat substandard (cf. Labov 1975), the use of anymore with the only appears to be wholly unremarkable, given attestations in such journals as the New York Times. Compare:

16 Note that Dutch is the only language of the three languages compared which allows a counterpart of anymore to combine with a restrictive adverb. In most cases, this involves a scalar use of only, in Dutch often expressed by maar ‘but’, as in the following, from André Demedts, De levenden en de doden, Davidsfonds, Leuven, z.j. [1st ed. 1959], p. 212: (i) Nu waren er maar twee middelen meer om de tijd te doden. Now were there but two means anymore for the time to kill ‘Now there were only two means for killing time’ Typical for this scalar use is the combination of maar ‘but’ or slechts ‘merely, just’ with a low numeral or a measure noun indicating a small quantity or amount. In Dutch, this usage is considered by some as substandard (however, the claim in Sassen 1977 that it is a dialectism from the northern Dutch area of Groningen has to be rejected, given examples like (i), from a Flemish novel), and an alternative construction, involving the non-polarity-item nog, is preferred: nog maar twee ‘yet but two’ is equivalent to maar twee meer ‘but two anymore’ (at least in most contexts – for more extensive discussion of the various factors involved, see Vandeweghe 1983). Differences between scalar and nonscalar uses of only-type adverbs have frequently been discussed in the literature (e.g. Jacobs 1983, Bayer 1996).

Polarity items in Strawsonian contexts – A comparison   

   59

(23)

The only reason I ever call someone anymore is if I don’t have their Twitter handle or e-mail address.17

(24)

Seeing their children was the only thing he seemed to care about anymore.18

There are more notable things in the table (e.g. the complete lack of questions as a host for the German and Dutch counterparts of anymore), but for the purposes of this paper, the main thing to note is the presence of a dissociation between the only on the one hand, and superlatives, the first, the last on the other in American English. If Strawson downward entailment were the only relevant property of the contexts to consider, we would expect all environments to either accept or resist the presence of anymore.

4 In weeks, months, years, ages Both English and Dutch (but not German) have a class of polarity items of the form in X, where X is a temporal noun indicating a contextually long stretch of time (hence they are maximizers, rather than minimizers, in the sense of Israel 1996). For some discussion of this class of items, see Hoeksema (2005). Their primary contexts of occurrence are negative sentences and comparative and superlative constructions. Some examples illustrating their use are: (25)

I haven’t seen Fred in weeks.

(26)

Fred had the best time in ages at the Zoo.

(27)

For the first time in years, Fred came to work in a suit.

Given that in ages etc. is acceptable in the restriction of a superlative, we might expect these items to also appear in the scope of only and other Strawson downward entailing contexts. However, sentences showing this type of combination of polarity item and environment are rather bad:

17 The New York Times, May 13, 2010. 18 Jonathan Franzen, The corrections, Farrar, Straus and Giroux, New York, 2001.

60   

   Jack Hoeksema

(28) * Only Fred has seen her in ages. (29) * Fred only wanted to kiss her in ages. (30) * Alleen Fred heeft haar in tijden gezien [Dutch] Only Fred has her in ages seen ‘Only Fred has seen her in ages’ Compare the above examples with sentences involving ever, another temporal polarity item:19 (31)

Only Fred has ever seen her.

(32)

Fred only ever wanted to kiss her.

Corpus data provide the picture in Table 4 below. Note that for the sake of brevity, all English items of the form in X have been bundled together, and similarly for Dutch.20 context

English

Dutch

N

%

N

%

superlative the only the first the last only XP negation question other

220 3 20 – – 250 – 32

42 0.6 4 – – 48 – 6

257 1 24 – – 692 – 84

24 0.1 2 – – 65 – 8

total

525

100

1058

100

Table 4: In X: English and Dutch data

19 Regarding ever in the restriction or focus of only, see Beaver and Clark 2003. 20 One difference between English and Dutch concerns the size of the category superlative. Some of this difference may be due to a slightly greater preference for comparatives in Dutch. Note that Seabiscuit is the fastest horse we have seen in years and Seabiscuit is faster than any horse we have seen in years are equivalent, and hence, for speakers and writers, competing options to choose from. Trade-off effects having to do with language-specific preferences for using either superlatives or comparatives may be noted with other polarity items as well (cf. Hoeksema 2008).

Polarity items in Strawsonian contexts – A comparison   

   61

Note once more that there is a dissociation between several types of context, tied together by Strawson downward entailment. While superlatives, as well as the only and the first provide acceptable contexts for polarity items of the type in X, restrictive adverbs such as only and questions do not.

5 Remotely – in de verste verte One of the lesser-known polarity items of English is the adverb remotely, as used in sentences such as:21 (33)

I am not remotely interested in your achievements.

(34)

Is an earthquake even remotely likely to happen here?

Of course, there is also another use, which is not polarity-sensitive in any way:

21 One of the reviewers is not convinced that remotely is a negative polarity item. I agree that it is not always that simple to distinguish among uses of words, in order to establish that one of them is polarity-sensitive, if other uses are sufficiently close in meaning. However, I hope the following paradigm will help make my point: (i) * That is remotely true. (ii) That is not remotely true. (iii) None of it is remotely true. (iv) * Some of it is remotely true. (v) If it is remotely true, I will eat my hat. (vi) Is that even remotely true? (vii) Anything that is remotely true, you should tell the police. (viii) The only thing that is remotely true is that we are understaffed. (ix) * Fred told me something that is remotely true. Note that in combination with true, the adverb remotely cannot be interpreted literally, but has a metaphorical sense as a degree modifier, indicating a minimal degree or extent. The reviewer offered as grammatical the following sentence: Someone who’s remotely interested in the matter could object. Indeed similar examples can be found by doing some googling on the Internet, e.g. (x) Someone who is remotely competent or sentient recognizes that 97% of the world’s scientists are right about climate change (http://my.firedoglake.com/cenkuygur/ 2011/10/28/why-republican-voters-cant-make-up-their-mind/, accessed on June 4, 2012) I take it that such cases are generic in nature, and differ from non-generic statements like (ix) and (xi) below in acceptability: (xi) ? Fred met someone who was remotely competent. The Oxford English Dictionary (s.v. remotely) notes as usage 3: “In the slightest degree, in any respect. Chiefly in negative contexts.”

62   

(35)

   Jack Hoeksema

The garage-door is controlled remotely from the car.

Dutch has an expression with a similar meaning, in de verste verte, meaning literally ‘in the most distant distance’: (36)

Ik ben in de verste verte niet tevreden I am in the furthest distance not satisfied ‘I am not even remotely satisfied / I am not satisfied by a long shot’

While both expressions may appear in the restriction of the only/de enige, as well as the first/de eerste, they may not appear with superlatives: (37)

Jones is the only teacher who is remotely interested in his students.

(38)

Jones is the first teacher who is even remotely qualified to teach.

(39) * Jones is the best teacher who is remotely interested in his students. (40) * Jones is the worst teacher who is remotely qualified to teach. (Similar judgments pertain to the Dutch counterparts.) Corpus data confirm our intuitive judgments: context

remotely

in de verste verte

N

%

N

%

superlative the only the first the last only XP negation question other

– 10 2 1 4 96 10 34

– 6 1 0.6 3 60 6 22

– 4 – – – 291 7 49

– 1 – – – 83 2 14

total

157

100

351

100

Table 5: Remotely and In de verste verte ‘in the most distant distance’

Given that superlatives are more common as a context than either the only or the first, as the data for any and ever clearly show, we must consider the fact that there are plenty of attestations with the latter class of contexts and none with

Polarity items in Strawsonian contexts – A comparison   

   63

the former to be a strong indication for a dissociation between superlatives and the only.

6 Minimizers Minimizers are one of the best-known classes of negative-polarity items, and most texts on the topic mention some of the colorful expressions in this class.22 Less well-known is the fact that minimizers have a rather different distribution than any or ever. Far more than these two expressions, minimizers prefer negation as their core context, and elsewhere I have claimed that this might be the reason why minimizers, but not indefinites like any, appear to be likely to develop an inherently negative interpretation in the course of the Jespersen cycle (Hoeksema 2009). Be that as it may, we may also ask whether minimizers are acceptable in Strawsonian contexts such as the scope of only, or the restriction of a superlative. As it turns out, minimizers are nearly nonexistent in most such contexts, and completely unacceptable in one, the restriction of superlatives. To see how clearly minimizers differ from any in superlative contexts, compare the following two sentences: (41)

She cooked the best meal anyone has ever served me.

(42) * She was the most beautiful woman I have cared a pin about. (43) * Fred is the smartest man who ever lifted a finger to help. Contexts such as the only or the first are not particularly common either, but attested: (44)

Over 150 other kids knew about it, and I was the only person to do a damn thing about it.23

22 The term is from Bolinger (1972). Typical minimizers are a word (He didn’t say a word about the proposal), a thing (He didn’t do a thing to protect our interests), a dime (Don’t give him a dime for that crappy job), as well as idiomatic expressions such as a red cent, a plugged nickel, a damn thing, etc. See Vallduví 1994, Israel 1996, 2001, 2011, Hoeksema 2002, Eckardt 2006 (chapter 5), among others. 23 http://www.songmeanings.net/songs/view/3530822107858719355/ , accessed on June 4, 2012.

64   

   Jack Hoeksema

This general picture is confirmed by our corpus data: context

English

Dutch

German

N

%

N

%

N

%

superlative the only the first the last only XP negation question other

– 4 1 – 2 931 20 156

– 0.4 0.1 – 0.2 84 2 14

– 6 1 1 9 14580 98 1470

– 0.04 0.006 0.006 0.06 90 0.6 9

– 1 – – – 218 3 33

– 0.4 – – – 85 1 13

total

1114

100

16165

100

255

100

Table 6: Minimizers in English and Dutch

Again, there is a clear dissociation between the only, the first etc., and superlatives, against the predictions of Hoeksema (1986a) and von Fintel (1999), which assume that these contexts pattern alike, at least as far as polarity sensitivity is concerned.24

7 Yet and as yet The English temporal adverb yet is a polarity item in one of its many uses. Together with already, still, anymore it forms a group of four adverbs that are sometimes referred to as adverbs of temporal perspective. While the first two are positive polarity items (Baker 1970), anymore and yet are negative polarity items. Some linguists (see especially Löbner 1989, 1999) have proposed to capture the relations between the four elements in terms of duality (analogous to the case of quantifiers by an Aristotelian Square of Opposition). We will have to ignore the details of how best to analyze these adverbs, noting only that any duality-based treatment primarily deals with the interaction of the four adverbs with negation, and does not say much about other environments. For example, Löbner’s account

24 Minimizers are only very marginally licensed by only. See Beaver 2004 for some discussion.

Polarity items in Strawsonian contexts – A comparison   

   65

does not make any predictions about occurrences with superlatives, or restrictive adverbs of the only-class. English yet has a related expression as yet/as of yet, which shows some interesting differences with yet (Hoeksema 1993). The most striking difference is that as yet may (and often does) appear to the left of the licensing element, whereas yet belongs to the more familiar kind of polarity items that appear to its right: (45)

As yet/*yet, there has been no news from the border.

(46)

There has been no news from the border as yet/yet.

(Note that there is a concessive use of yet which may appear in sentence-initial position in (45), being a discourse-linker, but it has a meaning that is obviously different from the meaning involved in (46), and the variant of (45) with fronted as yet.)

Context

Yet

%

As yet

%

superlative the only the first the last only XP negation question other

54 2 3 – 1 1812 148 71

3 0.1 0.1 – 0.05 87 7 3

– 3 – – 13 270 4 214

– 0.6 – – 3 54 0.8 42

total

2091

100

504

100

Table 7: Yet – As Yet

Note that there appears to be a nearly perfect complementary distribution of yet and as yet in the Strawsonian contexts of superlatives and their ilk on the one hand (these take yet) and only on the other (this one takes as yet). Here are some examples from the corpus: (47)

But Stalin knew that, as yet, the Americans possessed only one or two Bombs.25

25 Simon Sebag Montefiore, Stalin – The Court of the Red Tsar, Phoenix, London, 2004, p. 510.

66   

   Jack Hoeksema

(48)

[A]s yet Lewis was only halfway down his obligatory orange juice26

(49)

[F]or who, in his wildest dreams, could have foreseen in the Macedonian infant the greatest conqueror who had yet been born into the world?27

(50)

But the highlight of 1988 is perhaps our most brilliantly confounding headline/subhead yet: Bored? Try clubbing yourself.28

Again, we see that there is no guarantee that if an expression is triggered by one type of Strawsonian context, it is triggered by all. All 54 occurrences in superlative contexts involve yet, and of 14 occurrences with only, 13 involve as yet. Both differences are statistically significant (Fisher’s exact test, 2-tailed, p < 0.0001).

8 Alive One of the lesser-known negative polarity items of English is alive, in its use as a nominal modifier.29 When used as a predicate, alive is not a polarity item at all: (51)

I am glad to be alive.

(52)

He was alive and kicking.

However, as a post-nominal modifier30, alive acts as a polarity item. It appears in the restriction of negative, universal and superlative noun phrases, among others: (53)

Okonkwo was the greatest wrestler and warrior alive.31

26 Colin Dexter, The remorseful day. Pan Books, London, 2004 [1st ed 1999], p. 103. 27 J.B. Bury, A history of Greece, New York: The Modern Library, n.d. [1900], p. 675. 28 Philadelphia Citypaper, June 1, 2006, p. 16. 29 Other nominal modifiers that function as polarity items are worth his salt, to speak of and self-respecting. 30 Or as the coda element in an existential clause: There isn’t a man alive who loves you better than yours truly. 31 Chinua Achebe, Things fall apart, p 82. Heinemann, London, 1981 [1st ed. 1958]

Polarity items in Strawsonian contexts – A comparison   

(54)

Johnson knew more books than any man alive.32

(55)

Neither I nor any woman alive has ever performed them33

   67

However, occurrences in non-polarity contexts are ruled out: (56) # The doctor examined a woman alive. For our purposes, this is an interesting expression, being one of very few that occur with high frequency as the restriction of a superlative noun phrase, as well as the only and the last. However, among other Strawsonian contexts, restrictive adverbs appear to be absent, and sentences such as (57) are intuitively rejected: (57)

a. # Only Fred examined a woman alive. b. # Fred only examined a woman alive. c. # Fred examined only a woman alive.

Our corpus data are summarized in Table 8 below.

Context

N

%

superlative the only the first the last only XP negation question other

33 4 1 2 – 20 4 15

42 5 1 3 – 25 5 19

total

79

100

Table 8: Alive

I conclude that alive is one of a number of polarity items showing dissociation between restrictive adverbs and superlative contexts.

32 James Boswell, Life of Johnson, Oxford University Press, Oxford, 1985 [1791], p. 52. 33 Richard Adams, Shardik. Penguin 1977 [1974], p. 68.

68   

   Jack Hoeksema

9 Need, hoeven, brauchen Among the best-known polarity items are the modal verbs need, hoeven (Dutch) and brauchen (German), cf. Duffley and Larrivee (1998), van der Wouden (2001), Falkenberg (2001), van der Auwera and Taeymans (2009).34 One of their more unusual properties is a relatively high percentage of occurrences in the scope of a restrictive adverb such as only (or its Dutch and German counterparts). Some illustrative examples are: (58)

You need only press this button once.

(59)

Du brauchst lediglich dein Badezeug You need only your swim gear ‘You need only bring your swimming gear’

(60)

Je hoeft maar één keer te bellen You need but once to press ‘You need press only once’

mitzunehmen along-bring

[German]

[Dutch]

Occurrences in the restriction of superlatives are sharply ungrammatical: (61) * The tallest student we need consider is 6 foot 8. (62) * De sterkste drank die we hoeven te drinken The strongest beverage that we need to drink ‘The strongest beverage we have to drink is beer’

is bier. is beer

Again, we seem to have a difference in licensing behavior between two Strawsonian contexts. Our corpus data confirm our intuitions, as the data in Table 9 below show:

34 See Johannesen (2003) for Norwegian modal verbs, and De Haan (1997) for a typological study of the interaction of negation and modality. See also Iatridou and Zeijlstra (2010) and Homer (2011).

   69

Polarity items in Strawsonian contexts – A comparison   

Context

need

hoeven

brauchen

N

%

N

%

N

%

superlative the only the first the last only XP negation question other

– 2 – – 59 370 12 59

– 0.4 – – 12 74 2 12

– 36 – 4 466 1937 76 596

– 1 – 0.1 15 62 2 19

– 2 – – 69 307 3 14

– 0.5 – – 17 78 1 4

total

502

100

3115

100

395

100

Table 9: Need, hoeven, brauchen

Main verb occurrences of the three verbs in question were not included in these data, since the main verbs are not polarity items. Only auxiliary verb uses were included. Note that we have a further dissociation between superlatives and the only, indicating once more that these two contexts are not equivalent in their licensing behavior.

10 Summary of the distributional data In the previous sections, we have looked at a variety of negative polarity items, starting with the most familiar items any and ever, and moving to some of the less well-studied specimens of the family of polarity sensitive expressions. In each case, I have focused on a small set of nonstandard environments, restrictions of superlatives, the only, the first, the last and the scope of restrictive adverbs such as only. Whenever possible, I have tried to come up with cross-linguistic comparisons of English, Dutch and German data. We see that the nonstandard environments, even though they should pattern alike as being a Strawson downward entailing context in the sense of von Fintel (1999), they in fact show widely diverging behavior. Besides differences between restrictive adverbs such as only and the restriction of superlatives, we also found a great deal of evidence for differences between superlatives and the only, the first, the last (contra Hoeksema 1986a). Simplifying the data somewhat by reducing numbers to a simply binary +/– distinction, and ignoring some of the contexts in the tables, such as questions, we may summarize Tables 1–9 by Table 10 below. Note that this table differs from the previous ones in having the contexts displayed horizontally in the top row, and

70   

   Jack Hoeksema

the items vertically, in the leftmost column, whereas the other tables were set up the other way around. This was done for convenience, to make it possible to fit all data on one page.

Item↓

context→

Any Ever Anymore Meer Mehr In X [English] In X [Dutch] Remotely In de verste verte [Dutch] Minimizers [English] Minimizers [Dutch] Minimizers [German] Yet As yet Alive Need Hoeven [Dutch] Brauchen [German]

superl

the only

the first

the last

only

negation

+ + – – – + + – – – – – + – + – – –

+ + + – – + + + + + + + + + + + + +

+ + – – – + + + – + + – + – – – – –

+ + – – – – – + – – + – – – + – – –

+ + – + – – – + + + + – + + – + + +

+ + + + + + + + + + + + + + + + + +

Table 10: Overview

11 Most Gajewski (2010) discusses the case of polarity items in the restriction of the quantifier most as exemplified in the following sentence: (63)

Most people who have ever met him, agree the man is very charming

According to Gajewski’s account, the superlative nature of most and the fact that superlatives are Strawson Downward Entailing, are the two factors responsible for the acceptability of ever in sentences such as the above. If this is indeed the case, it would seem that certain interesting predictions can be made: (1) polarity items which appear in the restriction of a superlative, may also appear in the restriction of most, and (2) items which don’t appear in the restriction of a superlative, won’t appear in the restriction of most either.

Polarity items in Strawsonian contexts – A comparison   

   71

For a comparison of most with superlatives, I will confine myself to a few of the items discussed above, in particular the ones that commonly appear with superlatives. The results can be found in Table 11 below. Context →

Superlative

Most

Item ↓

N

%

N

%

any ever in X [English] in X [Dutch] alive yet

45 727 220 257 27 42

0.8 16 42 24 41 2

6 2 – – 2 –

0.08 0.05 – – 3 –

Table 11: Most and superlatives

As we may see from this table, occurrences of polarity items in the restriction of most are indeed attested, but rare. However, given the exceedingly large percentages of occurrences of ever and in X in the restriction of a superlative, we would expect, given the claim that most behaves like a regular superlative, at least internally (see Gajewski 2010 for details), a much higher number of occurrences. In the case of ever, we would expect to find a lot more occurrences in the restriction of most than we have found for any, given that ever also appears a lot more than any in the restriction of a superlative. The opposite is in fact the case. Note also that for polarity items of the form in X, the results of occurrences in the restriction of most are actually fairly bad: (64) ?*Most people I have met in ages agree Fred is a jerk. The same is true for their Dutch counterparts: (65) * De meeste mensen die ik in tijden ontmoet heb The most people that I in ages met have vinden Fred een griezel. find Fred a creep ‘Most people I have met in ages find Fred a creep’ My conclusion is that whatever the precise account of polarity items in the restriction of quantifier most is going to be, it can’t be based on the superlative character of most.

72   

   Jack Hoeksema

12 Conclusions I have provided evidence that polarity items that appear in some contexts that are Strawson Downward Entailing (but not regular Downward Entailing), do not necessarily appear in other such contexts. Two main conclusions can be drawn from the preceding sections: – Not all Strawson downward entailing contexts are alike, hence the semantic notion of Strawson downward entailment cannot be the one uniform foundation on which the theory of polarity licensing rests – Not all polarity items are the same in their licensing properties We have seen that polarity items may differ among each other in an absolute sense or in terms of likelihood: for example, any and ever differ greatly in their likelihood of appearing in the restriction of a superlative, but not absolutely; on the other hand, while any may appear in the restriction of a superlative, German brauchen (or English need) is completely barred from that position. The property of Downward Entailingness, postulated as the main criterion for the licensing of negative polarity items by Ladusaw (1979), while remaining theoretically attractive, has been shown to be flawed empirically by von Fintel (1999), following a slew of other linguists (beginning with Linebarger (1980, 1987). However, we must now view von Fintel’s alternative notion of Strawson Downward Entailment to be empirically flawed as well. For many polarity items, it does not provide a sufficient licensing condition, as we have seen in case after case in the preceding pages. One may also ask whether it provides a necessary condition for licensing. If we consider polarity items of the weakest kind, the ones that appear in nonveridical environments, such as the Greek indefinites of the kanenas-series discussed by Giannakidou (1997, 1998), or early modernDutch enig (see Hoeksema 2010a for discussion), we see cases for which Strawson downward-entailment is not necessary either (e.g. weak modals form nonveridical contexts that are not downward entailing).35 That means that Strawson downward-entailment is not the one notion underlying all instances of licensing.36 In

35 For an unusual case of a polarity item with a preference for nonveridical contexts, see Hoeksema and Sailer 2012, where the German expression aus Dummsdorf kommen (“come from Stupidville”) is discussed. Dutch has an expression het fijne weten van ‘know the fine [details] of’, which appears with negation, in questions, but also quite frequently with willen ‘want’, a typical nonveridical predicate. 36 For some other types of problems with Von Fintel (1999), see Giannakidou (2006) and Homer (2008). One interesting problem these papers point out is the status of polarity items in the restriction of universals. While the restriction of all and every may serve as a host to polarity

Polarity items in Strawsonian contexts – A comparison   

   73

light of the great and surprisingly subtle variation among polarity items, only some of which could be discussed here, we are not going to find such a notion either. It is possible that accounts such as those of Zwarts (1986, 1996) or Van der Wouden (1997), or the more recent proposal by Gajewski (2011), can be modified to provide a proper account of the variation at hand. At present, they are too simple because they distinguish only two or three classes of polarity items. In addition, Zwarts’ account has what one might call the concentric property: weak items are licensed in a superset of the contexts for strong items, and strong items are licensed in a superset of the superstrong items. In a picture: superstrong strong weak

Figure 1: The Zwarts Hierarchy

While this may be a pleasing picture to behold, and one that is relatively easy to formalize in one way or another, using a set of semantic conditions, we have seen that the concentric property does not hold for polarity licensing. The set of contexts for any two polarity items may overlap, without it being the case that one set is a superset of the other. We have shown this for in years and need, for example, and for American anymore (OK with the only, not OK with only) versus Dutch meer

items, the restriction of each may not (Seuren 1985): (i) Every student who has ever taken Linguistics 101, has heard this joke. (ii) All students who have ever taken Linguistics 101, have heard this joke. (iii) * Each student who has ever taken Linguistics 101, has heard this joke. It has been suggested (see e.g. Hoeksema 1986b) that each differs from every in having an existential presupposition. (Usually, it quantifies over a contextually-given group, unlike every.) However, under the Strawsonian account, this ought not make any difference.

74   

   Jack Hoeksema

(OK with maar ‘only/but’, not OK with de enige ‘the only’). Since only a small subset of all contexts have been looked at in detail in this paper, more such discrepancies are likely to crop up when the scope of the present study is expanded. In spite of decades of research on polarity items, the search for a proper account of licensing variation has only just begun. Using a variety of polarity items, rather than just the usual suspects any and ever, it is possible to pry apart various contexts that look, at first blush, to pattern alike, such as the only and superlatives, and to look for factors that might distinguish them.

13 References Atlas, Jay D. 1993. The importance of being only: Testing the new-Gricean versus neo-entailment paradigms. Journal of Semantics 10: 301–318. Atlas, Jay D. 1996. Only noun phrases, pseudo-negative generalized quantifiers, negative polarity items and monotonicity. Journal of Semantics 13: 265–328. Auwera, Johan van der and Martine Taeymans. 2009. The need modals and their polarity. In: Rhonwen Bowen, Mats Mobärg and Sölve Ohlander, Corpora and Discourse – and Stuff, 317–326. Gothenburg: University of Gothenburg. Baker, C. Lee. 1970. Double negatives. Linguistic Inquiry 1: 169–186. Bayer, Josef. 1996. Directionality and Logical Form: On the Scope of Focusing Particles and Wh-in-situ. Dordrecht: Kluwer. Beaver, David and Brady Clark. 2003. Always and only: Why not all focus sensitive operators are alike. Natural Language Semantics 11(4): 323–362. Beaver, David. 2004. Five only pieces. Theoretical Linguistics 30:45–64. Bhatt, Rajesh. 2002. The raising analysis of relative clauses: Evidence from adjectival modification. Natural Language Semantics 10(1): 43–90. Bolinger, Dwight. 1972. Degree words. The Hague: Mouton. Carlson, Gregory N. 1980. Polarity any is existential. Linguistic Inquiry 11(4): 799–804. Carlson, Gregory N. 1981. The distribution of free-choice any. In: Papers from the Seventeenth Regional Meeting of the Chicago Linguistic Society, 8–23. Chicago: CLS. Chierchia, Gennaro. 2004. Scalar implicatures, polarity phenomena, and the syntax/pragmatics interface. In: Adriana Belletti (ed.), Structures and Beyond, 39–103. Oxford: Oxford University Press. Chierchia, Gennaro. 2006. Broaden your views: Implicatures of domain widening and the “logicality” of language. Linguistic Inquiry 37: 535–590. Dayal, Veneeta. 1998. Any as inherently modal. Linguistics and Philosophy 21: 433–476. den Dikken, Marcel and Anastasia Giannakidou. 2002. From hell to polarity: “Aggressively non-D-linked” wh-phrases as polarity items. Linguistic Inquiry 33(1): 31–61. Duffley, Patrick J. and Pierre Larrivée. 1998. Need, dare and negative polarity. Linguistic Analysis 28: 1–19. Eckardt, Regine. 2006. Meaning Change in Grammaticalization: An Enquiry into Semantic Reanalysis. Oxford: Oxford University Press.

