Mood, Aspect, Modality Revisited: New Answers to Old Questions 9780226363660

Citation preview

Mood, Aspect, Modality Revisited

Mood, Aspect, Modality Revisited New Answers to Old Questions

Edited by Joanna Błaszczak Anastasia Giannakidou Dorota Klimek-Jankowska Krzysztof Migdalski The University of Chicago Press

Chicago and London

The University of Chicago Press, Chicago 60637 The University of Chicago Press, Ltd., London © 2016 by The University of Chicago All rights reserved. Published 2016. Printed in the United States of Amer ica 25 24 23 22 21 20 19 18 17 16

1 2 3 4 5

ISBN-13: 978-0-226-36352-3 (cloth) ISBN-13: 978-0-226-36366-0 (e-book) DOI: 10.7208/chicago/9780226363660.001.0001 Library of Congress Cataloging-in-Publication Data Names: Błaszczak, Joanna, editor. | Giannakidou, Anastasia, editor. | Klimek-Jankowska, Dorota, editor. | Migdalski, Krzysztof Marek, 1975– editor. Title: Mood, aspect, modality revisited : new answers to old questions / edited by Joanna Błaszczak, Anastasia Giannakidou, Dorota Klimek-Jankowska, Krzysztof Migdalski. Description: Chicago ; London : The University of Chicago Press, 2016. | Includes bibliographical references and index. Identifiers: LCCN 2016006708 | ISBN 9780226363523 (cloth : alk. paper) | ISBN 9780226363660 (e-book) Subjects: LCSH: Grammar, Comparative and general—Tense. | Grammar, Comparative and general—Mood. | Grammar, Comparative and general—Aspect. Classification: LCC P281 .M655 2016 | DDC 415—dc23 LC record available at http:// lccn .loc.gov/2016006708 The paper used in this publication meets the minimum requirements of the American National Standard for Information Sciences—Permanence of Paper for Printed Library Materials, ANSI Z39.48-1992.

Contents Preface

vii

PART I .

Tense, Aspect, and Modals: Their Categorial Status and Cross-linguistic Variation

chapter 1.

TAM Coding and Temporal Interpretation in West African Languages 6 Anne Mucha and Malte Zimmermann

chapter 2.

Modals: Meaning Categories?

45

Valentine Hacquard

chapter 3.

Epistemic Future and Epistemic MUST: Nonveridicality, Evidence, and Partial Knowledge 75 Anastasia Giannakidou and Alda Mari

PART I I .

Irrealis Moods: Subjunctive and Imperative

chapter 4.

On Finiteness and the Left Periphery: Focusing on Subjunctive 125 Manuela Ambar

chapter 5.

Evaluative Subjunctive and Nonveridicality Anastasia Giannakidou

chapter 6.

The Essence of a Category: Lessons from the Subjunctive 218 Martina Wiltschko

177

Contents

vi

chapter 7.

Imperatives as (Non-)modals

255

Mark Jary and Mikhail Kissine

chapter 8.

Approaching the Morphosyntax and Semantics of Mood 286 Ilse Zimmermann

PART I I I .

chapter 9.

Aspectual Recursion and Aspectual Coercion Aspectual Composition and Recursion

314

Henriëtte de Swart

chapter 10. Can Semantic Theories Be Tested Experimentally? The Case of Aspectual Coercion 346 Oliver Bott

chapter 11. Aspectual Coercion versus Blocking: Experimental Evidence from an ERP Study of Polish Converbs 381 Joanna Błaszczak and Dorota Klimek-Jankowska

About the Editors 437 About the Contributors 441 Index

445

Preface

T

he chapters in this volume were presented at the international TRAIT workshop “How Categorical Are Categories?” The workshop was organized at the University of Wrocław in January 2013 as part of the project “Understanding Linguistic Categories” funded by the Foundation for Polish Science within the FOCUS program. As the workshop title suggests, the driving force behind this volume is the question of what constitutes a linguistic category. Each chapter explores this question as it relates to the three traditional categories of tense, aspect, and mood. We call these categories traditional because they have long been acknowledged and studied by philologists and linguists and are by now classic in the sense that consensus has been reached that these categories have certain properties and manifest themselves in predictable ways that include patterns of expected interaction. Given this broad agreement, what new can be said about them? In fact, a lot can be said—if one addresses seriously what the labels subjunctive, imperative, future, aspect, and modality refer to. The literature tends to assume a straightforward mapping between grammatical category—for example, future ‘tense’, perfective ‘aspect’, modal ‘verb’— and semantic function; and the description of well-studied languages (e.g., Slavic with respect to aspect, Romance with respect to mood, English with respect to modals) cultivates a sense of stability and predictability in patterns that, however, vanishes once we branch out to lesser-known

viii

Preface

patterns and languages. Our goal in this volume is to reveal some of these lesser-known patterns and raise awareness about the consequences the study of these patterns have for our concept of linguistic category and the mapping between morphology and syntax on the one hand, and semantics and pragmatics on the other. Consider, as a brief illustration, the labels ‘tense’ and ‘mood’. In the literature, tense indicates the tense morphology used for temporal orientation. Tense is thus a morphological category—that is, the grammatical category realized in the system of a language. In a language such as English, the tenses are bound morphemes on the verbs; but in Mandarin morphemes that can be characterized as tenses are not attached to the verb, and they can be particles or adverbs, operating at the sentence rather than the verb level. In Modern Greek, the future tense appears to be a particle, in contrast with the other morphological tenses in Greek, which appear on the verb. Similar variation exists in the morphological realization of mood: in Romance languages, mood is a verbal category; but in Greek, mood distinctions appear external to the verb in the form of particles. In Native American languages, prospective aspect can play the role of mood or future; and in African languages, there is considerable syncretism in the morphological categories of tense and aspect. Often it is only by appealing to semantic functions that we can make meaningful distinctions. Hence we have already identified two important levels of analysis for what constitutes a linguistic category: the morphological level and the notional or semantic level. Crucially, the grammatical category of a par ticular expression—that is, whether it is a bound morpheme, a verb, an adverb, or a particle—does not predict its semantic function or its role within the notional system of the language. For instance, adverbs and bound morphemes can carry temporal information; likewise, aspectual markers can be particles, verbs, or suffi xes. In many languages, modality is expressed by both verbs and adverbs. In Greek and German, modal particles have meaning that is typically assigned to modal verbs in English, and in Italian an apparent tense suffi x (the future ‘tense’) gets a modal semantics. Such apparent mismatches are in fact quite common, suggesting that the mapping between morphological and notional categories is not isomorphic or universal, and that it needs to be addressed within the systems of specific languages. The volume provides a forum for discussion of the nature of linguistic diversity in the syntax and semantics of irrealis moods (subjunctive, im-

Preface

ix

perative), modality, and aspectual markers. Are these universally separate notional or morphological categories with clear- cut distinctions, or should they be understood as less discrete, conceptually related categories? If the latter option should turn out to be correct or more convincing, then how categorical is the semantics of the respective morphemes? The overarching question we are interested in is how the status of the traditional grammatical distinctions changes in light of linguistic variation, with emphasis on the growing body of evidence from languages that have previously been under-investigated. We approach these questions within a unified perspective (the question of what a linguistic category is) and by bringing together theoretical as well as experimental perspectives. The contributions in this volume can be grouped into three thematic sections: Section 1: Tense, Aspect, and Modals: Their Categorial Status and Cross-linguistic Variation Section 2: Irrealis Moods: Subjunctive and Imperative Section 3: Aspectual Recursion and Aspectual Coercion

We offer more commentary in separate introductions to each section. We want to thank the authors for their valuable contributions, as well as for working carefully with us to revise and fi nalize their chapters. We also want to thank our editor, Chris Rhodes, for his help and guidance, and three anonymous reviewers for reading the manuscript and offering generous and insightful feedback on each chapter. Finally, we acknowledge the support we received from the Foundation for Polish Science as part of the FOCUS program, without which this publication would not be possible. The Editors

chapter one

TAM Coding and Temporal Interpretation in West African Languages Anne Mucha and Malte Zimmermann

1

Introduction

T

his chapter discusses the temporal interpretation and the grammatical coding of tense, aspect, and modality (TAM) in two West African languages Hausa (Chadic, Afro-Asiatic) and Medumba (Grassfield Bantu, Niger- Congo). We chose these two languages because their tensemarking systems are radically different. As demonstrated in Mucha (2012, 2013), Hausa belongs to the typological class of tenseless languages, in which tense is not grammatically marked and temporal interpretation is pragmatically resolved, relying on aspectual and contextual cues. By contrast, Medumba belongs to the typological class of graded-tense languages (Comrie 1985; Cable 2013), which are able to express more fi negrained distinctions in past- and future- oriented interpretations, such as recent past or remote past. As a result, temporal interpretation in Medumba is expected to rely less on aspectual and contextual cues. Despite this striking difference in tense coding, we will show that the two languages are surprisingly similar in other respects. First, both languages have clearly identifiable linguistic categories of aspect and modality, which shows that the presence of these categories in a language is independent of the presence or absence of the category of tense. Second, both languages express future- oriented readings with a modal element, show-

TAM Coding and Temporal Interpretation

7

ing that there is no future tense in these languages and, perhaps, crosslinguistically (Enç 1996; Matthewson 2006; Copley 2009). Third, both languages rely on aspectual and contextual cues for temporal interpretation, providing support for the analysis of tense as a deictic category, relating the reference time (RT) to the utterance time (UT). The chapter is structured as follows. In the remainder of the introduction, we lay out the background assumptions concerning the semantic concepts underlying tense, aspect, and modality, as well as the notion of linguistic categories in natural language. We also identify a number of problems that are often encountered in semantic field research on TAM categories. Section  2 presents our analysis of Hausa as a tenseless language, which follows the more detailed account by Mucha (2013). The central claims are that Hausa does not have a linguistic category of tense, such that temporal interpretation is determined by the interaction of Aktionsart, aspect, and context, and that the so- called future marker is in fact a modal marker. Section 3 presents the fi rst ever sketch of a formalsemantic analysis of tense-aspect marking in Medumba, comparing it with another graded-tense language, Gĩkũyũ, as discussed by Cable (2013). We propose that some of the temporal markers in Medumba are indeed tense markers. In addition, Medumba has two sets of aspectual markers located in different structural positions. From a theoretical point of view this means that what is perceived as a unified typological class of graded-tense languages splits up into various subclasses that make use of different formal means for expressing temporal information. In addition to tense, Medumba has two aspectual categories, and, as in Hausa, future-oriented readings are expressed by means of a modal marker. Section 4 concludes.

1.1

Semantic Concepts Underlying Tense, Aspect, and Modality

We adopt the following defi nitions for the semantic concepts underlying the notions of tense, aspect, and modality, respectively. Following Reichenbach (1947) and Klein (1994), tense markers express the temporal relation between the utterance time (UT) and the reference time (RT) relative to which the proposition expressed is evaluated. There are three basic relations between UT and RT (Comrie 1985): simultaneity (UT = RT, or UT ⊆ RT) for present time reference, anteriority (RT < UT) for past time reference, and posteriority (UT < RT) for future time reference. As for grammatical coding, we follow Partee (1973) and Kratzer (1998) in treating tense as introducing a variable expression ti in T. This

chapter one

8

indexed tense variable gets a value g(i) by means of a contextual assignment, where g(i) corresponds to RT. Possible value assignment to ti in a given context is restricted by tense specifications. The tense specification adjoins to ti and denotes a partial identity function giving back a value if and only if g(i) stands in the relevant relation to UT as illustrated for the past tense in (1) (see Heim 1994): (1)

T 3 ti

PAST

λt: t < UT. t

Aspect refers to the temporal relation between RT and the event time (ET), which is the time at which the described event takes place (Klein 1994). The temporal relations involved are subset (⊆) or precedence relations ( past (RT < UT) >> future (UT < RT), where past interpretations are more complex than present interpretations, as they require a temporal shift, and future interpretations are more complex than past interpretations, as they require a modal displacement in addition to temporal shift. As we have stated repeatedly here, these default principles can be overruled by contextual information in accordance with the following principle: (20)

Contextual Reference Time Anchoring: Explicit temporal information may override pragmatic defaults. If the previous discourse context provides an RT alternative to the pragmatic default, this RT can serve as a temporal anchor for the time variable of the sentence.

We demonstrate the pragmatic resolution of RT with perfective and progressive sentences in (21a, b), which receive a default past and present interpretation, respectively, in the absence of further contextual information: (21)

a. Bill Bill

yaa 3sg.m.pfv

‘Bill played.’

yi do

wàasaa. play

TAM Coding and Temporal Interpretation

b. Bill Bill

ya-nàa 3sg.m-prog

19

wàasaa. play

‘Bill is playing.’ The semantic interpretations in (22a, b) reflect the semantic contribution of perfective and progressive aspect marking (cf. (2a, b)), but leave the value for RT open for pragmatic resolution. (22)

a. [[(21a)]] = λt.λw.∃e[ play(hawwa, e, w) ˄ time(e) ⊆ t ] b. [[(21b)]] = λt.λw.∃e[ play(hawwa, e, w) ˄ t ⊆ time(e) ]

Looking at (21a) first, a present interpretation is blocked by the BEC since the event is grammatically marked as temporally bounded by the perfective aspect. Of the remaining temporal interpretations, the past interpretation is chosen as the default interpretation over the future interpretation in accordance with the SPI, since the past interpretation is simpler. Again, this default past interpretation can be overridden by contextual considerations. By contrast, a present interpretation for (21b) is not blocked by the BEC, as progressive aspect marks the event as temporally unbounded. DP and SPI then conspire to yield the default present interpretation, which, again, can be overridden by contextual information.

2.5

Future- Oriented Readings with zaa

To conclude the discussion of Hausa, we show how future- oriented readings come about in sentences containing the future marker zaa. Following Mucha (2013), we argue that such future- oriented meanings consist of two components, a modal component expressed by zaa, and an aspectual component expressed by so- called subjunctive aspect, which shifts ET to a time after RT. Because of its ET-shifting nature, which is attested in a number of different structural environments (Schuh 2003; Mucha 2013), we reanalyze the so- called subjunctive marker as a marker of prospective aspect with the meaning in (23): (23) [[prosp]]g = λP.λe.λt.λw. [P(e)(w) & t < time(e)] Crucially, we treat the prospective aspect in (23) as semantically deficient. It differs from the run- of-the mill aspects PROG and PFV in (2a, b) in

chapter one

20

that it does not introduce existential closure of the event argument. Empirical evidence for this claim comes from the observation that the prospective is restricted to sentential environments that do not assert the existence of an ongoing or completed event. This is the case, for instance, with negated sentences or in hortatives; see Schuh (2003) and Mucha (2013) for extended discussion. We conclude that the restriction that the modal marker zaa combine with PROSP, but not with PROG or PFV, is due to the former’s semantic deficiency in the event domain. 5 The modal marker zaa itself is analyzed as a root modal that is realized in a structural position above Asp. It denotes a universal quantifier over possible worlds that are ranked highest with respect to a bouletic or inertial ordering source O (Copley 2009; Mucha 2013) given a set of facts provided by the realistic modal base MB in the utterance context: (24)

[[zaa]]g = λP. λt.λw.w’[w’∈MAXO(w)(t)(w)(MB(w)(t)) → ∃e[P(e)(t)(w’)]; only defi ned if MB is realistic and O bouletic or inertial

We follow Copley (2009) in assuming that bouletic and inertial orderings give rise to the future-oriented modal readings of intention and prediction, respectively, and Kratzer (2012a) in suggesting that all modal markers have a realistic modal base. The meaning components of prospective shift and modality have been argued to be lexicalized in one morpheme in the English future marker will (Enç 1996; Copley 2009), in St’àt’imcets kelh (Rullmann, Matthewson, and Davis 2008), and in Guaraní -ta (Tonhauser 2011b). Hausa differs from these languages in that the two meaning components are expressed by two independent grammatical markers. The overall future meaning can be compositionally derived as shown in (25), where g(i) is the value contextually assigned to the tense variable: (25)

a. [[Hàwwa zaa tà gudù]]g = [[zaa]]g ([[Prosp]]g ([[vP]])) Hawwa fut 3.sg.f.prosp run b. = [λP.λt.λw.∀w’[w’∈MAXO(w)(t)(MB(w)(t)) → ∃e[P(e)(t)(w’)]] ([λP.λe.λt.λw. P(e)(w) & t < time(e) ] (λe.λw. run(hawwa, e, w) )

TAM Coding and Temporal Interpretation

21

c. = [λP. λt.λw.∀w’[w’∈MAXO(w)(t)(MB(w)(t)) → ∃e[P(e)(t)(w’)]] (λe.λt.λw. run(hawwa, e, w) & t < time(e) ) d. = λt.λw.∀w’[w’∈MAXO(w)(t)(MB(w)(t)) → ∃e[run(hawwa,e,w’) & t < time(e) ]] e. = λw.∀w’[w’∈MAXO(w)(g(i))(MB(w)(g(i))) → ∃e[run(hawwa,e,w’) & g(i) < time(e) ]] f. = 1 iff in all worlds w’ that are accessible from w with bouletic/inertial ordering O on a realistic modal base MB (relative to the contextually provided RT g(i)): there is an event e of Hàwwa running and the time of e is subsequent to g(i). The analysis of the future modal zaa makes a number of correct predictions. First, the lexical restriction to bouletic and inertial ordering sources ensures that zaa is used for expressing intentions and predictions, which typically trigger future interpretations. Second, the fact that zaa is lexically restricted to co-occur with the prospective aspect (cf. (8)) accounts for the fact that sentences with zaa can never receive non-future modal readings, such as epistemic necessity. Finally, since zaa is a necessity modal expressing universal quantification over possible worlds, it cannot be felicitously used in contexts requiring a possibility modal: (26)

Context: It’s Sunday, so is Hàwwa going to have anything special for lunch? # Bà-n sanìi ba, àmmaa à yâu zaa tà dafàa shinkaafaa. neg-1sg know neg but today fut prosp cook rice intended: ‘I don’t know, but she might cook rice.’ Comment: “The certainty clashes with bàn sani ba.”

The analysis of modality and tense in future- oriented sentences in Hausa completes our examination of Hausa as a tenseless language. Hausa does not restrict the location of RT relative to UT in its grammatical system, either overtly or covertly. Instead, temporal interpretation in Hausa is pragmatically resolved, where pragmatic resolution is influenced by inner

chapter one

22

aspect (Aktionsart), grammatical aspect, temporal adjuncts, and the general context.

3

TAM Categories and Temporal Interpretation in a Graded-Tense Language: Medumba

This section presents the first steps toward a formal-semantic analysis of the TAM system of the graded-tense language Medumba. Drawing on original fieldwork, we show (i) that the graded-tense effects in temporal interpretation are not due to the presence of multiple formal tense distinctions but come about through the interaction of different linguistic categories; (ii) that the TAM system of Medumba is made up of markers belonging to four different categories, including tense, aspect, and modal markers; (iii) that the interpretation of temporally unmarked sentences (STSs) and the expression of future-oriented readings in Hausa (tenseless) and Medumba (graded tense) is subject to the same underlying principles.

3.1 Background Medumba (Grassfields Bantu, Niger- Congo) is a tone language spoken in the western region of Cameroon. While the basic word order of Medumba is SVO, the verb is frequently preceded by temporal markers. Previous descriptions of the Medumba TAM system are found in Nganmou (1991) and Kouankem (2012). Even though the two studies do not agree on the inventory of temporal markers, or on the exact meaning of these markers, they do assume a unified graded-tense paradigm expressing different remoteness distinctions between RT and UT. In this section, we argue that the temporal markers in question show diverse syntactic and semantic behavior, for which reason they should not be analyzed as a single category of (graded) tense.

3.2

A Formal Analysis of a Graded-Tense System: Gĩkũyũ

In his pioneering study of the tense system of Gĩkũyũ (Narrow Bantu, Niger- Congo), Cable (2013) reports that, descriptively, the language encodes three remoteness distinctions in the past, namely current past (27a), near past (27b), and remote past (27c), and two in the future, which are current future (27d) and remote future (27e).

TAM Coding and Temporal Interpretation

(27)

a. Mwangi Mwangi

23

nĩ-a-kũ-in-aga asrt-3sg.s - CUR- dance-pst.imp

‘Mwangi was dancing (earlier within the day).’ b. Mwangi Mwangi

ni-a-ra-in-aga asrt-3sg.s- NRP- dance-pst.imp

‘Mwangi was dancing (before today, but recently).’ c. Mwangi Mwangi

nĩ -a-a-in-aga asrt-3sg.s-REMP- dance-pst.imp

‘Mwangi was dancing (some time ago, not recently).’ d. Mwangi Mwangi

nĩ-a- kũ-⌀-in-a asrt-3sg.s- CUR-FUT- dance-fv

‘Mwangi will dance (within the day; or soon).’ e. Mwangi Mwangi

nĩ-a-ka- ⌀-in-a asrt-3sg.s-REMF-FUT- dance-fv

‘Mwangi will dance (sometime after today).’ The central result of Cable’s study is that the temporal remoteness morphemes (TRMs) involved in (27) are not tense markers in the sense of Klein (1994), Kratzer (1998), and others, as they do not modify RT, but the event time of a clause. This claim is supported with examples such as (28), in which RT (earlier today = current past) and ET (yesterday = near past) denote distinct time intervals, and in which the near past ET marker is obligatory. (28)

Context: Mwangi has been telling us for a while that he intends to travel to New York. Today we went to his house to say goodbye, but unbeknownst to us at the time, he had already left yesterday. Rĩĩria When

tũkinyire gwake, 1plS- CUR-arrive-P.PRV his

Mwangi Mwangi

nĩ-a-ra-thĩ-ĩte / * nĩ-a-kũ –thĩ-ĩte ASRT-3sgS-NRP -go-PERF/ ASRT-3sgS- CUR-go-PERF ‘When we arrived at his (house), Mwangi had already left.’

24

chapter one

From the mandatory occurrence of the near past form, which denotes a time interval including yesterday, Cable (2013) concludes that temporal remoteness morphemes in Gĩkũyũ always modify the event time, here the time of Mwangi’s leaving. If, by contrast, TRMs were to modify the reference time, which is here the time of visiting Mwangi’s house earlier on the day of utterance, the current past form should have been used. Another property of Gĩkũyũ TRMs is that the time intervals they denote are not distinct, but overlapping (Cable 2013). For instance, the remote past does not pick out a specific time interval prior to the utterance time but rather a large underspecified time interval reaching back a long time from the utterance time, and including the more specific time interval picked out by near past, which in turn includes the most specific current past time interval. This overlap hypothesis is based on the observation that speakers must use the least specific remote past if they are ignorant about the precise time at which an event occurred. Conversely, if they are fully informed about the event time, they have to make use of the most specific TRM consistent with their knowledge. Since the latter observation receives a principled account in Heim’s (1991) principle Maximize Presupposition, Cable (2013) concludes that the meaning contribution of Gĩkũyũ TRMs is presupposed, rather than asserted. According to Cable (2013), then, Gĩkũyũ does not make more fi negrained tense distinctions than English. Rather, the difference lies in the existence of a category of temporal markers that does not exist in familiar simple-tense languages, namely temporal remoteness morphemes, which denote partial identity functions over events, and which exist in addition to the well-known categories of tense, denoting a partial identity function over time intervals, and aspect, denoting properties of times. The following discussion of Medumba shows that there are also important differences within the typological class of graded-tense languages in the inventory of TAM categories. This suggests that graded-tense languages do not form a semantically unified class.

3.3 Temporal Markers in Medumba At fi rst glance, the paradigm of temporal markers in Medumba resembles that of Gĩkũyũ in that it seems to encode remoteness distinctions, albeit in an even more fine-grained fashion. The examples in (29) and (30) illustrate the meanings of the different past and future tense markers found in Medumba, according to Kouankem (2012).

TAM Coding and Temporal Interpretation

(29)

a. general/immediate past: Nana ⌀ ná ŋkwún Nana P0 cook beans ‘Nana just cooked beans.’ c. this morning: Nana cág ná ŋkwún Nana P 2 cook beans ‘Nana cooked beans.’ e. yesterday: Nana fə ná ŋkwún Nana P4 cook beans ‘Nana cooked beans.’ g. last month: Nana náɁ ná ŋkwún Nana P6 cook beans ‘Nana cooked beans.’

(30)

a. general future: Nana áɁ ná ŋkwún Nana F0 cook beans ‘Nana will cook beans.’

25

b. during the day: Nana yɔög ná ŋkwún Nana P1 cook beans ‘Nana cooked beans.’ d. last night: Nana zí ná ŋkwún Nana P3 cook beans ‘Nana cooked beans.’ f. last week: Nana lú ná Nana P5 cook

ŋkwún beans

‘Nana cooked beans.’ h. in the distant past: Nana luê ná ŋkwún Nana P 7 cook beans ‘Nana cooked beans.’ b. today (soon): Nana áɁ gɤúè 6 ná ŋkwún Nana F0+F1 cook beans ‘Nana will cook beans.’

c. today/this afternoon: d. tomorrow: Nana áɁ yɔög ná ŋkwún Nana áɁ cág ná ŋkwún Nana F0+F2 cook beans Nana F0+F3 cook beans ‘Nana will cook beans.’ e. remote future: Nana áɁ zí ná ŋkwún Nana F0 + F4 cook beans ‘Nana will cook beans.’

‘Nana will cook beans.’

chapter one

26

Earlier descriptions of Medumba assume a tense system with up to eight past distinctions and up to five future distinctions. Such claims are at odds with the fi ndings in our own fieldwork with native speakers of Medumba, which suggests that the markers in (29) and (30) should not all be treated on a par. More specifically, we propose that the TAM system of Medumba is made up of the following four distinct categories: (31)

i. ii. iii. iv.

Tense: Asp1: Asp2: Modal:

náɁ , fə 7 yɔög, cág, zí kéà, ⌀ áɁ

In our account, only the markers in the fi rst category are tense markers in the sense that they situate RT relative to UT. By contrast, the markers in the second category code a relation between ET and RT by temporally locating the event at a specific time of the day surrounding RT. As such, they express aspectual rather than tense information, and the corresponding category is therefore labeled Asp1. As the alert reader will have noticed in connection with (30), the markers yɔög, cág and zí show up in both past and future contexts, suggesting that they should not be considered to be past tense markers when occurring in isolation. While existing analyses treat the combinations of these particles with the future marker áɁ as lexicalized future tense forms that are unrelated to the past meanings, we propose a unified analysis in section  3.4. In addition to the time- of- day markers of the Asp1 class, Medumba also encodes the more familiar distinction between perfective and imperfective aspect (Asp2), which was introduced in (2) and did not show up in the paradigm in (29) and (30). Imperfective and perfective Asp 2 markers are discussed in section  3.5. Fi nally, we look at the modal marker áɁ, which is used for the expression of future- oriented and other modal readings in section 3.6.

3.4 Tense vs. Asp1 In this section, we put forward evidence against a unified analysis of the markers of the fi rst two categories in (31) as markers of tense, and for a distinct categorization as markers of tense and Asp1, respectively. Elements of the two categories show a different behavior in their distribution (section 3.4.1) and in their semantic interpretation (section 3.4.2).

TAM Coding and Temporal Interpretation

27

3.4.1 TENSE VS. ASP1: DISTRIBUTION AND (IN)COMPATIBILITIES.

Syntactically, markers of the first category can co-occur with markers belonging to the second category, as shown in (32a, b). When co-occurring, elements of the first category must always precede those of the second, suggesting that the former occupy a higher structural position. Moreover, markers of both classes cannot co-occur with elements of the same class within a sentence (cf. (33a, b)), suggesting that they belong to the same class. (32)

a. mə I

fə t

cág/zí/yɔög asp1

nɛn go

ntαn. market

‘I went to the market yesterday in the morning/before sunrise/ after noon.’ b. mə I

náɁ t

cág/zí/yɔög asp1

nɛn go

ntαn. market

‘(That day . . . ) I went to the market in the morning/before sunrise/after noon.’ (33)

a. * mə I

náɁ t

fə t

nɛn go

ntαn. market

intended: ‘I went to the market yesterday.’ b. * mə I

yɔög asp1

cág asp1

nɛn go

ntαn. market

intended: ‘I went to the market today in the morning.’ We conclude this subsection by observing that temporal markers of both classes are preferred when the speaker has the required knowledge of the relevant temporal facts (cf. (34)). In this, they resemble TRMs in Gĩkũyũ, which appear to be obligatory. At the same time, they seem to differ from temporal adjuncts in Medumba, which are optional in the presence of suitable contextual information. (34)

a. Context (adapted from Cable 2013): There is a special offer in a local store. If you bought a TV in the past year, you get a DVD player for free. You bought a TV this morning, so you ask for your DVD player:

chapter one

28

Mə I

?(cág)

asp1

ʒ←án yor buy my

télé TV

bin you

fá give

yor my

DVD mə DVD me

‘I bought my TV this morning, give me my DVD player!’ comment: “Without cág it doesn’t work, you have to be precise.” b. [..] You bought a TV yesterday, so you ask for your DVD: Mə ?(fə) ʒ←àn yor télé bin fá yor DVD mə I t buy my TV you give my DVD me ‘I bought my TV yesterday, give me my DVD player!’ 3.4.2 TENSE VS. ASP 1 : SEMANTIC INTERPRETATION. We propose that, semantically, markers of the fi rst category modify the location of RT relative to UT, suggesting that they encode tense in the sense of Klein (1994). This is illustrated for fə in (35), which expresses a precedence relation between Serge’s leaving time (ET) and the speaker’s arrival time (RT), on the one hand, and the speaker’s arrival time (RT) and UT on the other, resulting in a double-shifted interpretation ET < RT < UT. The past-shift of RT relative to UT is expressed by fə in the matrix clause, which can be repeated in the embedded clause. Crucially, the presence of fə in the embedded clause is insufficient for licensing the second past-shift of ET relative to RT, which requires the particle yǎ ‘already’ instead.

(35)

Context (adapted from Cable 2013): Serge has been telling us for a while that he intends to travel to New York. Today, we went to his house to say goodbye, but unbeknownst to us at the time, he had already left yesterday. bəg fə bɨàà kwúm ndá zə á (fə) yǎ tʃǎ we t aux arrive house pron he t already leave ‘When we arrived at his house, he had already left.’

As for the temporal restrictions expressed by náɁ and fə, the following generalizations emerge: The morpheme náɁ restricts RT to any time interval prior to UT that is sufficiently remote, as shown in (36). If the context provides a recent RT, náɁ is infelicitous (37). (36)

Q: What was Nana doing last year/last month/last week (when you visited her)?

TAM Coding and Temporal Interpretation

Nana Nana

náɁ t

29

k@ ná ŋkwún ipfv cook beans

‘Nana was cooking beans.’ (37)

Q: What was Nana doing yesterday/this morning (when you visited her)? # Nana náɁ k@ ná ŋkwún Nana t ipfv cook beans ‘Nana was cooking beans.’

The morpheme fǝ, by contrast, is mainly used in clauses that make temporal reference to the day prior to UT. Still, we do not agree with Kouankem (2012) that its meaning is ‘yesterday’. The temporal reference is more flexible than that, since fǝ also allows for reference to the day of utterance, as evidenced by (38) and the corresponding speaker’s comment: (38)

Q: Alice, you visited Marie today/yesterday. What was she doing? á fə k@ nǎ ŋkwún she t ipfv cook beans ‘She was cooking beans.’ comment: “The meaning of fə is not so strict. It can sometimes also be used when talking about today.”

In light of these facts, we conclude that náɁ and fə denote remote past tense and near past tense, respectively. Moreover, there is a striking contrast between Medumba and Gĩkũyũ in that the remote past marker in Medumba cannot be used if the exact time of reference is unknown (cf. (39)). In such cases, the temporally unmarked form introduced as general past in (29a) is required. 8 This suggests that, unlike in Gĩkũyũ, remote past and near past pick out distinct time intervals in Medumba, and that the distinction between náɁ and fə should be captured as temporal remoteness rather than as specificity. (39)

Context (adapted from Cable 2013): You are visiting your friend Serge, whom you haven’t seen in weeks. There is a brand new TV in his living room. You have no idea when he bought the TV. It could have been several days ago; it could have been yesterday; it could have been today. You report to your spouse:

chapter one

30

# Serge Serge

náɁ t

ʒ←àn buy

zə his

TV TV

nswó! new

‘Serge bought his new TV!’ comment: “That doesn’t work because I don’t know that it was long ago.” The temporal markers of the Asp1 category all restrict the event to a specific time of day: zí locates ET in the very early morning—that is, the time interval between midnight and sunrise, cág locates ET in the time span between sunrise and noon, and yɔög situates it in the afternoon—that is, between noon and sunset. Without additional marking and context, the temporal interpretation of these markers is typically relative to the day including UT, but closer scrutiny shows that the temporal interpretation is not deictic in that it is not necessarily relative to UT. In (40), zí locates ET relative to some past RT, the previous Sunday, as specified by the context and the remote past marker náɁ. (41) shows the same for fə and cág. (40)

Context: It is Tuesday and Alain and Daniel are talking about the impressive dinner that Nana gave on Sunday. Alain asks Daniel how Nana managed to prepare everything on her own. Daniel says: á náɁ zí ná bɔ kwaɁ ŋkɨbndjɨ she t asp 1 cook fufu early morning ‘She had cooked (the) fufu early in the morning.’

(41)

Q: Alice, you visited Mary yesterday, at 8 in the morning. What was she doing? á fə cág k@ nǎ ŋkwún she t asp1 ipfv cook beans ‘She was cooking beans (yesterday morning).’

Since the semantic function of the markers zí, cág, and yɔög consists in modifying ET in its temporal relation to RT, we treat them as a special class of aspectual Asp1 markers. Like the run- of-the-mill aspectual operators discussed in Kratzer (1998) (cf. (2)), Asp1 markers operate semantically at the level of truth conditions. In this, they differ not only from tense morphemes, but also

TAM Coding and Temporal Interpretation

31

from the temporal remoteness morphemes in Gĩkũyũ (cf. Cable 2013). The truth- conditional meaning of Asp1 markers accounts for the empirical observation that zí, cág, and yɔög can occur without temporal adjuncts in answers to explicit inquiries after the temporal location of an event (cf. (42)), which is incompatible with a semantic analysis of presupposed not-at-issue content. (42)

Q: When did Nana cook beans? Nana cág/zí/yɔög ná ŋkwún Nana asp1 cook beans ‘Nana cooked beans this morning/last night/this afternoon.’

By contrast, the morphemes of the first category cannot answer requests for temporal information according to the speakers we consulted (cf. (43)).9 (43)

Context: When did Nana cook beans? #Nana náɁ/fə ná ŋkwún Nana t cook beans intended: ‘Nana cooked beans some time ago/yesterday.’ comment: “This one does not work without an adverbial!”

Based on these observations we tentatively infer that the meaning of the Asp1 markers is truth- conditional, while that of the tense markers is presuppositional. More precisely, we propose that the meaning of the morphemes of the fi rst category are parallel to those of past tense markers found in European languages, except that the Medumba past tenses are more specific in adding a remoteness condition to the temporal presupposition. The Asp1 markers, by contrast, require ET to overlap with a specific subinterval of the day surrounding RT as part of the truth conditions. Tentative lexical entries for tense and Asp1 markers in Medumba are provided in (44) and (45): The tense markers náɁ and fə in (44) denote partial identity functions over time. They presuppose a contextually provided RT that is separated from UT by a longer (44a) and shorter (44b) temporal interval, respectively. Asp1 markers restrict the time property denoted by the progressive or perfective Asp2 phrase by mapping tensed propositions (of type ) onto tensed propositions (45).

chapter one

32

(44)

(45)

3.5

a. [[náɁ]]UT

= λt [t precedes UT by two days or more].t

b. [[fə]]UT

= λt [t precedes UT by one day or less].t

a. [[cág]]

= λP. λt. λw. ∃t’[t’ ∞ morning of the day of t & P(w)(t’) = 1]

b. [[yɔög]]

= λP. λt..λw. ∃t’[t’ ∞ afternoon of the day of t & P(w)(t’) = 1]

c. [[zí]]

= λP. λt. λw. ∃t’[t’ ∞ early morning of the day of t & P(w)(t’) = 1]

Aspect 2 : Imperfective and Perfective

In addition to the novel class of Asp1 markers, the TAM system of Medumba also expresses the more familiar distinction between imperfective and perfective introduced in (2). The imperfective marker k@, which also comes in the variants ŋg@ or a nasal prefi x, is compatible with tense (46) and Asp1 (47) marking, suggesting that it does not belong to either paradigm.10 (46)

á she

fə/náɁ t

k@ ipfv

naâ cook

ŋkwún beans

‘She was cooking beans (yesterday/some time ago).’ (47)

á she

cág/zí/yɔög asp1

k@ ipfv

nǎ cook

ŋkwún beans

‘She was cooking beans (this morning/last night/this afternoon).’ The imperfective marker k@ is preferably used in past and future contexts, but it is also compatible with present interpretations. (48)

Nana Nana

náɁ/ t/

áɁ / fut/

⌀ ⌀

k@ ipfv

ná cook

‘Nana was/will be/is cooking beans.’

ŋkwún beans

TAM Coding and Temporal Interpretation

33

Interestingly, the presence of an imperfective marker, including k@, seems to be required for a present tense interpretation of event predicates, as illustrated for an activity predicate in (49). Without the imperfective marker, the sentence typically gets an immediate past interpretation. Cognitive state predicates without imperfective marking, by contrast, get default present interpretations, as illustrated in (50). (49)

Context: What are the farmers doing? ?? (k@) bú ndjɨ nà they ipfv dig farm ‘They are plowing the farmland.’ comment: “Without k@ it would mean They have (just) plowed the farmland.”

(50)

Context: What does Carine think about Yves? à b!ní she hate.him ‘She hates him.’

In order to account for the obligatory past interpretations of eventive predicates in sentences without overt temporal or aspectual marking, we propose that Medumba has a covert perfective Asp2 marker, which contrasts with the overtly marked imperfective.11 Moreover, the temporal interpretation of temporally unmarked sentences (STSs) is pragmatically resolved in exactly the same way as has been proposed for Hausa STSs in section 2.4. In the eventive sentence in (49), the eventuality is interpreted as temporally bounded because it contains a covert perfective Asp2marker, and the BEC in (19) prevents the sentence from being interpreted as present tense.12 In the presence of overt imperfective aspect marking, ET is presented as unbounded (including RT), and the interaction of DP and SPI results in the default present interpretation of imperfective sentences. We conclude that in spite of the fact that Medumba differs from Hausa in overtly coding (past) tense distinctions, the temporal interpretation of sentences without overt tense marking relies on precisely the same interaction between aspectual coding and pragmatic reasoning in both languages. Summing up so far, we have identified the following properties of the TAM system of Medumba. Medumba does not have eight different past

chapter one

34

tenses, but at most two genuine tense markers, náɁ and fə, which distinguish between remote and near past. Second, the temporal markers that have been included in the same paradigm in previous accounts (cág, zí, yɔög) constitute a separate category of Asp1 markers, which locate ET at a specific time of day including RT. Third, in addition to tense and Asp1, Medumba has a category of Asp2 markers including an imperfective (k@) and a covert perfective marker. It follows that the syntactic structure of Medumba in (51) involves two aspectual layers. (51)

[TP T . . . [AspP1 Asp1 . . . [AspP2 Asp2 . . . [ vP]]]]

Finally, the temporal interpretation of temporally unmarked sentences relies on the same pragmatic principles that were argued are also at work in Hausa. On closer inspection, then, we found substantial parallels in the temporal interpretive systems of tenseless Hausa and graded-tense Medumba, despite all ostensible differences on the surface.

3.6

The Future Modal

Another important parallel in the TAM systems of Hausa and Medumba is that future- oriented readings in Medumba, too, are expressed by means of a modal expression. Like its Hausa counterpart zaa, the future marker áɁ is compatible with relative future would-readings (cf. (52)), which shows that áɁ does not encode deictic future time reference and is therefore not a marker of tense in the sense of Klein (1994).13 (52)

mə náɁ lǔ kɔ 2011 mbə mə áɁ nɛn cameroun ǎ 2012 I t take spear 2011 that I fut go Cameroon prep 2012 ‘In 2011, I took the decision that I would go to Cameroon in 2012.’

Moreover, áɁ is spontaneously used by speakers in order to express modal meanings other than those associated with simple future time reference—that is, bouletic or inertial necessity. In particular, áɁ can express future- oriented deontic necessity, cf. (53), and, strikingly, presentoriented epistemic necessity as well (cf. (54)). From these observations we tentatively conclude that the core meaning of áɁ consists simply in universal quantification over possible worlds, such that its meaning does not entail obligatory future shift, as illustrated in (55).14

TAM Coding and Temporal Interpretation

(53)

35

Context: Your sister is coming to your place and says that she would like to play with your children. You do not like the idea very much because it is quite late, so you say: bú áɁ zí they fut sleep ‘They have to sleep.’

(54)

Context: You want to visit your friend Elodie. When you arrive at her house you see that there is no light in her room, so you say: (á) áɁ bt zə Elodie nzi it fut cop that Elodie asp 2 .sleep ‘Elodie must be sleeping.’

(55)

[[áɁ]] = λp.λt.λw. ∀w’ [w’∈MAXO(w,t) (MB(w)(t)) → p(w’)]

As shown in section  3.3, áɁ frequently combines with Asp1 markers in order to specify a time of day in the future.15 Again, the ET location is specified relative to RT and independent of UT, whenever RT and UT are distinct. Since áɁ expresses quantification over possible worlds, and cág independently specifies the time of day at which the event occurred relative to the day containing RT, our analysis correctly predicts (56) to be felicitous on the interpretation given.16 If áɁ cág were instead a complex deictic future tense marker meaning ‘tomorrow’, as suggested in earlier descriptions, its meaning should clash with that of the temporal adverb ŋgab zə á səɁ lə ‘next week’. (56)

ŋgab week

zə that

á it

səɁ come

lə Marie def Marie

áɁ cág ná fut asp1 cook

ŋkwun beans

‘Next week, Marie will cook beans in the morning.’

3.7

Cross- linguistic Variation in Future Marking

We conclude this section with a few remarks on the observable crosslinguistic variation in future marking. The attentive reader will have noticed the following differences between the meaning of the Medumba future morpheme áɁ and that of its Hausa counterpart zaa, which was discussed in section 2.5. First, the modal marker áɁ encodes only quantification over

36

chapter one

possible worlds, whereas zaa also encodes existential closure of the event argument. Second, the meaning of áɁ is compatible with (at least) bouletic, inertial, deontic, and epistemic modality, whereas the interpretation of Hausa zaa seems to be restricted to bouletic or inertial modality—that is, to prediction or intention readings, in our analysis. And third, the meaning of áɁ does not entail a future shift but is compatible with present epistemic modality, whereas Hausa zaa obligatorily co-occurs with a prospective aspect marker forcing a future shift. While the fi rst difference appears to reflect a lexical idiosyncrasy of zaa, which we proposed in order to account for the mandatory co-occurrence of future and prospective aspect markers in Hausa, the other two differences exemplify the two major sources of variation in the meaning of future markers in natural language that were identified in Tonhauser (2011b). Cross-linguistically, there is one class of future markers that exhibit a future shift on all their occurrences and are mainly associated with predictive and intentional interpretations. This class includes the St’át’imcets future morpheme kelh (Matthewson 2006; Rullmann, Matthewson, and Davis 2008), the Guaraní future marker – ta (Tonhauser 2011b), and the Hausa future marker zaa. The other class includes the future morphemes of better-studied languages such as English will (Enç 1996; Copley 2009), German werden (Vater 1975; Zifonun et al., 1997) and Turkish –(y)EcEk (Yavaş 1982), as well as the future forms of Greek and Italian (Giannakidou and Mari 2014 and this volume). These markers behave more like the Medumba future morpheme áɁ in also allowing for non-future epistemic readings. Quite possibly, there is a causal connection between entailed future orientation and the possible modal flavors of future markers in a given language. As pointed out to us by Anastasia Giannakidou (pers. comm.), the modal analysis we adopt for the Hausa future marker zaa cannot be extended to Medumba in order to account for the epistemic readings of áɁ. Giannakidou and Mari (2014 and this volume) analyze the predictive and non-future epistemic readings of the Italian and Greek future markers by assuming epistemic ordering of metaphysical and epistemic modal bases, respectively. If this is the correct approach to less restricted future markers, such as Medumba áɁ, we could extend the analysis to the more restricted class of future markers by assuming that these always select a metaphysical modal base. At the same time, it has been proposed that the future orientation in modals always comes from the presence of a covert or overt prospective aspect marker; cf. Matthewson (2012, 2013) and Kratzer (2012b). Applying this view to future modals would make the assumption of metaphysical modality un-

TAM Coding and Temporal Interpretation

37

necessary, as argued, for instance, by Kratzer (2012b). Combining this idea with the analysis of future morphemes proposed by Giannakidou and Mari (this volume) would then allow us to reduce the meaning of future markers to epistemic modality entirely, a line of thought we will leave for further research.

3.8

Summary

The TAM system of Medumba comprises three distinct categories, a future modal áɁ, presuppositional tense markers restricting RT to near past (fə) or remote past (náɁ) relative to UT, and aspectual time of day markers (yɔög, cág, zí). These categories are supplemented by an imperfective aspect marker (k@), which contrasts with a phonologically covert perfective aspect marker. Despite the considerable complexity found in the TAM system of Medumba, some interesting parallels with more parsimonious temporal systems like that of Hausa were found, especially in the interpretation of future markers and of temporally unmarked sentences. Finally, we observed some interesting differences in the interpretive range of the Hausa modal marker zaa, which selects for a prospective marker and is hence restricted to future- oriented interpretations, and the less restricted modal marker áɁ in Medumba, which also allows for simultaneous epistemic interpretations.

4

Conclusions

This chapter investigates the temporal interpretation and the inventory of TAM categories in two typologically unrelated languages, the tenseless language Hausa (Chadic) and the graded-tense language Medumba (Grassfields Bantu). We have argued that tense distinctions are not grammatically marked in Hausa, and that the temporal interpretation of Hausa tenseless sentences is conditioned by aspectual and modal markers, the lexical Aktionsart of the verb, and, crucially, the pragmatic principles for tense resolution put forward by Smith, Perkins, and Fernald (2007). The graded-tense language Medumba, by contrast, was shown to have (past) tense markers in its inventory of TAM markers, in addition to aspectual and modal markers. However, we could only identify two past tense markers, namely near and remote past, instead of the up to eight past distinctions proposed in the earlier literature. The graded-tense

38

chapter one

flavor comes about through the complex interaction of the past markers and the modal future marker with a new category of aspectual time- ofday markers, which situate the event time within a certain part of the day containing the reference time. Crucially, these aspect markers differ from the temporal remoteness morphemes in another graded-tense language, Gĩkũyũ, as discussed by Cable (2013). Fi nally, it was shown that Smith, Perkins, and Fernald’s (2007) pragmatic principles are also at work in the temporal interpretation of Medumba sentences without tense marking. More generally, our investigation has implications for the crosslinguistic study of temporal interpretation and TAM inventories. First, the existence of the linguistic categories of (future) modal and aspect in a language does not depend on the presence of tense markers, as modal and aspectual markers are found in tenseless Hausa and graded-tense Medumba alike. Second, future-oriented readings are expressed by modal markers in both languages, supporting claims to the effect that future readings are of a modal nature cross-linguistically. At the same time, the future markers in the two languages are shown to differ in that the future marker in Medumba does not obligatorily express a future orientation. Third, the pragmatic principles put forward by Smith, Perkins, and Fernald (2007) are active not only in the temporal interpretation of tenseless languages, but also in a graded-tense language, pointing to their possibly universal nature. Finally, and most important, the differences between the individual TAM markers in Hausa and Medumba could not have been identified without careful formal-semantic analysis, which made crucial use of Matthewson’s (2004; 2011) methodological guidelines for formal semantic fieldwork. It is to be hoped that future research on tenseless and (graded) tensed African languages will shed yet more light on universals and variation in the temporal interpretation and the TAM inventory of natural languages.

Notes 1. The meanings in (2a, b) are modeled on the entries for imperfective and perfective aspect in Kratzer (1998). 2. Moreover, St’át’imcets STSs can refer to temporal intervals, including events before and during UT, as shown in (i). This observation provides evidence for the non-future analysis in (9) and against an analysis involving covert past and present morphology.

TAM Coding and Temporal Interpretation (i)

39

Context (abridged): Your white friends Theresa, Charlie and Marie got drunk at the bar. Theresa threw up at 10 pm, Charlie is in the process of throwing up. wat’k’ kw s-Theresa múta7 s- Charlie, vomit DET NOM-Theresa and NOM- Charlie (Matthewson 2006, p. 681f.) ‘Theresa and Charlie threw up/ are throwing up.’

3. When it comes to simultaneous past and present interpretations that were reported to exist in St’át’imcets in Matthewson (2006), see fn. 2, the data are not entirely clear in Hausa. Still, example (ii) below seems to show that Hausa behaves like St’át’imcets in allowing for simultaneous past and present interpretations when an activity predicate is marked for progressive aspect. Providing this example as evidence for the lack of covert past and present tense morphology in Hausa, we refer the reader to Mucha (2013) for more detailed discussion. (ii)

Context: John and Peter only have one guitar so they have to take turns playing. John practiced for an hour and then gave the guitar to Peter, who is now playing. You say: Su-nàa wàasaa (yâu). 3pl-prog play today ‘They are/were playing today.’

4. The effects of the BEC are also visible in languages that do overtly encode aspect and tense specifications. A case in question is Greek, in which perfective non-past sentences cannot refer to the utterance time. Instead, they invariably locate the described event in the posterior interval following the utterance time; see Giannakidou (2009; 2014) for relevant discussion. 5. As pointed out by Anastasia Giannakidou (pers. comm.), the deficiency of the prospective aspect is reminiscent of Abusch’s (1994) substitution operator and Giannakidou’s (2009, 2014) non- past, which she proposes as the meaning for the Greek subjunctive inflection. The non-past operator is semantically dependent on an embedding expression, such as the Greek future marker tha, which is reanalyzed as a temporal expression assigning the specific value n(ow) (= utterance time) to the open tense variable. This similarity is only apparent, though, as the Hausa prospective differs from the Greek subjunctive/non-past operator in an impor tant way: Unlike the combination of Greek future tha + subjunctive, which is tense-like in that it makes deictic reference to the utterance time, the combination of zaa + prospective in Hausa is aspect-like in also allowing for a relative future shift relative to an RT in the past or in the future; see examples (17) and (18). If one were to extend Giannakidou’s (2009, 2014) analysis to Hausa, one would have to assume that Hausa zaa can provide dif ferent values (past, present, future) for the t-variable of the substitution operator. Since equivalent

40

chapter one

relative temporal shifts are also observed in the absence of zaa, such a reanalysis would appear to be superfluous and unwarranted on empirical grounds. 6. Our consultants judged this form as marginal, and we will therefore ignore it in what follows. However, those speakers who at least recognize it as a possible form confi rm Kouankem’s (2012) description of an immediate future marker. 7. According to our consultants, there is no clearly identifiable difference between náɁ and lû, on the one hand, and féè and lú, on the other. It is probably due to this redundancy that lú and lû have the status of marginal forms: they are rarely volunteered, and sometimes they were not even recognized by our speakers. For this reason, we concentrate on the preferred forms náɁ and féè. 8. The past interpretation of formally unmarked sentences is accounted for in section 3.5, where it is argued that such unmarked sentences contain a covert perfective Asp2 marker ⌀PFV. 9. In later elicitations, other speakers accepted discourses of this kind, suggesting that these data are subject to variation (see Mucha 2015). We also tested for visibility to negation in order to bring out a potential at issue/non-at-issue contrast between the markers of the two classes. However, the relevant tests did not yield unequivocal results, which might be due to independent factors such as grammatical restriction on morphological focus marking in Medumba. 10. The morpheme cwɛn seems to have a similar meaning. However, since earlier studies disagree on whether it denotes present tense or progressive aspect, and since most of our consultants considered it marginal anyway, we set this marker aside for future study. 11. Covert marking of perfective aspect in opposition to overtly marked imperfective or progressive aspect is a well-attested pattern cross-linguistically and seems to be a general tendency in natural languages (Comrie 1976); see, e.g., Matthewson (2006) for St’át’imcets (Salishian), Hartmann and Zimmermann (2009) for Gùrùntùm (Chadic), and Hyman, et al. (2002) for Leggbo/Igbo (Niger- Congo). 12. Arguably, the BEC is blocked from applying to stative predicates like hate in (55) since perfective aspect with states requires ET to overlap with RT, rather than to be included in it (Kamp and Reyle 1993). 13. In order for this reading to arise, the future marker must occur in a clause embedded in a past tensed matrix clause. We leave it open whether the incompatibility of future and past markers within the same clause follows for structural reasons, such as competition for the same structural slot, or whether it is due to semantic reasons, such as confl icting presuppositions of the two markers, delegating the issue to future research. 14. Regarding the question of how the ET-shift in future- oriented interpretations of sentences with áɁ comes about if it is not coded in the meaning of the modal, there are two possibilities. The ET-shift could be the result of a pragmatic process of aspectual coercion when the context makes clear that a bouletic or inertial ordering source is required. Alternatively, the ET-shift could be due

TAM Coding and Temporal Interpretation

41

to the presence of a grammatically optional covert prospective marker, as has been proposed for English (Kratzer 2012b; see also Matthewson 2012). Again, we have to leave this question open. 15. A possible exception is the combination of áɁ and zí, which seems to lexicalize a remote future interpretation. According to our consultants, it means ‘some indefi nite time from tomorrow on’. The crucial factor governing its felicity, though, is that there be a night in between, which still relates it to the ‘night/early morning’ meaning of zí. 16. If the context provides no specific day within the RT interval of ‘next week’, (56) allows for a second interpretation on which Marie will cook beans on the morning of every day of the coming week. We propose that this reading is due to covert distributive quantification over the individual days of the week, as shown in the semantic representation in (iii). A parallel ambiguity can be observed for the English counterpart in (iv):

5

(iii)

∀w’[w’∈MAXO(w,UT)(MB(w)(UT))] → [∀t [day(t) & t ⊂ next_week] → ∃t’[t’ ∞ morning of t & Mary cooks beans at t’ in w’]]

(iv)

Next week, Mary will cook beans early in the morning.

References

Aboh, E. O. 2004. The Morphosyntax of Complement- Head Sequences: Clause Structure and Word Patterns in Kwa. Oxford, U.K.: Oxford University Press. Abusch, D. 2004. On the temporal composition of infi nitives. In The Syntax of Time, ed. J. Gueron and J. Lecarme. Cambridge, MA: MIT Press. Bohnemeyer, J. 2002. The Grammar of Time Reference in Yukatek Maya. Munich: Lincom. ———. 2009. Temporal anaphora in a tenseless language. In The Expressions of Time, ed. W. Klein and P. Li, 83–128. Berlin: Mouton de Gruyter. Cable,  S. 2013. Beyond the past, present and future: Towards the semantics of ‘Graded Tense’ in Gĩkũyũ. Natural Language Semantics 21(3): 219–276. Comrie, B. 1976. Aspect. Cambridge: Cambridge University Press. ———. 1985. Tense. Cambridge: Cambridge University Press. Condoravdi, C. 2002. Temporal interpretation of modals: Modals for the present and the past. In The Construction of Meaning, ed. D. Beaver, S. Kaufmann, B. Clark, and L. Casillas, 59–88. Stanford, CA: CSLI Publications. Copley, B. 2009. The Semantics of the Future. New York: Routledge. Enç, M. 1996. Tense and modality. In Handbook of Semantic Theory, ed. S. Lappin, 345–358. Oxford: Blackwell. Giannakidou, A. 2009. The dependency of the subjunctive revisited: Temporal semantics and polarity. Lingua 119: 1883–1908.

42

chapter one

———. 2014. The futurity of the present and the modality of the future: A commentary on Broekhuis and Verkuyl. Natural Language and Linguistic Theory 32: 1011–1032. Giannakidou, A., and A. Mari. 2014. The future in Greek and Italian: Truth conditional and evaluative dimensions. Manuscript. University of Chicago and Institut Jean Nicod, CNRS (submitted). Hacquard, V. 2009. On the interaction of aspect and modal auxiliaries. Linguistics and Philosophy 32(3): 279–315. Hartmann,  K., and  M. Zimmermann. 2009. Morphological focus marking in Gùrùntùm (West Chadic). Lingua 119: 1340–1365. Heim,  I. 1991. Artikel und Defi nitheit. In Semantics: An International Handbook of Contemporary Research, ed.  A. von Stechow and  D. Wunderlich, 487–535. Berlin: Mouton de Gruyter. ———. 1994. Comments on Abusch’s theory of tense. In Ellipsis, Tense and Questions, ed. H. Kamp, 143–170. Dyana 2: University of Amsterdam. Hoffmann, T. R. 1966. Past tense replacement and the English modal auxiliary system. In Mathematical Linguistics and Automatic Translation, ed. A. Oettinger. Report NSF-17, Harvard University Computation Laboratory. Hyman,  L.,  H. Narrog,  M. Paster, and  I. Udoh. 2002. Leggbó verb inflection: A semantic and phonological particle analysis. Proceedings of the 28th Annual Berkeley Linguistic Society Meeting, 399–410. University of California, Berkeley. Jaggar, P. J. 2001. Hausa. Amsterdam: John Benjamins. Jakobson, R. 1959. Boas’ view of grammatical meaning. American Anthropologist 61(5): 170–178. Jóhannsdóttir,  K., and  L. Matthewson. 2008. Zero-marked tense: The case of Gitxsan. Proceedings of NELS 37, 299–310. Amherst, MA: GLSA. Kamp, H., and U. Reyle. 1993. From Discourse to Logic. Dordrecht: Kluwer. Klein, W. 1994. Time in Language. New York, London: Routledge. Kouankem, C. 2012. The syntax of the Medumba determiner phrase. PhD diss., Université de Yaounde I. Kratzer, A. 1998. More structural analogies between pronouns and tenses. In The Proceedings of Semantics and Linguistic Theory (SALT) 8, ed. D. Strolovitch and A. Lawson, 92–110. Cornell University. Ithaca, NY: CLC Publications. ———. 2012a. Modals and Conditionals. Oxford: Oxford University Press. ———. 2012b. What “can” can mean. Lecture notes, LOT winter school, Tilburg, Netherlands. Matthewson, L. 2004. On the methodology of semantic fieldwork. International Journal of American Linguistics 70(4): 369–415. ———. 2006. Temporal semantics in a superficially tenseless language. Linguistics and Philosophy 29(6): 673–713.

TAM Coding and Temporal Interpretation

43

———. 2011. Methods in crosslinguistic formal semantics. In Semantics: An International Handbook of Natural Language Meaning. Vol. 1, ed. C. Maienborn, K. von Heusinger, and P. Portner, 268–284. Berlin: Mouton de Gruyter. ———. 2012. On the (non-)future orientation of modals. In Proceedings of Sinn und Bedeutung 16, ed. A. Aguilar Guevara, A. Chernilovskaya, and R. Nouwen, 431–446. Cambridge, MA: MIT Working Papers in Linguistics. ———. 2013. Gitksan modals. International Journal of American Linguistics 79(3): 349–394. Mucha, A. 2012. Temporal reference in a genuinely tenseless language: The case of Hausa. In Proceedings of Semantics and Linguistic Theory (SALT) 22, ed. A. Chereches, 188–207. Cornell University. Ithaca, NY: CLC Publications. ———. 2013. Temporal interpretation in Hausa. Linguistics and Philosophy 36(5): 371–415. ———. 2015. Temporal interpretation and cross-linguistic variation. PhD diss., Universität Potsdam (submitted). Newman, P. 2000. The Hausa Language. An Encyclopedic Reference Grammar. New Haven, CT: Yale University Press. Nganmou,  A. 1991. Modalités verbales: Temps, aspect et Mode en Medumba, PhD diss., Université de Yaounde I. Partee, B. 1973. Some structural analogies between tenses and pronouns in English. Journal of Philosophy 70(18): 601–609. Reichenbach, H. 1947. Elements of Symbolic Logic. New York: Macmillan. Ritter,  E., and  M. Wiltschko. 2004. The lack of tense as a syntactic category: Evidence from Blackfoot and Halkomelem. In Papers for the 39th International Conference on Salish and Neighbouring Languages, ed.  J.  C. Brown and T.  Peterson, 341–370. Working Papers in Linguistics 14, University of British Columbia. Rullmann, H., L. Matthewson, and H. Davis. 2008. Modals as distributive indefinites. Natural Language Semantics 16(4): 317–357. Schuh, R. G. 2003. The functional unity of the Hausa and West Chadic Subjunctive. UCLA Working Papers in Linguistics 9, p. 17–42. Smith, C. S. and M. S. Erbaugh 2005. Temporal interpretation in Mandarin Chinese. Linguistics 43(4): 713–756. Smith, C. S., E. T. Perkins, and T. B. Fernald. 2003. Temporal interpretation in Navajo. In Proceedings of SULA 2, 175–192. Amherst, MA: Amherst Working Papers in Linguistics. ———. 2007. Time in Navajo: Direct and indirect interpretation. International Journal of American Linguistics 73(1): 40–71. Thomason, R. 1984. Combination of tense and modality. In Handbook of Philosophical Logic: Extensions of Classical Logic, ed. D. Gabbay and F. Guenther, 135–165. Dordrecht: Reidel.

44

chapter one

Tonhauser, J. 2011a. Temporal reference in Paragyuan Guaraní. Linguistics and Philosophy 34(3): 257–303. ———. 2011b. The Paraguayan Guaraní future marker – ta: Formal semantics and crosslinguistic comparison. In Tense across Languages, ed. R. Musan and M. Rathert, 207–232. Tübingen: Niemeyer. Vater, H. 1975. Werden als Modalverb. In Aspekte der Modalität, ed. J. P. Calbert and H. Vater, 71–148. Tübingen: Narr. von Fintel, K., and L. Matthewson. 2008. Universals in semantics. The Linguistic Review 25(1–2): 139–201. Yavaş, F. 1982. Future reference in Turkish. Linguistics 20: 411–429. Zifonun,  G.,  L. Hoffmann,  B. Strecker, and  J. Ballweg.  1997. Grammatik der deutschen Sprache. Berlin: Walter de Gruyter.

chapter two

Modals: Meaning Categories? Valentine Hacquard

1

Introduction

I

n many languages, the same modal word can be used to express various flavors of modality (Bybee et  al. 1994; Palmer 2001; among others).1 English must, for instance, can express various kinds of necessities (epistemic, deontic, or teleological), as illustrated in (1). (1)

a. John is not in his office. He must be home. b. John parked illegally. He must pay a fi ne. c. John wants to get a PhD. He must write a thesis.

epistemic deontic teleological

Given the pervasiveness of this multiplicity of meanings across unrelated languages, a Kratzerian semantics—in which modals are not multiply ambiguous, but where they come in single lexical entries, with flavors determined by context—seems particularly appealing. Let us briefly review this system. Kratzer (1981, 1991) proposes that modals are quantifiers over possible worlds, whose restriction is contextually provided via ‘conversational backgrounds,’ responsible for the various flavors. Specifically, modals combine with a modal base f, which determines an initial set of worlds, and an ordering source g, which provides an ordering given certain ideals (such as deontic and bouletic). Modals quantify over some (possibility modals) or all (necessity modals) of the ‘best’ worlds (best in terms of the ordering source) among the worlds provided by the modal base:2

chapter two

46

(2)

a. [[must]]w = λf< s,st > λg< s,st > λq. ∀w’∈ Bestg(w)(f(w)): q(w’) = 1 b. [[can]]w

= λf< s,st > λg< s,st> λq. ∃w’∈ Bestg(w) f(w): q(w’) = 1

Kratzer identifies two broad classes of modals: epistemic modals (which take an epistemic modal base), and root modals (which take a circumstantial modal base). Root modals are further differentiated by combining with various ordering sources (deontic, bouletic, teleological). From this point on, I will ignore the ordering source and consider root modals as a group. (3)

a. fepistemic(w) = λw’. w’ is compatible with what is known in w b. fcircumstantial(w) = λw’. w’ is compatible with the circumstances in w

The same modal must is involved in (1a)–(1c), but it combines with different conversational backgrounds to yield different flavors. (4)

a. (John is not in his office.) He mustf-epis be home. epistemic In all worlds w’ compatible with what is known in w (e.g., the fact that he’s not in his offi ce), John is home in w’. b. (John parked illegally). He mustf-circ,g-deon pay a fi ne. deontic In all most ideal worlds w’, given the law in w, among those compatible with the circumstances in w (e.g., the fact that he parked illegally), John pays a fi ne in w’. c. John wants to get a PhD. He mustf-circ,g-boul teleological write a thesis. In all most ideal worlds w’ in which John gets a PhD, among those compatible with the circumstances in w, John writes a thesis in w’.

Beyond a difference in modal bases, epistemics and roots pattern differently in their interaction with tense and aspect. 3 First, modal flavors differ in temporal orientation. The time of evaluation of a root modal has to be the time of the embedded event—that is, the time provided by tense. I use the semi-modal have to, which can carry tense morphology, to illustrate the point.

Modals

(5)

47

John wanted to get a PhD . . . He had to write a thesis. Given John’s situation and goals then, it was necessary that he write a thesis then. *Given John’s current situation and goals, it is necessary that he wrote a thesis previously.

When have to receives an epistemic meaning, however, its time of evaluation cannot be back-shifted: in matrix contexts, the modal’s time of evaluation has to be the speech time.4 The past tense of have to is interpreted in the modal’s prejacent: (6)

(??Two days ago), John had to be in his office at the time of the crime. Given what we know now, it is necessary that he was in his office then. *Given what we knew then, it was necessary that he was in his office then.

Consider the following scenario from Hacquard (2010): Last week, John had a solid alibi. Mary said that they were meeting in his office at the time of the crime. But yesterday several eyewitnesses claimed they saw Mary at the library at the time of the crime, destroying John’s alibi. Here (6) is false: it cannot refer to the time when John’s alibi was solid (e.g., two days ago). Modal flavors also differ in their interactions with aspect. With perfective, a root modal triggers an ‘actuality entailment’ (Bhatt 1999), an uncancelable inference that the complement was actualized. This can be seen in languages that have an overt morphological distinction between perfective and imperfective aspect, such as French: (7)

a. Jean pouvait s’enfuir, mais il est resté. Jean couldroot-ipfv escape, but he stayed. b. Jean a pu s’enfuir, #mais il est resté. Jean couldroot- pfv escape, but he stayed. Jean was able to escape, (#)but he stayed.

The same modals with an epistemic interpretation are, however, immune to this effect; they do not force the realization of their complement with perfective aspect (Hacquard 2006, 2009):

48

(8)

chapter two

(A ce que je sache) Jean a (bien) pu s’enfuir, mais il est possible qu’il ne l’ait pas fait. (As far as I know) Jean couldepis - pfv escape, but it’s possible he didn’t. (As far as I know) Jean may (well) have escaped, but it’s possible he didn’t.

To sum up, the same words can be used to express root and epistemic modality in many different languages, suggesting that epistemic must and deontic must are one and the same. Yet we need to explain why they systematically differ in their interactions with tense and aspect, as a function of the flavor they express. Is there something about the meaning of each flavor that can explain these interactions? If the link between modal flavor and its interactions with tense and aspect were principled, we could make sense of the fact that we never fi nd the opposite patterns cross-linguistically, diachronically, or in language acquisition. If there were something inherent about the meaning of, say, epistemic, as opposed to root, modality, to make it outscope tense, we would expect it to do so across languages. Children would never hypothesize epistemic meanings for modals that appear in the scope of tense, and the pattern would repeat itself among speakers and hold up diachronically. What kind of inherent connections could we fi nd between modal meanings and tense and aspect? Could it be, for instance, that by virtue of being epistemic, a modal somehow expresses atemporal possibilities and necessities, 5 and thus cannot have a back- or forward-shifted evaluation time? Or could it be that by virtue of having a root meaning a modal is susceptible to actuality entailments? If circumstantial modality expresses potentials for events to happen, constraining such potentials to a very short time interval via the perfective might trigger a strong implication that the event must have been realized (see Enfeld 2003; Mari and Martin 2007; and Homer 2010 for such a take on actuality entailments with ability modals). There are reasons to believe that these interactions cannot be purely meaning-based. The modals we have been examining so far are all (semi-)auxiliaries. However, the meanings these modals express can also be expressed by verbs or adjectives, which do not seem to obey the same constraints. Adjectives (possible, likely) and verbs (seem, know, think) express epistemic meanings, and yet are easily interpreted in the scope of a past tense:

Modals

(9)

49

a. It seemed to John that Mary was home. b. John thought/knew that Mary was home. c. It was possible/likely that Mary was home.

Nouns (like means or permission) and verbs (like want) express root meanings, and yet they do not yield actuality entailments with perfective: (10)

a. Jean a voulu s’enfuir, mais il n’est pas parti. Jean wanted- pfv to escape, but he didn’t leave. b. Jean a eu les moyens de s’enfuir, mais il n’est pas parti. Jean had- pfv the means to escape, but he didn’t leave. c. Jean a eu la permission de partir, mais il n’est pas parti. Jean had- pfv permission to leave, but he didn’t leave.