Polarity items in Strawsonian contexts – A comparison   

   75

Falkenberg, Gabriel. 2001. Lexical sensitivity in negative polarity verbs. In Hoeksema et al., 79–98. Fauconnier, Gilles. 1978. Implication reversal in natural language. In: Franz Guenthner and Siegfried J. Schmidt, eds., Formal Semantics and Pragmatics for Natural Languages, 289–301. Dordrecht: D. Reidel. von Fintel, Kai. 1999. NPI-licensing, Strawson-entailment, and context-dependency. Journal of Semantics 16: 97–148. Gajewski, Jon. 2005. Neg-raising: Polarity and presupposition. Ph.D. dissertation, MIT. Gajewski, Jon. 2010. Superlatives, most and NPIs. Journal of Semantics 27(1): 125–137. Gajewski, Jon. 2011. A note on licensing strong NPIs. Natural Language Semantics 19: 109–148. Giannakidou, Anastasia. 1997. The Landscape of Polarity Items. Doctoral dissertation, Rijksuniversiteit Groningen. Giannakidou, Anastasia. 1998. Polarity Sensitivity as (Non)veridical Dependency. Amsterdam: John Benjamins. Giannakidou, Anastasia. 2001. The meaning of free choice. Linguistics and Philosophy 24: 659–735. Giannakidou, Anastasia. 2006. Only, emotive factive verbs, and the dual nature of polarity dependency. Language 82(3): 575–603. Giannakidou, Anastasia. 2011. Positive polarity items and negative polarity items: Variation, licensing, and compositionality. In: Claudia Maienborn, Klaus von Heusinger, and Paul Portner (eds.), Semantics: An International Handbook of Natural Language Meaning, 1660–1712. Berlin: Mouton de Gruyter. Guerzoni, Elena and Yael Sharvit. 2007. A question of strength: On NPIs in interrogative clauses. Linguistics and Philosophy 30: 361–391. Haan, Ferdinand de. 1997. The Interaction of Modality and Negation: A Typological Study. New York: Garland Press. Herdan, Simona. 2005. Who knows the only reading of only? Snippets 10: 7–8. Hoeksema, Jack. 1986a. Monotonie en superlatieven. In: Cor Hoppenbrouwers, Joop Houtman, Ineke Schuurman and Frans Zwarts (eds.), Proeven van taalwetenschap, 38–59. Groningen: Nederlands Instituut RUG. Hoeksema, Jack. 1986b. Monotonicity phenomena in natural language. Linguistic Analysis 16(1/2): 1986, 235–250. [Reprinted in Javier Gutierrez-Rexach (ed.). 2004. Semantics – Critical concepts, London: Routledge.] Hoeksema, Jack. 1993. As (of) yet. In: Ale de Boer, Jelly de Jong and Rita Landeweerd, eds., Language and Cognition 3. Groningen, 79–87. Hoeksema, Jack. 2002. Minimaliseerders in het standaard Nederlands. In: Tabu 32(3/4): 105–174. Hoeksema, Jack. 2005. In days, weeks, months, years, ages: A class of temporal negative polarity items. In: Dicky Gilbers and Petra Hendriks (eds.), Rejected Papers: Feestbundel voor Ron van Zonneveld, Groningen. Hoeksema, Jack. 2008. Distributieprofielen van negatief-polaire uitdrukkingen: een vergelijking van het Nederlands, Engels en Duits. Tabu 37(3/4): 111–195. Hoeksema, Jack. 2009. Jespersen recycled. In: Elly van Gelderen (ed.), 15–34, Cyclical Change. Amsterdam: John Benjamins. Hoeksema, Jack. 2010a. Dutch ENIG: from nonveridicality to downward entailment. Natural Language and Linguistic Theory 28(4): 837–859.

76   

   Jack Hoeksema

Hoeksema, Jack. 2010b. Negative and positive polarity items: An investigation of the interplay of lexical meaning and global conditions on expression. In: Laurence R. Horn (ed.), The Expression of Negation, 187–224. Berlin: Mouton de Gruyter. Hoeksema, Jack, Hotze Rullmann, Víctor Sánchez-Valencia, and Ton van der Wouden (eds.). 2001. Perspectives on Negation and Polarity Items. Amsterdam/Philadelphia: John Benjamins. Hoeksema, Jack and Manfred Sailer. 2012. Literal and nonliteral meaning in placename idioms. Yearbook of Phraseology 3: 127–142. Homer, Vincent. 2008. Presuppositions can be disruptors too: A case against Strawsonentailment. In: Natasha Abner and Jason Bishop (eds), Proceedings of the 27th WCCFL, 220–228. Somerville: Cascadilla Press. Homer, Vincent. 2011. Polarity and modality. Ph.D. dissertation, UCLA. Homer, Vincent. 2012. Domains of Polarity Items. To appear. Horn, Laurence R. 1969. A presuppositional analysis of only and even. Papers from the Fifth Regional Meeting of the Chicago Linguistics Society, 98–107. Horn, Laurence R. 1996. Exclusive company: Only and the dynamics of vertical inference. Journal of Semantics 13: 1–40. Horn, Laurence R. 2005. Airport ’86 revisited: Toward a unified indefinite any. In: Gregory N. Carlson and Francis Jeffry Pelletier (eds.), Reference and Quantification. The Partee Effect. Stanford: CSLI. Iatridou, Sabine and Hedde Zeijlstra. 2010. On the scopal interaction of negation and deontic modals. In: Maria Aloni, Harald Bastiaanse, Tikitu de Jager and Katrin Schulz (eds.), Logic, Language and Meaning: 17th Amsterdam Colloquium, Amsterdam, The Netherlands, December 16-18, 2009, 315–324. Berlin: Springer. Israel, Michael. 1996. Polarity Sensitivity as Lexical Semantics. Linguistics and Philosophy 19: 619–666. Israel, Michael. 1998. Ever: polysemy and polarity sensitivity. In: Linguistic Notes from La Jolla 19, 29–45. Israel, Michael. 2001. Minimizers, maximizers and the rhetoric of scalar reasoning. Journal of Semantics 18(4): 297–331. Israel, Michael. 2011. The Grammar of Polarity: Pragmatics, Sensitivity, and the Logic of Scales. Cambridge: Cambridge University Press. Jacobs, Joachim. 1983. Fokus und Skalen: Zur Syntax und Semantik von Gradpartikeln im Deutschen. Tübingen: Niemeyer. Johannessen, Janne Bondi. 2003. Negative polarity verbs in Norwegian. Working Papers in Scandinavian Syntax. Vol. 71, 33–73. Kadmon, Nirit and Fred Landman. 1993. Any. Linguistics and Philosophy 16: 353–422. Keenan, Edward L. and Leonard M. Faltz. 1985. Boolean Semantics for Natural Language. Dordrecht: D. Reidel. Klima, Edward S. 1964. Negation in English. In: Jerry A. Fodor and Jerrold J. Katz (eds.), The Structure of Language, 246–323. Englewood Cliffs, NJ: Prentice Hall. Krifka, Manfred. 1995. The semantics and pragmatics of polarity items. Linguistic Analysis 25: 209–258. Labov, William. 1975. What is a Linguistic Fact? Lisse: Peter de Ridder Press. Ladusaw, William A. 1979. Polarity sensitivity as inherent scope relations. Ph.D. dissertation, University of Texas at Austin. Lee, Young-Suk and Laurence Horn. 1994. Any as indefinite plus even. Ms. Yale University.

Polarity items in Strawsonian contexts – A comparison   

   77

Linebarger, Marcia. 1980. The grammar of negative polarity. Ph.D. dissertation, MIT, distributed by the Indiana University Linguistics Club. Linebarger, Marcia. 1987. Negative polarity and grammatical representation. Linguistics and Philosophy 10: 325–387. Löbner, Sebastian. 1989. German schon – erst – noch: An integrated analysis. Linguistics and Philosophy 12: 167–212. Löbner, Sebastian. 1999. Why German schon and noch are still duals: A reply to van der Auwera. Linguistics and Philosophy 22: 45–107. Rooy, Robert van. 2003. Negative Polarity Items in Questions: Strength as Relevance. Journal of Semantics 20, 239–273. Rullmann, Hotze. 2003. Additive particles and polarity. Journal of Semantics 20: 329–401. Sassen, Albert. 1977. Drie chroningismen: wachten, verlet, (maar) … meer. In: Hans Heestermans (ed.), Opstellen door vrienden en vakgenoten aangeboden aan Dr. C.H.A. Kruyskamp, 156–168. The Hague: M. Nijhoff. Seuren, Pieter A. M. 1985. Discourse Semantics. London: Routledge. Sharvit, Yael and Simona Herdan. 2006. Definite and non-definite superlatives and NPI-licensing. Syntax 9: 1–31. Simpson, John A. and Edmund S.C. Weiner 1989 The Oxford English Dictionary. 2nd ed., Oxford: Clarendon Press. Stefanowitsch, Anatol. 2008. Negative entrenchment: A usage-based approach to negative evidence. Cognitive Linguistics 19(3): 513–531. Szabolcsi, Anna. 2004. Positive polarity – negative polarity. Natural Language and Linguistic Theory 22: 409–452. Vallduví, Enric. 1994. Polarity items, n-words, and minimizers in Catalan and Spanish. Probus 6(2-3): 263–294. Vandeweghe, Willy. 1983. De partikels van de al-nog-pas-groep in het Nederlands. Doctoral dissertation, University of Ghent. Wouden, Ton van der. 1997. Negative Contexts: Collocation, Polarity and Multiple Negation. London: Routledge. Wouden, Ton van der. 2001. Three modal verbs. In: Sheila Watts, Jonathan West and Hans-Joachim Solms (eds.), Zur Verbmorphologie germanischer Sprachen, 189–210. Tübingen: Niemeyer. Zwarts, Frans. 1981. Negatief Polaire Uitdrukkingen I. GLOT 4(1): 35–132. Zwarts, Frans. 1986. Categoriale grammatica en algebraïsche semantiek. Doctoral dissertation, University of Groningen. Zwarts, Frans. 1995. Nonveridical contexts. Linguistic Analysis 25(3/4): 286–312. Zwarts, Frans. 1998. Three types of polarity. In: Fritz Hamm and Erhard W. Hinrichs (eds.), Plurality and Quantification, 177–238. Dordrecht: Kluwer.

Volker Gast

From æghwæðer to either: The distribution of a negative polarity item in historical perspective 1 Introduction Two major use types of either can be distinguished, those where it functions as a dual quantifier, and those where it functions as a particle.1 The attribute ‘dual’ is here used in a grammatical sense, i.e. as standing for a number category whose members denote sets with a cardinality of two. Within the quantifying use types we can moreover make a syntactic distinction between those cases where either has the distribution of a pronoun (cf. 1), and those were it takes the position of a determiner (cf. 2). (1)

“I’m not going to translate for either of you,” he said softly. [BNC CAO 1996]

(2)

What kind of redress does either kind of prisoner have against these conditions? [BNC H45 1049]

Within the family of particle uses, we can also distinguish two major types. First, either can be used as a contrastive disjunctive particle preceding the focus and correlating with or, as in (3) (cf. Haspelmath 2007 on contrastive disjunction). (3)

Any effect produced must surely be wishful thinking either on the part of the practitioner or the patient, or both. [BNC C9V 1138]

Second, either is used in a postfocal position, as a negative correlate of too (cf. König 1991: 61, who regards postfocal either as “a suppletive form of too”; cf. 4). (4)

My mother doesn’t like me either. [BNC EVC 1946]

1 I wish to thank the participants of the Workshop ‘Beyond any and ever’ as well as two anonymous reviewers for helpful comments and valuable suggestions. Any inaccuracies are my own.

80   

   Volker Gast

Postfocal either is also found following a disjunction in other types of nonveridical contexts, e.g. with constituent negation as in (5), and without negation as in (6) (cf. also Rullmann 2002 on such disjunctive postfocal uses). (5)

There were no men, or women either … [BNC]

(6)

If John had said so, or William either, I could believe it. (Rullmann 2002: 112)

This study focuses on quantifying uses of either as illustrated in (1) and (2) above, but occasional reference will be made to the particle uses as well. What makes the historical study of either particularly interesting is the fact that this element derives from a universal quantifier (Old English ægðer) while functioning as an existential quantifier (with a restriction to nonveridical contexts) in Modern English. The question arises how and why this change came about. Before considering the diachronic development of either, the ‘target state’ of this development – the distribution of either is Modern English – is outlined in Section 2. Section 3 contains an overview of the dual quantifiers of Old English. Section 4 describes and analyses the changes that took place in Middle English. Section 5 concludes with a summary and some general remarks on the study of historical developments in the domain of polarity-sensitive items.

2 Quantifying either in Modern English In descriptive grammars, either is commonly regarded as the dual form of (number-unspecified) any (see e.g. Quirk et  al.  1985:  391). This means that either is an existential quantifier which is restricted to a domain of quantification with a cardinality of two, and to ‘nonveridical’ contexts, i.e. contexts in the scope of a nonveridical operator as defined in (7) (e.g. Zwarts  1995,  1998;  Giannakidou 1998, 2011).2

2 The notion of ‘nonveridicality’ as opposed to, for instance, ‘downward entailment’ in the sense of Ladusaw (1979), has the advantage of covering negative polarity uses as well as free choice readings of either. Note that the entailment relations in (7) are assumed to hold relative to an ‘individual anchor’ in the sense of Farkas (1992), so common doxastic verbs do not create nonveridical contexts. For further details, cf. Zwarts (1995, 1998), Giannakidou (1998, 2011) and references cited in Giannakidou (2011).

From æghwæðer to either   

(7)

   81

A propositional operator O is a. veridical iff Op → p b. nonveridical iff ¬ [ Op → p ]

Either in (8a) and (9a) can be interpreted as shown in (8b) and (9b), respectively.3 (8c) and (9c) illustrate that (8a) and (9a) are nonveridical contexts. (8)

a. I don’t know either man. b. ¬∃x ∈⟦two men⟧ [ (I,x) ∈ ⟦know⟧ ] c. ¬ [ ¬p → p ] (true)

(9)

a. If I see either man, I will call you. b. ∃x ∈ ⟦two men⟧ [ (I,x) ∈ ⟦see⟧ ] → (I,you) ∈ ⟦call⟧ c. ¬ [[ p → q ] → p ] (true)

Like any, either is also used as a free choice element, as in (10) and (11). (10)

Anybody could do that!

(11)

You can pay with either card.

While the dual-of-any analysis is certainly not entirely mistaken, there are (at least) four respects in which either differs from any (cf. also Huddleston and Pullum 2002: 387–388). First, either invariably carries lexical stress and is thus, to an extent, inherently contrastive. Second, unlike any, either is also found in veridical contexts with a universal reading. Specifically, it occurs in prepositional phrases like on either side and in either hand, where it is equivalent to each and (distributive) both. A pertinent example is given (12). As will be seen below, universal uses of either in prepositional phrases as illustrated in (12) are a remnant of Old English.

3 Alternatively, either in (9a) can be interpreted as a wide-scope universal quantifier, cf. Hintikka (1980): ∀x [ x ∈ ⟦two men⟧ ∧ (I,x) ∈ ⟦see⟧ → (I,you) ∈ ⟦call⟧ ]. The relationship of quantification and scope will play a prominent role in the argument made in this paper.

82   

(12)

   Volker Gast

Sophie froze, a glass in either hand, her gaze fixed on the bathroom door, her lips parted in a silent cry. [BNC FP7 409]

The third difference between either and any is that either is only used with count nouns, while any (like some) can also be used with mass nouns (e.g. I don’t have any wine). This difference is obviously related to the fact that either requires a domain with a cardinality of two and, hence, a count noun to combine with. Finally, either differs from any in terms of the discourse pragmatic status of the domain of quantification. This difference, too, is obviously not totally unrelated to the number difference between either and any (dual vs. number-unspecified). While any can be used with both given and new domains, either requires a domain that is under discussion. Either-DPs behave like definites, and either can be regarded as a ‘strong’ determiner in the sense of Milsark (1977). This is illustrated by (13) and (14). (13)

a. There didn’t exist any trees before the world was created. (domain new) b. I got three books for Christmas, but I haven’t read any of them. (domain given)

(14)

a. * There didn’t exist either sex before the world was created. (domain new) b. I got two books for Christmas, but I haven’t read either of them. (domain given)

To summarize, Modern English either can be characterized as an existential quantifier with the attributes ‘dual’ and ‘strong’, and with a distributional restriction to nonveridical contexts. As we have seen, under specific circumstances it may also occur in veridical contexts with universal quantificational force. Given that these uses represent a remnant of Old English (cf. Section 3), it seems reasonable to regard them as exceptional. Having outlined the most important distributional properties of either in Modern English, we will now go back to Old English and try to reconstruct the developments that have led to this distribution.

From æghwæðer to either   

   83

3 The dual quantifiers of Old English Old English had a morphologically quite transparent paradigm of dual quantifiers, which is shown in Table 1. The paradigm contains the universal quantifier æghwæðer (contracted ægðer), the existential quantifier āhwæðer (~ awðer), and the negated existential quantifier nāhwæðer (~ nawðer). For the sake of simplicity, I will mostly use the contracted forms in the following when referring to any of these quantifiers.

a AND b a OR b NOT (a OR b)

neg

gen

con

‘which of two’ contracted

ge

n

æ ā ā

hwæðer hwæðer hwæðer

ægðer awðer nawðer

Table 1: Old English dual quantifiers

All dual quantifiers are based on the root hwæðer ‘which of two’ and contain the prefix ā-, before ge- ([je]) æ-. This prefix is widespread in the domain of quantitification and distributivity (cf. Germ. je-mand ‘someone’ < OHG êo-man, je ‘ever, each’). It derives from the Indo-European root *aiu̯-, aiu̯- ‘age, eternity’ (cf. Gr. aiō-, arch. aiwon, Lat. aev-, West Germanic aiwo[n] ‘age, eternity’), but its originally temporal meaning was generalized to the domain of referentiality at some point (e.g. ‘if there was ever [a] man’ ~ ‘if there was any man’). In the paradigm in Table 1, ā- can be assumed to be responsible for the interpretation of the three elements as quantifiers (rather than, say, question words). Æghwæðer differs from āhwæðer in terms of only one element, i.e. the multifunctional prefix ge-, which is also found in combination with non-finite forms in Old English and which has, in all likelihood, originally a sociative, comitative or conjunctive meaning (cf. Latin cum ‘with’). It is therefore glossed as ‘con’ for ‘conjunction’ in Table 1. Nawðer simply adds the negative prefix n- to awðer. In order to understand the developments leading from the paradigm shown in Table 1 to the situation in Modern English, we will have to consider the interpretation of the relevant quantifiers in different polarity contexts as well as their scope properties. Example (15) illustrates the use of (the universal quantifier) ægðer in a veridical context (we can assume a [veridical] assertion operator for such examples).

84   

(15)

   Volker Gast

Old English Ond se cyning æghwæðer-um þissa biscopa and det king each.of.two-dat.pl det.gen.pl bishop.gen.pl his gyfe sealde. his gift gave ‘And to each of these (two) bishops the king gave a gift.’ [Beda 504, 28; a900]

Ægðer is also used in nonveridical contexts. In this case its scope is clausebound, but ægðer seems to take scope over any clause-internal nonveridical operator, judging from the records available to us. (16) illustrates that ægðer does not take scope beyond the local clause. The context of this example makes it clear that ægðer is contained in the scope of (the conditional subjunction) gif. With an extra-clausal scope interpretation of ægðer, (16) would translate as ‘If either of them is foolish’ (≡ ‘for each x, if x is foolish …’). The scope of ægðer is here indicated by the position of the universal quantifier in the glosses. Ægðer itself corresponds to the variable bound by the universal quantifier. (16)

[Foolish teachers come for the pupil’s sins. Therefore often through the teacher’s folly the disciples come to grief, and often through the teacher’s wisdom foolish disciples are preserved.] Gif ðonne [ [ ægðer bið unwis …]], ðonne is to geðencenne … if then [ ∀x [   x is foolish …]] then is to think … ‘If then both (teachers and students) are foolish, we must consider [what Christ himself said in his Gospel.]’4 [Alfred, Gregory’s Pastoral Care 1, 29; a900]

In (17), ægðer takes wide scope relative to a clause-internal nonveridical operator. (17)

Hwa is þætte ariman mæge hwæt þær moncynnes who is that number can what there men.gen forwearðon on ægðere hand? perished on ægðer.dat hand ‘Who (is there that) can number those that fell on each side?’ [Alfred, Orosius I, 11; c893]

4 Translated by Henry Sweet (King Alfred’s West-Saxon Version of Gregory’s Pastoral Care, 1871, London, Trübner).

From æghwæðer to either   

   85

For the sake of simplicity, I will analyse the free relative clause in (17) as an indirect question, i.e. the sentence is taken to mean ‘Who knows how many men fell on each side?’5 (18) provides a paraphrase of the free relative clause as a (direct) question. The question operator Q can be interpreted as a request to provide a list of referents to which the properties assigned to the variable x apply. Given that there are two armies involved, and given that the universal quantifier binding the variable a ranging over armies takes scope over the question operator Q, in the form shown in (18) the question elicits two lists. With ægðer taking narrow scope, it would elicit a (single) list of warriors that fought in both armies (and died). (18)

∀a [ Qx ∈ ⟦men⟧ [ x ∈ ⟦die⟧ ∧ x ∈ a ]] ‘For both armies a, provide the set of all x from the set of men such that x died and x forms part of a!’

(17) also illustrates the fact pointed out by Einenkel  (1904:  66) that either was particularly common in prepositional phrases throughout the history of English: Æghwæðer, which was later heavily affected by ælc > each, is particularly remarkable in combination with healf, ende and hand. [my translation]6

As was mentioned in Section 2, this tendency is reflected up to the present day, as either is still used in veridical contexts with a universal interpretation when it occurs within PPs. This type of reflection of earlier language stages in later ones has been called ‘persistence’ by Hopper (1991). The existential quantifier awðer was typically used in nonveridical contexts, with narrow scope relative to the nonveridical operator – in (19), the conjunction ær ‘before’. (19)

7 þær hæfdon longsum gefeoht ær þara folca and there had long fought before det people aþer (= awðer) fluge. one.or.other fled ‘…and they fought long before either of the two parties fled.’ [Alfred, Orosius 198, 25; c893]

5 A free relative clause in the scope of a question operator as in (16) can be assumed to ‘inherit’ the question feature – and, hence, the nonveridicality – of the matrix clause. Free relative clauses are not per se nonveridical (e.g. I wonder/*know who has ever walked this way). 6 “Bemerkenswert ist das später von ælc > each so stark beeinträchtigte æghwæðer bei healf, ende und hand.”

86   

   Volker Gast

I have not found any examples of OE awðer in veridical contexts. This does probably not mean that awðer did not occur in such contexts, however, as there are some relevant examples from Early Middle English (cf. 20). (20)

Early Middle English Forrþi wendenn þeʒʒ full wel þatt owwþer off þa tweʒʒ therefore knew they full well that one.of.two of the two off dæþe wære risenn upp. from death was risen up ‘Therefore they knew very well that one of the two had risen from death.’ [Ormulum 10344; a1180]

Note moreover that there are occasional occurrences of OE oðer with the meaning ‘one of two’ in veridical contexts (without a correlate in the discourse environment as required by ModE other < OE oðer < Gm. *antharaz, cf. Germ. ander-). Bock (1887) assumes that such occurrences of oðer are contracted forms of awðer (cf. also Wülfing 1894: §359a for this position; but see Nusser 1913: 37ff for a counter-argument). From a semantic point of view, the Bock/Wülfing-hypothesis is certainly conceivable. (21) is a pertinent example from the Anglo-Saxon version of the Bible. (21)

Andreas, Simones broður Petres wæs oðer of þam Andreas, Simon.gen brother, Peter.gen was one of the twam, þa gehyrdon æt Iohanne, and him fyligdon. two rel heard prep John and him followed ‘Andrew, Simon Peter’s brother, was one of the two who heard what John had said and who had followed Jesus.’ [Anglo-Saxon Bible, John 1, 40; c990]

For the sake of completeness, an example of the negated universal quantifier nāhwæðer/nawðer is given in (22). (22)

naðer (= nawðer) ne mehte on oþrum sige geræcan neither not wanted on other victory reach ‘Neither wanted to grant the other the victory.’ [Alfred, Orosius 96, 33; c893]

All of the dual quantifiers mentioned above are commonly found in ‘correlative’ uses, i.e. as precursors of the modern particles either and neither. Ægðer correlates with the conjunction ge ‘and’ to introduce conjunctions (cf. 23), awðer correlates with oþþe ‘or’ in disjunctions (cf. 24), and nawðer cooccurs with the negator ne in negated conjunction (cf. 25).

From æghwæðer to either   

   87

(23)

se was ægðer, ge heora cyning, ge heora bishop det was both conj their king conj their biscop ‘He was both their king and their bishop.’ [Alfred, Orosius 238, 14; c893]

(24)

eala, wære he auðer, oþþe hat, oþþe ceald. alas were he one.of.two or hot or cold ‘Alas, were he either hot or cold’ [Alfred, Gregory’s Pastoral Care 445, 36; a900]

(25)

næron nawðer ne on Fresisc gescæpene, ne on Denisc not.were neither not on Frisian shaped not on Danish ‘They were neither in the Frisian nor in the Danish shape.’ [Sax. Chron. 897; c890]

Even though the history of the correlative constructions illustrated in (23)–(25) is largely parallel to the development of the individual quantifiers, there are some particularities that deserve further examination. For example, the Middle English forms of quantifying ægðer stopped being used in veridical contexts, while the particle uses are regularly found in such contexts up to the present day – with a disjunctive function as illustrated in (24) for awðer, however. For reasons of space, these differences cannot be addressed in the present study. Some information on the development of the particle constructions can be found in Einenkel (1904), Nusser (1913) and, somewhat more recently, in Rullmann (2002).

4 Changes in Middle English 4.1 Forms and contexts In Middle English, the system of dual quantifiers as shown in Table 1 above disintegrated, and the quantifiers changed in terms of both form and function. I will use ‘either’ (< ægðer) and ‘outher’ (< awðer) as generic labels for the (Late) Middle English successors of OE ægðer and awðer, with the small caps indicating that these labels are generalizations over the various forms found in Middle English texts. Moreover, I will distinguish two stages (within ME) in the development of either, i.e. either1 and either2. The Modern English quantifier will simply be termed ‘either’. Accordingly, we will be dealing with four stages as shown in Table 2. Note that these stages do not only relate to the forms and functions of ægðer/either/either, but to the entire systems of quantifiers. It should also be

88   

   Volker Gast

mentioned that both Stage II and Stage III are located in Late Middle English. The Early Middle English system of dual quantifiers seems to have been more similar to the Old English system than to the Late Middle English one.

Old English Stage I ægðer

Middle English Stage II Stage III either1 either2

Modern English Stage IV either

Table 2:  The four stages from OE ægðer to ModE either

The main objective of this study is to determine the development of ægðer/either/ either in terms of the interpretations that this quantifier could have in specific contexts. There are two possible readings, i.e. universal and existential, and two major context types, i.e. veridical and nonveridical. For reasons to become apparent, it is necessary to distinguish two types of nonveridical contexts, depending on the type of nonveridical operator. Nonveridicality may be triggered either by clause-internal operators (e.g. by modals, cf. 26), or by clause-external ones (e.g. conditional operators, cf. 27). The unit of ‘clause’ is represented as ‘TP’ here, and is assumed to correspond to a proposition at the level of interpretation. (26)

[TP You may take either card.] [Prop ◊[ you take either card ]]

(27)

If [TP you take either card ], you will lose. [Prop You take either card ] → you will lose

We will be concerned with four quantification/context-combinations: 1. universal quantification in veridical contexts, e.g. Both men were drunk; 2. existential quantification in the scope of a clause-internal nonveridical operator, e.g. You may take either card; 3. existential quantification in the scope of a clause-external nonveridical operator, e.g. If you take either card, you will lose; 4. existential quantification in veridical contexts, e.g. One of the two men died. Cases 2 and 3 can alternatively be regarded as expressing universal quantification, with the quantifier taking scope over the (clause-internal or clausal-external) nonveridical operator. The four quantification/context-combinations are summarized in Table 3.

From æghwæðer to either   

context

veridical

quantification scope

universal narrow

nonveridical clause-internal

clause-external

universal existential / wide narrow

universal existential / wide narrow

   89

veridical

existential narrow

Table 3: Quantification/context-combinations

4.2 The decline of outher The starting point of the developments in Middle English is probably the decline of outher, which gradually dropped out of use in Late Middle English. This development is described by Einenkel (1904: 64) as follows (my translations): … ahwæðer disappears as early as Early Modern English, not without having transferred its sense to æghwæðer.7

The ‘transfer’ of meaning mentioned by Einenkel (1904) is probably best understood as a replacement of outher by either in specific contexts. This process of replacement seems to have started in nonveridical contexts with a clause-internal nonveridical operator. In such contexts, a universal quantifier (like either1) is often equivalent to an existential quantifier (like outher) if the former is interpreted with wide scope and the latter with narrow scope relative to the nonveridical operator. This type of equivalence can be illustrated with examples containing a possibility modal. Consider (28). (28)

with luytel aise he miʒte sitte and onaisiliche ligge also with little ease he could sit … and uneasily lie also opon eþur side on either side ‘With little ease he could sit … and uneasily also lie on either side.’ [SLeg. Becket 2212; c1300]

Eþur in (28) can be interpreted as an existential quantifier in the scope of a circumstantial modal or, alternatively, as a universal quantifier taking scope over a dynamic modal. The relevant readings are paraphrased in (29) and (30), respec-

7 “…ahwæðer stirbt schon im NE. aus, aber nicht ohne seinen sinn vorher an æghwæðer abgegeben zu haben.”

90   

   Volker Gast

tively. Existential quantification is here represented as (internal) disjunction, universal quantification as (external) conjunction.8 (29)

a. ◊circ [ He lay on the left side ∨ he lay on the right side ] b. The circumstances allowed himi: PROi to lie on the left side or PROi to lie on the right side

(30)

a. ◊dyn [ he lay on the left side ] ∧ ◊dyn [ he lay on the right side ] b. Hei was able [ PROi to lie on the left side ] and hei was able [ PROi to lie on the right side ]

Given that the Middle English forms of may and might had both circumstantial and dynamic (as well as epistemic) readings, the modal in (28) could be interpreted in either way. Accordingly, the quantifier either/eþur could have a (narrow-scope) existential as well as a (wide-scope) universal reading. Such a relation of equivalence between wide-scope universal quantifiers and narrow-scope existential quantifiers interacting with a nonveridical operator can be stated more generally as in (31) (‘O↓’ stands for a nonveridical operator):9 (31)

For any predicate P, and for any set {a,b}: ∀x ∈ {a,b} [ O↓ [ P(x) ]]   ≡  O↓ [ ∃x ∈ {a,b} [ P(x) ]]

The Old English example in (32) illustrates the generalization in (31) for negation. The two interpretations are given in (33).

8 Cf. Zimmermann (2000) on the problem of the ‘choice effect’, i.e. the equivalence of ‘X may A or B’ and ‘X may A and X may B’. 9 As a reviewer points out, the equivalence in (31) holds generally only for antimorphic operators in the sense of Zwarts (1995, 1998) and Giannakidou (1998, 2011). It does not apply to examples like The IRS rarely audits anyone (from Ladusaw 1979: 102), as this example does not imply that everybody was audited rarely by the IRS. For either, the equivalence does seem to hold, however: Students rarely attended either lecture is equivalent to Both lectures were rarely attended by (any) students, as far as I can tell. This is probably related to the fact that either, unlike any, combines with given and finite domains; cf. Section 2.

From æghwæðer to either   

   91

(32)

ne eagena [hafað] ægðer twega neg eyes.gen has either two.gen ‘(It does not have) either of two eyes.’ [Anglo-Saxon Riddles 40, 11; a1100]

(33)

a. ∀e ∈ {left eye, right eye} [ (he,e) ∉ ⟦have⟧ ] b. ¬∃e ∈ {left eye, right eye} [ (he,e) ∈ ⟦have⟧ ]

Alternative scope construals are not the only reason why universal and existential quantifiers may be (near) equivalent. In some cases the exact type of quantification is simply not very relevant. Consider the passive sentence in (34), whose implicit subject can be interpreted either existentially or universally (i.e., ‘x was known’ can either imply that ‘everybody knew x’ or that ‘somebody knew x’). (34)

Two nomys had þat noble … knowen in his cuntre two names had that nobleman known in his country kyndly by aither. commonly by either ‘Two names had that nobleman … known in his country by one/both of them.’ [Destruction of Troy 10929–30, a1400]

(34) has several readings. For example, it could be taken to mean that each subject knew the king by both names (i.e. Pyrrhos or Neoptolemos).10 In that case, both the implied subject and aither have a universal interpretation, and scope relations do not matter. More likely, each subject knew the king by at least one of his names. This interpretation results from an existential reading of aither in the scope of a universally quantified implied subject, as shown in (35). (35)

For all people p: There is a name n such that p knows the king by n

Yet another reading results when aither is interpreted as a universal quantifier with wide scope, and the implied subject is existential (cf. 36). In this case, for both names n there are at least some subjects that know n. Unlike (35), (36) allows for the possibility that there are subjects that do not know either of the king’s name. (36)

For both names n: There is some person p that knows the king by n

10 In Greek mythology, the name Neoptolemos is used, but the Aeneid has Pyrrhos (for a son of Achilles and Deidamia).

92   

   Volker Gast

What the writer of (34) wants to say is, basically, that Pyrrhos/Neoptolemos had two names which were both used in his kingdom (unlike the two [given] names of, say, Barack Hussein Obama). How many of those names were known to exactly how many subjects is probably irrelevant. Even though wide-scope universal and narrow-scope existential readings of either are not identical in such cases, they were probably similar enough to be ‘confused’, thus contributing to a general feeling of equivalence of either and outher (in specific contexts). Such a feeling of equivalence also arose in reciprocal predications. Old English regularly used ægðer in combination with oþer to express reciprocity, as in (37): (37)

and heora þær ægðer oþerne ofslog and of.them there both/each other.acc killed ‘And there they killed each other.’ [Alfred, Orosius 68, 18; c893]

A similar construction is found in Middle English with outher (here realized as aþer): (38)

aþer askede of oþeres stat one.of.two asked about other.gen state ‘They enquired about each other’s health.’ [Bevis of Hampton, 1990; c1300]

The use of an existential quantifier in reciprocal contructions is familiar from Modern English, where one another is used (more or less) interchangeably with each other. Such strategies are pervasive in the languages of the world (cf. Heine and Miyashita 2008: 180–1) and well attested among the Germanic languages (cf. Plank 2008). According to Nusser (1913: 57), the use of outher in reciprocal predicates is the main reason for the ‘confusion’ of either1 and outher: At any rate, the recurrent reciprocal usage constitutes the transitory stage between the two meanings. [my translation]11

In my view, the reciprocal construction should be regarded as an independent development, as it comes with rather specific context conditions. Still, the occurrence of outher with (actually or apparently) universal meanings may have contributed to the ‘absorption’ of outher by either.

11 “Jedenfalls bildet die so häufige reziproke Verwendung die Übergangsstufe zwischen den beiden Bedeutungen …”

From æghwæðer to either   

   93

Note that there are also occasional examples of universal outher in nonreciprocal contexts, as in (39), testifying to the ‘confusion’ in (Late) Middle English. (39)

On outher side that day gret blood was shad. ‘A lot of blood was shed on each side that day.’ [Lydg. FP (Bod 263) 8.722, ?a1439]

While either took over great parts of the distribution of outher in nonveridical contexts, in veridical contexts outher was replaced by another item, i.e. the numeral oon ‘one’. This numeral had been in use as an existential quantifier since Old English times in non-dual contexts – especially in partitive DPs (e.g. aan of þam ‘one of them’). We can, again, use an example from the Bible to illustrate the rise of oon as an existential quantifier in dual contexts. In (21) above we saw an example of oþer with the meaning ‘one of two’ from the Anglo-Saxon version of the Bible. The Wycliffe-Bible, translated at the end of the 14th century, has oon in this context (cf. 40). (40)

And Andrewe, the brother of Symount Petir, was oon of the tweyne, that herden of Joon, and hadden sued hym. ‘Andrew, Simon Peter’s brother, was one of the two who heard what John had said and who had followed Jesus.’ [Wycliffe-Bible, John 1, 40; c1390]

To summarize, the developments sketched in this section led to a gradual replacement of outher by two competitors: in veridical contexts outher was replaced by oon, and in clause-internal nonveridical contexts it was gradually ousted by either1. This change is shown in Table 4. context

veridical clause-internal

quantification scope Stage I

nonveridical clause-external

universal universal existential universal / narrow wide narrow wide ægðer

Stage II

existential existential narrow narrow

awðer either1

oon outher

Table 4: From Stage I to Stage II

/

veridical

94   

   Volker Gast

4.3 Scope-widening: From either1 to either2 According to the developments sketched in Section 4.2, either1 had not undergone any lexical-semantic changes in comparison to ægðer. It was still a universal quantifier, and its scope properties were (basically) ‘canonical’. either1 could take scope over a clause-internal nonveridical operator but not over a clause-external one. Even so, the distribution of either1 seems to have changed in comparison to OE ægðer – basically because of the decline of a competitor, i.e. outher, in contexts where both quantifiers were in principle available. In the further course of its development, either seems to have completely ‘absorbed’ outher in nonveridical contexts. outher is only rarely attested after the 15th century in most varieties of English.12 As a consequence of this process, either1 extended its distribution further and took over outher-contexts in which it could previously not be used. Specifically, it came to be used in the scope of a clause-external nonveridical operator. The most typical cases of clause-external nonveridical operators are probably provided by conditional clauses, as in (41). (41)

Had eydur of hom byn to lacke full evyll we had ben steyd. had either of them been to lack full evil we had ben stayed ‘Had either of them been absent, great evil we would have incurred.’ [Gowther, Adv 19, 3, 1; a1500]

Eydur – the form of either2 used in the relevant text – is either interpreted as a narrow-scope existential quantifier, or as a universal quantifier which takes scope beyond the conditional clause. The two interpretations are shown in (42). (42)

a. narrow-scope existential ∃x [ x is absent ] → we incur great evil b. wide-scope universal ∀x [ x is absent → we incur great evil ]

(41) is a context where in Old English we would have expected awðer within the conditional. According to the information available to us, OE ægðer would have been interpreted as a universal quantifier within the nonveridical context in such cases (cf. 16 above). (41) consequently illustrates the distributional extension that either underwent in Middle English, thus turning into either2.

12 Some (especially Northern) varieties have preserved forms based on outher even in the Modern language, e.g. West Yorkshire: Tha can tak aather on ’em ‘You can take either of them’; cf. Wright (1889–1905: 243).

From æghwæðer to either   

   95

In veridical contexts, either2 was still interpreted as a universal quantifier for most of the Middle English period. Two examples from the 15th century are given in (43) and (44). (43)

And so ayther knyghtes made them redy with two grete spearys. ‘And so both knights prepared with two great spears.’ [Malory, Works, Win-C, 517/24; a1470]

(44)

Ayther were armed on a stede. ‘Both were armed on a stud/stallion.’ [Sir Triamour, Cambridge Ff.2.38; a1500]

Unless we regard either2 as polysemous – with a universal interpretation in veridical contexts and with an existential interpretation in nonveridical contexts – we are bound to conclude that it could take scope beyond the local clause in examples like (41). In general, the scope of universal quantifiers is taken to be clausebound in Modern English, however (cf. May 1988; Fox and Sauerland 1996). If this is true of Middle English, too – which I assume – either2 thus had ‘exceptional’ scope properties, insofar as it scope was ‘exceptionally wide’. As I will argue in Section  4.4, the exceptional scope behaviour of either2 – in conjunction with the emergence of a universal quantifier with ‘canonical’ scope properties, i.e. bothe – led to the reanalysis of either as a narrow-scope existential quantifier. The distribution of quantifiers at Stage III can be represented as shown in Table 5, where bothe – the Middle English precursor of Modern English both – is integrated into the picture. context

veridical

quantification scope

universal universal / existential narrow wide narrow

Stage I

nonveridical clause-internal clause-external

ægðer

Stage II

veridical

universal / existential existential wide narrow narrow awðer

either1

oon outher

Stage III

bothe either2

Table 5: From Stage I to Stage III

oon

96   

   Volker Gast

4.4 The rise of bothe Just like outher, either was affected by the rise of competing quantifiers which took over part of its distribution. The (non-dual) distributive quantifier eech ( either. Anglia 15: 62–71.    Farkas, Donka F. 1992. On the semantics of subjunctive complements. In Romance Languages and Linguistic Theory, ed. P. Hirschbühler and K. Koerner, 69–104. Benjamins.    Fox, Danny, and Uli Sauerland. 1996. Illusive scope of universal quantifiers. In Proceedings of NELS, ed. K. Kusumoto, volume 26, 71–85. Amherst: GLSA.    Gast, Volker, and Johan van der Auwera. 2011. Scalar additive operators in the languages of europe. Language 87:2–54.    Giannakidou, Anastasia. 1998. Polarity Sensitivity as (Non)veridical Dependency. Amsterdam: John Benjamins.    Giannakidou, Anastasia. 2011. Positive polarity items and negative polarity items: Variation, licensing, and compositionality. In Semantics: An international handbook of natural language meaning, ed. K. von Heusinger Maienborn, C. and P. Portner, 1660–1712. Berlin: de Gruyter Mouton.

From æghwæðer to either   

   101

   Haspelmath, Martin. 2007. Coordination. In Language Typology and Syntactic Description, vol. ii: Complex Constructions., ed. Timothy Shopen, 1–51. Cambridge: Cambridge University Press, 2nd edition.    Heine, Bernd, and Hiroyuki Miyashita. 2008. The intersection between reflexives and reciprocals. In König and Gast (2008), 169–223.    Hintikka, J. 1980. On the any-thesis and and the methodology of linguistics. Linguistics and Philosophy 4: 101–122.    Hopper, Paul. 1991. On some principles of grammaticalization. In Approaches to grammaticalization, ed. E. Traugott and B. Heine, 17–36. Amsterdam: John Benjamins.    Huddleston, Rodney, and Geoffrey Pullum. 2002. The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press.    König, Ekkehard. 1991. The Meaning of Focus Particles. A comparative perspective. London: Routledge.    König, Ekkehard, and Volker Gast, ed. 2008. Reciprocals and Reflexives. Theoretical and Typological Explorations, volume 192 of Trends in Linguistics. Berlin: de Gruyter Mouton.    Ladusaw, William. 1979. Polarity sensitivity as inherent scope relations. Doctoral Dissertation, University of Texas, Austin.    May, Robert. 1988. Ambiguities of quantification. Linguistic Inquiry 19: 118–135.    Milsark, Gary. 1977. Towards an explanation of certain peculiarities of the existential construction in english. Linguistic Analysis 3: 1–29.    Nusser, Oskar. 1913. Geschichte der Disjunktivkonstruktionen in Englischen. Heidelberg: Carl Winter.    Plank, Frans. 2008. Thoughts on the origin, progress, and pronominal status of reciprocal forms in Germanic, occasioned by those of Bavarian. In König and Gast (2008), 347–373.    Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik. 1985. A comprehensive grammar of the English language. London: Longman.    Rullmann, Hotze. 2002. A note on the history of either. In Proceedings of the 38rd Meeting of the Chicago Linguistics Society, ed. Maria Andronis, Erin Debenport, Anna Pycha, and Keiko Yoshimura, volume 38, 111–125. Chicago: Chicago Linguistics Society.    Wright, Joseph. 1889–1905. English dialect dictionary. Oxford: Oxford University Press.    Wülfing, Ernst. 1894. Die Syntax in den Werken Alfreds des Großen. Bonn: Hanstein’s Verlag.    Zimmermann, Thomas Ede. 2000. Free choice disjunction and epistemic possibility. Natural Language Semantics 8: 255–290.    Zwarts, Frans. 1995. Nonveridical contexts. Linguistic Analysis 25: 286–312.    Zwarts, Frans. 1998. Three types of polarity. In Plurality and quantification, ed. Fritz Hamm and Erhard Hinrichs, 177–238. Dordrecht: Kluwer.

Johan Brandtler

Evaluability – An alternative approach to polarity sensitivity *

1 Introduction Often overlooked in discussions on polarity phenomena is the fact that ‘nonnegative’ contexts, such as yes/no-questions and conditionals, tend to host both positive and (weak) negative polarity items in identical syntactic settings. Consider the distribution of the NPI ever and the PPI still in (1) and (2) below. (1)

a. Have you ever been to Paris? b. Have you still that item for sale?

(2)

a. If you ever go to Paris, you must see the Eiffel Tower. b. If you still have that item for sale, I’m interested in buying it.

The pattern illustrated above is distinct from the well-known complementary distribution of PPIs and NPIs in declaratives, as shown in (3): (3)

a. I have never ever / *still been to Paris. b. I still / *ever have that item for sale.

Since the pioneering work of Jespersen (1917), almost every proposed theoretical explanation of the distribution of polarity items has taken the occurrence of polarity items in negative contexts as primary, while regarding the licensing properties of the non-negative contexts in (1) and (2) as exceptional. It has been a prevalent idea that NPIs depend on some kind of formal licensing requirement in order to occur within a clause, such as being in the syntactic and/or semantic scope of a licensor of relevant properties (i.e. negation, universal quantifiers, superlatives, the restriction of only etc.); see e.g. Ladusaw (1979, 1980), Linebarger (1980, 1987),

* The current work has been funded by a grant from Riksbankens Jubileumsfond, Sweden. The ideas presented in this paper have been developed over a number of years, and culminated in the publication of Brandtler (2012). I am grateful to numerous people for discussions, and would especially like to thank Elisabeth Engdahl, Larry Horn, Dianne Jonas, Valéria Molnár and Christer Platzack. Needless to say, all remaining errors and shortcomings are entirely mine.

104   

   Johan Brandtler

Progovac (1994), Giannakidou (1998 and onwards), and Horn (2002) for different proposals along this line. As yes/no-questions and conditionals may host polarity items even in the absence of an overt licensor, the search has been directed at finding an underlying syntactic/semantic licensing feature common to both negative and non-negative licensing contexts. This paper changes perspective, and argues that the occurrence of polarity items in yes/no-questions and conditionals is actually the unmarked case. Based on Brandtler (2012), it is proposed that polarity items are sensitive to evaluability, a concept that refers to the possibility of accepting or rejecting an utterance in a communicative exchange. Utterances (or parts of utterances) are either evaluable or non-evaluable. According to the Evaluability Hypothesis, non-evaluable utterances are natural hosts for polarity items. Hence, the occurrence of polarity items in these environments does not require any kind of formal licensing, and this is the reason we find both PPIs and NPIs in the yes/no-questions and conditionals in (1) and (2) above. Evaluable utterances, in contrast, are restricted environments and may only host polarity items that are formally licensed. Hence, NPIs require the presence of a licensing element, as in (3a), and PPIs require the absence of such elements, as in (3b). According to the Evaluability Hypothesis, then, formal licensing is regarded as a means of ‘rescuing’ the polarity item from an otherwise hostile environment. If this characterization is correct, polarity items can be defined as semantically sensitive expressions that sometimes must rely on a formal dependency relation in order to fulfill their semantic requirement. It is further argued that the notion of evaluability has structural correlates in Swedish, the language on which the discussion is primarily based: evaluable clauses have [Spec,CP], while non-evaluable clauses lack [Spec,CP]. This connection is argued to exist because of an arbitrary (but fixed) association between the edge-feature in C (see Chomsky 2008) and evaluability. The concept of evaluability is distinct from the semantic notion of veridicality (see Giannakidou 1998), and it is argued that the current proposal renders better empirical results than theories building on veridicality – at least in relation to the Mainland Scandinavian languages (Swedish, Danish and Norwegian) and English.1 Furthermore, the present theory incorporates the notion of downward entailment (Ladusaw 1979, 1980) in a natural manner, although its applicability

1 Even though the Evaluability Hypothesis should in principle be extendable to the other Germanic languages, it is beyond the scope of this paper to discuss cross-Germanic variation of polarity item licensing in any detail.

Evaluability – An alternative approach to polarity sensitivity   

   105

is restricted to evaluable environments. As a consequence, the main insight of the DE-approach is maintained under the present hypothesis. The paper is organized in the following way. In the next section, I present a brief overview of previous theories, and the problems associated with accounting for the non-complementary distribution of NPIs and PPIs in yes/no-questions and conditionals. Section 3 gives the empirical background of this study, showing that the licensing of polarity items in Swedish correlates in a striking manner with the configuration of the C-domain. In section 4, the notion of evaluability is defined and related to Swedish clause structure, and in section 5 evaluability is related to polarity sensitivity. The connection between evaluability, polarity sensitivity and the Swedish C-domain is explicated in section 6.

2 The Problem Although the distribution of polarity items is both empirically complex and theoretically evasive, one may roughly distinguish three different NPI-licensing patterns: (4)

1. Licensing by an overt clausemate licensor: I have never ever been to Paris. 2. Licensing by an overt superordinate licensor: I regret that I ever talked to him. 3. Licensing in the absence of an overt licensor: Have you ever been to Paris?

Within syntactically oriented approaches to polarity item licensing, the idea of polarity items as dependent expressions has been prevalent ever since Klima 1964: polarity items must be within the syntactic scope of an operator with relevant licensing properties. The obvious problem for any syntactic approach is how to account for licensing pattern 3 above, i.e. licensing in the absence of an overt licensor. The solution, as put forward in e.g. Progovac (1994), is to assume a covert anteceding operator that binds the polarity item in its governing domain. Under this view, the feature bundle responsible for licensing may be either overt or covert, much like the overt/covert realization of case, number, or gender etc. in the world’s languages. The syntactic operator approach fails to account for the non-complementary distribution of PPIs and NPIs, however. That is, if a covert licensing operator is

106   

   Johan Brandtler

responsible for the licit occurrence of ever in (1a), then the very same operator should, in principle, rule out the occurrence of the PPI still in (1b) – contrary to fact.2 What is more, without providing a proper definition of the assumed operator, any purely syntactic account becomes close to non-explanatory, even if it were to attain descriptive adequacy. Within semantically based approaches to polarity sensitivity, the most influential is that of Ladusaw (1979, 1980), according to which polarity sensitivity can be understood in terms of logical inferencing from sets to subsets, from the general to the specific. The prediction is that NPIs are licensed in any environment or by any operator that is downward entailing (DE). The denotation of a DEexpression is a monotone decreasing function, i.e. an order reversing function that allows inferences from sets to subsets. An appealing feature of Ladusaw’s proposal is that the complementary distribution of NPIs and PPIs fall out naturally: NPIs are licensed in downward entailing environments, whereas PPIs are licensed in upward entailing environments. Similar to syntactic operator approaches, however, the DE-hypothesis cannot account for licensing pattern 3 above, albeit for entirely different reasons. Since yes/no-questions and conditionals are non-monotone – i.e. they neither allow entailments from the specific to the general (upward) nor from the general to the specific (downward) – their ability to host polarity items cannot be explained by Ladusaw’s theory.3 In part conceived as a reaction to Ladusaw’s hypothesis, Giannakidou’s (1998 and onwards) Veridicality Hypothesis offers an alternative account of polarity sensitivity that does account for NPI licensing in yes/no-questions and conditionals. According to the Veridicality Hypothesis, weak negative polarity items, such as

2 Progovac (1994) does address the problem of PPIs in ‘non-negative licensing contexts’ in some detail. Assuming that the covert operator is located in C, Progovac argues that NPIs must raise to C at LF, while PPIs remain low and hence is outside the scope of the operator. While we need not go into details here, there are a number of both conceptual and empirical problems with such an analysis; the reader is referred to Horn and Lee (1995) and Brandtler (2012: section 7.2) for a critical discussion. 3 The non-monotone status of yes/no-questions and optatives should be self-evident: they have no truth-value and are consequently not open to truth-based inferencing. With respect to conditionals, the story is somewhat more complex. The DE-status of the antecedent of conditionals is not obvious, as the following example shows (taken from von Fintel 1999: 136): i. If John subscribes to newspaper, he must be well informed If John subscribes to a newspaper that he can’t read, he must be well informed. Discussing a number of similar examples, von Fintel (1999: 135 ff.) notes that “in the modern semantic and philosophical literature on conditionals, it is now taken for granted that conditionals are not monotonic in their antecedent”. ⇍

Evaluability – An alternative approach to polarity sensitivity   

   107

ever, any are licensed by nonveridicality, that is, by undetermined truth-values. Strong NPIs, in contrast, have a more limited distribution and are licensed by anti-veridicality, i.e. by clauses or operators entailing the falsity of p.4 One major advantage of Giannakidou’s theory is that polarity items are defined as semantically sensitive expressions, and need not rely on any form of syntactic licensing in order to be licit within a clause; her theory also explains the occurrence of NPIs in nonveridical contexts in a non-stipulative manner. But there are some severe problems associated with the Veridicality Hypothesis as well. First, similar to the syntactic account of Progovac, it does not account for the non-complementary distribution of PPIs and NPIs in conditionals and yes/no-questions. In fact, Giannakidou explicitly refrains from discussing the distribution of PPIs (1998: 19), thus making the predictions of the Veridicality Hypothesis restricted to NPIs. Second, her theory cannot without further stipulation account for licensing pattern 2 above, i.e. long-distance licensing by emotive factive predicates like regret. Since complements to factive predicates are veridical, the occurrence of embedded NPIs is predicted to be impossible, contrary to fact.5 Although distinctly different, the theories of Progovac, Ladusaw and Giannakidou have at least two things in common: i) they build on the seemingly uncontroversial assumption that polarity items are dependent expressions, and ii) they have problems accounting for the non-complementary distribution of NPIs and PPIs. In what follows, I will argue that problem ii) stems from assumption  i). According to the present proposal, polarity items are defined as semantically sensitive expressions that must rely on a formal dependency relation in certain environments only, namely in evaluable utterances. But before developing this idea further, let us first review the relevant empirical data, taken from Swedish.