It thus appears that the constraints on modal flavors and their interactions with tense and aspect are limited to ‘grammatical modality’ (i.e., expressed by grammatical modal words), but not ‘lexical modality’ (i.e.,  expressed by lexical words such as adjectives or verbs; Traugott 2011). Adjectives and verbs that express similar meanings do not show the same patterns. Why should this be? Why should modals behave the way they do in the fi rst place? And what differentiates grammatical modals from other elements with similar meanings? These are the questions that this chapter addresses. In section  2, I review the proposal I made in Hacquard (2006, 2010) about (grammatical) modals and their interactions with tense and aspect. Section  3 asks how robust the constraints between modal flavor and tense and aspect are, and discusses counterexamples. Section  4 takes a closer look at verbs and adjectives on the one hand, and modals fully specified for flavor on the other, and discusses what might differentiate ‘grammatical’ from ‘lexical’ modality.

2

Event-Relativity of Modality

In Hacquard (2006, 2010), I proposed an account of modals with the goal of maintaining a Kratzerian semantics where modals come in single lexical

chapter two

50

entries, while providing scope-based constraints on the interpretations it can receive in order to derive the right patterns with tense and aspect.

2.1 Epistemics and Roots Differ in Structural Position The fi rst step in the account is to reduce the differing interactions of root and epistemic modals with tense and aspect to scope, with epistemics scoping above tense and aspect, and roots below, in line with a simplified Cinquean (Cinque 1999) organization of functional projections: (11)

Cinque’s Universal Hierarchy of Functional Projections . . . MODepis > T > Aspect > MODroot > VP

Assuming that epistemics scope above tense explains why their time of evaluation cannot be shifted (Groenendijk and Stockhof 1975; Picallo 1990; Stowell 2004; Hacquard 2006; Borgonovo and Cummins 2007; Demirdache and Uribe-Etxebarria 2008; Laca 2008; among others): they must be interpreted relative to the local time of evaluation. Similarly, assuming that root modals scope below tense explains why their time of evaluation has to be the time provided by tense. In Hacquard (2006, 2009), I proposed a way to explain why root, but not epistemic modals yield actuality entailments also in terms of scope. I argued that actuality entailments result from having perfective aspect take scope over a modal, which happens with roots, but crucially not with epistemics. Aspects are quantifiers over events that relate the running time of the VP event to a time of reference (provided by tense). Perfective aspect existentially quantifies over the VP event and places its running time within a reference time interval (cf. Kratzer’s 1998 formalization of Klein 1994), and, I argue, in a world of evaluation. (12)

[[Mary took-pfv the train]]w = true iff ∃e[τ(e) < t* & e in w & Mary took the train(e)]

When perfective takes scope under a modal (as happens with epistemics), its world of evaluation is provided by the modal, locating the event in the modal worlds. However, when perfective takes scope over the modal (as happens with roots), its world of evaluation has to be the matrix one, thereby forcing the event to occur in the actual world. This is sketched in (13) and (14):

Modals

(13)

a. Mary a pu prendre le train. ‘Mary may have taken the train.’

51

epistemic > perfective

b. [ModP can [TP past [AspP pfv 1 [VP Mary take the train e1] ] ] ] c. ‘There is a world w compatible with what is known in the actual world, such that there is a past event in w which is a train- taking by Mary.’ (14)

a. Mary a pu prendre le train. ‘Mary was able to take the train.’

perfective > root

b. [TP past [AspP pfv 1 [ModP can [VP Mary take the train e1 ] ] ] ] c. ‘There is a past event e1 in the actual world, which in some world compatible with the circumstances in the actual world is a train- taking by Mary.’ Example (14) is true if there is an actual event that in some circumstantial world is a train-taking by Mary. With a further default assumption that the event keeps its description across worlds, we obtain that this actual event is a train-taking by Mary. Following Bhatt (1999), I take the lack of actuality entailments with imperfective to be due to an additional layer of modality associated with the imperfective itself. In sum, I argue that epistemics and roots differ in actuality entailments based on the way they interact with aspect. What shields epistemics from actuality entailments is the fact that, unlike roots, they are interpreted above aspect. Assuming that modals’ peculiar interactions with tense and aspect can be derived in terms of scope, the puzzle to be explained is why modals with epistemic meanings scope above, but modals with root meanings scope below tense and aspect. In other words, why should Cinque’s hierarchy hold? Is it because epistemics and roots occupy dedicated positions by fi at (by having separate lexical entries that specify both position and flavor), or can we get Kratzerian modals that are unspecified for flavor to occupy different positions based on meaning?

chapter two

52

a. It rained. b. TP t 3 3
Past Asp




3 VP t
3

Perf λe2 e2 V rain Ti

Figure 2.1: Aspect Movement

2.2 Event-Relativity Assuming that a modal can appear in two positions, but without having any flavor specified (in Kratzerian fashion), the question that we need to explain is why a high modal is associated only with epistemic flavors, and a low modal with root flavors. First, we need to let the same modal appear in two positions. In Hacquard (2006, 2010), I derived two positions for modals by adopting a particular view of aspect. Traditionally, aspect is base-generated between tense and VP. I adopted instead von Fintel’s (2001) suggestion that aspect starts as an argument of the verb. Being a quantifier, it needs to move out for type reasons, leaving a trace (type v, for events) inside the VP. The trace of aspect allows the VP to have its event argument saturated. This crucially yields two nodes of type t: TP and VP, as shown in Figure 2.1 (I assume a referential analysis of tense, with tenses as pronouns of type i; Partee 1973; Kratzer 1998; among others): This allows a modal, which needs to combine with a proposition (after it combines with its modal base and ordering source), to appear either right above TP (a) or right above VP, with aspect moving right above the modal (b), as shown in Figure 2.2. Modals keep their Kratzerian lexical entries: they take a modal base f, an ordering source g, and a proposition q, and quantify over the best worlds of the modal base, given the ordering source. The only change from the Kratzerian view is that the conversational backgrounds are relative to an event, rather than a world of evaluation. As we will see in the next sections, different syntactic positions make available different events, which in turn make different modal bases available.

Modals

53

a. 3 TP t Mod

b.

3










T i
















Asp









3VP t 5 λe1


V e1



3

3 3 Asp









3 t
λe1 3VP t Mod 5










V e1

Ti

Figure 2.2: High and Low Modals

(15)

a. [[must]]w = λf< v,st > λg< v,st > λevλq< st > ∀w’∈ BESTg(e)(⋂f(e)): q(w’) = 1 b. [[can]]w = λf< v,st > λg< v,st > λevλq< st > ∃w’∈ BESTg(e)(⋂f(e)): q(w’) = 1

The event variable in the modal’s restriction needs to be bound locally by the nearest event binder.6 There are two possible event binders: aspects (as quantifiers over events), and a default topmost event binder, associated with a speech act projection (cf. Ross 1970; Rizzi 1997; Krifka 2001; among others). When a modal appears low within a clause (i.e., between aspect and VP), the closest event binder is the aspect quantifying over the VP event. The modal’s event variable gets bound by aspect, and thus the modal gets anchored to the VP event: (16) John had to escape. [CP speech e0 λe0 [TP T [AspP Asp1 λe1 [ModP Mod f e1 [VP escape e1 ]]] When a modal appears high in a matrix clause (i.e., above TP), the only event binder is the topmost event binder; the modal gets anchored to the speech event. (17) John had to have escaped. [CP speech e0 λe0 [ModP Mod f e0 [TP T [AspP Asp1 λe1 [VP escape e1 ]]] What happens in embedded contexts? If the modal is in the low position of the embedded clause, it will be anchored to the embedded VP event. However, if the modal is in the high position in the embedded clause, the closest event binder is no longer the speech event, but the aspect quantifying over the matrix event. The modal gets anchored to the matrix (think) event:

54

(18)

chapter two

Mary thought that John had to have escaped. [CP speech e 0 λe0 T Asp2 λe2 think e2 [CP Mod f e2 [TP T Asp1 λe1 [VP escape e1 ]]]

There are thus three possible events that a modal can be anchored to: the VP event (this happens when the modal appears in the low position and is bound by the aspect of its own clause); the speech event (this happens when the modal appears in the high position of a matrix clause); an attitude event (this happens when the modal appears in the high position of a clause embedded under an attitude verb). A first advantage of this system is that it explains why modals seem to be anchored to time and individual pairs, but why not all possible combinations of times and individuals are attested. Indeed, we fi nd that when a modal is anchored to the speaker, it has to be anchored to the speech time (19a) vs. (19b, c); when it is anchored to the subject, it has to be anchored to the VP time (19d) vs. (19e, f); when it is anchored to the attitude holder, it has to be anchored to the attitude time (20a) vs. (20b, c). These constraints are unexpected in a system where modals are anchored to a world. They, however, follow from an event-relative system: if a modal is anchored to an event, it will naturally be anchored to time and individual pairs, namely the running time of that event and its participants. A second advantage of this system is that it will allow us to derive correlations between modal meaning and modal height, as we will see in the next section. (19)

John had to be the murderer. a. Given what I know now, it is necessary that John was the murderer. b. *Given what I knew then, it was necessary that John was the murderer. c. *Given what John knows now, it is necessary that John was the murderer. d. Given John’s obligations then, it was necessary that John be the murderer. e. *Given my obligations then, it was necessary that John be the murderer. f. *Given John’s obligations now, it was necessary that John be the murderer.

Modals

(20)

55

Yesterday, Mary thought that John had to be the murderer. a. Given what Mary knew yesterday, it was necessary that John was the murderer. b. *Given what Mary knows now, it is necessary that John was the murderer. c. *Given what I knew yesterday, it is necessary that John was the murderer.

To sum up, a modal can appear in high (above TP) or low (above VP) position. When high, it is either anchored to the speech event (in matrix contexts) or an attitude event (in embedded contexts). When the modal is low, it is anchored to the VP event in its clause. What remains to be explained is why modals anchored to speech and attitude events (‘high’ modals) are associated with epistemic modal bases, and modals anchored to VP events (‘low’ modals) are associated with circumstantial ones. What sets speech and attitude events apart from regular events?

2.3

Epistemic Modality

Attitude predicates (think), unlike regular ones (escape), describe an information state (e.g., a state of belief for think; i.e., a set of beliefs that an individual holds). Speech acts can also be viewed as a kind of attitude. Assertions in par ticu lar are made relative to an information state, namely, the set of beliefs or commitments of the speaker. We can encode this relativity by assuming assertion operators tied to the speech act projection, which quantify over the speaker’s beliefs (cf. Krifka 1995; Kratzer and Shimoyama 2002; Menendez-Benito and Alonso- Ovalle 2008). In this view, speech and attitude predicates share the property of describing information states. This information state, I argue, is responsible for their special connection with epistemic modality. In a traditional view of epistemic modality, epistemics quantify over a state of knowledge, as in (21). An alternative view, however, argues that the quantificational domain of epistemics is determined by anaphoric reference to an embedding attitude (Veltman 1996; Hacquard 2006, 2010; Yalcin 2007), as shown in (22), where might quantifies directly over Mary’s belief worlds.

56

chapter two

(21)

Mary thinks that John might be the murderer. . . . in some world compatible with what M. knows in her belief worlds, J. is the murderer

(22)

Mary thinks that John might be the murderer. . . . in some world compatible with M.’s beliefs, J. is the murderer

Evidence for this view comes from the contrast between (23) and Yalcin (2007). In the traditional view, the contradictoriness of (23b) is unexpected. Example (23b) should simply ask you to imagine a state of affairs where it is raining, but where your state of knowledge does not rule out the possibility of rain (just as for (23a)). Under the anaphoric domain view, however, the contrast is expected: given that the modal quantifies directly over the supposition state, we run into a contradiction, with rain in all supposition worlds, but no rain in some supposition world. (23)

a. Suppose that [it is raining but you don’t {know/believe} that it is raining]. b. #Suppose that [it is raining but that it might not be raining].

In an event-relative framework, the anaphoricity of ‘epistemic’ modals is encoded by making the epistemic modal base anaphoric to the information state of an attitude event: (24)

fepistemic(e) = λw’. w’ is compatible with INFO STATE(e)7 where INFO STATE(e) is the information state quantified over by an attitude

Modals have to be relative to the nearest event. 8 If this event (or state) is an information state, an epistemic modal base makes the modal quantify over that information state. This happens when the modal is embedded under an attitude verb, as in (25), or when the modal is in the high position of a matrix clause, right under a speech event, as in (26). If the event that the modal is relative to is not an information state (as happens when the modal is anchored to a regular VP event), the epistemic modal base is not available.

Modals

57

(25)

Mary thinks that John might be the murderer.9 There is a thinking e by Mary, and in some world w’ compatible with the info state of e [i.e., Mary’s beliefs in w], that John is the murderer in w’.

(26)

John might be the murderer. There is an assertion e by me, and in some world w’ compatible with the info state of e [i.e., my beliefs in w], that John is the murderer in w’.

(27)

John had to escape. There is an event e, and in some world w’ compatible with the info state of e [#undefined!], that e is an escape by John in w’.

Thus, the epistemic modal base is anaphoric to the information state that the modal is relative to. Only speech and attitude events involve an information state. This predicts that only speech and attitude events can license an epistemic modal base. Anastasia Giannakidou (pers. comm.) suggests that there are other ways to account for Yalcin’s puzzle. In particular, she proposes that (23b) is odd because it is anchored to the speaker’s model (see Giannakidou 1998) and it posits confl icting requirements on that model—to be veridical by the first conjunct (because “speakers present themselves as conveying a veridical state”), and nonveridical in the second conjunct, (because “modals do not convey commitment to truth”): this is pragmatically impossible, hence the oddity. It is not clear to me why the imperative ‘suppose’ should trigger anchoring to the speaker’s model (Yalcin’s puzzle still stands when the speaker commits herself to a lack of rain: It’s not raining, but #suppose that that it is raining, but that it might not be.) However, we might be able to carry this proposal over to a confl ict in the addressee’s model. This is possible. I am certainly not arguing that event anaphoricity is the only solution to Yalcin’s puzzle. Rather, my point here is that this puzzle follows straightforwardly from the event-relativity account I proposed to derive the correlations between a modal’s interpretation and its interactions with tense and aspect. As discussed in section 2.2, the motivation behind anchoring a modal to an event, as opposed to an individual and her epistemic state (or “model” Giannakidou 1998), is that modals seem to be anchored to something more fi ne-grained. Take the sentence ‘John said that Mary must be the murderer, but he knows that his brother did it.’ In this sentence, epistemic must is anchored to John’s speaking

chapter two

58

event, not his epistemic state—hence the lack of contradiction with the following sentence. Event-relativity aims to explain why modals are relative only to certain time and individual pairs but not to others. Anchoring modals to an individual (or to an individual and a time independent of each other) does not suffice to derive the paradigms in (19) and (20).

2.4 Modal Height/Modal Flavor Correlations A modal can appear in two positions within a clause, either right above VP or right above TP. A modal has an event variable in its restriction, which needs to be bound locally. When low, the modal has to be bound by the aspect quantifying over the VP event. When high, the modal is bound either by the aspect quantifying over a matrix attitude event (in embedded contexts) or by the speech event (in matrix contexts). High modals (i.e., modals anchored to speech or attitude events) are associated with epistemic modal bases; low modals (i.e., modals anchored to VP events) are associated with circumstantial modal bases for the following reason: Epistemic modal bases are anaphoric; they require events that involve information states, which only speech and attitude events do. Low modals—that is, modals anchored to regular VP events, which typically do not involve information states10—cannot take an epistemic modal base. Instead, they take a circumstantial one. (28)

fcircumstantial(e) = λw’. w’ is compatible with the circumstances of e

(29)

John had to escape. There is an e which in all w’ compatible with the circumstances of e is an escape by John.

In sum, an event-relative account keeps a Kratzerian semantics where modals are not lexically specified for flavor. The correlation between modal flavor and modal height arises from the event- dependence of modals and from properties of the event that the modals are anchored to. This derives a Cinquean ordering of modals with respect to other functional projections (tense and aspect) without having to hard- code dedicated positions for specific flavors.

Modals

3

59

Robustness of the Constraints on Modal Flavor and Tense and Aspect Interactions

3.1 Temporal Interpretation of Epistemics In the previous section, I assumed that epistemic modals scope above tense (or rather that modals that scope above tense receive epistemic interpretations). But is it truly the case that the time of evaluation of an epistemic modal can never be shifted? First, with an overt restriction, it is possible to evaluate the modal at a past time: (30)

Given what we knew two days ago, John had to be in his office at the time of the crime.

Furthermore, some have argued that even in the absence of an overt restriction, epistemics are sometimes evaluated at a past time. Two kinds of contexts seem to lend themselves to such back-shifted epistemics. The fi rst are justifications for a past action. In example (31), for instance, von Fintel and Gillies (2007) argue that might expresses a past epistemic possibility, which held at the time that I looked in the drawer, but no longer at utterance time: (31)

A: Why did you look in the drawer? B: Because my keys might be in there.

The second kind of context is a narration, as illustrated by in example (32), from Rullman and Matthewson (2012), where had to describes an epistemic necessity given the facts as they stood this morning, but no longer at utterance time: (32)

This morning, I opened my phone bill and was shocked when I saw that I owed $10,000. This had to be a mistake! Unfortunately, it turned out to be correct. My husband had used my phone on his latest trip to Papua New Guinea, forgetting about the roaming charges.

In both types of scenarios, the shifting of the epistemic’s time of evaluation can be pinned on a particular operator, which can independently be

60

chapter two

argued to perform point- of-view shifting: ‘because’ for the former (see Stephenson 2007), and a free indirect discourse operator for the latter. Boogart (2007) argues that the past involved in examples similar to (32) is not a true past (i.e., back-shifted), but a kind of narrative past, where the viewpoint of the speaker is set back to ‘this morning’. The sentence describes a current possibility from that perspective, rather than a backshifted possibility from the present. In support of this, it seems that adding a ‘now’ at the beginning of the modal claim (Now, this had to be a mistake!) is possible, where ‘now’ does not refer to the utterance time, but is shifted to ‘this morning’. Absent such shifting operators, Kratzer (2009) points out that epistemics may not even need to be anchored to the speaker (or local attitude holder), when the context makes a particular source of information extremely salient: (33)

Nobody among us has had access to the information in this fi ling cabinet, but we know that it contains the complete evidence (including possibly forged evidence) about the murder of Philip Boyes and narrows down the set of suspects. We are betting on who might have killed Boyes according to the information in the fi ling cabinet. Harriet, who is innocent, says: ‘I might have killed him.’ According to the information in the fi ling cabinet, it’s possible I killed him.

Event-relativity does not preclude such cases. In the absence of an overt restriction, it explains why epistemics are anchored to the local attitude bearer and the attitude ‘now’ (speaker/speech time in matrix contexts; attitude holder and attitude ‘now’ in attitude contexts). However, it allows the binding of the modal’s event variable by overt or contextually salient events (so long as they involve an information state). An overt restriction (a) anchors the epistemic to the information state provided by the content of the restriction (‘what I knew then’ in (a)); a because- clause anchors the epistemic to the agent’s viewpoint at the time of the caused action; free indirect discourse (c) anchors the epistemic to the ‘now’ of the narrative; a contextually salient body of information or individual anchors the modal to its content/knowledge state (d):

Modals

(34)

a. b. c. d.

61

Given what Mary knew then, John had to be the killer. I ran off because John might be the killer. John had to be the killer! I was now convinced of that. ‘I might be the killer’ (according to John/the info in the fi ling cabinet).

The event-relative LFs in (35) provide a schematic illustration: (35)

a. [speech e : assertion] John hade to be the killer. In all w compatible with info state of e [my doxastic state], J. is the killer. b. [assertion] given what Mary knew then, John hade to be the killer. In all w compatible with info state of e [M.’s past knowledge state], J. is the killer. c. [assertion] according to the content of the cabinet, John hade to be the killer. In all w compatible with info state of e [the content of the cabinet], J. is the killer.

3.2 (Lack of) Actuality Entailments with Root Modals An appealing alternative approach to the scope account I proposed for actuality entailments argues that they arise from interpretive considerations and aspectual coercion (e.g., Mari and Martin 2007; Homer 2010).11 The very rough idea is that perfective limits temporally the ability to φ to the runtime of the φ- event. The most natural way to make sense of such an ephemeral ability is to assume that it was realized (yielding an actuality ‘entailment’). Such a line predicts that actuality entailments should be cancelable. And indeed, in examples (36) and (37), the ability need not have been realized: (36)

Notre nouveau robot a même pu repasser les chemises à un stade bien précis de son développement. Mari and Martin (2007) Our new robot could- pfv even iron shirts at a precise stage of its development.

chapter two

62

(37)

A plusieurs reprises, Olga a pu soulever un frigo, mais ne l’a pas fait. Homer (2010) On several occasions, Olga could- pfv lift a fridge, but didn’t do it.

Deriving actuality entailments through purely interpretive terms provides a natural explanation for why actuality entailments systematically distinguish epistemic from root modals. However, such an account also predicts that actuality entailments should not be idiosyncratic to modals. But as mentioned in the introduction, adjectives and verbs that express root meanings (such as ‘have the means’ or ‘have permission’) do not yield actuality entailments. Now, an interpretive account does not necessarily trigger actuality entailments in such cases: one can conceive of temporary permission, without suggesting that that permission was actualized. But why should a deontic possibility modal differ from the periphrastic ‘have permission’ on purely interpretive grounds, such that the former yields actuality entailments, but the latter does not? In section  4.2, I present a similar contrast between French and Italian ‘want’ which suggests that actuality entailments must involve structural factors. In response to the counterexamples presented here, I have proposed that the lack of actuality entailments with root modals is symptomatic of an additional layer of modality (as with imperfective), which may go undetected for morphological reasons (Hacquard 2009). Examples (36) and (37) use the passé composé, which spells out either a past perfective or a perfect. A perfect does not necessarily force perfective aspect, however, and is compatible with the additional presence of a generic (or other modal) operator, obscured in the morphology. Further support for this account over aspectual coercion comes from Hindi and Bulgarian, which overtly differentiate perfective and perfect aspects. In these languages, perfective (unlike a mere perfect) always forces an actuality entailment, even with sentences like those in (38) and (39), suggesting that the perfect is responsible for the lack of actuality entailments in (38) and (39) (cf. Hacquard 2014).

4

Event-Dependence and Grammatical Modality

Grammatical modals (e.g., verbs or adjectives) express meanings that other words express. However, these other words are not subject to the

Modals

63

same constraints. As we saw, while epistemic modals seem to scope above tense, ‘epistemic’ attitude verbs scope below: (38)

a. It seemed to John that Mary was home. b. John thought/knew that Mary was home. c. It was possible/likely that Mary was home.

Furthermore, unlike root modals, verbs and adjectives that have root meanings do not force actuality entailments when combined with perfective, in a language like French: (39)

a. Jean a voulu s’enfuir, mais il n’est pas parti. Jean wanted- pfv to escape, but he didn’t leave. b. Jean a eu les moyens de s’enfuir, mais il n’est pas parti. Jean had- pfv the means to escape, but he didn’t leave. c. Jean a eu la permission de partir, mais il n’est pas parti. Jean had- pfv permission to leave, but he didn’t leave.

Why do modals tend to be subject to these constraints? Why do verbs and adjectives tend not to? According to the event-relative account, the constraints on a modal’s interpretation arise from their event-dependence to complete their meaning. Lexical words are fully specified for meaning and are not eventdependent. In this view then, event-dependence is what differentiates grammatical from lexical modality. In this section, we test the generality of this claim and probe the lexical/grammatical distinction by turning to grammatical modals that are fully specified for meaning and to verbs that seem underspecified for meaning.

4.1

Cinque’s Hierarchy

One advantage of an event-relative account of modality is that it derives a Cinquean ordering of modals without hard- coding Cinque’s hierarchy. The positions of the functional elements under consideration are derived on purely semantic grounds. Epistemics scope high because they are speech- event-relative, roots low because they are VP- event-relative. There is no hard- coding of dedicated positions for flavor. Semantically, modals come in single lexical entries and end up with different meanings in different positions owing to their event- dependence.

64

chapter two

What about modals that are fully specified for meaning? Do they obey scoping restrictions similar to those of their underspecified counterparts? Given the proposed account, modals show scope restrictions by virtue of having flexible meanings. But if a modal does not have meaning flexibility, can an event-relative system derive meaning-based scope relations, without encoding dedicated structural positions for modal meanings? 4.1.1 ORDERING OF FULLY SPECIFIED MODALS. While modals in many of the world’s languages are reported to be underspecified for meaning, modals in many other languages are fully specified (van der Auwera and Ammann 2005). How do such modals pattern in their interactions with tense and aspect? Do they behave more like underspecified modals or like verbs and adjectives? A full answer to this question would take us well beyond the scope of this chapter, as we would need to look individually at each language carefully. One immediate challenge is to figure out whether such modals are ‘grammatical’ or ‘lexical,’ and what we take such a distinction to be, in a way that is not circular. Traugott (2011) provides the following distinctions, noting that they are “poles on a continuum as opposed to absolute contrasts” (p. 382): lexical categories are members of major word classes (noun, verb, adjective); they express referential meaning and can be used as predicates. We can assume further that lexical categories scope below tense and show characteristics of bi- clausality when they take a sentential complement. (Of course, such a classification gets murky once we consider elements that share properties with both, such as restructuring predicates; see section 4.2). Take Japanese kamosirenai and nitigainai, for instance, which are fully specified for meaning and express epistemic possibility and necessity, respectively.12 Unlike their French and English counter parts, they can scope under tense, as example (40) shows.

(40)

(Sensyu no jiten de wa) John-ga hannin-de aru {kamosirenakat-ta/nitigainakat-ta} (at an earlier point last week) John-Nom culprit-Pred be might-past/must-past (at an earlier point last week) John could/had to be the murderer.

However, these modals are arguably lexical, based on their adjectival morphology and the relative temporal independence of their complement:

Modals

(41)

65

Kinou John-wa zenjitu-ni hon- o yon-da (no) kamosirenakat/ nitigainakat-ta. Yesterday J-Top the.day.before-at book-Acc read-past (C) possible/necessary-past Yesterday, it was possible that John had read a book the day before.

Thus, while fully specified modals (like those in Japanese) may seem not to obey meaning-based temporal constraints, their morphological and syntactic properties may show that they are more lexical than functional in nature. Putting this worry aside, and assuming that there are still many languages with fully specified grammatical modals, do we have any evidence that they display the same constraints as their underspecified homologues? Investigating the interaction of fully specified modals with tense and aspect presents further challenges from the tense and aspect systems of the languages under investigation. Modals in Lilloet Salish, for instance, are fully specified for meaning (Rullman et al. 2008), but it is difficult to investigate their interactions with tense, given the lack of overt tense in that language. Beyond their interactions with tense and aspect, however, there may be one aspect in which fully specified modals obey structural constraints similar to those of their underspecified counter parts, namely the way they scope relative to one another. Nauze (2008) argues that modals from diverse language families show the same Cinquean ordering (epistemic > deontic > ability), regardless of whether they are under- or fully specified for meaning. The epistemic enclitic -kǝƚ, and the circumstantial circumfi x ka . . . a in Lilloet, for instance, are fully specified for flavor and only allow the epistemic to scope over the circumstantial: (42)

Ka-sǝk-s-as-á -kǝƚ ti sq’úm’c-a ti twéw’w’ǝt-a Ka-hit-TR-(3SG-)3SG-a -kǝƚ DET ball-DET DET boy-DET Demirdache (1997) ‘The boy will/might be able to hit the ball.’

It could be that ability and deontic modals do not scope over epistemic modals generally owing to meaning incompatibilities: it is strange to talk about an ability or obligation to bring about an epistemic possibility or necessity. Von Fintel and Iatridou (2004), however, argue against conceptual

chapter two

66

impossibility with the following example. Imagine a scenario in which an insurance company will only pay for an expensive test if there is a possibility that the patient may have Alzheimer’s. This state of affairs can be reported with the sentence in (43): (43)

For the test costs to be reimbursed, it has to (DEONTIC) be possible (EPISTEMIC) that the patient has Alzheimer’s.

Example (43) shows that a lexical epistemic can scope under a deontic modal, suggesting that this ordering cannot be ruled out on purely interpretive grounds. Note, however, that this embedding possibility does not seem available with modal auxiliaries, at least in English: (44) does not seem to have an interpretation where ‘may’ is read deontically and ‘have to’ epistemically.13 (44)

The patient may have to have Alzheimer’s.

Thus the ordering of (grammatical) modals with respect to one other may be rigidly fixed universally, though the matter is not entirely settled. But let us assume that it is, for the sake of argument. Can we derive this ordering without hard-coding dedicated positions, even in those languages with modals fully specified for meaning? 4.1.2 DERIVING CINQUE’S HIERARCHY. If modals are fully specified for meaning, they technically do not need to depend on an event to complete their meaning: their lexical entry could specify that they quantify over epistemically or circumstantially accessible worlds without being event-relative:

(45)

[[mustepis]] = λq ∀w’ compatible with what is known in w: q(w’) = 1

But if that is the case, why should epistemic modals appear high, and root modals low? Do we need to hard-wire a particular position for these fully specified modals in order to prevent unattested scope readings (e.g., where an epistemic would scope below a root modal)? Event-relativity can derive different positions for fully specified modals without any hard- coding of structural position. An alternative to the option given in (45) for fully specified modals lexically specifies a particular modal base, but specifies it in event terms, as in (46):

Modals

(46)

67

[[mustepis]]w = λevλq< st > ∀w’ compatible with INFO STATE (e): q(w’) = 1

Assuming that the event variable in a modal’s restriction still has to be bound by the nearest event binder, such a denotation will force an epistemic modal to appear above Aspect, so that it can be relative to an event that describes an information state—that is, a speech or attitude event. We would thus still predict that epistemics (which are speech/attitude event-relative) scope higher than roots (which are VP event-relative).

4.2 Lexical vs. Grammatical Modality Modals are subject to particular interactions with one another and with tense and aspect, based on the meaning they receive, a circumstance that verbs or adjectives expressing the same meanings are not subject to. We have argued that the particular behavior of modals can be blamed on their event- dependence for full meaning specification. But what about verbs and adjectives? Do they always behave uniformly with respect to their interactions with tense and aspect? Are they always fully specified for meaning? The answer to both questions seems to be no. As for the fi rst question, I showed earlier that, unlike bouletic modals, French ‘want’ (vouloir) does not yield actuality entailments. Italian volere does, however. A continuation that denies the truth of the complement comes out as a contradiction in Italian but not in French: (47)

a. Jean a voulu parler à Marie, mais il ne l’a pas fait. French b. Gianni ha voluto parlare a Maria, #ma non lo ha fatto. Italian ‘John wanted to talk to Mary, (#)but he didn’t do it.’

Why should Italian and French ‘want’ differ in this way? In Hacquard (2006, 2008), I argued that the crucial difference is that Italian volere is a restructuring predicate, unlike French vouloir. Assuming that restructuring verbs form a single clause with their complement (Wurmbrand 2001), I argued that volere sentences involve just one aspect, which quantifies over the VP event across volere, as with a root modal, yielding a single event that necessarily occurs in the actual world (as well as in the desired worlds):

chapter two

68

(48)

[[(48b)]]w = true iff in there is an event e in w, which all worlds w’ compatible with Gianni’s desires, is an event of Gianni talking to Maria.

That the ultimate cause for actuality entailments is structural and not merely meaning-based is further supported by the fact that Italian desiderare (‘desire’), which is not a restructuring predicate, does not yield actuality entailments. (49)

Gianni ha desiderato parlare a Maria, ma non lo ha fatto.

Italian

‘John desired to talk to Mary, (#)but he didn’t do it.’ Hence, while volere seems verbal in taking an individual argument (though see Grano 2012 for a proposal that want’s subject is not even selected by want itself, but raises from its complement), it behaves like a functional element in not requiring its own layer of aspectual quantification. This fundamental property that it shares with modal auxiliaries is responsible for its similar behavior with respect to actuality entailments. As for the second question, there appear to be verbs whose meaning also is flexible, though they are not as common as modals. Mandarin xiang (‘want’), for instance, sometimes expresses belief (‘think’) (Tardiff and Wellman 2000; Nguyen 2013): (50)

Zhangsan xiang Lisi zai Shanghai. Zhangsan {wants/thinks} Lisi to be in Shanghai.

The meaning that xiang receives is constrained by its syntactic environment. Is xiang genuinely ambiguous or should it receive a Kratzerian semantics? First, xiang seems verbal: it takes an individual argument and shows characteristics of bi- clausality, such as temporal independence (at least under its ‘think’ interpretation). Furthermore, the subcategorization constraints for its different interpretations are in line with those of unambiguous verbs that express similar meanings. It is thus not entirely clear that a verb like xiang is truly underspecified, with the full determination of its meaning dependent on its syntactic environment (as I proposed for modals). Verbs with flexible meanings (like xiang) may make a better case for ambiguity or polysemy than modals, given their relative cross-linguistic infrequency. Such verbs would be associated with multi-

Modals

69

ple lexical entries, with their own subcategorization specification, which would match that of their unambiguous synonyms. I leave an investigation of this question for future research. It is interesting to note that, cross-linguistically, ‘want’ seems to be particularly prone to modal behavior: it is often a core restructuring predicate (Wurmbrand 2001; Cinque 2004; Grano 2012), which makes it behave like a root modal (volere); it can be involved in polysemy (xiang), and it is often the source of grammaticalization (Heine 1993; Remberger 2010; among others). Looking at such predicates, which share syntactic and semantic properties with both regular verbs and modals, seems like fertile ground for an investigation of what distinguishes lexical from grammatical modality, and how a verb might turn into a modal. Given my proposal that event- dependence is what distinguishes grammatical from lexical modality, a process of grammaticalization might go as follows. What the verb ‘want’ and the modal ‘must’ have in common is their quantification over worlds. They differ in two respects: ‘want’ is an event sortal, implying participants; ‘must’ is not. The domain of quantification for ‘want’ is lexically provided and anchored to its experiencer, while for ‘must’ it is provided anaphorically via the event variable inside the modal’s restriction. If a verb turned into a modal, it would no longer be an event sortal and would lose its participants. It would, however, keep an event variable in its restriction that would need to be bound locally. This loss would result in a gain in both structural and meaning flexibility. A modal would be able to move freely to two positions, but the meaning that it could receive in these positions would depend on constraints imposed by the event binders that are locally available. If a modal still had a lexically specified modal base, it might be banned in certain positions. If it did not, it would be able to appear in either position and display the kind of meaning-based constraints with tense and aspect we have seen in this chapter. I take up some of these ideas in Hacquard (2013).

5

Conclusion

In many (unrelated) languages, the same modals can express various flavors of modality but show distributional constraints based on flavor (modal flavor/modal position correlations). I argue that these constraints are not solely due to the meaning they express. There is nothing inherent in epistemicity that requires atemporality, or in abilities or deontic

70

chapter two

permissions or obligations that requires actualizations. Instead, I argue that we can blame the correlations between modal flavor and modal position on the event- dependence of modals, without hard- coding particular syntactic positions dedicated to particular meanings. I hypothesize that this event- dependence is what ultimately differentiates grammatical from lexical modality.

Notes I am grateful to the organizers and audience of the TRAIT workshop for getting me started on the work that led to this chapter, and to Anastasia Giannakidou and two anonymous reviewers for their comments. For helpful discussion, many thanks also to Pranav Anand and Alexander Williams. 1. About half of the 200+ languages van der Auwera and Ammann (2005) examined across the world show overlap of form for either possibility or necessity modality, or both. 2. I assume that modal base and ordering source are explicit arguments of the modal, rather than parameters of the interpretation function. The function Best picks (Portner 2009) the most ideal worlds of the modal base given the ordering source. 3. They also differ in related ways in their interactions with negation and quantifiers. See Hacquard (2011) for a survey and further references. 4. Epistemic modals can be interpreted relative to a past with an overt restriction, as in the gloss in (cf. section 3). 5. This is, for instance, expected of so- called alethic (or logical) modality. See von Fintel (2006) for skepticism that alethic modality is ever attested in natural language. 6. The modal’s event argument is occupied by an event pronoun that must be bound locally. Syntactic considerations prevent aspect from appearing here, as it could not move out of a modal’s restriction, an island for movement (cf. Hacquard 2006). 7. In Hacquard (2006, 2010), I use the term content(e). I use info state here to avoid confusion with the term content, sometimes used to refer to the object of belief rather than the set of beliefs against which a belief claim is evaluated. 8. This is not a conceptual constraint, but rather has to be some kind of grammatical principle (cf. Percus 2000 and Hacquard 2006 for locality of binding in world/situation/event arguments). 9. An anonymous reviewer asks whether the event is still available if the attitude is under negation. It should be and yields the following: There is no e such that e is a thinking event by Mary and such that in some world w’ compatible with the info state of e [i.e., Mary’s beliefs in w], John is the murderer in w’.

Modals

71

10. See Hacquard (2006, 2010) for a discussion of cases where the VP event is an attitude itself. 11. Giannakidou and Staraki (2013) show that ability modals in Greek are always implicative when they appear in a causative coordinate structure, irrespective of the aspect that appears on the modal. Given that this structure also yields implicative readings with verbs like try, we can assume that the causative component is responsible for the actuality entailment in such constructions, above and beyond the contribution of modality and aspect. I thus focus here on actuality entailments that occur outside of this type of construction. 12. Thanks to Y. Izumi, Y. Ito, K. Funakoshi, and especially S. Funakoshi for judgments and helpful discussion. 13. Kratzer (1976) argues that in the following German example the embedded modal auxiliary können can be interpreted epistemically while the modal müssen is interpreted deontically. Nauze (2008), however, argues that the embedded modal cannot be interpreted epistemically. Similarly, it is not entirely obvious that ‘possible’ in (43) describes an epistemic rather than a circumstantial possibility. (i) Und auch in Zukunft muss diese Schnecke . . . Saugfüsse And also in future must this snail suction.feet haben können. have might And even in the future, this snail must possibly have suction feet. (Translation from Nauze (2008))

References Bhatt, R. 1999. Covert modality in non-fi nite context. PhD diss., University of Pennsylvania. Boogart R. 2007. The past and perfect of epistemic modals. In Recent Advances in the Syntax and Semantics of Tense, Aspect and Modality, ed.  L. de Saussure, J. Moescher, and G. Puskas, 47–69. Berlin: Mouton de Gruyter. Borgonovo, C., and S. Cummins. 2007. Tensed modals. In Coreference, Modality, and Focus: Studies on the Syntax- Semantics Interface (Linguistik Aktuell/ Linguistics Today 111), ed. L. Eguren and O. Fernández Soriano, 1–18. Amsterdam: John Benjamins. Bybee, J., R. Perkins, and W. Pagliuca. 1994. The Evolution of Grammar: Tense, Aspect and Modality in the Languages of the World. Chicago: University of Chicago Press. Cinque,  G. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford: Oxford University Press. ———. 2004. Restructuring and functional structure. In Structures and Beyond: The Cartography of Syntactic Structures, vol. 3, ed. A. Belletti, 132–191. Oxford: Oxford University Press.

72

chapter two

Demirdache,  H. 1997. ‘Out of control’ in Salish and event (de)composition. In Theoretical Issues at the Morphology- Syntax Interface, ed. A. Mendikoetxea and M. Uribe-Etxebarria, 97–143. Bilbao: University of the Basque Country. Demirdache, H., and M. Uribe-Etxebarria. 2008. Scope and anaphora with time arguments: The case of ‘perfect modals’. Lingua 118(11): 1790–1815. Enfeld, N. 2003. Linguistic Epidemiology. New York: Routledge Curzon. Giannakidou, A. 1998. Polarity Sensitivity as (Non)veridical Dependency. Amsterdam: John Benjamins. Giannakidou, A., and E. Staraki. 2013. Rethinking ability: Ability as modality and ability as action. In Genericity, ed. A. Mari et al., 250–275. Oxford: Oxford University Press. Grano, T. 2012. Control and restructuring at the syntax-semantics interface. PhD diss., University of Chicago. Groenendijk,  J., and  M. Stockhof. 1975. Modality and conversational information. Theoretical Linguistics 2(1/2): 61–112. Hacquard, V. 2006. Aspects of modality. PhD diss., Massachusetts Institute of Technology. ———. 2008. Restructuring and implicative properties of volere. In Proceedings of Sinn und Bedeutung 12, ed. A. Gronn, 165–179. Oslo: ILOS. ———. 2009. On the interaction of aspect and modal auxiliaries. Linguistics and Philosophy 32(3): 279–315. ———. 2010. On the event relativity of modal auxiliaries. Natural Language Semantics 32: 279–312. ———. 2011. Modality. In Semantics: An International Handbook of Natural Language Meaning, ed.  C. Maienborn,  K. von Heusinger, and  P. Portner, 1484–1515. Berlin: Walter de Gruyter. ———. 2013. The grammatical category of modality. In Proceedings of the 19th Amsterdam Colloquium, ed. M. Aloni, M. Franke, and F. Roelfosen, 19–26. Amsterdam: University of Amsterdam. ———. Forthcoming, 2014. Actuality entailments. In Companion to Semantics, ed.  L. Matthewson,  C. Meier,  H. Rullmann, and  T.  E. Zimmermann. New York: Wiley. Heine,  B. 1993. Auxiliaries. Cognitive Forces and Grammaticalization. New York: Oxford University Press. Homer,  V. 2010. French modals and perfective: A case of aspectual coercion. In Proceedings of the West Coast Conference on Formal Linguistics 28, ed. M. B. Washburn et al., 106–114. Somerville, MA: Cascadilla Proceedings Project. Klein, W. 1994. Time in Language. London: Routledge. Kratzer, A. 1976. Was “können” und “müssen” bedeuten können müssen. Linguistische Berichten 42: 128–160.

Modals

73

———. 1981. The notional category of modality. In Words, Worlds, and Contexts, ed. H. J. Eikmeyer and H. Rieser, 38–74. Berlin: Walter de Gruyter. ———. 1991. Modality. In Semantik: ein internationales Handbuch zeitgenössischer Forschung, ed. A. von Stechow and D. Wunderlich, 639–650. Berlin: Mouton de Gruyter. ———. 1998. More structural analogies between pronouns and tenses. In Proceedings of Semantics and Linguistic Theory (SALT) 8, ed.  D. Strolovich and A. Lawson, 92–110. Cornell University. Ithaca, NY: CLC Publications. ———. 2009. Modals and Context Dependency. Colloquium, Harvard University, February 20. Kratzer, A., and J. Shimoyama. 2002. Indeterminate pronouns: The view from Japanese. In Proceedings of the Third Tokyo Conference on Psycholinguistics, ed. Y. Otsu, 1–25. Tokyo: Hituzi Syobo. Krifka,  M. 1995. The semantics and pragmatics of polarity items. Linguistic Analysis 25: 209–257. ———. 2001. Quantifying into question acts. Natural Language Semantics 9(1): 1–40. Laca, B. 2008. On modal tenses and tensed modals. Manuscript, Université Paris 8/CNRS. Mari,  A., and  F. Martin. 2007. Tense, abilities, and actuality entailment. In Proceedings of the Amsterdam Colloquium, ed.  M. Aloni,  P. Dekker, and F. Roelofsen, 151–156. Amsterdam: ILLC/Department of Philosophy, University of Amsterdam. Menendez-Benito,  P., and  L. Alonso- Ovalle. 2008. Modal indefi nites. Natural Language Semantics 18(1): 1–31. Nauze,  F. 2008. Modality in typological perspective. PhD diss., Amsterdam University. Nguyen,  E. 2013. Observations on Mandarin xiang. Manuscript, University of Connecticut. Palmer, F. R. 2001. Mood and Modality. 2nd ed. Cambridge: Cambridge University Press. Partee, B. 1973. Some structural analogies between tenses and pronouns. Journal of Philosophy 70: 601–609. Percus,  O. 2000. Constraints on some other variables in syntax. Natural Language Semantics 8: 173–229. Picallo, C. 1990. Modal verbs in Catalan. Natural Language and Linguistic Theory 8(2): 285–312. Portner, P. 2009. Modality. Oxford: Oxford University Press. Remberger, E-M. 2010. The evidential shift of WANT. In Evidence from Evidentials, ed. T. Petersen and U. Sauerland, 161–182. Working Papers in Linguistics 28. Vancouver: University of British Columbia.

74

chapter two

Rizzi, L. 1997. The fi ne structure of the left periphery. In Elements of Grammar, ed. L. Haegeman, 281–337. Dordrecht: Kluwer. Ross, J. R. 1970. On declarative sentences. In Readings in English Transformational Grammar, ed. R. A. Jacobs and P. S. Rosenbaum, 222–272. Waltham, MA: Ginn. Rullmann, H., and L. Matthewson. 2012. Epistemic modals can scope under past tense. Texas Linguistics Society handout, Austin. Rullman, H., L. Matthewson, and H. Davis. 2008. Modals as distributive indefinites. Natural Language Semantics 16: 317–357. Stephenson, T. 2007. Toward a theory of subjective meaning. PhD diss., Massachusetts Institute of Technology. Stowell, T. 2004. Tense and modals. In The Syntax of Time, ed. J. Guéron and J. Lecarme, 495–537. Cambridge, MA: MIT Press. Traugott,  E. 2011. Modality from a historical perspective. Language and Linguistics Compass 5/6: 381–396. van der Auwera, J., and A. Ammann. 2005. Overlap between epistemic and situational possibility. In World Atlas of Language Structures, ed. M. Haspelmath, M.  S. Dryer,  D. Gil, and  B. Comrie, 310–313. Oxford: Oxford University Press. Veltman, F. 1996. Defaults in update semantics. Journal of Philosophical Logic 25(3): 221–261. von Fintel, K. 2001. Advanced Semantics Seminar, lecture notes, Massachusetts Institute of Technology. ———. 2006. Modality and language. In Encyclopedia of Philosophy, 2nd ed., ed. D. Borchert. Detroit: Macmillan Reference. von Fintel, K., and A. Gillies. 2007. An opinionated guide to epistemic modality. In Oxford Studies in Epistemology, vol. 2, ed. Z. Szabó, T. Gendler, and J. Hawthorne, 32–62. Oxford: Oxford University Press. Wurmbrand, S. 2001. Infi nitives. New York: Walter de Gruyter. Yalcin, S. 2007. Epistemic modals. Mind 116: 983–1026.

chapter three

Epistemic Future and Epistemic MUST: Nonveridicality, Evidence, and Partial Knowledge Anastasia Giannakidou and Alda Mari

1

Epistemic Future and MUST: Dutch, Greek, and Italian

T

he question of whether the notional category ‘future’ in natural languages is a tense or a modality has received a lot of attention in linguistic semantics, and it seems inescapable to conclude that the semantics of the future involves a modal component (e.g., see Bertinetto 1979; Enç 1996; Copley 2002; Squartini 2004; Kaufmann 2005; Mari 2009, 2010,to appear; Klecha 2014; Giannakidou 2012; Giannakidou and Mari 2012a, b; Broekhuis and Verkuyl 2014). Even a purely temporal analysis such as list by Kissine (2008) posits epistemic modality with the future. Future words, often and in many languages, exhibit purely modal readings. Consider as an example the English modal verb will.

(1)

a. The French will be on holiday this week. b. As far as I know, oil will float on water.

(Haegeman 1993)

The sentences here make no reference to the future, but seem to convey epistemic modality: for example, given what I know, the French are on holiday this week. Modal uses for will are common (see Palmer 1987; Tsangalidis 1980), and likewise for future expressions in other European languages—for example, they have been observed in Dutch (Broekhuis and Verkuyl 2014), German, Greek, and Italian.

chapter three

76

Broekhuis and Verkuyl treat the Dutch future verb zullen as an epistemic modal expressing that the prejacent proposition is the result of reasoning based on information judged to be “reliable and well-founded.” They go on to say that when using zullen “the speakers feel sufficiently confident to say p is true at n [now] or is to be made true at i [later than now]. This confidence relies on information judged as reliable and wellfounded. It may take all sorts of form dependent on the situation: as a hypothesis, a confident expectation, a reassurance, etc. What these circumscriptions have in common is that the speaker has entrance to sufficiently many worlds to be able to pick out the ones that seem convincing.” (Broekhuis and Verkuyl 2014, p. 1008). This passage renders Dutch zullen akin to a purely epistemic modal like must, as noted by Giannakidou (2014) and illustrated below: (2)

Context: I can’t see Hein. Hein zal (wel) in de/op see zijn. ‘He must be at sea (swimming/on a boat).’

(3)

Context: I know for sure: #Hein zal in de/op see zijn. #Hein must be at sea. #Hein ist wohl auf Zee.

We see here that zullen is being used epistemically, as an equivalent to must. We also see that a modal particle can be used (wel, wohl) in German (Zimmermann 2011) to convey this meaning equivalent to must. Though the inference that Hein is at sea is based on “reliable and well-founded” information (as Broekhuis and Verkuyl put it), if the speaker actually knows that Hein is at sea, she cannot use zullen or must. Zullen, importantly, receives purely epistemic readings also with past tense, as shown in (4) (example from Giannakidou 2014, attributed to J. Hoeksema): (4)

A: He is so grumpy. Hij zal wel slecht geslapen hebben! ‘He must have slept really badly!’

(5)

#He will have slept really badly.

Epistemic Future and Epistemic MUST

77

Epistemic zullen, like must, allows an epistemic reading with a past, but will with the past appears to force future shifting.1 We will call this must- equivalent use of the future ‘epistemic future’, and modal particles such as wel and wohl will be thought on a par (Giannakidou 2014). Here is the puzzle they all present: must, epistemic future, and wel, wohl are used to convey ‘strong’ statements, products of reasoning by the speaker; at the same time, they are incompatible with full knowledge of the proposition, thus making weaker statements than unmodalized assertions. In this chapter, we study the Greek and Italian future morphemes in their epistemic readings. Data from these languages, though relatively well known in mostly descriptive and typological works (Rocci 2001; Squartini 2004; Pietrandrea 2005; Mari 2009, 2010; Giannakidou and Mari 2013a; Tsangalidis 1998; Chiou 2014), are not very well known or widely discussed in the formal semantics literature, in contrast to English. Unlike will, which is a modal verb, the future markers (which we call FUT in this paper) in Greek and Italian are a bound morpheme (Italian) and a particle (Greek). Their use as epistemic futures is quite widespread and appears with both past and non-past. Epistemic future is also associated with the imperfective aspect, as we see in Greek, and with the progressive in Italian (6b). The combination of non-past and imperfective is the equivalent of present tense in Greek (Giannakidou 2014): (6)

a. I the

Ariadne Ariadne

tha FUT

troi eat.imperf.non-past3sg

tora. now

‘Ariadne must be eating now.’ b. Giacomo ora starà mangiando. Giacomo now FUT.be.3sg eating. ‘Giacomo must be eating now.’ (7)

a. I Ariadne tha ine arrosti (ji’afto dhen ine the Ariadne FUT be.3sg.nonpast sick (for-this not is edo). here) ‘Giovanni must/# will be sick (that’s why she’s not here).’

chapter three

78

b. Giovanni sarà malato. Giovanni FUT-be.3sg sick. ‘Giovanni must/# will be sick (that’s why he’s not here).’ (8)

a. I Ariadne tha itan arrosti xthes (ji’afto the Ariadne FUT be.past.3sg sick yesterday (for-this dhen irthe). not came.3sg ‘Ariadne must/# will have been sick yesterday (that’s why she didn’t come).’ b. Giovanni sarà stato malato ieri. Giovanni FUT.be.3sg been sick yesterday. ‘Giovanni must/# will have been sick yesterday (that why he didn’t come).’

(9)

a. I Ariadne tha milise xthes. the Ariadne FUT talk.past.3sg yesterday. ‘Ariadne must have spoken yesterday.’ b. Gianni avrà parlato ieri. Gianni have.fut.3sg spoken yesterday. ‘Gianni must have spoken yesterday.’

The past sentences contain the adverb meaning yesterday, which makes it clear that the reading is not about the future. The reading we get is, for all intents and purposes, equivalent to epistemic must, and it is fully productive. The predictive future of FUT is illustrated in (10): (10) a. Gianni arriverà domani. John arrive-FUT.3sg tomcorrow. ‘John will/#must arrive tomorrow.’ b. O Janis tha ftasi avrio. The John FUT arrive.nonpast.perf.3sg tomorrow. ‘John will/# arrive at 5 p.m./tomorrow.’

Epistemic Future and Epistemic MUST

79

Here we observe the expected reading of prediction, which involves a non-past that excluded the present. This is a perfective non-past that can be thought of as a prospective form (see Giannakidou 2009, this volume; also Giannakidou and Mari 2015 for more details on the predictive reading, which we do not discuss here). Regarding the epistemic reading, as with must, the speaker has a strong feeling that the proposition is most likely true, but at the same time she does not know that p is true. This is why she chooses to add the FUT. This difference becomes obvious if we compare the FUT/MUST sentence with unmodalized ones, in both Greek (11a, 12a) and Italian (11b, 12b): (11) a. I Ariadne ine/itan arosti, #ala dhen ime ke endelos sigouri. b. Giacomo è malato, #ma non sono sicura. ‘Ariadne/Giacomo is/was sick, #but I am not entirely sure.’ (12) a. I Ariadne tha ine arosti, ala dhen ime ke endelos sigouri. b. Giacomo sarà malato, ma non sono sicura. ‘Ariadne must be sick, but I am not entirely sure.’ In the unmodalized present and past sentences, the speaker is fully committed to the truth of the sentences. By asserting that Ariadne and Giacomo were sick, the speaker expresses knowledge that Ariadne and Giacomo were sick; hence a continuation that questions this knowledge is impossible because it would attribute to the speaker inconsistent knowledge. The epistemic future, on the other hand, is fi ne with I am not entirely sure. The sentences indicate an epistemic state in which the speaker considers a proposition to be very plausible while not being fully certain about it. This is in contrast to the positive unmodalized assertion in which the speaker has no doubt that p is true in the actual world, and we thus get a Moore effect. We can impressionistically describe this difference, following Karttunen (1972), by saying that the future/must sentences are ‘weaker’ than the unmodalized assertion. In (13)–(17) we give examples in Greek and Italian with the equivalents of must illustrating the same thing (in Greek prepi takes a subjunctive na- complement, like all modal verbs; Giannakidou 2009):

chapter three

80

(13)

a. I the ke and

Ariadne Ariadne endelos totally

prepi na troi tora, alla den ime must subj eat.present3sg now but not am sigouri. certain

b. Giacomo deve star mangiando #ma non sono sicura. . Giacomo must be eat-gerund but not am certain ’Giacomo/Ariadne must/will be eating now, but I am not entirely sure.’ (14)

I the ime am

Ariadne prepi na milise xthes, alla den Ariadne must subj talk.past.3sg yesterday, but not ke endelos sigouri. and totally certain

‘Ariadne must have spoken yesterday, but I am not entirely sure.’ (15)

a. Giovanni Giovanni totalmente totally

sarà stato malato, ma non sono FUT-be.3sg been sick, but not am sicura. certain.

b. Giovanni deve essere stato malato, ma non sono Giovanni must.3sg be been sick, but not am totalmente sicura. totally certain. ‘John must have been sick, but I am not entirely sure.’ The sentences with MUST, and the sentences with epistemic FUT are equivalent in the speaker’s intuition. (We use MUST for the equivalents of must cross-linguistically, including, of course, for must itself). They seem to be strong, but at the same time they are weaker than unmodalized assertions in that they are compatible with ‘I am not entirely sure’. FUT and MUST can also combine (Giannakidou 2012). The reading remains the same: (16)

I the

Ariadne tha Ariadne FUT

prepi must

na milise xthes. subj talk.past.3sg yesterday.

‘Ariadne must have spoken yesterday.’

Epistemic Future and Epistemic MUST

(17)

Giacomo Giacomo

81

dovrà aver parlato ieri. must-FUT.3sg have spoken yesterday.

‘Giacomo must have spoken yesterday’. Giannakidou (2012) and Giannakidou and Mari (2013, 2015) characterize this co- occurence as modal concord (see Huitink 2014 for an overview and references). Given the epistemic FUT and the parallel with MUST, it is appealing to argue that the epistemic future is an epistemic modal akin to must, and this what we pursue here, following our earlier work. 2 The current debate in the literature recruits the terms ‘weak’ and ‘strong’ (see von Fintel and Gillies 2010), but we fi nd these labels unsatisfactory. Here we argue that universal modals appear ‘weak’ because, like all modals, they are nonveridical (Giannakidou 1998, 1999; Beaver and Frazee 2011). We defi ne two kinds of (non)veridicality—objective and subjective (relying on epistemic commitment; for a fi rst approach see Giannakidou 1994, 1997). All modals are nonveridical objectively because they do not entail truth; and they are nonveridical subjectively because they do not entail knowledge of p by the speaker. At the same time, universal modals are biased modals: they entail p in the Best worlds, and this is what makes them appear strong. Our discussion, fi nally, implies that the category ‘modal’ is primarily a semantic category, not specifically tied to the morphosyntatic category ‘modal verb’. This is an important conclusion when we consider the larger cross-linguistic picture. The discussion proceeds as follows. In section 2, we present the notions of veridicality, nonveridicality, epistemic weakening, and biased modal that we will use for the analysis of FUT and MUST. In section 3, we discuss evidentiality and partial knowledge, and in section 4 we deal with the implications of our analysis as it relates to apparent Moore paradoxical effects.

2

FUT and MUST: Nonveridicality and Bias

In this section, we present the notions of veridicality and nonveridicality that serve as the foundation for our analysis of modality. After we clarify the core notions, we defi ne epistemic weakening as a weakening & commitment by the speaker—that is, the creation of a nonveridical epistemic space (Giannakidou 2014). Commitment weakening is decisive also for

chapter three

82

the analysis of subjunctive mood, as illustrated by Giannakidou (this volume) and we refer to that chapter for more on that topic.

2.1

Veridicality and Nonveridicality

Montague (1969) uses ‘veridicality’ to characterize sentences with direct perception verbs, such as see (see Giannakidou 2013a for a formal connection between truth and existence in relative clauses and with progressives). Authors have also used other labels, such as factivity and factuality to refer to veridicality (Karttunen 1971; Karttunen and Zaenen 2005; Kiparksy and Kiparsky 1970), as well as actuality (Bhatt 2006; Hacquard 2010). Veridicality in this second use is understood objectively as truth in the actual world: a sentence is veridical if the proposition it denotes is true in the actual world (Egré 2008). Zwarts (1995) and Giannakidou (1994, 1997) defi ne veridicality in terms of entailment as a property of expressions F, where such F entails the truth of its complement p: (18)

Def. 1. Veridicality. (based on Zwarts 1995, Giannakidou 1997, 1998, 1999). Let F be a unary sentential operator. The following statements hold: (i) F is veridical iff Fp → p is logically valid; other wise F is nonveridical. (ii) A nonveridical F is antiveridical, iff Fp → ¬p.

Functions that have veridicality and nonveridicality are propositional functions (but see Bernardi 2001 for type-flexible defi nitions). F is veridical iff Fp entails p—i.e., if whenever Fp is true, p is true too. F is nonveridical if Fp does not entail p—i.e., if when Fp is true, p may or may not be true. Note that nonveridical operators do not entail the falsity of p; this is a property of antiveridical operators such as negation (Giannakidou 1998, 1999, 2013). We call nonveridicality under this defi nition ‘objective’, because it does not make reference to parameters other than the propositions and the functions. There are no subjective parameters such as what individuals think or believe, though these will be needed for the treatment of modality and mood. For now, consider that if a verb such as know is our function F, know p, if true, entails p, and know is therefore veridical. But if want is the F, want p, even if it is true, does not entail p, and therefore

Epistemic Future and Epistemic MUST

83

want is nonveridical. Negation is also nonveridical, because not-p does not entail p. Objectively, veridicality appears close to factivity and actuality and characterizes operators such as know, which imply truth in the actual world. The sentences under the veridical or nonveridical operator can be called veridical and nonveridical too. Consider now temporal operators, such as past and present. Veridicality now has to be relativized with respect to the utterance time, as suggested by Giannakidou (2002) and can be defi ned as follows (Giannakidou 2002, (23)): (19)

Def. 2. Veridicality of temporal operators Let F be temporal function, t, an instant or an interval. F is veridical iff, if Fp is true at a time t, it entails that p is true at a (contextually given) time t’ ≤ t; other wise F is temporally nonveridical.

(20)

If a sentence of the form PAST(p) is true at tu , then it entails that p was true at a time prior to tu.

(21)

If a sentence of the form PRES(p) is true at tu , then it entails that p is true at tu.

Temporal veridicality is objective veridicality because the time of evaluation is an objective parameter and not a subjective one. Present and past, as we see, are temporally veridical. The future is clearly not, according to Def. 2, since FUT p does not entail that p is true at a time prior to or equal to tu. Moving now to modals, Giannakidou 1998, and Beaver and Frazee (2011) present nonveridicality as a defi ning property of the category modality. Consider: (22)

Nicholas might/must bring dessert.

(23)

Nicholas might/must have brought dessert.

Nicholas’s bringing of dessert is not, and cannot be, an actual fact under a modal. Logically, possibly p does not entail p. And must is also nonveridical, since must p does not entail that p either. MUST does not validate the veridicality principle T that holds for aleithic modality (see Giannakidou 1998, 1999).

84

chapter three

Objective veridicality appears equivalent to actuality, but veridicality is often also discussed in the context of commitment. As de Marneffe et  al. (2012) put it: declaratives like ‘Ariadne left’ convey fi rm speaker commitment, whereas qualified variants with modal verbs or embedded sentences imbue the sentence with uncertainty (de Marneffe 2012, p. 102). Similarly, Trnavac and Taboada (2012) use modals as nonveridical markers of uncertainty. The speaker is said to be ‘fully committed’ to the truth of an unmodalized sentence in the present or simple past, but is not fully committed when she uses a modal. Therefore, when we talk about the truth of a sentence, we talk about it in two ways: objectively, by appealing to what is true in the actual world at a given time, and subjectively by appealing to the speaker’s commitment to the truth of the sentence. Giannakidou tries to capture the connection between veridicality and speaker commitment by making the veridicality judgment relative to individual anchors. The truth of a sentence is now anchored to the individual asserting it. In main clauses the anchor is by default the speaker.3 Giannakidou defi nes models of evaluation to describe the information states of anchors. These models are sets of worlds, relative to i, corresponding to what i believes or knows.4 We call these models epistemic states in Def. 3: (24)

Def. 3. Epistemic state of an individual anchor i (Giannakidou 1999: (45)) An epistemic state M(i) is a set of worlds associated with an individual i representing worlds compatible with what i knows or believes.

Given the epistemic state, we can now identify (non)veridicality subjectively. Truth is defi ned with respect to knowledge (and belief): (25)

Def. 4. Subjective veridicality A function F that takes a proposition p as its argument is subjectively veridical with respect to an individual anchor i iff Fp entails that i knows or believes that p is true. This means that i’s epistemic state M(i) is such that: M(i) ⊆ p.

It follows that ∀w[w ∈ M(i) → w ∈ {w’|p(w’)}]. i knows/believes p, i is fully committed to p. Consider the following unembedded case:

Epistemic Future and Epistemic MUST

(26)

O Giannis The John

kerdise won.3sg

to the

85

agona. race.

‘John won the race.’ (27)

a. John won the race. b. [[John won the race]]M(speaker) = 1 iff ∀w[w ∈ M (speaker) → w ∈ {w’| John won the race in w’}]

If the speaker asserts John won the race, she must believe or know that John won the race, hence all worlds in M (speaker) are John-won-the race worlds: M (speaker) ⊆ p. The unmodalized sentence is therefore equivalent to the speaker knowing p. The indicative is therefore the mood that conveys homogeneity of a modal space M(i) as being a p space. This is useful also when we think of the direct evidential where it is simply an unmarked past or present. The simple past or present draws on ‘direct’ evidence in the sense that it gives the more reliable, undisputed knowledge. In the absence of indirect evidence, all worlds in M(speaker) are p worlds. Subjective nonveridicality, on the other hand, is epistemic uncertainty (or noncommitment): it indicates that the speaker does not know for sure that p is true. In this case, the epistemic state only intersects with p and therefore contains worlds where p is not true: (28)

Def. 5. Subjective nonveridicality A function F that takes a proposition p as its argument is subjectively nonveridical with respect to an individual anchor i iff Fp does not entail that i knows or believes that p is true. This means that i’s epistemic state M(i) is such that: M(i)–p ≠ ∅

From Def. 5, it follows that ∃w’ ∈M(i) : ¬p(w’). Hence a subjectively nonveridical function imposes nonhomogeneity on the epistemic state, since there is at least one not-p world. Modals are objectively nonveridical, as mentioned earlier, but also subjectively. A speaker asserting MUST/MAY p has, in her epistemic state, non-p worlds. Modals and the FUT are objectively nonveridical, as mentioned earlier, but also subjectively: the modal bases (which are subsets of M(speaker) have a p and a not-p space, and M(speaker) likewise is partitioned into p and not-p worlds. Like us, Condoravdi (2002) acknowledges the nonveridicality of

86

chapter three

modals in her diversity condition, which requires that there be one not-p world in the modal base. Subjective veridicality can thus be extended to characterize the epistemic states themselves. A veridical epistemic state is a nonpartitioned homogeneous epistemic state. A nonveridical epistemic state, on the other hand, is a space partitioned into p and ¬p worlds. (29)

Def. 6. Veridical and nonveridical epistemic states and commitment . a. An epistemic state (a set of worlds) M(i) relative to an individual anchor i is veridical with respect to a proposition p iff all worlds in M(i) are p-worlds. (full commitment) b. If there is at least one world in M(i) that is a ¬p world, then M(i) is nonveridical. (weakened commitment, uncertainty) c. If all worlds in M(i) are ¬p worlds, then M(i) is antiveridical. (counter- commitment)

A veridical epistemic state is a nonpartitioned, homogeneous epistemic state, a state of full commitment. A knowledge state is veridical; as we said, umodalized sentences in the past reveal veridical states. A nonveridical state M(i), on the other hand, is defi ned as one that contains at least one ¬p world. It is a nonhomogeneous partitioned state; it allows uncertainty, and in this case we talk about weakened commitment. All epistemic modals convey weakened commitment, and states of indirect evidentials are also nonveridical (as discussed in Giannakidou and Mari 2015). Importantly, with modals, weakened commitment and nonveridicality arise because the modal base is ordered, and p is true only in the Best worlds conforming to the ordering source (Portner 2009). Modal ordering creates nonveridical spaces generally. When all the worlds are ¬p, the state is antiveridical, as with negative and counterfactual assertions, which express counter- commitment of the anchor. Antiveridicality characterizes, generally, nonassertion—that is, optative and imperative sentences—since at the issuing of optative and imperative p clearly does not hold, and some may be invitations to bring about p (e.g., the imperative). Counter- commitment and weakened commitment are noncommitment to p, though only weakened commitment operators carry uncertainty. From the epistemic domain we can move to generalize veridicality and nonveridicality to all kinds of modal spaces (sets of worlds), includ-

Epistemic Future and Epistemic MUST

87

ing various kinds of modal bases. Veridicality and nonveridicality are now properties of modal spaces: (30)

Def. 7. Veridical and nonveridical modal spaces . a. A set of worlds M is veridical with respect to a proposition p iff all worlds in M are p-worlds. (homogeneity). b. A set of worlds M is non veridical with respect to a proposition p iff there is at least one world in M that is a ¬p world. (nonhomogeneity) c. A set of worlds M is antiveridical with respect to a proposition p iff M and p are disjoint.

Veridical spaces are homogeneous, whereas nonveridical spaces are nonhomogeneous. All modal bases are nonveridical spaces in this sense, since they are partitioned by their ordering in the Kratzerian semantics (see also Portner 2009; and Condoravdi’s 2002 diversity condition, which we mentioned earlier). Likewise, bouletic and deontic domains are nonveridical since they are also ordered. Ordering (Kratzer, 1981/1991) creates a partition, and therefore necessarily a nonveridical modal space. (Antiveridical states, on the other hand, are homogeneous. A typical example is the model of the speaker when interpreting a negative sentence.) Given nonveridical spaces, we must distinguish the cases in which we have ordering sources (Kratzer 1981, 1991; Portner 2009) from those in which we do not. Ordering sources characterize stronger modals such as must. With Portner we defi ne ‘ordering sources’ and ‘Best worlds’. (31)

Def. 8. Ordering of worlds (Portner 2009, p. 65) For any set of propositions X and any worlds w, v : w ≤x v iff for all p ∈ X, if v ∈ p, then w ∈ p.

(32)

Def. 9. Best worlds as per X. Best X: {w’ : ∀q ∈ X (w’ ∈ q)}

Now we defi ne ‘support set’: (33)

Def. 10. Support set of a proposition p. In a nonveridical modal space M, the support set W ⊂ M of a proposition p is the nonsingleton set of worlds that rank as Best, and is such that all worlds w’ in W are p-worlds.

88

chapter three

Given the notion of a support set, we can now defi ne projected truth with respect to that set. (34)

Def. 11. Actual truth. p is actually true iff p is true in the actual world.

(35)

Def. 12. Projected truth. p is projectively true iff p is true in all the worlds of the support set W: W ⊆ p.