4 A propositional operator F is veridical if and only if from the truth of Fp one can infer that p is true. Otherwise, it is nonveridical. Anti-veridical operators entail the falsity of p. Note that antiveridical operators form a subset of the nonveridical operators, since the logic inference Fp p is not valid for them either. 5 The Veridicality Hypothesis apparently makes the correct predictions for Greek, in which NPIs in factive complements are ungrammatical. Giannakidou (2006) suggests a solution to the grammatical English cases, which builds on additional contextual factors. ⇐

108   

   Johan Brandtler

3 Polarity sensitivity in Swedish In this section, I give a brief empirical presentation of the distribution of polarity items in Swedish. It is shown that Swedish displays a striking correlation between the structural configuration of the C-domain and the different ways polarity items may be licensed. As the Germanic languages in general (disregarding English), Swedish is a V2-language, meaning that the finite verb in declarative main clauses must be preceded by one clausal constituent only. Following standard theory, Swedish main clauses display generalized V-to-C movement: the finite verb must raise from its base position in V to C0. While V-to-C movement is found in all Swedish main clauses (with the exception of wh-exclamatives), [Spec,CP] need not be phonologically realized in all sentence types. Yes/no-questions, imperatives and optatives are linearly V1, meaning that the finite verb is not preceded by a clausal element, as shown in (6) below. The standard V2 word order is exemplified by the declarative in (5a), and its assumed structure illustrated in (5b): (5)

a. Antagligen gillar Sven inte hamburgare. Probably likes Sven not hamburgers ‘Sven probably doesn’t like hamburgers.’ b. [Spec,CP Antagligen [C0 gillar [Spec,TP Sven [NegP inte [VP Sven gillar hamburgare]]]]]

(6)

a. Gillar Sven inte hamburgare? likes Sven not hamburgers ‘Doesn’t Sven like hamburgers?’ b. [Spec,CP Ø [C0 gillar [Spec,TP Sven [NegP inte [VP Sven gillar hamburgare]]]]]

Subordinate clauses, in contrast, do not generally display V-to-C movement in Swedish. The complementizer is standardly assumed to reside in C0, effectively blocking the raising of the finite verb (see Platzack 1986). There are some systematic exceptions to this pattern, however, as complements to assertive and semi-factive predicates are well-known to allow so called embedded V2 (see e.g. Andersson 1975). Compare the standard (non-V2) subordinate clause in (7) with the embedded V2-clause in (8). Note that the raising of the finite verb affects the relative ordering between the verb and any clause adverbial.

Evaluability – An alternative approach to polarity sensitivity   

(7)

   109

a. att Sven inte gillar hamburgare. that Sven not likes hamburgers ‘…that Sven doesn’t like hamburgers.’ b. [Spec,CP Ø [C0 att [Spec,TP Sven [NegP inte [VP Sven gillar hamburgare]]]]]

(8)

a. att Sven that Sven

gillar inte hamburgare. likes not hamburgers

b. [Spec,CP att [Spec,CP Sven [C0 gillar [Spec,TP Sven [NegP inte [VP Sven gillar hamburgare]]]]]] Relating Swedish clause structure to polarity item licensing, I show in Brandtler (2012) that any Swedish clause overtly realizing [Spec,CP] displays licensing pattern 1: NPIs must be licensed by an overt clausemate licensor. This pattern is thus characteristic of declaratives (9a), wh-questions (9b), embedded V2-clauses (9c), non-restrictive relative clauses (9d), as well as wh-exclamatives:6 (9)

a. Jag vill *(aldrig) någonsin åka tillbaka. I want never ever go back ‘I never want to go back.’ b. Vilken känd artist turnerade *(aldrig) någonsin utanför USA? which famous artist toured never ever outside USA ‘Which well-known artist never ever toured outside the U.S.? c. Han sa, att han ville *(aldrig) någonsin åka tillbaka. he said that he wanted never ever go back ‘He said that he never ever wanted to go back.’

6 Two things are worth emphasizing here. First, wh-questions do allow NPIs even in the absence of an overt licensor, in which case they often have a rhetorical flavor. In Brandtler (2012: chapt. 10), I argue that NPI-licensing in such cases is dependent on the denotation of the whword: only when the denotation of the wh-word is a downward entailing function does it license polarity items. Second, wh-exclamatives do not allow polarity items at all. This behavior is curious in itself, but can perhaps be explained by the fact that exclamatives cannot be negated either. Hence, my claim that NPIs need to be formally licensed in wh-exclamatives might still hold, even though this requirement, for independent reasons, cannot be satisfied. See Zanuttini and Portner (2003) for a discussion on exclamatives, and Brandtler (2012: sections 4.2.2 and 6.1.4) for a discussion on the syntax and polarity sensitivity of Swedish wh-exclamatives.

110   

   Johan Brandtler

d. Sven, som för övrigt *(aldrig) någonsin lämnat Sverige, ska Sven who for other never ever left Sweden shall flytta till Kongo. move to Congo ‘Sven, who by the way never ever has left Sweden, is moving to Congo.’ Swedish linear V1-clauses (lacking overt realization of [Spec,CP]), on the other hand, display licensing pattern 3, i.e. they allow NPIs even in the absence of an overt licensor. This pattern thus holds for yes/no-questions, optatives and (the antecedent of) conditionals, similar to the English examples in (1) and (2) above. With regards to subordinate clauses, the story is somewhat more complex. In clauses introduced by the complementizer att ‘that’, NPIs are generally not grammatical unless licensed either by an overt clausemate or a superordinate licensor (licensing pattern 2). Clauses introduced by om ‘if’, however, readily allow NPIs even in the absence of an overt clausemate or superordinate licensor, and thus pattern with V1-clauses in adhering to pattern 3.7 This licensing difference can be mapped to structural configurations as well. Standard, non-V2 att-clauses in Swedish allow optional insertion of a cataphoric propositional pronoun det ‘it’ in between the finite verb and the complementizer. This possibility is indicative of [Spec,CP] in the subordinate clause, as illustrated in (10) below: (10)

a. Jag beklagar (det) att jag sårade dig. I regret it that I hurt you ‘I’m sorry for that I hurt you.’ b. Jag sa (det) att han förmodligen skulle I said it that he probably would ‘I told him he would probably lose.’

förlora. lose

It appears that the possibility of det-insertion is available in most assertive and factive predicate constructions in Swedish. A similar observation has been made for English, where complements to factive predicates may be introduced by the fact, which led Kiparsky and Kiparsky (1970) to assume a fact operator in

7 One may of course argue that the complementizer if is a lexical instantiation of a licensing operator. As already mentioned above, the fact that both NPIs and PPIs may occur in conditionals is highly problematic for such an analysis, as PPIs then must be allowed in spite of this very operator. An extensive critical discussion of the operator-approach is found in Brandtler (2012: 7.2.2).

Evaluability – An alternative approach to polarity sensitivity   

   111

[Spec,CP] in these clauses. In contrast, Swedish om-clauses does not allow cataphoric det-insertion, as the sentences in (11) show: (11)

a. Jag beklagar (*det) om jag sårade dig. I regret it if I hurt you ‘I’m sorry if I hurt you.’ b. Jag undrar (*det) om han kommer till festen. I wonder that if he comes to party.def ‘I wonder if he comes to the party.’

If the possibility of det-insertion is dependent on [Spec,CP], it seems reasonable to assume that [Spec,CP] is not available in om-clauses. Hence, att- and omclauses are structurally different with regards to the syntactic configuration of the C-domain: om-clauses cannot instantiate [Spec,CP], similar to (linear) V1-clauses. The structural similarity between V1-clauses and om-clauses is further strengthened by the fact that both may function as the antecedent of a conditional, as illustrated in (12) below: (12)

a. Om du ser Maria, hälsa henne från mig. if you see Maria greet her from me ‘If you see Maria, say hello from me.’ b. Ser du Maria, hälsa henne från mig. see you Maria greet her from me

Summarizing this discussion on Swedish, I have argued that att-clauses are structurally similar to V2 clauses in having a phonologically realizable [Spec,CP], while om-clauses are structurally similar to V1 clauses in lacking [Spec,CP]. Clauses with [Spec,CP] differ only with regards to whether the phonological instantiation of [Spec,CP] is obligatory or not. All in all, we may thus distinguish three configurations of the Swedish C-domain, based on the realization of [Spec,CP]. And intriguingly, these structural configurations can be mapped in a one-to-one fashion with the three distinct licensing patterns in (4) above. Consider (13): (13)

1. Licensing by an overt clausemate licensor: [Spec,CP] obligatorily realized

112   

   Johan Brandtler

2. Licensing by an overt superordinate licensor: [Spec,CP] optionally realizable 3. Licensing in the absence of an overt licensor: [Spec,CP] non-realizable Naturally, one must be careful when interpreting this correlation. It seems highly unlikely that the structural configuration per se affects the polarity sensitivity of the clause, as yes/no-questions, to mention one example, are well known crosslinguistically to host ‘unlicensed’ NPIs, irrespective of their syntactic structure. But the structural fact in Swedish that ‘unlicensed’ NPIs are licit in all clauses lacking [Spec,CP], but banned from all clauses with [Spec,CP] certainly calls for an explanation. The crucial question we need to focus on here is why: why does there exist a correlation between the structural configuration of the clause and its NPI-licensing properties? I propose that we should focus not so much on the structural configurations themselves, as on the very fact that there are structural differences between clauses with different licensing properties to begin with. If the syntactic classification mirrors a semantic distinction, we can assume that this semantic distinction has a bearing on polarity phenomena as well. In the next section, I argue that the relevant semantic property relevant for both polarity item licensing and the configuration of the C-domain is evaluability.

4 Evaluability Previous work on polarity phenomena has repeatedly explored the relation between negation and truth, and its relevance for polarity item licensing. The downward entailing approach of Ladusaw (1980) builds on the preserving/reversing of truth-value based inferences; the Veridicality Hypothesis as proposed by Giannakidou (1998) on the (non)availability of truth entailments. Even Progovac (1994) suggests that her polarity operator is somehow licensed by “unfixed truthvalues”. The notion of truth in formal logic builds on Frege’s classic conception of meaning: to know the meaning of a sentence is to know the conditions under which it is true. The judgment of a sentence as true or false thus hinges on the satisfaction of these truth-conditions. A slightly different picture of the relation between truth and falsity emerges if we change from the logico-semantic perspective to a pragmatic (or communicative) perspective.

Evaluability – An alternative approach to polarity sensitivity   

   113

Intuitively, what is at stake in conversation is not so much the logical or actual truth of a given proposition as the acceptance of that proposition as true. This distinction is important. According to Stalnaker (2002), accepting a proposition equals treating it as true, even in cases where one might suspect that it is actually not true. Observe the following passage from Stalnaker (2002: 716) (italics in original): To accept a proposition is to treat it as true for some reason. One ignores, at least temporarily, and perhaps in a limited context, the possibility that it is false. Belief is the most basic acceptance concept: the simplest reason to treat a proposition as true is that one believes that it is true. But there may be various reasons to ignore the possible situations in which some proposition is false even when one realizes that one of those possible situations may be the actual one. One may simplify or idealize in an inquiry, one may presume innocence to ensure fairness, one may make assumptions for the purpose of contingency planning, one may grant something for the purpose of an argument.

As Stalnaker points out above, it is perfectly possible to accept (i.e. to treat as true) a proposition that is demonstrably false in the actual world. This means that although the acceptance of a proposition is not entirely unrelated to the fulfillment of logical truth-conditions, it nevertheless builds on an additional set of contextual, social, and/or communicative considerations. At the core of discourse acceptance lies the notion of evaluability. That is, in order to accept or deny a proposition presented to us, we must evaluate the likelihood for it being true according to our beliefs, knowledge, inclination towards the speaker etc. In fact, we constantly evaluate all state-of-affairs presented to us – be it by observation, discovery, conversation or other kinds of perception. In reading this, you have probably already begun to evaluate whether my claim is correct or not. Consider the following passage from Lyons (1977: 777): As one rejects some physical entity that is offered (pushing it away so that it disappears or goes away (…)), so one may reject a proposition or a proposal. Looked at from this point of view (…) assent and dissent, rather than truth and falsity, would seem to be the notions with which we should operate in any account that we give of the difference between the assertion and the denial of p.

If we follow Lyons’s line of reasoning, the opposition between affirmative and negative propositions is directly related to our communicative need to evaluate and accept/reject information presented to us. Within a communicative exchange, however, one can only evaluate utterances used by the speaker to assert, presuppose, or entail the truth of p or ~p. Let us call such utterances evaluable.

114   

   Johan Brandtler

Importantly, both affirmative and negative utterances are equally evaluable. This claim rests on the Aristotelian view of negation as a mode of predication, and is at odds with the “standard” view within modal logic of negation as a truthfunctional connective. That is, instead of structurally treating ~p as “it is not the case that p”, ~p can be treated as “it is the case that not-p”. Under this view, a negative utterance does not assert the falsity of p but rather asserts (the truth of) not-p; cf. the discussion in Lyons (1977: 768) and Allen (2006: 5). This is why it is just as easy for the addressee to accept a negative utterance as true or false, as it is to accept an affirmative utterance.8 Contrasting with evaluable utterances, we find utterances by which the speaker does not assert, presuppose, or entail the truth of p or ~p. Such utterances (or part of utterances) cannot be evaluated, as the addressee cannot assent or dissent with something to which the speaker is not truthfully committed. We call such utterances non-evaluable. For example, it is not possible to accept or reject a yes/no-question, since by uttering such a question the speaker does not take responsibility for the truth of neither p nor ~p. In short, my notion of evaluability refers to the possibility of accepting or rejecting an utterance (or part of an utterance) as true. Note the focus on possibility here; the actual acceptance or rejection is of no relevance. Neither is the logical possibility of establishing the truth of a given proposition in relation to possible worlds or epistemic models. Hence, the concerns of modal logic are independent from the notion of evaluability. This is important, since it allows us to move away from truth-conditional inferencing. The question of whether a proposition is true or false does not arise, only the question of whether it is possible to accept that proposition as true in discourse. The actual evaluation process can take slightly different forms, however. One may distinguish between (at least) two kinds of evaluable utterances: i) those that are subjected to evaluation in the current (on-going) communicative exchange, and ii) those that are not subjected to evaluation in the on-going communicative exchange. I believe this distinction may capture the difference between, for instance, assertions and presuppositions in an intuitive and straightforward manner. Both assertions and presuppositions are evaluable notions, as they are used by the speaker to affirm the truth of p or ~p. On the view famously defended

8 Horn (1989 [2001]: chapt. 7) argues explicitly against the view of natural language negation as being an external truth-functional connective, based on the observation that “syntactically external (clause-peripheral) negation, as an iterating one-place connective on propositions, never – or hardly ever – happens [in the world’s languages]”, Horn (1989 [2001]: 471).

Evaluability – An alternative approach to polarity sensitivity   

   115

by Stalnaker (1978: 153), the essential effect of an assertion “is to change the presuppositions of the participants in the conversation by adding the content of what is asserted to what is presupposed.” If the assertion is accepted by the hearer/s, the speaker has succeeded in adding new information to the conversational common ground. In this way, a (successful) assertion reduces the set of possible worlds so that “all of the possible situations incompatible with what is said is eliminated” (ibid). According to the Stalnakerian view, the relation between assertion and presupposition can be summarized in the following way: a (successful) assertion becomes part of the conversational common ground (i.e. the presuppositions shared by speaker and hearer), therefore a presupposition is basically nothing but a ‘previously accepted assertion’. Naturally, this claim should not be taken too literally. A presupposed proposition need not have originated as an actual assertion in a given discourse, but it is crucially treated as though it had. In terms of evaluability, assertions and presuppositions can be said to differ only with regards to when and how the evaluative process takes place, as it were. Asserted information may be accepted or challenged by the hearer, and is as such subjected to evaluation in the on-going discourse. In contrast, presupposed information is treated as uncontroversial or already accepted information, and is as such not subjected to evaluation in the on-going discourse. But crucially, this does not mean that a presupposition is non-evaluable. It does affirm the truth of p or ~p. The basic difference between the two notions is simply that an assertion is subjected to evaluation at the time of utterance, whereas a presupposition has been subjected to evaluation at a time prior to the time of utterance, or at least functions as though it has been. This follows naturally from the Stalnakerian view that the presuppositions of the conversational common ground are but previously accepted assertions. Let us now illustrate the ‘evaluability classification’, according to which the evaluable category has been divided into two subcategories in order to capture inter alia the difference between assertions and presuppositions: (14)

Utterances Evaluable Subjected

Non-evaluable Not subjected

The Evaluability Classification

116   

   Johan Brandtler

Importantly, the two evaluable subcategories do not equal assertion and presupposition, respectively. In the “subjected to”-class, we find not only main clause declaratives, but also sentence types which are not generally used to make assertions, such as wh-exclamatives, wh-questions and non-restrictive relative clauses. Although distinctly different, these sentence types are all possible for the hearer to evaluate (i.e. accept or reject) in the on-going communicative exchange.9 The “not subjected to”-category contains that-clauses functioning as complements to factive and non-assertive predicates, as well as restrictive relative clauses and Swedish att-exclamatives.10 Hence, these sentence types express propositions that are either presupposed or backgrounded, in the pragmatic sense of being neither asserted nor presupposed. The non-evaluable category, finally, contains sentence types which do not express any commitment on part of speaker as to the truth of p or ~p, such as (the antecedent of) conditionals, yes/no-questions and optatives. For further discussion and motivation of the evaluability classification, the reader is referred to Brandtler (2012: ch. 6). As the attentive reader no doubt has already concluded, the evaluability classification can be mapped in a one-to-one fashion with the three structural configurations in Swedish as distinguished in the previous section. Consider figure (15):

9 A wh-exclamative may be challenged by the hearer, as a reply like I certainly have not! is felicitous in relation to the utterance What big feet you have!. Non-restrictive relative clauses display a number of syntactic and semantic properties associated with main clause declaratives (see Hooper and Thompson (1973)), including the possibility of being challenged by the hearer. According to the evaluability classification, the controversial issues surrounding the question of whether subordinate clauses can be asserted is avoided, which in itself is a theoretical argument for assuming the present categorization. Wh-questions, finally, implicate open propositions that are cancelled when answered negatively: the answer Nobody effectively cancels the implication of the question Who was late for work? that somebody was late for work. 10 Having no exact counterpart in English, Swedish att-exclamatives (‘that’-) have the form of subordinate att-complements, while functioning as main clause exclamatives: (1) Att du aldrig kan komma I tid! that you never can be in time ‘I’m amazed/surprised that you’re never on time!’

Evaluability – An alternative approach to polarity sensitivity   

(15)

   117

Sentences Evaluable

Subjected | Overt [Spec,CP]

Non-evaluable | No [Spec,CP]

Not subjected | Covert [Spec,CP]

The Evaluability Classification and Swedish Clause Structure Again, when analysing this correlation between clause structure and evaluability, one should probably focus more on the structural distinctions than on the actual structural configurations. The connection between [Spec,CP] and evaluability is in all likelihood language specific, and evaluability may be mirrored differently in other languages, much like the morpho-syntactic marking of such pragmatic concepts as topic/focus, politeness and modality is different crosslinguistically. This is important, as arguing that [Spec,CP] is inherently related to evaluability has rather far-reaching implications, and furthermore blurs the distinction between syntax, semantics and pragmatics. I will return to this issue in section 6 below. Having established the notion of evaluability and consequently mapped different sentence types to it, our next step is to relate evaluability to polarity sensitivity.

5 Evaluability as polarity sensitivity In section 3, we observed an intricate correlation in Swedish between polarity sensitivity and the structural configuration of the C-domain. In the previous section, I argued that the Swedish C-domain mirrors the evaluative status of the clause. Consequently, it seems reasonable to assume a connection between evaluability and polarity sensitivity as well, as this may account for our initial observation. By combining the evaluability classification, the configurations of the Swedish C-domain and polarity sensitivity, we get the following schema:

118   

   Johan Brandtler

(16)

Utterances Evaluable [+[Spec,CP]] Subjected | Licensing Pattern 1

Not subjected | Licensing Pattern 2

Non-evaluable [–[Spec,CP]] | Licensing Pattern 3

The Evaluability Classification and NPI-Licensing As illustrated in (16), each licensing pattern is related to a distinct syntactic configuration and is characteristic of a well-defined semantic class. Let us consider the implications of these connections. First, non-evaluable clauses host NPIs even in the absence of overt licensors, and PPIs and NPIs are not in complementary distribution. Second, evaluable clauses only allow formally licensed NPIs, and PPIs and NPIs are in complementary distribution. Taken at face value, these differences make it difficult to draw one generalized conclusion concerning the licensing and distribution of polarity items. From the illustration in (16), the correct conclusion to draw seems to be that polarity items only sometimes are syntactically and/or semantically dependent, namely in evaluable clauses. In non-evaluable clauses, their occurrence is free and hence does not rely on syntactic licensing. But this may come across as a rather void conclusion, as it is but a theoretically disguised description of the empirical observations. Furthermore, by opting for this interpretation, one must simultaneously abandon the ‘standard’ assumption that polarity items are necessarily dependent expressions. But perhaps this is not a bad thing after all, considering the problems previous theories have had in accounting for the noncomplementary distribution of PPIs and NPIs (recall section 2). The Evaluability Hypothesis presents an alternative approach that does account for the evasive syntactic and semantic dependency behavior of polarity items. Using a simple metaphor, we may equate polarity items in non-evaluable clauses with fish in water: they occur in their right biosphere, as it were. Therefore, they need not rely on any kind of formal (syntactic) licensing. The fact that NPIs and PPIs equally occur in these environments falls out naturally. In contrast, polarity items in evaluable clauses are like fish on land. In order to survive in their unnatural environment, the PI must be rescued much like a fish out of water must be put within the confinement of a water tank. This is where both syntactic scope and semantic truth-based inferencing in the sense of Ladusaw comes in.

Evaluability – An alternative approach to polarity sensitivity   

   119

It should be emphasized that the distributional difference between strong and weak NPIs is not problematic for the Evaluability Hypothesis. One of the keys to polarity phenomena lies in acknowledging both the nature of the hosting environment and the nature of the hosted item. Some environments are inherently hostile to all kinds of polarity items, some are inherently accepting them. Simultaneously, some polarity items are inherently choosy about their hosts (e.g. untilclauses), while others thrive in a large number of contexts (e.g. any, någonsin ‘ever’). This fact does not in itself lessen the explanatory adequacy of the Evaluability Hypothesis, since it applies to the hosting environment only. Thus, the observation that strong NPIs always require overt licensing points not to a weakness of the theory, but to a semantic property of the lexical item. Importantly, the current proposal constitutes a change of perspective from previous theories. Ever since Klima (1964), work on polarity has been set on finding one, single syntactic or semantic property relevant for licensing; preferably one that is associated to negation. According to my proposal, however, the distribution of polarity items is better thought of in terms of sensitivity. Furthermore, there is no immediate connection to negation or negative expectations. Syntactic or semantic licensing can in light of the present proposal be regarded as secondary: it is only required in those environments which do not inherently fulfill the sensitivity requirements for polarity items. This idea is explicated in the next subsection in relation to Ladusaw’s DE-hypothesis; in the following subsection, the current proposal is related Giannakidou’s notion of veridicality.

5.1 Evaluability and monotonicity As discussed in section 2, the main criticism against Ladusaw’s (1980) notion of downward entailment was that it cannot be extended to yes/no-questions and conditionals. But what if this ‘weakness’ is not really a weakness at all, but rather a natural consequence of the fact that the notion is not applicable to (or of no relevance for) these environments? In light of the present proposal, one may argue that the DE-hypothesis is non-applicable to a well-defined class of utterances. Intriguingly, the sentence types for which the DE-hypothesis does not account correlate in a one-to-one fashion with our non-evaluable category: the antecedent of conditionals, yes/ no-questions, and om-complements in general (e.g. the antecedent of counterfactual conditionals and embedded yes/no-questions). Hence, one might argue that the DE-hypothesis fails to explain the distribution of polarity items in nonevaluable utterances, simply because it only applies to evaluable utterances. And the reason for this limitation is straightforward.

120   

   Johan Brandtler

In order for an utterance to be evaluable, the speaker must take responsibility for the truth of the expressed proposition (be it affirmative or negative); otherwise an utterance is non-evaluable. Now, since only that which has a truth-value can be subjected to truth-based inferencing, it follows that only evaluable utterances contain propositions that are monotone. Non-evaluable utterances are necessarily non-monotone. The connection between evaluability and monotonicity can thus be captured as in (17) below. (17)

i. Non-evaluable utterances cannot be challenged in discourse, since the speaker does not take responsibility for their truth. ii. For this reason, non-evaluable utterances do not allow truth-based inferencing: they are non-monotone. iii. Consequently, the monotonicity principles are not applicable to nonevaluable utterances.

If this line of reasoning is on the right track, the fact that Ladusaw’s DE-hypothesis cannot account for the occurrence of NPIs in yes/no-questions should not be taken as a ‘weakness’, as it is ultimately a consequence of the fact that yes/noquestions are non-evaluable. That is, polarity items may occur freely in all nonevaluable environments precisely because they are non-monotone. Evaluable environments, in contrast, are either monotone increasing or decreasing, and, depending on which, only PPIs or NPIs will be licensed. Hence, the DE-hypothesis may be subsumed under the Evaluability Hypothesis, but its applicability is restricted to the evaluable class.

5.2 Evaluability and veridicality The most important difference between the concepts of evaluability and veridicality concerns the view on negated sentences. According to the Veridicality Hypothesis, a negative declarative asserts the falsity of p; hence a negative declarative pattern with other nonveridical clauses which do not assert/presuppose the truth of p. According to the Evaluability Hypothesis, a negative declarative asserts the truth of not-p, from which it follows that both p and ~p are semantically distinct from sentences which are not used to claim a truth-value, such as yes/no-questions and conditionals. These opposing perspectives ultimately make the Evaluability Hypothesis and the Veridicality Hypothesis incompatible with each other. With regards to the distribution of NPIs, the Veridicality Hypothesis presents a rather promising solution, as mentioned above. Under the assumption that

Evaluability – An alternative approach to polarity sensitivity   

   121

(weak) NPIs are sensitive to nonveridicality, it is expected to find them in both negative declaratives and yes/no-questions. What is forgotten, it seems, is that PPIs also occur in yes/no-questions and conditionals, but not in negative declaratives. Hence, one must conclude that PPIs are insensitive to veridicality, as they are found in both veridical and nonveridical environments. Put somewhat differently, one might perhaps say that PPIs are anti-licensed by anti-veridicality: they may occur in any environment unless it is anti-veridical. Furthermore, as discussed in section 2, the Veridicality Hypothesis incorrectly predicts complements to emotive factive predicates to disallow NPIs, as they are presupposed and therefore per definition veridical. Ultimately, this means that the distribution of polarity items cuts right across the veridicality border. In fact, if we assume that polarity items are semantically dependent expressions that are sensitive to the veridical status of the hosting clause, the distribution of PPIs in nonveridical contexts and NPIs in veridical contexts all but falsifies this hypothesis. In comparison, the Evaluability Hypothesis successfully explains both the non-complementary distribution of PPIs and NPIs as well as occurrences of NPIs in veridical contexts. Rather than seeing polarity items as dependent on the absence of a truth-entailment, I propose that they are sensitive to the evaluative status of the clause. Under this perspective, the truth of p is no longer primary; instead it is the act of claiming a truth-value that is argued to be the decisive factor in accounting for the distribution of polarity items. At least with regards to the Mainland Scandinavian languages, this seems to be a more correct characterization than that of the Veridicality Hypothesis.