In other words, a nonveridical modal space M supports a proposition if there is a support set W for the proposition in M. Since the support set is the set of Best worlds, this structure reveals a bias towards Best worlds. Modals that come with support sets, such as MUST and FUT are nonveridical, like all modals, but are biased. We thus defi ne the new category of biased modals: (36)

Def. 13. Biased modals. A modal operator F is biased if its modal base M contains a support set W ⊂ M : W ⊆ p.

We can think of the support set as the inner domain of the modal F, and the modal base as its outer domain. In this structure, it becomes clear that strength does not mean that the modal entails actual truth (veridicality). Rather, the biased modal entails p in the inner domain (i.e., in the support set), while p remains unsettled in the modal base, which is nonveridical and allows non-p worlds. There are two kinds of nonveridical epistemic modal spaces: those that contain a support set for a proposition (the biased modals), and the possibility modals that convey nonveridical equilibrium between p and ¬p (Giannakidou 2013b): (37)

Def 14. Nonveridical equilibrium (Giannakidou 2013b). An epistemic state M is in nonveridical equilibrium iff M is partitioned into p and ¬p, and there are no Best worlds.

A nonveridical state with equilibrium reveals no preference because there is no ordering. Consider, for example, It might rain tomorrow. This is a mere possibility statement, and there is no ordering that could create a support set for the proposition ‘it rains tomorrow’. Ordering sources add information restricting sets of possibilities and creating support sets,

Epistemic Future and Epistemic MUST

89

thus privileging one subset of the modal base over its complement (¬p). In terms of commitment, we propose the following scale of commitment strength, from strongest to weakest: (38)

Commitment strength More committed < unmodalized p, MUST p, POSSIBLY p> less committed

The modal space M of an unmodalized sentence contains only p worlds. When all worlds in M are p worlds, we have veridicality, and this conveys the strongest commitment. With a biased modal like MUST, we have a nonveridical space with a set of Best worlds where p is true (Giannakidou 2013b; Giannakidou and Mari 2013). In this case, the p worlds are the support set of p, but the modal base and M(speaker) still allow not-p worlds. The possibility sentence, on the other hand, conveys equilibrium between p and not-p (Giannakidou 2013; Giannakidou and Mari 2015); i.e., there is no preference for the p or non-p worlds, no Best worlds, no support of p. This is so because there is no ordering with the possibility modal. Whenever there is ordering, there are Best worlds; the universal modal will therefore give rise to stronger commitment. In other words, the universal modal is strong in the sense of partially supporting p in the Best worlds, but it is still weak because it is nonveridical objectively and subjectively. We defi ne here the subjective nonveridicality of epistemic modals: (39)

Subjective nonveridicality of epistemic modals MODAL epistemicp does not entail that the speaker knows that p is true.

All epistemic modals are subjectively nonveridical, possibility, as well as necessity modals. Epistemic modals, then, are nonveridical both objectively (because MODAL epistemic p does not entail p) and subjectively because MODAL epistemic p does not entail the speaker does not know that p. Hence, epistemic modals, even biased ones, are doubly weak (pace von Fintel and Gillies 2010). Recall the position of de Marneffe et  al. 2012 that modal verbs imbue the sentence with uncertainty (see also Trnvac and Taboada 2012).

chapter three

90

2.2

Epistemic Future and MUST

As noted at the beginning of the chapter, FUT has extensive epistemic use in Greek and Italian. We repeat the basic data. The epistemic use arises with non-past and with past, statives, and eventives: (40)

a. I Ariadne the Ariadne

tha FUT

milise xthes. talk.past.3sg yesterday.

‘Ariadne must have spoken yesterday.’ b. Gianni avrà Gianni have.fut.3sg

parlato ieri. spoken yesterday.

‘Gianni must have spoken yesterday.’ (41)

a. I Ariadne tha the Ariadne FUT

ine arrosti. be.3sg sick.

‘Ariadne must be sick.’ . b. Giovanni sarà malato. Giovanni FUT-be sick ‘Giovanni must be sick.’ (42)

a. I Ariadne tha the Ariadne FUT

itan was.3sg

arrosti. sick.

‘Giovanni/Ariadne must have been sick.’ b. Giovanni Giovanni

sarà must

stato malato. been sick.

‘Giovanni must have been sick.’ In all cases, the speaker is considering information she has and draws an inference based on that information. In accordance with what we said so far, the epistemic FUT/MUST modal base is nonveridical, as evidenced by the possibility of negative continuations, illustrated in (43) and (44):

Epistemic Future and Epistemic MUST

(43)

91

a. I Ariadne ine arosti, #ala dhen ime ke the Ariadne is sick, but not be.1sg and endelos sigouri. absolutely sure b. Giacomo è malato, #ma Giacomo is sick, but

non not

sono am

sicura. certain.

‘Ariadne/Giacomo is sick, #but I am not entirely sure.’ (44)

a. I Ariadne tha ine arosti, ala dhen the Ariadne FUT is sick, but not endelos sigouri. absolutely sure b. Giacomo Giacomo

sarà malato, be.3sg.fut, but

ma non not am

ime ke be.1sg and

sono sicura. sure.

‘Ariadne must be sick, but I am not entirely sure.’ In contrast to an unmodalized sentence, the FUT/MUST sentence is compatible with a continuation revealing weakened certainty. This indicates that not all worlds in the modal base are p worlds. With unmodalized past sentences, on the other hand, the not entirely sure continuation is not possible, as we see, because they convey unpartitioned, veridical, epistemic states. We come back to these sentences in section 4. In (45)–(48) we give examples, in Greek and Italian, with the verb equivalents of MUST, and note that they pattern with FUT and contrast with unmodalized assertions: (45)

a. I Ariadne prepi the Ariadne must ime ke endelos be.1sg and absolutely

na troi tora ala dhen subj eat.present3sg now but not sigouri. sure

b. Giacomo deve star mangiando, ma non sono totalmente Giacomo must be eat-gerund, but not am totally sicura. certain. ‘Giacomo/Ariadne must/will be eating now, but I am not entirely sure.’

chapter three

92

(46)

a. I Ariadne prepi na milise xthes, ala dhen the Ariadne must subj talk.past.3sg yesterday, but not ime ke endelos sigouri. be.1sg and absolutely sure ‘Ariadne must have spoken yesterday, but I am not entirely sure.’ b. Gianni deve aver parlato Gianni must have spoken ieri. yesterday, ma non sono totalmente sicura. ‘Gianni must have spoken yesterday, but I am not entirely sure.’

Sentences with MUST and those with epistemic FUT are equivalent in the speaker’s intuitions. FUT and MUST can combine in an instance of modal concord: (47)

I Ariadne tha prepi na milise xthes, ala the Ariadne fut must subj talk.past.3sg yesterday, but dhen ime ke endelos sigouri. not be.1sg and absolutely sure. ‘Ariadne must have spoken yesterday but I am not entirely sure.’

(48)

Giacomo Giacomo

dovrà aver parlato must-FUT.3sg have spoken

ieri. yesterday.

‘Giacomo must have spoken yesterday, but I am not entirely sure.’ Given the epistemic FUT and the parallel with MUST, we will pursue the same semantics for both. In Giannakidou and Mari (2015) we show that the predictive reading differs from the epistemic only in the modal base (which in the predictive reading is metaphysical). Here we focus strictly on the epistemic reading. Epistemic FUT is associated with an epistemic modal base. Specifically, the modal base is the set of propositions known by the speaker (w0 is the actual world): M = λw’.w’ is compatible with what is known by i (the speaker) in w 0. Note that ∩fepistemic(w 0). ⊂ M(speaker). (Our epistemic modality is thus subjective; see the distinction between objective and subjective by Papafragou (2006). In fact, given that we relativize with respect to individual anchors, there can be no objective modality, strictly

Epistemic Future and Epistemic MUST

93

speaking, in our system.) Given what the speaker knows, the modal base contains p worlds, but also ¬p worlds; it is nonveridical and nonhomogeneous. Let us defi ne the ordering and then Best worlds given the ordering. (49)

For any set of propositions 1 and any world w, w’: ≤ 1 w’ iff for all q ∈ 1, if w’ ∈ q, then w ∈ q.

(50)

Best worlds given the ordering 1. 1 : {w’ ∈ (∩ M): ∀q ∈ 1 (w’ ∈ q)}.

1 are a subset of worlds in the epistemic modal base, in which strange things do not happen (see Portner 1998; Mari 2014a). For instance, if I have red cheeks and an sneezing, then, under normal circumstances, I have the flu. However, circumstances are not necessarily normal. In extraordinary circumstances these symptoms are secondary and indeed indicative of a potentially much worse disease. The modal base is thus partitioned, and the modal space is subjectively nonveridical. One of the subsets of the modal base is ranked as the set of Best worlds given the ordering 1. FUT universally quantifies over the set Best 1 (which is a subset of the modal base, as we see). Hence: (51)

At the utterance time tu , 冀FUTepistemic(PRES(p))冁M will be defi ned only if the modal base M is nonveridical; if defi ned, 冀FUTepistemic(PRES(p))冁 1 = 1 iff ∀w’ ∈ Best 1 : p(w’, tu)

(52)

At the utterance time tu , 冀FUTepistemic(PAST(p))冁M will be defi ned only if the modal base M is nonveridical; if defi ned, 冀FUTepistemic(PAST(p))冁 1 = 1 iff ∀w’ ∈ Best 1 : ∃t’ ≺ tu Λ p(w’, t’)

(53)

At the utterance time tu , 冀MUSTepistemic(PRES(p))冁M will be defi ned only if the modal base M is nonveridical; If defi ned, 冀MUSTepistemic(PRES(p))冁 1 = 1 iff ∀w’ ∈ Best 1 : p(w’, tu)

(54)

At the utterance time tu , 冀MUSTepistemic(PAST(p))冁M will be defi ned only if the modal base M is nonveridical; if defi ned,

94

chapter three

冀MUSTepistemic(PAST(p))冁 1 = 1 if ∀w’ ∈ Best 1 : ∃t’ ≺ tu Λ p (w’, t’) This truth- conditional content of FUT is identical to MUST, to which we assign exactly the same truth conditions. By MUST we designate crossinguistically expressions that are the semantic cognates of English must, and of coure must itself: Following Giannakidou 2009 and Giannakidou and Mari 2015, we take it that the temporal information comes from the lower tense, which can be past or present (see also Staraki 2013 for an analysis of Greek modals as purely modal operators and not mixed modal/temporal as in Condoravdi 2002). How about the actual world? Best worlds are those in which strange things do not happen. Typically the actual world tends to be nonextraordinary (Portner 2009), but we also know that strange things happen. As a consequence, we do not claim that the epistemic agent knows that the actual world belongs to the set of Best worlds. Given that the accessibility relation is epistemic and therefore reflexive, it is ensured that the actual world is in the modal base (see Matthewson et al. 2007; Portner 2009), but it is not guaranteed that the actual world belongs to the p worlds. With universal quantification over the set of Best worlds, however, truth is projected within the support set and this is responsible for the sense of strength that comes with FUT and MUST; but it must not be confused with veridicality, which expresses full commitment to p and therefore does not allow for the possibility of not p within the modal base. 5 This account treats MUST as both strong (because of quantification within the set of Best worlds) and weak (because of the nonveridical modal base). p is entailed in the Best worlds over which MUST quantifies, although is it is not entailed in the modal base. It is useful to compare our account to von Fintel and Gillies (2010). They posit the existence of a kernel K, which is a set of worlds representing the privileged information. They then pose a presupposition of ‘direct unsettledeness’ in the kernel: (55)

Definition 5. (String must + evidentiality) Fix a c-relevant kernel K a. [[ mustф]]c,w is defi ned only if K does not directly settle [[ф]]c b. If defi ned, [[ mustф]]c,w = 1 iff BK ⊆ [[ф]]c

Von Fintel and Gillies say that K can fail to directly settle whether P, even though K entails whether P (von Fintel and Gillies 2010, p. 372).

Epistemic Future and Epistemic MUST

95

The modal base, in their account is a homogeneous space of p worlds; in ours it is not. However, because of the ordering source, we are able to carve out a set of worlds in which p is true, and it is on this set that MUST quantifies.

3

The Evidential Component: Partial Knowledge

In his seminal work, Kartunnen (1972) held that the weakness of MUST is intimately related to the weakness of the source of information. The view that we hold here is that epistemic weakening makes the biased modal statement compatible only with a partitioned nonveridical epistemic state consisting of a subspace of Best worlds that supports p (the inner domain), and a subspace that does not (the outer domain, the modal base). An important difference between our view and Kartunnen’s is that the epistemic weakening is not due to the fact that knowledge is indirect, but to the fact that knowledge is partial. Kartunnen ties ‘weakness’ to indirect evidence: when the speaker has indirect evidence that the prejacent is true, she uses the modal to signal that she is uncertain about the truth of the prejacent. Von Fintel and Gillies (2010, p. 361) challenge this position: “Our point is simple: weakness and indirectness are not two sides of a single coin at all. They are just different.” Their claim is that the epistemic modal must presupposes indirect evidence but that it is ‘strong’. In our view of epistemic weakening, the indirectness is reduced to a mere side effect, not a real phenomenon. The key is partial knowledge: epistemic weakness arises because the speaker is reasoning with partial knowledge, and she knows that she does not have all the facts (see also Mari 2010). Recall the partition that lies at the heart of the truth condition of universal, biased modals: the partition between best and non-best worlds. When the speaker reasons with a universal modal, she is aware that she does not have all the knowledge she needs to draw a valid conclusion in all worlds in the modal base. When she has complete knowledge, she cannot use a universal modal, as evidenced in direct visual perception contexts: (56)

Context: Direct visual perception of rain a. # It must be raining. b. # Tha vrexi. FUT rain.

chapter three

96

c. # Piovera. rain-FUT.3sg. d. # Tha prepi na vrexi. FUT must subjunctive rain. e. # Dovrà piovere. Must-FUT.3sg rain. Recall that, biased modals are similar to devices such as the German and Dutch modal particles, as we mentioned in the introduction. If I see the rain, I know that it is raining, and knowledge is veridical: if I know p, then all worlds compatible with my knowledge are p worlds. My epistemic space is not partitioned, but the opposite: it is homogeneously supporting p. Epistemic weakeners are incompatible with the state of complete, homogeneous knowledge that comes with direct perception. Why we need partiality rather than indirectness of knowledge is nicely illustrated in the contrast between the context above, where I see the rain, and in (57), where I only see a wet umbrella. (57)

I see a wet umbrella. a. It must be raining. b. Tha/Prepi na v rexi. FUT/Must subjunctive rain. c. Deve star piovendo. Must be raining. d. Pioverà. FUT-rain.3sg. e. Deve star piovendo, ma non sono sicura. ‘It must be raining, but I am not sure.’

In this context, I see a wet umbrella, but I do not see the rain, therefore I do not know that it is raining. The wet umbrella is an indication of rain and can support ‘It is raining’ by licensing the missing premise that the umbrella got wet because of the rain. The biased epistemic and FUT modals are fi ne in this case, in contrast to the direct perception of rain that we just saw, where they are bad. Continuation with ‘I am not sure’ is allowed here, as we see.

Epistemic Future and Epistemic MUST

97

Likewise, auditory perception is compatible with MUST and FUT because this too gives incomplete knowledge. (58)

Context: I am in a room with no windows, but I hear sounds of rain on the roof. a. It must be raining. b. Tha vrexi. FUT rain. c. Pioverà. rain-FUT.3sg. d. Tha prepi na vrexi. FUT must subjunctive rain.

I do not see the rain, so I do not know that it is raining; I only have the sound of rain. The sound might be caused by something other than the rain (hence I am also missing the premise: if sound, therefore rain). Only in the best worlds is the sound of rain due to rain. Auditory perception therefore provides only incomplete knowledge, and the modal is allowed. In other words, the apparent evidential effect of universal epistemic modals is due to the fact that they are indicators of reasoning with nonveridicality and incomplete knowledge. We summarize this in the following: (59) Evidential component of universal epistemic modals: partial knowledge a. Universal epistemic modals can only effectively weaken a proposition p if the speaker’s knowledge that supports p is not complete. b. Complete knowledge is knowledge of all the relevant facts for p. More technically, it is a set of propositions that entails p. c. All other knowledge is partial. The generalizations that we establish here for universal modals are very relevant for the discussion of evidentiality, especially in languages that have indirect evidential morphemes but do not mark direct perception (e.g., Native American languages such as Cheyenne (Murray, to appear), and Turkish and Bulgarian (Smirnova 2013a)). The ‘direct’

chapter three

98

evidential is typically an unmarked past or present, and the marked form is the so- called indirect evidential, which indicates that the source of information is not fi rst-hand knowledge of the speaker. The indirect evidential—like the universal in English, Greek, and Italian—can thus be understood as expressing reduced speaker commitment to p. Direct perception, on the other hand, as we discussed in the case of rain, entails full knowledge. If I see the rain, I know it is raining; hence direct perception gives veridical and complete knowledge. 6 Given partial knowledge, p is entailed only in the support set (the best worlds). Other wise, in the modal base p remains unsettled, as in von Fintel and Gillie’s kernel K. In fact, unsettledness in K is a presupposition for von Fintel and Gillies, just as nonveridicality is for us (60)

Definition 5 (String must + evidentiality) Fix a c-relevant kernel K. a. [[ mustф ]]c,w is defi ned only if K does not directly settle [[ф]]c b. If defi ned, [[ mustф ]]c,w = 1 iff BK ⊆ [[ф]]c

Current classifications of sources of evidence focus on the distinction between direct and indirect knowledge, with visual evidence counting as direct evidence entailing full knowledge, and reprobative and other internal evidence counting as indirect, thus implying incomplete knowledge. Our claim is that every type of source of information comes as either complete or partial. Complete knowledge is a set of propositions that entail p and partial knowledge is a set of proposition that is only compatible with p (see also Mari 2010). Greek and Italian do have a reportative evidential form (lei, ipan, si dice, dicono, pare): FUT can co- occur with it: . (61)

O Janis tha (prepi na) gini kala eipan. the John FUT (must SUBJ) become.3sg well, say.3pl.

(62)

Gianni guarirà, pare. Gianni recover-FUT.3sg, it seems. ‘John must recover, it seems.’

However, the reportative context by itself is not sufficient to trigger universal biased modals. We reproduce here an example from Smirnova (2013b), to show the contrast between the Greek/Italian FUT and the Bulgarian evidential, which is fi ne in this context.

Epistemic Future and Epistemic MUST

(63)

99

Reportative context: you and your sister were out of touch for a couple of years. Today she calls you on the phone to catch up. She tells you that her daughter Maria plays the piano. Later, you tell your husband: a.

Maria svirela na piano. Maria play.EV on piano Bulgarian evidential OK (Smirnova 2013b: (2))

b. # I Maria tha/prepi na pezi piano. #Maria must play the piano. c. # Maria suonerà il piano. #Maria must-FUT.3sg play the piano. The reason the universal modal is blocked in this context is that here the speaker has knowledge p provided by her sister’s utterance. The assertion ‘Maria plays the piano’ is part of the common ground due to the report, so the speaker’s epistemic state is veridical, and it contains no worlds that negate this information. FUT is incompatible with this state, and so is MUST. This example clearly shows that it is not indirectness that matters but whether the speaker knows. If one has full knowledge of p, even if this knowledge is due to something that someone else said, one cannot use an epistemic weakening device, not even a biased one (MUST, future, modal particles). The inferential context, on the other hand, is compatible with the universal modals because it presents partial knowledge. Again we use an example modeled after Smirnova (2013b). You and your sister were out of touch for a couple of years. Today you visit her for the fi rst time. As she shows you around her apartment, you see that there is a piano. Later, you tell your husband: (64)

a. Maria svirela na piano. Maria play on piano.

(Bulgarian evidential)

b. Maria tha pezi/prepi na pezi piano. Maria FUT play/must subjunctive play piano. c. Maria suonerà/deve suonare il piano. Maria play.FUT.3sg/must play the piano. d. Maria must play the piano.

chapter three

100

Here we have a piano, but we do not actually see Maria playing it; hence the knowledge is partial. An inferential context with missing premises is therefore an excellent environment for FUT, MUST, and the other weakening devices. Again, it is not a matter of indirectness, as we see the piano directly; it is simply reasoning with incomplete knowledge so that we can effectively partition the modal base into worlds that support the proposition and those that do not, as is required by epistemic weakening. Now, fi nally, consider FUT/MUST in a mirative use. Such uses have been reported for evidentials, for example, in Gitiksan. If the speaker sees John standing in the doorway, as in , he has complete evidence. Both Greek and Italian futures are banned. (65)

Gitkasan, Peterson (2010:143, ex. (30) = hiwitxw = t John EVID = CND arrive = PND

(66)

#Tha irthe o Janis! #Sarà Gianni!

Again, FUT cannot be used in this context of veridical direct perception because it gives complete veridical knowledge. We are saying that inferential knowledge cannot be used with the biased modal if complete, and in support of this, Mari (2009,forthcoming; see also previous discussion in Pietrandrea, 2005) notes that the italian future cannot be used in the following context of complete premises: (67)

a.

La palla è in A, B, o C. ‘The ball is in A, B or C.’ b. Non è nè in A, nè in B. ‘It is neither in A nor in B.’ c. # Sarà in C. ‘It will be in C.’

Here we have full knowledge of the situation, and c follows as a logical conclusion from the other two premises. Italian FUT is banned from this use, as expected under our analysis. However, Mari also notes that the epistemic modal dovere (must) is indeed acceptable, and so is must: (68)

a. La palla sarà in A, B, o C. b. Non è nè in A, nè in B.

Epistemic Future and Epistemic MUST

101

c. Deve essere in C ‘It must be in C.’ In Greek, both FUT and MUST are licensed in this complete knowledge context. (69)

a. The ball is in A, B, or C. b. The ball is in neither A nor B. c. I bala prepei na ine / tha ine sto C. the ball must subj be.3s / FUT be.3s in.the C ‘The ball must be/ FUT be in C.’

Do these data challenge our idea of incomplete knowledge? We think not, because they do not seem to involve pure epistemic reasoning. They present an “impure” (a term from Knobe and Szabo 2013) use of the universal modal, where epistemic reasoning is mixed with aleithic modality; or they are simply just aleithic because a conclusion is drawn based on the mathematical/logical disjunctive schema: p or q or r, ¬p and ¬q, therefore r. Hence the occurrence of universal modals in these contexts is indeed evidence for strength (as von Fintel and Gillies suggest) but not of strength of the epistemic MUST; rather, they signal aleithic use of MUST, whose strength is indisputable anyway. In support of this being a different use, we want to add that, often, this aleithic use of MUST/ FUT is signaled by focus on the modal—in contrast to the regular epistemic weakening case where MUST is typically unstressed and disallows focus: The ball MUST be in C! as opposed to # John MUST be a doctor, but I am not entirely sure, which is quite odd with the emphatic stress on MUST. In other words, there is indeed evidence, at the suprasegmental level at least, that this use of MUST is distinct from the epistemic one. To conclude, we have proposed a theory of FUT and MUST as biased universal modals. As modals, they are nonveridical (hence weak), both objectively (they do not entail p) and subjectively (they do not entail that the speaker knows that p). Reasoning with universal modals relies on partial knowledge, and they appear to be strong because in the worlds compatible with that knowledge p is true. It turned out that no recourse to a separate evidential component was necessary.

chapter three

102

4

Moore Effects with Biased Modals and Informational Flow

In this last section, we fi ne tune our analysis by examining some effects that appear to be Moore-paradoxical. The literature on the Moore paradox is vast, and we will not attempt a general analysis of it here, since our topic is not the paradox itself. Our new observations are that we fi nd Moore effect with biased modals, but different variants of Moore’s paradox affect the modals in dif ferent ways. To explain the variation, we propose that Moore effects do not necessarily reveal a veridicality confl ict, but manifest also sensitivity to informational flow that previously has escaped attention. The classical Moore paradox itself arises with sentences like (70) and (71): (70)

# It is raining and I don’t know that it is raining.

(71)

# It is raining and I don’t believe it.

The usual reaction is that the sentences above are odd, contradictorysounding, and unassertable. In the literature, the sentences are treated as defective in that they involve the speaker in some kind of epistemic confl ict. As we said in section  2.1, a positive unmodalized assertion is subjectively veridical; that is, the speaker is typically understood as knowing that p is true. If this is so, then in the sentences above the speaker’s epistemic state M(speaker) is presented as both being included in p and allowing ¬p worlds. This is a contradictory epistemic state, and the sentences are defective because of this veridicality confl ict imposed by the two conjuncts. In a more recent discussion Yalcin (2007) coins the term epistemic contradictions for Moore variants with logical forms such as ф and it is not possible that ф. For example: (72)

# It is raining and it might not be raining.

(73)

#It is raining and it is possible that it is not raining.

Yalcin calls these epistemic contradictions. Again the confl ict appears to be between a veridical epistemic state established by the unmodalized fi rst conjunct (where all worlds are raining worlds), and a nonveridical

Epistemic Future and Epistemic MUST

103

state, allowing raining and nonraining worlds in the second conjunct. So, both Yalcin’s examples and the classic Moore paradox examples involve an epistemic confl ict that reveals a subjective veridicality confl ict. Interestingly from our perspective, future and MUST sentences give rise to what appears to be a Moore paradoxical effect. We observe it in (74)–(77). (We replace and with but to make the sentences more natural sounding, but as can be seen, the effect is observed): (74)

# It must be raining, but it might not be raining.

(75)

#Tha vrehi, ala ine pithano na min. FUT rain.present.3sg ala is possible subj not vrehi rain.present.3sg ‘#It must be raining, but it is possible that it it might not be raining.’ (epistemic)

(76)

Gianni sarà arrivato, #ma è possibile che non sia John be.3sg.fut arrived, but is possible that not be.subj.3sg arrivato. arrived. ‘John must have arrived, but it is possible that he has not arrived.’ (epistemic)

(77)

#Prepi na vrehi, ala ine pithano na min must SUBJ rain.present.3sg ala is possible subj not vrehi. rain.present.3sg ‘#It must be raining, but it is possible that it it might not be raining.’ (epistemic)

If biased modals are nonveridical, as we argue, and convey a partitioned state, how can the data above be explained? If the nonveridical spaces associated with FUT and epistemic MUST are nonveridical (thus allowing ¬p worlds), as we are arguing, why are they not compatible with a continuation that raises that possibility? As we proceed to show how the Moore effect can be explained in our account, we want to recall first the new set of data, of similar structure,

chapter three

104

that we mentioned earlier and that support the nonveridical analysis. Recall that we used them as evidence for it. Here is a sample of the sentences with but I am not entirely sure. (78)

a. I Ariadne tha ine arosti, ala dhen ime ke endelos the Ariadne FUT is sick, but not am and completely sigouri. sure b. Giacomo sarà malato, ma non sono completamente Giacomo be.3sg. fut sick, but not be.1sg entirely sicura. certain. ‘Ariadne/Giacomo must be sick, but I am not entirely sure.’

(79)

a. I Ariadne prepi na troi tora, alla den ime the Ariadne must subj eat.present3sg now, but not be.1sg ke endelos. sigouri and completely sure. b. Giacomo deve Giacomo must completamene completely

star mangiando, ma non sono be eat-gerund, but not be.1sg sicura. sure

‘Giacomo/Ariadne must/will be eating now, but I am not entirely sure.’ With continuations like but I am not entirely sure, the Moore effect seems to be removed. Crucially, the effect remains with an unmodalized veridical sentence: (80)

I Ariadne ine/itan arosti, #ala dhen ime ke endelos the Ariadne is/was sick, but not be.1sg and completely sigouri. sure ‘Ariadne is/was sick, but I am not entirely sure.’

Epistemic Future and Epistemic MUST

(81)

105

Giacomo è/era malato, #ma non sono Giacomo is/was sick, but not be.1sg completamente sicura. completely certain. ‘Giacomo is/was sick, #but I am not entirely sure.’

The present and past sentences are subjectively and objectively veridical. When the speaker utters them, as far as she knows, Ariadne and Giacomo were sick, and knowledge of that cannot be canceled by a continuation that questions it. All worlds in M(speaker) are worlds in which Ariadne and Giacomo are sick, so the second conjunct induces epistemic contradiction that comes from this veridicality confl ict (all worlds in M(speaker) are p worlds in the fi rst conjunct, while not all worlds in the same space are p worlds in the second conjunct). The epistemic future and MUST, on the other hand, are fi ne with but I am not entirely sure. Why? Because both conjuncts are not veridical, and they are both of equal informational strength, so there is no confl ict between them. The fi rst conjunct establishes a nonveridical modal base that allows ¬p worlds—i.e., the worlds that are not Best. In the second conjunct, we move from the modal base, to the larger space—i.e., the speaker’s epistemic state. The speaker is in a nonveridical epistemic state: I am not entirely sure that p is equivalent to I am committed to p but not fully, which means that my epistemic state is also biased toward p but allows ¬p worlds. The presence of entirely is crucial in the sentence as it reveals the bias (as opposed to I am not sure, which is a neutral sentence with no bias toward p. Hence the two conjuncts make reference to nonveridical modal spaces that are in agreement and not in confl ict, since they are both nonveridical and biased. We call this situation informational harmony. We do not get a Moore-paradox with informational harmony, unlike with the positive unmodalized assertion where I am not entirely sure that p creates exactly the kind of contradiction the classical Moore continuation gives (a contradictory epistemic state). Hence, I am not entirely sure fully supports our nonveridical treatment of epistemic FUT and MUST. What we just said relies on the notion of informational strength. We said that the two sentences are of equal informational weight. But in the classical Moore cases and in our FUT/MUST variants of them, the sentences do  not have the same informational weight. This creates informational

chapter three

106

conflict that manifests itself in two ways: (a) as breakdown of information flow, which normally proceeds from weaker to stronger (as we define it below), and (b) as an informational contradiction. The Moore effects with FUT and MUST are due to these conflicts. Consider first how the classical case illustrates breakdown of information flow: (82)

# It is raining and/but it might not be raining.

The first conjunct It is raining presents the rain as an actual fact (objective veridicality), or as knowledge of the speaker of this fact (subjective veridicality) that is, in all worlds compatible with the speaker’s knowledge it is raining. The second conjunct conveys weaker information: the speaker considers it possible that it is not raining. This discourse makes the hearer conclude that the speaker is not being cooperative. She said something false in either the first or the second conjunct, in both cases violating quality and thus being misleading. We end up with a confl ict, as well as a breakdown of what can be thought of as normal information flow. Proceeding from weaker to stronger is the normal course of information flow, as expected by Gricean pragmatics: (83)

Normalcy conditions on information flow Information flow is considered normal iff: (i) Information goes from weaker information A to stronger information B. Or, (ii) A and B do not informationally contradict each other.

These conditions are nothing extraordinary, but are a mere summary of run- of-the-mill versions of Gricean views of how information normally proceeds. Weaker and stronger are the informational alternatives compared; that is, they are the propositions denoted by the sentences. The problem, crucially, in the second conjunct, comes from the fact that the stronger information was established fi rst: a veridical sentence is informationally stronger than a nonveridical sentence. And within nonveridical sentences, S with bias is stronger that S with equilibrium. Recall that the ordering source reveals bias. With equilibrium, there is no ordering source. With ordering sources the domain of quantification is more restricted and is thus informationally richer (à la Stalnaker). Below we give the relevant scale:

Epistemic Future and Epistemic MUST

(84)

107

Informational strength ordering relevant for Moore’s contrasts weaker 〈might ¬p, MUSTp, p〉 stronger

Let us represent Moore’s sentences S as a pair of alternatives 〈S1, S2〉: (85)

Moore’s variant: p and might ¬p Alternatives: 〈S1 :p, S2 :might¬p〉

S1 is a stronger alternative than S2 because the veridical epistemic state is not partitioned: all worlds are p worlds. S2 says something that is both informationally weaker and in veridicality confl ict with S1. The reverse order, from weaker to stronger, is predicted to be fi ne, and this prediction is borne out: (86)

It might not be raining, but in fact it is raining.

Here, conversation proceeds normally, because the weaker S2 precedes S1 (and we added in fact to help the sentences connect; note that in fact has no effect on the other order: It is raining but in fact it might not be raining remains odd). In the order above, the second Phrase seems to correct the fi rst one, to strengthen it; and because information proceeds normally, the difference in veridicality is in harmony with the strengthening. Strengthening always seems to serve a discourse function, but it will lead us too far astray to develop this in more detail (see Geurts 2010 for a neo- Gricean pragmatic theory that addresses informational strengthening in detail, while also arguing that it does not always serve the same function.) For now, suffice it to raise awareness that the ill-formedness of the classic Moore sequence reveals both a veridicality confl ict and a violation of informational normalcy. If the latter gets fi xed, the difference in veridicality becomes innocuous.7 To go now to our FUT/MUST sentences, consider fi rst the case of a negated possibility modal in the second conjunct: (88)

a. #Tha vrehi, ala ine pithano na min FUT rain.present.3sg but is possible subj not vrehi. rain.present.3sg

chapter three

108

b. #Starà piovendo, ma è possibile che non Stay.3sg.fut rain.gerund, but is possible that not piova. rain.3sg.subj. ‘#It must be raining, but it might not be raining.’ (89)

#Prepi na vrehi, ala ine pithano na min must SUBJ rain.present.3sg ala is possible subj not vrehi. rain.present.3sg ‘#It must be raining, but it is possible that it it might not be raining.’

In the cases where a universal modal is followed by the negation of a possibility modal, we have utterances of modalized sentences in both conjuncts. But we do not have the same informational weight in both conjuncts. Must p conveys bias toward p worlds, while the possibility utterance conveys equilibrium; hence S1 is the stronger alternative, while S2 is informationally weaker: we go from a stronger S1 (with ordering sources revealing bias) to a weaker S2 (with no ordering sources and equilibrium): (90)

Moore’s variant: MUST p and might ¬p Alternatives: 〈S1 :MUST p, S2 :might ¬p〉

According to the normalcy condition, information flow requires the stronger statement to be second. The information flow in the Moore sentence is thus not normal, and the sentence is defective for this reason. Notice, crucially, that if we reverse the order as we do below, the sequence is improved and shows no Moore effect: (91)

a. Ine pithano na min vrexi, alla malon tha is possible subj not rain.present.3sg, but probably FUT vrexi. rain.present.3sg. ‘It is possible that is not raining, but most likely it must be raining.’

Epistemic Future and Epistemic MUST

109

b. È possibile che non stia piovendo, ma Is possible that stay.3sg. subj not rain.gerund, but starà piovendo, vedrai. stay.3sg.fut rain. gerund, see.2sg. ‘It might not be raining, but I am pretty certain it is raining, you’ll see.’ These discourses are normal because the stronger sentence follows the weaker one. This fact illustrates that the problem with the Moore sentences with strong modals in the fi rst conjunct followed by negations of weaker modals are due to a breakdown of normal information flow. The two alternatives are not epistemically inconsistent but are informationally not-normal. Consider, fi nally, the continuations with I don’t believe it: (92)

Gianni sarà arrivato, #ma non lo credo. John be.3sg.fut arrived, but not that believe.1sg. ‘#John must have arrived, but I do not believe it.’

Here, the alternatives are: (93)

Moore’s variant: MUST p and I do not believe that p Alternatives: 〈S1 :MUST p, S2 : I do not believe that p〉

The speaker establishes bias toward p with the use of a universal modal in the fi rst conjunct. She continues by saying that she does not believe that John arrived. Importantly, believe is a neg-raising verb, so not believe that p typically strengthens believe that not p; (see Horn 1979 for a classical piece on neg-raising with belief verbs). In the strengthened reading, both conjuncts appear to be informationally equally strong, but, crucially, in confl ict with each other: the first conjunct conveys bias toward the p set (John arrived) and the second alternative strengthens the belief that John did not arrive (there are no p worlds in the speaker’s epistemic state, counter- commitment to p). This creates a conflict in the joint utterance because the speaker is required to have both bias toward p and counter- commitment to it. This is an informational ‘contradiction’, a conflict that cannot be repaired. As we see, change of order has no effect:

chapter three

110

(94)

a. Den not tha FUT

pistevo oti irthe o Janis, #alla believe.1sg that FUT.came.3sg the John, but irthe. came.3sg.

b. Non credo che John sia arrivato, #ma Not believe.1sg.pres that John be.3sg.subj arrived, but sarà arrivato. be.3sg.fut arrived. ‘#I do not believe that John arrived, but he must have arrived.’ Here the speaker remains in an informationally contradicting state where she is required to both have bias toward p and believe p to be false. This is an impossible informational state. Notice that in sentence; (95) without negation in the second conjunct, we have again informational harmony (though perhaps also a bit of redundancy): (95)

It must be raining, and I believe that it is raining.

We close our discussion here by summarizing the three cases of Moore continuations for FUT/MUSTp that we found: (a) A continuation that creates no effect, revealed by I am not entirely sure; this continuation illustrates informational harmony. (b) A continuation that violates informational normality from weaker to stronger information (with might not p); this effect can be fi xed by reversing the order of conjuncts. (c) A continuation that creates informational contradiction (with I do not believe p). Order has no effect on this one.

Certainly further study is required to understand better the interactions between modals in Moore sentences, and to refi ne how the notion of informational strength we proposed interacts with the semantics. Here we offered but a few initial, and we hope helpful, observations about the behavior of FUT/MUST in Moore-like sentences.

Epistemic Future and Epistemic MUST

5

111

Conclusion

In this chapter, we identiy three new categories. The fi rst is the category epistemic future. Epistemic future is equivalent to the epistemic modal verb MUST, and this fact is evidence that the grammatical category ‘future’ cannot be a purely temporal one, in agreement with a substantial amount of recent work on Dutch, Greek, and Italian futures. We offered an analysis of epistemic future as a universal epistemic modal identical to MUST, and showed that in Greek and Italian the two can even co- occur. The second category is the biased modal. The epistemic future and MUST are biased modals, we argued, and biased modals are both strong and weak. They are strong because they entail p in the set of Best worlds (the support set), but they are weak because their modal base is nonveridical and therefore contains non-p worlds too. Biased modals express weaker epistemic commitment of the speaker to the propositional content and function generally as epistemic weakeners. The category of biased modals also includes, besides FUT and MUST, modal particles such as wohl and wel in German and Dutch. Though we did not offer an explicit analysis, we want to suggest that certain indirect evidentials, in languages exhibiting a binary opposition between evidential and unmarked, are also epistemic weakeners. The function of epistemic weakening is to create a nonveridical epistemic space—that is, a partitioned space with at least one non-p world. The third category, then, is the category of (non)veridicality. Veridicality is a property of natural language expressions entailing the truth of their complement (objective veridicality) or knowledge of an individual of the truth of a proposition (epistemic, or subjective, veridicality). Veridicality and nonveridicality are fundamental in characterizing the truth conditions of modal expressions (modal verbs and particles), but also propositional attitudes and mood choice (as is further discussed by Giannakidou in this volume and in earlier work). The licensing of negative polarity items also seems to be sensitive to nonveridicality. We developed a framework for (non)veridicality of modal expressions that is flexible enough to allow analysis of new phenomena within a number of limited and well-understood premises. Finally, concerning the evidential component of MUST, we suggested that, contrary to what has been claimed in the literature about indirectness, the key is partial knowledge. Partial knowledge represents a partitioned, nonhomogeneous, and therefore nonveridical epistemic

112

chapter three

state, and explains nicely why the use of universal epistemic modals is constrained to such states. We also showed that visual perception of an event is privileged because it leads to complete knowledge, and therefore FUT and MUST, being nonveridical, cannot be used with direct visual perception. In other words, what much of the literature characterizes as the evidential component of MUST is simply an epiphenomenon, reflecting the nonveridical nature of MUST and its incompatibility with full knowledge. We presented this material at the workshop on categories in Wroclaw, at the International Congress of Linguists in Geneva, and in colloquia at the Universities of Brussels, Groningen, and Chicago, as well as Northwestern University. We thank the audiences for their insightful feedback, and the organizers of the workshop for giving us the opportunity to develop these ideas. We are also very grateful for comments and helpful discussion to Kai von Fintel, Itamar Francez, Jack Hoeksema, Chris Kennedy, Mikhail Kissine, Jason Merchant, and Malte Willer, as well as the anonymous reviewers of this volume, for their help and suggestions. Alda Mari’s research was funded by the ANR-10-LABX-0087 IEC and ANR-10-IDEX-0001-02 PSL. She also gratefully thanks the CNRS-SMI 2015.

Notes 1. It must also be noted that there is considerable variability in judgments among native English speakers, and some accept purely epistemic readings of will, as in the Greek and Italian (6) and (7). We have also encountered speakers that accept epistemic will with past adverbs. It remains true, however, that although the Greek and Italian epistemic futures are unexceptional and widely attested, purely epistemic will is harder to fi nd, and its existence has been contested in the literature (see, e.g., Copley 2002). Another relevant fact is that in English there is competition between will and must, but in Greek and Italian, MUST equivalents can co- occur with FUT (Giannakidou 2012; Giannakidou and Mari 2013). 2. In previous works we argued that the epistemic interpretation holds with both statives and eventives (and more precisely with perfective non-past in Greek), an issue that we do not discuss here. 3. Individual anchoring of truth should be seen on a par with other kinds of anchoring of propositional content, such as temporal anchoring or event anchoring (Hacquard 2010). The individual anchor is a parameter of evaluation similar to Lasersohn’s (2005) judge. In embedded sentences, the main clause subject is also a potential anchor, and this has repercussions for mood, as shown in Giannakidou’s work.

Epistemic Future and Epistemic MUST

113

4. The difference between knowledge and belief is not so impor tant for our purposes here, and in many other cases, e.g., for mood choice, it does not matter either, as because verbs of knowledge and belief both select the indicative in many languages. Belief makes a difference for an agent typically when it is contrasted with knowledge—i.e. when the agent is aware that she does not have enough information to support a proposition. In this case, we can say that we have semantic narrowing (Geurts and van Tiel 2013). 5. In a recent work, Mari (2014b) argues that some other modals are veridical with respect to a restricted domain that contains the actual world, and nonveridical with respect to an outer domain. This, she argues, is what happens with past modals with present orientation, triggering the veridicality (a.k.a. actuality entailments). What is common to MUST/FUT and these modals is that the modal domain is divided into two subdomains, with the inner one being veridical and the outer one being nonveridical. 6. Note, at the same time, that direct access in the sense of Willett (1988) does not count per se as a case of complete knowledge. There is a difference between ‘seeing some facts that are compatible with p being true’ and ‘knowing that p is true in virtue of visual evidence’. As Lee 2012 explains, witnessing p is not equivalent to have visual evidence for p. Likewise, visual evidence can be incomplete evidence for assessing the truth of p (in (57), I see the umbrella, but I do not see the rain). Similarly, inferential knowledge can be complete or incomplete: it is complete if all the premises allowing to conclude that p is true are given and incomplete if there are some missing premises for concluding that p is true. 7. We think it is worth mentioning that the normalcy conditions we posit here are not specific to modality, but are general. Consider, e.g., quantifiers: (87)

a. # Every linguistics student came to the Halloween party, and some linguistics students came to the party. b. Some linguistics students came to the Halloween party; in fact, every linguistics student came to the party.

The odd sequence is not normal because it proceeds from strong (with universal quantification) to weak (with existential quantification). Here violation of normalcy leads to redundancy.

References Bertinetto, P. M. 1979. Alcune ipotesi sul nostro futuro (con alcune osservazioni su potere e dovere). Rivista di Grammatica Generativa 4: 77–138. Broekhuis, H., and H. Verkuyl. Forthcoming. Binary tense and modality. Natural Language and Linguistic Theory #32(3): 973–1009. Ciardelli,  I.,  J. Groenendijk, and  F. Roelofsen. 2013. Inquistitive semantics: A new notion of meaning. Language and Linguistic Compass 7: 459–476.

114

chapter three

Chiou, Michael. 2014. What is the future of Greek? Towards a pragmatic analysis. Research in Language 12(4). DOI: O.ISIS/rela-2015-0004. Condoravdi, C. 2002. Temporal interpretation of modals: Modals for the present and for the past. In The Construction of Meaning, ed. D. Beaver, Luis D. Cassillas Maritinez, Brady Z. Clark, and S. Kaufmann, 59–88. ———. 2003. Moods and modalities for will and would. Invited communication at Amsterdam Colloquium. Copley, B. 2002. The semantics of the future. PhD diss., Massachusetts Instiute of Technology. de Marneffe, M., Christopher D. Manning, and Christopher Potts. 2012. Did it happen? The pragmatic complexity of veridieaity assessment. Computational Linguistics 38(2): 301–333. Giannakidou, A. 1997. The landscape of polarity items. PhD diss., University of Groningen. ———. 1998. Polarity Sensitivity as (Non)veridical Dependency. Amsterdam: John Benjamins. ———. 1999. Affective dependencies. Linguistics and Philosophy 22: 367–421. ———. 2002. Licensing and sensitivity in polarity items: From downward entailment to nonveridicality. In CLS 38: Papers from the 38th Annual Meeting of the Chicago Linguistic Society, ed., Maria Andronis, Anne Pycha and Keiko Yoshimura. ———. 2009. The dependency of the subjunctive revisited: Temporal semantics and polarity. Lingua 120: 1883–1908. ———. 2012. The Greek future as an epistemic modal. In Proceedings of ICGL 10. Komotini. ———. 2013a. (Non)veridicality, epistemic weakening, and event actualization: Evaluative subjunctive in relative clauses, In Nonveridicality and Evaluation: Theoretical, Computational, and Corpus Approaches, ed. R. Trnavac and M. Taboada, Emerald, Studies in Pragmatics. ———. 2013b. Inquisitive assertions and nonveridicality. In The Dynamic, Inquisitive, and Visionary Life of ф, ?ф and Possibly ф: A Festschrift for Jeroen Groenendijk, Martin Stokhof, and Frank Veltman, ed. Maria Aloni, Michael Franke, and  F. Roelofsen, 115–126. ILLC Publications, University of Amesterdam ———. 2014. The futurity of the present and the modality of the future: A commentary on Broekhuis and Verkuyl. Natural Language and Linguistic Theory, 1011–1032. Giannakidou,  A., and  A. Mari. 2012a. Italian and Greek futures as evidential operators. To appear in Proceedings CLS 48. ———. 2012b. The future of Greek and Italian: An epistemic analysis. Proceedings of Sinn und Bedeutung 17: 255–270. http://semanticsarchive.net /Archive / Dk3NGEwY/GiannakidouMari.pdf.

Epistemic Future and Epistemic MUST

115

Giannakidou, A., and A. Mari. 2013. A two- dimensional analysis of the future: Modal adverbs and speaker’s bias. Proceedings of the Amsterdam Colloquium 2013, 115–122. ———. 2015. The future as epistemic modality: The view from Greek and Italian. Manuscript University of Chicago and Institut Jean Nicod. Geurts,  B. 2010. Quantity Implicatures. Cambridge: Cambridge University Press. Geurts, B., and B. Tiel. 2013. Embedded scalars. In Semantics and Pragmatics 6, Vol. 9: 1–37. Hacquard,  V. 2006. Aspects of modality. PhD diss., Masschusetts Institute of Technology. ———. 2011. Modality. In Semantics: An International Handbook of Natural Language Meaning, ed.  C. Maienborn,  K. von Heusinger, and  P. Portner, 1484–1515. Berlin: Mouton de Gruyter. Huitink, J. 2012. Modal concord. A case study in Dutch. Journal of Semantics 29(3): 403–437. ———. 2014. Modal concord. Forthcoming in to The Blackwell Companion to Semantics, ed. Lisa Matthewson, CŽcile Meier, Hotze Rullman, and Thomas Ede Zimmermann. Karttunen, L. 1971. Some observations on factivity. Papers in Linguistics 4(1): 55–69. ———. 1972. Possible and must. Syntax and Semantics, vol. 1, 1–20  J. Kimball. New York, Academic Press. Kaufmann,  S. 2005. Conditional Thruth and Future Reference. Journal of Semantics. 22(3): 231–280. Kaufmann,  S.,  C. Condoravdi, and  V. Harizanov. 2006. Formal approaches to modality. In The Expression of Modality, ed.  W. Frawley, 71–106. Berlin: Mouton de Gruyter. Kiparksy, P., and C. Kiparsky. 1970. Fact. In Progress in Linguistics, ed. M. Bierwich and K. Heilolph, 143–173. Berlin: Mouton de Gruyter. Kissine,  M. 2008. From predictions to promises. Pragmatics and Cognition 16:169–189. Kratzer,  A. 1981. The Notional Category of Modality. In Words, Worlds, and Contexts. New Approaches in Word Semantics, ed.  H.  J. Eikmeyer and  H. Rieser, 38–74. Berlin: Mouton de Gruyter. ———. 1991. Modality. In Semantics: An International Handbook of Contemporary Research, ed. A. von Stechow and D. Wunderlich, 639–650. Berlin: Mouton de Gruyter. Lassiter, D. 2011. Measurement and Modality: The Scalar Basis for Modal Semantics. PhD diss., New York University. ———. 2013. The weakness of must: In defense of a mantra. CSLI Workshop Perspectives on Modality. Stanford.

116

chapter three

Mari, A. 2009. Disambiguating the Italian Future. In Proceedings of Generative Lexicon, 209–216. ———. 2010. On the evidential nature of the Italian future. Manuscript IJN http:// jeannicod.ccsd.cnrs.fr/docs/00/67/85/49/ PDF/ Evidential _ Future_ Italian.pdf. ———. 2014a. Each Other, asymmetry and reasonable futures. Journal of Semantics, 31(2): 209–261. ———. 2014b. Actuality Entailments: broadening the space of the possibilities. Ms. Institut Jean Nicod and Talk at Linguistics Colloquium University of Chicago, Dec. 2014. ———. Forthcoming. Modalités et Temps. Bern: Peter Lang AG. McCready, E and N. Ogata. 2007. Evidentiality, modality, and probability. Linguistics and Philosophy 30:147–206. Montague, R. 1969. On the nature of certain philosophical entities. The Monist 53: 159–194. de Marneffe,  M.,  C. Manning, and  C. Potts. 2012. Did it happen? The pragmatic complexity of veridicality Assessment. Computational Linguistics 38: 301–333. Matthewson, L. 2010. Crosslinguistic variation in modality systems: The role of mood. Semantics and Pragmatics 3: 1–74. Matthewson,  L.,  H., Rullmann, and  H. Davis. 2007. Evidentials as epistemic modals: Evidence from St’at’imcets. The Linguistic Variation Yearbook 6: 161–201. Murray, S. Forthcoming. Evidentials and Illocutionary Mood in Cheyenne. International Journal of American Linguistics. Palmer, F. R. 1987. Mood and Modality. Cambridge: Cambridge University press. Pietrandrea, P. 2005. Epistemic modality: Functional properties and the Italian system. Amsterdam: John Benjamins. Portner, Paul. 1998. The progressive in modal semantics. Language 74(4): 760–87. ———. 2009. Modality. Oxford: Oxford University Press. Prior, A. 1957. Time and Modality. Oxford: Oxford University Press. Smirnova,  A. 2013a. The meaning of the Bulgarian and Turkish evidentials. Contrastive Linguistics 2/3: 205–223. ———. 2013b. Evidentiality in Bulgarian: Temporality, epistemic modality, and information source. Journal of Semantics 30(4): 479–532. Squartini,  M. 2004. Disentangling evidentiality and epistemic modality in Romance. Lingua 114: 873–895. ———. 2012. Evidentiality in interaction: The concessive use of the Italian future between grammar and discourse. Journal of Pragmatics 44(15): 2116–2128. Thomason, R. 1984. Combination of tense and modality. In Handbook of Philosophical Logic, ed. D. Gabbay and F. Guenthner, 135–165. Dordrecht: Reidel. Trnavac , R., and M. Taboada. 2012. The contribution of nonveridical rhetorical relations to evaluation in discourse. In Language Sciences 34: 301–318. Tsangalidis, A. 1998. Will and tha: A Comparative Study of the Category Future. Thessaloniki, Grecee: University Studio Press.

Epistemic Future and Epistemic MUST

117

von Fintel, K., and A. Gillies. 2010. Must . . . stay . . . strong ! Natural Language Semantics eighteen: 351–383. von Fintel, K., and I. Heim, 2007. Intensionality. Ms., Massachusetts Institute of Technology. von Fintel, K., and S. Iatridou, 2006. How to say ought in foreign: The composition of weak necessity modals. Ms., Massachusetts Institute of Technology. Willett, T. 1988. A cross-linguistic survey of the grammaticization of evidentiality. Studies in Language 12: 51–97. Yalcin, S. 2007. Epistemic modals. Mind 116, 464: 983–1026. Zimmermann,  M. 2011. Discourse particles. In Semantics: An International Handbook of Natural Language Meaning, ed. C. Maienborn, K. von Housinger, and P. Portner, 2011–2038. Berlin: Mouton de Gruyter. Zwarts, F. 1995. Nonveridical contexts. Linguistic Analysis 25: 286–312.

chapter four

On Finiteness and the Left Periphery: Focusing on Subjunctive Manuela Ambar

1

Introduction

T

he study of subjunctive has a long tradition. Over the past decades numerous works have contributed significantly to improving our understanding of this mood; yet aspects of its syntax remain intriguing and challenging for the theory of grammar. As the traditional labels “subjunctive” and “conjunctive” (the current term in the Portuguese grammatical tradition) suggest, dependency has been seen as the hallmark property of subjunctive. As a corollary it is generally considered the mood of subordination par excellence, being selected by given classes of matrix predicates (2b) and not occurring freely in matrix clauses (1b), as opposed to indicative, which is the independent mood of unembedded assertions (1a): (1)

a. Ele he

vai ao cinema. go.ind.prs.3sg to.the movies

‘He goes to the movies.’ b. *Ele vá ao cinema. he go.sbjv.prs.3sg to.the movies (2)

a. O João disse que ele vai the John say.ind.pst.pfv. 3sg that he go.ind.prs. 3sg

chapter four

126

ao cinema to.the movies ‘John said that he goes to the movies.’ b. O João quer the John want.ind.prs.3sg ao cinema. to.the movies

que ele vá that he go.sbjv.prs.3sg

‘John wants him to go to the movies.’ The idea that the subjunctive is dependent goes back to traditional grammarians. Beyond the common dichotomy realis (indicative) vs. irrealis (subjunctive), whose empirical problems were described (Farkas 1992; Giannakidou 1998), other views were explored. In his philosophical grammar of Portuguese, Soares-Barbosa (1822) observed that the subjunctive occurs dependent on a verb visible or hidden,1 recalling what in the early generative grammar was known as Ross’s (1970) performative approach. Inspired by those works, Faria (1974) tried to defi ne the properties of such overt or covert verbs in Portuguese through the following properties: [±declarative, ±communication, ±thought, ±factive, ±emotive]. After establishing different combinations among the features in this set, Faria concluded that, plausibly, there are no true classes of predicates selecting subjunctive, 2 a conclusion certainly motivated by the difficulty of identifying the unifying property shared by all its uses, which appear to be more chaotic than the use of indicative. Probably it was this difficulty that also led researchers to take the subjunctive as the unmarked mood, or the default mood (Portner 1997; Marques 2004, 2009; Schlenker 2005). According to Marques (2004, p. 99), “indicative is the strong, marked mood, the mood associated with a particular modal value, while subjunctive, associated with a richer variety of modalities, becomes the default mood, occurring in those contexts that do not require the presence of indicative.” Note, however, that the most common view took the subjunctive as marked and the indicative as unmarked. Clearly the concept of markedness would have to be defi ned. But whatever its precise defi nition, when we turn to the syntactic encoding of subjunctive and indicative we have to have a calculus of the properties of their triggers: we have to know at least in what positions they are licensed, what properties are represented in the

On Finiteness and the Left Periphery

127

syntactic structure to be read by the semantic interface, how licensing is done. Subjunctive has to be seen in its relation to indicative. If indicative is the marked option, associated with one modality, the question is why the subjunctive is not associated with that modality and how the apparent default behavior is derived in a system where optionality is not a choice. Although it is commonly assumed that indicative is not subject to licensing, it is plausible that both moods have to be licensed. I argue for this hypothesis here. Moreover, I assume that the subjunctive, like any other mood, involves licensing of tense and that different projections of the left periphery, relating the propositional content to the discourse, do the rest. The difficulty of detecting the unifying property may then be the effect of the interaction of different values provided by discourse projections. Nonveridicality as defi ned by Giannakidou seems, however, to be the semantic unifying property; here I investigate how the syntactic structure can derive it. Section  2 deals with the characterization of subjunctive: distribution and selection, dependency of tense, and obviation. Indicative and subjunctive have their own properties; both require licensing, with root subjunctives being particularly revealing. The volitional-epistemic divide extends to main clauses. Section  3 proposes an analysis of the tense of subjunctive through valuation of t(ense)-features in the probe-goal system. A distinction between tev-features (event-related) and tt-features (morphology-related) solves the paradox created by the need for the tense of subjunctive to be dependent and independent. Section 4 considers concepts discussed in the literature—veridicality, worlds and logophoricity, models of evaluation and anchors—that support my proposal. Section 5, in line with insightful works highlighting the role of evaluation in the account of subjunctive, introduces EvaluativeP and AssertiveP, which I also propose to account for other phenomena. Roughly, EvaluativeP is the projection of subjunctive, AssertiveP the projection of indicative. Tense valuation is also needed. Section 6 deals with other puzzles whose derivation involves those projections, needed as a window to the world, beyond tense. Section 7 presents some concluding remarks.

2

The Characterization of Subjunctive

A challenging point is how semantic properties are encoded into the syntactic structure. The main research on subjunctives has inevitably

chapter four

128

focused on their distribution in subordinate contexts (see section 2.1.1), where both semantic and syntactic properties are involved. In this section I highlight the relevance of matrix subjunctive clauses in the characterization of subjunctive (2.1.2) and the correlation of main- embedded properties (2.1.3).

2.1

Distribution and Selection: Main Patterns in Portuguese

In Portuguese the subjunctive can occur in both subordinate and root contexts. 3 I concentrate on complement and matrix clauses. 2.1.1 COMPLEMENT CLAUSES. The core indicative-subjunctive divide is established between assertive, epistemic predicates (such as ‘say’ and ‘believe’) and volitional, directive, or emotive factive predicates (such as ‘want’, ‘order’, and ‘regret’), as in most Romance languages.

A. Subjunctive. Volitional verbs such as querer ‘want’ (see (2b)) require subjunctive and exclude indicative; likewise, directive and emotive factive predicates take subjunctive and do not accept indicative, as illustrated in (3) and (4) respectively: (3)

O Pedro ordenou/pediu que eles the Peter ordered/asked that they se calassem / *se calaram. be.sbjv.pst.3pl /be.ind.pst.3pl quiet. ‘Peter ordered/asked them to be quiet.’

(4)

A Madalena lamenta que o fi lho the Madeleine regrets that the son chegue/ *chega tarde. arrive.sbjv.prs.3sg/arrive.ind.prs.3sg late. ‘Madeleine regrets that her son arrived late.’

B. Indicative. Inversely, assertive and epistemic predicates generally occur with indicative (adjectival predicative structures follow the pattern of the corresponding verbs): (5)

A Teresa afi rma que ele é the Teresa asserts that he be.ind.prs.3sg

On Finiteness and the Left Periphery

129

simpático /*seja simpático. kind/be.sbjv.prs.3sg kind ‘Teresa asserts that he is kind.’ (6)

O António sabe que a Maria the António knows that the Mary vem /*venha à festa. come.ind.prs.3sg /come.sbjv.prs.3sg to the party ‘Antonio knows that Mary is coming to the party.’

(7)

O Pedro acha que a Ana foi / *fosse the Peter thinks that the Ana go.ind.pst.pfv3sg /go.sbjv.pst.3sg ao teatro. to-the theater ‘Peter thinks that Ana went to the theater.’

C. Indicative- Subjunctive. Given predicates in group B can take either indicative or subjunctive: acreditar ‘believe’, imaginar ‘imagine’, presumir ‘presume’, admitir ‘admit’ (among others): (8)

Acredito que ele tem / tenha razão. believe. ind.prs.1sg that he be.ind.prs.3sg/ be.sbjv.prs.3sg right ‘I believe that he is right.’

(9)

Imagino que queres / queiras Imagine.ind.prs.1sg that want.ind.prs.2sg /want.sbjv.prs.2sg descansar. to rest ‘I imagine that you want to rest.’

(10)

Presumo que ele presume.ind.prs.1sg that he conhece/conheça a situação. know.ind.prs.3sg/know.sbjv.prs.3sg the situation. ‘I assume he knows the situation.’

chapter four

130

(11)

Admito / suponho que ele admit/suppose.ind.prs.1sg that he vem / venha come.ind.prs.3sg /come.sbjv.prs.3sg ‘I admit / suppose that he is coming.’

And others raise doubts. The oddity of (12b) disappears if an overt modal is introduced (12c): (12) a.

Penso que ele vem. think.ind.prs.1sg that he come.ind.prs.3sg ‘I think that he is coming.’

b.

?? Penso que ele venha. think.ind.prs.1sg that he come.sbjv.prs.3sg

c.

Penso que ele possa think.ind.prs.3sg that he may.sbjv.prs.3sg

aparecer. appear

‘I think he may appear.’ The indicative-subjunctive patterns in A– C raise the question of what is selection; if it is the obligatory requirement a head imposes on its complement for a given property, then verbs in group A obligatorily select subjunctive in Portuguese, but verbs in group C, taking either indicative or subjunctive, by defi nition do not. Two well-known behaviors of subjunctive in this sense are tense and obviation (see section 2.2). In turn, if subjunctive under verbs in group C (or group B) is not selected, then licensing might be of the same type as in root clauses, weakening the matrix- embedded divide (see Torrego and Uriagereka 1992). Though the properties of subjunctive main clauses have been subject to less systematic scrutiny, they raise relevant questions about the characterization of subjunctive. The ban on subjunctive in root clauses (1b) has contributed to notions such as marked vs. unmarked mood and to the idea that the subjunctive is somehow defective or dependent. Sentences like (1a) are assertions, where the speaker reports his knowledge about a given state of affairs based on common ground (shared

2.1.2 MAIN CLAUSES.

On Finiteness and the Left Periphery

131

knowledge). The subjunctive is ruled out in this context. However if we turn to sentences expressing an attitude of the speaker, or an evaluation of the state of affairs described, the subjunctive occurs on its own, and with the intended meaning it is the indicative that cannot occur in these matrix clauses: (13)

a. Vá ele às aulas! (e /assim terá êxito Go.sbjv.prs.3sg he to-the courses (and/so he will succeed nos seus estudos)4 in his studies) ‘Let him take the courses!’ a’. Assim ele vá às aulas! so he go.sbjv.prs.3sg to.the courses ‘Let him take the courses!’ b. # Vai ele às aulas! Go.ind.prs.3sg he to.the courses ‘He takes the courses!’ b’. # Assim ele vai às aulas! so he go.ind.prs.3sg to the courses ‘So he takes the courses!’

In (13a) and (13a’) the speaker wants him to attend the courses and believes doing so is enough for him to succeed in his studies; she wishes the unrealized event to become a fact, to belong to the common ground. This meaning does not obtain in (13b and b’). Thus one could assume that it is the indicative that does not occur on its own in this context. The marked or unmarked status of mood becomes vacuous; rather it seems that each mood has its own specific properties: roughly, the subjunctive is the mood of evaluation; the indicative is the mood of assertion. 5 However, the interplay between evaluation and assertion appears to be more sophisticated. Examples (13b), (14b), and (15b) can also express an evaluation by the speaker of the fact described by the indicative, but in this case the evaluation results in a relation of contrariness concerning that fact (veridical): he does take the courses, they do follow your

chapter four

132

advice, or they do study, but the results are not the expected ones: they do not succeed; in (13a, a’)–(15a, a’), the attitude of the speaker is that if the event which is not a fact (nonveridical) becomes a fact (veridical), the effects will match the speaker’s expectations: (14) a. Sigam eles o teu follow.sbjv.prs.3pl they the your conselho![e/assim conseguirão o que pretendem] 6 advice [and/so they will get what they wish] ‘Let them follow your advice!’ a’. Assim eles sigam o teu conselho! So they follow.sbjv.prs.3pl the your advice [e conseguirão . . . ] [then they will succeed . . . ] ‘Let them follow your advice!’ b. Seguem eles o teu conselho! follow.ind.prs.3sg they the your advice [e/mas não conseguem o que pretendem] [and/but they don’t get what they pretend] ‘They do follow your advice, but . . .’ (15)

a. Estudem eles! [e/assim conseguirão o que pretendem, study.sbjv.3pl they [and/so they will get what they wish, e.g. um trabalho] e.g., a job] ‘Let them study!’ a’ Assim eles estudem! So they study.sbjv.prs.3pl ‘Let them study!’

[then they will succeed . . . ]

On Finiteness and the Left Periphery

133

b. Estudam eles! [e/mas não conseguem o que pretendem, Study.ind.3pl they [and/but they don’t get what they pretend, e.g. um trabalho] e.g., a job] ‘They do study, but . . .’ But things become more puzzling if emphatic strategies of negation and affi rmation interact with the speaker’s presupposition based on the common ground. For concreteness: if in (14a) the speaker takes ‘your advice’ as (i) good (or appropriate to a given purpose), then accomplishing the event will lead to a good result (she will succeed); but if she takes it as (ii) bad (inappropriate), then the consequence will be bad. Curiously, emphatic affirmation is compatible with (ii), though not with (i): by emphasizing affi rmation the speaker disapproves of the realization of the event; inversely, emphatic negation is incompatible with (ii), but compatible with (i): the speaker expresses approval.7 The presence of assim ‘so’ neutralizes the diversity of readings: only the presupposition of type (i) is available, consistent with the speaker’s wish that the action take place (they should do . . . ), and emphatic strategies of affi rmation or negation do not apply. Unfortunately, I cannot pursue this topic here and leave it for a future work. I would just note that by using the subjunctive the speaker not only concedes to bring the event to the common ground but also analyzes its integration therein. Apparently, evaluation of a non-fact that can become a fact (subjunctive) interacting with the speaker’s presupposition and with negative or positive features renders the puzzle of contrariness more intricate than evaluation of a fact (indicative). Note further that in these evaluative structures inversion is required, unless an operator such as assim ‘so’ is merged. Plausibly, inversion is the effect of verb movement to EvaluativeP (see section 5). Subjunctive main clauses can also simply express a wish (16–17) 8 or a direct (18) or indirect commitment (19); as above, in all these uses the speaker wants a non-fact (a nonveridical proposition, in Giannakidou’s terms; see below) to become a fact (a veridical proposition): (16)

(Que) (that)

venham as chuvas! come.sbjv.prs.3sg the rains

‘That the rains come!’

chapter four

134

(17)

Oxalá (que) ele venha! hope (that) he come.sbjv.prs.3sg ‘Hope that he comes!’

(18)

a. Não digas/digam isso not say.sbjv.prs.2sg/3pl that ‘Don’t say that.’ b. Apague/Apaguem a luz, se faz favor turn.sbjv.prs.3sg/3pl.off the light please ‘Turn off the light, please.’

(19)

O João que apague a luz the John that turn.sbjv.prs.3sg.off the light ‘John should turn off the light.’

The subjunctive obligatorily substitutes for the imperative under negation (18a), deferent treatment singular, and (non)-deferent plural (18b). Dislocation of the subject over the complementizer is a root phenomenon: (19) cannot be embedded, unlike (18) (see n. 3). Main subjunctives can also express possibility if talvez ‘maybe’ precedes the verb ((20a) vs. (20b)), behaving like an epistemic modal parallel to the Greek case discussed in Giannakidou, this volume;9 talvez can be replaced by it is possible (probable) (21a), though not by adverbs derived from these adjectives (21b); the ban on (21b) seems to extend to all adverbs with the suffi x -mente, a fact probably due to the cognitive origin of this suffi x: (20)

a. Talvez (que) ele venha / *vem Maybe (that) he come.sbjv.prs.3sg /come.ind.prs.3sg ‘Maybe he comes.’ b. Ele *venha / vem talvez

(21)

a. É possível/provável que ele Is possible/probable that he

On Finiteness and the Left Periphery

135

venha / *vem come.sbjv.prs.3sg / come.ind.prs.3sg ‘It is possible/probable that he is coming.’ b. Possivelmente/provavelmente ele Possibly/probably he *venha / vem come.sbjv.prs.3sg / come.ind.prs.3sg ‘Possibly/probably he is coming.’ So, in main clauses the subjunctive indicates nonassertive illocutionary force, evaluation, and epistemic possibility. Sentences in (16)–(18)10 could be complements of volitional verbs (see group A), as subjunctive triggered by talvez ‘maybe’ could occur with epistemic and assertive verbs (groups B and C), though not vice versa; saber ‘know’, the typical indicative selector, resists the subjunctive choice:11 2.1.3 THE MAIN- EMBEDDED DIVIDE.

(22)

a. O João acredita/pensa/?sabe/diz que ela the John believes/thinks/knows/says.ind.prs.3sg that she talvez apareça. maybe appear.sbjv.prs.3sg b. * O João quer/pede que ela talvez the John want/ask.ind.prs.3sg that she maybe aapareça. ppear.sbjv.prs.3sg

The absence of the overt subjunctive trigger talvez reinforces the threshold of acceptability: (23)

O the ela she

João acredita /?pensa /*sabe /*diz que John believes /?thinks /*knows /*says.ind.prs.3sg that apareça. appear.sbjv.prs.3sg

The contrasts in (23) are weakened in the presence of a modal verb of possibility, taking over the role of talvez (24):

chapter four

136

(24)

O the ela she

João acredita /pensa /??sabe /diz que John believes /thinks /knows /says.ind.prs.3sg that possa aparecer. may.sbjv.prs.3sg appear

The variation in (23) clearly depends on the lexical properties of the main verbs and is also cross-linguistic: believe accepts subjunctive in Portuguese or Italian, though not in Spanish or Greek. Giorgi and Pianesi (1997) observed that the subjunctive seems to exist on a continuum from [–realistic] (believe) to [+realistic] (know). I have nothing interesting to add. I would just note that when uttering sentences with believe the speaker knows that by choosing the subjunctive she will refer to an event taken as not being a fact (nonveridical), but she also knows everything about the lexical properties of believe and solves the puzzle. One possibility is that given verbs, in their bundles of features, have at least one appropriate property that can be used to evaluate facts or non-facts, while other verbs do not. In (25) the subjunctive is used if the speaker assumes that John believes in the possibility of her being sick (talvez could be inserted); the indicative is chosen if the belief is in the fact that she is sick. The oddity of (26a) has been widely discussed in the literature; for me (among other native speakers), the sentence is possible with a given stress on the verb,12 meaning that the speaker distinguishes between what she knows (because it is common ground) and what she believes or wants to believe, as confirmed by (26b), where overt focus on the subject (= speaker) produces an acceptable sentence: (25)

O João acredita que ela esteja /está The John believes that she be.sbjv.prs.3sg /be.ind.prs.3sg doente. sick

(26)

a. (#) Não acredito que ela está doente. I don’t believe that she be.ind.prs.3sg sick b. Só eu não acredito que ela está doente. Only I do not believe that she is.ind.prs.3sg sick

Let us return to (23). The ungrammaticality respecting dizer ‘say’ disappears under a reported command meaning (27a), or under the insertion of talvez ‘maybe’, leading to different readings:

On Finiteness and the Left Periphery

(27)

137

a. O Joãok disse-lhei que proi/*k the Johnk said-him.dat that proi/*k viesse/venha à festa. come.sbjv.pst.3sg/ come.sbjv.prs.3sg to.the party ‘John told him to come to the party.’ que proi/k talvez b. O Joãok disse-lhei the John k said-him.dat that proi/k maybe cviesse/venha à festa. ome.sbjv.pst.3sg/come.sbjv.prs.3sg to.the party ‘John told him that maybe he would come to the party.’

As pointed out to me by Anastasia Giannakidou, these cases of dual mood choice have parallels in Greek (Giannakidou and Staraki 2013).13 The question is whether the subjunctive is selected in this case. I maintain that whenever indicative and subjunctive alternate, selection is not at stake. The contrasts in 27a and 27b occur in this sense: as with other verbs expressing commitment (order, ask); the subject of the embedded clause is controlled by the matrix object, not by the matrix subject, but the embedded tense is not dependent on the matrix tense (27a), unlike selected subjunctive; moreover, if talvez ‘maybe’ is inserted (27b), the reported order meaning is lost, the sentence is grammatical with the reading brought by talvez, and control of the embedded subject by the main subject becomes possible (preferential reading).14 Thus: (i) an operator cannot intervene between the subjunctive and the verb that selects it (22b), a simple case of intervention effect: a given feature on the verb in need of valuation would remain unvalued; (ii) insertion of operators in unselected domains does not block the relation with the main predicate, because there is none; in this case the clause will receive the interpretation bestowed by the operator; the intervention of other operators produces a relativized minimality effect, proving their own existence in the structure. The volitional- evaluation- epistemic divide in main clauses fi nds correspondence in embedded subjunctive domains. I conclude that whenever subjunctive is obligatory it is selected by the main predicate, where selection means the presence of a given unvalued feature in the main predicate, which inspects its c- commanding domain for a goal to get valued; whenever subjunctive and indicative alternate (groups B and C), the

chapter four

138

subjunctive is licensed much as in main clauses, by an operator, and plausibly in EvaluativeP, as will become clearer in what follows.15

2.2 Tense and Obviation A well-known phenomenon that characterizes subjunctive clauses is obviation, or the ban on co-reference between the embedded and the matrix subject. Obviation is illustrated across languages by sentences like (28): (28)

* Eui quero que proi venha à festa I want.ind.prs.1sg that pro come.sbjv.prs.1sg to-the party ‘I want that I come to the party.’

Numerous studies attributed the disjoint reference effect in (28) to the defective status of tense in subjunctive clauses (Jakubowicz 1984; Picallo 1984; Raposo 1985; Ambar 1988; and others): the embedded tense of subjunctive is dependent on the matrix tense, but the indicative tense is not: (29)

a. O Pedro quer que o João the Peter wants that the John vá/ *fosse/ *for ao cinema. go.sbjv.prs.3sg/ go.sbjv.pst/ go.sbjv.fut to-the movies b. O Pedro pensa que o João the Peter thinks that the John vai / foi / irá ao cinema. go.ind.prs. 3sg / go.ind.pst. 3sg /go.ind.fut.3sg to-the movies

By virtue of this defectivity, embedded and matrix tenses form a chain; since binding relations are defi ned with respect to tense, the binding domain for the embedded pronominal subject is extended to the matrix clause. According to principle B of binding theory, the pronominal embedded subject cannot be co-indexed with the matrix subject. In Ambar (1988, 1992) I used the binding approach with a slightly different implementation: T and Agr were seen as [±strong, ±overt]; from the different possible combinations of those features and their valuation were derived the different moods on a scale of fi niteness.16

On Finiteness and the Left Periphery

139

It has been pointed out that the binding view is misleading, namely the opposition between anaphoric and independent tense. On the one hand, works on the syntax of tense (Zagona 1990; Stowell 1993; Demirdache and Uribe-Etxebarria 2000; among others) have shown that indicative clauses also exhibit restrictions on the sequence of tense (or consecutio temporum). For a work on subjunctive in line with Demirdache and Uribe-Etxebarria, see Christodoulou and Wiltschko (2012), who argue for the need to dissociate the function of functional categories from their content; in this view Infl is an anchoring category, but it is not universally associated with tense, as in the case of Greek na (see also Wiltschko, this volume).17 On the other hand, it is now widely acknowledged that not all subjunctive clauses behave alike: either the strict sequence of tense illustrated in (29a) or obviation in (28) can be violated, as is the case when subjunctive occurs in unselected clauses under epistemic and assertive verbs. Focus induces a violation of obviation also in selected complements (see (56) in Ambar 1988), as well as modals (Ruwet 1991). Dobrovie- Sorin (2001) argues for an alternative approach to obviation: Romance languages (though not Romanian) do defi ne a binding domain for the embedded subjunctive subject. When a coreference between matrix and embedded subject is required, the choice of control infi nitives over subjunctive would have the effect of a principle of preference to choose anaphors rather than pronouns whenever possible. Yet subjunctive clauses in complements of volitional verbs exhibit stronger restrictions on consecutio temporum than indicative clauses (29b), leading us to hypothesize that some version of dependent tense is to be maintained. The matter is intricate, however. Opposite views have been advocated with respect to the status of tense in complements to volitional predicates. Landau (2004) distinguishes between anaphoric tense [−T] (complements of begin, continue, stop, manage, etc.), dependent tense [+T] (complements of want, ask, decide, believe, etc.), and independent tense [+T] (indicative, with no [T] in C, as opposed to anaphoric and dependent tenses). For Martin (1996) and Bošković (1997), the tense of control infinitives is not defective; it is [+T], in contrast to the tense of exceptional case-marking (ECM) verbs [–T]; [+T] assigns null case. The proposal relies on Stowell’s (1982) observation that the tense of control infinitives is independent, unrealized future with respect to the matrix tense and in the analysis of

chapter four

140

eventive verbs in infinitives suggested by Enç (1991), who argues that those predicates contain a temporal argument that needs to be bound: tense, adverbs of quantification, and have and be are anchoring elements. Thus, in control infinitives like (30), tense is available to anchor the temporal argument, whereas in ECM complements it is not (31): (30)

John tried to buy the book.

(31)

* John believed Peter to buy the book.

Enç’s proposal reinforces Stowell’s observation, which confl icts with the idea that the tense of complements of volitional verbs is dependent or anaphoric, in a sense defective. In minimalism, crucially after Chomsky (2001), the notion of defectiveness took on particular importance: a functional category (T, v) is defective if it contains an incomplete set of φ-features, the exclusion of C being solved by the introduction of feature inheritance in Chomsky (2007): T inherits its features from C (Tcomp); if there is no C, T will be defective (Tdef). For a detailed discussion of defectiveness see Gallego (2007),18 who renders the system uniform: C is always present, either as C def or C comp. Subjunctives in Catalan (as in Portuguese) support C –Tdef: they show an overt complementizer, but behave like defective domains (obviation), joining ECMs. Uriagereka and Gallego (2007) analyze subjunctive clauses in Romance languages as ECMs, rejecting the truncation approach of C (“an ill-understood process”) and assuming Pesetsky and Torrego’s (2001, 2004) (henceforth P&T’s) dissociation of tense and agreement features. The C-T of subjunctives is φ- complete, but Tense/ Case defective. My previous analyses of (non)fi nite structures (e.g., Ambar 1998, 1999, 2005, 2007) and the system presented here share with Uriagereka and Gallego (2007) the claim that C is always present in the structure, the differentiation of φ-features and tense features in regard to valuation, and the idea that some features get valued in the same domain in which others remain unvalued, in which case the latter will fi nd a value in the higher domain (as in ECM). Here, I further hypothesize that the speaker’s projections (see section 5) complete the licensing of (non)fi nite structures.

On Finiteness and the Left Periphery

3

141

Tense Valuation in the Derivation of Subjunctive

I tried to solve the apparent paradox created by the need to consider tense as dependent and independent, through the valuation of tense features Ambar (2005, 2007), reinterpreting my previous analyses according to the probe-goal system of Chomsky (2001). In Chomsky (2001) Agree acts on Tense only through its φ-features. My proposal shares with P&T (2004) the idea that the valuation of tense features is involved in clause formation but diverges from their system on two crucial points: (i) features enter the derivation as valued or unvalued, much as with Chomsky’s (2001, 2008) interpretable vs. uninterpretable features; (ii) Chomsky’s bundles of φ-features are extended to tense features: t-features enter C, T, v (plausibly also To(bj) as proposed in Ambar 1998, 1999; and P&T 2001, which, however, I omit here), and V as bundles that include different features codifying properties related to the morphological tense (responsible for nominative)—the tt-features—and to tense related to event and Aktionsart properties, the so- called aspectual properties (plausibly responsible for accusative)—the tev-features. Roughly, t-features are gathered into two groups, but others may prove to be relevant. Thus, like Nouns and φ-features, Verbs come from the Lexicon with valued or unvalued t-features and φ-features, unvalued features are valued by Agree. I will reserve the abbreviation T1 for the highest Tense category and T or t-features for tense features in each category (C, T, v, and V), whenever the distinction between tt -features and tev -features is irrelevant. Let us discuss two possible approaches: (i) the fi rst considers T1 to be the locus of semantic tense interpretation (Pollock 1989; P&T 2004) and has its t-features valued (interpretable) before Agree applies—that is, T1 inherits its t-features from C and C has valued t-features; thus T in T1 is valued (interpretable). Assuming further that the features of indicative V are valued, there is a redundancy in this option: if the tfeatures of V are valued, and the t-features of C inherited by T1 are also valued, a probe would be missing; if the t-features of V are unvalued, the probe (in V) does not c- command the goal (in C-T), against Chomsky’s 2001 condition for Agree to apply, but according to Adger’s (2003) defi nition of Agree ((68), p. 135). Note that this problem does not come up in P&T’s theory, where the combination of [±interpretable, ±valued] solves the problem; (ii) the second possible approach considers T in C and in T1, inherited from C, to be unvalued (uninterpretable): plausibly,

142

chapter four

the t-features on functional heads are unvalued, generally (in a sense, then, the functional projections have no content until they are valued, in partial agreement with Wiltschko’s proposal, this volume). For concreteness, consider the indicative: the t-features (tt and tev) are unvalued in C and T1, but valued in V; consequently: the unvalued t-feature in vP probes the t-feature in V and gets valued; the unvalued t-feature in T1 probes the valued t-feature in vP and gets valued; the unvalued t-feature in C probes the valued t-feature in T1 and gets valued; C becomes t- complete and has EPP, which is satisfied by merging the complementizer. Take the subjunctive: the t-features (tt and tev) are unvalued in C and T1, but now only the tev-features are valued in V, and the tt-features are unvalued; arriving at C, only the tev-features related to an event will be valued, everything else being equal. Thus the tt-features have to be valued somewhere else, plausibly either by matrix Tense or by an operator. Suppose further that matrix verbs selecting subjunctive have an unvalued β-feature, which acts as a probe inspecting its c- commanding domain for a goal. The label β for this feature means ‘a given feature of the bundle of features of V responsible for selection’ (plausibly related to evaluation); I further assume this probe will fi nd its goal in the valued tev-feature in the embedded domain (C in a non-split CP or a given position in the left periphery; see below). At this point the tt-features remain unvalued until the valued t-features in the matrix domain value them. We have to assume that somehow the relevant valued features in the matrix domain will value the unvalued tt-features under its c- command through either Adger’s (2003, example (68)) or P&T’s (2004) version of Agree. Let me clarify two points: Concerning the issue of the independent/dependent tense and obviation, my analysis solves the apparent paradox referred to in the preceding section. In a sense Stowell is right: the temporal interpretation of clauses under volitional verbs is independent, unrealized future because the tense of the event (given by tev -features of the bundle of features) is valued in the embedded domain and thus is independent of the matrix; those who argue for the deficiency of tense (Picallo 1984, among others) in a sense are also right: by virtue of selection (β-feature valuation) the embedded domain extends to the matrix; note that this should be enough to extend the binding domain for the subject, predicting obviation; I argue this is the case in Russian (cf. (34)–(35)), which exhibits obviation, though not tense dependency; but in languages with subjunctive morphology, there is in addition a

On Finiteness and the Left Periphery

143

subset of unvalued t-features (tt) that can match a value only in T of the matrix T1, predicting strict sequence of tense. We can then keep the binding view. Thus, in languages of the Russian type, only obviation is predicted (no subjunctive morphology, no t-features in need of valuation), but both obviation and tense dependency are predicted to exist in selected contexts in languages of the Portuguese type; thus we are dissociating tense from obviation, differently from Picallo’s first insight (in the representations, absence of un(valued) means ‘valued’): (32)

a. O Joãoi quer/*quis que pro*i /j venha the John i wants/wanted that pro*i/j come.sbjv.prs.3sg à festa. to-the party ‘John wants you to come to the party.’ b. [O Joãoi T1 [tt, tev] [vP [tt, tev] [VP quer [βu ] [tt, tev] [CP (que) [tt u, tev] [TP proi T1 [tt u , tev] [vP [v’ [tt u , tev] [VP venha [tt u, tev] à festa ]]]]]]]]

I suggested above that whenever the subjunctive alternates with the indicative there is no selection (see B and C above). Assuming that subjunctive morphology on the verb always has the same properties, also in unselected contexts the tt-features in the embedded domain should be unvalued. Suppose then that an operator in the embedded domain takes over the role of valuing those features,19 with the consequence that valuation does not extend to the main domain and the prediction that neither obviation nor strict sequence- of-tense surfaces: (33)

a. O Joãoi admite/admitiu que proi /j venha/possa the John i admits/admitted that proi /j come/can.sbjv.prs.3sg vir à festa. come to.the party ‘John admits/admitted that he comes/can come to the party’ b. [O Joãoi T1 [tt, tev] [vP [tt, tev] [VP admite/admitiu [tt, tev] [CP OP tt [(que) [tt u, tev] [TP pro*i/j T1 [tt u , tev] [vP [v’ [tt u , tev] [VP venha[tt u, tev] à festa]]]]]]]]

chapter four

144

Plausibly, what I am labeling here OP (operator) is but features of a dedicated projection of the left periphery (EvaluativeP), the projection of subjunctive as already mentioned; but see section 5, which integrates in a cartographic approach the t-features valuation analysis briefly presented so far, corresponding essentially to what I proposed in earlier works (Ambar 2005, 2007). For ease of exposition I continue to refer to OP. As far as the subjunctive/indicative shift is concerned, a question that might be addressed is why the selected subjunctive cannot shift into indicative, in contrast to the shift of indicative into subjunctive in unselected contexts.20 Consider two pictures: (i) Take an indicative verbal form; merge its CP with a matrix volitional verb; the embedded CP will be t- complete and will constitute a phase. Therefore the unvalued β-feature on the matrix verb would remain unvalued, leading to ungrammaticality. (ii) Take a subjunctive verbal form, merging its CP with a predicate typically selecting indicative. No ungrammaticality is induced: there is no unvalued β-feature in the matrix V; tev enters the derivation valued on V, and the ttfeature will be valued by merging an OP.

Now, if the ungrammaticality referred to in (i) is due to nonvaluation of the feature on the matrix verb, we should ask if there is any other way to value this feature across languages, with the prediction that indicative would be possible in complements to verbs selecting subjunctive. Russian is one of those languages. Russian has no subjunctive morphology, and predicates of the volitional type have Past (Indicative) in their complements (identical to Past selected by epistemic or declaratives); but then the complementizer incorporates a morpheme by (a residue of the old aorist). Example (34) below exemplifies a Russian clause embedded in a volitional predicate: (34)

Ja xoču čtoby ty menya liubil. I.nom want čto+by you.nom me loved.ind.pst.3sg ‘I want that you love me.’

I argue that by in the embedded domain is a goal for the probe in the matrix V (the unvalued β-feature). This probe-goal relation extends the embedded domain to the matrix. Therefore obviation is derived as desired: Russian exhibits obviation, like Portuguese, though not strict

On Finiteness and the Left Periphery

145

sequence- of-tense (see discussion above under point 1), owing to the absence of unvalued t-features: (35)

Ivani xočet čtoby on*i/j potseloval Nadju Ivan wants that-sbj he kissed.ind.pst.3sg Nadju ‘Ivan wants to kiss Nadja’

Complements of epistemic and declarative verbs are headed by čto: (36)

Ivan skazal čto on potseloval Nadju Ivan said that he kissed.ind.pst Nadju ‘Ivan said that he kissed Nadja.’

If in Russian complements to volitional predicates the t-features are all valued, as in Portuguese and Russian indicative clauses, CP is t- and φcomplete and has EPP valued by merging čto. Suppose then that by is merged in a higher projection and that it qualifies as a goal for the probe on the matrix verb. The question then is in what position by is merged. I propose that it is merged in a position on the left periphery, EvaluativeP (see (58)), which is relevant for the semantic and syntactic characterization of subjunctive, as we have been seeing (see Giannakidou, this volume, and section 5 in this chapter), and where čto will move to adjoin by. Joanna Błaszczak asks what motivates this movement. Notice that in root subjunctives by may occur in different positions, but not the initial position. Mezhevich (2006, p. 135) argues that by is a clitic, and as such must be attached to another element; 21 this element is čto, which adjoins it. Another hypothesis I would like to pursue relies on the relation between complementizers and illocutionary force (Evaluative and Assertive, in our analysis). Merging čto low in the structure (in Rizzi’s FinP) satisfies EPP of a domain t- φ- complete, though not the illocutionary force; plausibly, complementizers have to check force, responsible for clause typing; either they move to the relevant projection in the left periphery or Agree applies (for a similar idea see Giannakidou 2009). It is tempting to argue that čto- by incorporation has the effect of neutralizing the (indicative) Past interpretation in the embedded clause (the head of the embedded domain is now by, not čto). Thus all these intriguing cross-linguistic facts follow straightforwardly from the analysis above. In Portuguese, as in most Romance languages, the subjunctive is expressed through specific morphology on the verb;

chapter four

146

in Russian it is expressed with the particle by, external to the verb, but the verb exhibits an invariable morphology, typical of past (indicative). The phenomenon also exists in other languages. Balkan languages also use particles (akin to complementizers) to express the subjunctive, and again a typical verbal morphology appears. Ancient Greek had verbal morphology for indicative and subjunctive (Giannakidou 2009; San Martin 2007). Modern Greek follows the pattern of Balkan languages: the subjunctive is expressed with the particle na, and the verb exhibits a perfective non-past (Giannakidou 2009). Why specific morphology on the verb appears when particles are used has been a point scarcely addressed in the robust literature on subjunctive. Giannakidou (2009) is an exception in the study of this link. She notes that Greek “na is accompanied typically by a verbal form—the perfective nonpast—which is itself also dependent” (p. 1884). The deficiency of the tense of the subjunctive is a temporal dependency anchored to n(ow) (the variable introduced by na). In my analysis, to express the subjunctive, Portuguese, Greek, and Russian take verbal forms whose tev-features (event) enter the derivation as valued (subjunctive, perfective non-past, past); however, the tt-features enter the derivation unvalued in Portuguese and Greek, but valued in Russian (as in indicative); in Portuguese those features are valued by the matrix tense (or by a given operator); in Greek they are valued by na, as Giannakidou (2009) proposes. Observe that both Portuguese (28) and Russian (35) exhibit obviation, while Greek lacks it (38): (37)

Thelo na kerdisi o Janis. want.inp.1sg sbjv win.pnp.3sg the John (Giannakidou 2009, p. 1887) ‘I want John to win.’

(38)

Thelo na pao stin Eladha. ‘I want to go to Greece.’

(Felix 1989, p. 113)

If thelo, like xoču and quero ‘want’, select the subjunctive clause, they should bear an unvalued feature; as before, the binding domain for the embedded subject extends to the matrix clause, with obviation being derived, a good prediction for Portuguese and Russian, though not for Greek. One possibility might be to look at Greek in line with Martineau (1994): Modern French exhibits obviation; Older French did not, suggesting that in one stage of a language what is provided by lexical se-

On Finiteness and the Left Periphery

147

lection is given by other means in another stage of the same language. Martineau suggests that lack of obviation in Older French relies on the visibility of a Mood Phrase, 22 which has been argued in numerous works to be involved in Balkan subjunctives (see Giannakidou 2009, and section 4 in this chapter). Although I cannot go into this topic further here, as in previous works, I assume that discourse properties can be encoded into lexical items or not, in which case they are represented in discourse projections: in Older French volitional verbs do not bear the unvalued feature in need of valuation; that property is encoded in a discourse projection. The richness of the C-system in Greek, with a variety of particles looking like complementizers, is a symptom that in Greek also represents that property (na) in a projection, say MoodP, which is plausibly linked to the highest projection (establishing a relation with the world), either by movement, as proposed in Giannakidou (1998) or by some version of Agree (Giannakidou 2009). Thus na would have a function similar to Portuguese operators (and modals) in unselected contexts, one of the consequences being that it would defi ne a binding domain for the embedded subject and would value the unvalued tt-features of the perfective non-past. The issue is tricky, though: while in Older French “the choice between the indicative or the subjunctive is not lexically determined” (Martineau 1994, p. 59), and “it is only in discourse that the mood is defi ned” (idem, quoting Moignet 1984, p. 226), it is unclear that the same is true in Greek; volition verbs occur with both indicative and subjunctive in Older French, but in Greek as far as I know they do not (they take na, not oti). Some version of selection seems then to be at stake. It is tempting to relate lack of obviation in Greek with cases of lack of obviation in Portuguese in volition contexts (see discussion in examples (56) and (57)). The status of na and the position the subject occupies become relevant to explain the phenomenon (whether this correlates with the lack of infi nitives in Greek is unclear and remains open, but see San Martin 2007). Although the tense-valuation approach outlined so far appears to be necessary for an account of the subjunctive-indicative divide, it is not sufficient. The speaker’s relation with the world, represented in syntactic structure by discourse projections, seems to complete the puzzle.

4

Worlds, Speaker, and Left Periphery

Elaborating on Marques’s (2009) remark that the subjunctive “does not seem to be associated with a specific kind of modality” (p. 153), Gianna-

chapter four

148

kidou (Chap. 5, this volume) observes: “[I]f we are in search of the holy grail of one modality for the subjunctive, we are not in luck. But if we look closer, we see that there is indeed one underlying property shared by all subjunctive contexts: they are all nonveridical.” As she writes, the property of nonveridicality is related to the speaker (or the clause subject), a point that revives old proposals on the role of speech acts and illocutionary force in the derivation of subjunctive structures (e.g., Searle 1969). According to Giannakidou (1994, 1998, 1999, 2009, and this volume), indicative is selected if the complement clause is taken to be true by some individual in some possible world,23 or strictly speaking, in some individual’s model, which is a set of worlds, an epistemic state, compatible with what a person knows; 24 subjunctive occurs in nonveridical contexts. How (non)veridical spaces relate to worlds is clear in Giannakidou (Chap. 5, this volume, (30)): a. An epistemic state (a set of worlds) M(i) relative to an individual anchor i is veridical with respect to a proposition p iff all worlds in M(i) are p worlds (full commitment to p). b. If there is at least one world in M(i) that is a ¬p world, then M(i) is nonveridical (weakened commitment to p). c. If all worlds in M(i) are ¬p worlds, then M(i) is antiveridical (countercommitment to p).

All verbs that select the subjunctive come with spaces that follow (ii); that is, they are partitioned into p and non-p worlds. The spaces need not be epistemic, they can be bouletic (as with deontic modals), emotive factives, and evaluative nonepistemic uses of subjunctive in general. Thus a veridical epistemic space is homogeneous; no partition is created (all worlds are p worlds). a nonveridical space is not homogeneous (it contains p and non-p worlds). But see Giannakidou (Chap. 5, this volume) for semantic details. Quer (1998) also uses this model (set of worlds) when he talks about model shift. Thus Giannakidou’s (non)veridicality covers the simple dichotomy realis- irrealis and avoids its empirical problems by introducing worlds, evaluation, and anchors to that evaluation (the speaker or the matrix subject): truth is relativized to an epistemic model anchored to the speaker or to the subject of the main clause. Quer (1998) also assumes that the reliable contribution of the subjunctive is to signal a shift in the model of evaluation of the truth of the proposition. Bianchi (2001) shares a similar

On Finiteness and the Left Periphery

149

view. As she observes, her conception of logophoric centers (perspectives or cognitive states introducing the event) is close to the notion of models of evaluation anchored to agents. An external logophoric center establishes a relation with the common ground, and thus with the speaker; in an internal logophoric center the perspective is defi ned by the matrix subject (the internal speaker). It was also logophoricity that led Speas (2004) to propose that worlds are encoded into the syntactic structure. Her analysis of evidentiality in languages such as Quechua shows that “Evidential morphemes encode information about the type of evidence the speaker has for the truth of what s/he is saying,” thus “the argument specified for evidential agreement is a world argument rather than a nominal argument” (Speas 2004, p. 253, italics mine).25 Speas discusses the possible correspondence between logophoric categories and the highest four projections of Cinque (1999),26 concluding that the overlap between them is not as straightforward as it might appear; rather her account of the evidential paradigm involves features “from all four projections combining.” Establishing a parallel with subjunctive, it is tempting to see a correlation between the variable modalities this mood conveys and the type of evidence the speaker has for the truth of what she is saying (depending on her relation with the world); if visible evidential morphemes such as those of Quechua were available in the languages we are concerned with, things would surely be simpler. Conceivably, the derivation of subjunctive involves properties of different projections, as Speas concludes for the evidential paradigm, and as I also argued for other phenomena analyzed under the structure I proposed for the left periphery (section 5): clause typing and interpretation are compositional; they involve properties of different projections (Ambar 2003). Now the question is which projections are activated and which pragmatic properties of subjunctive can be encoded in those projections. This recalls the old controversy over the viability of encoding pragmatic factors in syntactic representations, as was fi rst attempted by Ross (1970). The reason for the debate has always been that discourse properties are potentially infi nite, thus not sufficiently constrained to be accounted for by syntactic mechanisms. I explore here for subjunctive what I have claimed before, based on other facts: the compositional perspective of the left periphery highly constrains the infi nite character of pragmatic factors. For concreteness, I argue that the interplay between evaluation and assertion is derived by EvaluativeP and AssertiveP, projections of the left periphery (Ambar 1999b, 2000, 2003):

chapter four

150

(39)

(XP*) EvaluativeP AssertiveP (XP*) WhP FocP (XP*) TP (Ambar 2003)27

AssertiveP encodes into the left periphery what Searle (1969) defi ned as “what the speaker knows”; it is related to the common ground (see Ambar 2003 for details); it is the locus of the speaker’s epistemic attitude, the place of individual anchoring in the truth conditions. EvaluativeP is a projection codifying the speaker’s evaluative attitude toward the state of affairs described; it is the not- at- issue level, the non-true conditional part, the emotive (bouletic or other wise) attitude, in Giannakidou’s terms. This explains the order Evaluative-Assertive and is consistent with Giannakidou’s words (Chap. 5, this volume): Subjunctive selection with emotives is thus consistent with nonveridicality in that, in this case, a nonveridical space is created at the non at-issue level, as opposed to nonveridical verbs, where nonveridicality characterizes at-issue content.

The fact that in Greek we fi nd, at this high position, different complementizers supports these ideas, as Giannakidou also observes.28 Both projections are the speaker’s projections. It is tempting to establish a parallel with the indicative-subjunctive divide: roughly, AssertiveP is the projection of indicative; EvaluativeP is the projection of subjunctive. This hypothesis retrieves the old idea that the choice of mood is related to properties of speech acts: indicative, though not subjunctive, is the mood of assertions (Quer 2009). Assume then that the derivation of subjunctive clauses involves discourse projections. Many studies, particularly on the Balkan subjunctive, have explored the cartographic approach of Rizzi (1997, 2004), 29 to which MoodP is often added (Terzi 1992; Rivero 1994; Roussou 2000; Cornilescu 2000; Alboiu 2002; Giannakidou 1998, 2009; Hill 2013; among others). In Rizzi (1997, 2004), FinP is associated with finiteness and modality. An issue that remains unsettled is whether there is any overlap between MoodP and FinP;30 MoodP is usually assumed to head the inflection system. D’Alessandro and Ledgeway (2010) discuss aspects of Abruzzese that seem to suggest that MoodP can be dispensed with. In contrast, Hill (2013) argues that negation in Romanian provides evidence for MoodP: the Romanian complementizer ca can co- occur with the subjunctive

On Finiteness and the Left Periphery

151

marker să (Dobrovie- Sorin 1987, 1994): ca is in Fin, like ca in Abruzzese; but unlike Abruzzese ocche, Romanian să cannot be in T, since it precedes negation. A similar argument is used by Giannakidou (1998, p. 52; 2009, p. 1883): in Greek, the subjunctive marker na precedes the negation operator that is compatible with it (min); thus it occupies MoodP above NegP; Giannakidou derives the position of the subject in Greek subjunctive clauses from the presence of this projection in the structure: subjects move to spec,MoodP, explaining why they precede na. Thus MoodP seems to be required to account for the considerable cross-linguistic variation in subjunctive clauses, yet whether it can be dispensed with remains an open question. Another point is the relation between mood and modality. Different works propose a relation between Fin and/or MoodP and ForceP. In Giannakidou (1998), na originating in MoodP moves to C, or does not move but is still linked to C (Giannakidou 2009); in Roussou (2000) the Greek subjunctive marker originates in C M(odal) (equivalent to Rizzi’s FinP) and moves to C op (corresponding to Rizzi’s ForceP); Bianchi (2001) links Fin to the speech point S (the external logophoric center); Kempchinsky (2009) introduces W(orld) in ForceP (WR for indicative, WSU for subjunctive); W is linked to Fin. (I cannot, of course, discuss all the analyses of subjunctive here.) The main goal of this section is to bring to the discussion syntactic, semantic, and pragmatic factors that are involved in the accurate characterization of subjunctive and some of the attempts to represent them in discourse projections. Coming back to the observation by Giannakidou with which I started this section, one factor seems to be crucial in the determination of (non)veridicality: the speaker. The question turns out to be how the speaker and the property of nonveridicality are represented in the syntactic structure.

5

The Speaker. EvaluativeP and AssertiveP

In some proposals regarding the different phenomena involving the relation between speaker and hearer, it has been claimed that the left periphery of sentences should include a speech act layer: Speas and Tenny (2003), adapted by Haegeman and Hill (2011), propose the following structure:

152

(40)

chapter four

a. [saP SPEAKER [sa sa [sa* (UTTERANCE CONTENT) [sa* sa* (HEARER) (Speas and Tenny 2003) b. [SAP SPEAKER [SA’ SA [saP HEARER [sa’ sa [CP = utterance Specifier [C’ C Φ allocutive probe [TP . . . (Haegeman and Hill 2011)

In my proposal (39), which is repeated below for convenience: (39)

XP* EvaluativeP AssertiveP XP* WhP FocP XP* TP (Ambar 2003)

EvaluativeP and AssertiveP are the speaker’s projections. Peggy Speas suggested to me that EvaluativeP and AssertiveP, being the speaker’s projections, might form a phase. I do not think they do (but see Speas and Tenny 2003). Another point is whether a projection representing the hearer should be encoded in the structure of the left periphery; my suspicion is that what regards the hearer is introduced in the utterances by the speaker (through either the common ground (AssertiveP) or evaluation (EvaluativeP)); however, the rich range of cross-linguistic facts provided by Speas and Tenny (2003) would have to be examined in detail under this hypothesis. But this is a topic for another work. Finally, a comment on putting this structure of the left periphery in embedded clauses, a controversial assumption: in unselected subjunctive clauses, no problem arises according to Emonds’s (1976) defi nition of root; as for selected clauses, it has been shown that phenomena such as topic and focus occur in some languages under given conditions, though not in others (e.g., in Portuguese and Spanish, though not in Brazilian Portuguese or English); it seems then that independent properties (correlating with other phenomena) and/or properties of the root clause itself regulate the crash or the valuation of those projections. We have seen that EvaluativeP-AssertiveP, in combination with tense features and with properties of other projections, derive a considerable number of subjunctive facts. In this proposal evaluation generalizes to all uses of subjunctive. Interestingly, Giannakidou (this volume) independently reaches a similar conclusion: “Evaluative subjunctive, in fact, manifests itself in three functions identified here are epistemic, emotive, and preference subjunctive. Evaluative subjunctive, in all cases, creates

On Finiteness and the Left Periphery

153

a nonveridical space, that is, a modal space partitioned into  p  and  not p worlds.” In this view evaluation underlies the property of nonveridicality. If EvaluativeP is involved in the derivation of nonveridicality, we should wonder whether evaluation acts in other subjunctive contexts, namely in those involving verbs like want. Interestingly, want, like regret and ask, accepts degree operators, which is typical in evaluative contexts, suggesting that these verbs carry evaluative features, 31 with consequences for the derivation of their subjunctive complements: (41)

a. Quero tanto/muito que chegue want.ins.prs.1sg so much/very that arrive.sbjv.prs.3sg a Primavera. the spring b. Lamento tanto/muito que não regret.ins.prs.1sg so much/very that you tenhas vindo. have.sbjv.prs.2sg come c. Pedi-lhe tanto/muito que viesse. ask.ins.prf.1sg-him so much/very that come.sbjv.pst.3sg

(42)

a. ?? Não acredito tanto/muito que ele leia I don’t believe so much/very that he read.sbjv.prs.3sg o livro. the book b. * Admito tanto/muito que ele vá. I admit so much/very that he come.sbjv.prs.3sg c. * Suponho tanto/muito que ele chegue a tempo. I suppose so much/very that he arrive.sbjv.prs.3sg ontime

Paradigms (41) and (42) also trace the cut between subjunctive selected by matrix predicates (only subjunctive is possible in (41)) and unselected subjunctive (both subjunctive and indicative can be chosen in (42)), as in the intensional-polarity divide of Stowell (1993). Matrix clauses are not lexically selected either, yet the evaluation puzzle does exist: the relevant

154

chapter four

evaluative properties are encoded into EvaluativeP acting on AssertiveP, based on the common ground, and confi rming what I suggested above about discourse properties being encoding into the lexicon or into discourse projections. The proposal that EvaluativeP is involved in the derivation of subjunctive clauses thus fi nds support in Giannakidou’s observations on evaluative subjunctive and Quer’s observation that subjunctive signals a shift in the model of evaluation of the truth of the proposition. I hypothesize that Evaluative derives nonveridicality; Assertive derives veridicality. But is AssertiveP involved in the derivation of subjunctive clauses (see n. 5)? Root subjunctive clauses under (13)–(15) lead to the suspicion that it is. That at least some features of Assertive may be involved in subjunctive is suggested by the fact that, under given restrictions and less freely than in indicative contexts, topic and focus in selected subjunctive complements are relatively possible in languages like Portuguese and Spanish (Ambar and Jiménez-Fernández 2014 (henceforth A&J). According to Emonds (1970, 1976), these discourse phenomena creating new positions at the root of the sentence are root transformations (not structure-preserving transformations). Interestingly, Hooper and Thompson (1973) (henceforth H&T) limit these types of root phenomena to assertive contexts and observe that they can appear in embedded clauses that qualify as assertions (e.g., complements of declarative and epistemic verbs); still, according to H&T (1973, p. 479), complements to emotive verbs are among the embedded clauses that are incompatible with root phenomena. The correspondence between my projections and the selectedunselected divide I have been talking about is obvious. I will assume that Assertive plays a role in selected subjunctive clauses (emotive in H&T) in languages of the Portuguese or Spanish type, but is defective in a sense to be made clearer below, 32 rendering topic and focus moderately acceptable, but subject to given constraints (e.g., verb movement over the focus element as shown in A&J 2014). In unselected subjunctive clauses, Assertive is less defective (embedded assertions in H&T), rendering topic and focus more acceptable. I maintain my characterization of AssertiveP as codifying “what the speaker knows” (Ambar 2003). Evidence can be provided by subjunctive clauses under epistemic verbs (see discussion of (25) and (26)) and by the interplay between evaluative and assertive in main clauses. Interestingly Giannakidou independently arrives at a similar conclusion:

On Finiteness and the Left Periphery

155

“In other words, when the speaker chooses to utter the subjunctive version, she is making a point to distinguish between what she believes and what she knows for sure. When she chooses the indicative version, the speaker is in a veridical state and the belief is justified (‘higher degree of certainty’), but when she chooses the subjunctive, she seems to be aware that she her belief might not be justified.” Further support for the hypothesis that AssertiveP plays a role in the derivation of subjunctive may come from a fact noted by Quer (2009, p. 1782): Again, the indicative interpretation accords with the presupposition or assertivity associated with this mood, but the subjunctive variant does not simply have a ‘‘neutral’’ reading. On the contrary, the embedded proposition expressed in subjunctive is presupposed, that is it gets interpreted as integrating the common ground of the conversational context.

Quer (2009, p. 1782) observes that (43) exhibits a “concessive nuance,” which can be interpreted as conceding that the proposition can become part of the common ground (n. 11): (43)

a. J’admets que vous avez raison (French) ‘I admit that you are.ind right.’ b. J’admets que vous ayez raison ‘I admit that you be.sbjv.prs.2pl right.’ (Grevisse 1993, p. 1618)

AssertiveP was intended to account for presupposition and factivity, based on the common ground (Ambar 2003). Take one of the facts analyzed therein: the factivity effect provided by the shift of wh- questions into wh- exclamatives: (44)

O que comprou o João? the what buy.ind.pst.pfv.3sg the John ‘What did John buy?’

(45)

O que o João the what the John

comprou! buy.ind.pst.pfv.3sg

chapter four

156

Abstracting from the inversion phenomenon here, in (44) the wh- phrase inspects for a value for the variable: that is, the speaker does not know what John bought; in the exclamative (45) she does know and evaluates. This exclamative factive effect (Grimshaw 1979) is interpreted as the activation of AssertiveP. Evaluation accomplished in EvaluativeP implies that the speaker knows what he is evaluating. Thus, if the subjunctive involves evaluation, it should also somehow involve knowledge. Observe further that the activation of Evaluative in exclamatives in (45) also suggests contrariness with respect to the questions in (44): in (45) the speaker knows what John bought; in (44) she does not know, recalling the contrariness effect characterizing the interplay Evaluative–Assertive in root subjunctive clauses (described in section 2.1). Again we fi nd support in Giannakidou’s insightful observations. Discussing the subjunctive-indicative divide in complements to emotive factive verbs, she observes (Giannakidou, this volume): “Baker (1970) suggested that emotives express a negativity, a ‘contrariness’ between a perceived fact and some mental or emotional state.” Note that evaluation has an emotive component. If evaluation generalizes to other subjunctive contexts, then the contrariness effect will be expected generally, as we have been seeing. Until now I have omitted any discussion of emotive factive verbs, except in section 2.1, where we saw they obligatorily select subjunctive. The problem raised by those verbs is well known: presupposing the truth of their complements (veridicality), they should select indicative, the case of Greek and Russian, among other languages, though not the case of Portuguese and most Romance languages. Since Kiparsky and Kiparsky (1970) factivity has been seen as a nominal effect. In Portuguese, emotive-factives select subjunctive (46a) or inflected infinitive (46b), which in given contexts can be introduced by the defi nite determiner o ‘the’; in the subjunctive clause, o cannot occur: (46)

a. Lamento (*o) que eles tenham dito isso I regret (*the) that they have.sbjv.prs.3pl said that b. Lamento (o) terem eles dito isso. I regret (the) have.inf.3pl they said that

To ensure the nominal status of inflected infi nitival complements to factive verbs, Raposo (1987) proposes that they select IP whose head (Agr)

On Finiteness and the Left Periphery

157

is nominal (Infl lacking Tense). In earlier works claiming that inflected infi nitives do have tense, I proposed (Ambar 1992, 1998) that these verbs select (DTu), D with an unvalued tense feature (on tense in nominals see Grimshaw 1990). That proposal had the following effects on obviation: (47)

a. O Pedroi lamenta que pro*i/j diga The Peteri regrets that pro*i/j say.sbjv.prs.3pl b. O Pedroi lamenta que proi/j the Peteri regrets that proi/j

isso that

tenha dito isso have.sbjv.prs.3pl said that

It is the presence of tenha ‘have.sbjv’ that prevents obviation in (47b), contra (47a). I assumed that ter ‘have’, valuing the t-feature of D, creates a binding domain for the embedded subject; other wise the unvalued t-feature of D remains unvalued until it probes a value in the matrix, predicting obviation as before (47a). Adapting this analysis to what I propose here, and adding Rizzi’s FinP to the structure in (40), (47a) would look like (48): (48)

[EvalP [AssertP πT [TP Tns tevi, tt] o Pedro lamenta [tev, ttu] [βui] [EvalP Dtu[βi] que [AssertP πT [tev, ttu] [FinP que[tev, ttu] [TP Tns [tev, ttu] pro diga isso].

By selecting DT in Evaluative, two properties of these constructions are encoded in the structure: [+fact], represented by DT, [+evaluative/ emotive], represented by evaluative selection ([β] in EvaluativeP and on the verb). According to Giannakidou (this volume), the emotive status of these structures in Greek is provided by “expressive agreement” between pu (the factive complementizer) and the emotive property on the verb. My view is similar: this agreement relation is the evaluative selection represented by [β]. Suppose now we also assume that pu values DT. The nominal properties of pu described in Giannakidou (2009) and in Roussou (2010) seem to come in this sense. 33 If this were the case, then we would understand why the subjunctive occurs in Portuguese and the indicative occurs in Greek: not only would pu establish the relation with the emotive property of regret, but it would also value DTu. Portuguese has no such specialized complementizers, which is why que cannot take over the role of pu; thus the subjunctive has to be chosen because the choice of indicative would lead to ungrammaticality: all t-features

chapter four

158

would be valued in the Fin-TP domain; therefore the unvalued t-feature on D would remain unvalued and the structure would crash.

6

More Puzzles

So far I have interpreted the deficiency of the tense of subjunctive as the effect of embedded tt-features being valued in the matrix domain, whereas tev-features are valued in the embedded domain. Now let us examine how Assertive relates to tense. In a sense, AssertiveP is a window to the relation between speaker and world. We also know that indicative is the mood of assertions (Quer 2009). Assume then that Assertive encodes w(orld) properties and tense properties. Assume further that AssertiveP is “complete” if indicative is chosen—let us say if πT is satisfied—where πT means: proposition with all T-features valued, that is iff its c- commanded tense domain is T- complete. This is not the case for subjunctive clauses: with Assertive, only features related to the event are valued. For concreteness, consider the representation of a selected subjunctive (also simplified for clarity): (49)

a. O João quer que ele venha ‘John wants him to come.’ b. [EvalP [AssertP πT [TP T1 tevi, tt] o João quer [βui] [EvalP [βi] que [AssertP πT [tev, ttu] [FinP [tev, ttu] [TP Tns [tev, ttu] ele venha ]]]]]]].

When Assertive is not complete in its own domain it has to reach the higher AssertiveP. In a sense, the proposition expressed in subjunctive is brought to the common ground, as Quer observed and as I think is the case of subjunctive in general.34 Note that this interpretation does not come up in indicative clauses. In selected subjunctive there is a chain formed by the need for valuation of the tt-features and [βu] on the volitional verb (see section 3), which is valued by the tev-feature. Embedded Assertive accesses the matrix Assertive. As is well known since Picallo (1984), valuation of tense of selected subjunctive is done in the immediate higher domain: (50)

O Pedro lamenta que a Ana queira que eles a Peter regrets that Ana want.sbjv.prs.3sg that they her destruam. destroy.sbjv.prs.3pl

On Finiteness and the Left Periphery

159

Kempchinsky (2009, p. 1805, ex. (44)), quoting Borgonovo (2003), notes the intriguing fact translated here as (51), which contrasts with (50): (51)

a. A imprensa quer que a gente (não) acredite The press wants that we (not) believe.sbjv.prs.1pl que o governo *minta/mente. that the government lie.sbjv.prs.3sg/ind.prs.3pl

Kempchinsky then observes that the ungrammaticality of (51) is maintained if suspect, which according to the author is not a “strong intensional,” replaces believe in (51).35 In Portuguese, suspect takes either indicative or subjunctive. Note that in (51) the t-features of subjunctive are valued in each immediate higher domain. Suppose then that the problem involves Assertive, not tense valuation. One could conceive that the intervention of a selected Assertive (complement of want) between two instances of unselected AssertiveP (matrix Assertive and complement of believe) blocks licensing. But then, we would not be able to explain (52): (52)

a. Acreditem que eles Believe.sbjv.prs.3pl that they *mintam/mentem lie.sbjv.prs.3pl / lie.ind.prs.3pl b. Acredito que eles I believe.ind.prs.1sg that they mintam/mentem lie.sbjv.prs. 3pl / lie.ind.prs.3pl

Let us suppose that the mismatch is due to operators occurring in unselected domains. Suppose they have to be identified.36 More precisely, OP in the embedded clause of believe in (51) has to be identified by the immediate valued head (i.e., believe with all its t-features valued (T-complete). However, this subjunctive verbal form (acreditemos ‘believe.sbjv’) has unvalued tt-features: OP is not identified and the structure crashes. The reasoning applied to (51) also derives (52a), which contrasts with (52b), where all of the features of acredito are valued. According to the judgments of a large number of speakers (some speakers get confused by the data), sentences like (53) are grammatical:

160

(53)

chapter four

Eles não acreditam que admitas que eles they don’t believe that admit.sbjv.prs.2pl that they mintam/mentem lie.sbjv.prs.3pl / lie.ind.prs.3pl

If this is the case, then we cannot maintain the previous hypothesis: in (53) the embedding verbal forms are not T- complete; thus the operators would not be identified and the sentence should be ungrammatical, contrary to the facts. Plausibly, operators have to be identified, but without the requirement that V has to be T- complete. We come back to the idea that the ungrammaticality of (51) relies on Assertive. Recall that Assertive should be more defective in selected subjunctive clauses than in unselected subjunctive domains (see observations on topic and Focus and on H&T’s insight). Given that in both types of Assertive t-features are similar (subjunctive), I must conclude that Assertive has other properties of its own, beyond tense. Clearly those properties are world-related. Thus assume that Assertive has W(orld) and T(ense) properties. I suggest then that Assertive in unselected contexts has W properties different from W properties in selected contexts. Suppose now that embedded defective Assertive reaches the matrix Assertive under a matching requirement: if the given properties of W (set of worlds) do not match in successive Assertive domains, an intervention effect obtains: therefore the chain cannot be formed, defective embedded assertive does not reach matrix Assertive, and the structure crashes. Thus (50) is well formed because a chain is formed: no Assertive with different W-features intervenes; (51) is bad because Assertive in the complement of want has W-features distinct from those of Assertive in the complement of believe. Consequently, the most embedded Assertive cannot reach matrix Assertive, and it still needs to by virtue of its defectivity due to T-features; the ungrammaticality in (52a) is also explained: no matching exists between embedded and matrix Assertive; but note that if the embedded Assertive were of the selected type, the sentence would be acceptable: a chain would be formed and both embedded and matrix Assertive would be valued in the conversational context, mediated by Evaluative (there is an addressee): e.g., não queiram que eles mintam, ‘don’t want they lie sbjv.prs.3sg’; (53), similar to (50), is well formed: W and T-features are of the same type in all domains. Note that this hypothesis simultaneously shows that there is no overlap between Tense and Assertive, reinforcing the pertinence of the latter:

On Finiteness and the Left Periphery

161

Assertive (as well as Evaluative) links the speaker to the world, but imposes restrictions on tense. This is my assumption for the time being; other facts will have to be considered in future work. An apparent problem is raised by (54): (54)

a. Peçam/ordenem/digam-lhes que venham Ask/order/say.sbjv.prs.3pl-them.dat that come.sbjv.prs.3pl b. Pedi-lhes Ask.ind.pst.1sg-them.dat

que venham. that come.sbjv.prs.3pl

At fi rst glance, the well-formed nature of (54) is not a problem: (54) seems to follow the obligatory selection pattern in (50): a chain is formed and embedded Assertive reaches matrix Assertive, respecting matching. A closer look reveals problems I left open in section 2.1: if (54) follows the selection pattern, how o explain the unrestricted sequence of tense in (54b)37 and the control of the embedded subject by the matrix object? Note that one of the verbs in (54a), dizer ‘say’, contrary to pedir ‘ask’, allows dual selection in the embedded clause (cf. the occurrence of other operators in (27)); thus, in this case, according to the proposal outlined so far, the subjunctive should not occur. Something else has to be said (my suspicion is that the point relies on other properties of object control). For the time being, I maintain the idea of an operator in these structures. But now the operator is a commitment operator, characterized by establishing a link with the addressee (lhes) in (54) (the hearer in direct commands). I argue that this operator is identified by the matrix complement. Note that lack of obviation in object control was also a problem for the binding approach. There have been different attempts to solve it by claiming that the matrix object does not c- command the subjunctive clause, an approach difficult to maintain if we consider the antisymmetry of Kayne (1994). Still, note that (51) is possible with indicative: indicative is T- complete; thus Assertive is not defective and does not need to reach the matrix Assertive (no intervention obtains). I argue that the same explanation holds for clauses in which a modal is introduced, improving the presence of the subjunctive: (55)

a. A imprensa quer que não acreditemos que o The press wants that not believe.sbjv.prs.1pl that the

chapter four

162

governo possa estar a mentir. government may.sbjv.prs.3sg be prep. lying. b. Acreditem que eles possam vir a Believe.sbjv.prs.2pl that they may.sbjv.prs3pl come to passar no exame. pass in.the exam The explanation relies on the modal capacity to value tense. In section 2.1 we observed that the modal strategy facilitates the subjunctive in complements of epistemic and assertive predicates. This regular behavior of modals leads us to hypothesize that MoodP is activated. Plausibly, both modal verbs and the modal operator of possibility (talvez ‘maybe’) are merged in this position and establish a link with Evaluative and Assertive. The behavior of topic and focus in these structures can be revealing about the adequateness of this hypothesis. To conclude, let us examine a case of selected subjunctive in which coreference between the matrix and the embedded subject is possible. So far we have assumed that an operator in an unselected domain defi nes this domain as a binding domain (for redefi nitions of binding domain see Lasnik and Uriagereka 2005; San Martin 2007). Focus operators or inversion also seem to defi ne a binding domain in selected clauses in Portuguese, as observed in Ambar (1988, p. 777), with the effect that obviation is canceled:38 (56)

a. Os chefesi querem que só elesi mandem ‘The bosses want that only they command.’ b. A Ana quer que seja ela a fazer o jantar the Ana wants that be.sbjv.prs3sg she to make the dinner ‘Ana wants her to make the diner.’

As discussed under (38), Greek does not exhibit obviation in subjunctive complements to volition verbs; moreover, in this context overt subjects either follow the verb (37) or are left dislocated, preceding na—adjacency between na and the verb being required (57):

On Finiteness and the Left Periphery

(57)

163

a. *Thelo na o Pavlos erthi. want.1SG sbjv the Paul.nom come.pnp.3sg (Giannakidou 2009, p. 1891) b. Thelo o Pavlos na erthi. want.1sg the Paul.nom sbjv come.pnp.3sg ‘I want Paul to come.’

Comparing Greek with Portuguese, it is tempting to argue that in both languages the mechanism rendering lack of obviation possible creates a similar configuration for the subject (post-verbal or left dislocated), defi ning a binding domain, but the trigger of the operation is certainly different: in Portuguese, focus is the reason why pro is unavailable; in Greek, it is licensing of the subjunctive itself (plausibly relying on properties of na and its relation with perfective nonpast). The operation seems to involve verb movement. 39 In any case, adjacency (no lexical material, apart from clitics and negation, may intervene between na and the verb) and the position of the dislocated subject preceding na have to be accounted for. I cannot pursue this further here, but see Giannakidou (2009).40 Finally, volitional predicates allow constructions like (58), as I have noted in previous work (Ambar 2005): (58)

Quero os meninos a dormir (?-em) às 9h. I want the kids to go to sleep (-Agr.3pl) at 9 o’clock.

In these structures, adjacency between a and the verb is required, and the overt subject has to precede a, recalling Greek na subjunctive clauses. Moreover, a can also occur with perception verbs (eu vi os alunos a sair ‘I saw the students leaving’), as studied by Raposo (1989), who attributes an aspectual status to a, correctly I think. Interestingly, Giannakidou (2009, n. 2) observes that na can also occur in this context in Greek, distinguishing, however, this case from subjunctive na structures. I argued that a is a deontic element originating low in the structure (in v or To) where it values the tev-features of V; in the absence of valued tt-features the embedded subject rises to matrix v-ToP where the accusative case is valued, an exceptional case-marking effect, close to the proposal of Uriagereka and Gallego (2007) for Romance subjunctives in general.

chapter four

164

Limitations of space preclude pursuing these topics here. The study of these correlations, including the relation with Greek na, with English deontic to explored in Kayne (2012), and with subjunctive object control will be the subject of future study. In section 3 I suggested that by in Russian (35), repeated below, was merged in Evaluative. Now we are in a position to provide a representation (59): (35)

Ivani xočet čtoby on*i/j potseloval Nadju Ivan wants that-sbj he kissed.ind.pst.3sg Nadju

(59)

[EvalP [AssertP πT [TP Tns tevi, tt] Ivani xočet [βui] [EvalP čto-by [βui] [AssertP [tev, ttu] [FinP [tev, ttu] čto [TP Tns [tev, tt] on*i/j potseloval Nadju ]]]]]]]

FinP is T- complete, where EPP is satisfied by merging čto; by is merged in Evaluative, where the probe (βu) on matrix V probes by and gets valued; obviation will obtain; čto moves to Evaluative, adjoining by.

7

Conclusions

In this chapter I proposed a syntactic mechanism for deriving subjunctive clauses, hypothesizing that it encodes both tense features in the heads and semantic-pragmatic properties in discourse projections. The traditional idea that the subjunctive is dependent is viewed here as the effect of tense features valuation, in line with Pesetsky and Torrego (2004); differently, however, I maintain the notion of bundles of features and assume that features enter the derivation as valued or unvalued, much as in Chomsky (2001). The anaphoric tense relation in Picallo’s approach, deriving obviation by extending the binding domain to the matrix clause, is here restricted to a subset of the bundle of features: the tt -features; those related to event (the tev -features) enter the derivation valued. I argued that this differentiation is responsible for the independent temporal interpretation observed by Stowell (1982). Binding extension occurs only in selected contexts, owing to an unvalued β-feature on the main verb (as proposed in Ambar 2005, 2007); in unselected clauses an operator does the job, in line with Stowell’s concept of an intensional-polarity divide. I hypothesize that features on functional heads are unvalued generally. I explored the viability of encoding in syntactic representation pragmatic

On Finiteness and the Left Periphery

165

factors that accurately characterize the indicative-subjunctive divide. Giannakidou’s (non)veridicality and its relation with evaluation were particularly inspiring. I argued that the speaker’s projections EvaluativeP and AssertiveP, independently motivated, are the projections of subjunctive and indicative respectively. Defective Assertive also intervenes, capturing Quer’s observations on presupposition and Hooper and Thompson’s observations on root phenomena in Assertive contexts. Tense-valuation and these projections encoding a W-feature do not overlap. Space limitations precluded discussing other issues, such as objections to considering indicative as assertion (it occurs in questions); but see Kartunnen (1977). Trying to understand the behavior of subjunctive in main clauses, I traced a parallel with the embedded volitional- epistemic divide and discovered how sophisticated the interplay is between evaluative and assertive. This topic and many questions omitted or left open will be the subject of future work.

Notes 1. SoaresBarbosa (1822): “O Subjunctivo, ou Conjuntivo he um modo dependente de outro verbo claro, ou occulto . . . ; sem o qual não faz sentido, nem póde estar na oração” (‘The subjunctive, or conjunctive is a mood dependent on another verb visible, or hidden . . . without which neither it has any sense nor it can occur in the clause’). 2. Faria (1974, p. 88): “não há classes defi nidas de verbos que implicam Conjuntivo. Existem, sim, graduações entre os verbos que implicam Conjuntivo” (‘there are no defi ned classes of verbs that imply conjunctive. There are rather gradations among the verbs that imply conjunctive’). 3. Unlike subjunctive, which is usually regarded as an embedded mood, the imperative mood is taken to be restricted to root clauses. The Portuguese imperative mood only occurs in root clauses, but its use is restricted to 2SG; in deferent 2SG, 1PL, (non)deferent 2PL, and negative imperatives the subjunctive is chosen (cf. (18) below), in which case it can be embedded. Other languages allow root subjunctives, e.g., Russian and Greek (Mezhevich 2006) and embedded imperatives, e.g., Slovenian (Dvořák and Zimmermann 2007). 4. The translations provided are approximate. In both (13a) and (13a’), where assim ‘so’ occurs, rendering inversion optional, the speaker believes the realization of the event will bring the expected consequences, much as in conditionals (the only thing he needs to be successful in his studies is to take the courses). This type of structure appears in idiomatic expressions: (i)

Queira Deus! want.sbjv.prs.3sg God ‘God willing!’

chapter four

166

(ii)

Assim seja! So be.sbjv.prs.3sg ‘So be it!’

5. In section 5 I argue that defective assertive also plays a role in the licensing of subjunctive. 6. The evaluative interpretation in the (a) sentences of (13)–(15) is not to be confused with focus (Ambar 1999b). It could be focus, but then the sentences would lose the strong evaluative intonation (exclamative-like) and the speaker would not express her own convictions about the realization of the event: just the subject would be focused, being contrasted with other possible agents, very close to subjunctive imperatives. For example, with a focus interpretation (14a) would stand for ‘THEY should follow your advice, not me/not Peter.’ 7. Here emphatic affi rmation is given by repeating the verb at the end of the sentence (i); emphatic negation by repeating the negation operator (ii): (i)

Sigam eles o teu conselho Follow.SUBJ3PL they the your advice sigam! follow.SUBJ.3PL [speaker’s disapproval; presupposition: bad advice)] ‘They should not follow your advice.’

(ii)

Não sigam eles o teu conselho não! Not follow.SUBJ.3PL they the your advice not. [speaker’s approval; presupposition: good advice)] ‘They should follow your advice!’ [it is mandatory that they do]

8. The presence of the complementizer strengthens the wish, rendering the clause closer to commitment. 9. (i) illustrates Giannakidou’s paradigm (Chap. 5 in this volume): (i)

a. Efije o Nicholas. left.3sg the Nicholas

(veridical, perfective past)

‘Nicholas left. ’ b. *Na efije o Nicholas. c. Isos/pithanon (na) efije oNicholas. Maybe/possibly subj left.3sg the Nicholas ‘Maybe Nicholas left.’

On Finiteness and the Left Periphery

167

10. For imperatives in this group, see Kempchinsky (1986, 2009). 11. Some speakers still fi nd that (22) and (24) are not fully acceptable. Rita Veloso observes that an adversative contrast improves subjunctive in this context: (i)

? Sei que talvez vos know.ind.prs.1sg that maybe (it) you.dat parecer arrogância, mas . . . seem arrogance, but . . .

possa may.sbjv.3sg

Plausibly, this is a case of contrariness provided by evaluation. Conjunctions in concessive clauses seem to lexically codify this property, conceding that a fact is integrated in the conversational context (cf. Quer’s [1998] example in (43)): (ii)

Ainda que/embora estej a chover, yet that/though be.sbjv.prs.3sg prep. raining, vou sair go.ind.prs.1sg outside

There is an evidential-flavor associated with these structures (Allegedly it is raining, but I’ll go outside). 12. Portuguese speakers (regional use) produce sentences like: Ela está doente. Não m’acredito ‘She is sick. I don’t believe myself’, where m’ (me) is the clitic 1sg. Quer (2009) reports equivalent facts in Catalan. 13. Giannakidou and Staraki (2013) observe that Greek lei ‘say’ is an assertive verb if it co- occurs with oti (assertive complementizer), but acquires a volitional meaning when associated with na (subjunctive) (see discussion under their (50)). The authors also observe that “these might be explained in terms of lexical ambiguity,” but they “believe it is more profitable to envision these systematic correlations . . . as interactions where the syntax affects the meaning” (p. 266). 14. Control by the object is possible, but null subject restrictions related to discourse show up. 15. Throughout the text I assume the presence of an operator, but see the observation under (33). Operators have been claimed to be involved in the analysis of subjunctive also in selected contexts (e.g., Kempchinsky 1986; 2009). For Manzini (2000) the subjunctive is an indefi nite T licensed in a head-to-head dependency bound by an intensional operator, in line with Tsoulas (1995). 16. In Picallo (1984) Infl is [±T, ±Agr]. In Raposo (1987) T is not always present in Infl, but a Tense operator occurs in C in complements to epistemics. As for Agr, my proposal was in line with Borer (1989). 17. Mezhevich (2006) examines Russian subjunctive based on the works cited in the text and on Iatridou (2000). Tense, aspect, and mood are dyadic predicates, each of which relates two temporal arguments through a relation of (non) coincidence ([±coin]). Features can enter the derivation as [±interpretable] and

chapter four

168

[±valued], as in Pesetsky and Torrego (2004). I am unable to discuss the details of this thesis, to which I only had access in the review process. 18. See also Adger (2003) and Pires (2006), among others. 19. I am assuming these operators have t-features, a natu ral assumption if one thinks of the relation between Tense and, for instance, negation operators (as in Kayne 1975; also Laka 1990; Zanuttinni 1991). 20. Marques (2004) notes that with duvidar ‘doubt’ the subjunctive can shift into indicative if the sentence is negated: Duvido que ele venha vs. Não duvido que ele vem ‘I doubt that he come.sbjv.3sg’ vs. ‘I don’t doubt that he come. ind.3sg’. Note, however, that duvidar ‘doubt’ can take indicative in the absence of negation, whereas ‘want’ never can: (Até) duvido de que existo. (David Mourão Ferreira) ‘I (even) doubt that I exist.ind1sg.’ Yet Marques’s facts are interesting, recalling the positive- negative puzzle of evaluation-assertion we observed for matrix clauses above. 21. But Mezhevich (2006, pp.  152–154) concludes that čtoby is a single lexical item (no element can intervene between čto and by). In my analysis, if čto does not move to EvaluativeP (the highest projection responsible for clause typing), adjoining by, the illocutionary force feature remains unvalued, leading to a crash, thus predicting that they cannot be stranded. 22. Martineau (1994, p. 57) observes: “This hypothesis is in the line of other proposals made for Romance languages: among others, Motapanyane (1992) and Rivero (1987) for Rumanian, Terzi (1992) for the Balkan languages, Ambar (1993) for Portuguese and Pollock (1993) for French, all of them arguing for an extended structure of the embedded functional categories.” 23. Not necessarily the real world (as is the case of fiction verbs; see Farkas 1992; Quer 1998; among others). 24. The defi nition is given in Giannakidou (2009, (45)): (1)

Epistemic model of an individual I An epistemic model M(i) ∈ M is a set of worlds associated with an individual i representing worlds compatible with what i believes or knows.

(2)

Truth in an epistemic model A proposition p is true in an epistemic model M(i) iff M(i) ⊆ p: ∀w [w ∈ M (i) → w ∈ λw’. p (w’)

(3)

a. John won the race. b. [[John won the race]] = 1 iff ∀w: w ∈ M(speaker) →w 0w’. John won the race in w’.

25. In Quechua the source of the information conveyed by a sentence (personal experience, various types of evidence, or hearsay) is marked by evidential morphemes (Speas, 2004, p. 253, her example (2)):

On Finiteness and the Left Periphery

169

(i) wañu-nqa-paq-mi ‘It will die (I assert)’ (ii) wañu-nqa-paq-shi ‘It will die (I was told)’ (iii) wañu-nqa-paq- chi ‘It will die (perhaps)’. 26. The supposed correspondence would be the following (Speas 2004, p. 256): (i) Cinque’s projections: Speech Act MoodP > Evaluative MoodP > Evidential Mood > Epistemic ModP; (ii) Logophoric category: speech > thought > know > direct perception; (iii) Evidential category: hearsay > indirect evidence > direct evidence > personal experience. 27. EvaluativeP was first proposed at the “International Workshop on Focus,” held in Paris in 1996 and partially published in Ambar 1999b; the label was inspired by Barwise and Cooper’s (1981) pure quantifiers vs. evaluative quantifiers. Though independently motivated, (39) shares with Rizzi’s left periphery the position of XPs*–topic-like positions, which I did not label topic owing to my suspicion that different properties correspond to the different positions –, Wh, and Focus ordering. Rizzi’s Fin was absent from my system (whether Assertive and Fin overlap is still unclear, and I have been adopting Fin); Rizzi’s ForceP is split in (39), both interpretation and clause typing being compositional. 28. I thank Anastasia Giannakidou for these observations. As for whether AssertiveP could be MoodP, the answer involves other issues, discussed below: do MoodP-ForceP and MoodP-FinP overlap? I leave this open. 29. Though Rizzi’s left periphery dispense with any introduction, its representation is given in (i): (i) FORCE (TOP*) INT (TOP*) FOC (TOP*) FIN TP 30. In Rizzi (1997, 2004), Finº is associated with fi niteness and modality; finiteness stands for phi-features and tense. For some scholars, phi-features are associated with Force (see Alboui 2010, for Romanian), and tense is a property of Fin. 31. Ambar and Vasconcelos (2012) argue that verbs come from the lexicon as bundles of features: predicates that select subjunctive carry a given feature (labeled ∫) that verbs taking indicative do not carry. I am assuming here that this feature is evaluative, represented by the β-feature throughout the text. For a similar approach see Baunaz (2014). 32. Emonds (1976) acknowledges the observations of Hooper and Thompson (1973) on root phenomena in embedded assertions. A matter for research is whether the notion of root “unselected fi nite IP are always Discourse Projections” and the generalized Tensed S Constraint (“A trace inside an IP sister of Zº must be bound within a maximally extended projection of Zº)” (Emonds 2012, pp. 27 and 31, respectively) can be made compatible with my approach, accounting for cross-linguistic variation. 33. In line with Kayne (1982), I have long assumed (Ambar 1992) that complementizers are nominal, as does Roussou (2010); still, she notes a difference between pu and other complementizers: pu has operator-like localizing properties

chapter four

170

(similar ‘to the relative/wh- adverb ‘where’’). We could hypothesize that this property enables pu to value T on D. 34. In this analysis, the incompleteness of Assertive in subjunctive is responsible for restrictions on topic and focus (recall H&P’s observations above). 35. The author concludes that it is not in fact intensionality that is relevant: “Rather, it is simply that the matrix predicate expresses an alternative world or set of worlds from that obtaining at the temporal reference point for that matrix predicate.” 36. Much as operators in relatives are identified by their antecedents (Chomsky 1986). 37. The sequence present-past is ungrammatical: *Peço-lhes que viessem. SUBJ.past. A possible explanation can rely on the relation of embedded Assertive with matrix Assertive (see Iatridou 1993). 38. The phenomenon exists in Spanish as well; see Quer 2006, ex. (33), and Kempchinsky 2009, n9. 39. Varlokosta (1993) implements the tense dependency of F-subjunctives by requiring that the embedded V rise (through T) to C, as Landau (2004) observes. According to Landau (2004), the subjunctives we have been dealing with are F-subjunctives, those where independent temporal adverbs may occur in the matrix and in the embedded domain. Landau (2014) considers these to be NC (non- control), [+T+Agr] structures, while sentences like (i) are OC (obligatory control) [−T+Agr]: (i) O Yanis the John.NOM Giorgos) George.NOM)

kseri knows

na sbjv

kolimbai swim.3SG

(*o (*the

‘John knows how to swim.’ 40. In Giannakidou (2009), TP is an unsaturated inflectional domain, the highest inflectional category being Mood, where na is merged; subjects either move to spec.MoodP (predicting they precede na) or remain in situ, VP internal (predicting the post-verbal position); Neg (min) may intervene between MoodP and TP.