6 Syntactic arguments In the previous sections, I have defined the notion of evaluability and argued that it can be successfully used to understand polarity sensitivity. A reasonable question to ask at this point is what independent evidence there is for assuming that a notion like evaluability is relevant for grammar. In this section, I suggest a syntactic account of the connection between evaluability, polarity sensitivity and the configuration of the Swedish C-domain. Let us begin by discussing the position claimed to be relevant for evaluability, that is, [Spec,CP]. So far, I have assumed without arguments that linear V1-clauses and om-clauses lack [Spec,CP] altogether. A reader familiar with the ‘standard’ syntactic account of Scandinavian Languages may find this analysis unorthodox. Under the assumption that C hosts an uninterpretable EPP-feature, the presence of [Spec,CP] is normally assumed also in V1-clauses. Under this view,

122   

   Johan Brandtler

linear V1- and V2-clauses only differ with regards to the phonological realization of [Spec,CP]. Structurally, the finite verb is always in second position, being preceded either by an overt clausal element or a covert syntactic operator (see Holmberg and Platzack 2005). In Brandtler (2012), I argue against this view in line with Platzack (2008, 2009), who proposes abandoning the EPP-analysis of the Swedish C-domain in favor of an edge-feature analysis. According to Chomsky (2008), the edge-feature in C can be understood as a syntactic requirement that a phase head must have a specifier. Consequently, the realization of [Spec,CP] is not motivated by the valuation of a particular set of features (such as the EPP), and this affects the visibility condition of [Spec,CP]: “EPP is a demand that an Agree-relation must be visible at the SM interface, the edgefeature a demand that a phase head must have an Ā specifier. Hence, visibility at SM holds for EPP but not necessarily for the edgefeature” Platzack (2008: 7). Following the edge-feature analysis, the syntactic derivation of [Spec,CP] is thus dependent on the presence of an edge-feature in C. Unless the edge-feature is instantiated, [Spec,CP] cannot be realized. The critical reader may object to the present characterization, arguing that the theoretical status of the notion edge-feature remains vague and therefore stipulative. While I am the first to admit that the explanatory adequacy of the EFanalysis is debatable, it is descriptively more accurate than the EPP-analysis. And the very fact that EF is semantically or pragmatically void should not be seen as a weakness, however, since we should not attribute meaning to a mechanism, but to its output.11 Being in itself semantically vacuous, the edge-feature may instead be associated to semantic content within a given language. And this, I would argue, is the reason evaluability is reflected in the presence/ absence of [Spec,CP] in Swedish: it is a direct consequence of the association between the pragmatic-semantic notion of evaluability and the edge-feature in C. And this assumption, in turn, also accounts for the observed correlation between polarity sensitivity and the configuration of the Swedish C-domain. Assuming that [Spec,CP] is dependent on the instantiation of the Edge-Feature in C, the connection between polarity sensitivity, evaluability and the structural configuration of the C-domain can be summarized in the following way.

11 One should keep in mind that the EPP may be criticized on similar grounds for being a rather abstract concept with dubious explanatory power.

Evaluability – An alternative approach to polarity sensitivity   

(18)

   123

i. Polarity items are sensitive to evaluability; ii. Evaluability is associated to the edge-feature in C; iii. The edge-feature in C is responsible for the syntactic derivation of [Spec,CP]. iv. Conclusion: The NPI-licensing properties of a Swedish clause correlate with the structural realization of [Spec,CP].

If this hypothesis is correct, the observed correlation between [Spec,CP] and polarity item licensing is language specific. It comes about as a result from the arbitrary association in Swedish between the edge-feature in C and evaluability. Hence, we cannot draw any universal implications from this observation. However, the semantic connection between evaluability and polarity sensitivity may be tested against any other language. The strict modular approach as outlined here thus gives that the syntactic association is language-specific, whereas the semantic connection may be universal (or at least applicable to a large number of languages).

7 Summary Let me summarize the theoretical implications of the Evaluability Hypothesis, and its relation to previous accounts of polarity item licensing. The Evaluability Hypothesis suggests an alternative explanation to polarity sensitivity as a phenomenon, which does not build on the opposition between affirmative and negative sentences, or on the distinction between truth and falsity. Instead, it builds on the possibility of an utterance being accepted as true or false within a communicative exchange. Evaluable utterances are truly polar, since they express either the truth of p or the truth of ~p. It is irrelevant whether the expressed proposition is true or false in the absolute sense of propositional or modal logic; it must only be possible to accept it as true in a communicative exchange. In contrast, non-evaluable utterances are non-polar, since they neither express the truth of p nor the truth of ~p. For this reason, they cannot be accepted or rejected in discourse. Since non-evaluable utterances are non-polar, they are also non-monotone. This means that both NPIs and PPIs may occur in identical syntactic and semantic environments. From this characterization, so called ‘non-negative licensing contexts’ (such as yes/no-questions and conditionals) have nothing in common with ‘negative

124   

   Johan Brandtler

licensing contexts’ (such as negative declaratives) – in fact, they display quite opposing semantic properties. This conclusion is in stark contrast with former theories, which have sought to unite negative and ‘non-negative’ licensing contexts. According to the Evaluability Hypothesis, non-evaluable utterances are bona fide environments for polarity items, whereas evaluable utterances are exceptional. We have seen that the distinction between evaluable and non-evaluable utterances is structurally mirrored in Swedish. Evaluable clauses are endowed with an edge-feature in C, while non-evaluable clauses lack an edge-feature in C. At ‘surface structure’, this difference is reflected by the presence or absence of [Spec,CP]. Hence, the Evaluability Hypothesis straightforwardly accounts for the observed correlation between clause structure and the distribution of polarity items in Swedish – the start of the entire study. None of the previous theories of polarity sensitivity provides a viable account of this observation. Admittedly, the general applicability of the Evaluability Hypothesis remains to be tested against other languages. However, I believe that the present study provides a promising alternative account of polarity phenomena, that both challenges and incorporates previous theories put forward in the literature.

8 References Allen, Keith. 2006. Clause-type, primary illocution, and mood-like operators in English. Language Sciences 28: 1–50. Andersson, Lars-Gunnar. 1975. Form and function of subordinate clauses. Doctoral Dissertation, University of Gothenburg, Gothenburg. Brandtler, Johan. 2012. The Evaluability Hypothesis. The Syntax, Semantics and Pragmatics of Polarity Item Licensing. Amsterdam: John Benjamins. Chomsky, Noam. 2008. On phases. In Foundational Issues in Linguistic Theory: Essays in honor of Jean-Roger Vergnaud, ed. by Robert Freidin, Carlos P. Otero, and Maria Luisa Zubizarreta, 133–166. Cambridge, MA: MIT Press. von Fintel, Kai. 1999. NPI licensing, Strawson entailment and context dependency. Journal of Semantics 16: 97–148. Giannakidou, Anastasia. 1998. Polarity sensitivity as a (non)veridical dependency. Number 23 in Linguistik Aktuell. Amsterdam: John Benjamins. Giannakidou, Anastasia. 2006. Only, emotive factive verbs and the dual nature of polarity dependency. Language 82: 575–603. Holmberg, Anders, and Christer Platzack. 2005. The Scandinavian languages. In Comparative syntax, ed. by Guglielmo Cinque and Richard S. Kayne, 420–458. Oxford: Oxford University Press. Hooper, Joan, and Sandra Thompson. 1973. On the applicability of root transformations. Linguistic Inquiry 4: 465–497.

Evaluability – An alternative approach to polarity sensitivity   

   125

Horn, Laurence R. 1989 [2001]. A natural history of negation. Stanford: CLSI Publications. Horn, Laurence R. 2002. Assertoric inertia and NPI-licensing. Chicago Linguistic Society 38: 55–82. Horn, Laurence R., and Young-Suk Lee. 1995. Progovac on polarity. Journal of Linguistics 31: 401–424. Jespersen, Otto. 1917. Negation in English and other languages. Copenhagen: Bianco Lunos Bogtrykkeri. Kiparsky, Paul, and Carol Kiparsky. 1970. Fact. In Progress in linguistics, ed. by Manfred Bierwisch and Karl Erich Heidolph, 313–341. The Hague: Mouton. Klima, Edward. 1964. Negation in English. In The structure of language, ed. by J Fodor and J Katz, 246–332. Englewood Cliffs, NJ: Prentice-Hall. Ladusaw, William. 1979. Polarity sensitivity as inherent scope relations. Doctoral Dissertation, University of Texas at Austin, Austin, Tx. Ladusaw, William A. 1980. On the notion Affective in the analysis of negative-polarity items (2002). In Formal semantics. The essential readings, ed. by Paul Portner and Barbara H. Partee, 457–470. Blackwell Publishing. Linebarger, Marcia. 1980. The grammar of negative polarity. Doctoral Dissertation, MIT, Cambridge, MA. Linebarger, Marcia C. 1987. Negative polarity and grammatical representation. Linguistics and Philosophy 325–87. Lyons, John. 1977. Semantics, volume 1, 2. Cambridge: Cambridge University Press. Platzack, Christer. 1986. The position of the finite verb in Swedish. In Verb second phenomena in Germanic languages, ed. by Hubert Heider and Martin Prinzhorn, 27–47. Dordrecht: Foris Publications. Platzack, Christer. 2008. The Edge Feature on C. Unpublished manuscript, October 2008. Platzack, Christer. 2009. Old wine in new barrels: The Edge Feature on C, topicalization and stylistic fronting. Talk presented at the Mailing Seminar, University of Reykjavik, November 30–December 1, 2009. Progovac, Ljiljana. 1994. Negative and positive polarity. Cambridge: Cambridge University Press. Stalnaker, Robert C. 1978. Assertion. In Syntax and semantics, volume 9, 315–332. New York: Academic Press. Stalnaker, Robert C. 2002. Common ground. Linguistics and Philosophy 25: 701–721. Zanuttini, Raffaella, and Paul Portner. 2003. Exclamative clauses: at the syntax-semantics interface. Language 79: 39–81.

Luka Crnič

How to get even with desires and imperatives *

1 The meaning and distribution of even The primary semantic import of the focus particle even, which we assume is adjoined at a clausal level at LF, is a scalar presupposition that orders the prejacent of even with respect to the alternatives on which it operates.1 There have been various proposals about the flavor and the quantificational strength of this presupposition (Karttunen & Peters 1979, Kay 1990, Merin 1999, Herburger 2000 and many others). The choice between them is inconsequential for the discussion in this paper. For concreteness, we assume that the scalar presupposition triggered by even requires the likelihood of its propositional argument to be lower than that of a relevant alternative (Bennett 1982, Kay 1990): (1)

⟦ even ⟧g,c(C, p, w) is defined only if ∃q ∈ C [p ⊳c q]. If defined, ⟦ even ⟧g,c(C, p, w) = 1 iff p(w) = 1.

The sentence in (2a) has the structure in (2b) where even associates with the focused element John (focal stress is indicated by capitalization). The sentence presupposes that there is an alternative to the sister of even that is more likely than the proposition that John made one video, say, that Paul made one video, while its assertive meaning is that John made one video. (2)

a. Even JOHN made one video. b. [even C1] [JohnF made one video]

* Special thanks to Gennaro Chierchia, Kai von Fintel, Danny Fox and Irene Heim for discussion as well as to the audiences at the Göttingen workshop on polarity, WCCFL 29 and SALT 21. Thanks also to two anonymous reviewers of the present volume on NPIs. 1 Another inference that is often assumed to be triggered by even is the additive (or existential) presupposition. Since additivity is largely tangential to the purposes of this paper, we leave it aside and refer the reader to Rullmann 1997, Guerzoni 2003 and the references cited therein.

128   

   Luka Crnič

c. ⟦ (2b) ⟧g,c(w) is defined only if ∃q ∈ {that x made one video | x is a relevant individual}: that John made one video ⊳c q. If defined, ⟦ (2b) ⟧g,c(w) = 1 iff John made one video in w.

1.1 Scalarity, entailments, and the scope of even The scalar presupposition triggered by even is subject to the principle in (3), which follows from basic probability theory.2 The principle imposes a hard condition on the distribution of even: if even is adjoined to a clause whose alternatives entail it, it will trigger a presupposition that violates (3) and is thus unsatisfiable. (3)

Scalarity and entailment If a proposition p entails a proposition q, q cannot be less likely than p.

An illustration of the principle is in (4a) where even associates with the weak element one in its immediate surface scope, i.e. it associates with an element that is entailed by its alternatives. We call such occurrences of even ‘weak even’ for brevity. The sentence presupposes that it is less likely that John made one video than that he made some other number of videos (4c). Since making n videos entails making one video for all n>0, all the alternatives in the domain of even entail its prejacent. According to (3), none of them can be more likely than it. This clashes with (4c) and explains the deviance of (4a). (4)

a. # John made even ONE video. b. [even C1] [John made oneF video] c. ⟦ (4b) ⟧g,c is defined only if ∃q ∈ {that John made n videos | n ∈ ℕ>0}: that John made one video ⊳c q iff for some n>1: that John made one video ⊳c that John made n videos. (# in all contexts)

A puzzle emerges when we look at certain embedded occurrences of even – in particular, at occurrences of even in the scope of downward-entailing operators.

2 More precisely, it follows from Kolmogorov’s third axiom that states that the likelihood of a union of mutually exclusive propositions equals the sum of the likelihoods of the propositions. Applied to (3), if a proposition p entails a proposition q, it holds that the sum of the likelihoods of p and q\p equals the likelihood of q. Since the likelihood of q\p is greater or equal to zero, it holds that the likelihood of p is at most as great as the likelihood of q.

How to get even with desires and imperatives   

   129

A pertinent example is in (5). If even were interepreted in situ in this sentence, it would trigger the same scalar presupposition as in (4); the sentence would accordingly be pragmatically deviant. (5)

John didn’t make even ONE video.

To explain the contrast between (4) and (5), Lahiri (1998) builds on the approach to even by Karttunen & Peters (1979), according to which even may move at LF. A rescue hatch is available to even in negative sentences like (5) – it may covertly move above negation where it triggers a consistent scalar presupposition. That is, Lahiri assumes that the sentence in (5) can have the LF in (6a) where there is an intervening entailment-reversing operator between the moved even and its associate one. The presupposition of this structure is that it is less likely that John didn’t make one video than, say, that John didn’t make two videos (6b). Since the former proposition is logically stronger than the latter, the presupposition is compatible with (3) and may very well be correct. (6)

a. [even C1] [not [even C1] [John make oneF video]] b. ⟦ (6a) ⟧g,c(w) is defined only if for some n>1: that John didn’t make one video ⊳c that John didn’t make n videos. (P in plausible contexts)

1.2 A more general prediction The core ingredient of Lahiri’s (1998) proposal is the assumption that even may move to avoid triggering an incorrect presupposition. He has shown that if even that associates with a weak element in its immediate surface scope moves above negation or other downward-entailing operators, the resulting structure can have a consistent and plausible interpretation since downward-entailing operators reverse entailments. However, moving even across an operator that reverses entailments is not a necessary but a sufficient condition for configurations containing weak even to comply with (3). The necessary condition is that even moves across an operator that is not upward-entailing, i.e. a downward-entailing or a non-monotone operator.3

3 As pointed out by Rullmann (1997), a patent issue for an approach that assumes that even can move at LF is that it must allow for non-canonical movement of even, e.g. even may move out of the antecedent clause of a conditional. A possible way to defuse this issue has been suggested by Lahiri (2006): the scalar presupposition that accompanies sentences with even

130   

(7)

   Luka Crnič

A prediction of the movement approach to even A sentence with a weak even is acceptable only if even is at surface structure in the scope of a non-upward-entailing operator.

2 Weak even in desire statements and imperatives Weak even may occur in the immediate surface scope of non-negative desire predicates and in imperatives. This is at first sight at odds with the prediction in (7): there appears to be no non-upward-entailing operator in these sentences across which even could move at LF. Accordingly, it should trigger an incorrect scalar presupposition.

2.1 Non-factive desire predicates Weak even may occur in the scope of non-negative desire predicates like hope and would like. A few naturally occurring examples of this kind are given in (8) where we boldface the embedding desire predicate and capitalize the focused element. (8)

a. I hope to someday make even ONE video of that quality. b. In fact, I would like to find even ONE person who hasn’t had at least one paper-cut this year. People without hands do not count. c. Everyone wants to prevent even “ONE bad incident” as BellaQuest points out.4

Another conspicuous instance of a non-factive desire predicate that licenses weak even is wish. Two naturally occurring examples are given in (9).

is triggered by an even-like operator (EmphAssert in Krifka 1995, E in Chierchia 2006) that is attached at a clausal level to check the relevant feature of even; even itself is truth-conditionally vacuous, it stays in situ, and primarily conditions the insertion of the even-like operator. In the following we retain for perspicuity the assumption that even itself moves. 4 http://vimeo.com/8918647, http://www.murdershewrites.com/tag/natalie-r-collins/http:// www.people.com/people/archive/article/0,,20141718,00.html

How to get even with desires and imperatives   

(9)

   131

a. I wish I wrote even ONE riff featured on ANY Death album. b. This Haley chick wishes she could sing even ONE note close to Janis.5

In languages that have a dedicated scalar particle (or a collocation of particles) that realizes weak even – e.g. auch nur in German (e.g.  Guerzoni  2003) – that scalar particle is used in examples that correspond to the above data. This is illustrated in (10) where auch nur associates with 20%, an expression denoting a relatively low percentage. (10)

Ich hoffe auch nur 20% von deiner Begeisterung, die du ihm I hope even 20% of your enthusiasm that you him geschenkt hast, bei ihm wach halten zu können. given have with him sustain to be able ‘I hope to sustain even 20% of his excitement that you caused.’6

2.2 Factive desire predicates It is well-known at least since Kadmon & Landman  (1993) that weak even may occur in the scope of non-downward-entailing factive desire predicates like glad (11a). Weak even is also licensed in the scope of other factive evaluatives (11b). (11)

a. John is glad that he read even ONE paper on this topic. b. It’s good/great/fantastic that John solved even ONE exercise.

2.3 Imperatives The final class of non-negative modal environments in which weak even may occur are imperatives. Some examples of imperatives with weak even are given in (12). Translations of these imperatives into German contain auch nur, as illustrated in (13).

5 http://www.johnsoncitypress.com/News/article.php?id=91361, http://www.youtube.com/ watch?v=c64DHj9yEI 6 http://www.inahallermann.de/index.php?page=rueckmeldungenvonklienten

132   

(12)

   Luka Crnič

a. Show me even ONE party that cares for the people! b. Break even ONE record that I can’t! c. Give me even just ONE reason not to hurt you!

(13)

Zeig mir auch nur eine Partei die sich wirklich ums show me even one party that self really for Volk kümmert! people cares ‘Show me even one party that really cares for the people.’7

These imperatives may be given natural paraphrases with the embedding predicates challenge, dare and defy (14). Importantly, if we make the paraphrases nonperformative, weak even continues to be licensed (15). (14)

a. I challenge you to show me even ONE party that cares for the people. b. I dare you to score even ONE goal against my team.

(15)

John challenged Steve to show him even ONE party that cares for the people.

To summarize: we have presented felicitous sentences in which weak even and its German counterpart auch nur occur in the scope of non-factive and factive desire predicates and in imperatives, i.e. in three modal environments that tend to be analyzed as upward-entailing.

2.4 Three challenges Any approach to the above data faces three challenges. First: it needs to explain why weak even in modal environments does not trigger an infelicitous scalar presupposition. Second: it needs to explain the restriction of weak even to only a subset of modal environments. Third: it needs to explain why the above sentences are accompanied by a distinct bias. We describe these three challenges in turn.

7 http://www.youtube.com/all_comments?v=VWTmA5oGkeQ

How to get even with desires and imperatives   

   133

2.4.1 Consistency of the scalar presupposition The occurrence of weak even in the examples above is unexpected on the common assumption about the semantics of desire predicates and the imperative operator – they are upward-entailing (e.g. Hintikka 1962, Schwager 2006). That is, they are commonly taken to license inferences along the lines of (16a): for all n>0, that I hope to make n videos entails that I hope to make one video. In accordance with the condition in (3), any likelihood relation between the propositions in (16a) satisfies the condition in (16b): for all n>0, that I hope to make n videos is at most as likely as that I hope to make one video. (16)

a. that I hope to make one video ⇐ that I hope to make two videos ⇐ … b. … ⊴c that I hope to make two videos ⊴c that I hope to make one video

Now, the sentence in (17a), repeated from above, may be assigned the structures in (17-b,c). In one structure even stays in situ, while in the other it moves above the desire predicate hope. (17)

a. I hope to someday make even ONE video of that quality. b. [I hope [[even C1] [PROI to make oneF video of that quality]]] c. [even C1] [I hope PROI to make oneF video of that quality]

We have seen in the introduction that if even stays in situ (17b), its scalar presupposition is illicit. The same holds if even scopes above hope: the scalar presupposition triggered by even, given in (18), is at odds with the condition in (16b) – it cannot both hold that every alternative is at most as likely as the prejacent (16b) and that there is an alternative that is more likely than the prejacent (18). We are at an impasse. (18)

⟦ (17c) ⟧g,c is defined only if for some n>1: that I hope to make one video of that quality ⊳c that I hope to make n videos of that quality.

2.4.2 Constrained distribution of weak even in modal environments Weak even is not licit in every type of modal environment. More to the point, it is not licit in epistemic and doxastic modal environments, e.g. in the scope of epistemic and doxastic attitude predicates like know, believe and think:

134   

(19)

   Luka Crnič

# John knows/thinks that he will make even ONE video of that quality.

Furthermore, weak even is also illicit in the scope of certain non-doxastic attitude predicates. This holds for intend and for the directive attitude predicates command and order (20). As we have pointed out above, certain directive predicates differ from command and allow for a licit occurrence of weak even in their scope (21). (20)

a. # John intends to make even ONE video of that quality. b. # John commanded me to make even ONE video of that quality.

(21)

a. John challenged me to make even ONE video of that quality. b. I urge you to plant even ONE of these tubers.8

2.4.3 Extra inference The occurrences of weak even in desire statements and imperatives are accompanied by a distinct bias. For example, the sentences in (22) induce the inferences that I take it to be unlikely that the addressee will find any party that cares for the people and that I take it to be unlikely that I will make one video of that quality, respectively. (22)

a. Show me even ONE party that cares for the people! b. I hope to someday make even ONE video of that quality.

More generally, contexts in which non-negative desire statements or imperatives with weak even are used are such that the attitude holder or, in the case of imperatives, the speaker is taken to assign a low probability to the alternatives invoked by the sentential complement of the attitude predicate.9

8 http://www.cherrymenlove.com/gardening_flowers/2011/03/story-from-a-spring-garden.html 9 The sentences in (i) appear not to trigger a low probability bias: it is very likely that Tyson will inspire a lot of kids and due to his well-known self-confidence he probably believes in it too (ia); (ib) may be addressed to a very generous philanthropist who is expected to donate a lot of money.

How to get even with desires and imperatives   

(23)

   135

Low probability bias generalization a. A context satisfies a low probability bias wrt an individual and a set of alternatives if the individual takes the alternatives to be unlikely to obtain. b. Desire statements with weak even are accompanied by a low probability bias wrt the attitude holder and the alternatives invoked by the sentential complement of the desire predicate. c. Imperatives with weak even are accompanied by a low probability bias wrt the speaker and the alternatives invoked by the imperative.

A closely related restriction has been discussed by Kadmon & Landman (1993) with respect to the occurrence of weak even in the scope of factive desire predicates. For example, they have claimed that (24) is licit only in contexts in which we did not get better tickets than these tickets (Kadmon & Landman 1993:385). That is, their characterization differs from (23) in that it requires the attitude holder to believe that the alternatives have not obtained and not that they were unlikely to obtain. (24)

I’m glad we even got THESE tickets!

However, their characterization is too strong in light of the data in (25). In contrast, the weaker characterization in (23) can be used to correctly describe the contexts in which (25) and its ilk may be used – the attitude holder only needs to believe that it was unlikely that the relevant alternatives will obtain rather than believe that they did not obtain.

(i)

a. Tyson, who said he hopes to inspire even ONE kid to turn his life around, has given similar speeches around the world. (http://www.lvrj.com/sports/15870627.html)   b. If you can’t come, PLEASE donate even ONE DOLLAR! (http://iam.bmezine.com/?rebekahsxrevenge) We suggest that even these sentences are accompanied by a low probability bias – the difference between them and the cases in the main text is that bias in (i) has the flavor of polite pretense. For example, it arguably holds that although Tyson may in fact be certain that he will inspire many kids, (ia) presents him as being excessively humble and falsely modest – i.e. a low probability bias obtains under pretense. This allows us to stick to (23) as a uniform characterization of all the data.

136   

(25)

   Luka Crnič

Just be glad that the Eastern networks could even FIND Elko, let alone have a camera there to record the speech.10

To summarize: there is a puzzle concerning the distribution of weak even – it may occur in modal environments that have traditionally been classified as upwardentailing. Two further questions concerning the occurrence of weak even in modal environments were presented: weak even is restricted to a subset of modal environments (desire statements, imperatives) and its occurrence is accompanied by a distinct bias.

3 Steps towards a resolution Two resolutions of the above puzzles are presented. The first is based on Heim’s (1992) non-monotone desire semantics, while the second is based on von Fintel’s (1999) monotone desire semantics. In the latter case, an additional mechanism needs to be assumed to account for the felicity of weak even.

3.1 Non-monotone desire 3.1.1 Negation-related semantics of desire and non-monotonicity Heim  (1992) proposes that to want p is, roughly, to believe that p is desirable (good). The belief component of this meaning is fleshed out as a doxastic accessibility function that returns for an attitude holder x and a world w a set of worlds compatible with the beliefs of x in w (cf.  Hintikka  1962), while the desirability component is characterized relationally: a proposition is desirable iff it is preferred to its negation, which is the case iff every world in the proposition is better than every world in the complement set of the proposition; in the cases at hand, these propositions are constituted by the closest belief worlds of the attitude holder in which the proposition denoted by the sentential complement of want and its negation, respectively, hold (Heim 1992:197):11

10 http://www.wayiplay.com/news/why-wont-news-networks-air-complete-political-speechrallies-from-start-to-finish 11 A preference relation among propositions is defined on the basis of a preference relation among worlds as in (ia) (cf. von Wright 1963:31); the maximal similarity function is defined as in (ib) (Lewis 1973). We leave the similarity relation ≤ out of our representations.