References Adger, D. 2003. Core Syntax: A Minimalist Approach. New York: Oxford University Press. Alboiu, G. 2002. The Features of Movement in Romanian. Bucharest: EUB. ———. 2010. A-Probes, Case, and (In)Visibility. http://ling.auf.net/lingBuzz/001163. Ambar, M. 1988. Para uma sintaxe da inversão sujeito-verbo em Português. PhD diss., University of Lisbon. ———. 1992. Temps et structure de la phrase. In Structure de la Phrase et Théorie du Liage, ed. H. G. Obenauer and A. Zribi-Hertz, 29–49. Paris: Presses Universitaires de Vincennes.

On Finiteness and the Left Periphery

171

———. 1993. Functional categories, selection, control and checking theory. Paper presented at the Second Workshop on the Syntax of Central Romance Languages–Abstracts, Autonomous University of Barcelona, Spain. ———. 1998. Inflected infi nitives revisited Genericity and single event. Journal of Canadian Linguistics 43: 5–36. ———. 1999a. Infi nitives vs. participles. In Semantic Issues in Romance Syntax. Current Issues in Linguistic Theory, ed. E. Treviño E and J. Lema, 1–20. Amsterdam: John Benjamins. ———. 1999b. Aspects of focus in Portuguese. In The Grammar of Focus, ed. L. Tuller and R. Rebuschi, 23–53. Amsterdam: John Benjamins. ———. 2000. Wh- questions and wh- exclamatives: Unifying mirror effects. In Romance Languages and Linguistic Theory, ed.  C. Beyssade,  R. BokBennema, F. Drijkoningen, and P. Monachesi, 15–40. Current Issues in Linguistics Theory. Amsterdam: John Benjamins. ———. 2003. Wh- asymmetries. In Asymmetry in Grammar, ed. A. M. Di Sciullo, 209–249. Amsterdam: John Benjamins. ———. 2005. V-to- C, tense domains and conditions on merge and move. Paper presented at GLOW in Asia V Abstracts. Jawaharlal Nehru University, New Delhi, India. ———. 2007. Verb movement and tense. EPP and t- completeness. In Proceedings of the XXXII Incontro di Grammatica Generativa, ed. C. Picchi and A. Pona, 1–20. Firenze: Edizione dell’ Orso. Ambar, M., and A. Jimenez. 2014. Subjunctives: How much left periphery do you need? Paper presented at the Olomouc Linguistics Colloquium, Olomouc, Czech Republic. Ambar,  M., and  M. Vasconcelos. 2012. Conjuntivo em Português: Revisitando a análise de Isabel Hub Faria. In Nada na linguagem lhe é estranho— homenagem a Isabel Hub Faria, ed. A. Costa and I. Duarte, 533–551. Porto: Edições Afrontamento. Baker, C. L. 1970. Double negatives. Linguistic Inquiry 1: 169–186. Barwise, J., and R. Cooper. 1981. Generalized quantifiers and natural language. Linguistics and Philosophy 4(2): 159–219. Baunaz,  L. 2014. Zooming on predicates selecting the subjunctive mood in French: The emotive factor. Paper presented at LSRL44, London, Ontario, Canada, May 2–4. Bianchi, V. 2001. On person agreement. Manuscript, Scuola Normale Superiore, Pisa, Italy. Borer, H. 1989. Anaphoric AGR. In The Null Subject Parameter, ed. O. Jaeggli and K. J. Safi r, 69–109. Dordrecht: Kluwer Academic Borgonovo,  C. 2003. Mood and focus. In Romance Language and Linguistic Theory, ed. J. Quer, M. Scorretti, P. Sleeman, and E. Verheugd, 17–30. Current Issues in Linguistic Theory. Amsterdam: John Benjamins.

172

chapter four

Bošković, Z. 1997. The Syntax of Nonfi nite Complementation: An Economy Approach. Cambridge, MA: MIT Press. Chomsky, N. 1986. Barriers. Cambridge, MA: MIT Press. ———. 2001. Derivation by phase. In Ken Hale: A Life in Language, ed.  M.  J. Kenstowicz, 1–52. Cambridge, MA: MIT Press. ———. 2007. Approaching UG from below. In Interfaces + Recursion = Language? Chomsky’s Minimalism and the View from Syntax- Semantics, ed. U. Sauerland and H. M. Gärtner, 1–29. Berlin: Mouton de Gruyter. ———. 2008. On phases. In Foundational Issues in Linguistic Theory: Essays in Honor of Jean- Roget Vergnaud, ed. R. Freidin, C. P. Otero, and M. L. Zubizarreta, 133–167. Cambridge, MA: MIT Press. Christodoulou, C., and M. Wiltschko. 2012. Function without content: Evidence from Greek Subjunctive na. In Towards a Biolinguistic Understanding of Grammar: Essays on Interfaces, ed. A. M. Di Sciullo, 117–140. Amsterdam: John Benjamins. Cinque,  G. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective. New York: Oxford University Press. Cornilescu, A. 2000. The double subject construction in Romanian. In Comparative Studies in Romanian Syntax, ed. V. Motapanyane, 83–134. Oxford: Elsevier. D’Alessandro, R., and A. Ledgeway. 2010. At the C-T boundary: Investigating Abruzzese complementation. Lingua 120: 2040–2060. Demirdache,  H., and  M. Uribe-Etxebarria. 2000. Tenses, aspects and time adverbs as spatio-temporal predicates. Paper presented at the International Round Table on the Syntax of Tense and Aspect, Université Paris 3, November. Dobrovie- Sorin, C. 1987. Syntaxe du Roumain. PhD diss., Université Paris 7. ———. 1994. The Syntax of Romanian. Berlin: Mouton de Gruyter. ———. 2001. Head-to-Head merge in Balkan subjunctives and locality. In Comparative Syntax of Balkan languages, ed. M. L. Rivero and A. Ralli, 44–73. New York: Oxford University Press. Dvořák,  B., and  I. Zimmermann. 2007. The Slovenian imperative in embedded clauses. In Linguistic Investigations into Formal Description of Slavic Languages, ed. P. Kosta and L. Schürcks, 171–184. Contributions to the Sixth European Conference held at Potsdam University, November 30–December 2, 2005 (Potsdam Linguistic Investigations 1). Frankfurt am Main: Peter Lang. Emonds,  J. 1970. Root and structure-preserving transformations. PhD diss., Massachusetts Institute of Technology. ———. 1976. A Transformational Approach to English Syntax. New York: Academic Press. ———. 2012. Augmented structure preservation and the Tensed S Constraint. In Main Clause Phenomena– New Horizons, ed.  L. Aelbrecht,  L. Haegeman, and R. Nye, 23–46. Amsterdam: John Benjamins.

On Finiteness and the Left Periphery

173

Enç, M. 1991. On the absence of the present tense morpheme in English. Manuscript, University of Wisconsin, Madison. Faria, I. H. 1974. Conjuntivo e a restrição da frase mais alta. Boletim de Filologia 23: 81–188. Farkas,  D. 1992. On the semantics of subjunctive complements. In Romance Languages and Modern Linguistic Theory, ed.  P. Hirschbühler and  K. Koerner, 71–104. Current Issues in Linguistic Theory. Amsterdam: John Benjamins. Felix, S. 1989. Finite infi nitives in Modem Greek. In Syntactic Phrase Structure Phenomena in Noun Phrases and Sentences, ed. C. Bhatt, E. Löbel, and C. Schmidt, 113–131. Amsterdam: John Benjamins. Gallego, Ángel J. 2007. Phase theory and parametric variation. PhD diss., University of Barcelona. Giannakidou, A. 1994. The semantic licensing of NPIs and the Modern Greek subjunctive. Language and Cognition 4. Yearbook of the Research Group for Theoretical and Experimental Linguistics. University of Groningen. ———. 1998. Polarity Sensitivity as (Non)veridical Dependency. Amsterdam: John Benjamins. ———. 1999. Affective dependencies. Linguistics and Philosophy 22: 367–421. ———. 2009. The dependency of the subjunctive revisited: Temporal semantics and polarity. Lingua 120: 1883–1908. Giannakidou, A., and E. Staraki. 2013. Ability, action, and causation: From pure ability to force. In Genericity, ed.  A. Mari,  C. Beyssade, and  F. Del Prete, 250–275. Oxford: Oxford University Press. Giorgi, A., and F. Pianesi. 1997. Tense and Aspect: From Semantics to Morphosyntax. Oxford: Oxford University Press. Grevisse, M. 1979. Complement selection and the lexicon. Linguistic Inquiry 10: 279–326. ———. 1990. Argument Structure. Cambridge, MA: MIT Press. ———. 1993. Le bon usage: Grammaire française. Paris: Duculot. Haegeman, L., and V. Hill. 2011. The syntacticization of discourse. Manuscript, Ghent University and University of New Brunswick Saint John. Hill,  V. 2013. The emergence of the Romanian subjunctive. Linguistic Review 30(4): 1–37. Hooper, J. B., and S. A. Thompson. 1973. On the applicability of root transformations. Linguistic Inquiry 4: 465–497. Iatridou,  S. 1993. On nominative case assignment and a few related things. MIT Working Papers in Linguistics 19: 175–198, Massachusetts Institute of Technology. ———. 2000. The grammatical ingredients of counterfactuality. Linguistic Inquiry 31: 231–270.

174

chapter four

Jakubowicz, C. 1984. Do binding principles apply to INFL? In NELS 15, ed. S. Berman,  J.  W. Choe, and  J. McDonough, 188–206. Amherst, MA: GLSA Publications. Karttunen, L. 1977. Syntax and semantics of questions. Linguistics and Philosophy 1: 3–44. Kayne, R. 1975. French Syntax. The Transformational Cycle. Cambridge, MA: MIT Press. ———. 1982. Arguments and predicates. GLOW Newsletter 8: 24. ———. 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press. ———. In press. Comparative syntax and English is. Linguistic Analysis. Kempchinsky, P. 1986. Romance subjunctive clauses and logical form. PhD diss., University of California, Los Angeles. ———. 2009. What can the subjunctive disjoint reference effect tell us about the subjunctive? Lingua 119: 1788–1810. Kiparsky, P., and C. Kiparsky. 1970. Fact. In Progress in Linguistics: A Collection of Papers, ed.  M. Bierwisch and  K.  E. Heidolph, 143–173. The Hague: Mouton de Gruyter. Laka,  I. 1990. Negation in syntax: On the nature of functional categories and projections. PhD diss., Massachusetts Institute of Technology. Landau, I. 2004. The scale of fi niteness and the calculus of control. Natural Language and Linguistic Theory 22: 811–877. ———. 2014. A Two-Tiered Theory of Control. http:// ling.auf.net / lingbuzz /001937. Lasnik,  H., and  J. Uriagereka. 2005. A Course in Minimalist Syntax: Foundations and Prospects. London: Blackwell. Manzini, M. R. 2000. Sentential complementation: The subjunctive. In Lexical Specifi cation and Insertion, ed. P. Coopmans, M. Everaert, and J. Grimshaw, 241–267. Amsterdam: John Benjamins. Marques,  R. 2004. On the system of mood in Eu ropean and Brazilian Portuguese. Journal of Portuguese Linguistics 3: 89–109. ———. 2009. On the selection of mood in complement clauses. In Crosslinguistic Semantics of Tense, Aspect and Modality, ed. L. Hogeweg, H. de Hoop, and A. Malchukov, 179–204. Amsterdam: John Benjamins. Martin, R. A. 1996. A minimalist theory of PRO and control. PhD diss., University of Connecticut. Martineau, F. 1994. The expression of the subjunctive in older French. Catalan Working Papers in Linguistics (CatWPL) 3(2): 45–69, Universitat Autònoma de Barcelona. Mezhevich, I. 2006. Featuring Russian tense: A feature-theoretic account of the Russian tense system. PhD diss., University of Calgary. Moignet, G. 1984. Grammaire de l’Ancien Français. Paris: Klincksieck Motapanyane, V. 1992. On subjunctive morphology and its interaction with syntax. Manuscript, University of New Brunswick.

On Finiteness and the Left Periphery

175

Pesetsky, D., and E. Torrego. 2001. T-to- C movement: Causes and consequences. In Ken Hale: A Life in Language, ed. M. Kenstowicz, 355–426. Cambridge, MA: MIT Press. ———. 2004. The syntax of valuation and the interpretability of features. Manuscript, University of Massachusetts-Boston. Picallo, C. 1984. The infl node and the null subject parameter. Linguistic Inquiry 15: 75–102. Pires, A. 2006. The Minimalist Syntax of Defective Domains: Gerunds and Infi nitives. Amsterdam: John Benjamins. Pollock, J. Y. 1989. Verb movement, universal grammar, and the structure of IP. Linguistic Inquiry 20: 365–424. ———. 1993. Notes on clause structure. Manuscript, Université Paris. Portner, P. 1997. The semantics of mood, complementation, and conversational force. Natural Language Semantics 5: 167–212. Quer, J. 1998. Mood at the Interface. The Hague: Holland Academic Graphics. ———. 2006. Subjunctives. In The Blackwell Companion to Syntax, ed. M. Everaert and H. Riemsdijk, vol. 4, 660–684. Oxford: Blackwell. ———. 2009. Twists of mood: The distribution and interpretation of the indicative and the subjunctive. Lingua 119: 1779–1787. Raposo, E. P. 1985. Some asymmetries in the binding theory in romance. Linguistic Review 5: 75–109. ———. 1987. Case theory and infl-to- comp: The inflected infi nitive in Eu ropean Portuguese. Linguistic Inquiry 18: 85–109. ———. 1989. Prepositional infi nitival constructions in Eu ropean Portuguese. In The Null Subject Parameter, ed. O. Jaeggli and K. Safi r, 277–305. Dordrecht: Kluwer. Rivero, M. L. 1987. Barriers and the null subject parameter in Modern Greek. In Proceedings of the North Eastern Linguistic Society 18, ed. J. Blevins and J. Car ter, vol. 2, 412–425. Amherst: University of Massachusetts. ———. 1994. Clause structure and V-movement in the languages of the Balkans. Natural Language and Linguistic Theory 12: 63–120. Rizzi, L. 1997. The fi ne structure of the left periphery. In Elements of Grammar, ed. L. Haegeman, 281–339. Dordrecht: Kluwer. ———. 2004. Locality and left periphery. In Structures and Beyond: The Cartography of Syntactic Structures 3, ed. A. Belletti, 223–252. Oxford: Oxford University Press. Ross, J. R. 1970. On declarative sentences. In Readings in English Transformational Grammar, ed. R. A. Jacobs and P. S. Rosenbaum, 222–272. Washington: Georgetown University Press. Roussou, A. 2000. On the left periphery; modal particles and complementizers. Journal of Greek Linguistics 1: 63–93. ———. 2010. Selecting complementizers. Lingua 120(3): 582–603.

176

chapter four

Ruwet, N. 1991. Je veux partir./*Je veux que je parte. On the distribution of fi nite complements and infi nitival complements in French. In Syntax and Human Experience, ed. and trans. J. Goldsmith, 1–55. Chicago: University of Chicago Press. San Martin,  I. 2007. Beyond the infi nitive vs. subjunctive rivalry: Surviving changes in mood. In Coreference, Modality, and Focus: Studies on the Syntax– Semantics Interface, ed. L. Eguren and O. Fernández Soriano, 171– 190. Amsterdam: John Benjamins Publishing. Schlenker, P. 2005. The lazy Frenchman’s approach to the subjunctive (speculations on reference to worlds, presuppositions, and semantic defaults in the analysis of mood). In Romance Languages and Linguistic Theory 2003: Selected Papers from Going Romance 2003, ed.  T. Geerts,  T. Ginneken, and H. Jacobs, 269–309. Amsterdam: John Benjamins. Searle, J. 1969. Speech Acts. New York: Cambridge University Press. Soares-Barbosa,  J. 1822. Grammatica Philosophica da Lingua Portugueza ou Principios da Grammatica Geral applicados á nossa linguagem. Lisbon: Typographia da Academia Real das Sciencias. Speas, M. 2004. Evidentiality, logophoricity and the syntactic representation of pragmatic features. Lingua 114: 255–277. Speas, M., and C. Tenny. 2003. Configurational properties of point of view roles. In Asymmetry in Grammar, ed. A. M. Di Sciullo, 315–344. Amsterdam: John Benjamins. Stowell, T. 1982. The tense of infi nitives. Linguistic Inquiry 13: 561–570. ———. 1993. Syntax of tense. Manuscript, University of California, Los Angeles. Terzi, A. 1992. PRO in fi nite clauses: A study of the inflectional heads of the Balkan languages. PhD diss., New York: City University of New York. Torrego, E., and J. Uriagereka. 1992. Indicative dependents. Manuscript, University of Massachusetts-Boston and University of Maryland. Tsoulas, G. 1995. The nature of the subjunctive and the formal grammar of obviation. In Grammatical Theory and Romance Languages, ed.  K. Zagona, 293–306. Amsterdam: John Benjamins. Uriagereka, J., and A. J. Gallego. 2007. Subjunctive dependents. Paper presented at XVII Colloquium on Generative Grammar, June 13–15, Girona, Spain. Varlokosta, S. 1993. Control in modern Greek. In University of Maryland Working Papers in Linguistics 1, ed. C. Mason, S. Powers, and C. Schmitt, 144–163. College Park, MD. Zagona, K. 1990. Times as temporal argument structure. Manuscript, University of Washington. Zanuttinni, R. 1991. Syntactic properties of sentential negation. A comparative study of romance languages. PhD diss., University of Pennsylvania.

chapter five

Evaluative Subjunctive and Nonveridicality Anastasia Giannakidou

1

Introduction: The Landscape of Subjunctive

T

he study of grammatical mood has a long tradition in philology and linguistic semantics. Typologically, we fi nd morphological distinctions such as indicative, subjunctive, optative, and imperative. Focusing on the contrast between the subjunctive and the indicative, traditional grammars typically identify the indicative with realis and the subjunctive with irrealis, and describe the contrast in terms of selection. Observe the basic contrast with attention to French: (1)

a. Marc sait que le printemps est/ *soit arrivé. Marc knows that the spring be-ind-3sg / be-sbjv.3sg arrived ‘Marc knows that spring has arrived.’ b. Marc veut que le printemps soit/ *est long. Marc wants that the spring be-sbjv-3sg / be-ind-3sg long ‘Marc wants spring to be long.’ c. Le printemps est/ *soit arrivé. The spring be-ind-3sg/ be-subj-3sg arrived ‘Spring has arrived.’

178

chapter five

The verb savoir ‘know’ in (1a) is said to select the indicative, but the volitional verb vouloir ‘want’ in (1b) selects the subjunctive. The indicative is the default mood of unembedded declaratives, as in (1c). In both unembedded clauses and in the complements of ‘knowledge’ verbs, the indicative refers to a true event; hence the characterization realis. The complement of a desire-verb, on the other hand, merely expresses a desire, and the content of a desire is not a fact (hence irrealis). This is a typical pattern, and in strict selection, the indicative and subjunctive are in complementary distribution: one mood excludes the other. Though the indicative-subjunctive pattern has been most extensively described for Indo-European languages, it is by no means restricted to these, and it appears in many of the world’s languages, including Native American languages (see Matthewson 2010 for Salish; and Wiltschko, this volume). The contrast between subjunctive and indicative also correlates with evidentiality, especially in languages that have only one indirect evidential morpheme (Murray, forthcoming; Smirnova 2013). In this case, the indirect evidential is used when the speaker has reduced commitment to the truth of the sentence; therefore the indirect evidential form appears to be parallel to the subjunctive. I do not discuss indirect evidentials in this chapter, but the framework I establish, in particular the category of epistemic subjunctive, is very relevant for the indirect evidential. When we compare the subjunctive and indicative, we observe two patterns: (a) one that involves selection by particular classes of verbs and other elements (e.g., sentential connectives such as those meaning without, before), and (b) one in which the speaker has a choice between indicative and subjunctive. In this chapter I study cases that fall under (b). My goal is to show that while selection manifests sensitivity of the subjunctive to the logical property of nonveridicality, the optional cases reveal a major function of the subjunctive itself to create nonveridical modal spaces. Most of the optional subjunctives I discuss here are translated into English with possibility modals, and with similar modal particles in Dutch and German. My main language of illustration is (Modern) Greek. Unlike French and other Romance languages, and in contrast to Ancient Greek, the mood contrast in contemporary Greek is manifested not as a verbal morphology, but in the form of particles. This pattern is observed also in Balkan (Slavic) languages and Romanian (Farkas 1985; Rivero 1994; Giannakidou 1998, 2009; Roussou 2000; Bulatovic 2008; Todorovic 2012; among others). The subjunctive particle is na, which precedes the tensed

Evaluative Subjunctive and Nonveridicality

179

verb. The indicative is unmarked in main clauses; that is, past tenses (simple past, imperfective past, perfect tenses) and the present are ‘indicative’. (One cannot fail to note the parallel with the ‘direct’ evidential, which also is typically an unmarked past or present). In embedded clauses, the indicative particles oti and pu are used as complementizers. The na clause contains the so- called verbal dependent form (Holton et al. 1997) ‘perfective non-past’: (2)

Thelo na kerdisi o Janis. want.1sg sbjv win-prf-nonpst-3sg the John ‘I want John to win.’

(3)

O Pavlos kseri oti efije i Roxani. the Paul knows-3sg that-ind left-3sg the Roxani ‘Paul knows that Roxanne left.’

(4)

Efije/ Fevgi/ *fiji i Ariadne. left.3sg/leave-ipfv-nonpst-3sg/prf-pst-3sg the Ariadne ‘Ariadne left.’ ‘Ariadne is leaving.’

The complement sentence in (2) is in the subjunctive mood, signaled by na. The verbal form used is glossed as ‘perfective non-past’. It is a form that cannot occur without na, as we see in (4). This form designates future orientation, occurring also with the future particle tha (Giannakidou 2009, 2012; Giannakidou and Mari 2014, this volume). I do not discuss the properties of this form here, but focus instead on the choice of particle. Greek possesses another mood particle—as for optative mood— which is used in main clauses only. Na also has this use: (5)

As fiji/ efevge o Janis! opt leave-perf-nonpst-3sg/ left-imprf-3sg the John ‘Let John leave!’ ‘I wish John had left!’

(6)

Na fiji o Janis! subjv leave-prf-nonpst-3sg the John

chapter five

180

‘Let John leave!’ ‘John, leave!’ The use of as is reminiscent of if only optatives with past (see Grosz 2012 for discussion). In main clauses, then, only the non-indicative is marked; the indicative is the absence of mood particles. This correlates, as I said earlier, with the use of the so- called indirect evidential marking in Turkish, Bulgarian, and Cheyenne, where the absence of an indirect evidential correlates with the unmarked ‘direct evidential’ form. With emotive verbs, Greek uses the indicative complementizer pu: (7)

O Pavlos lipate {pu/*oti} efije i Roxani. the Paul is-sad-3sg that left-3sg the Roxani ‘Paul regrets that Roxanne left.’

Pu marks the complement formally as distinct from oti. Pu follows emotive verbs lipame, metaniono ‘regret’, and xerome ‘be glad’ (see Christidis 1981; Varlokosta 1994), which are claimed to be factive; however, epistemic factive verbs such as ksero ‘know’ select the indicative, as we saw in (1). I will say more about pu later in the chapter. Importantly, in many Romance languages, such as Spanish and Italian, the emotive verb requires subjunctive (example (8) is from Villalta 2008, p. 470): (8)

Marcela se alegra de que la hayan invitado. Marcela SE glad-3sg of that her have-pst-subjv-3pl invited. ‘Marcela is glad that they invited her.’

As Quer (2001, 2009), Marques (2004, 2010), and Ambar (this volume) discuss, the use of the subjunctive after emotive verbs in Romance languages is subject to considerable variation, with some languages being stricter (Spanish, Italian), others allowing both subjunctive and indicative (Catalan, French, Portuguese), and Romanian using the indicative, like Greek. When we consider the subjunctive after emotive verbs in Romance languages, therefore, it seems reasonable to understand it as a case of ‘variable’ subjunctive, not strict selection. Regarding strict selection, and putting emotive verbs aside, the main Greek patterns are very similar to the ones we fi nd in Romance languages, as follows:

Evaluative Subjunctive and Nonveridicality

(9)

181

Indicative verbs assertives: leo ‘say’, dhiavazo ‘read’, isxirizome ‘claim’ fiction verbs: onirevome ‘to dream’, fandazome ‘imagine’ epistemics: pistevo ‘believe’, nomizo ‘think’.1 factive verbs: xerome ‘be glad’, gnorizo ‘know’, metaniono ‘regret’

Thus in terms of mood, complements of belief and fiction verbs behave like unembedded assertions and complements of knowledge verbs; they select indicative: (10)

O Nicholas onireftike/ nomize oti/*na efije i the Nicholas dreamt-3sg / thought-3sg that-ind left-3sg the Ariadne. Ariadne ‘Nicholas dreamt/thought that Ariadne left.’

This pattern is challenging if we believe that the indicative implies ‘truth in the actual world’, because complements of belief, fiction, and assertive verbs are not true in this sense. Of the indicative complements, only complements of know refer to facts (Karttunen 1971; Kiparsky and Kiparsky 1970). But the grammar of mood selection appears to make no distinction between actual events and imagined or believed facts. Verbs selecting subjunctive belong to the following classes: (11)

Subjunctive verbs volitionals: directives: modal verbs: permissives: negative:

thelo ‘want’, elpizo ‘hope’, skopevo ‘plan’ dhiatazo ‘order’, simvulevo ‘advise’, protino ‘suggest’ prepi ‘must’, bori ‘may’ epitrepo ‘allow’; apagorevo ‘forbid’ apofevgho ‘avoid’, arnume ‘refuse’

Some of these verbs contain a volitional component, but not all of them do; for example, modal verbs in both epistemic and dynamic uses select the subjunctive (and this holds for Greek as well as for Romance languages). We also observe selection patterns with adjuncts. Nonveridical adjuncts such as prin ‘before’ and xoris ‘without’ select na, but meta ‘after’ selects the indicative:

chapter five

182

(12)

a. Prin na kimithi, epline ta dontia tu. before subjv sleep-prf-nonpst-3sg washed-3sg the teeth his ‘Before sleeping, he brushed his teeth.’ b. *Prin kimithike, epline ta dontia tu. before sleep-prf-pst-3sg wash-prf-pst-3sg the teeth his

(13)

a. *Afu na kimithi, figame. after subjv sleep-prf-nonpst-3sg left-prf-pst-3sg b. Afu kimithike, ‘After he slept we left.’

(14)

figame.

a. Ekane ti metafrasi xoris na xrisimopiisi Did-3sg the translation without subjv use-prf-nonpst-3sg lexiko. dictionary ‘He did the translation without using a dictionary.’ b. * Ekane ti metafrasi xoris xrisimopiise lexiko. Did.3sg the translation without use-prf-pst-3sg dictionary

The use of the subjunctive with ‘without’ and ‘before’ challenges the idea that subjunctive is dependent on preference of contextual alternatives (Villalta 2008). There is no preferential component in these connectives and the subjunctive is selected, as we noted, also with mere possibility modals that contain no order and no preference at all. In selection, the subjunctive itself does not appear to add to the semantics, and it is generally accepted that it reflects syntactic dependency (see, e.g., Ambar, this volume, and earlier works). But in some instances a speaker can choose the subjunctive, outside these selection contexts, to produce a semantic effect. In these cases the subjunctive does make a semantic contribution. A famous case of such optional subjunctive is the subjunctive in relative clauses; I summarize below some well-known examples: (15)

Optional subjunctive: subjunctive adds to the meaning (i) ‘Polarity’ subjunctive, after negation of an other wise indicative selecting verb (see Quer 1998, 2001, 2009)

Evaluative Subjunctive and Nonveridicality

183

(ii) In relative clauses to convey uncertainty of existence (see Farkas 1985; Quer 1998; Giannakidou 1998, 2013a) (iii) In free relatives, adding the dimension of free choice (Quer 1998, 2001; Marques 2010; Giannakidou and Cheng 2006) (iv) With modal adverbs of possibility (Giannakidou 2009) (v) In questions, producing epistemic meaning (vi) With fi rst-person belief, with a change of meaning (vii) With verbs compatible with both moods such as elpizo ‘hope’ (dual ‘selection’) In this chapter, I examine the last three cases, which have not been studied in depth before. Let me provide an initial look at the data. Consider fi rst the subjunctive in questions: (16)

Ti na theli? What subjv want-3sg ‘What might he want?’

(17)

Na tou arese to fagito? subjv he- gen liked-3sg the food ‘Might it be the case that he liked the food?’

These examples were fi rst mentioned by Rouchota (1994), who called them dubitative. As indicated, the na-version behaves as if it contains a possibility epistemic modal might. Similar use of the subjunctive exists in St’át’imcets (Matthewson 2010). Importantly, these questions are ‘weaker’ than without the subjunctive, as expected if the subjunctive is a possibility modal, which is what I argue for here. Consider now the subjunctive with first-person belief: (18)

Pistevo na kerdisi o Janis. Believe-1sg that-subj win-prf-nonpst-3sg the John ‘I hope for John to win.’

(19)

Pistevo oti tha kerdisi o Janis. Believe-1sg that-ind fut win-prf-nonpst-3sg the John ‘I believe that John will win.’

chapter five

184

With the subjunctive, we observe a change in the meaning of the verb: pistevo is not interpreted as a verb of belief, but as ‘hope’, as can be seen in the comparison with the canonical selected indicative. A related phenomenon is the dual mood pattern, for example, with the verb elpizo ‘hope’: (20)

Elpizo na kerdisi/kerdise o Janis. Hope-1 sg subj win-prf-nonpst/nonpst-3sg the John ‘I hope for John to win/to have won.’

(21)

Elpizo oti kerdise o Janis. Hope-1sg that-ind won-3sg the John ‘I hope that John won.’

Equivalents of hope are also flexible in other languages (see, e.g., a discussion of French hope in Portner and Rubinstein 2012). These cases of ‘relaxed’ mood alternation also suggest a meaning for the subjunctive, and are further discussed in sections 5 and 6. Before I proceed to the specifics of the analysis, and given the topic of this volume, I want to offer some brief typological remarks on the general picture. It has been observed that the subjunctive is hard to unify as a notional category. For instance, we read that: Thus the selection of indicative or subjunctive for complement clauses in Portuguese seems to follow from two factors: nonveridicality and epistemic modality. The indicative is selected for veridical contexts, or if the attitude towards the complement proposition is of epistemic nature. The subjunctive is selected other wise. It does not seem to be associated with a specific kind of modality. (Marques 2010, p. 153)

I agree that it is impossible to unify the subjunctive with one modality, and Wiltschko draws a similar conclusion in this volume. The subjunctive cannot be simply identified with a single label ‘epistemic modality’, or ‘bouletic modality’, since it is used with both epistemic and dynamic modals; and the fact that the optional subjunctive has a semantic effect, but the selected one does not, adds to the complexity. The semantic effect sometimes is epistemic, as we saw in the questions (also with relative clauses; Giannakidou 2014), but it can also be bouletic (as with subjunc-

Evaluative Subjunctive and Nonveridicality

185

tive belief). It seems impossible, thus, to say that the subjunctive itself is associated with one modality, because it is not. Hence, if we are in search of the holy grail of one modality for the subjunctive, we are not in luck. But if we look closer, we see that there is indeed one underlying property shared by all subjunctive contexts: they are all nonveridical. As I discuss in the next section, nonveridical domains are sets of worlds partitioned into p and not-p worlds, and the partition could (but does not have to) be the result of ordering (e.g., ordering sources with modals). And though not all subjunctive contexts can be identified with an ordering, they are all nevertheless partitioned into p and not-p spaces. And the subjunctive, when it has meaning, induces an evaluation, which is in all cases the creation of a nonveridical space. I propose that there are three kinds of evaluation. The fi rst is epistemic, Epistemic subjunctive is akin to a possibility modal, as we saw in questions. The second is emotive subjunctive, which appears with emotive verbs in Romance languages, I argue, imposes nonveridical partitioning between p and not-p worlds as a presupposition. Third, we have the evaluation observed with fi rst-person belief, verbs meaning hope, and in this case the subjunctive gives a preference ordering between p and not-p worlds. Finally, I discuss the Greek emotive complementizer pu and argue that it carries a negative expressive index (Potts 2007; Giannakidou and Yoon 2011). The chapter proceeds as follows. In section 2, I introduce the concepts of veridicality and nonveridicality as a partitioning of the modal space. In section 3, I discuss selection to illustrate the role of nonveridicality. In section 4, I examine the epistemic subjunctive; in section 5, I examine the emotive subjunctive and the expressive complementizer pu. In section 6, I discuss dual mood patterns.

2

Veridicality and Nonveridicality

Here, using the framework I developed in earlier work, I present and defi ne the main notions of veridicality and nonveridicality and the way they relate to modality and mood (Giannakidou 1994; see also Giannakidou 1997, 1998, 1999). In this chapter I rely primarily on more recent work (Giannakidou 2013, 2014). Giannakidou and Mari (this volume) discuss the relevance of nonveridicality to the future and epistemic modality, and I use some of those observations in section 4.

186

chapter five

Montague (1969) uses ‘veridicality’ to characterize sentences with perception verbs such as see (see Giannakidou 2013a for a formal connection between truth and existence, especially as it reveals itself in relative clauses and with progressives). Authors have also used other labels, such as factivity and factuality, to refer to what I call objective veridicality (Karttunen 1971; Kiparksy and Kiparsky 1970; Egré 2008), as well as actuality (Bhatt 2006; Hacquard 2010). In my own work, veridicality is a property of expressions that entail the truth of their complement sentence (see also Zwarts 1995; Beaver and Frazee 2011): (22)

Def 1. Objective veridicality (based on Zwarts 1995; Giannakidou 1997, 1998, 1999). Let F be a monadic sentential operator. The following statements hold: (i) F is veridical just in case Fp → p is logically valid; other wise F is nonveridical. (ii) A nonveridical F is antiveridical iff Fp → ¬p.

Functions that have veridicality and nonveridicality are propositional functions (see Bernardi 2001 for type-flexible defi nitions). F is veridical iff Fp entails p—that is, if whenever Fp is true, p is true too. F is nonveridical if Fp does not entail p—that is, if when Fp is true, p may or may not be true. Note that nonveridical operators do not entail the falsity of p; this is a property of antiveridical operators, as indicated in (ii); antiveridical operators are a subset of nonveridical ones. Nonveridicality under this defi nition is objective, because it does not make reference to parameters other than the propositions and the functions. There are no subjective parameters such as what individuals think or believe, but these will be needed for the treatment of modality and mood (see below). For now, consider that if a verb such as know is our function F, know p entails p, and know is therefore veridical; but if want is the F, want p does not entail p, and therefore want is nonveridical. Hence the basic mood opposition can be readily described in terms of objective veridicality. However, we do have to explain why ‘believe’ and ‘dream’ meanings select the indicative. I will go now right to the characterization of modal verbs as nonveridical. Beaver and Frazee (2011) present nonveridicality as a defining property of the category modality. Consider:

Evaluative Subjunctive and Nonveridicality

(23)

Nicholas might/must bring dessert.

(24)

Nicholas might/must have brought dessert.

187

Logically, possibly p does not entail p; and must is also nonveridical, since must p does not entail that p either. Must does not validate the veridicality principle T that holds knowledge and aleithic modality (see Giannakidou 1998, 1999; discussion in Giannakidou and Mari, this volume; and Portner 2009 on the weakness of must). As de Marneffe et al. 2012 put it: “declaratives like Ariadne left convey firm speaker commitment, whereas qualified variants with modal verbs or embedded sentences imbue the sentence with uncertainty” (de Marneffe 2012, p. 102). Similarly, Trnavac and Taboada (2012) use modals as nonveridical markers of uncertainty. Veridicality and nonveridicality are often also discussed in the context of commitment. The speaker is said to be ‘fully committed’ to the truth of an unmodalized sentence, but is not fully committed in the case of a modal sentence. Smirnova (2012, 2013) uses ‘epistemic commitment’ as her criterion for mood choice and evidentiality. In my earlier work, I argued that a fully committed speaker is in a veridical epistemic state, which is a state with only worlds where the proposition is true. Thus when we talk about the truth of a sentence, we talk about it in two ways: objectively, by appealing to what is the case in the actual world, and subjectively, by appealing to the commitment that individuals have to the truth of the sentence. The connection between veridicality and speaker commitment, I suggest, can be captured by making veridicality relative to individual anchors (see also Farkas 1992). Anchors are the individuals asserting the sentence, or to the main clause subject if the sentence is embedded. The veridicality inference is now relativized to the individual anchor, specifically, to the anchor’s ‘model’ of evaluation (Giannakidou 1997, 1998, 1999, 2011). These models are sets of worlds, representing what the individual anchor believes or knows. 2 We can think of models as ‘modal bases’ associated with individuals, or epistemic states (see (25)). In main assertions, the default is set to the epistemic space of the speaker. I proceed now with the defi nitions (based on Giannakidou 2013, 2014): (25)

Def. 2. Epistemic state of an individual anchor i An epistemic state M(i) is a set of worlds associated with an individual i representing worlds compatible with what i knows or believes.

chapter five

188

Given the epistemic state, we can now identify (non)veridicality subjectively—that is, with respect to the individual anchor’s epistemic state. In Giannakidou (2009), I proposed the following defi nition: (26)

Veridicality A propositional operator F is veridical iff from the truth of Fp we can infer that p is true according to some individual i (i.e., in some individual i’s epistemic model) (Giannakidou 2009, p. 1889)

I rephrase this now as subjective veridicality, as follows: (27)

Def. 3. Subjective veridicality A function F that takes a proposition p as its argument is subjectively veridical with respect to an epistemic state M(i) of an individual anchor i iff: (i) Fp entails or presupposes that i knows/believes that p is true. (ii) If i knows/believes that p, then i’s epistemic state M(i) is such that: M(i) ⊆ p.

From Def. 3, it follows that ∀w[w ∈ M(i) → w ∈ { w’| p(w’)}]. Subjectively veridical functions require in their truth conditions homogeneous epistemic states, included in p. This is the state of full commitment. Consider, for example, an unembedded sentence: (28)

a. O Giannis kerdise to agona. The John won-3sg the race ‘John won the race.’ b. [[ John won the race ]]M(speaker) = 1 iff ∀w [w ∈ M(speaker) → w ∈ { w’| John won the race in w’}]

If the speaker asserts ‘o Giannis kerdise ton agona’ John won the race, she must believe or know that John won the race; hence all worlds in M(speaker) are John-won-the-race worlds: M(speaker) ⊆ p. The unmodalized sentence is therefore equivalent to the speaker knows that p. The indicative is therefore the mood that conveys homogeneity of a modal space M(i) and typically characterizes knowledge—and note that with a negated indicative we also have homogeneity (all worlds are not-p).

Evaluative Subjunctive and Nonveridicality

189

When we think of unmarked past or present as a ‘direct evidential’, we have exactly this veridical (positive) epistemic state. The simple past or present draws on ‘direct’ evidence in the sense that it represents the more reliable knowledge. In the absence of an indirect evidential, all worlds in M(speaker) are p worlds. Subjective nonveridicality, on the other hand, indicates that i does not know or believe p. The epistemic state now intersects with p and contains ¬p worlds: (29)

Def. 4. Subjective nonveridicality A function F that takes a proposition p as its argument is subjectively nonveridical with respect to an individual anchor i iff: (i) Fp does not entail that i knows or believes that p is true. (ii) i’s epistemic state M(i) is such that: M(i) – p is not ∅, which means that (iii) ∃ w’ ∈ M(i) : ¬p(w’).

A subjectively nonveridical function imposes nonhomogeneity on the epistemic state, since there is at least one not-p world. Modals are objectively nonveridical, as mentioned earlier, but also subjectively. A speaker asserting MUST/MAY p allows in her epistemic state not-p worlds (see Giannakidou and Mari, this volume, for more discussion). This is the state of reduced, or ‘weakened’, epistemic commitment. Veridicality can be extended to characterize the epistemic states themselves. A veridical epistemic state is a homogeneous epistemic state that fully supports p. A nonveridical epistemic state, in contrast, is partitioned into p and ¬p worlds: (30)

Def. 6. Veridical and nonveridical epistemic states and commitment a. An epistemic state (a set of worlds) M(i) relative to an individual anchor i is veridical with respect to a proposition p iff all worlds in M(i) are p worlds. (full commitment to p). b. If there is at least one world in M(i) that is a ¬p world, then M(i) is nonveridical (weakened commitment to p). c. If all worlds in M(i) are ¬p worlds, then M(i) is antiveridical (counter- commitment to p).

A veridical epistemic state is a nonpartitioned, homogeneous epistemic state, a state of knowledge or belief (full commitment). An individual in

190

chapter five

a veridical epistemic state has no doubt about p. A nonveridical state, on the other hand, is defi ned as one that contains at least one ¬p world; it therefore conveys weaker commitment to the proposition than a veridical state—that is, only partial commitment at best. All epistemic modals convey nonveridical epistemic states, as do states of indirect evidentials (Giannakidou and Mari 2014, this volume). A speaker asserting MUST/ MAY p, allows in her epistemic state not-p worlds. When all the worlds are ¬p, the state is antiveridical, as with negation and counterfactual assertions, which express counter- commitment of the anchor. Antiveridicality characterizes also optative and imperative sentences, since in these cases i has no commitment to p. Counter-commitment and weakened commitment are noncommitment to p, though only weakened commitment operators are partitioned into p and not-p spaces. From the epistemic domain, we can move to generalize veridicality and nonveridicality to all kinds of modal spaces (sets of worlds), including various kinds of modal bases. Veridicality and nonveridicality are now properties of modal spaces: (31)

Def. 7. and nonveridical modal spaces (i) A set of worlds M is veridical with respect to a proposition p iff all worlds in M are p worlds. (Homogeneity) (ii) A set of worlds M is nonveridical with respect to a proposition p iff there is at least one world in M that is a ¬p world. (Nonhomogeneity) (iii) A set of worlds M is antiveridical with respect to a proposition p iff M and p are disjoint.

All modal bases are nonveridical spaces. Condoravdi (2002) imposes a diversity condition on modals to produce exactly the same effect. Bouletic and deontic domains are also nonveridical because they are ordered. Ordering (Kratzer 1991) always creates a partition, therefore necessarily a nonveridical modal space. The ordering is also responsible for the appearance of ‘strength’ in a nonveridical domain; this is discussed in more detail in Giannakidou and Mari (this volume) in the context of universal epistemic modals. I do not repeat that discussion here, but come back to its main observations in section 4.

Evaluative Subjunctive and Nonveridicality

3

191

Mood Choice in Selection Patterns Is Regulated by (Non)veridicality

In this section, I illustrate how veridicality and nonveridicality account for the basic selection pattern, so that we can then discuss the new cases. In the work I mentioned earlier (Giannakidou 1994, 1997, 1998, 1999, 2009, 2011), I advocated the view that mood choice is regulated by nonveridicality: indicative verbs are veridical and subjunctive nonveridical. In more recent work by others, it has been shown that the relevance of (non)veridicality for mood holds in Greek diachronically, in all stages of the language since Homeric Greek (Chatzopoulou 2012). Similarly, Marques (2004) writes that “while in European Portuguese the selection of mood is sensitive to both veridicality and epistemic modality, Brazilian Portuguese is becoming a language where the truth-value of the proposition is the main factor responsible for mood selection.” Sarigul (2015) shows that nonveridicality is also the key factor in selection of complements also in Turkish. In what follows, I offer key observations about the basic verb categories, concentrating on Greek.

3.1

The Indicative as an Indication of a Veridical Epistemic State

As I said in section  2, main assertions without modals express full speaker commitment and are therefore veridical. (32)

a. John won the race. b. [[ John won the race ]]M(speaker) = 1 iff ∀w[w ∈ M(speaker) → w ∈ { w’| John won the race in w’}]

If the speaker asserts that John won the race, she is committed (i.e., she knows or believes) that John won the race; hence all worlds in M(speaker) are John-won-the-race worlds: M(speaker) ⊆ p. We thus conclude that the ‘unmarked’ indicative is the mood chosen by a speaker when she is the veridical state. In the case of ‘know’, both speaker and knower are committed to the truth of the embedded proposition: (33)

[[Nicholas knows that Ariadne left ]] = 1 iff ∀w [w ∈M(Nicholas) →w ∈ {w’. Ariadne left in w’}] and ∀w [w ∈M(speaker) →w ∈ {w’. Ariadne left in w’}]

chapter five

192

With propositional attitudes there are two potential anchors: the speaker (as in the unembedded case) or the main clause subject. Knowledge verbs are veridical with respect to both anchors, and in Giannakidou (1998, 1999) I called them ‘strongly’ veridical. But how about believe and dream? How are these veridical? Following classic treatments of belief, for the evaluation of p in i believes that p, it must be the case that i, the main clause subject, is committed to p. We have, as with knowledge, two potential anchors: the speaker and the main clause subject. Unlike with knowledge, the speaker’s and the subject’s belief spaces need not coincide: the speaker need not believe that p is true, but the truth condition for belief requires that the believer not have not-p worlds in her belief space. In the example below, Nicholas’s belief space (i.e., the set of worlds compatible with what Nicholas believes) is a subset of p: M(Nicholas) ⊆ p. The speaker may believe or even know that what Nicholas believes is false, but this is irrelevant to Nicholas’s beliefs and mood choice: (34)

O Nicholas pistevi oti efije i Ariadne. the Nicholas believe.3sg that-ind left- 3sg the Ariadne ‘Nicholas believes that Ariadne left.’

(35)

a. [[Nicholas believes that Ariadne left ]] M(Nicholas) = 1 iff ∀w [w ∈M(Nicholas) →w ∈ {w’. Ariadne left in w’}] b. Veridicality of the belief verb If believe(i,p) is true, then M(i) ⊆ p

Since all worlds in M(Nicholas) are p worlds, the belief verb is veridical with respect to that model. The complement of belief is not a fact known, but a fact believed or imagined by the agent of the belief. Now consider dreams: (36)

O Nicholas onireftike oti efije i Ariadne. the Nicholas dreamt-3sg that-ind left-3sg the Ariadne ‘Nicholas dreamt that Ariadne left.’

(37)

[[Nicholas dreamt that Ariadne left ]] M(Nicholas) = 1 iff ∀w [w ∈M(Nicholas) →w ∈{w’. Ariadne left in w’}]

Evaluative Subjunctive and Nonveridicality

193

When I dream or imagine something, as a dreamer, I am fully committed to the fictional reality of my dream. As Farkas (1985) noted: fictional reality replaces the actual one, and we can understand this as a kind of context shift: dream shifts the model of evaluation from the model of the speaker to the M(i), where i is the dreamer; and from knowledge or belief worlds to worlds of dreams. Dreams are private spaces, and mood markers, like evidentials in various languages, can be indicators of shifts to such spaces (see also Giorgi and Pianesi 1997). All worlds in that space are p worlds, since reality no longer plays a role.3 It is interesting to note that indirect evidential marking also disappears in dreams and storytelling (e.g., in Turkish; Ozge Sarigul, pers. comm.). This suggests that the grammar treats fictional contexts as shifted, nonpartitioned states where veridicality holds as if in the real world. Hence dream and fiction verbs are subjectively, but not objectively, veridical. The indicative thus is an indicator of objective and subjective veridicality. We can summarize this as follows: (38)

Indicative as an indicator of objective and subjective veridicality The indicative is the indicator of veridical epistemic states and is selected by expressions that are at least subjectively veridical.

We can view this generalization as a licensing condition on the indicative: the indicative will be licensed only in the scope of an expression that is at least subjectively nonveridical. This formulation renders mood selection akin to polarity licensing (in line with Giannakidou 1998, 2009; Quer 2009), and it is no accident that both polarity items and mood choice are sensitive to the property of (non)veridicality. We can further say that the property of veridicality is lexically represented, for selection at least, as a grammatical feature on the verbs selecting the mood, and the relation between the higher veridical verb and the indicative is an agreement relation of veridical features (see also Ambar, this volume; Baunaz 2014 offers a different approach, still based on feature matching between the verb and complementizer features). I offer a parallel account for the subjunctive next.

3.2

The Subjunctive as an Indicator of Nonveridicality

The subjunctive follows verbs that come with partitioned, nonveridical spaces. These contain p and not-p worlds. The partition can be created by

194

chapter five

an ordering. Take want verbs as a representative case. The truth condition for thelo ‘want’ only requires that the intersection between M(subject) and p be nonempty. In addition, the volitional verb imposes an ordering indicated below as >more desirable such that the worlds in which p is true are more desirable than the worlds in which p is not true. (39)

[[Nicholas wants that Ariadne leave]] M(Nicholas) = 1 iff i. {w| w ∈ M (Nicholas)} ⋂{w’| Ariadne leave in w’} is not ⌀, and ii. {w| w ∈M (Nicholas) ⋂p} >more desirable { w’| w’ ∈M (Nicholas) -p}

If i wants p, not all worlds in M(i) are p worlds. In fact, the ‘wanter’ considers both possibilities, p and not-p worlds, and prefers the former. This places nonveridicality at the heart of the truth condition for desire. If i wants p, in the doxastic model of i there are p and not-p worlds, as indicated above. It is easy to see how this picture generalizes to other directive verbs such as ‘ask’, ‘suggest’, and ‘order’: (40)

[[Nicholas asked Ariadne to leave]] M(Nicholas) = 1 iff i. {w| w ∈ M (Nicholas)} ⋂{w’| Ariadne leave in w’} is not ⌀, and ii. {w| w ∈M (Nicholas) ⋂p} >more desirable { w| w 0M (Nicholas) -p}

(41)

[[Nicholas suggested that Ariadne leave]] M(Nicholas) = 1 iff i. {w| w ∈ M (Nicholas)} ⋂{w’| Ariadne leave in w’} is not ⌀, and ii. {w| w ∈M (Nicholas) ⋂p} >more desirable { w’| w’ ∈M (Nicholas) -p}

This analysis is faithful in spirit to Heim’s treatment of desire reports (Heim 1992). Heim poses that a wants p is true if John prefers p to not-p, as can be seen in (42) (Heim 1992, p. 193): (42)

“α wants that ф’ is true in w0 iff for every w ∈ Dox (α) (w0): every ф-world maximally similar to w is more desirable to α in w0 than any non- ф world maximally similar to w.”

Dox (α) (w) is the accessibility function giving doxastic alternatives for α, M(i) in our case. The opposition between p and not-p is crucial in creating nonveridicality in Dox (α) (w). The semantics of modal verbs follow the same pattern; see Giannakidou (1998, Chap. 3) and Giannakidou and Mari (2014 and this volume) for epistemic modals: the modal base (MB) is part of M(i). With neces-

Evaluative Subjunctive and Nonveridicality

195

sity modals it is not the case that M(i)⊂ p, and it is not the case that MB ⊂ p, as only the Best worlds intersect with p (Portner 2009). (43)

[[prepi/ MUST]] w,f,g = λq . ∀w’∈Best g(w) (⋂f(w)): q(w’) = 1; where Bestg(w)(X) selects the most ideal worlds, given the ordering given by g(w)

Thus a universal modal, irrespective of the kind of ordering, comes with a partitioned, therefore veridical, modal base, and it also comes with a partitioned M(speaker). It is quite obvious that the same holds for the possibility modal: (44)

[[bori/ can]] w,f,g = λq. ⋂f(w) ⋂ q(w’) is not ⌀.

In (44), (⋂f(w) is a nonveridical space. The possibility modal does not induce ordering and creates relatively weak statements. I argue in section 4 that there is a species of epistemic subjunctive in Greek that functions exactly as a possibility modal. Importantly, the subjunctive with possibility modals, and the ability of the subjunctive itself to serve as such, indicates that ordering or preference is not the crucial factor (pace Villalta 2008), but nonveridicality. Villalta (2008), in her treatment of Spanish subjunctive, replaces the nonveridicality opposition between p and not-p with an preference relation between p and contextually given q. (45)

Semantics of want based on comparison of alternatives (Villalta 2008) [[wantC]]g (p)(a)(w) = 1 iff ∀q: q ≠ p & q ∈ g(C): Simw’ (Dox a(w) ⋂ p) >a,w Simw’(Dox a(w) ⋂ q)

This move, apart from being unmotivated (since, among others q entails not-p if p is not identical to q), loses the insight of p versus not-p, which is a very useful one, not just for mood, but also for the licensing of negative polarity items (NPIs) (see Giannakidou 1998). Nonveridicality is also at the heart of modality (Giannakidou 1998; Giannakidou and Mari, this volume; recall Condoravdi’s (2002) diversity presupposition of modals that requires p and not-p in all modal bases). Hence nonveridicality allows us to unify mood selection, NPI licensing, and modality—and by giving it up as Villalta does, we lose this broader generalization.

chapter five

196

We can summarize the condition for subjunctive as follows: (46)

Subjunctive as an indicator of nonveridicality (i) The subjunctive is an indicator of a nonveridical epistemic state or modal base and is selected by expressions that are at least subjectively nonveridical. (ii) Subjunctive sentences indicate epistemic weakening.

Given what I have said so far, it seems obvious that epistemic weakening is the creation of a nonveridical modal space, as I discuss next (following Giannakidou 2014): (47)

Epistemic weakening Epistemic weakening is the creation of a nonveridical epistemic space.

If the subjunctive contributes a nonveridical space, then the relation between the higher selecting verb and the subjunctive is one of agreement, or matching—as I suggested earlier about the indicative. If the veridicality properties of the verbs are encoded as syntactic features, selection can be understood as veridical (indicative) or nonveridical (subjunctive) agreement. Semantically, the subjunctive in selection is thus vacuous;4 for more ideas about how to capture the syntactic dependency in the nonveridical framework see the chapter by Ambar in this volume.

4

Epistemic Subjunctive: A Possibility Modal

I start with an observation about Portuguese. Marques (2010) notes that in Portuguese, belief and assumption verbs such as suspect and assume may occasionally allow the subjunctive: (48)

a. Acredito que a Maria está doente. Believe-1sg that the Maria is-ind-3sg ill b. Acredito que a Maria esteja doente. believe.1sg that the Maria is.subj.3sg ill ‘I believe that Maria is ill.’

Evaluative Subjunctive and Nonveridicality

197

Note that the use of the fi rst person neutralizes the difference between speaker and believer. Marques says that “the selection of one or another mood is related to the degree of belief being expressed. The indicative signals a high degree of belief, the subjunctive a lower degree. . . . The concept of veridicality accounts for this case of mood variation. With the indicative, the inference follows that the relevant proposition is true (according to the subject of the main clause), contrary to what happens if the subjunctive is selected” (Marques 2010, p. 145). In other words, when the speaker chooses to utter the subjunctive version, she is making a point to distinguish between what she believes and what she knows for sure. When the speaker chooses the indicative version, she is in a veridical state and the belief is justified (there is a higher degree of certainty); but when the speaker chooses the subjunctive, she seems to be aware that her belief might not be justified. The speaker, in the subjunctive version, has some uncertainty in her epistemic state and her commitment to Maria is ill is weaker; that is, she allows for the possibility of not-p. The presence of subjunctive after fi rst-person belief indicates precisely this weakening of veridicality. This is strikingly similar to what happens when we add a modal verb in the embedded clause: (49)

a. I believe that Maria might be sick. b. I believe that Maria is sick.

In contrast to the unmodalized versions, commitment to the truth of embedded sentence is weakened by adding might. The use of the subjunctive, I claim, is exactly like adding might. I defi ne a new species of subjunctive, epistemic subjunctive, which is akin to a possibility modal: (50)

[[Epistemic Subjunctive]] w,f,g = λq . ⋂f(w) ⋂ q(w’) is not ⌀;

In (50), (⋂f(w) is a nonveridical epistemic modal base because not all worlds are p worlds. The possibility modal does not induce ordering and creates relatively weak statements. I speculate that the Portuguese subjunctive in Marques’s examples is of that kind, and I provide evidence below that Greek makes extensive use of epistemic subjunctive. It is a matter of cross-linguistic variation whether a language has this type of subjunctive.

chapter five

198

Before we move on to the specifics of Greek, I want to clarify that in commitment and epistemic weakening, we have a scale of commitment strength that goes as follows, from strongest to weakest (Giannakidou and Mari, this volume): (51)

Commitment strength ( ≫ is ‘more committed’) Non-modalized p (speaker knows p, added to the common ground) ≫ MUST p (speaker does not know p, but is biased towards p) ≫ POSSIBLY p (speaker does not know p, and there is no bias)

When all worlds in M are p worlds, we have veridicality, and this conveys the strongest commitment. With MUST, we have a nonveridical space with bias toward the p worlds. In Chapter 3 in this volume, Mari and I characterize stronger modals such as MUST and FUT as biased modals; they are stronger in terms of commitment than possibility modals. With biased modals there is a non-singleton support set p, but the modal base and M(speaker) still allow not-p worlds (nonveridicality). The possibility sentence, in contrast, conveys equilibrium between p and not-p (Giannakidou 2013); that is, there is no preference for the p or not-p worlds. In this case we have the weakest commitment, and the epistemic subjunctive is a modal of that kind. There is substantial evidence in Greek that the epistemic subjunctive is a possibility modal. Often the subjunctive appears with possibility adverbs (Giannakidou 2009): (52)

Isos/pithanon (na) efije o Nicholas. Maybe/possibly subjv left-3sg the Nicholas ‘Maybe Nicholas left.’

Na, crucially, is incompatible with modals of probability and necessity. In this case, the future particle tha is used (Giannakidou 2012; Giannakidou and Mari 2013 and this volume). (53)

* Malon/Sigoura na kimate o Nicholas. Probably/certainly subjv sleep-3sg the Nicholas

Evaluative Subjunctive and Nonveridicality

(54)

199

Malon/Sigoura tha kimate o Nicholas. Probably/certainly fut sleep-3sg the Nicholas ‘Probably/Certainly Nicholas is asleep.’

The FUT sentence is equivalent to MUST. Given the availability of tha, and the ill-formed nature of na with stronger adverbs, the truth conditions of the subjunctive sentence must be delegated to possibility. Both types of sentences (with na and tha) are epistemically weaker than the unmodalized positive assertion. But the future particle is a universal modal, whereas the subjunctive is an existential. Greek thus has two strategies for commitment weakening with the particles—a stronger one with the universal modals and a weaker one with the subjunctive. When co-occurring, na and the possibility adverbs isos and pithanon enter modal concord; thus the reading of the sentence contains one possibility modal. Similar examples can be reproduced with the present tense: (55)

Isos/pithanon (na) kimate o Nicholas. Maybe/possibly subjv sleep-3sg the Nicholas ‘Maybe/Possibly Nicholas is asleep.’

(56)

◊ (Nicholas is asleep at the present time)

We can now view our initial data on questions in this light: when a subjunctive is added to a question, it adds might. (57)

Pjos irthe sto party? Who came-3sg to the party ‘Who came to the party?’

(58)

Ti na ipe? What subjv said.3sg ‘What might he have said?’

(59)

Pjos na irthe sto party? Who subjv came-3sg to the party ‘Who might have come to the party?’

chapter five

200

(60)

Na tou milise (arage)? subjv him talked-3sg Q-particle

Polar questions

‘Might she have talked to him?’ (61)

Tou milise? him talked-3sg ‘Did she talk to him?’

Here the speaker asks the hearer about the possibility of p rather than p itself. The subjunctive has a similar use in Salish, as reported by Matthewson (2010), where it is said that such questions are ‘conjectural’. Mathewson argues that the subjunctive has an overall weakening effect, very much in the spirit outlined here. The na/might question, as can be seen, is about questioning the possibility of p rather than p itself. The German so- called modal particles are reported to have a similar use (the example is from Zimmermann 2011 with his translation): (62)

Hat Hans wohl Maria eingeladen? has Hans prt Mary invited ‘What do you reckon: Has Hans invited Mary?’

Zimmerman says: “The question above is not about whether or not Hans has invited Mary, but by using wohl the speaker indicates her awareness that the addressee may not be fully committed to her answer” (Zimmermann 2011, p. 2020). In agreement with what I just said about Greek modal particles create a question that does not require a factual answer (and see Matthewson 2010 for Salish). Given the meaning of epistemic subjunctive as might that I am suggesting, questions with the subjunctive are equivalent, literally, to the translated questions containing might. (63)

na = might

(64)

Who might have come to the party?

Without na, we have a regular information question, with the set of answers below:

Evaluative Subjunctive and Nonveridicality

(65)

201

[[Who came to the party?]] = {Bill came to the party, Marina came to the party, Ariadne came to the party, Nicholas came to the party, . . . }

With the might question, the answer set is the following: (66)

[[Who might have come to the party?]] = {◊ Bill came to the party, ◊ Marina came to the party, ◊ Ariadne came to the party, ◊ Nicholas came to the party, . . . }

The answer set contains modalized propositions about who possibly came to the party, and it does not tell us much about who actually came to the party. In further support of this analysis, consider the following pair: (67)

Poso xrono na ine o Agios Vasilis? How old might Santa Claus be?

It would be rather bizarre, for an adult, to ask the question without the subjunctive, as in (68): (68)

# Poso xrono ine o Agios Vasilis? How old is Santa?

The oddity comes from the fact that the nonsubjunctive question presumes that the answerer will give an answer—and unless you are a child and you believe in Santa Claus, an answer to this question is not possible. In sum, I defi ned in this section a new species of subjunctive that I called epistemic. This subjunctive is equivalent to might, and its evaluative function is to weaken the veridicality of the sentences. Languages may differ with respect to whether they allow their subjunctives to function in this way. The Greek na and the Portuguese subjunctive systematically use epistemic subjunctives. I proceed now to examine the emotive subjunctive.

5

Emotive Subjunctive: Nonveridicality in the Pragmatic Dimension

The species I identify as ‘emotive subjunctive’ occurs after emotive factive verbs in some Romance languages, and it is also manifested in fi rst-person

chapter five

202

belief alternations and dual mood patterns observed with the verb meaning hope. I argue that in this case the subjunctive again functions modally, this time not as a modal in the assertion, but at the level of presupposition: consistent with its use as a possibility epistemic modal, the emotive subjunctive introduces the presupposition that the individual anchor considered not-p possible, at a time prior to the assertion. Thus in this case too, the subjunctive is an indicator of a nonveridical epistemic state, but prior to the assertion. Finally, we contrast the emotive subjunctive with the emotive Greek complementizer pu, which, I argue, contributes negativity in the expressive dimension.

5.1

The Subjunctive with Emotive Verbs

As I noted earlier, factive verbs (know) are veridical; emotive factives should therefore not select the subjunctive. This is indeed the case with epistemic factives know, as I showed. But regarding emotive verbs, there are three patterns: (i) languages that require subjunctive (Spanish, Italian, maybe French); (ii) languages that allow both subjunctive and indicative ((Brazilian) Portuguese, Catalan, Turkish); (iii) languages in which emotives select indicative (Greek, Hungarian, Romanian, Bulgarian); the emotive complement may be distinguished in some other way.

Given this variation, it becomes clear that, cross-lingustically, the emotive class is not a typical selection context. In nonveridical contexts, emotives should simply not allow the subjunctive since they are factive and therefore veridical; hence the languages in (iii) are well-behaved. But we still need to explain the option of subjunctive in types (i and ii), and why there is a special marking with the emotive verb in Greek. We will see that, although we may deny the factive nature of the emotives, we cannot deny their veridical nature (thanks to Johan Rooryck and Paul Egré for discussing this question with me). Huddleston and Pullum (2002) call emotives not entailing and give examples like the following: (69)

Falsely believing that he had infl icted a fatal wound, Oedipus regretted killing the stranger on the road to Thebes. (Klein 1975, quoted in Gazdar 1979, p. 122)

Evaluative Subjunctive and Nonveridicality

203

Here, it is not entailed (i.e., it is not true in the actual world) that Oedipus infl icted a fatal wound. Egré (2008) offers similar examples: (70)

John wrongly believes that Mary got married, and he regrets that she is no longer unmarried. (Egré 2008, (30), citing earlier work by Egré and Schlenker)

These examples show that one can have an emotive attitude toward something that one believes to be a fact but may not actually be a fact. In the normal case, we are happy or sad about something that we know happened; but one may also believe that something happened (a believed fact) and then feel happy or sad about it. Hence emotive verbs need not be veridical in the objective sense (as know is); but they are subjectively veridical, since emotive verbs rely on the emotive subject’s belief of p. In this sense they are like belief and fiction verbs. But these verbs, as I showed in section 3, select indicative. Why, then, are emotives compatible with the subjunctive? Baker (1970) suggested that emotives express a negativity, a “contrariness” between a perceived fact and some mental or emotional state. According to Baker, we say that we are surprised when a certain fact does not conform to our expectations; relieved when it does not conform to our fears; disappointed when it is not in line with our hopes. Likewise, we say that a certain fact is odd or strange if it seems counter to our view of what is logical. Emotives, as a class, convey this “contrary” component, via which they can also trigger NPIs, something that veridical verbs normally do not do: (71)

a. b. c. d.

*Ariadne believes/dreams that she talked to anybody. *Ariadne knows that she talked to anybody. Ariadne regrets that she talked to anybody. Ariadne is amazed that we got any tickets at all!

Very much in agreement with Baker (and later Linebarger 1980), I have argued that the appearance of NPIs with emotive verbs is due to accessing, in the pragmatics of the emotive verb, a negative inference (Giannakidou 1997, 2006). Here I build on this idea. I suggested that the component of emotives responsible for voiding veridicality is a counterfactual conditional: (72)

John regrets that I bought a car. → John would prefer it if I had not bought a car.

204

chapter five

The nonveridical proposition with regret is a counterfactual conditional with a negative protasis, and is noncancelable: (73)

John regrets that I bought a car; #in fact he wouldn’t want me to buy a car.

Negating John would want me to buy a car creates an oddity, suggesting that this inference is “not merely a conversational implicature, as argued in Linebarger, but rather something stronger, perhaps a presupposition or a conventional implicature in the sense of Potts (2005). In fact, since emotive factives convey an expressive attitude toward the propositional content of their complement, it makes sense to argue that they all encode conventionally this attitude” [emphasis added] (Giannakidou 2006, p. 595). Here I argue that the negative component is a presupposition: the main clause subject has a belief or expectation that not-p was true prior to the assertion. It is because of this presupposition that we get the perceived contrariness, and it is this proposition that the NPI accesses in order to be triggered: (74)

Negative presupposition of emotive verbs a. [[ i Vemotive p ]] is defi ned only if i believed or expected that not-p, at a time t’< tu (the utterance time). b. At t’< tu: ∀w [w ∈M(i) (t’’) →w ∈λw’. ¬p (w’)] c. If defi ned, [[ i Vemotive p ]] M(i) = 1 iff ∀w [w ∈M(i) (tu) →w ∈λw’. p (w’)]

In other words, Nicholas is surprised that Ariadne talked to him can only be felicitous in a context where, prior to the utterance, Nicholas believed that Ariadne would not talk to him. This is what it means to be surprised. Likewise, if Ariadne is amazed that we got any tickets at all, Ariadne must have believed that we would not get any tickets at all; hence the NPI. If Ariadne regrets that she talked to anybody, then prior to the assertion she preferred not to talk to anybody. The availability of this negative presupposition is crucial to the lexical meaning of the emotive verb and is responsible for rescuing the NPI and for licensing the subjunctive. The subjunctive after the emotive verb is thus also an NPI, licensed by the emotive verb:

Evaluative Subjunctive and Nonveridicality

(75)

205

[[SUBJemotive (p)]] is defi ned iff: there was a time t’ < tu (the utterance time) such that the main clause subject i believed or expected that not-p at t’.