How to get even with desires and imperatives   

(26)

   137

If defined, ⟦ want ⟧g,c(≽, p, i, w) = 1 iff ∀w’ ∈ Dox(i, w): sim(w’, Dox(i, w) ∩ p) ≻i,w sim(w’, Dox(i, w) \ p).

The presupposition triggered by want is that its propositional argument is logically independent of the beliefs of the attitude holder. Accordingly, the sentence in (27a) presupposes that John neither believes that he will make one video of that quality nor that he won’t make one video of that quality; its assertive meaning is computed in (27b): each of John’s belief worlds is such that the closest belief worlds to it in which John makes one video of that quality are better than the closest belief worlds to it in which he does not. That is, all else being equal, John would rather be in a world in which he makes one video of that quality than in a world in which he makes no videos of that quality. (27)

a. John wants to make one video of that quality. b. If defined, ⟦ [John [wants [PROJ  to make one video …] ⟧g,c(w) = 1 iff ∀w’ ∈ Dox(John, w): sim(w’, Dox(John, w) ∩ that John makes one video of that quality) ≻John,w sim(w’, Dox(John, w) ∩ that John makes no videos of that quality).

The semantics of desire predicates like hope, wish and glad is built up in a similar fashion, though some modifications might be needed with respect to their presuppositions and the relata of the preference relation (cf. Heim 1992 for details). Furthermore, an analogous meaning can also be assigned to the imperative operator. We simplistically assume that there are two differences between it and desire predicates: first, the imperative operator does not quantify over the doxastic alternatives of an attitude holder but over the worlds in the context set, cs(c); second, the imperative operator triggers additional presuppositions that are responsible for its performativity (cf. Schwager 2006 for a thorough discussion of performativity of imperatives). (28)

If defined, ⟦ IMP ⟧g,c(≽, p, w) = 1 iff ∀w’ ∈ cs(c): sim(w’, cs(c) ∩ p) ≻sp(c),w sim(w’, cs(c) \ p).

The imperative in (29a) has the meaning in (29b): given the speaker’s preferences, it is better that you show me one party that cares than that you show me

(i)

a. p ≻i,w q ≡df ∀w’,w”(p(w’) = 1 ∧ q(w”) = 1 → w’ ≻i,w w”) b. sim(≤, w, p) ≡df λw’. p(w’) = 1 ∧∀w”(p(w”) = 1 → w’ ≤w w”)

138   

   Luka Crnič

no parties that care. And since the speaker is an authority, the imperative effects a command. (29)

a. Show me one party that cares! b. If defined, ⟦ IMP [you show me one party that cares] ⟧g,c(w) = 1 iff ∀w’ ∈ cs(c): sim(w’, cs(c) ∩ that you show me one party that cares) ≻sp(c),w sim(w’, cs(c) ∩ that you don’t show me one party that cares).

The negation-related semantics of desire predicates and the imperative operator described above is non-monotone. For example, it predicts that the propositions denoted in (30) are logically independent. Namely, assume that it holds (i) that I prefer making two videos of that quality to making no or exactly one video of that quality and (ii) that I prefer making no videos of that quality to making exactly one video of that quality. Clearly, (30b) is true in the scenario due to (i). (30a) is false in the scenario if there are belief worlds where I make exactly one video of that quality: the closest belief worlds to them in which I make one video of that quality are due to (ii) not better than those in which I make no videos of that quality. This demonstrates the non-upward-entailingness of desire predicates on the negation-related analysis. Their non-downward-entailingness can be demonstrated in a similar way. (30)

a. I hope to make one video of that quality. b. I hope to make two videos of that quality.

3.1.2 Consistency of the scalar presupposition The sentence in (31a) where weak even occurs in the scope of hope may have the LF given in (31b) where even scopes above the desire predicate. The scalar presupposition that it triggers is given in (31c): there is an alternative that is more likely than that I hope to make one video of that quality. (31)

a. I hope to make even ONE video of that quality. b. [even C1] [I hope [even C1] PROI to make oneF video of the quality] c. For some n>1: that I hope to make one video of that quality ⊳c that I hope to make n videos of that quality.

How to get even with desires and imperatives   

   139

Since due to the non-monotonicity of hope it holds that for no n>1, the proposition that I hope to make n videos of that quality entails the prejacent of even, the scalar presupposition is compatible with the condition in (3). This is in line with the prediction in (7): weak even may be acceptable in the scope of a non-upwardentailing operator, in particular, in the scope of a non-monotone operator.

3.1.3 Plausibility of the scalar presupposition However, the condition that there are alternatives in the domain of even that do not entail the prejacent of even is only a necessary and not yet a sufficient condition for the felicity of even. Since the scalar presupposition described in (31c) is not trivial, it is expected to be satisfied only in certain contexts. Specifically, it is satisfied in contexts in which it holds (i) that if there are preferences that distinguish between a proposition that one makes a certain number of videos of that quality and a proposition that one makes another number of videos of that quality, then the proposition that describes one as making a greater number of videos of that quality will be preferred to the proposition that describes one as making a lower number of videos of that quality12 – ‘the more, the better’ – and (ii) that some relevant alternatives are preferentially distinguishable from the proposition that I make one video of that quality. In such a context, the proposition that I hope to make one video of that quality being true necessitates the distinguishable alternative, say, the proposition that I hope to make two videos of that quality to be true as well (if defined). Furthermore, it holds that the reverse relation is not true in such a context: if I hope to make two videos of that quality, it does not follow that I hope to make one video of that quality. For example, it may be that I hope to make two videos of that quality but that I prefer making no videos to exactly one video of that quality. It consequently does not hold that I hope to make one video of that quality. Thus, relative to the information state that satisfies (i) and (ii), there is an alternative that is more likely than that I hope to make one video of that quality – this is ex hypothesi the proposition that I hope to make two videos of that quality. There are contexts that do not satisfy the two conditions discussed above: contexts in which people are indifferent between making a greater and making a lower number of videos of that quality, contexts in which they prefer making a lower to making a greater number of videos of that quality, and contexts in which it is open what people’s preferences are. Since besides its desiderative compo-

12 See e.g. Hansson 2001:67–70 on how to define preferences among compatible propositions.

140   

   Luka Crnič

nent, which is not helpful in the just described contexts, there is nothing in the content of the sentence in (31a) that could be used to support the scalar presupposition in (31c), the sentence is either perceived as pragmatically odd in these contexts or an appropriate preference relation is accommodated. The same considerations apply to the derivation of the plausibility of weak even in the scope of factive desire predicates and in imperatives.13

3.1.4 Restricted distribution Weak even is illicit in the scope of doxastic attitude predicates. This is expected on the approach to attitude predicates entertained in this section: desire but not doxastic attitude predicates have a negation-related semantics; the semantics of doxastic attitude predicates is upward-entailing (Hintikka  1962). Accordingly, the scalar presupposition triggered by the sentence in (32a), given in (32c), is unsatisfiable: due to the upward-entailingness of believe it holds that for every n>0, the proposition that I believe that John made n videos of that quality entails the proposition that I believe that John made one video of that quality and thus none of the former propositions can be more likely than the latter proposition, contra (32c). (32)

a. # I believe that John made even ONE video of that quality. b. [even C1] [I believe John made oneF video of that quality] c. For some n>1: that I believe that John made one video of that quality ⊳c that I believe that John made n videos of that quality.

The infelicity of weak even under intend and command has a different source. Although both predicates arguably share the non-monotone semantics of want, which suffices for even that scopes above them to trigger a presupposition compatible with (3), this presupposition is not plausible. Unlike with hope and its kin, the desiderative component of intend and command cannot be used in tandem with the preferences in the context to support the scalar presupposition in (33c):

13 There is a prominent approach to even that assumes that even is ambiguous and that weak even spells out the item with a more restricted distribution (e.g. Rooth 1985, Rullmann 1997). Applied to the cases above, such an approach has a difficulty (i) explaining why weak even may not occur in positive episodic sentences but may occur in desire statements and imperatives and (ii) accounting for the glaring context-dependence of weak even in these environments.

How to get even with desires and imperatives   

   141

if I intend to make one video of that quality and prefer making two videos of that quality to making one video of that quality, it does not necessarily hold that I intend to make two videos of that quality (similar reasoning applies to command). Furthermore, there is nothing else in the meaning of intend and command that would make it plausible to treat weaker intentions and commands as less likely than stronger ones. (33)

a. # I intend to make even ONE video of that quality. b. [even C1] [I intend [PROI to make oneF video of that quality]] c. For some n>1: that I intend to make one video of that quality ⊳c that I intend to make n videos of that quality.

This is different with challenge. Its semantic import includes the requirement that the challenged task is difficult, e.g. (34a) gives rise to a requirement that it is difficult to make one video of that quality. This information can be used in supporting an appropriate likelihood relation among the alternatives: it may very well be less likely that one is challenged to perform an easy task – making one video of that quality – than to perform a harder alternative task – making two or more videos of that quality. (34)

a. I challenge you to make even ONE video of that quality. b. [even C1] [I challenge you [ PROyou  to make oneF  video of that quality]] c. For some n>1: that I challenge you to make one video of that quality ⊳c that I challenge you to make n videos of that quality.

3.1.5 Extra inference Desire statements containing weak even occur in contexts that exhibit a low probability bias with respect to the attitude holder and the alternatives invoked by the sentential complement of the desire predicate, as described in (23), repeated below. For example, any context in which (35a) is used satisfies the condition in (35b): John, the attitude holder, takes it to be unlikely that an alternative to the sentential complement of hope will obtain, i.e. that he will make one or more videos of that quality.

142   

(23)

   Luka Crnič

Low probability bias generalization a. A context satisfies a low probability bias wrt an individual and a set of alternatives if the individual takes the alternatives to be unlikely to obtain. b. Desire statements with weak even are accompanied by a low probability bias wrt the attitude holder and the alternatives invoked by the sentential complement of the desire predicate. c. Imperatives with weak even are accompanied by a low probability bias wrt the speaker and the alternatives invoked by the imperative.

(35)

a. John hopes to make even ONE video of that quality. b. John believes it is unlikely that he will make one video of that quality.

We tentatively submit that the low probability bias accompanying desire statements with weak even emerges from the interaction of the scalar presupposition triggered by even, which is satisfied in contexts in which stronger alternatives to the sentential complement of the desire predicate are better than the proposition denoted by the sentential complement, and the link between one’s desires and one’s actions. Since a more thorough and formal investigation of these issues goes beyond the scope of this paper, we only outline the reasoning that might be behind the bias described in (35): Having a desire to make a certain number of videos of that quality is linked to how the attitude holder chooses to act. Different acts result in different outcomes and the attitude holder’s choice is guided by optimizing her chances of being happy with the outcomes. Now, having the desire to make one and the desire to make, say, two videos of that quality is linked to the pursuit of outcomes corresponding to making one and to making two videos of that quality, respectively. In a context in which (35a) is felicitous, the latter outcomes are more desirable to the attitude holder than the former (or at least as desirable). The pursuit of the former is thus legitimate only if the attitude holder takes the latter to be sufficiently less likely to obtain. Namely, only in this case is she optimizing her chances of being happy.

3.1.6 A problem for the approach The negation-related approach to desire and imperatives successfully explains the potential felicity of weak even in their scope. However, it faces the problem that its treatment of desire predicates and the imperative operator as non-mono-

How to get even with desires and imperatives   

   143

tone functions does not seem to be warranted in light of the data in (36) and (37),14 as argued by von Fintel (1999). Namely, if the semantics of desire predicates were non-monotone, the inference pattern in (36) should be invalid and the discourse in (37) should be acceptable since the conjuncts would be independent. This appears not to be the case. (36)

a. John hopes to make more than five videos of that quality. b. ⇒ John hopes to make more than one video of that quality.

(37) # John wants a free flight on the Concorde but he doesn’t want a flight on the Concorde. To summarize: we have presented an approach to desire predicates and the imperative operator according to which these are non-monotone functions that allow the embedding of weak even in appropriate contexts. Although the approach does shed light on the distribution of weak even in desire statements and imperatives, it faces the problem that desire predicates and the imperative operator appear to be monotone functions. In the following subsection, we present an approach that is not affected by this problem. However, to explain the distribution of weak even, it needs to employ mechanisms that the approach in this subsection can remain agnostic about.

3.2 Monotone desire 3.2.1 Doubly-relative modal analysis and upward-entailingness The modal semantics of desire predicates is relativized to two conversational backgrounds (von Fintel 1999). The first conversational background – the modal base – delivers a set of doxastically accessible worlds of the attitude holder, while the second conversational background – the bouletic ordering source – provides the propositions that are used in ordering this set. For perspicuity, we assume that among the doxastically accessible worlds one can always find a set of worlds that

14 Apparently convincing examples that do support the idea that desire predicates are nonmonotone and that do not crucially involve context-shifting discourses are the so-called Ross’s paradox examples (i) (cf. Aloni 2007 and others). We defer an investigation of them to another occasion. (i) I hope to send this letter ⇏ I hope to send or burn this letter

144   

   Luka Crnič

are not worse with respect to the given bouletic ordering source than other doxastically accesible worlds (limit assumption); these are the desire-best worlds.15 A desire statement of the form i hopes that p then states that all the desire-best worlds of i are such that p is true in them (38). As before, desire predicates presuppose that their propositional argument is independent of the attitude holder’s belief state. The semantics of imperatives is identical to (38), modulo the authority presupposition (cf. Schwager 2006). (38)

If defined, ⟦ hope ⟧g,c(f, g, p, i, w) = ∀w’∈best(∩f(i,w), g(i,w)) [p(w’)=1].

Desire predicates and the imperative operator are clearly upward-entailing on this approach: if I hope to make two videos, then all my desire-best worlds are such that I make two videos in them; it follows that all my desire-best worlds are such that I make one video in them and thus that I hope to make one video.

3.2.2 The puzzle If desire predicates and the imperative operator are upward-entailing, the occurrence of weak even in their scope is unexpected. Namely, a sentence like (39a) may have the two structures in (39-b,c) (see (17-b,c) above) and in both structures the domain of even contains only alternatives that entail the prejacent of even, causing its scalar presupposition to be illicit. For example, the presupposition of (39c) is given in (40). It is deviant since it holds that for all n>1, the proposition that I hope to make n videos of that quality entails the proposition that I hope to make one video of that quality and thus cannot be more likely than it. (39)

a. I hope to make even ONE video of that quality. b. [I hope [[even C1] PROI to make oneF video of that quality]] c. [even C1] [I hope to make oneF video of that quality]

(40)

⟦ (39c) ⟧g,c is defined only if for some n>1: that I hopeue to make one video of that quality ⊳c that I hopeue to make n videos of that quality. (# in all contexts)

15 A world w’ is not worse than a world w” with respect to an individual i, a world w and an ordering source g iff for all p ∈ g(i,w): if w” ∈ p, then w’ ∈ p. The best worlds with respect to f, g, i, w are best(∩f(i,w), g(i,w)) = {w’ | w’ ∈ ∩f(i,w) and there is no w” ∈ ∩f(i,w) such that w” 0}

This domain is obtained by inserting a covert exhaustification operator exh that associates with one into the structure and moving even above it: (42)

[even C1] [exh C0] [I hope to make oneF video of that quality]

A simplified meaning of exh is given in (43): its import is that all the alternatives in its domain that do not entail the prejacent are false (see Fox 2007 for a more sophisticated analysis). The meaning of the sister of even in (42) is computed in (44) and is equivalent to the proposition that I hope to make one video of that quality and I am okay with making exactly one video of that quality. The alternatives in the domain of even in (42) are determined analogously and the domain has thus the form given in (41). (43)

⟦ exh ⟧g,c(C, p, w) = 1 iff p(w) = 1 ∧∀q ∈ C [p ⊈ q ∧ q(w) = 0]

(44)

⟦ [exh C0] [I hope to make oneF  video of that quality] ⟧g,c(w) = 1 iff I hopeue to make one video of that quality in w & for n > 1, it is not the case that I hopeue to make n videos of that quality in w.

16 The paraphrase that I am okay with making exactly one video of that quality stands for the existential modal proposition that given my attitudes there are some desire-best worlds in which I make exactly one video of that quality.

146   

   Luka Crnič

3.2.4 Consistency and plausibility of the scalar presupposition The scalar presupposition triggered by even in (42) is given in (45): there is an alternative that is more likely than that I hope to make one video of that quality and I am okay with making exactly one video of that quality. Since the alternatives in the domain of even are mutually exclusive, the presupposition in (45) complies with (3). (45)

∃q ∈{that I hopeue to make n videos of that quality & I am okay with making exactly n videos of that quality | n ∈ ℕ>0}: that I hopeue to make one video of that quality & I am okay with making exactly one video of that quality ⊳c q.

Although due to their mutual exclusivity any likelihood relation may obtain on the alternatives described in (45), not every likelihood relation on them is plausible. More concretely, the presupposition in (45) is satisfied in contexts in which it jointly holds (i) that if there are preferences that distinguish between making more and making fewer videos of that quality, then making more videos of that quality is preferred to making fewer – ‘the more, the better’ – and (ii) that some alternatives are preferentially distinguishable from the proposition that I make one video of that quality. Namely, given this information, the expectation is that one hopes to make a notable number of videos of that quality rather than that one hopes to make one video of that quality and is okay with just one. Thus, in such a context there is an alternative that is more likely than the prejacent.

3.2.5 Restricted distribution and extra inference The infelicity of weak even in the scope of believe, intend and command is a consequence of an implausible scalar presupposition. That is, although the exhaustification mechanism rescues moved even from triggering a presupposition that would clash with (3), the presupposition it does trigger is implausible. An example with a doxastic predicate is given in (46). The sentence in (46a) has the structure in (46b): even scopes above the doxastic predicate and the exhaustification operator. The assertive meaning of the sentence is expressed in (46c).

How to get even with desires and imperatives   

(46)

   147

a. # John believes that he will make even ONE video of that quality. b. [even C1] [exh C0] [John believes he will make oneF  video of that quality] c. John believes that he will make one video of that quality & John believes that he might make exactly one video of that quality.

The scalar presupposition of the sentence is that there is an alternative in the domain of even that is more likely than its prejacent (47). Unlike with desire predicates where their desiderative component and the relevant preferences in the context were shown to play a decisive role in making the scalar presupposition plausible, there is no component in the semantics of believe that could perform a similar function. (47)

For some n>1: that John believes that he will make one video of that quality & John believes that he might make exactly one video of that quality ⊳c  that John believes that he will make n videos of that quality & John believes that he might make exactly n videos of that quality.

Similar considerations apply to intend and command. Namely, there is no component of the meaning of these predicates that would support the scalar presupposition triggered by weak even (see discussion in the preceding subsection). Accordingly, weak even is pragmatically deviant under intend and command. This is different for challenge that has a component that can be utilized to support the respective scalar presupposition: it requires the challenged task to be difficult. Since it naturally holds that more difficult tasks qualify as more reasonable challenges, it may very well be less likely that someone is challenged to do an easy task rather than a hard task. The extra inference is derived as in the preceding subsection.

3.2.6 A problem for the approach The exhaustification operator exh may apply in the absence of intensional operators. Accordingly, it could be used to rescue weak even in simple episodic sentences. For example, the sentence in (48a) could be parsed as (48b) where even takes scope above exh. The scalar presupposition of (48b) is given in (48c). It complies with (3) and may be true – it may be the case that it is less likely that John read exactly one book than, say, that he read exactly two books. However, the sentence in (48a) is judged as deviant even in contexts in which this presup-

148   

   Luka Crnič

position is satisfied and this is problematic for the account. Although we do not currently have a solution to this problem, we hope that an investigation of the conditions on the insertion of exh will prove helpful in explaining the unavailability of the parse in (48b) (see e.g. Fox & Spector 2009). (48)

a. # John read even ONE book. b. [even C1] [exh C0] [ John read oneF book] c. ⟦ (48b) ⟧g,c is defined only if for some n>1: that John read exactly one book ⊳c that John read exactly n books.

To summarize: if we adopt an approach that treats desire predicates and the imperative operator as monotone functions, an additional grammatical mechanism needs to be assumed to account for the felicity of weak even in their scope. The mechanism that we assumed and that allowed us to explain the distribution of weak even is the covert exhaustification operator. An issue for employing this mechanism is that, all else being equal, it leads us to expect a broader distribution of weak even than what we find. A more thorough study of the interaction of even and exh is required.

4 Negative polarity items in desire statements and imperatives Certain negative polarity items (NPIs), such as stressed any and ever, are licensed in non-negative desire statements and imperatives. (49)

a. I am glad that ANYONE likes me. (Kadmon & Landman 1993) b. Find me a politician that EVER cared for us AT ALL. c. He wished ANYONE AT ALL would look at him with that same love and sweetness.

This is unexpected on the characterization of NPI licensing condition in (50) (von Fintel  1999). This is because desire predicates and the imperative operator are either non-monotone or upward-entailing functions, as we have extensively discussed in preceding sections, and definitely not (Strawson) downward-entailing ones.

How to get even with desires and imperatives   

(50)

   149

NPI licensing condition NPIs are only grammatical if they are in the scope of a Strawson downward-entailing operator.

A possible response to the data in (49) is to redefine the NPI licensing condition, e.g. so that it requires NPIs to occur under non-upward-entailing operators (cf. Progovac 1994:279). However, any such response leaves a crucial property of the NPIs in (49) unexplained: their context-dependence. For example, Kadmon & Landman (1993:388) point out that (49a) is only felicitous in contexts in which the attitude holder has a preference for a certain subset of people in the domain of anyone to like him. In this respect, the sentences in (49) closely resemble desire statements and imperatives with weak even that we have discussed in this paper. The resemblance between the context-dependent distribution of weak even and stressed NPIs like any and ever suggests that a successful way of explaining the data in (49) might be to assume that the distribution of the latter expressions is governed by a covert even (Krifka 1995). More precisely, we may assume that stressed any and ever denote existential quantifiers that are obligatorily accompanied by a covert even. Their alternatives are existential quantifiers that differ from any and ever only in that their domains are subsets of the domains of any and ever, i.e. any and ever are entailed by their alternatives (Krifka 1995, Chierchia 2010). (51)

a. ⟦ any D ⟧g,c = λPet.λQet. ∃x∈D [P(x) = Q(x) = 1] b. ALT([any D]) = {λPet.λQet. ∃x∈D’ [P(x) = Q(x) = 1]: D’⊆D}

Accordingly, if covert even and the existential quantifier cannot take split scope in a sentence, the sentence will be pragmatically deviant: it will trigger a scalar presupposition that the prejacent of even, which is entailed by all the alternatives, is less likely than some of the alternatives. (52)

a. # ANYONE likes me b. [even C1] ⟦[any D one] likes me] c. ⟦ (52b) ⟧g,c is defined only if ∃q ∈ {that someone from D’ likes me: D’⊆D}: someone from D likes me ⊳c q iff for some D’⊂D: someone from D likes me ⊳c someone from D’ likes me. (# in all contexts)

150   

   Luka Crnič

If covert even and the existential quantifier are separated by a non-monotone quantifier at LF, as exemplified in (53) (if we adopt Heim’s treatment of desire predicates), we obtain configurations in which even may trigger a correct presupposition: e.g. if there is a specific group of people that I want to like me, it is plausible that me being glad that someone from a bigger group likes me is less likely than me being glad that someone from the specific group likes me. (53)

a. I am glad that ANYONE likes me b. [even C1] [I am glad that [anyoneF likes me]] c. ⟦ (53b) ⟧g,c is defined only if for some D’⊂D: I am glad that someone from D likes me ⊳c I am glad that someone from D’ likes me.

Following Krifka (1995), we have effectively reduced configurations with stressed any and ever to those with weak even discussed above. The derivation of the correctness of the scalar presuppositions of the sentences in (49) as well as of other inferences accompanying these sentences can proceed along the lines described in Section 3. Since stressed any and ever are ‘licensed’ if the covert even that associates with them triggers a correct scalar presupposition, their felicity and context-dependence in (49) is explained.

5 Conclusion Weak even may be felicitous in non-negative desire statements and in imperatives. If we assume that even can move at LF, this is expected on the negationrelated approach to desire and imperatives. Namely, since the approach assigns the respective operators a non-monotone semantics, moving even above them at LF allows even to trigger a presupposition that is correct in appropriate contexts. The main problem for the negation-related approach is that non-negative desire predicates and the imperative operator appear to be upward-entailing rather than non-monotone. On the more standard modal approach to desire and imperatives, the occurrence of weak even is at first unexpected in these environments – this is because the environments are upward-entailing on this approach and weak even cannot occur in upward-entailing environments. We dealt with this puzzle by assuming that an exhaustification mechanism may apply in the scope of moved even that allows it to trigger a correct presupposition. The main problem for the approach is restricting the application of this rescuing mechanism. A further investigation into both of the above-mentioned problems is mandated.

How to get even with desires and imperatives   

   151

We conclude by pointing out two additional avenues for future research. First: An issue that we set aside in this paper is the additive inference that often accompanies even, i.e. the inference that an alternative other than the prejacent of even is true (see footnote 1). If we were to encode this inference as a presupposition of even, we would obtain wrong predictions for the sentences discussed in this paper. For example, the prediction for (54a), where even moves above glad at LF, would either be that it conveys that John is glad that he read two papers (on Heim’s semantics of desire predicates) or that it is inconsistent (on a more standard modal semantics of desire predicates coupled with covert exhaustification). However, the sentence is consistent and lacks any such entailment. (54)

a.

John is glad that he read even ONE paper on this topic.

b. ⇏John is glad that he read two papers on this topic. Interestingly, if even takes scope above the desire predicate at surface structure, an additive inference does appear to be generated, as illustrated in (55). We face a dilemma: if we encode an additive presupposition into even, we fail to derive a correct prediction concerning (54); if we do not encode an additive presupposition into even, we fail to derive a correct prediction concerning (55) (see Rullmann 1997 for a third alternative). This needs to be explored further. (55)

a.

I am even glad that we got THESE tickets.

b. ⇒I am glad that we got some tickets other than these. Second: The prediction described in Section 2 leads us to expect that weak even should in appropriate contexts be able to occur in the scope of non-monotone nominal quantifiers. This prediction is borne out, as illustrated in (56a). Strikingly, the distribution of weak even is again mirrored by the distribution of certain NPIs (56b) (cf. Linebarger 1987). A uniform explanation of the felicity of sentences like (56) is called for and we plan to pursue it at another occasion. (56)

a. Exactly two congressmen read even ONE book. b. Exactly two congressmen read ANY book AT ALL.