SUBJemotive does not affect the truth conditions of the complement, but contributes a nonveridical presupposition. The subjunctive appears to be sensitive to the presupposition, just as NPIs can be sensitive to assertion or presupposition. From this perspective, the subjunctive after emotives strengthens the connection between NPIs and mood morphemes. As with epistemic subjunctive, languages will parametrize as to whether they possess the emotive subjunctive or not. Spanish and Italian have it, but Greek and the languages of type (b) do not. Languages of type (b) that optionally allow the subjunctive also have it. This analysis, and especially the reference to a previous epistemic state, echoes earlier observations in my treatment of the implicative manage. In order to explain why manage p selects the subjunctive, despite its veridical inference (i managed p entails p), I suggested that belief states must be relativized to times (Giannakidou 2011). Mari (2014) argues that the possibility of not-p must be part of what she calls an “extended modal base” of ability modals. This extended modal base is required to be nonveridical (like all modals); therefore by presupposition it must contain not-p worlds. This helps Mari explain actuality entailments with ability modals (Last night, John was able to drink ten beers) while preserving the nonveridical analysis of ability (CAN (John drink ten beers)) does not entail that John drinks ten beers is true). These are very useful observations that relate to the discussion of emotives. However, the presupposition of emotivity is stronger than i simply believing that it was possible that not-p. If the attitude holder considered it merely possible that not-p, he would be in equilibrium with not-p, which would not justify the contrariness observed. Note also that, importantly, Greek NPIs are not licensed by the negative presupposition; therefore the fact that the Greek subjunctive is not licensed in this context is consistent with its analysis as an NPI, as I suggest. The Greek subjunctive is selected (in fact, strictly selected) by Greek manage, while NPIs are blocked (Giannakidou 1998). The contrast clearly suggests that the analysis of negativity in implicatives and emotives cannot be the same.

chapter five

206

5.2 Greek Emotive Complementizer pu In Greek, recall that we do not have the emotive subjunctive, but a special complementizer, pu: (76)

O the i the

Nicholas lipate/xerete pu/*na/*oti efije Nicholas is-sad-3sg /is happy-3sg that-emotive left-3sg Ariadne. Ariadne

‘Nicholas regrets/is happy that Ariadne left.’ (77)

O Nicholas kseri oti/*pu efije i Ariadne. ‘Nicholas knows that Ariadne left.’

I argue that pu carries expressive content, in line with other expressive complementizers that we know Greek possesses, such as the metalinguisitic comparative complementizer para (Giannakidou and Stavrou 2009; Giannakidou and Yoon 2011). According to Potts (2007), an expressive indicates that the speaker is in a heightened emotional state, and offers a suitable framework to understand the class of emotive verbs altogether. To formalize the claim, Potts uses expressive indices: (78)

An expressive index is a triple , where a,b ∈ De and I ∈ [−1, 1].

Expressive indices are the foundation for expressive domains and are contained in expressives such as damn and bastard. These indices encode the degree of expressivity and the orientation of the expressive, and they are defi ned via numerical intervals I ⊆ [−1, 1]. We can read as conveying that individual a is at expressive level I for an individual b. Mapping emotional attitude onto expressive intervals has the advantage of allowing flexibility from very neutral (if I = [−1, 1])—in Potts’s words, “a has no feelings for b”—to very negative ones. Emotive relations emerge as we narrow down I to proper subintervals of [−1, 1]; the more positive the numbers, the more positive the expressive relationship, and conversely.

Evaluative Subjunctive and Nonveridicality

207

Giannakidou and Yoon (2011) suggest that individuals can have emotion about propositions. The motivating data were expressive metalinguistic comparatives such as I’d rather die than marry him! In Korean and Greek, these involve special complementizers, nuni and para; English simply has than. We can now say that the class of Vemotive contains expressive indices: (79)

Emotive verbs contain expressive indices An emotive verb contains an expressive index , where a is the individual anchor, q the proposition it embeds; and I ranges between [−1, 1].

The expressive index is a contribution of Vemotive at the not-at-issue level. These indices can have morphosyntactic realization, and may actually trigger agreement. It is not uncommon for expressives to do that; for example, Potts and Kawahara (2004) claim this for honorific agreement. I claim that the relation between Vemotive and pu is expressive agreement; the emotive verb carries a morphosyntactive feature +expressive and selects a C that agrees with this feature. If this sounds like a reasonable analysis, then one can claim a parallel analysis for the emotive with the subjunctive: the emotive has a morphosyntactic expressive feature like Greek and selects the subjunctive as an agreeing form. Greek opts for the C position because this is a productive strategy in the language. Romance languages typically do not exploit the C position (though perhaps they do covertly; see Baunaz 2015); expressive agreement targets the next available head: Mood. Languages vary as to whether they have the lexical items ‘emotive subjunctive’ or ‘emotive C’. Languages in the Romance family (group (i)) have emotive subjunctive; languages in group (iii) have emotive C. Languages in the middle group, (ii) are in transition—either developing or discontinuing the emotive subjunctive (Portuguese, Turkish). These ideas are quite new, and certainly more detailed study is needed. I want to offer here a framework useful for addressing the crosslinguistic variation observed with the emotives. We proceed next to dual mood patterns, which reveal a third function of evaluative subjunctive.

chapter five

208

6

The Subjunctive as Preference Ordering

Recall the initial examples from Greek: (80)

Pistevo na kerdisi o Janis. Believe-1sg that-subjv win-prf-nonpst-3sg the John ‘I hope John to win.’

As indicated, the verb pistevo is not interpreted as a verb of belief, but it seems to be akin to ‘hope’. Note also the importance of fi rst person: (81)

* I Maria pistevi na kerdisi The Maria believe-3sg that-subjv win.prf-nonpst-3sg o Janis. the John

In the third person, na is impossible. The sensitivity to fi rst person suggests that the use of na is tied to the speaker; it is therefore distinct from the emotive subjunctive, which concerns the main clause subject. This subjunctive cannot be of the epistemic kind either, since epistemic subjunctive does not affect the meaning of the attitude; there is a contrast between (80) with the subjunctive and the sentence in (82) with indicative and a possibility modal: (82)

Pistevo oti bori na kerdisi Believe-1sg that-ind is-possible-3sg. win.prf-nonpst-3sg o Janis. the John ‘I believe that it is possible for John to win.’

Here, with the indicative and an embedded epistemic modal, the speaker asserts his belief that it is possible for John to win. This is a very different meaning from (80), where the verb meaning appears to be affected and we no longer have a belief. Given the similarity to ‘hope’, it seems reasonable to assume that in this case the subjunctive introduces a preference ordering that (a) creates a nonveridical partitioning in the speaker’s epistemic state by introducing worlds in which John does not win, and (b) says that the worlds in which John wins are preferred over the worlds in which John does not win.

Evaluative Subjunctive and Nonveridicality

(83)

[[Pistevo na kerdisio Janis]] M(speaker) = 1 iff i. ¬ ∀w’ [w’ ∈ M (speaker) → John wins at w’] and ii. {w| w ∈ (M (speaker) ⋂p) } >more desirable { w’| w’ ∈ (M (speaker).-p)}

(84)

[[SUBJpreference]] = λp. p is more desirable to the speaker than ¬ p.

209

The belief verb is interpreted as akin to ‘hope’. The SUBJpreference has a dif ferent meaning from the emotive subjunctive I defi ned earlier— and which Greek lacks. The subjunctive of preference gives the preference in the assertion, thereby necessitating not- p worlds at the time of the assertion. This changes the meaning of the verb. With the emotive subjunctive, the not- p worlds were entertained at the time prior to the assertion—but at the assertion, the complement of the emotive verb is taken to be a fact (actual or perceived). I suggest that the subjunctive of preference is also at work with the verb meaning hope itself: (85)

Elpizo na kerdisi o Janis. Hope-1sg that-subjv win-prf-nonpst-3sg the John ‘I hope for John to win.’

(86)

Elpizo oti tha kerdisi o Janis. Hope-1sg that-ind fut win. perf.nonpast.3sg the John ‘I hope that John will win.’

Hope counter parts are flexible in European languages (e.g., French; see Portner and Rubinstein 2012), and we observe here that English allows infi nitival and that complements with hope. The dual mood patterns correlate again with a change in the verb meaning. I argue that the subjunctive we fi nd with hope is the subjunctive of preference. The meaning is parallel to what I indicated earlier with fi rst-person belief: (87)

[[Elpizo na kerdisi o Janis]] M(spekaer) = 1 iff i. ¬ ∀w’ [w’ ∈ M (speaker) → John wins at w’] ii. {w| w ∈ (M (speaker) ⋂p)} >more desirable { w’| w’ ∈ (M (speaker) – p)}

chapter five

210

(88)

[[Elpizo oti tha kerdisi o Janis]] M(speaker) = 1 iff ∀w’ [w’ ∈ M (speaker) → ∃t [tu < t & John wins at t at w}’]

The indicative complement with the future is a stronger statement—and though the truth conditions designated may be a bit too strong, the indicative expresses certainty about the existence of winning times. In closing, I want to mention some additional data that can be understood to follow from linking the mood choice to a change in the meaning of the verb. Quer observes that, with a choice between indicative and subjunctive, a mixed assertive- emotive verb loses its assertive meaning with the subjunctive: “When indicative is an option, the predicate yields an assertive reading which is absent with a subjunctive argument clause” (Quer 2001, pp. 106–107): (89)

Es queixava que li posessin Refl complain-imprf-3sg that her/him put-subjv-imprf-3pl males notes. bad marks ‘S/he complained that (subjunctive) they gave her/him bad grades.’

(90)

Es queixava que li posaven Refl complain-imprf-3sg that her/him put-ind-imprf-3pl males notes. bad marks ‘S/he complained that (indicative) they gave her/him bad grades.’

The effect is lost in English, but is visible in Greek with the complementizer. Now we have alternation between oti and pu: (91)

a. O Janis paraponethike oti ton ksexasa. The John complained-3sg that-ind him forgot-1sg ‘John complained that I forgot him.’ b. O Janis paraponethike pu ton ksexasa. The John complained-3sg because-emotive him forgot-1sg ‘John complained because I forgot him.’

Evaluative Subjunctive and Nonveridicality

211

While the oti-version asserts that I forgot him, and this proposition can be negated, the pu-version cannot be negated: (92)

a. O Janis paraponethike oti ton ksexasa; ala kani lathos: dhen ton ksexasa. ‘John complained that I forgot him; but he is wrong, because I didn’t. b. O Janis paraponethike pu ton ksexasa; # ala kani lathos: dhen ton ksexasa. ‘John complained that I forgot him; but he is wrong, because I didn’t.

Examples (91) and (92) support my analysis of pu as nonassertive, since expressive content cannot be negated. The pu-version is emotive: in choosing it, I, the speaker, bring in my perspective and feel bad about forgetting John. This speaker orientation is characteristic of expressives as a class, and the fact that we fi nd it with pu is encouraging for the approach I suggest here. We fi nd this in the pairing with remember, for example (see Christidis 1981): (93)

a. Thimithika oti ton sinandisa sto Parisi. I remembered that I met him in Paris. b. Thimithika pu ton sinandisa sto Parisi. I remembered (as-if-I-were-there) that I met him in Paris.

The pu-version brings about an emotive reading in thimithika ‘remember’. Overall, I think that opening the discussion of the interaction between verbal meaning and mood choice provides a useful perspective from which to address the other wise mysterious occurrence of complementizer and mood switches that have been known in the literature for quite a while. Portner and Rubinstein (2012) have also written about flexible patterns (in Romance languages), as I mentioned earlier; and though I was not able to address their work in detail here, there is considerable common ground between their ideas and the analysis of the subjunctive I defend in this chapter, in particular their notion of contextual commitment.

chapter five

212

7

Conclusion: Subjunctive, Evaluation, Categories

In this chapter, I make the claim that the rich landscape of subjunctives cross-linguistically becomes quite manageable if we acknowledge two factors. The fi rst is sensitivity of the subjunctive to the property of nonveridicality (objective with reference to the actual world, or subjective with reference to an individual’s knowledge or beliefs). The second factor is the evaluative function of the subjunctive. Evaluative subjunctive is manifests in three functions identified here: epistemic, emotive, and preference. Evaluative subjunctive, in all cases, creates a nonveridical space—that is, a modal space partitioned into p and not-p worlds. Evaluative subjunctive, therefore, is itself a nonveridical expression, unlike strictly selected subjunctive, which depends on the existence of a nonveridical licenser higher up and does not contribute much in truth conditions. The connection between nonveridicality and evaluation is insightfully discussed also in the recent work of Trnavac and Taboada (2012, 2013); I have also argued that the epistemic subjunctive is an epistemic possibility modal in relative clauses (see Giannakidou 2013a). The preference subjunctive is an ordering, otherwise typically given by preference/bouletic attitude verbs. That a particle takes up the functions of modals and verbs is not an unexpected fi nding—especially for Greek, a language in which the future particle also functions as an epistemic modal (Giannakidou 2012; Giannakidou and Mari, this volume). We must conclude therefore that the question of the morphosyntactic category of an item is distinct from the question of its semantics, and this is not a novel conclusion (see Roussou and Tsangalidis 2010). Particles, like the Greek subjunctive and the future, and modal particles in Dutch and German, can perfectly well perform modal or evidential functions that are other wise attributed to modal verbs or adverbs. In the larger scheme of things, the two important lessons to extract from the work presented here are (a) that the categories ‘particle’, ‘modal verb’, and ‘attitude verb’ are just labels and that linguistic items form notional categories cross-linguistically based on their meaning, and (b) that nonveridicality seems to be a decisive component in the meaning of subjunctive, in both strict selection and choice.

Evaluative Subjunctive and Nonveridicality

213

Notes The impetus for the work that became this chapter was a presentation at the TRAIT workshop on “categories” at the University of Wroclaw in 2013. I then presented this material at a seminar at the Institut Jean Nicod in Paris, and at colloquia at Leiden University and the University of Groningen. I am very grateful to the audiences for valuable comments, especially to Manuela Ambar, Oliver Bott, Philippe de Brabanter, Jack Hoeksema, Francois Recanati, and Malte Zimmermann. Many thanks also to Paul Egré, Mikhail Kissine, Rui Marques, Salvador Mascarenhas, Johan Rooryck, and Anastasia Smirnova for discussion, suggestions, and comments. A special thanks goes to Alda Mari for our many conversations about nonveridicality, epistemic modality, and virtually everything else discussed here. Finally, my appreciation goes to the anonymous reviewers of this volume for their comments, which helped me in both rethinking and reorganizing the discussion. 1. In Italian belief verbs can take indicative or subjunctive (Giorgi and Pianesi 1997; Portner 1992; Mari 2015). In Portuguese the subjunctive can optionally be triggered too, and the choice correlates with the ‘degree of commitment’ (Marques 2010). 2. Individual anchoring of truth should be seen on a par with other kinds of anchoring of propositional content—such as temporal anchoring or event anchoring (see, e.g., Hacquard 2010). The individual anchor is a subjective parameter of evaluation similar to Lasersohn’s (2005) judge. The difference between knowledge and belief is not so impor tant for our purposes here, as verbs of knowledge and belief both select the indicative. Belief makes a difference for an agent typically when it is contrasted with knowledge—i.e., when the agent is aware that she does not have enough information to support a proposition. In this case, we can say that we have semantic narrowing (Geurts and van Tiel 2013); see the later discussion on epistemic subjunctive. 3. A dreamer can have moments of awareness that ‘shift’ her back to reality—as is the case, for instance, of being aware that one is dreaming (thanks to Oliver Bott for raising this possibility). We must say, however, that these moments are probably excluded by some generalized domain restriction mechanism or narrowing. 4. Note that in Villalta’s account too the selected subjunctive has no semantics; it is merely an identity function, as indicated in Villalta’s (111): [[SUBJ IP]] = [[IP]]. Therefore, despite an initial claim that a semantics for the subjunctive will be given, Villalta in fact does not give a semantics for the subjunctive.

References Ambar, M. 1999. Infi nitives vs. participles. In Semantic Issues in Romance Syntax, Current Issues in Linguistic Theory, ed. Esthela Trevino and José Lema, 7–32. Amsterdam: John Benjamins. Baker, C. L. 1970. Double negatives. Linguistic Inquiry 1: 169–186.

214

chapter five

Baunaz, L. 2015. On the various sizes of complementizers. Probus 27(2): 193–236. Beaver,  D., and  J. Frazee. Forthcoming. In The Handbook of Computational Linguistics, edited by Ruslan Mitkov. Bernardi, R. 2002. Reasoning with polarity in categorial type logic. PhD thesis, University of Utrecht. Bhatt, R. 2006. Covert Modality in Non- finite Contexts. Berlin: Mouton de Gruyter. Bulatovic, G. 2008. Modality, futurity, and temporal dependency: The semantics of the Serbian perfect nonpast and future 2. PhD diss., University of Chicago. Chatzopolous,  K. 2012. Negation and nonveridicality in the history of Greek. PhD thesis, University of Chicago. Christidis,  A.  F. 1981. Oti/pos – pu: complementizer choice in Modern Greek. Studies in Greek Linguistics 2: 113–177. Den Dikken, M., and A. Giannakidou. 2002. From Hell to polarity: ‘Aggressively non-D-linked’ wh-phrases as polarity items. Linguistic Inquiry 33: 31–61. Egré,  P.  2008. Question embedding and factivity. Gratzer Philosophische Studien 77: 85–125. Farkas,  D.  F. 1985. Intensional Descriptions and the Romance Subjunctive Mood. New York: Garland. ———. 1992. Mood choice in complements clauses. In Romance Languages and Modern Linguistic Theory, ed. P. Hirschbühler and E. F. K. Koerner, 69–105. Amsterdam: John Benjamins. Gazdar,  P.  1979. Pragmatics: Implicature, Presuppostion, and Logical Form. New York: Academic Press. Geurtz, Bart, and Bob van Tiel. Embedded scalars. Semantics and Pragmatics 6(9): 1–37. http://dx.doi.org /10.3765/sp.6.9. Giannakidou, A. 1994. The semantic licensing of NPIs and the Modern Greek subjunctive. In Language and Cognition 4: 55–68. ———. 1997. The landscape of polarity items. PhD diss., University of Groningen. ———. 1998. Polarity Sensitivity as (Non)veridical Dependency. Amsterdam: John Benjamins. ———. 1999. Affective dependencies. Linguistics and Philosophy 22: 367–421. ———. 2006. Only, emotive factive verbs, and the dual nature of polarity dependency. Language 82: 575–603. ———. 2009. The dependency of the subjunctive revisited: Temporal semantics and polarity. Lingua 120: 1883–1908. ———. 2011. Non-veridicality and time: The dependency of the subjunctive revisited. In Tense across Languages, ed. R. Musan and M. Rathert, 59–90. Berlin: Mouton de Gruyter. ———. 2012. The Greek future and modal concord. In Selected Papers from 10th International Conference on Greek Linguistics, 46–61. Komotini, Greece: Democritus University of Thrace.

Evaluative Subjunctive and Nonveridicality

215

———. 2013a. (Non)veridicality, evaluation, and event actualization: Evidence from the subjunctive in relative clauses. In Nonveridicality and Evaluation: Theoretical, Computational, and Corpus Approaches, ed. M. Taboada and R. Trnavac, 17–47. Studies in Pragmatics. Leiden, The Netherlands: Brill. ———. 2013b. Inquisitive assertions and nonveridicality. In The Dynamic, Inquisitive, and Visionary life of φ, ?φ and possibly φ: A Festschrift for Jeroen Groenendijk, Martin Stokhof and Frank Veltman, ed. Maria Aloni, Michael Franke, and F. Roelofsen, 115–126. University of Amsterdam. ———. 2014. The futurity of the present and the modality of the future: A commentary on Broekhuis and Verkuyl. Natural Language and Linguistic Theory 32: 1011–1032. Giannakidou, A., and A. Mari. 2013. A two dimensional analysis of the future: Modal adverbs and speaker’s bias. Proceedings of the Amsterdam Colloquium 2013, 115–122. University of Amsterdam. ———. 2014. The future in Greek and Italian: Metaphysical and epistemic dimensions. Manuscript, University of Chicago and Institut Jean Nicod. Giannakidou, A., and S. Yoon. 2011. The subjective mode of comparison: Metalinguistic comparatives in Greek and Korean. Natural Language and Linguistic Theory 29: 621–655. Giorgi, A., and F. Pianesi. 1997. Tense and Aspect: From Semantics to Morphosyntax. Oxford: Oxford University Press. Grosz, Patrick. 2012. On the Grammar of Optative Constructions. Amsterdam: John Benjamins. Hacquard,  V. 2010. On the event relativity of modal auxiliaries. Natural Language Semantics 18: 79–114. Heim,  I. 1992. Presupposition projection and the semantics of attitude verbs. Journal of Semantics 9: 183–221. Hintikka, J. 1962. Knowledge and Belief. Ithaca, NY: Cornell University Press. Holton, D., P. Mackridge, and I. Philippaki-Warburton. 1997. Greek: A Comprehensive Grammar of the Modern Language. London: Routledge. Huddleston, C., and G. Pullum. 2002. The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press. Karttunen, L. 1971. Some observations on factivity. Papers in Linguistics 4(1): 55–69. Kiparksy, P., and C. Kiparsky. 1970. Fact. In Progress in Linguistics, ed. M. Bierwich and K. Heilolph. Berlin: Mouton de Gruyter. Kratzer, A. 1991. Modality. In Semantics: An International Handbook of Contemporary Research, ed. Arnim von Stechow and Dieter Wunderlich, 639– 650. Berlin: Mouton de Gruyter. Laca, B. 2010. On the temporal orientation of subjunctive complements in Spanish. Manuscript, Unversité Paris 8.

216

chapter five

Lasersohn,  P.  2005. Context dependence, disagreement, and predicates of personal taste. Linguistics and Philosophy 28: 643–686. Mari, A. 2014. Actuality entailments: Broadening the space of the possibilities. Manuscript, IJN and University of Chicago. Marneffe, M. de, C. Manning, and C. Potts. 2012. Did it happen? The pragmatic complexity of the veridicality judgement. Computational Linguistics 38: 300–333. Marques,  R. 2004. On the system of mood in Eu ropean and Brazilian Portuguese. Journal of Portuguese Linguistics 3: 89–109. ———. 2010. On the selection of mood in complement clauses. In Crosslinguistic Semantics of Tense, Aspect, and Modality, ed. L. Hogeweg, H. de Hoop, and A. Malchukov. Amsterdam: John Benjamins. Matthewson, L. 2010. Crosslinguistic variation in modality systems: The role of mood. In Semantics and Pragmatics 3: 1–74. Montague, R. 1969. On the nature of certain philosophical entities. The Monist 53: 159–194. Murray, S. Forthcoming. Evidentials and illocutionary mood in Cheyenne. International Journal of American Linguistics. Philippaki-Warburton, I, 1993. The subjunctive mood and the syntactic status of the particle na in Modern Greek. Folia Linguistica 28: 297–326. Portner, P. 1997. The semantics of mood, complementation, and conversational force. Natural Language Semantics 5: 167–212. Portner, P., and A. Rubinstein. 2012. Mood and contextual commitment. In The Proceedings of Semantics and Linguistic Theory (SALT) 22, ed. A. Chereces, 461–487. Cornell University. Ithaca, NY: CLC Publications. Potts, C. 2007. The expressive dimension. Theoretical Linguistics 7: 1–32. Potts,  C., and  S. Kawahara. 2004.  Japanese honorifics as emotive defi nite descriptions. In Proceedings of Semantics and Linguistic Theory 14, 235–254, ed. Kazuha Watanabe and Robert B. Young. Ithaca, NY: CLC Publications. Quer, J. 1998. Mood at the interface. PhD diss., University of Utrecht. ———. 2001. Interpreting mood. Probus 13: 81–111. ———. 2009. Twists of mood: The distribution and interpretation of the indicative and the subjunctive. Lingua 119: 1779–1787. Rivero,  M.  L. 1994. The structure of the Clause and V-movement in the languages of the Balkans. Natural Linguistic Theory 12: 63–120. Rouchota,  V. 1984. On the interpretation of na clauses in Greek: A relevance theoretic approach. Working Papers in Linguistics 3, 255–277, University College London. Roussou, A. 1994. The syntax of complementisers. PhD diss., University College London. ———. 2000. On the left periphery: Modal particles and complementizers. Journal of Greek Linguistics 1: 63–93.

Evaluative Subjunctive and Nonveridicality

217

Roussou, A., and A. Tsangalidis. 2010. Reconsidering the “modal particles” in Modern Greek. Journal of Greek Linguistics 10: 45–73. Sarigul, O. 2015. (Non)veridicality and complement clauses in Turkish. MA thesis, University of Chicago. Smirnova, A. 2013. Evidentiality in Bulgarian: Temporality, epistemic modality, and information source. Journal of Semantics 30(4): 479–532. ———. 2014. The semantics of mood in Bulgarian. Chicago Linguistic Society 48. Todorovic, N. 2012. Two kinds of subjunctive clause in Serbian. PhD thesis, UIC. Trnavac, R., and M. Taboada. 2012. The contribution of nonveridical rhetorical relations to evaluation in discourse. Language Sciences 34: 301–318. Varlokosta, S. 1994. Issues on Modern Greek sentential complementation. PhD diss., University of Maryland, College Park. Veloudis, J. 1983/84. I ipotaktikí stis anaforikés protásis [The subjunctive in relative clauses]. Glossologia 2–3: 111–135. Villalta,  E. 2008. Mood and gradability: An investigation of the subjunctive mood in Spanish. Linguistics and Philosophy 31: 467–522. von Fintel, K., and A. Gillies. 2010. Must . . . stay . . . strong! Natural Language Semantics 18: 351–383. Yoon, S. 2010. Not in the mood: The syntax and semantics of evalautive negation. PhD diss., University of Chicago. Zimmermann,  M. 2011. Discourse particles. In Semantics: An International Handbook of Natural Language Meaning, ed. C. Maienborn, K. von Heusinger, and P. Portner, 2011–2038. Berlin: Walter de Gruyter. Zwarts, F. 1995. Nonveridical contexts. Linguistic Analysis 25(3–4): 286–312.

chapter six

The Essence of a Category: Lessons from the Subjunctive Martina Wiltschko

1

Introduction

T

he subjunctive is not a homogeneous category across languages. I argue that we are inclined to label a language-specific category as a subjunctive if it contrasts with an independent main assertive clause type (sometimes labeled as indicative), as illustrated in Figure 6.1. In this chapter I show that this contrast can be established in dif ferent ways. Thus subjunctive is not a homogeneous natu ral class: it cannot be defi ned on the basis of form or interpretation. Consequently, any analysis that aims to treat the subjunctive as a primitive category is bound to fail. It is for this reason that, in the typological literature, categories are often treated as prototypes “with fuzzy boundaries” (Comrie 1989, p. 38). At the same time, the fact that there is no universal subjunctive category cannot be taken as evidence that there is no universal grammar, as, for example, Evans and Levinson (2009) do. I show here that there are universal under pinnings for the construction of a subjunctive. However, across languages this construction plays out in dif ferent ways. I conclude that language-specific categories are always constructed (Wiltschko 2014). The chapter is organized as follows: in section 2, I introduce a common analysis of subjunctive as a defective tense. This analysis defines the main analytical puzzle I address here: if indeed subjunctive is characterized as a defective tense, then what is its status in tenseless languages? To develop this analysis, I introduce in section  3 my assumptions about the status

The Essence of a Category

219

Halkomelem clause-types independent nominalized clauses clauses

imperative

subjunctive

Figure 6.1

of grammatical categories more generally. In particular, I review Ritter and Wiltschko’s (2014) analysis of tenselessness. This analysis is based on the premise that universal categories cannot be defined by their content. In section  4, I review the analysis of subjunctive in a tensed language, Cypriot Greek, as developed by Christodoulou and Wiltschko (2012). In section 5, I proceed to the original empirical contribution of this chapter: the description and analysis of an unusual subjunctive found in Upper Austrian German. What is unusual about the subjunctive in this language is, among other things, the fact that it may appear in matrix independent clauses. I argue that Upper Austrian German uses subjunctive instead of tense. However, I show in section 6 that while tenselessness is a necessary condition for this type of subjunctive to be constructed, it is not a sufficient condition. There are tenseless languages in which the subjunctive has a more regular distribution than in Upper Austrian German. In section 7, I show that the unified semantics for subjunctives as developed in Giannakidou (this volume) does not succeed in providing either the necessary or the sufficient conditions for subjunctive marking. I conclude in section 8 with a formal typology of subjunctives, as well as some remarks about the essence of grammatical categories.

2

What Is the Categorial Status of Subjunctives?

It is common for the subjunctive to be analyzed as a defective tense. For example, Picallo (1984, 1985) argues that subjunctive is featurally characterized as [–tense, +agr], in contrast with indicative clauses, which are [+tense, +agr]. A more recent incarnation of this view is developed in Giannakidou (2009), who argues that the essence of the subjunctive is its encoding of a dependency (see also Ambar, this volume). The defective-tense analysis has two major consequences. First, it requires the subjunctive to

chapter six

220

be dependent on a higher structure; and second, dependency results in the transparency of the subjunctive clause (i.e., embedded CP and TP do not create an opaque domain). (1)

[TP T . . . [CP [TP Tdefective . . . .]]]

2.1

Diagnosing Defective Tense

Evidence of its dependency stems from the fact that the temporal specification of the subjunctive clause depends on the matrix tense, as shown in (2), based on Catalan; and independent temporal reference cannot be established, as shown in (3), based on Greek. (2)

Catalan a. Sabia que telefona / telefonava know.ipfv.1sg that phone.ind.prs.3sg / phone.ind. ipfv.3sg ‘I knew that s/he calls/that she used to call.’ b. Desitjja que telefoni / *telefonés Desire.prs.3sg that phone.sbjv.prs.3sg / phone.sbjv.ipfv.3sg ‘S/he wishes that s/he calls/called.’

(3)

(Quer 2006, p. 2)

Greek *O eaftos to arxizi na ton anisixi avrio det self his.nom begin-3sg sbjv pc worry.3sg tomorrow ‘He started being worried about himself tomorrow.’ (Alexiadou and Anagnostopoulou 1999, p. 26 (31b))

Additional evidence for its dependency is the fact that, typically, the subjunctive cannot be used in matrix clauses. This is shown in (4), based on Catalan. (4)

Catalan a. *Daniel haya llamado Daniel has call.sbjv.prf.3sg ‘Daniel has called.’

The Essence of a Category

221

b. *Ahir plogués Yesterday rain.sbjv.ipfv.3sg ‘Yesterday it rained.’

(Quer 2006, p. 3)

As a result of this dependency, embedded clauses are transparent for different types of relations in the matrix clause. For example, subjunctive clauses may trigger obviation effects such that the embedded subject must be interpreted as disjoint from the matrix subject. This is shown in (5), based on Spanish. (5)

Spanish a. *Queremosi que ganemosi want.1pl that win.sbjv.prs.1pl ‘We want to win.’ b. Queremosi want.1pl

que ganenk that win.sbjv.prs.3pl

‘We want them to win.’

(Quer 2006, p. 29)

And fi nally, in some languages, subjunctives allow for long- distance anaphors, as shown in (6), based on Icelandic. (6)

Icelandic a. Jóni veit að Péturj rakar sig*i/j Jón know.3sg that Pétur shave.ind.3sg self ‘Jón knows that Pétur shaves himself.’ að Péturj raki sigi/j b. Jóni segir Jón say.3sg that Pétur shave.sbjv.3sg self ‘Jón says that Pétur shaves him(self).’

(Quer 2006, p. 29)

2.2 Subjunctives in Tenseless Languages The particular problem I address in this chapter concerns the categorial status of subjunctives in tenseless languages. As is well known, not all languages make use of the morphosyntactic category tense. This is evident from the fact that the temporal interpretation of matrix clauses is

chapter six

222

Blackfoot clause-types ORDER MODE

independent independent

conjunct

imperative

subjunctive

unreal

Figure 6.2

not dependent on overt tense marking. Instead, an unmarked predicate is compatible with both a present and a past interpretation. This is true for both Blackfoot (Algonquian) and Halkomelem (Salish) (Ritter and Wiltschko 2009, 2014). (7)

Blackfoot Oma píítaawa áípaawaniwa. om-wa píítaa-wa a-ipaawani-wa dem-prox eagle-prox ipfv-fly.AI-prox ‘That eagle is/was flying up.’ (adapted from Reis Silva and Matthewson 2008, p. 8)

(8)

Halkomelem í qw’eyílex tútl’ò aux dance he ‘He is/was dancing.’

What is interesting, from the present perspective, is that these tenseless languages have a category that is labeled as a subjunctive in the relevant reference grammars. The Blackfoot language has five different clausetypes, known as orders and modes, as schematized in Figure 6.2. The clause-type classified as subjunctive is found, for example, in conditional clauses such as (9). (9)

Blackfoot Nitsiikaakaahsi’taki aotoyaakihtsiniki nit-iik-aak-yaahssi-i’taki a- oto-yáakihts-iniki 1-int-fut-be.good.vti-feel.emotion ipfv-go.to.do-go.to.bed-sbjv ‘It will make me happy when you go to bed.’ (BB)

The Essence of a Category

223

Halkomelem clause-types independent nominalized clauses clauses

imperative

subjunctive

Figure 6.3

Interestingly, however, there is another clause-type, which, based on its interpretation, might be classified as a type of subjunctive. It refers to unreal situations and occurs in counterfactual conditionals. This clausetype, illustrated in (10), is labeled as unreal in Frantz’s (1991) grammar of Blackfoot. (10)

Blackfoot Kitsaotoyaakihts-htopi nitsiikaak- ok-i’taki kit-saw- oto-yáakihts-htopi nit-iik-aak- ok-i’taki 2-neg-go.do.do-go.to.bed-unreal 1-int-fut-bad-feel.emotion ‘If you hadn’t gone to bed, I would have been mad.’ (BB)

Turning to Halkomelem, we also observe the existence of a clause-type that has been labeled as a subjunctive by Galloway (1993). It exists alongside three other clause-types: independent, nominalized, and imperative, as shown in Figure 6.3. The subjunctive in this language has a distribution that is commonly associated with subjunctives: it occurs in negative sentences such as (11), as well as in conditionals (12). (11)

Halkomelem a. éwe i- s lhemexw neg aux-3ss rain ‘It is not raining.’ b. éwe í- s neg aux-3s

kw’éts-l-àlèm see-tr-1sg.pass

‘I wasn’t seen.’

chapter six

224

Upper Austrian clause-types indicative

infinitive

imperative

subjunctive

Figure 6.4

(12)

Halkomelem a. we li- s l-stl’i kw- el-s lam lam-tsel if aux-3sg.sbjv 1sg.poss-want comp-1sg.poss-nom go go-1sg.sbj ‘If I want to go, I will go.’

(Galloway 1993, p. 451)

b. li t’wa iyolem we i-l- elh ts-xelce aux prt alright comp aux-1sg.sbjv – pst vbl- catch ‘It would be good if I caught a fish.’

(Thompson 2012, p. 45 (43))

Finally, Upper Austrian German, another tenseless language (see section 5.1 for evidence), also has a subjunctive. In this language, subjunctive clauses contrast with other clause-types, namely the indicative, the imperative, and the infi nitive, as illustrated in Figure 6.4. The subjunctive occurs in some of the typical subjunctive environments: optatives (13a) and reportatives (13b). (13)

Upper Austrian German a. Wonn a nua ham gang-at. If he only home go-sbjv ‘If only he went home.’ b. Ea hot gsogt . . . He has said . . . . . . ea gan-at gean ham. . . . he go-sbjv preferably home ‘He said he would like to go home.’

If the subjunctive is indeed defi ned as a defective tense, then the question arises, what is its categorial status in tenseless languages? If there is

The Essence of a Category

225

no tense, what does it mean to be a defective tense? To answer this question, we fi rst need to establish the syntax of tenselessness.

3

The Syntax of Tenselessness

Recall that the main question I am pursuing here is whether there is a category ‘subjunctive’ that can be universally characterized. ‘Defectiveness of tense’ features prominently in a number of analyses of subjunctives and may thus be viewed as a universal property of subjunctives. If that is so, the question arises as to whether this property also characterizes subjunctives in tenseless languages. To answer this question, we must have in place a clear understanding of tenselessness. In particular, there appear to be two types of tenseless languages: superficially tenseless languages (Matthewson 2006) and genuinely tenseless languages (Zimmermann and Mucha, this volume). The former are characterized by the presence of a silent morpheme restricting the temporal interpretation of the clause, whereas the latter are characterized by the absence of any such morphemes, overt or covert. It is especially in the genuinely tenseless languages that the question regarding the identifying criteria for subjunctives becomes critical. In the absence of tense, there cannot be a defective tense. Hence defectiveness of tense cannot be the defi ning property of subjunctive, since, as the examples here show, at least some tenseless languages have constructions that are labeled as subjunctive. But if there are genuinely tenseless languages that do indeed have a subjunctive, then we need another way to characterize its defi ning property. Thus, to compare categories across languages that differ in the inventory of grammatical categories, we need a formal typology. To develop a typology for subjunctives, including those of tenseless languages, I adopt the framework I introduced with Elizabeth Ritter (see Ritter and Wiltschko 2014 (henceforth R&W)). Our main thesis is that functional categories are not intrinsically associated with substantive content, but instead, content is associated with universal abstract categories on a language-specific basis (see also Wiltschko 2014). Here, I briefly review our proposal.

3.1

Tense

If functional categories are not intrinsically associated with substantive content, their label should also not be based on the substantive content

chapter six

226

they spell out. Thus R&W use the label INFL instead of Tense. Building on work by Demirdache and Uribe-Etxebarria (1997), we propose that INFL is intrinsically equipped with an unvalued coincidence feature [ucoin].1 This coincidence feature is responsible for ordering two abstract arguments: the utterance situation in SpecIP and the event situation in SpecVP.2 Following minimalist assumptions, unvalued features must be valued in the course of the derivation. R&W depart from standard minimalism in assuming that it is the substantive content of the linguistic objects (henceforth LOs) associated with INFL that is responsible for valuation. In particular, present tense marking values [ucoin] as [+coin] (14a), while past tense marking values INFL as [– coin] (14b). (14)

a. Yoshi likes his ball.

b. Yoshi lik-ed his ball

IP

IP

I

Utt-time INFL [+coin] | present

I

Utt-time

Ev-time

INFL [–coin] | past

Ev-time

The valuation of [ucoin] via the substantive content associated with the morphological marking is referred to as m(orphological)-valuation. Note further that the substantive content associated with the valuation strategy not only determines the interpretation of the functional head, but it also influences the interpretation of the abstract situation arguments that INFL orders. In particular, in the case of temporal content, these abstract arguments are interpreted as times. This is consistent with the assumption that situations contain temporal coordinates, in addition to spatial coordinates as well as participants.

3.2

Tenseless Languages

In this view of categories, tenselessness arises as follows: if the function of INFL is dissociated from its substantive content, we may expect [ucoin] to be m-valued by LOs of different substantive content. In particular, R&W argue that in Halkomelem locative auxiliaries serve to mvalue [ucoin]: proximate auxiliaries value INFL as [+coin], asserting that the event location coincides with the utterance location and distal aux-

The Essence of a Category

227

iliaries value INFL as [– coin], asserting that the event location does not coincide with the utterance location. (15)

a. í qw’eyílex tútl’ò prox dance he ‘He is/was dancing [here].’ b. lí qw’eyílex tútl’ò dist dance he ‘He is/was dancing [there].’

(16)

a. Proximate auxiliaries

b. distal auxiliaries

IP

IP

I

Utt-loc INFL [+coin] | proximate

I

Utt-loc

Ev-loc

INFL [–coin] | distal

Ev-loc

As for Blackfoot, R&W argue that in this language [ucoin] is valued by participant marking. In par ticu lar, local person marking (fi rst and second person) values INFL as [+coin], asserting that the event participant coincides with the utterance participant (speaker or addressee), while nonlocal person marking (third person) values INFL as [– coin], asserting that the event participant does not coincide with the utterance participant. (17)

a. Kitsinóóhpoaawa kit-ino- o-hp- oaawa 2-see-1:2-local-2pl ‘I saw you (pl).’ b. Ana póókaawa inoyííwa ani imitááyi an-(w)a pookaa-wa ino-yii- Ø-wa an-(y)i imitaa-yi dem-prox child-prox see-dir-3-prox dem- obv dog- obv ‘The child saw the dog.’

chapter six

228

TENSE PERSON 



LOCATION 



vP

temporal


participant spatial Figure 6.5

(18)

a. local person

b. nonlocal person

IP

IP

I

Utt-part INFL [+coin] | 1st/2nd

I

Utt-part

Ev-part

INFL [–coin] | 3rd

Ev-part

Thus, in this view, the universal category INFL is the universal basis for the construction of three language-specific grammatical categories: tense, location, and person. This is schematized in Figure 6.5. Tenseless languages are defi ned as those in which INFL is valued by nontemporal content.

3.3

Tenseless Constructions

In addition to providing a framework for analyzing tenseless languages, R&W also explore tenseless clause-types within other wise tensed languages, such as infi nitives and imperatives, as well as constructions in which tense appears to be fake, as in counterfactuals. Consider fi rst

The Essence of a Category

229

infi nitives. There are at least two types of infi nitival complements: simultaneous infi nitives (19) and future irrealis infi nitives (20): (19)

Yoshi is starting to play.

(20)

Yoshi wants to play.

In the absence of tense marking, [ucoin] in INFL of the embedded predicate is valued by the substantive content of the embedding predicate. Aspectual verbs selecting for simultaneous infi nitives value [ucoin] as [+coin], whereas future- oriented verbs selecting for future irrealis infi nitives value it as [– coin]. In this case, the abstract situation argument associated with SpecIP is pronominal. Its deictic character in matrix clauses is a default interpretation in the absence of an appropriate antecedent (Enç 1987). Thus a simultaneous infi nitive asserts that the embedded event coincides with the event denoted by the embedding predicate (21), while a future irrealis infi nitive asserts that the embedded event does not coincide with it (21). (21)

a. simultaneous infi nitive

b. future irrealis infi nitive VP

VP

Ev-sit

Ev-sit

V

I [+coin]

I

Pro-sit

I

Pro-sit

IP

V futurate

IP

V aspectual

V

VP

Ev-sit

I [–coin]

VP

Ev-sit

The valuation of [ucoin] via the substantive content associated with the embedding predicate is referred to as pred(icate)-valuation. The third valuation strategy is via a higher functional head. In the case of INFL, the head is C, and thus this strategy is labeled C-valuation. R&W argue that in imperatives [ucoin] is valued by the substantive content associated with imperative C (22). The directive force values it as [+coin]. R&W follow Han (2006) in assuming that directive force introduces a plan set, which R&W assume to be associated with SpecCP. 3

chapter six

230

Thus an imperative directs the addressee to put the event situation into their plan set (24). The [– coin] value is argued to be derived via the counterfactuality associated with C (23). In this configuration (i.e., within the context of the if-clause), INFL asserts that the event situation (having a car) does not coincide with the evaluation world associated with SpecCP (24). (22)

(You) be quiet!

(23)

If I had a car right now, I would drive.

(24)

a. directive force

b. counterfactuals CP

CP

Plan-set

Eval-sit

C

C counterfactual

IP

C directive

I [+coin]

IP

I

Pro-sit

I

Pro-sit

C

VP

Ev-sit

I [–coin]

VP

Ev-sit

In sum, according to R&W’s framework, there is a universal category INFL, which is associated with an unvalued coincidence feature. This feature is valued by substantive content, but there are three different strategies for valuation: m-valuation, pred-valuation, and C-valuation. Tensed clauses are defi ned as those in which morphological tense marking serves to value INFL via m-valuation. Tenseless clauses have in common that their [ucoin] feature is valued by substantive content associated with other syntactic heads. This is summarized in Figure 6.6.

4

The Syntax of Subjunctives: The Case of Cypriot Greek

The typology for INFL-valuation, developed by R&W, is the basis for the analysis of the subjunctive in Cypriot Greek by Christodoulou and Wiltschko (2012) (henceforth C&W). Cypriot Greek is a tensed language. The subjunctive marker na is found in all of the tenseless constructions identified in Figure  6.6: simultaneous and future irrealis events as well as imperatives and counterfactuals. Here I discuss each of them in turn.

The Essence of a Category

231

[u coin] m-valuation

[+coin] present tense

[-coin] past tense

pred- valuation

[-coin] [+coin] simultaneous future infinitives irrealis infinitive

C-valuation

[-coin] [+coin] counterimperative factual

Figure 6.6

Aspectual verbs, such as arxis- (‘start’) and katafer- (‘manage’) embed complement clauses introduced by na (25). (25)

a. O Kostas arxis- e . . . det Costas start.prf-pst.3.sg . . . . . . na pez-i kithara . . . sbjv play.ipfv-prs.3.sg guitar ‘Costa has started playing the guitar.’ b. Katafer -a . . . manage.prf-pst.1.sg . . . . . . na parados- o ti diatrivi mu. . . . sbjv submit.prf-dep.1.sg det dissertation 1sg.gen ‘I managed to submit my dissertation.’

In (25), the embedded event is interpreted as occurring simultaneously with the matrix event consistent with the analysis according to which the matrix predicate values INFL as [+coin]. Future- oriented verbs like thel(‘want’) also embed complement clauses introduced by na. (26)

a. thel-is . . . want.ipfv-prs.2.sg . . . . . . na par-ume liga frut -a? . . . sbjv take.prf-dep.1.pl little fruit-n.pl.acc ‘Would you like us to get some fruit?’

chapter six

232

b. i-thel-a . . . past-want.ipfv-past.1.sg . . . na pernus-ame ap’ ti vivliothiki. . . . sbjv pass.ipfv-pst.1.pl from det library ‘I wanted us to pass by the library.’ In (26), the embedded event is interpreted as following the matrix event, if it occurs at all. This is consistent with the analysis according to which the matrix predicate values INFL as [– coin]. Contexts of C-valuation are also consistent with na. Both imperatives (27) and counterfactuals (28) are marked as subjunctive via na. (27)

na mas grap-s- ete. sbjv 1.pl.gen write-prf-dep.2.pl ‘(do) write to us!’

(28)

An kerdiz-es to laxio . . . if win.ipfv-pst.2.sg det lottery . . . . . . na anakeniz -es to spiti . . . sbjv renovate.ipfv-pst.2.sg det house ‘If you had won the lottery, you would have been able to renovate the house.’

This establishes that na is used in all contexts where INFL is valued from a higher head: via predicate valuation or via C-valuation. Conversely, na cannot be used in the context of m-valuation (i.e., in indicative clausetypes with present or past marking). This is shown in (29)–(31). (29)

a. O Petr-os kolimb-ai kaθe proi. det Petr-nom swim.ipfv-pres.3.sg every morning ‘Peter swims every morning.’ b. O Petros kolimb-ai tora. det Petr-nom swim.ipfv-pres.3.sg now ‘Peter is swimming right now.’

The Essence of a Category

(30)

233

a. O Petr-os kolimb-use kaθe proi. det Petr-nom swim.IPFV-PST.3.SG every morning ‘Peter was swimming/used to swim every morning.’ b. O Petr-os kolimb-is- e xθes to proi. det Petr-nom swim-prf-prs.3.sg yesterday det morning ‘Peter swam yesterday morning.’

(31)

a. *O det

Petr-os na kolimb-ai tora.4 Petros-nom sbjv swim.ipfv-prs.3.sg now

b. */??O Petr-os na kolimb-is-i tora. det Petros-nom sbjv swim-prf-dep.3.sg now ‘Peter should swim right now.’ It is not immediately clear what these clause-types that are introduced by na have in common. They do not form an obvious natural class. However, in the typology of INFL-valuation, the distribution of na can be straightforwardly stated: na is used when INFL is not valued via m-valuation (i.e., in the context of predicate- and C-valuation). To unify the use of na across predicate- and C-valuation, C&W suggest that na spells out the unvalued coincidence feature in INFL. In other words, na lexicalizes [ucoin]. As per UG, unvalued features must be valued, and hence the presence of na indicates that INFL is yet to be valued. At this point in the derivation this must be achieved via a higher head, and thus na appears to be dependent on a higher head. Na marking in Cypriot Greek qualifies as a subjunctive because it contrasts with the independent clause-type. This contrast comes about because it is impossible to spell out an unvalued feature in the context of m-valuation. Accordingly, the Cypriot Greek subjunctive qualifies as the elsewhere case (see also Portner 1997; Schlenker 2005; Siegel 2009). In as much as this analysis of na is successful, it provides us with further support for the dissociation of the function of a functional category from its substantive content. If function and content were intrinsically related, we would not expect there to be exponents of functional categories that lack content. But that is precisely what na is: a functor without content. It is therefore not surprising that meaning-based analyses are not successful (see section 7 for further discussion).

chapter six

234

As mentioned earlier, several traditional analyses of the subjunctive have treated it as an instantiation of defective tense. In the case of na, this is unsurprising given the typology of INFL-valuation: na occurs in precisely those environments where m-marking, in the form of tense, does not. However, the typology developed here has the added advantage that it can straightforwardly be extended to subjunctives in tenseless languages, such as Halkomelem, Blackfoot, and Upper Austrian German, as I will now show. In contrast, it is not clear what predictions the defective-tense analysis would make in light of subjunctives in tenseless languages.

5 Subjunctive in Upper Austrian German The subjunctive in Upper Austrian contrasts with the indicative more directly. In particular, I argue that subjunctive marking values [ucoin] in INFL as [– coin] via m-valuation, while indicative values it as [+coin]. Thus, in this dialect, the indicative/subjunctive contrast replaces the present/past contrast found in Standard German. The latter is a contrast based on tense, whereas the former, I suggest, is a contrast based on realis: whether or not the event is part of the utterance world (i.e., the real world). I begin by offering evidence that Upper Austrian is tenseless (5.1). I then show that it has a productive subjunctive (5.2), which can be analyzed as valuing INFL as [– coin] via m-valuation (5.3). And fi nally, I discuss the predictions of this analysis, showing that the Upper Austrian subjunctive, unlike those of other languages, does not show the usual dependency effects (5.4).

5.1

Upper Austrian German Is Tenseless

According to the criteria developed by R&W, the Upper Austrian dialect of German is tenseless: there is no obligatory morphosyntactic contrast between present and past. The fi rst argument to this effect comes from the fact that there is no dedicated form for the simple past. To see this, compare the Upper Austrian verbal paradigm (Table 6.1) with the paradigm found in Standard German (Table 6.2). The forms corresponding to the Standard German simple past are unattested in Upper Austrian German. This holds true for both strong verbs, where past in Standard German is marked via ablaut, as well as for weak verbs, where past in Standard German is marked by the suffi x – te. Note that it is not just the

The Essence of a Category

235

table 6.1 Standard German Present and Past Strong Verbs

Weak Verbs

Feature Specification

present

simple past

present

simple past

1sg 2sg 3sg 1pl 2pl 3pl

Ich komm- e Du komm- st Er komm- t Wir komm- en Ihr komm- t Sie komm- en

Ich kam Du kam- st Er kam Wir kam- en Ihr kam- t Sie kam- en

Ich koch- e Du koch- st Er koch- t Wir koch- en Ihr koch- t Sie koch- en

Ich koch- te Du koch- te- st Er koch- te Wir koch- te- n Ihr koch- te- t Sie koch- te- n

table 6.2 Upper Austrian Present; Past Forms Not Attested Strong Verbs

Weak Verbs

Feature Specification

unmarked

*simple past

unmarked

*simple past

1sg 2sg 3sg 1pl 2pl 3pl

I kum Du kum- st Ea kum- t Mia kum- en Ia kum- ts Si kum- en

*I kam *Du kam- st *Ea kam *Mia kam- en *Ia kam- ts *Si kam- en

I koch Du koch- st Ea koch- t Mia koch- n Ia koch- ts Si koch- n

*I koch- te *Du koch- te- st *Ea koch- te *Mia koch- te- n *Ia koch- te- ts *Si koch- te- n

(* indicates unattested forms)

case that a particular morphological exponent is lost; both allomorphs (ablaut and – te suffi xation) are equally affected: neither of them exists as a marker of past tense. Second, the unmarked verb form is compatible with present, past, and future time references, as indicated by the use of temporal adverbials. The compatibility of the unmarked form with present, past, and future interpretations is consistent with an analysis of Upper Austrian as a genuinely tenseless language (see Zimmermann and Mucha, this volume). This is shown for strong verbs in (32) and for weak verbs in (33). (32)

a. I kum grod ham. I come now home

present

‘I am coming home right now.’ b. I kum gestan ham. I come yesterday home ‘I came home yesterday.’

past5

chapter six

236

c. I kum moagn ham. I come tomorrow home

future

‘I will come home tomorrow.’ (33)

a. I koch grod. I cook now

present

‘I am cooking right now.’ b. I koch gestan. I cook yesterday

past

‘I was cooking yesterday.’ c. I koch moagn. I cook tomorrow

future

‘I will cook tomorrow.’ To unambiguously establish the temporal interpretation, a periphrastic construction with an auxiliary and a nonfi nite form of the verb are used. Specifically, a past interpretation is achieved with a present perfect construal with be (34a) or have (34b), depending on the main verb and a past participle. A present interpretation is achieved with a prepositional construction (‘I am at verb-ing’; (35)). And a future interpretation is achieved with the modal verb werden and the main verb realized as an infi nitive (36). (34)

a. I bin ham kuma I am home come.ptcp

past (via present perfect)

‘I came home yesterday.’ b. I hob kocht I have cook.ptcp ‘I came home.’ (35)

a. I bin am ham kuma I am at home come.ptcp ‘I am coming home.’

present

The Essence of a Category

237

b. I bin am kochn I am at cook.inf ‘I am cooking.’ (36)

a. I wead ham kum-a I will home come-inf

future (via modalization)

‘I will come home.’ b. I wead koch-n I will cook-inf ‘I will cook.’ These examples confi rm general patterns of temporal interpretations: a past interpretation in tenseless languages is often accomplished aspectually (see, e.g., Lin 2006 for Chinese). For completeness, note that there is a simple past for the auxiliary sein (‘to be’) as shown in (37). But even in the paradigm for be, we observe the absence of past as a category. The evidence here has to do with the absence of a past perfect form in Upper Austrian German (38). This contrasts with Standard German, where such a form exists (39). (37)

a. I bin kronk. I be.prs.1sg. sick ‘I am sick.’ b. I woa kronk. I be.pst.1sg. sick ‘I was sick.’

(38)

a. I bin kronk gwen. I be.pres.1sg. sick be.ptcp ‘I was/have been sick.’ b. *I woa krank gwen. I be.pst.1sg. sick be.ptcp ‘I had been sick.’

chapter six

238

(39)

a. Ich bin krank gewesen. I be.prs.1sg. sick be.ptcp ‘I was/have been sick.’ b. I war krank gewesen. I be.past.1sg. sick be.ptcp ‘I had been sick.’

5.2

Upper Austrian German Has a Subjunctive

As noted above, neither ablaut nor suffi xation is attested as a past marker in Upper Austrian. However, both of them exist as markers of subjunctive. Weak verbs have the suffi x – at, as shown in Table 6.3. For strong verbs, the morphological properties of Upper Austrian subjunctive marking differ from Standard German past. Specifically, as illustrated in Table 6.4, there are three morphological strategies to mark strong verbs as subjunctive: ablaut, – at suffi xation (same as with weak verbs), and ablaut combined with – at suffi xation.6 The subjunctive in Standard German differs in form from its Upper Austrian counterpart. Specifically, Standard German has two types of subjunctives, known as Konjunktiv 1 and Konjunktiv 2, with different distributional properties. As illustrated in Table  6.5, in weak verbs the former is marked with a suffi x - e, while the latter is identical in form with past tense marking. The homophony between past tense and Konjunktiv 2 has been argued to be responsible for the decline in the use of the synthetic subjunctive in favor of a periphrastic construction (Fabricius-Hansen and

table 6.3 Subjunctive Marking in Upper Austrian Weak Verbs Weak Verb: koch ‘cook’ Feature Specification

unmarked

subjunctive

1sg 2sg 3sg 1pl 2pl 3pl

I Koch Du koch- st Ea koch- t Mia koch- n Ia koch- ts Si koch- n

I koch- at Du koch- at- st Ea koch- at- t Mia koch- at- n Ia koch- at- ts Si koch- at- n

The Essence of a Category

239

table 6.4 Subjunctive Marking in Upper Austrian Weak Verbs Strong Verb: kum ‘come’ Feature Specification

unmarked

subjunctive: ablaut

subjunctive: unmarked -at

subjunctive: ablaut -at

1sg 2sg 3sg 1pl 2pl 3pl

I kum Du kum- st Ea kum- t Mia kum- en Ia kum- ts Si kum- en

I kam Du kam- st Ea kam ?Mia kam- en Ia kam- ts ?Si kam- en

I kum- at Du kum- at- st Ea kum- at- t Mia kum- at- n Ia kum- at- ts Si kum- at- en

I kam- at Du kam- at- st Ea kam- at Mia kam- at- n Ia kam- at- ts Si kam- at- n

table 6.5 Subjunctive Marking in Standard German Weak Verbs Weak Verb: koch ‘cook’ Feature Specification

present

past: -te

Konjunktiv 1: - e

konjuntiv 2: te

1sg 2sg 3sg 1pl 2pl 3pl

Ich koch- e Du koch - st Er koch - t Wir koch - en Ihr koch - t Sie koch - en

Ich koch- te Du koch- te- st Er koch- te Wir koch- te- n Ihr koch- te- t Sie koch- te- n

Ich koch- e Du koch- e- st Er koch e Wir koch- en Ihr koch- e- t Sie koch- en

Ich koch- te Du koch- te-st Er koch- te Wir koch- te- n Ihr koch- te- t Sie koch- te- en.

table 6.6 Subjunctive Marking in Standard German Strong Verbs Strong Verb: komm ‘come’ Feature Specification

present

past: ablaut

Konjunktiv 1: - e

Konjuntiv 2: ablaut - e

1sg 2sg 3sg 1pl 2pl 3pl

Ich komm- e Du komm- st Er komm- t Wir komm- en Ihr komm- t Sie komm- en

Ich kam Du kam- st Er kam Wir kam- en Ihr kam- t Sie kam- en

Ich komm- e Du komm- e- st Er komm- e Wir komm- en Ihr komm- e- t Sie komm- en

Ich käm- e Du käm- e-st Er käm- e Wir käm- en Ihr käm- e- t Sie käm- en.

Sæbø 2004). As illustrated in Table  6.6, for strong verbs, Konjunktiv 1 is based on the form for the present tense with an additional suffi x – e, whereas Konjunktiv 2 is formed by means of ablaut, albeit a different type of ablaut than past tense and the subjunctive – e suffi x. Thus, in the case of strong verbs, past and subjunctive marking are not homophonous (though the periphrastic construction is still preferred). In sum, we have seen that the form of subjunctive marking in Upper Austrian is more similar to past tense marking in Standard German than

chapter six

240

it is to subjunctive marking. This is consistent with the claim that Upper Austrian subjunctive and Standard German past are associated with the same universal category INFL. This is despite the fact that they are classified as different categories according to traditional grammatical analysis (mood vs. tense, respectively).

5.3

The Subjunctive in Upper Austrian German Is Associated with INFL

I propose that Upper Austrian subjunctive marking is associated with INFL, thus instantiating another language-specific guise of INFL. Its morphosyntactic distribution is consistent with this proposal. Consider again the fourth column of the paradigm in Table 6.4. Crucially, the subjunctive suffi x – at precedes the agreement suffixes (kam- atsubj - stagr). Assuming that agreement marking may attach to whatever LO is associated with INFL, we may conclude that subjunctive marking is associated with INFL. Interestingly, agreement marking for second person in Upper Austrian is associated with (finite) verbs and with complementizers (Bayer 1984). Specifically, in the absence of a complementizer, subject agreement is realized on the verb, as shown in (40). In the presence of a complementizer, the same subject agreement marker is associated with both the verb and the complementizer (41). (40)

a. nua du kumm- st only you come-2sg ‘only you come are coming . . .’ b. nua es kumm-ts only you.2pl come-2pl ‘if only you guys come . . .’

(41)

a. Wenn- st nua du kumm-st if-2sg only you come-2sg ‘if only you come . . .’ b. Wenn-ts nua es kumm-ts if-2pl only you.2pl come-2pl ‘if only you guys come . . .’

The Essence of a Category

241

Crucially, subjunctive marking is restricted to verbs (42) and cannot attach to complementizers (43). (42)

a. nua du only you

kumm-at-st come-sbjv-2sg

‘Only you would come.’ b. nua es kumm-at-ts only you.pl come-sbjv-2sg ‘Only you guys would come.’ (43)

a. wonn-st nua du kumm-at-st if-2sg only you come-sbjv-2sg ‘If only you would come.’ b. *wonn-at-st nua es kumm-at-ts if- subj-2sg only you.pl come-sbjv-2sg ‘Only you guys would come.’ c. *wonn-at-st nua es kumm-ts if- subj-2sg only you.pl come-sbjv-2sg ‘Only you guys would come.’

This suggests that subjunctive marking is associated with a position lower than that for complementizers. At the same time, there is evidence that subjunctive marking must be associated with a position higher than that of viewpoint aspect. We know this because subjunctive marking may co- occur with viewpoint aspect marking. In particular, subjunctive marking is compatible with both imperfective (unmarked) and perfective (marked) constructions. And crucially, it is associated with the fi nite auxiliary verb rather than with the participle. This holds for both strong verbs (44) and weak verbs (45). This pattern is reminiscent of tense marking, which is generally assumed to be associated with INFL. (44)

a. E kam-at eh. he come.sbjv-sbjv prt ‘He would come anyhow.’

chapter six

242

b. Ea war-at kumma. he was.sbjv-sbjv come.ptcp ‘He would have come anyhow.’ (45)

a. Ea koch-at eh. he cook-sbjv prt ‘He would cook anyhow.’ b. Ea hed-at he has-sbjv

eh kocht. prt cook.ptcp

‘He would have cooked anyhow.’ Thus the morphosyntactic patterning of subjunctive marking is consistent with the claim that it is associated with INFL. Expanding on R&W’s analysis, I propose that it serves to value [ucoin] as [– coin], asserting that the event situation does not coincide with the evaluation world (46). (46)

Upper Austrian subjunctive marking values INFL as [– coin] CP

Eval-world

C

IP

C

Pro-sit I [–coin] -at SUBJUNCTIVE

I

VP

Ev-sit

Note that Upper Austrian is a verb-second language, and as such the evaluation world argument in SpecCP is always available as an antecedent for the pronominal situation argument associated with SpecIP (see Wiltschko 2014 for more elaborate discussion).

The Essence of a Category

5.4

243

Predictions

If indeed in Upper Austrian, the indicative/subjunctive contrast replaces the present/past contrast found in Standard German, I predict that the subjunctive should behave like an independent clause-type. This prediction is borne out. The Upper Austrian subjunctive shows none of the dependent properties characteristic of other subjunctives (see section 2.1). First, and probably most important, it may be used in matrix clauses (47).7 (47)

a. Ea ruaf-at o. He call-sbjv prt ‘He (would) call.’ b. Es regn-at. it rain-sbjv ‘It would rain.’

Second, the temporal specification of the embedded subjunctive clause is independent of the temporal specification in the embedding clause (48). That is, the embedded clause may be (periphrastically) specified as present, perfect, or future, even though the matrix clause is specified as perfect. (48)

a. Ea hot gsogt, du kumm-at-st He has said.perf you come-sbjv-agr ‘He said you would come.’ b. Ea hot gsogt, du war-at-st kumma He has said.perf you was-sbjv-agr come.ptcp ‘He said you would have come.’ c. Ea hot gsogt, du wuat-at-st kumma. He has said.perf you will-sbjv-agr come.ptcp ‘He said you would come.’

The temporal independence correlates with the fact that it allows for independent temporal reference, as shown in (49).

chapter six

244

(49)

Ea hot gestan gsogt dass a moagn hamgang-at. he has yesterday said that he tomorrow home go-sbjv ‘He said yesterday that he would go home tomorrow.’

And fi nally, the independence of the subjunctive further correlates with the fact that it does not create a transparent clause. Thus the subjunctive does not create obviation effects (50); nor does it allow for long- distance anaphors (51). (50)

Ea vasuach-(at)t (eh) dass a gwinn-at He try- sbjv prt that he win-sbjv ‘He is trying to win.’

(51)

a. Da Honsj hot gsagt dass da Pedai si i/*j rasiert. det H has said that det peter refl shave ‘Hans said that Peter shaved himself.’ b. Da Honsj hot gsagt dass da Pedai si i/*j rasier-at. det H has said that det peter refl shave-sbjv ‘Hans said that Peter would shave himself.’

Thus the Upper Austrian subjunctive differs in many respects from the subjunctives found in Romance and Balkan languages. However, it shares in common with these subjunctives that it is associated with INFL and that it contrasts with assertive clauses. In fact, I argue that it does so in the most direct way, namely by being a realis-based category that serves to m-value INFL in independent clauses. In this way, the tenselessness of Upper Austrian is a necessary condition for this type of subjunctive to be constructed, and so is the presence of a CP to introduce the evaluation world. In the next section, however, I will show that tenselessness is not a sufficient condition.

The Essence of a Category

6

245

Subjunctive in Other Tenseless Languages

Here, I briefly show that in two tenseless languages, Halkomelem and Blackfoot, the subjunctive has a more usual distribution than it does in Upper Austrian (see R&W 2014). Consider fi rst Halkomelem. One criterion of a genuinely tenseless language is the compatibility of an unmarked predicate with either a present, past, or future interpretation (Matthewson 2006; Zimmermann and Mucha, this volume). This is the case for Halkomelem. The sentence in (52) can be interpreted as present, past, or future. (52)

tsel í:mex. 1sg.s walk ‘I am/was/will be walking.’

(Wiltschko 2002, p. 687 (62))

As section 2.2 shows, Halkomelem has a subjunctive, which is restricted to negative clauses and conditionals. These are dependent clauses. This characteristic sets it apart from Upper Austrian subjunctives, which is expected, given that INFL is substantiated by location and thus cannot simultaneously be realis based. However, its distribution also differs from the subjunctive in Cypriot Greek in that it cannot be used for commands or in clauses embedded under verbal predicates. Instead, a command is either marked with a dedicated imperative marker – lha (53) or realized as an independent clause (53). And embedded clauses are always nominalized (Galloway 1993; Thompson 2012, as in (54). (53)

a. omet = lha sit.down = imp ‘(You) sit down.’

(Galloway 1993, p. 310)

b. xó:lh-me-thet = chexw look.after-red-refl2sg.s ‘(You) take care of yourself.’ (54)

(Galloway 1993, p. 311)

a. tsel tot-et kwels xwemxolem] 1sg.sbj try-tr-3o [comp = 1sg.poss = nom run-mid] ‘I tried to run.’

(Thompson 2012, p. 158 (24))

246

chapter six

b. tsel iyo-thet [kwels xwemxolem] 1sg.sbj start-refl [comp = 1sg.poss = om run-mid] ‘I started running/to run.’ (Thompson, 2012, p. 159 (25)) We can conclude that the Halkomelem subjunctive must be licensed by the presence of either the negation ewe or the conditional complementizer we. Since both of these elements have been independently argued to be associated with C (Wiltschko 2002), this suggests that Halkomelem subjunctive is restricted to contexts of C-valuation. I therefore submit that the Halkomelem subjunctive instantiates the [– coin] value of C-valuation. I turn now to Blackfoot. We have seen that this tenseless language has a subjunctive as well. Regarding Blackfoot’s tenselessness, a few remarks are in order. First, Blackfoot has been characterized as a superficially tenseless language by Reis Silva and Matthewson (2008). In particular, they postulate that there are two silent tense morphemes: one with a present and one with a past temporal interpretation. If that is so, we may conclude that Blackfoot is not really tenseless, and thus, everything else being equal, its subjunctive may be characterized as defectiveness of tense. However, everything is not equal. In particular, Louie (2015) concludes that the silent temporal (and aspectual) morphemes of Blackfoot merge much lower than they do in, for example, English, namely within the VP domain. In contrast, however, subjunctive marking appears in the same hierarchical position as tense marking in English (i.e., in INFL). Thus, even if Blackfoot is only a superficially tenseless language, its subjunctive is, nevertheless, not straightforwardly amenable to analyzing it as a defective tense. Still, Blackfoot subjunctive has a distribution that is not unusual for a subjunctive (see section 2.2). In particular, the clause-type labeled subjunctive is restricted to future- and present- oriented conditionals. Counterfactual conditionals are realized by means of a dedicated clause-type; the unreal and negative clauses do not trigger subjunctive marking. 8 Given that conditional clauses may be analyzed as instantiating Cvaluation, we may conclude that, just as in Halkomelem, Blackfoot subjunctive marking may be analyzed as manifesting a [– coin] value derived by C-valuation. The reason it is not triggered by negation, unlike in Halkomelem, has to do with the fact that the syntax of negation differs (see Déchaine and Wiltschko 2003). And the reason it cannot occur in counterfactual conditionals is that there is a more specific clausetype available that serves this function, namely the unreal. According to Frantz (1991), the unreal is a special type of independent clause (see

The Essence of a Category

247

also Déchaine and Wiltschko, forthcoming). I propose that the unreal may be analyzed as a clause-type modifier. Thus, in Blackfoot, both the subjunctive and the unreal derive clause-types that contrast with the independent assertive type, albeit in different ways. Note that a traditional classification of clause-types based on the division between indicative and subjunctive mood does not capture the Blackfoot pattern. While the independent order may be characterized as a form of indicative, the fact that the unreal is a special form of the independent appears contradictory: according to substance-based criteria, the unreal behaves like an irrealis (or subjunctive), but according to Blackfoot internal morphosyntactic criteria, it behaves like an indicative. If so, the classical distinction between indicative and subjunctive mood (or between realis and irrealis for that matter) will not lead to an empirically adequate classification of Blackfoot clause-types. Any current theory that bases its categories on this distinction is equally bound to fail.9

7

A Common Semantics for Subjunctives?

One goal of this chapter is to come to terms with the fact that the subjunctive is a heterogeneous category. It cannot be clearly delineated according to either form or meaning. This fact contrasts with the argument by Giannakidou (this volume) that all subjunctives create a nonveridical space, either at the level of at-issue meaning (epistemic subjunctives) or at the expressive level (emotive subjunctives) (see also Ambar, this volume). However, while it may be possible to characterize many instances of subjunctives in this way (unsurprisingly, since it is the clause-type that contrasts with indicative assertions), it is neither a necessary nor a sufficient condition for the use of subjunctives. For example, Matthewson (2010) observes that in St’át’imcets Salish subjunctive marking is sometimes syntactically obligatory but semantically redundant. A case in point is given in (55), where the subjunctive is obligatory in the context of the complementizer i (‘when’). (55)

i = kél7 = at tsicw, when.pst = fi rst = 1pl.sbjv get.there áts’x- en-em i = cw7ít = a tsitcw see-dir-1pl.erg det = many = exis house ‘When we fi rst got there, we saw lots of houses.’ (Matthewson 2005, p. 74)

chapter six

248

This example establishes that nonveridicality is not a necessary condition for the subjunctive to be used in St’át’imcets. In a similar vein, the Upper Austrian data discussed in section  5.3 (examples (44) and (45)) pose some problems for Giannakidou’s unified analysis. Consider the example in (56). (56)

Context: Strang (Martina’s husband) is not at home. Sonja and Martina are at Martina and Strang’s house talking about whether Strang could help out with some drawings. Martina suddenly hears Strang’s footsteps and says to Sonja: Du, da Strang kam-at you, det Strang come.sbjv-sbjv . . . do kunnt-st = n glei . . . prt could.sbjv.2sg = him prt

eh grod . . . prt just.now . . . frogn. ask

‘You know, he is coming just now . . . so you could ask him.’ From the perspective of the speaker, there is direct evidence for the truth of the proposition and so there is no subjective nonveridicality at play. What licenses the subjunctive in this case is the fact that the addressee does not know that the proposition holds in the real world (Wiltschko 2014). This is evidenced by the fact that the subjunctive is incompatible with the discourse particle jo, which, like Standard German ja, indicates that the proposition is taken to be uncontroversial (by the speaker and the addressee) at the time of the utterance (Lindner 1991, p. 173). (57)

a. Ea komt jo grod. he come prt just.now ‘He is coming just now (as you know).’ b. *Ea kam-at jo grod. he come.sbjv-sbjv prt just.now ‘He is coming just now.’

(Wiltschko 2014, p. 182 (71))

To accommodate this type of data, Giannakidou’s (2009) analysis would have to be enriched with an addressee- oriented nonveridical space. I will leave the question as to whether this is possible—and if so, how—for future research (see Wiltschko 2014 for a more detailed analysis).

The Essence of a Category

249

In addition to not being a necessary condition for the use of subjunctive, nonveridicality is also not a sufficient condition. This can be seen on the basis of the German data involving the modal particle wohl cited in Giannakidou (2014). (58)

a. Context: I can’t see Hein. Er ist wohl auf See. He be.3.sg prt at.det sea ‘He may be at sea.’ b. Context: I know for sure: #Hein ist wohl auf See. Hein be.3sg prt at.det sea (Giannakidou 2014, p. 1027 (ex. 36))

Giannakidou suggests that Greek subjunctives behave like the German modal particle wohl, in that both serve to weaken the epistemic commitment of the speaker, thus creating a nonveridical space. Crucially, the sentences in (58) are not marked for subjunctive, which in turn establishes that nonveridicality is not a sufficient condition for subjunctive marking. In fact, in Upper Austrian, subjunctive marking is not compatible with woi (the counterpart of Standard German wohl), as shown in (59). (59)

Context: I can’t see Hein. *Ea war-at woi am Mea. he be.sbjv-sbjv prt at.det sea ‘He would be at sea.’

Thus, while Giannakidou makes an interesting attempt to provide a unified analysis of subjunctive (see also Ambar, this volume), it cannot delineate the subjunctive as a category by identifying the necessary and sufficient conditions for its use.

8

Toward a Typology of Subjunctive

This chapter establishes that the subjunctive is not a uniform category. It cannot be defined on the basis of sound or meaning. But if subjunctive is

chapter six

250

Cypriot Greek

INFL [u coin] m-valuation m-valuation

pred-valuation

not real [+coin]

[+coin]

C-valuation

[–coin]

[+coin]

[–coin]

[–coin]

Upper Austrian

Blackfoot "unreal"

Halkomelem Blackfoot

Figure 6.7

not a universal category, then mood cannot be a primitive category either, because the subjunctive is typically conceived of as a subcategory of mood. This observation highlights the need for a formal typology for categories that is not based on substantive content or traditional grammatical categories. The framework developed by R&W provides the basis for such a typology. It has allowed us to analyze the clause-types classified as subjunctives in four unrelated languages, their common properties as well as their differences. The result of this analysis is summarized in Figure 6.7. What all of these subjunctives have in common is that they contrast with the independent assertive clause-type instantiated by the [+coin] value derived via m-valuation. But this contrast may be derived in different ways. Nevertheless, the construction of the different types of subjunctive is constrained by the syntax of INFL. It manifests different valuation strategies including [– coin] derived via C-valuation (in Halkomelem and Blackfoot), or derived via m-valuation (in Upper Austrian); or it may be a clause-type modifier (the unreal in Blackfoot). And finally, the best-known and most diverse type of subjunctive, namely the one found in Cypriot Greek, can be analyzed as instantiating the unvalued coincidence feature.10 Given this diversity of the subjunctive category within and across languages, it comes as no surprise that it has been described as an “Irrgarten für Grammatiker” (a grammarian’s maze) (Jongeboer 1985). But this maze appears only if one assumes that what is labeled subjunctive in a

The Essence of a Category

251

given language must necessarily correspond to the category labeled subjunctive in the next language. More generally, it points to the conclusion that the categories we fi nd in the languages of the world are necessarily language-specific. This conclusion does not, however, mean that there cannot be a universal grammar (cf. Evans and Levinson 2009). Rather, language-specific categories are constructed. In particular, they are constructed from abstract universal categories (such as INFL) and language-specific LOs. Crucially, there is no universal association between the categories (such as INFL) and their substantive content, just as there is no universal association between categories and sounds. Both types of associations are necessarily language-specific. But if universal categories are dissociated from meaning, then we expect to fi nd categories that cannot be defi ned on the basis of their meaning, such as, for example, the subjunctive.

Notes 1. For a more extensive discussion of the coincidence feature see Ritter and Wiltschko 2014; Wiltschko 2014; and Welch 2012. 2. For ease of exposition, I will abstract away from the role of reference time associated with aspect. 3. Just as in complements of future- oriented verbs, the future orientation of the command derives from the future orientation of the valuing predicate (i.e., directive force). 4. The example in (31) is grammatical when the main focus of the utterance falls on the subject o Petr-os. However, in this case the sentence receives a modal interpretation, such that ‘Petros should (be the one to) swim now’, or an imperative reading in which Petros is indirectly instructed to swim. Therefore I suggest that in such cases it is the modal force that values the [ucoin] feature of INFL. 5. In isolation, the past time interpretation is not as readily available. It is facilitated with an introduction by you know and a locational particle in sentenceinitial topic position. Furthermore, the sentence seems to require a continuation, with a sentence describing what happened next. i) Wasst eh, do kumm I gestan ham und da hund Know prt loc come I yesterday home and det dog woa ned do was not there . . . ‘You know. I came home yesterday and the dog wasn’t there.’ I must leave the exact conditions of use for the past of the unmarked verb for another occasion. See Cowper (1998) for the dif ferent uses of the simple present in English.

252

chapter six

6. The distribution of these forms has not yet been explored. 7. For a discussion of the semantic properties associated with these independent subjunctives, see Wiltschko 2014. 8. See Louie (2015) for a detailed discussion of the semantic differences between the two types of conditionals. 9. For a more comprehensive analysis of Blackfoot clause-types within the R&W framework, see Wiltschko 2014. 10. The difference between C-valuation and pred-valuation mirrors the difference between operator and predicate selection (Stowell 1993; see Ambar, this volume, for discussion).

References Alexiadou, A., and E. Anagnostopoulou. 1999. Raising without infi nitives and the nature of agreement. Proceedings of the West Coast Conference on Formal Linguistics 18, 14–26. Bayer, J. 1984. COMP in Bavarian syntax. Linguistic Review 3: 209–274. Christodoulou,  C., and  M. Wiltschko  M. 2012. Function without content. Evidence from Greek Subjunctive na. In Towards a Biolinguistic Understanding of Grammar: Essays on Interfaces, ed. A-M di Sciullio, 117–140. Amsterdam: John Benjamins. Comrie,  B. 1989. Language Universals and Linguistic Typology. 2nd ed. Chicago: University of Chicago Press. Cowper, E. 1998. The simple present tense in English: A unified treatment. Studia Linguistica 52: 1–18. Déchaine, R-M, and  M. Wiltschko. 2003. Negation at the left periphery: Evidence from Algonquian and Salish. In Proceedings of the 15th Western Conference on Linguistics (WECOL 2001), 104–117. Fresno: California State University. ———. Forthcoming. Micro-variation in agreement, clause-typing, and fi niteness: Comparative evidence from Plains Cree and Blackfoot. Proceedings of the 42nd Algonquian Conference. Memorial University Newfoundland. Demirdache,  H., and  M. Uribe-Extebarria. 1997. The syntax of temporal relations: A uniform approach to tense and aspect. In Proceedings of the West Coast Conference on Formal Linguistics 16, ed.  E. Curtis,  J. Lyle, and  G. Webster, 145–159. Stanford, CA: CSLI Publications. Enç, M. 1987. Anchoring conditions for tense. Linguistic Inquiry 18: 633–657. Evans, N., and S. Levinson. 2009. The myth of language universals: Language diversity and its importance for cognitive science. Behavioral and Brain Sciences 32: 429–492. Fabricius Hansen, C., and K. Sæbø. 2004. In a meditative mood: The semantics of the German reportive subjunctive. Natural Language Semantics 12: 213–257. Frantz, D. 1991. Blackfoot Grammar. Toronto: University of Toronto Press.

The Essence of a Category

253

Galloway, B. 1993. A Grammar of Upriver Halkomelem. Berkeley: University of California Press. Giannakidou, A. 2009. The dependency of the subjunctive revisited: Temporal semantics and polarity. Lingua 120: 1883–1908. ———. 2014. The futurity of the present and the modality of the future: A commentary on Broekhuis and Verkuyl. Natural Language and Linguistic Theory 32: 1011–1032. Han, C-H. 2000. The Structure and Interpretation of Imperatives: Mood and Force in Universal Grammar. New York: Routledge. Jongeboer, H. A. 1985. Im Irrgarten der Modalität. Ein Kapitel aus der deutschen Grammatik. Groningen: Wolters-Noordhoff. Lin, J-W. 2006. Time in a language without tense: The case of Chinese. Journal of Semantics 23: 1–53. Lindner,  K. 1991. Wir sind ja doch alte Bekannte: The use of German ja and doch as modal particles. In Discourse Particles: Descriptive and Theoretical Investigations on the Logical, Syntactic and Pragmatic Properties of Discourse Particles in German, ed.  W. Abraham, 163–201. Amsterdam: John Benjamins. Louie, M. 2015. The temporal semantics of actions and circumstance in Blackfoot. PhD diss., University of British Columbia. Matthewson, L. 2005. When I was small—i wan kwikws: Grammatical Analysis of St’át’imcets Oral Narratives. Vancouver: University of British Columbia Press. ———. 2006. Temporal semantics in a supposedly tenseless language. Linguistics and Philosophy 29: 673–713. ———. 2010. Cross-linguistic variation in modality systems: The role of mood. Semantics and Pragmatics 3: 1–74. Picallo, C. 1984. The INFL node and the null subject parameter. Linguistic Inquiry 15: 75–102. ———. 1985. Opaque domains. PhD diss., City University of New York. Portner,  P.  1997. The semantics of mood, complementation and conversational force. Natural Language Semantics 5: 167–212. Quer,  J. 2006. Subjunctives. In The Blackwell Companion to Syntax, vol. 4, ed. M. Everaert and H. van Riemsdijk, 660–684. Oxford: Blackwell. Reis Silva,  A., and  L. Matthewson. 2008. An instantaneous present tense in Blackfoot. Semantics of Under- Represented Languages in the Americas 4. Amherst, MA: GLSA. Ritter,  E., and  M. Wiltschko. 2009. Varieties of INFL: TENSE, LOCATION, and PERSON. In Alternatives to Cartography, ed.  J. van Craenenbroeck, 153–201. Berlin: Mouton de Gruyter. ———. 2014. The composition of INFL: An exploration of tense, tenseless languages and tenseless constructions. Natural Language and Linguistic Theory. http:// link.springer.com /article/10.1007/s11049 - 014- 9248 - 6.

254

chapter six

Schlenker, P. 2005. The lazy Frenchman’s approach to the subjunctive. In Romance Languages and Linguistic Theory 2003, ed. T. Geerts I. van Ginneken, and H. Jacobs, 269–309. Amsterdam: John Benjamins. Siegel, L. 2009. Mood selection in Romance and Balkan. Lingua 119: 1859–1982. Thompson, J. 2012. Syntactic nominalization in Halkomelem Salish. PhD diss., University of British Columbia. Welch, N. 2012. The bearable lightness of being: The encoding of coincidence in Two- Copula Languages. PhD diss., University of Calgary. Wiltschko, M. 2002. The syntax of pronouns: Evidence from Halkomelem Salish. Natural Language and Linguistic Theory 20(1): 157–195. ———. 2014. The Universal Structure of Categories. Towards a Formal Typology. Cambridge: Cambridge University Press.

chapter seven

Imperatives as (Non-)modals Mark Jary and Mikhail Kissine

1

I

Introduction

mperatives constitute one of the greatest challenges for contemporary model-theoretic semantics. There are two related reasons for this. First, imperatives are mostly (although, as we will see, not exclusively) used for the performance of directive speech acts, such a commands, requests, orders, pleas, and the like—that is, for speech acts that are not intended to convey information. Second, and partly because of this prototypical illocutionary function, imperatives do not seem to be open to truthjudgments, which raises doubt that they are the kind of objects that can be modeled in truth- conditional terms. Two main trends have emerged in the recent literature that seek to resolve this tension: one is to analyze imperatives as semantic objects distinct from what may be found in declaratives (Portner 2007; Mastop 2005); the other consists in assigning imperatives a declarative-like semantics, combining this with pragmatic machinery that blocks truth-judgments. We have criticized the fi rst trend at length elsewhere (Jary and Kissine 2014, chaps 4 and 6). Here we focus on what is probably the strongest incarnation of the latter line of thought: Kaufmann’s (2012) theory of imperatives as (performative) modals. In the next section, we outline Kaufmann’s theory in detail. In section 2, we argue that, despite its sophistication, it fails to predict the unacceptability of judging an imperative as true or false. An important lesson that we draw from this discussion, in section 3, is that, in one way or another, a proper semantic treatment of imperatives has to do justice to their ‘potentiality’—that is, to the fact that they do not claim anything about the current situation, while

chapter seven

256

nevertheless being constrained by it. We briefly consider, and reject, the way Giannakidou (2009, 2012) proposes to capture this kind of intuition. Section 4 opens the positive part of the chapter. We provide a dynamic semantics for imperatives, which captures their inherent potentiality. We use the Discourse Representation Theory (DRT) framework (Kamp and Reyle 1993), as we believe it provides the most intuitive and transparent way to illustrate our claim, but the approach we advocate can be easily implemented in any dynamic framework. Before proceeding, we should point out that we do not wish to claim that potentiality is all that there is to imperatives: with a few qualifications, imperatives are also inherently second-person oriented and are restricted to agentive interpretation (see, e.g., Zanuttini 2008; Jary and Kissine 2014, chap.  2). However, these features are orthogonal to the topic of this paper—more specifically, whether imperatives are modal. The resistance of imperatives to truth-judgments and the way in which potentiality could be handled formally, by contrast, are crucial if one wishes to distinguish imperatives from modals.

2

Kaufmann: Imperatives as Performative Modals

Directive speech acts clearly lie on the performative side of Austin’s (1975) constative/performative divide, just like the ‘explicit’ performative sentence of the type in (1): (1)

I order/command/beg ‘etc.’ that p.

Jary (2007) argues that a good reason for thinking that explicit performatives like (1) are not assertions is that they do not require acceptance in order that the common ground (CG) be updated with their propositional content. In other words, while an assertion such as (2) can be denied and hence not update the common ground (if, for instance, the addressee (A, henceforth) says, ‘No you didn’t: you ordered Smith’), the utterance of (3) automatically updates the common ground with its propositional content. That is to say, as long as the appropriate ‘preparatory’ conditions are met (i.e., the speaker (S, henceforth) has the requisite authority, etc.), the utterance of (3) results in the proposition that S orders A to clean the latrines automatically becoming part of the common ground: no acceptance by S is necessary; nor is rejection possible.

Imperatives as (Non-)modals

(2)

I ordered you to clean the latrines.

(3)

I hereby order you to clean the latrines.

(4)

Clean the latrines!

257

This feature of explicit performatives has led to them being described as ‘self-verifying’, but the common-ground perspective on communication invites us to think of them as incontestable additions to the CG. It is notable that the same type of update occurs when the order is given by means other than an explicit performative: with the same assumptions of felicity (and allowing the context to determine the type of directive force intended), an utterance of (4) will also result in the proposition that S orders the hearer to clean the latrines automatically becoming part of the common ground. Given that any felicitous performative utterance results in an incontestable update of the CG (to the effect that that act has been performed), a form that leads to incontestable updates of the CG might be an efficient way of carrying out performative utterances. All that is needed is to specify a form whose utterance would have the required incontestable update results. Utterances of such a form would not be open to truth judgments because acceptance or rejection would not be an issue. Kaufmann (2012) seeks to pursue this line by arguing that imperatives are disguised modal declarative sentences. The core idea is thus that the imperative mood should be thought of as a necessity modal, very similar (though not identical) to deontic must. To see where this intuition comes from—and how it relates to the notion of an incontestable (and hence not apt for truth judgments) use of a declarative—consider the fact that, in most contexts, (5) would constitute a close alternative to (4). (5)

You must clean the latrines.

Moreover, when used this way, there are similarities between imperatives and deontic uses of must that do not hold between imperatives and other deontic modal sentences. The assertion that the hearer will not carry out the action described by the imperative or must sentence creates an infelicitous discourse that does not result with a should or ought to sentence: (6)

#

Clean the latrines, but you’re not going to.

chapter seven

258

(7)

#

(8)

You should/ought to clean the latrines, but you’re not going to.

You must clean the latrines, but you’re not going to.

Ninan (2005) uses this difference between the deontic must, one the one hand, and should and ought to, on the other, to argue that deontic must has only a performative use, which is to say that it can only be used to perform a directive speech act. In contrast, should and ought to can be used either performatively or descriptively. In other words, we could utter (9) either to advise the hearer to clean the latrines or simply to specify his obligations. This is why, according to Ninan, should and ought to have past deontic uses, as in (10), whereas must has none. One cannot direct someone to bring about a past state of affairs; hence, only an epistemic reading is possible for (11). (9)

You should/ought to clean the latrines.

(10)

You should/ought to have cleaned the latrines.

(11)

You must have cleaned the latrines.

Clearly, the infelicitous nature of the discourse we observe in (6) and (7) is a result of the directive force that the fi rst sentences of these utterances convey: other means of conveying directive force result in the same pattern. (12)

#

Would you mind cleaning the latrines, please? But you’re not going to.

The reason is obvious: it is irrational to direct someone to do something that you believe that they are not going to do. What makes the must case interesting, as Ninan points out, is that the pattern appears to hold even if the subject is third-person, as we see in (13). (13)

#

Sam must go to confession, but he’s not going to.

This suggests that the directive use of must might be primary, and that third-person cases such as (13) are interpreted as directives to the hearer to bring about the state of affairs described. However, there is also evi-

Imperatives as (Non-)modals

259

dence that deontic must can be used without a performative interpretation. Ninan points out that (14) cannot be plausibly analyzed as any kind of request directed at A or at the pope. Because, according to Ninan, the pragmatic unacceptability of (6) and (7), or of (13), is a hallmark of performativity, he goes on to say that if you fi nd this sentence acceptable, then you must be giving it a descriptive interpretation, and you would also fi nd (15) acceptable. (14)

The pope must change his stance on contraception.

(15)

The pope must change his stance on contraception, but he is not going to.

As we noted earlier, Kaufmann wants to argue that imperatives are disguised must sentences (albeit, as we will soon see, with some unique presuppositional characteristics). What the data in (14) suggest is that must is not inherently performative, but achieves its directive force by pragmatic means. In other words, semantically speaking, performative modals have a bonafide truth- conditional content, of the same kind as their descriptive cognate. Since Kaufmann takes imperatives and performative modals to have the same semantic meaning, and since she takes performative modals to have truth- conditional content, she concludes that imperatives have the same declarative semantics. Her challenge is to explain how they nevertheless result in an incontestable update of the CG.

2.1

Kaufmann’s Theory

Kaufmann uses the standard modality framework developed by Kratzer (1981, 1991). Kaufmann’s core idea is that imperative mood contributes a necessity operator whose modal base is constituted by the CG,1 and whose ordering source is determined contextually, and—importantly— determines the kind of speech act that the imperative constitutes: if the imperative is uttered as a command, then the ordering source singles out among the worlds of the common ground those that best conform to what S commands; if the imperative is uttered as a wish, then the ordering source singles out among the worlds of the common ground those that best conform to what S wants. For the time being, let us limit the discussion to imperatives used as commands. The CG will be viewed as

chapter seven

260

the set of possible worlds consistent with what is mutually accepted by the participants in the conversation (Stalnaker 1978, 2002). Informally, in Kaufmann’s view, the truth- conditions of a commanded imperative with the content p are: (I) Imperative(p), meant as a command, is true iff p is true in all the worlds of the CG that conform best to what S commands in the context of an utterance.

In order to distinguish imperative utterances from descriptive modal assertions, Kaufmann posits a number of presuppositions. The fi rst, and probably most important, is the presupposition of Epistemic Authority. For any imperative to be felicitous, argues Kaufmann, S must have authority over the ordering source (we use the formulations I, II, III, and IV as given in Kaufmann and Kaufmann 2012, pp. 152–154): (II) For every proposition p, S believes that p is necessary with re spect to the worlds singled out from the CG by the ordering source g iff p is necessary with re spect to the worlds singled out from the CG by g.

In other words, whenever S believes that a proposition is true in every possible world of the CG that has been singled out by the ordering source, this proposition is, in fact, true in every possible world of the CG singled out by that ordering source. By combining (I) with (II), we get (III): (III) S believes that imperative(p), meant as a command, is true iff p is true in every possible world of the CG that conforms best to what S commanded in the context of utterance.

At this point, Kaufmann (2012, pp. 152–154) invokes Gricean considerations of quality: all the participants in the conversation may reasonably surmise that S will not say things she believes to be false. In relation to (III), this means that if A believes that S believes that imperative(p) is true, eo ipso, A believes that p is true in every possible world of the common ground that conforms best with what S commands: (IV) If S utters imperative(p), meant as a command, and A believes that S is sincere and cooperative, then A believes that p is true in every world of the CG that conforms best to what S commanded in the context of utterance.

Imperatives as (Non-)modals

261

With (IV) Kaufmann’s theory reaches a stage where the utterance of an imperative results in an unchallengeable update of the common ground. Unless the imperative is infelicitous because S’s epistemic authority on the ordering source is violated, it is automatically taken as true.2 Finally, imperatives also come with the presupposition of Epistemic Uncertainty, according to Kaufmann (2012, pp.  155–157): any felicitous imperative(p) triggers the presupposition that S believes that p is neither ruled in nor ruled out by the common ground. At this point, then, Kaufmann’s theory predicts the following: if S utters imperative(p) and if S is sincere and her utterance felicitous, it becomes mutually accepted that p is true in all the worlds of the common ground singled out by some ordering source over which S has epistemic authority, and that S does not believe that p is true in every world of the common ground. We can now consider how the ordering source is chosen in Kaufmann’s account, and how she ensures that it corresponds to a variety of directive force. An initial problem is that, unless the ordering source is restricted, (16) may mean something like (17) (cf. Kaufmann 2012, p. 157): (16)

Get up!

(17)

Those alternatives that are most plausible according to what I take to be the usual course of events are such that you get up.

This is because, as Kaufmann points out, S has epistemic authority over what she considers to be the most plausible course of events; furthermore, even though S takes A’s getting up as very plausible, she does not (necessarily) believe that it is necessary, with respect to the common ground, that A will get up. So the interpretation in (16) respects the semantics of the imperative, as well as both presuppositions of Epistemic Authority over the ordering source and of Epistemic Uncertainty with respect to the content of the imperative. To circumvent such issues, Kaufamann (2012, pp. 159–161) proposes that imperatives come with the presupposition that the propositional content constitutes one solution to some contextually salient decision problem, and that the ordering source is mutually believed to constitute the relevant criteria for resolving the decision problem. As for good wishes, Kaufmann argues that if the content of the imperative is not a proposition whose truth is under A’s control, then it does not constitute a solution to any decision problem. In such a case, the presupposition is canceled, and the only restriction there is on the

chapter seven

262

ordering source is that it is relative to S’s preferences (2012, p. 160). For instance, (18) would mean, according to this view, that in every possible world of the CG that conforms best to what S desires, A enjoys her meal. (18)

Enjoy your meal.

So, despite adopting a declarative-like semantics, Kaufmann’s account seems to predict that imperatives are not used to assert: because they are unchallengeable additions to the CG, they do not put forth a proposition for acceptance. Moreover, her analysis predicts, by the same token, that imperatives are used performatively: whenever an imperative is interpreted as a command, S is taken to have performed the command at hand. In the next section, however, we will argue that this elegant theory is nevertheless flawed.

3

Is Absence of Truth- Judgement Really Explained?

In Kaufmann’s account, accepting an utterance of an imperative as felicitous entails accepting it as true. This is because imperatives come with a presupposition of Epistemic Authority on the ordering source: an imperative(p) is felicitous and S believes that imperative(p) is true if, and only if, p is true in all the worlds of the common ground that are singled out by the ordering source. As a result, accepting an imperative as felicitous—again, presuming that S believes the imperative is true— amounts to accepting that the imperative is true. This feature is exploited by Kaufmann (2012, pp. 163–165) to explain the fact that imperatives cannot be judged as true or as false: (19)

S: Get up! A: #That’s true/ #that’s false (you’re lying).

The truth-judgments in (19) are infelicitous, argues Kaufmann, because they target the truth-value of a proposition whose truth is made common ground by S’s epistemic authority on the imperative’s ordering source. She claims that (20) is odd for exactly the same reasons as (19) (2012, p. 166): because it is reasonable to assume that S knows whether she is hungry; then, if S is sincere, it is odd to put the truth-value of S’s utterance under discussion. Consequently, as long as her assertion is felicitous and she is sincere, it becomes common ground that she is hungry.

Imperatives as (Non-)modals

(20)

263

S: I’m hungry. A: #That’s (not) true. / #What you’re saying is (not) true.

However, Kaufmann (2012, pp.  167–168) does allow for the possibility that utterances such as I’m hungry can be rejected. She accepts that (21) is felicitous, but argues that, while A’s answer discounts S’s utterance as a lie, it leaves her epistemic privilege untouched. (21)

S: I’m hungry. A: That’s not true. You are lying.

In other words, A does not reject the presupposition that S knows whether she is hungry, but makes it clear that he believes that S says something she knows to be false. A number of objections can be made at this point. To begin with, not all rejections of I’m hungry leave S’s epistemic privilege untouched. For instance, in the following exchange one of us had with his son, what is targeted is the accuracy of the addressee’s evaluation of his own state of hunger: (22)

S: I’m hungry. A: No, you’re not. You had a full bowl of ice- cream fi fteen minutes ago.

More important, Kaufmann’s explanation of (21) raises the question of why is it impossible to discard a directive as a lie, leaving, in the same fashion, S’s epistemic privilege untouched. (23)

S: Close the door! A: #That’s (not) true. You are lying.

Assuming that the ordering source in (23) is what S commands, A’s disagreement would constitute an accusation to the effect that S does not believe that A closes the door in all worlds in the common ground that conform best to what she commands. Notably, A would not be denying S’s epistemic authority, but since S knows what she commands, her insincerity would mean that she knows that her utterance did not constitute a command to close the door. This is because to reject the sincerity of an imperative, on Kaufmann’s account, amounts to rejecting that p is true in all worlds picked out by the ordering source. If this is the case,

264

chapter seven

then it follows that S did not, in fact, command that p, for the truth of p in all worlds picked out by the ordering source is a necessary condition for the imperative to constitute a command that p. An accusation of insincerity, in the case of an utterance of an imperative, should therefore, in Kaufmann’s account, amount to a challenge to the felicity of the command. This problem arises because the theory relies on a maxim of quality: any norm of truthfulness must be open to exploitation. This puts Kaufmann in a dilemma: either she must drop the maxim of quality and leave A with no reason to accept that p is true in all worlds picked up by the ordering source; or she must keep the maxim and accept that an utterance of an imperative can be rejected as a lie, which fl ies in the face of empirical observation. There is another way to argue that Kaufmann’s theory wrongly predicts that one can challenge the felicity of a command by rejecting the imperative as false. In the following example given by Kaufmann (2012, p. 150) herself, A clearly rejects the felicity of S’s order; as a result, A also rejects that in every possible world of the CG that conforms best with what S commands, A goes home—for S did not command anything in the first place.3 (24)

S: Go home immediately! A: Hey wait a minute, you are in no position to give me orders.

But if so, it is unclear why a rebuttal like the one in (23) could not serve the same function. That is, A’s challenge in (23) should be equivalent to a rejection like the one in (24). To see why, consider how, in Kaufmann’ account, (24) could fail to be a felicitous directive speech act. Recall that, in her view, any imperative (that is not interpreted as a good wish) comes with the presupposition that its content is a solution to some decision problem and that the ordering source consists in relevant criteria to solve this problem. So, plausibly, rejecting an imperative as in (24) amounts to making clear that the ordering source cannot serve as a solution for any decision problem (what S may order is irrelevant to A’s action planning, as S cannot order him to do anything). But if in (23) S does not believe what she says, then—provided that she still has epistemic authority over what she commands—it follows that she does not believe that her utterance constituted a command that A go home immediately. Therefore rejecting her utterance as false should just mean that A points out that S knows that her command is infelicitous. That is, A’s challenge in (23) should be equivalent to a rejection like the one in (25).

Imperatives as (Non-)modals

(25)

265

S: Go home immediately. A: You are not in the position to give me orders, and you know it.

Another possibility that Kaufmann’s analysis allows is that S may be sincere, but mistaken about her epistemic authority. To see the consequences of such an error, it is useful to consider once again deontic must (as in (5)): (5)

You must clean the latrines.

Recall that Kaufmann claims that such uses are semantically identical to imperatives. The difference between modal sentences and imperatives is that, because the latter come with the presupposition of epistemic authority on the ordering source (and with a restriction on the ordering source) they cannot be used descriptively: if the imperative is felicitous, it is true. Modals like must, by contrast, may be used descriptively. When they are, as for instance in (26), S is not taken to have epistemic authority over the ordering source. (26)

You must clean the latrines. That’s what the sergeant said.

This is why, as we saw in (26), a descriptively used must can be combined with the assertion that content under its scope is false. (27)

You must clean the latrines. That’s what the sergeant said, but I know that you won’t.

(15)

The pope must change his stance on contraception, but he is not going to.

Kaufmann and Kaufmann (2012) suggest that while “responses like That’s (not) true! are generally infelicitous with performative uses,” when an utterance allows for both descriptive and performative interpretations “such a response can retroactively disambiguate and ‘lock in’ the descriptive interpretation.” This is what would happen in (28): (28)

A: You must empty the trash! B: That’s not true. The sergeant didn’t tell me anything.

Equivalent and attested examples are relatively easy to come by:

chapter seven

266

(29)

A: Jesus can save you . . . but you must believe it! All you have to do is accept him as your saviour and learn from his teachings . . . only that way will you be saved in the coming end . . . you must listen! . . . B: Oh no I must not. Stop pushing rubbish down other people’s reading space. (http://www/politicalforum.com /religion /219052-jesus - can-save-you-but-you-must-believe.html)

In (28) and (29), A’s rebuttal of the must sentence shows that S has no epistemic authority in the modal’s domain; that is, S does not know what is necessary given A’s obligations. But recall that, with imperatives, denying S’s epistemic authority over the ordering source amounts to treating the imperative as infelicitous. In the exchange in (19), the content of S’s imperative is that in every possible world of the CG that conforms best to what S commands, A gets up. Now, imagine that S is sincere but has no epistemic authority over the ordering source. This may be because S is not fully aware of the kind of directive speech acts that she is entitled to perform. For instance, it could be that, although she thinks that she is allowed to issue commands, she is actually in no position to do so. In such a case, it is possible that the imperative is false. There is no obvious reason why, in Kaufmann’s account, in (19) A should not be able to reject the directive as infelicitous. So, again, Kaufmann’s theory proves unable to explain the infelicity of responding That’s not true to an imperative.

4

Potentiality

A reaction that seems natural to us, at this point, is the following. From a semantic point of view, imperatives cannot be equated with modals. Modals say something true or false about the actual world: what is necessary or possible given a set of propositions. Imperatives cannot be used to make claims about the state of the world. Yet imperative use does appear to be restricted by the way the world is. It has been noted by several authors that utterances of the imperative are constrained by a need for the proposition expressed not to be ruled in or out by background information. In other words, the imperative appears to be limited to presenting potentialities (see, e.g., Davies 1986;

Imperatives as (Non-)modals

267

Wilson and Sperber 1988; Kissine 2013, chaps. 2 and 4; Jary and Kissine 2014, chap. 2). This is most evident in predetermined cases, such as these from Wilson and Sperber (1988): (30)

[A child sent to apologize to someone, as she approaches his door:] Please, be out.

(31)

[A mother, whose notoriously badly behaved child has been sent to apologize to someone, as the child arrives home:] Please, don’t have made things worse.

Although the state of affairs is decided, S is ignorant of it, and hence the  state of affairs described by the utterance is compatible with what she knows. The potentiality restriction means that counterfactual uses of the imperative are not possible. So one cannot, for example, use (32) as an alternative to (33). (32)

Don’t have done that!

(33)

If only you hadn’t done that!

To insist that the content of an imperative has to be potential does not mean that, at the utterance time, the truth-value of the content is not objectively settled. What it means is that the truth-value of the content has to remain unsettled with respect to a relevant set of information, a relevant contextual background. For instance, Dominicy and Franken (2002) observe that the following imperative may be uttered by an archaeologist who is about to unwrap a mummy, and for whose theory the date of birth of the mummified king is crucial. (34)

Please, be born before 4000 B.C.!

The reason (34) is felicitous is that the relevant background (the information the archaeologist and his audience have at the utterance time) is unknown; it is unknown whether the mummified king was born before 4000 B.C. (even though, it is, of course, objectively settled). This why uttering (34) would be infelicitous once the mummy has been properly dated. According to this view, the imperative presents as potential a state of affairs in which A performs some action but does not assert that this

chapter seven

268

state of affairs is potential. Directive force could then be argued to result from pragmatic considerations: the hearer seeks to identify the point of a non-assertoric utterance that presents him as the agent of an action, and a reasonable hypothesis is that the utterance is offered as a reason for him to take that action. As for the precise nature of the directive act, it would be determined by pragmatic considerations similar to those that, according to the view that directive force is encoded by the imperative, must be brought to distinguish, say, orders from advice. The challenge is to capture, in formal terms, potentiality (and in particular, the distinction between presenting a state of affairs as potential and asserting that a particular state of affairs is potential), while allowing a uniform semantic interpretation of declarative and imperative clauses. Imperatives are not the only form to exhibit potentiality. Subjunctive clauses, in many languages, have the property of denoting states of affairs that are neither ruled in nor ruled out by the CG (this is why subjunctive is often said to be a ‘non-assertoric’ mood (see, e.g., Bybee, Perkins, and Pagliuca 1994, pp.  212–237). For instance, in French, main- clause (present tense) subjunctive can only express contents whose truth-value is undetermined—that is, only potential contents. The utterance of (35) would be infelicitous if it is known that Jean leaves the city anyway or that he will not leave the city (see Schlenker 2005; Kissine 2013, pp. 47, 52). (35)

Que Jean quitte la ville! That Jean leave- SBJV.3SG.PR the city! (= Let Jean leave the city.)

Giannakidou (2009, 2012) claims that Greek subjunctive particle na, as well as the imperative mood, are nonveridical. Because what she means by nonveridical is synonymous with our use of ‘potential’, it is instructive to have brief look at her formalization. Informally, a propositional operator φ is said to be veridical iff φ(p) entails or presupposes that p is true according to some relevant information state M; φ is antiveridical iff φ(p) entails that p is false according to some relevant information state M. Accordingly, φ is said to be nonveridical iff φ(p) does not entail nor presuppose that p is true or false according to some relevant information state M (Giannakidou 2009, p. 1889). So far it thus seems that imperatives are indeed nonveridical. The problem, however, is to defi ne this notion formally without getting back to a modal account. And, in fact, we

Imperatives as (Non-)modals

269

part company with Giannakidou (2012) when she formalizes nonveridicality in Kratzerian terms. According to her account, a nonveridical operator, such as the subjunctive, should be seen as a universal epistemic modal with a nonveridical modal base. A modal base W is nonveridical relative to p iff it contains at least one ¬p world. Because φ(p) is said to be a necessity operator with a modal base that is nonveridical relative to p, an ordering source g is needed. As a result, φ(p) is true iff p is true in every possible world of the modal base that conforms best to g. Let f be an epistemic conversational background, which selects the CG as a modal base: (36)

∩f(w) = λw’.w’ is compatible with what is mutually accepted in w

Given an ordering source g, here is what the semantics of a nonveridical operator looks like (Giannakidou 2012): (37)

For any possible world w, a conversational background f and an ordering source g: ⟦φ⟧w, f, g = λp.∀w’ ∈ Bestg(w)(∩f(w)): p(w’) = 1, where Bestg(w)(CG) selects from CG the most ideal worlds given g(w)

Although we are very sympathetic to the idea of nonveridicality, we have several worries about applying such an account to imperatives. To begin with, it is unclear what the ordering source should be. One solution would be to treat it in the same way that Kaufmann does; but then the account would be open to the objections raised in the previous section. A related problem is that the semantics in (37) predicts that an imperative sentence can be assigned a truth-value. To these objections, Giannakidou could respond that in her view the imperative introduces an illocutionary operator (generated under Mood 0), which ensures that imperatives cannot be used assertorically, and hence that their truth-value cannot be challenged (cf. Giannakidou 2009). However, it is unclear how this solution—which seems ad hoc to us in relation to imperatives, at least—should be implemented. Furthermore, we have argued elsewhere that there are strong methodological and theoretical reasons for not equating imperative mood and illocutionary force (Kissine 2012; Kissine 2013, chaps. 2 and 5; Jary and Kissine 2014). Here, let us just say that we are looking for an account that extends to instances in which the imperative is used without directive force.

chapter seven

270

The fi rst class of such cases is constituted by good wishes, as in (18): (18)

Enjoy your meal.

To be sure, it may be argued that in English the use of imperatives in good wishes is not very productive and is close to idiomatic (Davies 1986, p. 51). However, there are languages in which casting good wishes in imperative is much more productive. For instance, Georgian has an optative mood especially for the expression of third-person wishes (with nonstative verbs); however, for second-person good wishes, the imperative is used instead (Boeder 2010, p. 626). A second class of nondirective uses of the imperative mood is constituted by imperative-like conditional (ILC) constructions such as the following: (38)

Come down with the flu and you’ll be in bed for weeks.

Some scholars have denied that the fi rst clause in constructions like (38) is imperative (Clark 1993; Han 2000, pp. 188–197; Russell 2007). However, a consensus is emerging that constructions like (38) are imperatives (Iatridou 2009; Jary and Kissine 2014, chap. 3). To mention only one very compelling reason, such constructions are found in languages with a clear morphological imperative, such as French and Spanish:4 (39)

Sachez être juste et can-IMP.2PL. be-INF fair and débarrasseront get.rid-IND.FUT.SIMPLE.3PL

ils they de of

se PR.REFL vous. you

(= Be fair and they’ll get rid of you.) (40)

Sé un ladron, y todo te saldrá be-IMP.2SG thief and all you turn- out.FUT.3SG bien. well ( = Be a thief and everything will turn out fine. ) (Grande Alija 1997, p. 59)

Imperatives as (Non-)modals

5

271

Imperatives in Direct Representation Theory

The challenge posed by the semantics of imperatives remains one of capturing the fact that they do not seem to be true or false in the actual circumstance of evaluation without lapsing into a modal theory, where the imperative would boil down to some kind of necessity operator. The intuitive idea is thus that, somehow, imperatives shift the perspective from what is actually the case to something that could be the case, but without introducing a claim about what is possible or necessary. Asher and Lascarides (2003a) propose thinking of imperatives as shifting the possible world parameter of interpretation, viz. that the imperative causes the discourse be evaluated at some other possible world. The problem with this approach is, fi rst, that it is doubtful that an imperative sentence causes the shift of the possible world parameter relative to the whole discourse, and, second, that how this shift should be constrained remains a moot point (Kaufmann 2012; Jary and Kissine 2014, chap. 4). We have decided to capitalize on another ‘mobile’5 parameter of truthconditional interpretation: value assignment to variables. In order to implement a nonmodal semantics of imperatives, we will rely on a minimal enrichment of the most standard version of DRT (Kamp and Reyle 1993). The first reason for this choice is that the intuition about the potentiality of imperative is best captured from a dynamic perspective; the second is that, in DRT, dynamic relations are precisely ensured by value-assignment functions. For simplicity’s sake, we adopt a highly naïve Davidsonian semantics of imperative verbs. We assume that each verb comes with at least an individual variable, introduced by the verbal subject, and an event variable. This conception is, of course, far too crude. However, our objective here is to formalize the potentiality of imperative clauses, and as will become obvious in a moment, nothing in our account hinges on the details of verbal semantics, so long as each verb introduces at least one eventuality variable. In this chapter we also ignore all the temporal and aspectual issues because, again, they are not important for our main objective. The basic terms of traditional versions of DRT are Discourse Representation Structures (DRSs), which consist of a universe of individual discourse referents and a set of conditions. Because we treat events as (unrepeatable) particulars, individual discourse referents may denote individuals and objects (1.1.1), as well as events (1.1.2). Furthermore, in order to be able to account for the impossibility of judging imperatives true or

chapter seven

272

false, we follow Geurts (1998) in introducing propositional discourse referents within the basic terms of DRS (1.2). 1.

DRS terms 1.1 R i = R E ∪ RO 1.1.1 RO = {x, y, z, . . . x’, x’’ . . . } 1.1.2 R E = {e, e’ . . . } 1.2 Rp = {p, q, . . . p’, p’’ . . . }

The major innovation we bring to the syntax of DRSs is the inclusion, in 2.3, of !K within the set of possible conditions, the exclamation mark standing for the imperative, and in fact, any potential, mood. The defi nition of the truth-predicate in 2.7 is needed in order to account for truthjudgments (and the lack thereof). 2.

Syntax of DRSs 2.1 A DRS K is a pair of a universe of discourse referents UK ⊆ R i∪Rp and of a set of conditions Con K; 2.2 If α1 , . . . α n ∈ R i, and P is a n-place predicate, then P(α1 , . . . αn) is a condition; 2.3 If K and K’ are DRSs, then ¬K, !K, K⇒K’, K∨K’ and N(K) are conditions; 2.4 If α ∈ R i and β = R i, then α = β is a condition; 2.5 If K and K’ are DRSs, then K(∀)K’ is a condition; 2.6 If α ∈ Rp and K is a DRS, then α = K is a condition. 2.7 If α ∈ Rp, then True(α) and N(α) are conditions.

The explicit objective of DRT is to model the incremental construction of discourse. A DRS may thus be thought of as an information slate, akin to the CG, updated with new information. For this reason, an essential component of the theory is its capacity to merge DRSs. 3.

Merging of DRSs K⊕K’ = < UK ∪ UK’, Con K ∪ Con K’>

4.

Semantics of DRSs

The model described in 4.1 is standard, with events conceived as (unrepeatable) particulars (4.1.1).

Imperatives as (Non-)modals

4.1

273

Model M = , where 4.1.1 D is the set of particulars, viz. individuals, objects, and events; 4.1.2 I is the interpretation function that maps pairs of possible worlds and predicates onto n-tuples of particulars; 4.1.3 W is the set of possible worlds; 4.1.4 R is the accessibility relation on possible worlds, such that R(W) ∈ Pow(W)

Essential to the semantic interpretation of DRSs is the embedding function, which does the job of value assignments in traditional static models. In 4.2 and 4.3, we follow Geurts (1998) in order to allow straightforward interpretations of propositional terms. 4.2 4.3

4.4 4.5

4.6

A 0- order embedding function f is a partial function from individual referent markers onto particulars of D; an n- order embedding function is the union between a 0- order embedding function and a partial function that maps propositional referent markers onto pairs , where w ∈ W and f is an n–1- order embedding function; an embedding function g extends f with respect to K, f [K]g iff Dom(g) = Dom( f ) ∪ UK and g ⊇ f. Given an embedding function f, a model M, a possible world w, and a DRS K, f verifies K, M⊨ f, w K, iff f verifies all the conditions of Con K relative to w; Given a condition γ ∈ Con K and Γ ⊆ F0 ∪ Fn, where F0 is the set of 0- order embedding functions and Fn is the set of n- order embedding functions, then, for any f ∈ Γ: 4.6.1 if γ = P(α1 , . . . αn), such that α1 , . . . αn ∈ R i, then, M⊨ f, w γ iff ∈ I(P)(w); 4.6.2 if γ = (α = β), and α , β ∈ R i, then, M⊨ f, wγ iff f(α) = f(β); 4.6.3 if γ = (α = β), and α ∈ Rp and β = K’, then M⊨ f, wγ iff, for every w’ and every g, ∈ f(α) iff M⊨ g, w’K’; 4.6.4 if γ = ¬K, then M⊨ f, wγ iff ∄g such that f[K]g, g ∈ Γ and M⊨ g, wK; 4.6.5 if γ = !K, then M⊨ f, wγ iff • there is a g ⊃ f, such that f [K]g, g ∈ Γ and M⊨ g, wK; • there is a discourse referent e ∈ UK , such that e ∈ R E; • and for all h ⊇ f, such that h ∈ Γ and h ⊉ g, h(e) is undefi ned;

chapter seven

274

if γ = K ⇒ K’, then M⊨ f, wγ iff for all g such that g ∈ Γ, f [K]g, and M⊨ g, wK, there is an h, such that h ∈ Γ and g[K’]h and M⊨ h, wK’; 4.6.7 if γ = K∨K’, then M⊨ f, wγ iff there is a g such that g ∈ Γ and f [K]g, and M⊨ g, wK or M⊨ g, wK’; 4.6.8 if γ = K(∀α)K’ and α ∈ R i, iff for all g, such that g ∈ Γ and f [K]g, and, for all d ∈ D, g(α) = d and M⊨ g, wK, for all h, such that h ∈ Γ and g[K’]h, M⊨ h, wK’; 4.6.9 if γ = N(α), α ∈ Rp then M⊨ f, wγ iff for all g⊇f, such that g ∈ Γ, and every w’ ∈ R(W), ∈ f(α) 4.6.10 if γ = True(α), and α ∈ Rp, then M⊨ f, wγ iff for every g ⊇ f, and every h, such that ∈ f(α), and g, h ∈ Γ, g ⊇ h. A DRS K is true with respect to a possible world w and an embedding function f ∈ Γ ⊆ F0 ∪ Fn iff for every g ⊇ f, such that g ∈ Γ, M⊨g, wK. 4.6.6

4.7

First, a DRS K under the scope of the imperative operator may be verified by f only if f is strictly extendable relative to K. This means that the content of the imperative can neither already be part of the main DRS nor be ruled out by it. Note, however, that 4.6.5 does not say that the content of an imperative sentence cannot be true or false. The idea, rather, is that an imperative sentence cannot be used to update the CG represented by the main DRS, because its content cannot be true under the same value assignments as the rest of the discourse (unless any further discourse update presupposes the truth of this content). Take (41a), as an illustration. (41)

a. There is a red fi le in the closet. Take it. b. [x, y, z: red-fi le(x), closet(y), addressee(z) ![e: takes(z, x, e)]]

Informally, what the DRS in (41b) shows is that any further assertion should presuppose that there is a red fi le in the closet, but not that the addressee takes it. By defi nition, once the function that verifies the fi rst, declarative sentence has been extended to another function that verifies the imperative sentence, the truth of all subsequent nonimperative clauses will presuppose the truth of the imperative one. For this reason, the event variable introduced by the imperative verb remains under the scope of the imperative (!). In this way, the discourse may be further updated with declaratives, without requiring that the imperative be true by the same token. This is exactly what happens in (42).

Imperatives as (Non-)modals

(42)

275

a. There is a red fi le in the closet. Take it. The map is in there. b. [x, y, z, x’: red-fi le(x), closet(y), addressee(z), map(x’), contained-in(x’, x), ![e: takes(z, x, e)]]

The information that the map is in the red fi le can be true or false independently of whether A takes it, because every assignment function that extends from the fi rst (declarative) to the second (imperative) sentence cannot be extended to subsequent declaratives. However, an embedding function may extend from the fi rst to the third sentence, both being declaratives. Accordingly, both declaratives can be true (or false), while the fact that A takes the red fi le remains unsettled. Compare now (42) with (43). Here, under the most natural interpretation, the third, declarative sentence can be verified only if the imperative is too. This is what is captured by (43b). (43)

a. There is a red fi le in the closet. Take it. You’ll fi nd the map in there. b. [x, y, z, x’: red-fi le(x), closet(y), addressee(z), map(x’), ![e, e’: takes(z, x, e), fi nd-in(z, x’, x, e’)]]

Note that the discourse referent introduced by the map is still accommodated at the uppermost level. This new discourse referent cannot be bound, and as argued, for instance, by Geurts (1999, p.  57), pragmatic principles favor accommodation at the highest level. That is, the assignment of values to the variables introduced by the fi rst sentence can be extended to that introduced by the map without necessarily assigning value to the event variable introduced by the VP in the third sentence.6 Accommodating the latter within the main DRS would imply that both declarative sentences could be true without the imperative being verified. Although not impossible, such an interpretation is hardly salient.7 Contrast now the semantics of imperative clauses with that of negation given in 4.6.4. According to this defi nition, whenever a DRS contains a negated DRS as a condition, no embedding function can verify the main DRS and the negated DRS at the same time. This captures the idea that the material under the scope of negation is ruled out of the CG. By contrast, when a DRS contains an imperative as a condition, no embedding function can extend both to this imperative and to other declaratives; this, however, does not mean that no embedding function can verify both the main DRS and the imperative. Compare (41) with (44).

chapter seven

276

(44)

a. I opened the closet. There is no fi le in there. b. [y, z, e: closet(y), speaker(z), opened(z, y, e) ¬[x: fi le(x), contained-in(x, y)]]

In (44b), the negated DRS cannot be verified at the same time as the conditions in the main DRS without getting a contradiction. In (41b), by contrast, the DRS embedded under ! can be verified, but not by an embedding function that could verify other conditions in the main DRS. It is important to realize that the merging of DRSs is taken here as a representation of the state of the common ground. A DRS that stands under the scope of ! represents a piece of information that cannot be integrated within or rejected from the common ground. Now, the common ground is inherently dynamic, and a proposition that is potential relative to the common ground at some moment may become nonpotential— because it has been accepted or challenged—at another. It is a crucial test for a formal theory of imperatives to predict such phenomena. Take (45a): (45)

a. S: Take the garbage bin out. A: I already did. b. [x, y, z, e’: S(x), A(y), garbage-bin(z), take- out(x, y, e’) ! [e: take- out(e)]]

A’s response challenges the potentiality of S’s imperative; as his response introduces an event discourse referent, which cannot be linked with that under the scope of the imperative (see (45b)). In other words, either A’s assertion has to be rejected or the content of S’s utterance should be integrated within the CG. If S follows up by repeating the imperative Take the garbage bin out, it would be assumed that she intends to reject A’s assertion as false. A different line of explanation is required for (46): (46)

S: Take the garbage bin out. A: OK, I will.

In contrast to (45), here it seems that A’s response does not have the effect of rendering the content of the imperative nonpotential. Rather, it seems that the effect of A’s answer is to make it mutually accepted that, provided some normalcy conditions hold, the content of the imperative sentence will be made true. At the utterance time, however, it cannot be accepted as true, or rejected as false, that A took the garbage out. Never-

Imperatives as (Non-)modals

277

theless, A’s answer does not seem to denote a potentiality, for it does not exhibit the same resistance to truth-judgments as the imperative: (47)

S: Take the garbage bin out. A: OK, I will. S’: That’s true, he will do it.

There is debate as to whether will should be considered a modal, but all parties assume that contents of assertions about the future are located under the scope of a necessity modal (e.g., Copley 2009; Del Prete 2014; Kissine 2008; 2014; Klecha 2014). It makes sense, therefore, to model A’s answer in (46) as involving a modal restricted to possible worlds where everything goes as expected; see (48a). Because modals operate on propositions, the propositional variable under the scope of will has to be assigned a value. The most likely candidate is the content of the imperative; this pragmatic accommodation is modeled in (48b): (48)

a. [x, y, p: addressee(x), garbage-bin(y), ![e: takes- out(x, y, e)], will(p)] b. [x, y, p: addressee(x), garbage-bin(y), ![e: takes- out(x, y, e)], will(p), p = [e: takes- out(x, y, e)] ]

What (48b) says is that the proposition under the scope of will is true under the same embedding functions and at the same possible worlds as the content of the imperative sentence. Assuming that will is an instance of N, the defi nition in 4.6.9 entails that will(p) will be verified by f at w iff for every possible world w’ in the domain of will, and every g ⊇ f ∈ Γ, the DRS [e: open(x, y, e)] is verified by g at w’. Accordingly, what is targeted by the truth-judgment in (47) is not the content under the scope of will, but rather the truth of will(p) at w. This is consistent with the general agreement that future- oriented propositions do not have a truth-value at utterance time (see, e.g., Thomason 1984; Belnap, Perloff, and Xu 2001; MacFarlane 2008; Stojanovic 2014). Assuming that can is the dual of N, a similar analysis applies to (49)a. (49)

a. A: Open the door. B: I can do it. b. [x, y, p: addressee(x), door(y), ![e: open(x, y, e)], CAN(p)]] c. [x, y, p: addressee(x), door(y), ![e: open(x, y, e)], CAN(p), p = [e: open(x, y, e)]

278

chapter seven

Another potentially challenging case consists in sequences involving quantification, such as (50a). (50)

a. Every student will pick an apple. [Addressing one of the students:] Pick an apple. b. [x’, y’: [x: student(x)](∀x)[e, y: apple(y), pick(x, y, e)], addressee(x’), apple(y’), ![e’: pick(x’, y’, e’]]

The defi nition in 4.6.8 only requires that whenever an assignment function that maps x on any individual that satisfies the predicate student is extended by another assignment function h relative to the DRS [e, y: apple(y), pick(x, y, e)], h verifies this DRS. This is compatible with h being a function that does not verify anything else in the main DRS, as required for embedding the imperative Pick an apple. As we have stressed throughout this chapter, an important test for the adequacy of the formal account is whether it correctly predicts that imperatives cannot be judged true or false. Let us start by modeling truthjudgments with declaratives. The DRS in (51b) models the dialogue in (51a), in which A judges S’s assertion to be true. (51)

a. S: The fi le is in the closet. A: That’s true. b. [p, y, x, e: closet(y), fi le(x), contained-in(x, y), True(p), p = [y, x, e: closet(y), fi le(x), contained-in(x, y)]]

Informally, what (51b) says is that A said that a proposition is true, and this proposition is identical to what S said. Look fi rst at 4.6.3. Embedding functions map propositional markers on pairs of embedding functions and possible worlds. So, for a propositional marker p to be equivalent to the DRS K given the embedding function f, it is required that for every that f maps p on, K be true at w under g (and vice-versa). That is, p is true under the same assignment of values and at the same possible worlds as K. Now consider the defi nition of the truth-predicate in 4.6.10. What it means, informally, is that the condition True(p) is verified under f at w, iff f maps p on a set of embedding function/possible worlds pairs, such that every assignment of values g, such that ∈ f(p), g can be extended by any extension of f (truth is monotonic). Consider now what would be the DRS resulting from A’s answer in (52a).

Imperatives as (Non-)modals

(52)

279

a. S: There is a red fi le in the closet. Take it. A: # That’s true. b. [x, y, z, x’, p: red-fi le(x), closet(y), addressee(z), ![e: takes (z, x, e)], True(p), p = [e: takes(z, x, e)] ]

As we just saw, the condition True(p) is verified by the embedding function f in a possible world w iff every assignment function g, such that ∈ f(p), can be extended by any extension of f. Now, what does it take for the condition p = [e: takes(z, x, e)] to be verified by that same embedding function f? According to 4.6.3, it must be the case that f maps p to those, and only those possible worlds/embedding function pairs under which the DRS [e: takes(z, e)] is verified. But remember that, according to our defi nition of imperatives, given in 4.6.5, if an embedding function verifies the DRS under the scope of the imperative in ![e: takes(z, x, e)], then it cannot be extended by any extension of f. Therefore, if p = [e: takes(z, x, e)] and ![e: takes(z, x, e)] are both verified by f, then f(p) necessarily contains a pair , where g cannot be extended by any extension of f (but only by those that verify [e: takes(z, x, e)]). This is incompatible with what is needed for True(p) to be verified by f. Therefore, our account correctly predicts that any attempt to judge an imperative as true would result in a noninterpretable DRS, such as (52b). We leave it as an exercise for the interested reader to show that the same prediction is correctly borne out for (53): (53)

a. S: There is a red fi le in the closet. Take it. A: # That’s not true.

To conclude our discussion of truth-judgments, observe that our analysis straightforwardly explains why judgments relative to the directive speech act performed by the imperative are felicitous. Because the performance of a speech act is mutually obvious to all the participants in the conversation, the fact that it has been performed automatically becomes common ground. In DRT, this can be represented by adding an event to the main DRS. Let us represent the fact that S ordered A to bring about the truth of p as a four-place relationship between S, A, p, and an event. It is this event that can stand as the target of comments on the speech act performed, as in example (54): (54)

a. S: There is a red file in the closet. Take it. A: That’s ridiculous. b. [x, y, z, x’, p, e’: red-file(x), closet(y), addressee(z), speaker(x’) ! [e: takes(z, x, e)], order(x’, z, p, e’), p = [e: takes(z, x, e)], ridiculous(e’)]

280

chapter seven

Before concluding, let us sketch how our account could deal with the imperative in imperative-like conditional constructions (ILCs), such as (55), whose syntactic subject, we suggest, could be modeled as a generic pronoun Gen. (55)

a. Catch a cold and you’ll be off work for weeks. b. [x: Gen(x), ![e, e’: catch-a-cold(x, e), off-work-for-weeks(x, e’)]]

In DRT, conjunction corresponds to the merger of two DRSs (cf. the defi nition 3). Accordingly, the most natural interpretation of a discourse string like (55a) would be (55b). Compare (55) with its conditional paraphrase in (56). (56)

a. If you catch a cold, you’ll be off work for weeks. b. [x: Gen(x), [e: catch-a-cold(x, e)] ⇒ [e’: off-work-for-weeks(x, e’)]]

In our account, what (55b) represents is that, given an embedding function f that verifies the main DRS, there is a g that extends f relative to ‘catch-a- cold(x, e)’ and also verifies ‘off-work-for-weeks(x, e’)’. According to the defi nition of conditionals in 4.6.6, any function that verifies and extends f relative to ‘catch-a- cold(x, e’)’ can also be extended to verify ‘off-work-for-weeks(x, e’)’. So, it follows that (55) entails (56), but that the reverse does not hold. That ILCs entail, but are not equivalent to, their conditional paraphrases is a welcome result, as not all conditionals can be rephrased as ILCs: (57)

a. If miss your train, there is a waiting room on platform 1. b. ? Miss the train, and there is a waiting room on platform 1. (from Clark 1993)

Unlike main- clause imperatives, nondirective ILCs may be targeted by truth-judgments, as in (58). (58)

a. S: Catch a cold, and you’ll be off work for weeks. A: That’s true.

It makes sense to suppose that what the truth-judgement targets is, in fact, the consequential relationship entailed by the ILCs. And our account predicts that such truth-judgments are felicitous. Consider (59):

Imperatives as (Non-)modals

(59)

281

a. S: If you catch a cold, you’ll be off work for weeks. A: That’s true. b. [x, p: Gen(x), [e: catch-a- cold(x, e)] ⇒ [e’:off-work-forweeks(x, e’)], True(p), p = [e: catch-a- cold(x, e)] ⇒ [e’: offwork-for-weeks(x, e’)]]

The condition p = [e: catch-a- cold(x, e)] ⇒ [e’:off-work-for-weeks(x, e’)] is verified by f iff for every g and w, such that ∈ f(p), g verifies [e: catch-a- cold(x, e)] ⇒ [e’:off-work-for-weeks(x, e’)] at w, viz. iff for all h ⊇ g, such that h verifies [e: catch-a- cold(x, e)], there is a j ⊇ h, such that j verifies [e’:off-work-for-weeks(x, e’)]. The condition True(p) is verified by f at w iff for ∈ f(p), g can be extended by any extension of f. One question that we leave untouched here is why exactly in ILCs the declarative clause is not incorporated within the main DRS. One reason is probably pragmatic, as the resulting interpretation of, for instance, (55) would be that it is true you will be off work and in bed for weeks, whether or not you catch a cold. Another possible reason is that ILCs appear to be single intonational units. Finally, examples like (55) could be instances of left-subordinating and (Culicover and Jackendoff 1997; cf. Russell 2007; Kaufmann 2012, pp. 237–241). What our account does explain are the sequential constraints on ILCs. Because the fi rst clause of (60a) is declarative, it will be incorporated within the main DRS. The subsequent imperative then cannot receive a consequential interpretation. (60)

6

a. ? You’ll be off work for weeks and catch a cold. b. [x, e: Gen(x), off-work-for-weeks(x, e), ![e’: catch-a-cold(x, e’)]]

Conclusion

Imperatives have much in common with modals. The temptation to reduce the former to the latter is all the more understandable because modals are easily implemented within truth-conditional, modeltheoretic semantic analyses, while such an account still remains a challenge for imperatives. One of the main reasons for this is the fact that imperatives are not amenable to truth-judgments. Accordingly, any modal theory of imperatives must explain this resistance. We have discussed in detail what is, to our mind, the most successful attempt of this kind, that of Kaufmann (2012). We showed that, despite its sophistication and elegance, Kaufmann’s

282

chapter seven

theory still fails to explain adequately the relationship between truthjudgments and imperatives. We argue that the reason imperatives cannot be judged true or false is that their content is inherently potential. In the formal rendering of this intuition, we showed that, unlike modals, imperatives do not amount to a claim about the truth of their contents relative to some possible world or another. Instead, from a dynamic point of view, they are modeled as requiring an assignment value function different from that required for the interpretation of propositions that belong to the common ground. This captures the distinction between claiming that p is potential and presenting p as a potentiality: the job of imperatives, we claim, is to do the latter.

Notes 1. This idea does not apply to Kaufmann’s treatment of advice (see Kaufmann 2012, 141–143), but we will leave these complications aside here. 2. Kaufmann (2012, p. 84) also endorses the Limit Assumption, according to which, if the modal base—here the common ground—is not empty, the set of possible worlds singled out by the ordering source will not be empty either. 3. This is true provided that Epistemic Uncertainty is preserved. 4. Kaufmann is immune to this objection, as she does not build directive force within the imperative semantics. However, her own account of ‘conditional imperatives’ in terms of modal subordination (Kaufmann 2012, p. 212) faces serious objections (von Fintel and Iatridou 2012; Jary and Kissine 2014, chap. 4). 5. Cf. Belnap et al. (2001, pp. 145–149). 6. For ease of exposition, we are treating will fi nd as a single verbal unit. 7. Most probably, the relationships between imperatives and declaratives in sequences like and could be captured in fi ner detail by incorporating our semantics of imperatives within a dynamic theory enriched with rhetorical relations (such as Elaboration and Narration), of the sort put forth by Asher and Lascarides (2003b).

References Asher, N., and A. Lascarides. 2003a. Imperatives in dialogue. In The Semantics and the Pragmatics of Dialogue for the New Millennium, ed.  P. Kuehnlein, H. Rieser, and H. Zeevat, 1–24. Amsterdam: John Benjamins. ———. 2003b. Logics of Conversation. Cambridge: Cambridge University Press. Austin,  J.  L. 1975. How to Do Things with Words. 2nd ed., ed.  J.  O. Urmson and M. Sbisà. Oxford: Oxford University Press. Belnap,  N.  D.,  M. Perloff, and  M. Xu. 2001. Facing the Future: Agents and Choices in Our Indeterminist World. Oxford: Oxford University Press.

Imperatives as (Non-)modals

283

Boeder, W. 2010. Mood in Modern Georgian. In Mood in the Languages of Europe, ed. B. Rothstein and R. Thieroff, 603–633. Amsterdam: John Benjamins. Bybee, J. L., R. D. Perkins, and W. Pagliuca. 1994. The Evolution of Grammar: Tense, Aspect, and Modality in the Languages of the World. Chicago: University of Chicago Press. Clark, B. 1993. Relevance and ‘Pseudo-Imperatives’. Linguistics and Philosophy 16: 79–121. Copley, B. 2009. The Semantics of the Future. New York: Routledge. Culicover, P. W., and R. Jackendoff. 1997. Semantic subordination despite syntactic coordination. Linguistic Inquiry 28: 195–217. Davies, E. 1986. The English Imperative. Beckenham, U.K.: Croom Helm. Del Prete,  F. 2014. The interpretation of indefi nites in future tense sentences: A novel argument for the modality of will? In Future Times, Future Tenses, ed P. De Brabanter, M. Kissine, and S. Sharifzadeh, 44–71. Oxford: Oxford University Press. Dominicy, M., and N. Franken. 2002. Speech acts and relevance theory. In Essays in Speech Act Theory, ed. D. Vanderveken and S. Kubo, 263–283. Amsterdam: John Benjamins. Geurts, B. 1998. Presuppositions and attitude contexts. Linguistics and Philosophy 21: 545–601. ———. 1999. Presuppositions and Pronouns. Amsterdam: Elsevier. Giannakidou, A. 2009. The dependency of the subjunctive revisited: Temporal semantics and polarity. Lingua 119: 1883–1908. ———. 2012. The Greek future: Epistemic modality and modal concord. In Selected Papers of the 10th International Conference on Greek Linguistics, ed. Z. Gavriilidou, A. Efthymiou, E. Thomadaki, and P. Kambakis-Vougiouklis, 48–61. Komotini, Greece: Democritus University of Thrace. Grande Alija,  F.  J. 1997. Los imperativos condicionales. Contextos 15(29–30): 49–65. Han, Ch-H. 2000. The Structure and Interpretation of Imperatives: Mood and Force in Universal Grammar. New York: Garland. Iatridou, S. 2009. De modo imperativo. Lecture Notes for a Course on the Syntax and Semantics of Imperatives at Ealing6, École Normale Supérieure, Paris. Jary, M. 2007. Are explicit performatives assertions? Linguistics and Philosophy 30: 207–34. Jary, M., and M. Kissine. 2014. Imperatives. Cambridge: Cambridge University Press. Kamp, H., and U. Reyle. 1993. From Discourse to Logic: Introduction to Modeltheoretic Semantics of Natural Language, Formal Logic and Discourse Repre sentation Theory. Dordrecht: Kluwer. Kaufmann, M. 2012. Interpreting Imperatives. Dordrecht: Springer.