152   

   Luka Crnič

6 References Aloni, Maria. 2007. Free choice, modals, and imperatives. Natural Language Semantics 15(1). 65–94. Bennett, Jonathan. 1982. Even if. Linguistics and Philosophy 5(3): 403–418. Chierchia, Gennaro. 2006. Broaden your views. Implicatures of domain widening and the ‘logicality’ of language. Linguistic Inquiry 37(4): 535–590. Chierchia, Gennaro. 2010. Meaning as an inferential system: Polarity and free choice phenomena. Harvard University. von Fintel, Kai. 1999. NPI licensing, Strawson entailment, and context dependency. Journal of Semantics 16(2). 97–148. Fox, Danny. 2007. Free choice and the theory of scalar implicatures. In Uli Sauerland & Penka Stateva (eds.), Presupposition and Implicature in Compositional Semantics, 71–120. Basingstoke: Palgrave Macmillan. Fox, Danny & Benjamin Spector. 2009. Economy and embedded exhaustification. Handout from a talk at Cornell. MIT & ENS. Guerzoni, Elena. 2003. Why even ask? On the pragmatics of questions and the semantics of answers. MIT dissertation. Hansson, Sven Ove. 2001. The Structure of Values and Norms. Cambridge: Cambridge University Press. Heim, Irene. 1992. Presupposition projection and the semantics of attitude verbs. Journal of Semantics 9(3): 183–221. Herburger, Elena. 2000. What Counts: Focus and Quantification, vol. 36. Cambridge (MA): The MIT Press. Hintikka, Jaakko. 1962. Knowledge and Belief: An Introduction to the Logic of the Two Notions. Ithaca: Cornell University Press. Kadmon, Nirit & Fred Landman. 1993. Any. Linguistics and Philosophy 16(4). 353–422. Karttunen, Lauri & Stanley Peters. 1979. Conventional implicature. In Choon-Kyu Oh and David A. Dinneen (eds.), Syntax and semantics, vol. 11, 1–56. New York: Academic Press. Kay, Paul. 1990. Even. Linguistics and Philosophy 13(1). 59–111. Krifka, Manfred. 1995. The semantics and pragmatics of weak and strong polarity items. Linguistic Analysis 25. 209–257. Lahiri, Utpal. 1998. Focus and negative polarity in Hindi. Natural Language Semantics 6. 57–123. Lahiri, Utpal. 2006. Scope, presuppositions and dimensions of meaning: Some observations on scalar additive particles in English, Hindi and Spanish. Handout from Sinn und Bedueutung 11, Universitat Pompeu-Fabra, Barcelona. Lewis, David. 1973. Counterfactuals. Wiley-Blackwell. Linebarger, Marcia C. 1987. Negative polarity and grammatical representation. Linguistics and Philosophy 10(3). 325–387. Merin, Arthur. 1999. Information, relevance, and social decisionmaking: some principles and results of decision-theoretic semantics. In Larry Moss, Jonathan Ginzburg & Maarten de Rijke (eds.), Logic, Language and Computation, vol. 2, Palo Alto: CSLI Publications. Progovac, Ljiljana. 1994. Negative and Positive Polarity: A Binding Approach. Cambridge: Cambridge University Press. Rooth, Mats. 1985. Association with focus: University of Massachusetts, Amherst dissertation.

How to get even with desires and imperatives   

   153

Rullmann, Hotze. 1997. Even, polarity, and scope. Papers in Experimental and Theoretical Linguistics 4. 40–64. Schwager, Magdalena. 2006. Interpreting imperatives: University of Frankfurt/Main dissertation. von Wright, Georg H. 1963. The Logic of Preference. Edinburgh: University of Edinburgh Press.

I-Ta Chris Hsieh

On NPI licensing in possibility conditionals 1 Introduction This paper concerns the licensing of NPIs in a possibility conditional. Examples of this kind are given in (1a, b). In (1a, b), the possibility modals may/might signal the compatibility between the antecedent and the consequent; as shown in these two examples, weak NPIs such as any and ever are grammatical in the antecedent (i.e. the if-clause) of a possibility conditional.1 (1)

a. If John had ever been to Paris, he might have seen the Arc de Triomphe. b. If John subscribes to any newspapers, he may be well-informed.

In the literature (Lewis 1973, 1986, Kratzer 1981, 1991; a.o.), it has been assumed that modal elements quantify over relevant possible worlds; while necessity modal elements such as must have been taken to be a universal quantifier, possibility modals like may/might/be allowed to have been taken to be an existential one. Based on this assumption, the meaning of the possibility modal statements in (2a, b) can be roughly represented as (2a′, b′) respectively. (2)

Epistemic Possibility a. John might have parked here yesterday. a′. ∃w′[w′ is epistemically accessible to the world of evaluation w ∧ John parked here yesterday in w′] Deontic Possibility b. (According to the school regulation,) John may take this course. b′. ∃w′[w′ is deontically accessible to the word of evaluation w ∧ John takes this course in w′]

Furthermore, in a Lewis-Kratzer-von Fintel style semantics of conditionals (Lewis 1973, 1986; Kratzer 1986; von Fintel 1994; a.o.), a conditional is a modal state-

1 The focus of this paper is only limited to weak NPIs such as any and ever.

156   

   I-Ta Chris Hsieh

ment, and the if-clause in a conditional serves as the restrictor of the modal. Based on this assumption, the meaning of the possibility conditionals in (3a, b) can be represented as (3a′, b′). In (3a′, b′), the if-clause serves to restrict the possibility modals may and might. (3)

a. If John had had ice cream for dessert, he might have had port for nightcap. a′. ∃w′[w′ is accessible to the word of evaluation w and John had ice cream for dessert in w′ ∧ John had port for nightcap in w′] b. If John has finished his assignment, he may have a piece of candy. b′. ∃w′[w′ is accessible to the word of evaluation w and John has finished his homework in w′ ∧ John has a piece of candy in w′]

These two assumptions together with the DE-based approach of NPI licensing (Fauconnier 1975, 1979; Ladusaw 1979; von Fintel 1999; a.o.), as I will show in the following however, leads to the wrong prediction on the licensing of NPIs in the antecedent of a possibility conditional. Following the proposal in Rullmann, Matthewson and Davis (2008), I further suggest that this problem can be solved if we assume that possibility modals, instead of being existential quantifiers, are universal quantifiers whose quantificational domain is further restricted by a modal choice function. The rest of this paper is structured as follows. In Section 2 I will first discuss NPI licensing in conditionals and show that the traditional assumption, namely that possibility modals are existential quantifiers, together with a Lewis-Kratzervon Fintel style semantics of conditionals, leads to the wrong prediction on the distribution of NPIs in the if-clause of a possibility conditional. I will then suggest in Section 3 that the problem can be solved if we extend the idea of modal choice functions proposed by Rullmann et al. (2008) to English possibility modals. The discussion in Section 4 is to show how the distinction between necessity and possibility modals can be recast under the proposal made here. Some alternative solutions to the proposed analysis are examined in Section 5. The conclusion and some open questions are in Section 6.

On NPI licensing in possibility conditionals   

   157

2 NPI licensing and conditionals 2.1 Conditionals and the SDE condition Weak NPIs, such as any and ever, have a limited distribution; (4) shows that they are grammatical when negation is present but ungrammatical when it is absent; (5) shows that they are grammatical in the restrictor of the universal quantifier every but not in that of the existential quantifier some. (4)

a. * John ate any potatoes. b. John didn’t eat any potatoes.

(5)

a. Every student who had ever been to Paris became a good chef. b. * Some student who had ever been to Paris became a good chef.

Fauconnier (1975, 1979) and Ladusaw (1979) suggest that the distribution of NPIs can be captured by the Downward Entailing (DE) condition, according to which NPIs are only grammatical in a context in which an inference from a set to its subset is valid.2 (6)–(7) show that the distribution of NPIs in (4)–(5) follows from the DE condition; the scope of negation and the restrictor of every support a DE inference and hence license NPIs. (6)

a. kale⊆vegetables b. John did not eat vegetables. ⇒ John did not eat kale. c. John ate vegetables. =/⇒ John ate kale.

(7)

a. linguistics students⊆students b. Every student passed the exam. ⇒ Every linguistics student passed the exam. c. Some student passed the exam. ⇏ Some linguistics student passed the exam.

2 A notion of cross-categorial entailment (⇒) is defined as in (i). (i) Cross-categorial entailment (⇒) a. for any p, q of type t, p⇒q iff p=0 or q=1 b. for any f, g of type , f ⇒g iff for all x of type σ, f(x)⇒g(x)

158   

   I-Ta Chris Hsieh

The antecedent of conditionals licenses NPIs as well (see (8)). Examples of this kind, however, pose a problem to the DE condition, for the antecedent of a conditional is not intuitively a DE context. This is evidenced by the failure of strengthening the antecedent (see (9); see von Fintel (1999, 2001) and the references cited therein). (8)

a. If John had ever been to Paris, he must have seen to the Arc de Triomphe. b. If John subscribes to any newspapers, he will become well-informed.

(9)

a. wet matches ⊆ matches If I struck a match, there would be a fire. ⇏ If I struck a wet match, there would be a fire. b. newspapers John cannot read ⊆ newspapers If John subscribes to a newspaper, he will become well-informed. ⇏ If John subscribes to a newspaper he cannot read, he will become wellinformed.

To solve this problem, there are two possible ways. The first possibility is to give up the DE approach and turn to other proposals for NPI licensing (e.g., Linebarger 1980, 1987; Giannakidou 1998, 1999; a.o.). On the other hand, one can implement an entailment-based approach of NPI licensing by appealing to a weaker licensing condition of NPIs (e.g., Ladusaw 1979; Heim 1984; Kadmon and Landman 1993; von Fintel 1999). Throughout this paper, I am taking the second position to the theory of NPIs; in particular, I assume the SDE condition of NPI licensing, which is proposed by von Fintel (1999). von Fintel (1999) suggests that NPI licensing is subject to a weaker notion of entailment, which he dubbed as Strawson entailment; an NPI is grammatical only if it is in a Strawson Downward Entailing (SDE) context (see the definition in (10)).3 (10)

a. The SDE condition of NPI licensing: An NPI is only grammatical if it is in the scope of α such that [[ α ]] is SDE.

3 As noted in von Fintel (1999), the SDE condition can only be seen as a necessity condition of NPI licensing. As pointed out in Progovac (1993), Lahiri (1998), Guerzoni and Sharvit (2007), and others, crucially, a licensing environment cannot be Strawson upward-entailing (SUE).

On NPI licensing in possibility conditionals   

   159

b. Strawson Downward Entailingness: A function f of type is Strawson downward entailing (SDE) iff for all x, y of type σ such that x⇒y and f(x) is defined: f(y)⇒f(x) Furthermore, he suggests a monotonic semantics of conditionals. In a nutshell, in his proposal, a conditional is evaluated with respect to a quantificational domain W’ and is presuppositional in the sense that the quantificational domain of a conditional W’ has to be compatible with the antecedent (see (11)).4 Based on this idea, a conditional if p, would-q is defined only if the quantificational domain W’ is compatible with the antecedent; the modal operator would, which is a universal quantifier over worlds, then operates on the intersection of W′ and the antecedent.5 (11)

For any w∈W, p⊆W, q⊆W and W’⊆W, [[ would ]]w,W′(if p)(q) is defined only if W′∩p≠∅; if defined, [[ would ]]w,W′(if p)(q)=1 iff ∀w′∈W′∩p: w′∈q

The licensing of NPIs in the antecedent of a conditional and the failure of strengthening the antecedent are captured by the SDE condition and the semantics in (11)

4 In von Fintel’s (1999) proposal, the quantificational domain W’ is constrained in the way that it must be an admissible sphere in a modal base ∩A(w) with respect to an ordering source R(w). A representation from Gajewski and Sharvit (2009) that is more faithful to von Fintel’s semantics is given in (i). Based on (i), all the worlds in the modal base ∩A(w) are arranged into nested spheres (i.e. ‘Lewis spheres’) with respect to the ordering source R(w). The quantificational domain W′ not only has to be one of the ‘Lewis’-spheres but also has to be compatible with the antecedent (see (ib)). (i) For any w∈W, p⊆W, q⊆W, A, and R, W′⊆W, [[ would ]] A,R,w,W′(if p)(q) is defined only if   a. W′is an admissible sphere in the modal base ∩A(w) with respect to the ordering source R(w); (Admissibility Presupposition)   b. W′∩p≠∅ (p is compatible W′) (Compatibility Presupposition) if defined, [[ would ]] A,R,w,W′(if p)(q)=1 iff ∀w′∈W′∩p: w′∈q (where A is a function that maps a world w to a set of true propositions in w the conjunction of which (i.e. ∩A(w)) serves as the modal base; R

This pair serves as the argument of the recursive exhaustification operator. So, the first coordinate of (68) becomes the proposition that Bill is a minimal exception to the statement ‘no boy runs’. This means that Bill is a boy and does run. The second coordinate becomes a set of propositions each of which says that an alternative set is a minimal exception to the generalization, see (70). (69)

exhR (68)= exh

(70)

exh

Now, because no happens to be a determiner that guarantees unique minimal exceptions, the truth of the first coordinate in (70) actually entails the falsity of every alternative proposition that is a member of the second coordinate, except

An analogy between a connected exceptive phrase and polarity items   

   207

the proposition itself. So, all of the alternatives are excluded by the exhaustifier. But since their falsity was already entailed by the first coordinate, in this case, the second layer of exhaustification has no effect. Now we must turn to one of the cases that motivated the inclusion of a second layer of exhaustification. Recall that, unlike no, not all determiners guarantee that a minimal exception is unique. Non-AA LDE quantifiers and non-monotone determiners allow non-unique minimal exceptions. Quantifiers headed by these determiners cannot felicitously host a but-EP. So, our goal is to rule out the combination through our new approach involving exhR. Below I work through the case of a non-monotone determiner, exactly three. For convenience, I again abbreviate the expression exactly three as ‘3!’. (71)

a. *Exactly three boys but Bill run. b. run exhR 3! boys

but Bill

Again, the denotation of the complement of exhR can be found in (72) and the result of recursive exhaustification in (73). The exhaustified first coordinate is the proposition that Bill is a minimal exception to the statement that exactly three boys run. This implies that exactly four boys run and that Bill is one of them. The set of exhaustified alternatives contains propositions stating that other sets constitute minimal exceptions to the generalization that exactly three boys run. None of these alternatives is entailed by the first coordinate – again, except for the proposition itself. But not all of them are entailed to be false by it either, unlike the case in (69). So, the exclusion of these alternatives is contentful. (72)

, {: R∈D}>

(73)

exhR(71)= exh

208   

   Jon Gajewski

(74)

exh (λw.⟦3!⟧(⟦boy⟧w–{Bill})(⟦runs⟧w), {λw.⟦3!⟧(⟦boy⟧w–S)(⟦runs⟧w): S⊆{Bill}}) = Bill is a boy and Bill and exactly three other boys run (Abbreviated: ‘3+Bill boys run’)

(75)

exh

This additional content, however, leads to a contradiction. While no particular alternative set is entailed to be a minimal exception, it is entailed that there must exist three other minimal exception sets. So, if exhaustification requires that all alternative propositions are false, since they are not entailed, this contradicts the exhaustified first coordinate of (75). Thus, we are able to rule out the co-occurrence of but-EPs with quantifiers headed by determiners that allow multiple minimal exceptions to a generalization. The second layer of exhaustification, which enforces the uniqueness of minimal exceptions, leads to contradictions in these cases. In the other cases, the second layer of exhaustification is harmless. The combination of LUE determiners and but-EPs is ruled out by a presupposition at the first layer, since no alternative propositions are eliminated. In the case of AA, LDE determiners every and no, the correct semantics is derived at the first level and the second level’s contribution is redundant.

4 The scope of exhR A question arises as to what are the possible scopes for the exhaustifier. The answer affects the possible interpretations of but-EPs and, hence, its distribution. Recall that in order to account for uses of but-EPs with any, it must be possible for exhR to take scope at a position that is not local to the EP associate. In other words, exhR need not occur at the first scope site c-commanding the but-EP’s associate quantifier. For example, in principle, exhR could take scope below no professor, but must be allowed to scope above it for the story in section 2.2 to work. (76)

exhR No professor saw any student but[+exh] Bill

4.1 Constraints on scope for exhR We want to know now whether there are any other cases in which non-local scope is allowed. In (77), there are two possible scopes for exhR, (77a) and (77b).

An analogy between a connected exceptive phrase and polarity items   

(77)

   209

Every professor saw no student but Bill a. Every professorx [exhR no student but Billy [ x saw y ] ] ‘Bill is a student, every prof saw Bill, no other student was seen by any prof.’ b. exhR Every professorx [ no student but Billy [ x saw y ] ] ‘Bill is a student, some prof saw Bill, no other student was seen by any prof.’

Reading (77a) is available. The weaker reading (77b), however, which does not entail that every professor saw Bill, is not available. We should also keep in mind other possible relative scope of the quantifiers – even though it is often difficult for negative quantifiers like no student to scope over universals. (78)

exhR [ no student but Billy [ every professorx [ x saw y ] ] ] ‘Bill is a student, every prof saw Bill, no other student was seen by every prof.’

(77a) asymmetrically entails both (77b) and (78) Note also that (77b) and (78) are independent. So, the judgment is solid that (77b) is unavailable. The tentative conclusion, then, seems to be the following. (79)

but-EPs require their exhaustifier (exhR) to be as local as possible.

A possible explanation for the perceived markedness of (77b) might be a preference for placements of the exhaustifier that yield stronger readings. Several researchers have recently proposed such a constraint on the placement of exhaustifiers. I include one below. (80)

Let ϕ be a logical form. Let ϕ’s competitors be all the LFs that differ from ϕ only with respect to where exhaustivity operations occur. Then everything else being equal, ϕ is dispreferred if one of its competitors is stronger. (Chierchia, Fox, Spector 2008)

Assuming that exhR must take scope very locally may help to explain some aspects of but-EPs distribution. For example, even though a consistent meaning could be derived for (81a), if exhR took scope over no professor, such a meaning is not available. This could be accounted for if exhR was forced to scope below no professor, as in (81b).

210   

(81)

   Jon Gajewski

a. * No professor saw a student but Bill. b. No professorx [exhR a student but Billy [ x saw y ] ]

But, of course, this leaves us with a question. Why is the exhaustifier exhR allowed to take wider scope when the EP is attached to a DP headed by any?

4.2 Relativized minimality Following Chiechia’s (2006, 2013) approach to intervention, I assume that the exhaustifier triggered by but belongs to the same system as the exhuastifier that Chierchia (2004) uses to account for the distribution of any. Because they belong to the same class of operators, crossing dependencies are not allowed. An exhaustifier is not allowed to skip an (unchecked) exh-feature. Nesting dependencies are allowed. (82)

exhR exh No professor saw any[+exh] student but[+exh] Bill

(83)

exh [ No professorx exhR [ any[+exh] student but[+exh] Billy [x saw y]]]

In this way we can reduce the problem of the scope of but’s exhaustifier to the problem of any’s. Licensing of any requires its exhaustifier to scope above its negative licenser. Relativized minimality forces but’s exahustifier to scope above any’s. Obviously, for this to work the locality conditions on but’s exhaustifier must be violable and ranked below the conditions on any’s. An additional prediction of this story is that an unlicensed any should improve a but-EP’s co-occurrence with a c-commanding non-NPI existential associate. In some cases this is a good prediction, in others it is less clear, cf. (84) and (85). (84) (85)

Mary doesn’t like a relative of any politician but Billy Carter. ??

Mary doesn’t know a person who owns any ferrets but Fred.

At this point, this account remains a sketch of an approach to how to limit but-EPs to associating with NPIs in the scope of DE operators.

An analogy between a connected exceptive phrase and polarity items   

   211

5 Conclusion In this paper I have tried to argue for a parallel between the semantics/pragmatics of the exceptive marker but and that of polarity/free choice items. The idea is simple. The basic semantics of EM but is that of set subtraction; it removes the individuals denoted by its complement from the restrictor of the quantifier it is attached to. In addition, but is an alternative-introducing expression. In fact, I argue that a but-EP introduces second order alternatives, alternatives to a proposition p and to p’s set of alternatives. These second-order alternatives then are predictably exploited by a particular exhaustifier, which I labeled exhR. This exhaustifier applies recursively, imposing minimality on the semantics of but-EPs at one level and then imposing uniqueness at the second level. The analysis finds antecedents in the work of Fox (2007) and Chierchia (2006, 2013). There are still areas of the account that need work. In Section 4, we looked in a preliminary way at possible scopes that exhR could take with respect to the surface position of but-EPs. A fuller understanding of the constraints on scope awaits further research. So does our precise understanding of why but-EPs can associate with any in the scope of a DE operator, but not with non-NPI indefinites like a.

6 References Carlson, Greg. 1980. Polarity any is existential. Linguistic Inquiry 11: 799–804. Chierchia, Gennaro. 2004. Scalar Implicatures, Polarity Phenomena, and the Syntax/Pragmatics Interface. In A. Belletti (ed.) Structrues and Beyond. Oxford. Chierchia, Gennaro. 2006. Broaden you views: Implicatures of domain widening and the logicality of language. Linguistic Inquiry. Chierchia, Gennaro. 2013. Logic in Grammar: Polarity, Free Choice, and Intervention. Oxford University Press. Chierchia, Gennaro, Danny Fox and Benjamin Spector 2008. The grammatical view of scalar implicatures and the relationship between semantics and pragmatics. to appear in Handbook of Semantics, Paul Portner, Claudia Maienborn et Klaus von Heusinger (Eds), Mouton de Gruyter. Crnič, Luka. 2011. Getting even. Ph.D. thesis. MIT Dayal, Veneeta. 1998. Any as inherently modal. Linguistics and Philosophy 21, 433–476. Dayal, Veneeta. 2004. The universal force of Free Choice any. Linguistic Variation Yearbook 4, 5–40  von Fintel, Kai. 1993. Exceptive constructions. Natural Language Semantics 1, 123–148. von Fintel, Kai. 2000. Notes on exceptives. Ms. MIT. http://web.mit.edu/fintel/www/exceptives. notes.pdf

212   

   Jon Gajewski

Fox, Danny. 2007. Free choice and the theory of scalar implicatures. In U. Sauerland & P. Stateva (eds.), Presupposition and implicature in compositional semantics, 537–586. New York: Palgrave Macmillan. Gajewski, Jon. 2002. L-analyticity and natural language. Ms. MIT. Gajewski, Jon. 2008. NPI any and connected exceptive phrases. Natural Language Semantics. García-Álvarez, Ivan. 2009. Generality and exception: a study in the semantics of exceptives. Diss. Stanford University. Hoeksema, Jacob. 1987. The logic of exception’, in ESCOL IV, pp. 100–113. Hoeksema, Jacob. 1996. The semantics of exception phrases’, in J. V. der Does and J. V. Eijck (eds.), Quantifiers, Logic and Language, pp. 145–177. Kratzer, Angelika and Junko Shimoyama 2002. Indeterminate pronouns: The view from Japanese. In Yukio Otsu (ed.): The Proceedings of the Third Tokyo Conference on Psycholinguistics. Tokyo: Hituzi Syobo, 1–25. Krifka, Manfred. 1995. The semantics and pragmatics of polarity items. Linguistics Analysis 25, 209–257. Ladusaw, William. 1979. Polarity Sensitivity as Inherent Scope Relations. U. of Texas disseration. Linebarger, Marcia. 1981. Polarity any as an existential quantifier. CLS 16, 211–19. Menéndez-Benito, Paula. 2010. On Universal Free Choice Items. Natural Language Semantics.  Moltmann, Friederike. 1995. Exception sentences and polyadic quantification. Linguistics and Philosophy 18, 223–280. Nathan, Lance. 1999. Either: Negative polarity meets focus sensitivity. Ms. MIT Peters, Stanley and Dag Westerståhl 2006. Quantifiers in Language and Logic. Oxford: Oxford University Press. Postal, Paul M. 2004. Skeptical Linguistics Essays. Oxford: Oxford University Press. Sauerland, Uli. 2004. Scalar implicatures in complex sentences. Linguistics and Philosophy, 27: 367–391. Zwarts, Frans. 1998. Three types of polarity items. In F. Hamm and E. Hinrichs, eds., Plurality and Quantification.

Elena Herburger and Simon Mauck

The chance of being an NPI 1 The licensing question It is widely believed that negative polarity items (NPIs) are licensed in contexts that invert the normal direction of inference or that license downward entailments (Fauconnier 1975; Ladusaw 1980). NPIs like any and ever can thus occur in the scope of sentential negation, as in (1a), in the restriction of universal quantifiers, as in (1b), and in the scope of negative propositions like without, as in (1c). As the pairs in (2) – (4) show, these expressions all license downward inferences from a superset to a subset case: (1)

a. I do*(n’t) see anything strange. b. Every/*some student(s) with any sense asked for help. c. Nancy crossed the border without/*with any problem.

(2)

a. Monica didn’t walk home. |= b. Monica didn’t walk home quickly.

(3)

a. Every dog was barking. |= b. Every brown dog was barking.

(4)

a. Nancy traveled without her sons. |= b. Nancy traveled without her youngest son.

The generalization that NPIs are licensed in downward entailing (DE) contexts is a major result of the research on negation and quantifiers. This, however, does not mean that it is free of empirical limitations. One that will remain largely undiscussed in this paper is that not all NPIs are licensed in the same kind of DE contexts (e.g. van der Wouden 1997; Zwarts 1998). Another issue the Ladusaw/ Fauconnier generalization is known to face is that not all contexts that license NPIs are strictly speaking DE. As (5) shows, if and only, for instance, license NPIs but, as we can see in (6) and (7), they do not license downward inferences:

214   

(5)

   Elena Herburger and Simon Mauck

a. If anyone asks for more coffee, we’ll get some. b. Antonella only ever buys Italian shoes.

(6)

a. If this match were struck, it would light. |≠ b. If this match were struck after having been soaked in water overnight, it would light. (Stalnaker 1968)

(7)

a. Sid only eats ORGANIC vegetables. |≠ b. Sid only eats ORGANIC dandelion greens.

Other expressions that license NPIs but require extra assumptions to qualify as creating DE contexts are adversative predicates like regret and be sorry that (Linebarger 1987; Kadmon and Landman 1993; von Fintel 1999, a.o.). Clearly, (9a) can be true without (9b) being true also, for the first example entails (presupposes) that the speaker bought a bike – and it could have been a mountain bike – but the second example requires that he bought a race bike. (8)

I’m sorry/regret I mentioned anything.

(9)

a. I’m sorry/regret I bought a bike before I found out about the bike-sharing program. |≠ b. I’m sorry/regret I bought a race bike before I found out about the bikesharing program.

The approximatives hardly and barely may also be included in this list (Horn 2002 and references therein). (11a), for instance, entails that Mary passed the test and that she came close to not doing so. That, however, does not mean that she passed the first part of the test and that she came close to not doing so, which is what (11b) requires for it to be true: (10)

Andrea hardly/barely ate anything.

(11)

a. Mary barely passed the test.  |≠ b. Mary barely passed the first part of the test.

And, finally, although few and most also license NPIs in their restrictions, their restrictions actually do not license downward inferences in the general case:

The chance of being an NPI   

   215

(12)

Few/most people who knew anything about the case were willing to come forward.

(13)

a. Few linguists work on Hungarian. |≠ b. Few Hungarian linguists work on Hungarian.

(14)

a. Most Americans have a driver’s license. |≠ b. Most Americans under 16 have a driver’s license.

Yet, despite problems such as these, the Fauconnier/Ladusaw generalization seems so powerful and appealing that the issues just mentioned are generally set aside, with the idea that they can be explained in a way that does not substantially subtract much from the main observation (cf. e.g. von Fintel’s 1999 “Strawson downward entailment”). But even if we manage to address the troublesome cases, the generalization that NPIs are licensed in DE contexts still leaves us with a puzzling question. Why are some expressions only allowed to occur in contexts with particular logical properties? It is not easy to see what lexical choice should have to do with logical inference. In a series of papers Kadmon and Landman (1993), Krifka (1995), Lahiri (1998), Chierchia (2006) and, in a somewhat different form also Israel (1996) argue that the key to the puzzle is to be found in the meaning of the NPIs themselves. Appealing and elegant though these proposals are, in what follows we argue that they do not fully explain the distribution of NPIs. We argue that while the distributional restrictions on NPIs are indeed related to their meanings, ultimately, the distributional restrictions on NPIs cannot be derived from those meanings. Whether a particular, semantically predisposed element is an NPI is at the end of the day a matter of chance, as also proposed by Eckardt (2006, 2008/t.a.). In the absence of a semantic property that fully determines their distribution, we conclude that what tells NPIs apart from non-NPIs is but a simple, semantically meaningless feature. We then explore several ramifications of our feature theory of NPIs. Our characterization of NPIs in terms of an NPI-feature is shown to tie in well with the historical development of NPIs, including that of ‘even’ NPIs, the NPIs that would seem most amenable to a semantic account. We also discuss how the NPI feature is licensed, relying in this respect heavily on Sánchez Valencia (1991), see also Dowty (1994), Ludlow (2002). We, however, highlight the fact that NPI licensing is only sensitive to what

216   

   Elena Herburger and Simon Mauck

we call Local Polarity and not to Global Polarity. We conclude with some interesting distributional predictions of our feature-based account.1

2 A semantic answer to the licensing question 2.1 The basic idea Abstracting away from important details, the accounts that aim to derive the distribution of NPIs from their semantics subscribe in some form or another to the following claim: (15)

Semantic answer to the licensing puzzle: NPIs are bottom scalar elements that associate with/contain an operator that requires them to be the semantically stronger than their scalar alternatives.

Bottom scalar elements occupy the bottom rung on a Horn scale (Horn 1972), e.g. some, or. As is evident from the entailments below, in non-negative contexts some and or represent the weakest of these scalar alternatives and every/all and and the strongest: (16)

a. Gwen edited all of the articles in the magazine. |= b. Gwen edited most of the articles in the magazine. |= c. Gwen edited some of the articles in the magazine.

(17)

a. Tom brought wine and beer to the barbecue. |= b. Tom brought wine or beer to the barbecue.

1 Earlier versions of our arguments for a feature-based account of NPI licensing were presented at the 2006 Swarthmore Workshop on Negation and Polarity, the 2007 Workshop on Negation and Polarity at the University of Tübingen, and the Parasession on Negation of the 2009 Annual Meeting of the Berkeley Linguistics Society, and were written up in much abbreviated form in Herburger and Mauck (2007, 2009), see also Herburger (2011). The analysis we present in this paper is largely new. We are grateful to the reviewers for their helpful comments and for encouraging us to provide a more detailed account. We would also like to thank Ruth Kramer for helpful discussion of the syntactic issues.

The chance of being an NPI   

   217

In DE contexts, in contrast, the entailments of course go in the opposite direction and every/all/and now function as the weakest of the alternatives and some (any)/ or the strongest: (18)

a. It is not true that Gwen edited some (any) of the articles. |= b. It is not true that Gwen edited most of the articles. |= c. It is not true that Gwen edited all of the articles.

(19)

a. It is not true that Tom brought wine or beer to the barbecue. |= b. It is not true that Tom brought wine and beer to the barbecue.

If, as (15) claims, NPIs correspond to the weakest elements of a Horn scale and come with an operator requiring them to be semantically stronger than any of their scalar alternatives it follows that NPIs are restricted to DE contexts because it is precisely in these contexts where the normally weakest are the strongest. The licensing puzzle seems solved, and elegantly so. As will be discussed next, what requires those low scalar elements that are NPIs to be stronger than their higher ranked alternatives varies somewhat on the different proposals. While the issues discussed in section 3 can be appreciated without knowing the details (and a reader already familiar or not interested in the details can directly skip to that section), we thought it appropriate to describe the specifics of the various accounts.

2.2 The various ‘even’ accounts The earliest proponents of what can be considered a semantic answer to the licensing puzzle summarized in (15), Kadmon and Landman (1993), claim, first, that NPIs have wider quantificational domains than their non-NPI counterparts and, second, that this widening has to lead to strengthening. Any potatoes is held to differ from some potatoes in quantifying also over potatoes that might normally be considered irrelevant, e.g. old potatoes, very small potatoes, etc. The following dialogue is said to support this (p. 360):

218   

(20)

   Elena Herburger and Simon Mauck

YOU: Will there be French fries tonight? ME: No, I don’t have potatoes. YOU: Maybe you have just a couple of potatoes that I could take and fry in my room? ME: Sorry, I don’t have ANY potatoes.

Since the restrictions of existential quantifiers are upward entailing contexts, in non-negative environments a smaller domain results in a semantically stronger claim than an alternative claim involving a larger quantificational domain. Kadmon and Landman’s analysis exploits the fact that in DE contexts the opposite holds; by claiming that NPIs have a wider than usual domain and imposing a requirement that widening must lead to strengthening they predict that NPIs are restricted to negative contexts. Krifka’s (1995) proposal also builds on the assumption that NPIs like any are low scalar expressions, meaning that any has the semantics of an existential quantifier. For him the difference between NPIs and their non-NPI counterparts is not how wide their respective domains are, but rather that NPIs (but not nonNPIs) trigger more specific alternatives. For the NPI anything these alternatives would be specific things like a pencil, a book, etc. The second part of his proposal is the assumption that NPIs like anything appear in the scope of an operator (Scal.Assert) that generates the implicature that no semantically stronger (i.e. more specific) alternative holds. As a result, *Mary saw anything is ruled out as a contradiction between the meaning of the sentence (‘Mary saw something’) and the implicatures generated by Scal.Assert and the specific alternatives that are posited, namely ‘Mary didn’t see a pencil’, ‘Mary didn’t see a book’, etc. When the NPI appears in a negative context, no such contradiction ensues and the NPI is acceptable. We are glossing over Krifka’s proposal for strong NPIs like ANY and at all, for which he posits a scalar operator with a semantics similar to that of even. In relevant respects this part of the account foreshadows Lahiri’s (1998) analysis. The beauty of Hindi – and Lahiri’s (1998) analysis of it – is that no silent operator needs to be posited, nor is it necessary to claim that NPIs induce widening or more specific alternatives. This is so because the NPIs in question consist of a focused low scalar element (e.g. kuch), which denotes a minimum quantity, say one, and the particle bhii, which Lahiri (1998) analyses as a wide scope ‘even’ that generates two conventional implicatures, namely an existential one, saying that some alternative to the focused element (two, three, etc.) makes the sentence true, and a scalar one, saying that the actual amount described in the sentence (one) is the least likely to make the sentence true (Karttunen and Peters 1979). In positive contexts the semantics of the low scalar kuch etc. and the conventional implicatures due to bhii are in conflict. If two people came, one person

The chance of being an NPI   

   219

did, so it cannot be that ‘one’ or ‘some’ is the less likely and hence semantically stronger than its alternative ‘two’. If, on the other hand, the NPI is embedded in a DE context, the demands of the wide scope ‘even’ that associates with ‘one’ do not result in contradiction, because now ‘one’ does indeed represent the semantically strongest and hence least likely of the alternatives.2 Finally, Chierchia (2006) aims to give a general account of NPIs, free choice items and scalar expressions in a framework where the pragmatics interfaces dynamically with the syntax. In the spirit of Kadmon and Landman (1993), Chierchia maintains that any denotes an existential quantifier that induces widening in its domain of quantification. And similar to Krifka (1995), Chierchia posits that NPIs come with alternatives, in particular, alternative domains of quantification. In the case of the NPI any (but not a free choice any) these alternative domains involve subsets of the widened domain of quantification of any. He then posits that any bears an uninterpretable feature [+s], which needs to be checked by a s-operator. This operator amounts to an instruction to ‘lock in’ the pragmatic implicatures at a given point in the derivation of the sentence. Given the particular alternatives that are associated with the NPI any, Chierchia’s account triggers the presence of an operator Ec, which amounts to a tacit ‘even’. Similar to Krifka (1995)’s Emphatic. Assert, this operator requires that the proposition expressed by the sentence entail all alternative propositions derived with the help of the alternative domains: (21)

Ec(p) = p ∧ ∀q ∈ C [p ⊆ q]

2 Lahiri’s description resembles that in Lee and Horn (1994), who claim that the distribution of any is a consequence of it being an indefinite with an incorporated ‘even’, where ‘even’ picks out the least likely of the alternatives. Lee and Horn make use of the fact that the implicatures generated by ‘even’ require the existence of a scale of likelihood associated with the sentence. They then posit that any is licensed only where such a scale can be constructed. Whether or not a scale can be constructed can be determined by whether or not even a single or even + a superlative can appear in a sentence. Thus, pointing to examples like (i) and (ii), Lee and Horn predict that any can occur in a sentence if even a single or even + a superlative can occur in the same position in that sentence: (i) a. * There is any boy running around in the garden. b. * There is even a single boy running around in the garden. (ii) a. There isn’t any boy running around in the garden. b. There isn’t even a single boy running around in the garden. Lee and Horn also note that many languages, including Hindi, Modern Hebrew, Korean, and Turkish include an instance of even as part of their NPIs.

220   

   Elena Herburger and Simon Mauck

As the alternative domains are smaller than the original, widened one, a proposition involving an element in the original (i.e. wider) domain will only entail the alternative propositions when embedded in a downward entailing context. Thus, *I saw anyone is synonymous with I saw someone, except that the domain of quantification is wider in the NPI sentence. If the widened domain of quantification consists of {a,b,c}, the alternative domains are {a, b}, {a, c}, {b,c}. The Ec operator then says that the sentence must entail all alternatives generated with the help of the alternative domains triggered by any. But it will fail to do so; if it was a that the speaker saw, the alternative involving {b,c} will be false. This, combined with the action of the s operator in ‘freezing’ the pragmatics, rules out *I saw anyone. By the same token, I didn’t see anyone is ruled in. The alternatives will be the same as before. Unlike before, however, applying the Ec operator will not result in contradiction, as all the alternative propositions are entailed by the original proposition; if the speaker did not see any individual from the set {a, b, c} then she will not have seen any from the sets {a, b}, {a, c}, or {b, c} either. The widened domains are said to satisfy the requirements of the even-like Ec operator in downward entailing environments, similar to what we saw in Krifka’s and Lahiri’s accounts.

3 Issues raised by the ‘even’ analyses Despite their undeniable appeal, the analyses just summarized raise a number of issues, the most obvious perhaps being that they explain distribution of NPIs in terms of felicity but unlicensed NPIs do not so much give rise to infelicity but to ungrammaticality (Giannakidou 2001). Unlike the examples in (22), which are pragmatically bizarre but grammatically well-formed, our initial examples would not be pragmatically bizarre without the licensors but grammatically ill-formed: (22)

a. # Even Dick Cheney wanted to invade Iraq. b. # Even one person attended the reception.

(23)

a. * I see anything. b. * Some student(s) with any sense asked for help. c. * Nancy crossed the border with any problem.

A more theory-internal problem stems from the assumption that the ‘even’-like operator the advocates of the semantic answer to the licensing puzzle postu-

The chance of being an NPI   

   221

late (or exploit) needs to take wide scope over the element responsible for the DE context. If it took narrow scope then for the right truth conditions to follow it could not pick out the least likely or semantically strongest alternative but would need to pick out the most likely or semantically weakest one. While this in itself would not be problematic – it would just mean there are two ‘evens’ (cf. Rooth 1985 a.o.) – it would also mean that the relevant ‘even’ operator is itself an NPI. This would render the analysis circular. Therefore, for the analysis to retain its explanatory force the ‘even’-like operator must necessarily take scope over the element creating the DE context. What speaks against this, however, are instances of NPIs that are licensed across domains that bar scopal movement. It is not clear, for instance, how a tacit ‘even’-like operator lexically associated with anyone could move out of the because clause and above the negation in (24a), or how the two instances of any could move out of the relative clause and above the negation in (24b):3 (24)

a. Mary didn’t send the e-mail [because anyone had asked her to] but because she wanted to. b. They wouldn’t confirm the appointment of a judge [who had voiced any opinion about any controversial matter].

A fourth, quite different concern the semantic answer to the licensing puzzle raises is that English even – and thus perhaps ‘even’ in other languages, including those with overt ‘even’ NPIs – may not really pick out the least likely, semantically strongest alternative but the most informative or noteworthy one, which at times may not be the semantically strongest one. This observation regarding even is due to Kay (1990) and supported by the following example of his: (25)

A: It looks as if Mary is doing well at Consolidated Widget. George [the second vice president] likes her work. B: That’s nothing. Even Bill [the president] likes her work.

3 This would not be a problem under Chierchia’s account, since for him the ‘even’ operator can be generated anywhere in the syntax as can the s-operator that checks the [+s] feature on the NPI. This free generation of s, however, means that Chierchia cannot exploit what could be an advantage of his proposal: If the feature were not checked, it would actually lead to ungrammaticality rather than the unsuitability that all lexicalist accounts seem to predict.

222   

   Elena Herburger and Simon Mauck

B’s reply indicates that Bill’s approval of Mary’s work is more significant than George’s, even if it is not necessarily less likely (cf. also Herburger 2000). If, however, even really picks out the most noteworthy or contextually significant alternative it becomes difficult to derive the distribution of NPIs from the presence of an ‘even’-like operator where the semantics of this operator crucially exploits the inverse relation between likelihood and semantic strength. Of course, one can posit a more abstract operator that is defined to pick out the semantically strongest alternative rather than the pragmatically most noteworthy one. But such a move would seem to undermine the argument from languages like Hindi, where NPIs seem to contain an overt instance of ‘even’. We may at this point also note that also faces a purely descriptive problem, as also noted Szabolcsi (2004). The proposal is predicated on the assumption that NPIs denote bottom scalar elements. Though this is true in a great many cases, it is not true in all. As discussed in detail in Israel (1996), not all NPIs are low scalar expressions: (26)

La película *(no) fue gran cosa. the film not was big thing ‘The film was not great (= bad)’

(27)

a. He is*(n’t) much of a cook. b. It won’t/*(will) be long before they notice. c. Writers do*(n’t) accept suggestion much these days. (Klima 1964: 283)

Clearly, gran cosa, much of a, long and much here are not low scalar expressions but rather relatively high scalar ones. But they function as NPIs nonetheless. A final reason to remain skeptical of the semantic answer to the licensing puzzle is the existence of pairs like some and any. Arguably the (unstressed) NPI any and the PPI some have the same meaning (e.g. Krifka 1995, Lahiri 1998). That is, (28b) is the negation of (28a) and (29b) that of (29a): (28)

a. I have some potatoes. b. I don’t have any potatoes.

(29)

a. I have potatoes. b. I don’t have potatoes.

The chance of being an NPI   

   223

The domain widening that can be observed in (Kadmon and Landman 1993) seems to be due to stress on the determiner rather than the semantics of the determiner itself, a point also made in Krifka (1995) and Lahiri (1998): (30)

I don’t have ANY potatoes.

If, however, some and the NPI any are semantically equivalent existential determiners, it cannot be the meaning of any that is responsible for its distribution as an NPI. Rather, as we have seen, two additional assumptions are needed, namely (i) the assumption that any and its kin but not some induce ‘widening’, ‘more specific alternatives’ or ‘higher ranked focus alternatives’, and (ii) the assumption that any comes with an operator requiring it to be stronger than its alternatives. Absent strong independent evidence for (i) or (ii), we might as well say that the difference between any and some is that any is an NPI and some is not. Or, put differently, we might say that the NPI any carries an NPI-feature that must be checked by an element that bears a negation-feature. In contrast with any, some does not bear this feature.

4 The pragmatics of scales and the NPI-feature Our conclusion that what makes a particular expression an NPI is but a semantically empty NPI-feature raises a number of non-trivial questions, the most obvious of which is perhaps this: If NPI-hood just amounts to carrying a feature, why are so many NPIs low scalar elements? And why are the others relatively high scalar expressions and so much less frequent? It seems significant that the strength of low scalar items in negative contexts makes them pragmatically useful if what one is after is a strong or emphatic claim (Jespersen 1971; Israel 1996 a.o.). This is clearly true of NPIs that lexically express something of little value (a red cent, etc.), but it also holds true of NPIs that are merely low scalar expressions and synchronically, if not diachronically, lack the colorful content of minimizers, e.g. the NPI any-. We surmise that it is this pragmatic usefulness that leads some low scalar elements to bear an NPI-feature. Which ones do, however, seems to be ultimately up to chance. Relatively high scalar elements are also pragmatically valuable in DE contexts. But instead of contributing to semantically strong, emphatic, and in some instances hyperbolic claims, they do the exact opposite; they lend themselves to weak assertions, and the attending equivocation, and also understatement (cf. Israel 1996). (31), for example, can be taken to mean that the speaker has some

224   

   Elena Herburger and Simon Mauck

but not much time (weak, equivocal claim) or that she has no time at all (understatement): (31)

I don’t have much time.

The weak claim arises from a regular scalar implicature due to Quantity, which implicates the negation of stronger alternatives. Since negation reverses the direction of entailment, the stronger alternative to much is some. (31) therefore implicates that it is not the case that the speaker does not have some time, which means she has some time. It seems reasonable to assume that an understated claim arises when Quantity is suspended and the speaker is taken to not have said everything there was to say. Since without Quantity no scalar implicature is generated in (31), the sentence no longer implicates that the speaker has some time. The prevalence of bottom scale NPIs as opposed to high scalar NPIs, which has all but obscured the existence of the latter, suggests that we more often put a premium on clarity and emphasis when negation is involved than on weakness, vagueness or understatement. In both instances, however, the scalar rung has useful pragmatic consequences in negative environments. So useful, we argued, that some of the relevant expressions come to be restricted to negative environments.

5 The historic development of NPIs, including ‘even’ NPIs 5.1 Acquisition, change or loss of the NPI-feature Evidence for the “chanciness” of NPI-hood that we claim to exist comes from the curious fact that a scalar element can be an NPI in one language without being so in another, cf. for instance the English NPI any with its merely existential German counterpart einig-. It also ties in with the observation that there can be two semantically equivalent elements in one language only one of which is an NPI. As we already saw, English any vs. some fit this description, as do German brauchen vs. müssen (‘must’). Finally, if NPI-hood is nothing deep but only a feature, we expect that NPI-hood may not be stable across time. One way to understand the Jespersen cycle is to say that the NPI-feature sometimes acquires semantic content and turns into a negation-feature. For instance, French pas (‘step’) famously developed from an NPI-like companion of

The chance of being an NPI   

   225

ne to a negation in its own right in colloquial French. Similarly, English never can be viewed as having an NPI in Old English and Middle English, licensed by ne, but now in standard English it functions as a negative expression in its own right. In our terms this means that in both instances the NPI-feature has acquired semantic content and turned into a negation-feature (cf. Herburger 2001).4 (32)

He nevere yet no vileynye ne seyde In al He n-ever yet n- vile-thing not said in all his Lif unto no maner wight his life to n- manner man ‘He never said any vile thing in all his life to any man’ (Middle English, 14th century; Chaucer)

The opposite change, namely that of a low scalar NPI becoming a regular indefinite (i.e. that of an NPI losing its NPI-feature feature) does not seem to be as well attested. Jäger (2006, 2007), however, argues that in Old High German, einigfunctioned as an NPI in the scope of negation, doubt, without, in yes-no questions, and in the antecedent of a conditional. In (33) for instance, it appears in the scope of without. (33)

buuzssan einigan zuuiuun ist dhanne archennit, dhazs… without any doubt is then recognized that ‘It is recognized without any doubt that…’ (Isidor VI.5, cited after Jäger 2007)

In its journey from Old High German through Middle High German to Modern German einig then must have lost its NPI-featrue since now it is used as an existential with free distribution. We take this to indicate that just as NPI-hood can be acquired and transformed it can also be lost. This makes sense if NPI-hood is a semantically inert feature born by some semantically predisposed expression. It makes less sense if NPI-hood is intrinsically linked to the semantics of an expression.

4 We assume that the Negative Concord that never and other currently negative expressions participated in their previous history (and, in some dialects, in current practice) is a phenomenon closely related to NPI licensing. This is a widely held view (cf. e.g. Dowty 1994, Herburger 2001, Szabolcsi 2004 and many others.) It is, however, not a universally held position. Zeijlstra (2004) argues that Negative Concord and NPI licensing are two separate phenomena, see also Penka (2010). Limitations of space prevent us from addressing their arguments here.

226   

   Elena Herburger and Simon Mauck

5.2 Analogous development of overt ‘even’ NPIs Interestingly, the acquisition, change or loss of the NPI feature seems to affect ‘even’ NPIs in much the same way it affects other NPIs. This would be surprising if the distribution of ‘even’ NPIs were determined by the ‘even’ they contain, but is less surprising if ‘even’ NPIs are just like other NPIs and characterized by an NPI-feature. The change in the NPI-hood of an ‘even’ NPI can be illustrated with data from Modern Hebrew. Like Hindi, Modern Hebrew has morphologically complex NPIs containing an instance of ‘even’ (af) that generally require negation to be licensed. This, however, is not the case in elliptical answers, which may suggest that they are already on the path of becoming negative expressions (NEs) in their own right (cf. Herburger 2001): (34)

A: mi ba? who came ‘Who came?’ B: af exad (*lo) even one not ‘No one’

What further lends credence to the suspicion that af exad and similar expressions are at this point not only NPIs in Hebrew but also NEs is the existence of expressions where the NPI-feature has acquired semantic content. In particular, in the speech of some children and teenagers af exad is starting to be used without negation in preverbal position: (35)

a. lo ra… af exad amar she-zeh ra not bad… even one said that-this bad ‘It’s not bad. No one said that it was bad.’ http://www.facebook.com/BTV.OMRI/posts/10150169100767529 b. …kimat af exad ba im kova… …almost even one came with hat ‘…almost no one came with a hat…’ http://www.mynet.co.il/articles/0,7340,L-3715469,00.html

If the limited distribution of NPIs depended entirely on the internal morphological make-up of the relevant expressions (‘even’ + ‘one’), then, given that this morphological make-up presumably does not change for individual speakers

The chance of being an NPI   

   227

depending on syntactic environment, the change we described above would be hard to explain. On the other hand, on the present view of NPIs we can say there currently two af exad items in the lexicon of Hebrew, one bearing a contentless NPI-feature, meaning ‘anybody’, the other bearing a contentful negation-feature meaning ‘nobody’. In relevant respects af exad thus resembles Spanish nadie, which has been argued to be currently both an NPI and an NE (Herburger 2001). Like other NPIs, even ‘even’ NPIs can also lose their NPI-hood. This arguably happened in the historical development of Greek kanenas. According to Kiparski and Condoravdi (2006), kanenas derives from kai (‘and’), an (‘even’), and henas (‘one’). They report that it was used as an emphatic NPI in early Medieval Greek but ceased to be used as such in later Medieval Greek, where it came to be employed as regular, non-emphatic NPI: (36)

kai den eglutōsen kanenas and not escaped anyone ‘And not one of them escaped’ (Makhairas 16.15, cited after Kiparsky and Condoravdi 2006)

What is of special significance here is Kiparski and Condoravdi’s observation that in some dialects of Modern Greek, notably Cretan, kanenas is now used as a simple positive indefinite, as shown in (37). It seems then that the ‘even one’ expression kanenas in Cretan resembles German einig in being a low scalar NPI that has ceased to be an NPI and become a ‘regular’ low scalar expression instead. (37)

kianenas perastikos da perase ki ēkopse ta portokalia some passerby here passed-by and cut the oranges ‘Some passerby must have passed by and cut the oranges’

Finally, the history of some Indo-Aryan languages may offer another example of an ‘even’ NPI reverting to a simple indefinite. Sanskrit formed both indefinites and NPIs by combining a wh-word with a focus particle or a logical connective such as va (‘or’). For instance, the combination of kas ‘who’ plus chid ‘even’ resulted in a word much like English any, usable as an NPI as well as a free choice item. The NPI use is illustrated in : (38)

pravishantam na maam kas chid apashyan enter-pres.ppl.acc not me-acc who even see-impf.3p ‘No one (not anyone) saw me entering’ (Mahabharata, Vanaparva, Nala 2,18)

228   

   Elena Herburger and Simon Mauck

Curiously, in Bhojpuri and Hindi, both descendants of Sanskrit, the neuter form of kas chid, kim chid, developed into a regular indefinite kuch ‘something’ (cf. the kuch that appears in the Hindi NPI and free choice element kuch bhii discussed by Lahiri 1998): (39)

ham kuch leb (Bhojpuri) I something will-take ‘I’ll take something’

The development of kas chid > kuch offers another instance of an ‘even’ NPI losing its sensitivity to DE contexts, i.e. its losing its NPI-feature, and turning to a simple indefinite, which, in turn, may combine with an instance of ‘even’, i.e. bhii, to form an ‘even’ NPI.

6 Licensing the NPI feature 6.1 NPI licensing and local polarity Since we concluded that the limited distribution of NPIs is best attributed to a semantically vacuous and, as we just saw, historically unstable feature, we should now address how this feature is licensed. Our task would be more straightforward if NPIs were only licensed under c-command. As, however, Ladusaw’s (1979) discussion of Klima (1964) shows, c-command between licensor and licensee is not the relevant relation as it fails to capture the difference between every and no; the former indeed licenses NPIs in its syntactic c-command domain (its restriction), but the latter also licenses NPIs in its scope, the syntactic correlate of which the determiner does not c-command. Luckily, in our search for an account with the requisite syntactic properties we can borrow extensively from Sánchez Valencia’s (1991) theory regarding the relation between the monotonicity properties of lexical items and those of the sentences they appear in. The relevance of this theory for NPI licensing is also pointed out in Dowty (1994) and Ludlow (2002), though in a somewhat different way. Working within categorial grammar, Sánchez Valencia (1991) develops a calculus of the DE-ness or monotonicity of a sentence and its parts. An essential ingredient of his theory is the claim that lexical items are marked for the monotonicity of their arguments. Assuming that determiners denote functions from sets of individuals to sets of sets of individuals, every, some and no carry the lexical features in (40a–c). Every, for instance, marks its syntactic sister as downward

The chance of being an NPI   

   229

monotonic and the sister of the QP that dominates it as upward monotonic. Sentential negation bears the feature marking in (40d). (If the determiners are interpreted in situ they denote functions of type