284

chapter seven

Kaufmann, M., and S. Kaufmann. 2012. Epistemic particles and performativity. In Proceedings of Semantics and Linguistic Theory (SALT) 22, ed. A. Chereches, 208–225. Kissine,  M. 2008. Why will is not a modal. Natural Language Semantics 16: 129–155. ———. 2012. Sentences, utterances, and speech acts. In Cambridge Handbook of Pragmatics, ed. K. Allan and K. M. Jaszczolt, 169–90. Cambridge: Cambridge University Press. ———. 2013. From Utterances to Speech Acts. Cambridge: Cambridge University Press. ———. 2014. Will, scope and negation: A reply to Broekhuis and Verkuyl. Natural Language and Linguistic Theory 32: 1421–1431. Klecha,  P.  2014. Diagnosing modality in predictive expressions. Journal of Semantics 31: 443–455. Kratzer, A. 1981. The notional category of modality. In Words, Worlds, and Contexts, ed. H-J Eikmeyer and H. Rieser, 38–74. Berlin: Mouton de Gruyter. ———. 1991. Modality. In Semantics: An International Handbook of Contemporary Research, ed. A. von Stechow and D. Wunderlich, 639–650. Berlin: Walter de Gruyter. MacFarlane, J. 2008. Truth in the garden of forking paths. In Relative Truth, ed. M. Kölbel and M. García- Carpintero, 81–102. Oxford: Oxford University Press. Mastop, R. 2005. What can you do? Imperative mood in semantic theory. PhD diss., Amsterdam: Amsterdam University. Ninan, D. 2005. Two puzzles about deontic necessity. In New Work on Modality, ed.  J. Gajewski,  V. Hacquard,  B. Nickel, and  S. Yalcin, 149–178. MIT Working Papers in Linguistics 51. Cambridge, MA: Massachusetts Institute of Technology. Portner,  P.  2007. Imperatives and modals. Natural Language Semantics 4: 351–383. Russell, B. 2007. Imperatives in conditional conjunction. Natural Language Semantics 15: 131–166. Schlenker, P. 2005. The lazy Frenchman’s approach to the subjunctive. (Speculations on reference to worlds and semantic sefaults in the analysis of mood). In Proceedings of Going Romance XVII, ed. T. Geerts, I. van Ginneken, and H. Jacobs, 269–309. Amsterdam: John Benjamins. Stalnaker, R. C. 1978. Assertion. In Syntax and Semantics 9, ed. P. Cole, 315–322. New York: Academic Press. ———. 2002. Common ground. Linguistics and Philosophy 25: 701–721. Stojanovic, I. 2014. Talking about the future: Unsettled truth and assertion. In Future Times, Future Tenses, ed. P. De Brabanter, M. Kissine, and S. Sharifzadeh, 26–43. Oxford: Oxford University Press.

Imperatives as (Non-)modals

285

Thomason,  R.  H. 1984. Combinations of tense and modality. In Handbook of Philosophical logic. Vol. 2, ed. D. Gabbay and F. Guenthner, 135–165. Dordrecht: Reidel. Von Fintel, K., and S. Iatridou 2012. Imperative puzzles. Slides for a paper presented at the linguistic colloquim. Amherst. Wilson, D., and D. Sperber. 1988. Mood and the analysis of non-declarative sentences. In Human Agency, Language, Duty and Value: Philosophical Essays in Honour of J. O. Urmson, ed. J. Dancy, J. M. Moravcsik, and C. C. W. Taylor, 77–101. Stanford, CA: Stanford University Press. Zanuttini, R. 2008. Encoding the addressee in the syntax: Evidence from English imperative subjects. Natural Language and Linguistic Theory 26: 185–218.

chapter eight

Approaching the Morphosyntax and Semantics of Mood Ilse Zimmermann

1

The Issue

W

ithin a minimalist framework of sound-meaning correlation, in this chapter I articulate my hypotheses on the difference between sentence mood and verbal mood. Sentence mood relates sentence types to illocutionary types of root sentences, while verbal mood relates the propositional content of root and embedded clauses to worlds. I focus on declarative speech acts in the indicative and the subjunctive verbal moods in Russian. I am especially interested in the paradigmatic and syntagmatic distribution of the subjunctive marker and in its invariant, grammatically determined meaning. I show how verbal mood and tense are interrelatedand pay some attention to conditional modifiers. In the sound-meaning correlation of utterances, we are accustomed to reckoning with reference to eventualities to their participants x and to time spans t. In view of much of the work on event semantics, one could content oneself with extensional semantics. Nevertheless, we have to ask ourselves how intensional factors of meaning come into play (cf. Montague 1970a−c, 1973 [all reprinted in Thomason 1974]; Heim and Kratzer 1998; von Fintel and Heim 2011; von Stechow 2012; and many others). I would like to ask the question, “Where are the worlds?” (cf. Zimmermann 2010, 2013, 2015b, c). Clearly people’s worlds are connected with their mental states. But how is this fact reflected in the structure of linguistic utterances? By what means of expression do we refer to worlds—that is, to people’s mental states?

Approaching the Morphosyntax and Semantics of Mood

287

Within a minimalist framework of sound-meaning correlation, I concentrate on mood in its relation to worlds and time, and I argue for differentiating between sentence mood and verbal mood. The analysis here offers some improvements on my earlier work (Zimmermann 2009, 2010, 2013). The Russian system of tense and mood markers and the meaning associated with them deserve special attention, in view of their pecularities in contrast with other Slavic and non- Slavic languages. I start with declarative root sentences with the indicative and subjunctive moods, as shown in (1) and (2), respectively. (1)

V poslednie gody v Potsdame during last years in Potsdam vosstanavl-iva-l = sja gorodskoj dvorec. reerect-IPFV-PTCP.PST[M.SG] = REFL town castle ‘During the last few years in Potsdam the town castle was reerected.’

(2)

Pri GDR byα gorodskoj dvorec ne during GDR would α town castle not vosstanavl-iva-l = sja by−α . reerect-IPFV-PTCP.PST[M.SG] = REFL would−α ‘During the GDR the town castle would not have been reerected.’

While the preterite indicative in (1) is expressed by the inflectional affi x −l, the subjunctive in (2) is composed of this suffi x and the clitic particle by. I will show how the lexical entries and syntactic configurations determine the modal and temporal interpretation of these sentences.

2

Basic Assumptions

My considerations are built on a conception of minimalism and on the differentiation between grammatically determined semantic form (SF) and conceptual structure (CS) (Bierwisch  1983, 1985, 2007, 2011; Bierwisch and Lang 1987; Dölling 1997; Lang and Maienborn 2011). In the correlation between sound and meaning in linguistic expressions, the lexicon plays a central role. Every lexical entry contains a phonetic characterization (except for zero morphemes), a morphosyntactic

chapter eight

288

categorization, and the SF of the pertinent lexical item. I show that the association of certain grammatical formatives with their SF is delayed and conveyed only in the functional domains on the left periphery of the clause. Syntactic representations are of a purely syntactic nature. For clauses, I assume the following hierarchical domains.1 (3)

(ForceP) CP MoodP TP PolP vP*VP

As I show later in the chapter, functional domains can be fused. ForceP delivers various illocutionary types of sentences. It is absent in embedded clauses. CP characterizes the various sentence types (Brandt et al. 1992; Zimmermann 2009, 2010, 2013, 2015b, c). In generative grammar, it is not unusual to assume a functional projection MoodP (cf. Lohnstein 2000; Giannakidou 2009, 2011a, b, 2014, and this volume). I show which kind of meaning is contributed by this functional phrase. TP relates the topic (reference) time t to the utterance time t0 . In PolP, the decision between affi rmation and negation takes place. As for aspect (Klein 1994), I assume that it is delivered by the verb. 2 The main verb of clauses is the head of VP. Its arguments are merged in vP and/or VP. In general, I assume that the syntactic domains vP and VP serve to describe a situation with its participants and modifiers, while the functional domains ForceP, CP, MoodP, and TP relate vPs to discourse or to matrix clauses. 3 As regards morphology, I adhere to a conception according to which the lexicon brings in fully derived and inflected word forms (Wunderlich 1997). Thus the fi nite verb in (1) is represented in the lexicon with its word structure (a), morphosyntactic categorization (b), and semantics (c), as indicated in (4).4 (4)

a. [[[[voz[ stanavl’]] iva] l] sja] b. +V−N−pf+part+l-part+pret−imp−subj−fi n+max c. λyλtλe [[τ(e) ⊇ t ] ˄ [e INST [REERECT y x]]] with τ ∈ , ⊇, REERECT ∈ , INST ∈

The clitic sja marks the passive voice of the imperfective verb and blocks its external argument position. 5 Imperfectivity of the verb in (1) and (2) is expressed by the suffi x −iva. The suffi x −l in bold type expresses preterite tense and indicative—that is, nonimperative and nonsubjunctive mood.6

Approaching the Morphosyntax and Semantics of Mood

289

The word-structure feature +max characterizes the verb form in (4a) as capable of being merged in syntax. The SF of the verb consists of the argument structure (thematic grid) and the predicate-argument structure. The argument position λy is associated with selectional agreement requirements −neut, −fem, and −pl to the pertinent argument expression, gorodskoj dvorec ‘the town castle’. The person features αI, βII associated with the external argument position of a verb would characterize it as a fi nite verb form (Pitsch 2014, p. 129). Correspondingly, the verb form in (4) is categorized as −fi n. The l-participle does not inflect for person, like participles in general. As in Pitsch (2013, 2014), inflectional affi xes are considered to be formal reflexes of their meaning represented by functional (zero) heads.7 These heads c- command the verb form and select certain of its morphosyntactic features (cf. Sternefeld 2006). 8 While (5) represents the lexical entry of the suffi x – l in the word structure (4), the functional zero heads T 0 in (6) and Mood0 in (8) deliver the respective temporal and modal meaning components.9

3

Tense Morphology and Tense Meaning

Russian has a nonhomogeneous tense system. While the preterite is expressed by an ancient Slavic participle form, the l-participle, which agrees with the subject in gender and number and is based on the infi nitival stem, nonpreterite verb forms agree with the subject in person and number and are based on the present stem. Thus the suffi x −l in (1) and (2) has the following lexical representation: (5)

a. /-l/ b. +part+l-part+pret−imp(−subj) β−fi nαmax c. λP λx()α λtλe [P x t e] with P ∈

The suffi x −l selects the infi nitival verb stem and adds the features + part, +l-part, +pret, −imp, and optionally −subj and −fi n.10 The wordstructure feature αmax co-varies with the presence of masculine agreement features in the external argument position. In (1) and (2), the lparticiple agrees with the masculine (−neut−fem−pl) subject and counts as a +max verb form. The semantic representation of the l-participle in

chapter eight

290

(5c) amounts to an empty function, at the level of word structure. Its temporal semantic contribution is delivered by a zero formative in T0 .11 (6)

a. /⌀/ b. +T c. λP λtλe [([t < t0] ˄ ) α [P t e]] with P ∈

The functional zero head T0 selects a preterite complement and relates its topic time t to the utterance time t0. This relation characterizes the topic time as being before the utterance time, but only in the indicative (−imperative, −subj). In the subjunctive (−imperative, +subj), this relation is absent (see section 4.3).

4

The Analysis of Mood

In the analysis of mood, the left periphery of clauses with its functional domains TP, MoodP, CP, and ForceP is important.

4.1

Verbal Mood: The Indicative

Verbal mood relates propositions to mental models (Lohnstein 2000) by reference to worlds w and/or situations σ and binds the referential argument e of verbs. In situation semantics, Kratzer (1989, 2004, 2011) regards worlds as maximal situations, σ ≤ w, and propositions of type as sets of possible situations. In this sense, one could understand intensionalization of propositions of type as in (7a) or (7b) (see n. 23). (7)

a. ˄p = λw [ . . . w . . . ] b. ˄p = λσ [ . . . σ . . . ]

In the following I take (7a) for granted. In Russian, verbal mood is realized as indicative, subjunctive, or imperative. Their respective SFs are brought in by the functional zero head Mood 0. In (8), the indicative verb form is associated with its meaning. (8)

a. /⌀/ b. +Mood

Approaching the Morphosyntax and Semantics of Mood

291

c. λP (λw)∃e [[P t e ](w)] with w ∈ , P ∈ The functional zero head Mood 0 selects an indicative complement, blocks its topic-time argument position,12 binds the referential eventuality argument e, and optionally relates the pertinent proposition to worlds. The latter operation takes place in cases of intensionalization. Since in example (1) MoodP is the complement of the declarative operator, which selects ˆp (see section  4.2), MoodP must deliver a worldrelated complement.

4.2 Sentence Mood While Brandt et al. (1992) and Reis (1997, 1999) regard CP as the highest functional projection for root and embedded clauses, I have deviated from this conception since 2004 with my contribution “Satzmodus” (published in 2009). I accept Krifka’s (2001, 2004, 2013) assumption of illocutionary type operators for root clauses converting propositions of type into illocutionary types. This takes place in ForceP, as shown in (9). (9)

a. /⌀/ b. +Force c. λˆp [DECLAR/QUEST/EXCLAM/DIR ˆp] with DECLAR, QUEST, EXCLAM, DIR ∈

Example (9) contains lexical entries for +Force zero morphemes with their respective SF. The corresponding illocutionary-type operators are parameters that vary according to the social- cultural and linguistic context. They allow one to derive the commitments and modal mental states (see (11)–(13)) connected with the pertinent speech act. (For parameters in SF and their specification, see Dölling 1997.) As a rule, Force0 and C 0 are fused in root clauses. Example (10) is the lexical entry of the fused functional heads Force 0 and C 0 in declarative root clauses. Syntactically, it represents declarative root clauses as the unmarked sentence type by the feature −wh. Semantically, it selects a nonimperative (indicative or subjunctive) intensional propositional argument and combines it with the declarative illocutionary type.13 The choice between indicative and subjunctive is free (unselected) in declarative and interrogative root clauses.

chapter eight

292

(10)

a. /⌀/ b. +Force+C−wh c. λˆp [DECLAR ˆp]

For my considerations of root declarative clauses, the following meaning postulates (MPs) are important: (11)

MP 1: ∀ˆp [[DECLAR ˆp] → ∀w [[w ∈ ˆp ⋂ CG] → [w ∈ MVOLIT sp]]]

For all assertions, it follows that the speaker wants the pertinent proposition to be in the common ground.14 The same is true with questions. But assertions and questions differ with respect to the following MP: (12)

MP 2: ∀ˆp [[DECLAR ˆp] → ∀w [[w ∈ M EP sp] → [w ∈ ˆp]]]

Meaning postulate (2) relates declarative root sentences to the epistemic mental model of the speaker M EP sp. As will become clear, indicative declarative root clauses open a veridical modal space with respect to the mental state of the speaker (cf. Giannakidou 2014, this volume). I believe that these characterizations coincide with Truckenbrodt’s (2006a, b) semantic interpretation of the syntactic feature +C, −wh, but with one difference from my system of assumptions. Truckenbrodt does not assume illocutionary-type operators. For me, there is an important difference between a small universal set of mental models and their specification in the SF of language-specific lexical entries. Therefore, I assume that the following MP applies to veridical verbs like assume and believe, which describe particular modal mental states: (13)

MP 3: ∀ˆp∀x [[{assume’/believe’/ . . . } ˆp x] → ∀w [[w ∈ M EP x] → [w ∈ ˆp]]]

With these components of morphosyntactic expressions and empty functional categories, the semantic structure of (1), whose syntax is represented in (14),15 will be (15). (1)

V poslednie gody v Potsdame during last years in Potsdam

Approaching the Morphosyntax and Semantics of Mood

293

vosstanavl-iva-l = sja gorodskoj dvorec. reerect-IPFV-PTCP.PST[M.SG] = REFLtown castle ‘During the last few years in Potsdam the town castle was reerected.’ (14)

[ForceP/CP V poslednie godyi [ForceP/CP ⌀ [MoodP ti [MoodP ⌀ [TP ti [TP ⌀ [PolP ⌀ [VP v Potsdame [VP [V vosstanavlivalsja][DP ⌀ [NP gorodskoj dvorec]]]]]]]]]]

(15)

DECLAR λw∃e [[[t < t 0] ˄ [[∃!y [[TOWN CASTLE y] ˄ [[τ(e) ⊇ t] ˄ [e INST [REERECT y x]]]] ˄ [loc(e) ⊂ POTSDAM]] ˄ ∃z [[LAST YEARS z] ˄ [t ⊆ z]]]]w]16

4.3

Verbal Mood: The Subjunctive

Now let us turn to the subjunctive (cf. Quer 1998; Hacquard 2006; Giannakidou 2009, 2011a, b, 2014, and this volume; and von Fintel and Heim 2011) and to the morphosyntactic and semantic representation of sentence (2), with its syntactic structure (16).17 (2)

Pri GDR byα gorodskoj dvorec during GDR would α town castle vosstanavl-iva-l = sja reerect-IPFV-PTCP.PST[M.SG] = REFL

ne not by−α . would−α

‘During the GDR the town castle would not have been reerected.’ (16)

[ForceP/CP Pri GDR i [ForceP/CP ⌀ [MoodP ti [MoodP byα ⌀ [TP ⌀ [PolP [DP ⌀ [NP gorodskoj dvorec]]j [PolP ne [VP [V’ [V[V vosstanavlivalsja] by−α]tj ]]]]]]]]]

In Russian, the subjunctive is expressed by the preterite form of the verb or the infi nitive18 and the enclitic particle by. As shown in (16), the particle is adjacent to the verb or to the fi rst constituent of the clause.19 In complement or conditional clauses, it is cliticized to the respective complementizer čto ‘that’ or esli ‘if’. I assume that by is merged in V and can move to Mood.20 Its lexical entry is given in (17).

chapter eight

294

(17)

a. /by/, V__, C__, XP__ b. +subj c. λP [P]

The particle selects a maximal nonfi nite preterite or nonparticipial (infi nitival) verb form and adds the feature +subj to it.21 Its subjunctive semantics is delivered by the functional head Mood0 in (18). The meaning in (17c) is an empty function.22 (18)

a. /⌀/ b. +Mood c. λP λw∃e [[~[w Rconf wu]] : [[P t e]w]] with w, wu ∈, P ∈ , Rconf ∈

The semantic contribution of the subjunctive is comparable to the indicative and adds the restriction that the worlds w do not conform to the world of the modal subject u. 23 This characteristic restriction of the subjunctive corresponds with semantic properties of prospective predicate expressions like trebovat’ ‘require’ and dlja togo ‘in order to’, which embed clauses with by.24 The modal subject u will be specified as the speaker in the context of the declarative sentence mood. As the equivalences in (19) show, I regard the proposition λw [[~ [w Rconf wu]] : [w ∈ˆp]] as tantamount to λw~∀w’ [[[w’ ∈ M EP u] ˄ [w’≤ w]] → [w’ ∈ˆp]]—that is, it is similar to what Giannakidou (this volume) characterizes as nonveridicality, in contrast to λw∀w’ [[[w’ ∈ M EP u] ˄ [w’≤ w]] → [w’ ∈ˆp]] (cf. the meaning postulates in (12) and (13)). Moreover, with u = speaker, we are dealing here with a negated presupposition.25 (19)

a. ∀w∃ˆp [[[~ [w Rconf wu]] : [w ∈ˆp]] ↔ ~∀w’ [[[w’ ∈ M EP u] ˄ [w’≤ w]] → [w’ ∈ˆp]]] b. ∀w∃ˆp∀w’ [[[[w’ ∈ M EP u] ˄ [w’≤ w]] → [w’ ∈ˆp]] ↔ [[w ∈ M EP u] ˄ [w ∈ˆp]]] c. ∀w∃ˆp Genx [[[[PERS x] ˄ [w ∈ M EP x]] ˄ [w ∈ˆp]] ↔ [w ∈ CG]] 26

Thus, in contrast to Wiltschko (this volume) and Christodoulou and Wiltschko (2012), I assume that functional categories do have content

Approaching the Morphosyntax and Semantics of Mood

295

and that the semantic characterization of the subjunctive in (18c) is the SF pendant of the syntactic feature [−coin(cident)] in Infl 0 in the analysis of the subjunctive by the authors.27 Now, the preterite form of the selected verb with the subjunctive marker by deserves special attention. The preterite functional zero head T0 varies in its meaning contribution depending on the value of the feature αsubj (cf. (6c)). The temporal characterization is absent in the subjunctive. This phenomenon is called “fake preterite” and corresponds to the infinitive, lacking temporal specification.28 Thus the preterite verb form vosstanavlivalsja in its temporal meaning in (1) is characterized by (6c) as referring to a time span before the utterance time. This is not the case for this verb form in (2), where it is accompanied by the subjunctive marker by. In (2) the relation of the topic time t to the utterance time is unspecified in SF and resides in the knowledge of the speaker, including the linguistic context. If examples (1) and (2) are considered as a coherent text, the temporal interpretation of the topic time in (2) in CS is inherited from (1), where t is expressed as being before the utterance time. If (2) is considered independently of (1), t is existentially bound, leaving the relation to t0 unspecified in CS—that is, whether t is to be interpreted as being before t0 or if it is not.29 Thus, the SF of the inner MoodP in (16) will be (20). (20)

λw∃e [[~ [w Rconf wu]] : ∃!y [[[TOWN CASTLE y] ˄ ~[[τ(e) ⊇ t] ˄ [e INST [REERECT y x]]]]w]]

The semantic contribution of the presence of the subjunctive particle by is marked by bold type. It characterizes the worlds w to which the proposition expressed by the MoodP applies as not conforming with the world of the modal subject u. Observe that in (20) there are two different occurrences of negation. The fi rst one is brought in by Pol0 and negates the propositional kernel of the clause. The second one is delivered by the restriction connected with the subjunctive mood (cf. (18c)). It is relevant that the subjunctive meaning represented by the empty functional category Mood 0 takes scope over the so- called sentence negation in Pol0 (cf. (16)). In a sense, one could regard the zero heads T0 and Mood 0 together as a covert auxiliary verb that conveys discourse- oriented meaning components of the clause. I leave open the question whether T0 and Mood 0 are fused in syntax, as assumed by Pitsch (2014).

chapter eight

296

4.4

Conditional Modifiers

First, the conditional adverbial modifier pri GDR in example (2) will be integrated into the sentence structure. Conditional modifiers are merged and interpreted as adjuncts of MoodP (see the trace ti in (16)). The conditional meaning of the preposition pri is represented in (21).30 The semantic template (22), which relates DP meanings to worlds (cf. Schwarz 2011), applies to the proper name GDR with the resultant representation (23). (21)

a. /pri/ b. −V−N c. λpλqλw∀w’ [[[p w’] → [q w’]]w] with p, q ∈

(22)

λuλw∃z [[z ≤ w] ˄ [z = u]] with z, u ∈

(23)

λw∃z [[z ≤ w] ˄ [z = GDR]]

With the semantic components of Force 0 /C 0 in (10), of pri in (21), and of its complement in (23), the SF of sentence (2) is (24). (24)

DECLAR λw∀w’∃z [[[[z ≤ w’] ˄ [z = GDR]] → ∃e [[~ [w’Rconf wu]] : ∃!y [[[TOWN CASTLE y] ˄ ~[[τ(e) ⊇ t] ˄ [e INST [REERECT y x]]]]w’]]]w]

The implication in the scope of the declarative sentence–type operator can be true if both the antecedent and the consequent are not true in the speaker’s epistemic mental model. In the SF (24), the nonconformity of w’ and Wu is not represented for the antecedent. But since—according to the knowledge of the speaker—the GDR does not belong to the world wu in the given temporal context and, consequently, w’ does not conform to wu, the implication in (24) is true. In contrast to example (2) with the conditional PP pri GDR, the correponding conditional clause in (25) with the conjunction esli ‘if’ expresses the restriction that the GDR does not exist in the world of the speaker. 31 (25)

Esli by GDR suščestvova-l-a, gorodskoj if would GDR exist.IPFV-PTCP.PST-F.SG the.town

Approaching the Morphosyntax and Semantics of Mood

297

dvorec ne . vosstanavl-iva-l = sja by castle not reerect-IPFV-PTCP.PST[M.SG] = REFL would ‘If the GDR had existed, the town castle would not have been reerected.’ The meaning of the conjunction esli corresponds to the meaning of pri in (21c). Thus the SF of the conditional clause in (25) will be (26). (26)

λqλw∀w’∃e’ [[[~ [w’ Rconf wu]] : [[[τ(e’) ⊇ t] ˄ [e’INST [EXIST GDR]]]w’]] → [qw’]]w]

Examples (27)−(29) illustrate the possibility of variation in the expression of Russian conditional constructions (RG 1980, vol. 2, pp. 104ff). (27)

Esli by syn uči-l = sja , if would the.son learn.IPFV-PTCP.PST[M.SG] = REFL mat’ by ne ogorča-l-a = s’. the.mother would not worry.IPFV-PTCP.PST-F.SG = REFL ‘If the son would {learn/have learned}, the mother would not {worry/have worried}.’

(28)

Uči-l = sja by syn, mat’ by ne ogorčalas’. learn.IPFV-PTCP.PST[M.SG] = REFL . . .

(29)

Uč-i = s’ (by) syn, mat’ by ne ogorčalas’. learn.IPFV-IMP.2.SG = Refl . . .

In (28), the subjunctive verb seems to substitute for the conditional conjunction. 32 In the colloquial variant (29), the substituting entity is the second-person singular form of the imperative, 33 with or without the subsequent subjunctive particle by. There is a close semantic relationship between the subjunctive and the imperative that allows for their mutual substitution in many cases (see Zimmermann 2009, 2015c; Dvořák and Zimmermann 2007). The temporal unspecificity of the subjunctive and its characteristic restriction are valid for the imperative too. 34 While in (28) and (29) there is a conditional zero conjunction, to which the raised verb is adjoined, (31) and (32) syntactically are not sentences

chapter eight

298

with a conditional clause, in contrast to (30). Nevertheless, the referential parallelism of the subject in (31) and (32) and the anaphoric pronoun ėto ‘this’ in (30), as well as its relation to the proposition of the antecedent, deserve special attention. (30)

Esli if ėto this

by would menja me

Sergej menja poseti-l , Sergej me visit.PFV-PTCP.PST[M.SG] ob-radova-l- o by. PFV-please-PTCP.PST-N.SG would

‘If Sergej would {visit/have visited} me, it would {please/have pleased} me.’ (31)

Poseščenie Sergeja menja ob-radova-l- o by. a/the.visit of.Sergej me PFV-please-PTCP.PST-N.SG would ‘{A/The} visit from Sergej would {please/have pleased} me.’

(32)

Kniga menja ob-radova-l-a by. a/the.book me PFV-please-PTCP.PST-F.SG would ‘{A/The} book would {please/have pleased} me.’

The semantic interpretation of the subject phrases in (31) and (32) involves reference to possible worlds (situations) that must be related to the worlds (situations) the respective sentences refer to. Moreover, the far-reaching semantic synonymy of constructions like (30) and (31) with a nominalization must be taken into account. First, I propose the application of the template (22) and of a conditional template, (33), to the noun phrases in (31) and (32). The resultant semantic representation for example (32) is (34). (33)

λpλqλw∀w’[[[p w’] → [q w’]]w]

(34)

DECLAR λw∀w’∃x i [[[[BOOK x i] ˄ ∃z [[z ≤ w’] ˄ [z = x i]] → [[~ [w’ Rconf wu]] : ∃e [[τ(e) ⊆ t] ˄ [e INST [PLEASE sp x i]]]w’]]]w] with BOOK ∈ , PLEASE ∈ , α ∈ {e, t}

A tacit assumption in this analysis is the movement of the subject phrase of (32) to MoodP, and then to ForceP/CP (see (35)). The last occurrence of its trace ti is semantically interpreted as x i.

Approaching the Morphosyntax and Semantics of Mood

(35)

299

[F/CP [DP ⌀ kniga]i ⌀ [MoodP ti [MoodP ⌀ [TP ⌀ [PolP ⌀ [vP ti menja obradovala by]]]]]]

The syntactic derivation and semantic interpretation of (31) proceeds analogously. Thus the semantic representation of (31) will be (36). (36)

DECLAR λw∀w’∃x i [[x i INST [VISIT sp Sergej]] ˄ ∃z [[z ≤ w’] ˄ [z = x i]] → ∃e [[~[w’ Rconf wu]] : [[τ(e) ⊆ t] ˄ [e INST [PLEASE sp x i]]]w’]]w] with VISIT ∈

It must be mentioned that DPs in general and as such the nominalization in (31), poseščenie Sergeja ‘Sergej’s visit’, are not specified for tenseor for the reference to mental models. Therefore the antecedent of the implication in (34) and (36) lacks the modal specification ~[w’ Rconf wsp], which can only be added in CS. Thus the implication in the scope of the declarative operator is true and is in accordance with MP 2 in (12). (12)

MP 2: ∀ˆp [[DECLAR ˆp] → ∀w [[w ∈ M EP sp] → [w ∈ ˆp]]]

In (30), both the antecedent and the consequent have the modal specification ~[w’ Rconf wsp], which is expressed by the subjunctive, represented on the level of SF. Now, we must ask how the pronominal subject of (30) is related to the conditional clause, which is merged and interpreted as an adjunct of MoodP and moved to ForceP/CP. 35 As in Zimmermann (2015a) for German es ‘it’, I propose the SF (37) for the Russian anaphoric pronoun ėto ‘this’. (37)

λP∃x [[x = y] ˄ [P x]] with P∈ , α ∈ {e, t, . . . }

In CS, the parameter y in this generalized quantifier is co-indexed with a coreferent proposition in the conditional antecedent in (30). This is shown in (38). (38)

DECLAR λw∀w’∃e’ [[[~ [w’ Rconf wu]] : [[[τ(e’) ⊆ t’] ˄ [e’ INST [VISIT sp Sergej]]]i w’]] → ∃e [[~ [w’ Rconf wu]] : [[[τ(e) ⊆ t] ˄ [e INST [PLEASE sp x i]]]w’]]w]

chapter eight

300

4.5

The Flexible Temporal Interpretation of Subjunctive Clauses

Finally, I make some observations on the flexible temporal interpretation of subjunctive clauses. Without context, the examples (39) and (40) with the subjunctive do not have any specified temporal relation to the utterance time t0. I assume that the topic-time argument t, unbound in SF, is existentially quantified in CS or specified in dependence on the context, as in (2), on the basis of (1). The clauses in (39) and (40) can refer to the past or to the non-past. The fi rst has to do with a counterfactual interpretation of the subjunctive, and the second with the so- called conditional. 36 (39)

Boris kupi-l by mašinu. Boris buy.PFV-PTCP.PST[M.SG] would a/the.car ‘Boris would {buy/have bought} {a/the} car.’

(40)

Ja izvini-l-a = s’ by. I apologize.PFV-PTCP.PST-F.SG = Refl would ‘I would {apologize/have apologized}.’

In contrast to German, which marks counterfactivity with subjunctive pluperfect verb forms, Russian counterfactive interpretations reside exclusively in the context. (41)

A: Počemu Nina ne why Nina not (na-) α pisa-l-a mu? (PFV-) αwrite(.IPFV)−α -PTCP.PST.IND -F.SG him ‘{Why did Nina not write him/Why has Nina not written him}?’ B: Esli by if would ona by she would

Nina zna-l-a ego adres, Nina know.IPFV-PTCP.PST-F.SG his address emu (na) α pisa-l-a. him (PFV-) αwrite(.IPFV)−α -PTCP.PST.-F.SG

‘If Nina would have known his address, she would have written him.’

Approaching the Morphosyntax and Semantics of Mood

(42)

301

A: Počemu Nina ne (na-) α piš-et why Nina not (PFV) α−write(.IPFV)−α -PRS.IND. 3.SG emu? him ‘Why {does/will} Nina not write him?’ B: Esli by Nina zna-l-a ego adres, if would Nina know.IPFV-PTCT.PST-F.SG his address ona by emu (na-) α pisa-l-a. she would him (PFV-) αwrite(.IPFV)−α -PTCP.PST.-F.SG ‘If Nina knew his address, she would write him.’

From the questions, speaker B knows that Nina has not written/did not write him, in case (41), and that she does not/will not write him, in case (42).

5 Summing Up In this chapter I show how the morphosyntactic and semantic components of root and conditional clauses determine their semantic form (SF), which is enriched by meaning postulates and on the level of conceptual structure (CS). Root clauses are characterized by illocutionary-type operators in Force, fused with C. The reference of root and embedded clauses and of noun phrases to worlds (situations) is anchored in the functional categories Mood and D, respectively. Concerning verbal mood, I have concentrated on the question of what subjunctive is in Russian. I believe that the subjunctive always expresses the restriction that the pertinent world does not conform to the world of the respective modal subject u. It can be shown that this is valid for all occurrences of the subjunctive, in both root and embedded clauses. Furthermore, subjunctive clauses are temporally underspecified. Their topic time t is not related to the utterance time t0, in SF. The subjunctive shares these two semantic properties with the imperative. Therefore subjunctive verb forms and the imperative can occur in complementary distribution in many cases. Conditional modifiers, PPs or clauses, are considered to be adjuncts of MoodP. They describe circumstances that restrict the set of worlds

chapter eight

302

the respective matrix clause refers to. I proposed two templates that accommodate the subject phrase of certain nonembedded clauses to this function. In the sound-meaning correlation the differentiation between SF (as the grammatically determined meaning of the overt and covert morphosyntactic components of clauses) and CS (as a representation of nonlinguistic knowledge) proved essential. In particular, the treatment of the topic-time argument as a parameter in SF explains the flexibility of the temporal interpretation of Russian subjunctive clauses. The treatment of the functional categories T and Mood as zero categories with the meaning contribution of selected verb forms opens a window for interesting comparative analyses of languages and language types. Whereas the selected morphosyntactic formatives and their categorizations may vary considerably, the categorial and semantic contribution of the selecting functional categories tends to be universal.

Notes I would like to thank the organizers of the TRAIT workshop and its participants for the opportunity to present some ideas about the morphosyntax and semantics of time and mood. I profited from discussion of these topics at the ZAS in Berlin; at the Seminar für Slavische Philologie, University of Göttingen;and at the Institute of Slavic Languages and Literatures, University of Potsdam. For critical remarks, I am indebted to two anonymous reviewers, and especially to Anastasia Giannkidou and Hagen Pitsch. For consultation concerning the examples and their interpretation, I thank Anatoli Strigin. 1. Cf. Ambar’s (this volume, (39), (48), (59)) split categories of the left periphery: (i)

XP* EvaluativeP AssertiveP XP* FinP TP . . . (with XP* for dislocated DPs)

Ambar assumes that EvalP and AssertP are pragmatic categories that allow for semantic decomposition of modality and constrain the influence of pragmatic factors on the semantic interpretation of clauses. 2. Rus sian deverbal nominalizations like razrabatyvanie vs. razrabotka ‘elaboration’ are derived lexically and based on the corresponding verb stem plus aspectual markers and their semantics. Therefore, I depart from de Swart (this volume), who assumes the existence of aspect phrases below the tense phrase. 3. In Slavic grammatical tradition, the factors represented in these functional domains are characterized as components of “predikativnost” (‘predicativity’). Cf. Pitsch (2014, p. 177ff.).

Approaching the Morphosyntax and Semantics of Mood

303

4. For reasons of explicitness, redundant morphosyntactic features are not omitted. Elementary semantic types are for various sorts of individuals, for propositions, and for worlds. Argument positions can be associated with morphosyntactic conditions on the pertinent argument expression. 5. Passivization as an operation on the argument structure of the verb takes place in the lexicon. Other functions of the formative sja are left out of consideration here. Cf. Fehrmann, Junghanns, and Lenertová (2010). 6. Morphosyntactic features are chosen in correspondence with their phonetic realization. In Russian, the subjunctive and the imperative are signaled by special overt morphemes. This is not the case with respect to the indicative. Likewise, the neuter and the feminine gender have explicit markers, in contrast to the masculine. 7. On this split between semantics and morphology, cf. von Stechow (2012, section 8). 8. My treatment of selection departs from the common system of feature checking in generative syntax (cf. von Stechow 2012, section 8 and Ambar, this volume). It deserves additional study to compare the dif ferent approaches to selection, especially with regard to minimality. 9. I do not use terms like PAST in syntax. Instead the involved constituents have corresponding features like +pret, and the functional head is phonetically zero. 10. I do not agree with Ambar’s analysis (in this volume ) of the Russian subjunctive. She assumes that in Russian subjunctive clauses the particle by is combined with the past indicative form of the verb. The past verb form selected by the subjunctive particle by is left unspecified with respect to mood, in my system (see (5b), (6c), and (17b, c). 11. In modern generative syntax, the feature +pret in (5b) would be u(ninterpretable)past, which is checked by the feature i(nterpretable)past in T 0, instead of the selection feature +pret in (6c). 12. Unbound variables can be activated by lambda abstraction in SF, or they are specified, co-indexed, or existentially bound in CS. In indicative root clauses like (1), t is a parameter and gets existentially bound in CS. 13. Corresponding unmarked complement clauses are introduced by the complementizer čto ‘that’. It has the lexical entry (i). Semantically, it is an empty function (Zimmermann 2015a, b). (i)

a. /čto/ b. – Force+C−wh c. λp [p] with p ∈ {t, }

14. As for the common ground, see (19c) and n. 23. 15. In root clauses, movements to ForceP/CP as in (14) and (16) take place in PF in order to mark the beginning of a clause and are not visible in LF.

chapter eight

304

16. The unbound variables t and x in (15) are parameters and will be existentially bound in CS. As regards the modifiers and the subject in (1), I assume that they are merged and interpreted as adjuncts of TP and VP and in the complement position of V, respectively. 17. Isačenko (1962), Barnetová et al. (1979), and RG (1980), describe the Russian subjunctive in detail. Concerning generative approaches to mood, see Ambar (this volume). 18. Infi nitival control constructions with the subjunctive marker by are not considered here. 19. In colloquial Russian, the particle by can occur twice, after the verb and on the left periphery of the pertinent clause. 20. Migdalski (2006) and Tomaszewicz (2012) assume for Polish that by is merged as the head of a special modal functional phrase between TP and NegP (my PolP) and moved to the left periphery. In Ambar (this volume, (59); cf. n. 1, here), the Russian subjunctive marker by is merged as the head of EvalP and the complementizer čto is raised from FinP to this position. 21. Observe that the selected verb form is unspecified for mood (cf. (6c)). 22. As with tense (see n. 11), the feature +subj in (17b) could be u(ninterpretable) subj, which is checked by the feature i(nterpretable)subj in Mood, instead of the selection feature +subj in (18c). 23. The relation Rconf is comparable to the accessibility relation between worlds described by Kratzer (1991a, b). Taking into account developments in situation semantics, (8c) and (18c), equivalently, could be represented as (i) and (ii), respectively. (i) (ii)

λP(λσ) α∃e [([e ≤ σ] ˄) α [P t e]] λPλσ∃e [[[~[σ Rconf σu]] ˄ [e ≤ σ]] ˄ [P t e]] with P ∈ , σ, σu ∈ , ≤ ∈ α, α∈ { λw. λt. λ P ∃e[e is in w & τ (e) ⊆ t & P(e) IMPERFECTIVE ∼> λw. λt. λP.∃e[e is in w & t ⊆ τ (e) & P(e)]

In Slavic languages, there is a strong contrast between perfective and imperfective aspect in their morphological and semantic markedness. Only perfective aspect is marked both morphologically and semantically. The meaning of the perfective operator is fairly specific, and it invariably denotes an atomic event; see (4). In contrast, imperfective aspect may give rise to many interpretations; see (5). Unlike progressive aspect in English, imperfective aspect in Slavic can be used to denote both unbounded episodic (5-i) and quantified/frequentative/iterative/habitual eventualities (5-ii). In both uses the eventualities in the denotation of imperfective aspect are unbounded. (4)

Interpretation of perfective sentences Jan z-gubi- ł klucz. John pfv-lose-pst key (i) ‘John lost a key (once)’ (ii) *‘John lost a key (more than once).’

✓episodic *non- episodic

chapter eleven

386

(5)

Interpretation of imperfective sentences Jan kosi trawnik. John mow.prs.ipfv lawn (i) ‘John is mowing the lawn.’ (ii) ‘John mows the lawn.’

✓episodic/progressive ✓non- episodic/habitual

Klimek-Jankowska (2012) notes that the semantics proposed by Kratzer (1998) for perfective aspect is the same for Polish and French and that it does not capture one important difference in its distribution in these two languages. Boneh and Doron (2010, p. 341) following Kleiber (1987, p.  216 (45)) claim that in French habituals of the type presented in (6) perfective aspect expresses a habit that is limited within the boundaries of a given time span. By contrast, perfective aspect in Polish cannot express habits included within the boundaries of a given time span, as evidenced by the ungrammaticality of a corresponding Polish example in (7). Note that this expression should be possible according to the defi nition of perfective given in (3). Thus the perfective aspect in Polish imposes an additional restriction on an event argument, namely that it is atomic (noniterated). To sum up, perfective aspect in Polish, apart from locating the temporal trace of an event within a reference time, requires that an event e that satisfies a predicate P does not consist of the sum of proper subevents that also satisfy P. This modification is included in the semantic entries of Polish perfective and imperfective aspect in (8). (6)

French (Kleiber 1987: 216) Paul est allé-pfv à la messe le dimanche pendant trente ans. ‘Paul went to church on Sundays for 30 years.’

(7)

*Paweł po-szed- ł do kościoła w niedzielę przez 30 lat. Paweł pfv-go-pst to chuch on Sunday for 30 years. intended: ‘Paul went to church on Sundays for 30 years.’

(8)

PERFECTIVE ∼> λw. λt. λ P.∃ e[e is in w & τ (e) ⊆ t & P(e) & e = σe’[P(e’) & e’ ⊂ e]] Paraphrase: IMPERFECTIVE ∼> λw. λt. λP.∃ e[e is in w & t ⊆ τ (e) & P(e)]

Aspectual Coercion versus Blocking

387

Coming back to the layered representation in (1), the next projection above the grammatical Aspect in Polish is Tense. In Polish one can distinguish between past and non-past tenses (see also Whaley 2000; Borik 2006; Mezhevic 2006). In the past tense, aspect plays a purely aspectual role in the sense that it only shows how an eventuality unfolds in time but does not locate it with respect to the speech time; see (9). In contrast, in non-past tense it is the aspect that locates an eventuality with respect to the speech time. More precisely, non-past imperfective verbs can only receive a present tense interpretation (10a), while non-past perfective verbs can only be interpreted as future (10b). 2 (9)

a. Jan pisa- ł artykuł. John write.ipfv-pst article ‘John was writing an article.’ b. Jan na-pisa- ł artykuł. John pfv-write.pst article ‘John wrote an article.’

(10)

a. Jan pisze artykuł. John write.impf.non-pst article ‘John is writing an article.’ b. Jan na-pisze artykuł. John pfv-write.non-past article ‘John will write an article.’

With this background in mind, we can now formulate the basic observation, namely: while imperfective aspect seems to be not fully specified (as it allows for more than one interpretation), perfective aspect has just one possible interpretation in Polish, namely that of a single atomic event; hence its semantics is fixed. In the literature on aspect there is much discussion of whether, in addition to visible morphological aspectual operations, there is space for “interpretational adjustments” required by context (internal or even external). This question is especially interesting from the Polish (or Slavic) point of view, given that in this language there is a lot of

chapter eleven

388

morphology regulating aspectual interpretation. The question that arises at this point is whether “aspectual adjustment operations” are possible in Polish, and if so, of what kind? In order to answer this question, let us look first at what kinds of tacit aspectual adjustment operations exist in other languages.

3

Coercion

3.1

Coercion in Theoretical Research

When it is necessary to adjust the aspect of a verb or one of its projections to the requirements of the context, linguists speak of aspectual coercion. Usually this kind of aspectual adjustment is necessary when there is a mismatch between the aspect of a verbal expression and tej aspectual constraint of its adverbial modifier; see (11) (from Dölling 2014, examples (1)–(3)). (11)

a. #Fred played the sonata for one day. b. #Fred played the sonata for one minute. c. #Fred played the piano for one year (habitual interpretation)

In (11a) the eventuality denoted by the VP played the sonata is inherently bounded. The adverbial modifier for one day is durative, and as such it requires that the eventuality it modifies does not have a natural end. Hence there is a clash between the aspectual properties of the VP and the requirements of the adverbial modifier. To prevent this clash, the VP meaning undergoes an adjustment and coerces an iterative meaning. Sentence (11b) provides an example of a coercion emerging from a sentence-internal source. The adverbial modifier for one minute triggers a subtractive interpretation, meaning that the eventuality of playing the sonata cannot have any terminal point. Sentence (11c) is an example of a coercion triggered by an external source. Dölling (2014) points out that sometimes coercion emerges from the incompatibility of the eventuality with our experiential knowledge. For instance, in (11c) we arrive at the meaning that Fred played the piano repeatedly. However, in order to get this habitual interpretation we rely on our world knowledge: one cannot play the piano for one year without stopping.

Aspectual Coercion versus Blocking

389

One important point is that all of the adjustments demonstrated in (11a– c) are morphologically invisible. The question is how these coercion operations can be explained. Originally, scholars such as Pustejovsky (1995), Jackendoff (1997), de Swart (1998), and Rothstein (2004) assumed that aspectual coercion is always triggered by a mismatch between two expressions that are to be composed. For example, according to de Swart (1998, p.  361), in order to resolve the confl ict of the type illustrated in (11a) a coercion operator C e, changing the bounded eventuality played the sonata into a homogeneous one, is inserted in the compositional derivation, right above the eventuality denoted by the VP and before it serves as an input to the adverbial modifier for one day, which imposes its own aspectual constraints. 3 This means that coercion operators must be inserted incrementally into the compositional derivation. This aspect is regarded by Dölling (2014) as problematic for two reasons. First, he points out that aspectual coercion need not emerge from a sentenceinternal source, as was demonstrated in (11c). Second, as Dölling (2014) emphasizes, it is not always the verbal expression that is coerced as a result of an aspectual confl ict; sometimes it is the adverbial modifier that can be adjusted, as illustrated in (12). (12)

#Rob ran to the summit for thirty minutes.

As Dölling observes, this sentence can be interpreted such that the adverbial specifies the duration of the state brought about by the running event. In other words, the meaning conveyed by (12) is that Rob ran to the summit and the resulting state of his being at the summit lasted thirty minutes. Dölling points out that an appropriate coercion operator cannot be inserted incrementally (compositionally). Rather he takes this observation to be a strong argument in favor of a pragmatic approach to coercion. Following Egg (2005), he assumes that the meaning of some linguistic elements is strongly underspecified and that adjustments of aspect are context-driven enrichments, with no impact on semantic compositionality. Egg (2005; see also Pulman 1997) claims that the semantic structure of expressions does not contain any coercion operator; rather, the semantic construction introduces gaps or blanks that buffer potential aspectual conflicts and into which relevant operators (or a combination of them) can be inserted if required. In other words, our semantic system can somehow anticipate possible aspectual coercions and “prophylactically” include such buffer positions in the structure. If any aspectual incompatibility arises, the

chapter eleven

390

buffer position will be filled; otherwise it will be deleted. Dölling agrees in general, but says that the proposal is “too coarse-grained.” Instead, he proposes a two-stage approach. In the first stage a structure called semantic form (SF) models the context-independent meaning of expressions composed strictly compositionally. At the same time, SF comprises parameters—that is, free variables—that indicate where particular constants of the respective type are to be inserted into the linguistically determined meaning. This instantiation of parameters and, thereby, specification of meaning according to what is understood to be world knowledge, is the task of pragmatics. When the SF parameters are instantiated, a parameterfixed structure for the meaning of expression arises, which constitutes the second stage in Dölling’s approach. In sum, “this two-stage approach considers that grasping the full meaning of an utterance always includes pragmatic enrichments of its SF in the course of interpretation” (Dölling 2014, p. 222). In the next section we look at coercion in the context of experimental research.

3.2

Coercion in Experimental Research

Some of the first experiments involving aspectual coercion were conducted by Piñango et al. (2006), Pylkkänen and McElree (2006), Pickering et al. (2006), Bott (2008), Brennan and Pylkkänen (2008), and Pylkkänen (2008), who focused on a very small subset of possible kinds of coercion— specifically, complement coercion and iterative coercion—as exemplified in (13a and (b) respectively. (13)

a. John began the book. b. John jumped for ten minutes.

In (13a) began requires an event-denoting complement (as it describes the beginning of an event), but the book denotes an entity, not an event. Hence there is a mismatch: the book is of the wrong semantic type to combine directly with begin. To repair it, the meaning of an entity-denoting direct object must be shifted (coerced) into an event-denoting complement, resulting in a coerced interpretation such as ‘John began some activity involving the book (reading or writing)’. This operation or phenomenon is called complement coercion (Pustejovsky 1995). In (13b) there is a mismatch between the temporal (durative) modifier for ten minutes and the verb jumped, which denotes a near-instantaneous

Aspectual Coercion versus Blocking

391

punctual event (Brennan and Pylkkänen 2008). Hence the meaning of a punctual event must be coerced into an iterative meaning, resulting in the interpretation ‘John jumped several times for ten minutes’ (repetitive meaning). This operation or phenomenon is called aspectual coercion (Moens and Steedman 1988; Pustejovsky 1991) (or more precisely iterative coercion; see Bott 2010). Bott (2010) in his experiments includes, in addition to iterative coercion, two further types: subtractive and additive coercion, exemplified in (14a and b) respectively. (14)

a. John wrote a letter for hours. b. John found the keys in ten minutes.

In (14a) a mismatch arises because of the bounded character of the eventuality denoted by the VP wrote the letter and durative character of the adverbial modifier for hours. In order to resolve this conflict, the accomplishment is stripped of its culmination, resulting in a simple activity meaning. In (14b) an achievement is coerced into an accomplishment by adding a preparatory phase (e.g., the activity of searching) to the culmination (fi nding the key) because the achievement itself denotes a punctual eventuality that is too short to last ten minutes. The overall picture emerging from psycholinguistic research is rather blurry. Even the same type of coercion—for example, iterative coercion— can yield mixed fi ndings depending on the research question, the design, or the method used in the experiment. So for instance, some studies (e.g., Piñango et al. 2006; Brennan and Pylkkänen 2008) have found that sentences with iterative coercion such as that in (13b) take longer to read than sentences involving no aspectual mismatch. At the same time, there are other studies (e.g., Pickering et al. 2006) that did not fi nd any such processing consequences. Brennan and Pylkkänen (2008) conducted a magnetoencephalography (MEG) study aimed at fi nding neural correlates of the processing cost elicited by aspectual mismatch resolution in sentences like (13b); they managed to identify two distinct effects: an earlier right-lateral frontal, anterior temporal, and posterior temporal/cerebellar effect at 340–380 milliseconds (ms) after verb onset, and a later anterior midline effect at 440–460 milliseconds (ms).4 To our knowledge, Bott (2010 and this volume) has undertaken the broadest psycholinguistic research on aspectual coercion. In a nutshell,

chapter eleven

392

his behavioral experiments show that the different kinds of coercion lead to different results. For instance, iterative coercion based on accomplishments (as in (15a)) differs from that based on semelfactives (as in (15b)) in that accomplishments are harder to coerce, which in Bott’s view results from the fact that, first, an implausible subtractive coercion reading is computed, and only afterwards is this initial choice revised in favor of an iterative meaning. (15)

German (Bott: this volume) a. Hans durchtauchte das Schwimmbecken dreißig Jahre lang, Hans dived through the pool for thirty years dann wurde ihm das Tauchen vom Arzt verboten then was him the diving by his doctor forbidden ‘Hans dived through the pool for thirty years before his doctor forbade him to continue.’ b. Den ganzen Morgen nieste der Junge recht laut . . . the whole morning sneezed the boy rather loudly . . . ‘The whole morning, the boy sneezed rather loudly . . .’

To the best of our knowledge, Bott (2010, this volume) is so far the only one to conduct an ERP (event-related potentials) study of aspectual coercion (or more precisely, additive coercion). The experiment was conducted in German; see (16). (16)

German (Bott: this volume) In zwei Stunden hatte der in two hours had the obwohl sie gut versteckt although it well hidden

Förster die Falle entdeckt, ranger the trap discovered war. was

‘In two hours, the ranger had discovered the trap, although it was hidden well.’ For this type of coercion Bott reports a negativity that emerged rather late and persisted for at least one second. He takes this negativity to reflect increased working memory demands and hence to be an instance of a sustained working memory LAN (left anterior negativity). Bott fo-

Aspectual Coercion versus Blocking

393

cused on additive coercion resulting from adding a preparatory phase to an achievement, yielding an accomplishment. In some sense the socalled inchoative coercion, reported by Brennan and Pylkkänen (2010) and illustrated in (17), might also be regarded as an instance of additive coercion because an inception is added to a state. (17)

Within a few minutes, the child cherished the precious kitten.

In (17) the predicate cherish denotes a state. To match the semantics of the completive adverbial within a few minutes the context in (17) must be enriched with a “begin-to- cherish” (inceptive) component. This type of inceptive coercion has been recently tested in Brennan and Pylkkänen’s (2010) self-paced reading and a subsequent magnetoencephalography (MEG) study for psych verbs in English. They found a reading delay associated with inchoative coercion and a distributed fronto-temporal effect of around 300–500 ms for coercion.

3.3

Taking Stock So Far: Factors That Determine Coercion

In this section we present new experimental evidence from an ERP study investigating how aspectual mismatches are processed in Polish, a language with a clear aspectual morphology. It will be shown that, when confronted with the Polish data, the existing theories of coercion seem too permissive. They claim that “coercion is triggered if there is a confl ict between the aspectual character of the eventuality description and the aspectual constraints of some other element in the context” and that “[t]he felicity of an aspectual reinterpretation is strongly dependent on linguistic context and knowledge of the world” (de Swart 1998, p. 360). It seems, however, that even in the case of imperfective aspect in Polish, which (as was pointed out in section  2) is semantically underspecified and as such should be expected to leave some space for tacit aspectual adjustments in contexts in which there is a clash between the semantic properties of two elements to be compositionally combined, there is not necessarily an automatic repair mechanism. It turns out that in many cases of such clashes the imperfective form is not coerced; instead, the parser compares the problematic imperfective form with other existing aspectual forms in this language (Polish) in accordance with pragmatic principles, as formulated in (18) (Sauerland 2008, p. 598 (52)).

chapter eleven

394

(18)

Maximize presupposition (Percus 2006): Do not use S in context set c if there is an S′ such that: a. c ⊂ domain([[S]]′) b. you believe S′ to be true c. S′ is derived from S by a single or multiple replacements of one item of a scale with another item higher on the same scale

The principle in (18) can be paraphrased as in (19) (from Kagan 2008, (13)): (19)

Maximize assertion: Use the most informative assertion that is true. Maximize presupposition: Use the most informative presupposition that is satisfied.

For our purposes here, these principles should be interpreted in such a way that, in cases in which two competing forms (perfective and imperfective verbs) satisfy the truth- conditional requirements of some context, the form that wins the competition is the one that is semantically (and morphologically) more marked—that is, the one that has a more specific (informative) semantic content. To anticipate the discussion to follow, it turns out that the perfective form that is more marked blocks the use of an imperfective form, which as such could also satisfy the interpretational needs of a given context, if it were coerced. In sum, the following factors matter for the process of aspectual interpretation: (20)

a. Overt aspectual morphology and compositional operations b. Tacit adjustments (coercion) c. Blocking (resulting from the existence of a competition between two forms)

In section 4 we provide a background for converbs, which served as the basis for our ERP study. As will be briefly shown, converbs provide the most fruitful ground for testing the limits of aspectual coercion in Polish.

4

Basic Facts about Converbs in Polish

Converbs (also referred to as conjunctive participles or gerunds, depending on the linguistic tradition) are temporally anchored participial

Aspectual Coercion versus Blocking

395

clauses functioning in a sentence as optional (i.e., adjunct-like) adverbial modifiers. In temporal relations between the matrix and the embedded event, two types of converbs can be distinguished: (i) simultaneous converbs, or SIM (see (21a)), and (ii) anterior converbs, or ANT (see (21b)). 5 We use the following abbreviations in the glosses: sim.ptcp = simultaneous converb, ant.ptcp = anterior converb. For clarity, we separate converbial morphemes and perfective prefi xes from the rest of the verb with a hyphen, although this is not a convention of Polish orthography. (21)

a. Anna szła do pokoju pal-ąc papierosa Ann walk.pst.ipfv to room smoke.impf-sim.ptcp cigarette w pośpiechu. in hurry ‘Ann was walking to the room (while) smoking a cigarette in a hurry.’ b. Anna we-szła do pokoju za-pali-wszy papierosa Ann pfv-walk.pst to room pfv-smoke-ant.ptcp cigarette w pośpiechu. in hurry. ‘Ann entered the room (after) having smoked/lit a cigarette in a hurry.’

Polish converbs impose two kinds of restrictions: (i) specific morphological selectional requirements, and (ii) specific semantic/pragmatic constraints on temporal ordering.

4.1 Morphological Selectional Requirements A simultaneous converb like paląc in (21a), meaning ‘while smoking’, consists of a converbial morpheme - ąc attached to a verbal stem. An anterior converb like zapaliwszy in (21b), meaning ‘after having smoked’, consists of a converbial morpheme -wszy attached to a verbal stem. What is particularly relevant for our study is the fact that these two converbial morphemes impose specific selectional restrictions on the kind of aspectual verbal stem they can be combined with: the simultaneity morpheme - ąc selects for an imperfective verbal stem while

396

chapter eleven

the anteriority morpheme -wszy selects for a perfective verbal stem. Given that in Polish every verb is morphologically marked by either perfective or imperfective aspect, and given the morphological selectional requirements of converbial morphemes, the following aspectual mismatches arise; see (22). In (22a) the simultaneity morpheme - ąc semantically selecting for an imperfective verbal stem is combined with a perfective verbal stem. In (22b) the anteriority morpheme -wszy semantically selecting for a perfective verbal stem is combined with an imperfective verbal stem. (22)

a. *Anna szła do pokoju za-pal-ąc Ann walk.pst.ipfv to room pfv-smoke-sim.ptcp papierosa w pośpiechu. cigarette in hurry ‘*Ann was walking to the room (while) having smoked/lit a cigarette in a hurry.’ b. *Anna we-szła do pokoju pali-wszy Ann pfv-walk.pst to room smoke.impf-ant.ptcp papierosa w pośpiechu. cigarette in hurry Lit. ‘Ann entered the room (while) having smoked a cigarette in a hurry.’

Table 11.1 summarizes the possible matching and mismatching combinations of converbial morphemes and verbal stems in Polish.

4.2

Semantic and Pragmatic Constraints on Temporal Ordering

In addition to the fact that converbial morphemes impose specific morphological selectional restrictions, they also impose specific temporal orderings between the main clause eventuality and the converbial eventuality. Obviously, the anterior converb zapaliwszy ‘after having smoked’ in (21b) requires that the main clause eventuality follow the converbial eventuality, whereas the simultaneous converb paląc ‘while smoking’ in (21a) requires an overlapping temporal relation between the main clause eventuality and the converbial eventuality. Ideally, the perfect match for an anterior converb would be a perfective matrix verb (as in (21b)), and

Aspectual Coercion versus Blocking

397

table 11.1 Converbs: Morphological (Mis)matches Verbal Stem

Converbial Morpheme

Examples

Imperfective

-ąc (SIM) -wszy (ANT)

✓pal-ąc ‘smoking’ *pali-wszy

Perfective

-ąc (SIM) -wszy (ANT)

*za-pal-ąc ✓za-pali-wszy ‘having smoked’

Note: SIM = simultaneity; ANT = anteriority; ✓ = grammatical; * = ungrammatical

the perfect match for a simultaneous converb would be an imperfective matrix verb (as in (21a)). This is so because a perfective denotes an eventuality with clear boundaries, which in turn makes it possible to locate such a perfective eventuality after the converbial eventuality. In contrast, an imperfective verb denotes an unbounded eventuality, which as such can overlap with the converbial eventuality. The examples in (23) illustrate a situation in which these ideal matches are not respected—that is, when an anterior converb is combined with an imperfective matrix verb (23a) and a simultaneous converb is combined with a perfective matrix verb (23b). (23)

a. ?Anna szła do pokoju za-pali-wszy Ann walk.pst.ipfv to room pfv-smoke-ant.ptcp papierosa w pośpiechu. cigarette in hurry Lit.: ‘Ann was walking to the room (after) having smoked/lit a cigarette in a hurry.’ Intended: ‘Ann started walking to the room after she had smoked a cigarette in a hurry.’ b. ?Anna we-szła do pokoju paląc Ann pfv-walk-pst to room smoke.ipfv-sim.ptcp papierosa w pośpiechu. cigarette in hurry ‘Ann entered the room while smoking a cigarette in a hurry.’

Table 11.2 summarizes the possible matching and mismatching combinations of main clause verbs, marked for perfective or imperfective aspect,

chapter eleven

398 table 11.2 Converbs: Semantic and Pragmatic (Mis)matches Main Clause Eventuality

Converbial Eventuality

Examples

Imperfective verb

V-ąc (SIM) V-wszy (ANT)

✓ cf. (21a) ? cf. (23a)

Perfective verb

V-ąc (SIM) V-wszy (ANT)

? cf. (23b) ✓ cf. (21b)

Note: SIM = simultaneity; ANT = anteriority; ✓ = acceptable; ? = marginally acceptable

with converbs, expressing a simultaneous or anterior relation between the main clause and the embedded clause eventualities.

4.3

Converbial Contexts in Polish: Expectations with Respect to Coercion

As pointed out in the two preceding sections, there are four types of mismatches in converbial contexts: two morphological clashes (recall Table  11.1), and two semantic/pragmatic mismatches (recall Table  11.2). Given the earlier discussions of coercion and of imperfective action (more precisely, its underspecified interpretation), we would expect that in all cases with a clash involving an imperfective verbal form it should be possible to carry out a tacit adjustment operation, resulting in acceptable sentences. To be precise, we would expect the following clashes to be resolved: (i) one of the morphological clashes—the combination of an imperfective verb with a converbial anteriority morpheme -wszy ((22b)), and (ii) one of the semantic/pragmatic clashes—the combination of an imperfective main clause eventuality and the anterior converb ((23a)). In contrast, given what we said about the perfective aspect in section 2, more precisely its semantic specificity, we would expect that no repair mechanisms should be possible in clashes involving a perfective verb. To be precise, we expect no resolution of the following clashes: (i) one of the morphological clashes—the combination of a perfective verb with a converbial simultaneity morpheme - ąc ((22a)), and (ii) one of the semantic/pragmatic clashes—the combination of a perfective main clause eventuality with the simultaneity converb ((23b)). These expectations were the testing ground for the ERP study, which was combined with an online acceptability rating study and a corpusbased analysis of the frequency of simultaneous and anterior converbs in Polish.

Aspectual Coercion versus Blocking

5

399

Experimental Research on Converbs

5.1

Relevant Conditions and Comparisons

As mentioned in section  4, there are two kinds of morphological mismatches, *perfective + - ąc sim and *imperfective + -wszyant, and two kinds of semantic/pragmatic mismatches, anterior converb + imperfective main verb and simultaneous converb + perfective main verb. Given this, we decided to create two sets of data, each set consisting of three conditions: Condition 1: Control, Condition 2: Morphological mismatch, and Condition 3: Semantic/pragmatic mismatch. In the fi rst set (Set A) we only have imperfective main verbs, and in the second set (Set B) we only have perfective main verbs. The design is summarized in Table 11.3. Examples: Set A Condition 1: Control ✓Anna szła do pokoju pal-ąc papierosa Ann walk.pst.ipfv to room smoke.ipfv-sim.ptcp cigarette w pośpiechu. in hurry ‘Ann was walking to the room (while) smoking a cigarette in a hurry.’ Condition 2: Morphological mismatch *Anna szła do pokoju za-pal-ąc Ann walk.pst.ipfv to room pfv-smoke-sim.ptcp papierosa w pośpiechu. cigarette in hurry ‘*Ann was walking to the room (while) having smoking a cigarette in a hurry.’

table 11.3 Experimental Design Set

Main Clause

Converbial Clause

Condition

A

Imperfective

✓ simultaneous *simultaneous ✓ anterior

1. Control 2. Morphological mismatch 3. Semantic/pragmatic mismatch

B

Perfective

✓ anterior *anterior ✓ simultaneous

1. Control 2. Morphological mismatch 3. Semantic/pragmatic mismatch

Note: ✓ = grammatical; * = ungrammatical

400

chapter eleven

Condition 3: Semantic/pragmatic mismatch ?Anna szła do pokoju za-pali-wszy Ann walk.pst.ipfv to room pfv-smoke-ant.ptcp papierosa w pośpiechu. cigarette in hurry Lit.: ‘Ann was walking to the room (after) having smoked/lit a cigarette in a hurry.’ Intended: ‘Ann started walking to the room after she had smoked a cigarette in a hurry.’ Examples: Set B Condition 1: Control ✓Anna we-szła do pokoju za-pali-wszy papierosa Ann pfv-walk.pst to room pfv-smoke-ant.ptcp cigarette w pośpiechu. in hurry ‘Ann entered the room (after) having smoked/lit a cigarette in a hurry.’ Condition 2: Morphological mismatch *Anna we-szła do pokoju pali-wszy Ann pfv-walk.pst to room smoke.ipfv-ant.prt w pośpiechu. in hurry

papierosa cigarette

Lit. ‘Ann entered the room (while) having smoked a cigarette in a hurry.’ Condition 3: Semantic/pragmatic mismatch ?Anna we-szła do pokoju paląc Ann pfv-walk.pst to room smoke.ipfv-sim.prt w pośpiechu. in hurry.

papierosa cigarette

‘Ann entered the room while smoking a cigarette in a hurry.’ Condition 1 is a control condition since the simultaneous converb in Set A is derived from an imperfective (durative/unbounded) verb, thus obeying the selectional restrictions of the simultaneity morpheme - ąc;

Aspectual Coercion versus Blocking

401

accordingly, the anterior converb in Set B is derived from a perfective (bounded) verb, thus obeying the selectional restrictions of the anteriority morpheme -wszy. Condition 2 represents a morphological mismatch since in Set A the simultaneity morpheme - ąc is combined with a perfective (bounded) verb, and hence its selectional restrictions are violated; accordingly, in Set B the anteriority morpheme -wszy is combined with an imperfective (unbounded) verb. Condition 3 is an instance of a semantic/pragmatic mismatch at the discourse level. In Set A the converbial clause contains an anterior converb which requires that the eventuality denoted by the main verb should be temporally located (hence it should begin) after the eventuality in the converbial clause. For this to happen, the eventuality in the main clause must have (at least) an initial boundary in order to be located after the eventuality in the converbial clause. As already mentioned, bounded eventualities in Polish are morphologically manifested by means of perfective aspect. However, the main verb in Condition 3/Set A is imperfective (i.e., semantically unbounded) hence there is a mismatch between the requirements of the anteriority morpheme and the unbounded interpretation of the main verb. In order to process and understand this sentence, there is a need for some repair of this mismatch. The unbounded eventuality in the main clause has to be reinterpreted as having a clear inception. In Condition 3 in Set B the converbial clause contains a simultaneous converb that imposes a simultaneity interpretation—that is, the eventuality denoted by the main verb should be temporally overlapping with the eventuality in the converbial clause. Ideally, an imperfective verb denoting an unbounded eventuality should be used in the main clause to guarantee the overlapping temporal relation between two events, as required by the simultaneous converb. However, in this case a perfective verb is used in the main clause which, by denoting a bounded eventuality, cannot be simultaneous with the converbial eventuality, but rather is understood as enclosed within it. For “morphological mismatch” conditions we used a within-set comparison with the control condition. In other words, sentences in Condition 1/Set A served as a control for sentences in Condition 2/Set A, and accordingly, sentences in Condition 1/Set B served as a control for sentences in Condition 2/Set B. For the “semantic/pragmatic mismatches” we used an across-set comparison. More precisely, the control condition for sentences in Condition

chapter eleven

402 table 11.4 Relevant Comparisons Comparison

Morphological Mismatches

1 2

Condition 1/Set A (control) vs. Condition 2/Set A (morphological violation) Condition 1/Set B (control) vs. Condition 2/Set B (morphological violation) Semantic and Pragmatic Mismatches

3 4

Condition 1/Set B (control) vs. Condition 3/Set A (inception) Condition 1/Set A (control) vs. Condition 3/Set B (inclusion)

3/Set B were sentences in Condition 1/Set  A. Accordingly, the control condition for sentences in Condition 3/Set A were sentences in Condition 1/Set B. Sentences in Condition 1/Set A served as a control condition for sentences in Condition 3/Set B because the sentences were identical except for the fact that those in Condition 1/Set A used imperfective main verbs. Likewise, sentences in Condition 1/Set B served as a control condition for sentences in Condition 3/Set A because the sentences were identical except for the fact that those in Condition 1/Set B used perfective main verbs. Table 11.4 summarizes the relevant comparisons between the conditions.

5.2

Online Acceptability Rating Studies

5.2.1 ACCEPTABILITY

OF

CONVERBS

IN

SENTENTIAL

CON-

Material, Participants, and Experimental Task: The material consisted of 300 stimuli sentences that were divided into five lists, each containing ten sentences per condition. Each participant saw altogether ninety sentences: 10 sentences x 6 conditions = 60 + 30 fi ller sentences. The conditions were the following: sentences with simultaneous converbs (no violations), sentences with simultaneous converbs with morphological violations, sentences with anterior converbs (no violations), sentences with anterior converbs with morphological violations, sentences with semantic/pragmatic violations (inception), sentences with semantic/pragmatic violations (inclusion). Forty fi rst- and second-year students from the University of Wrocław, all of them native speakers of Polish, took part in this online questionnaire study. The participants were asked to evaluate the acceptability of presented sentences on a scale from 0 (absolutely unacceptable) to 100 (perfectly acceptable). Only the minimal and maximal numerical values (0 and 100) were presented.

TEXTS.

Aspectual Coercion versus Blocking

403

67.81

inclusion

100 incepƟon

45.71 100

Figure 11.1: Pragmatic mismatches in converbs used in sentences: Acceptability rating on the scale from 0 to 100.

Results and Discussion: Semantic/Pragmatic Mismatches. For sentences with semantic/pragmatic mismatches we obtained the results displayed in Figure 11.1. Recall from section 4.2. that anterior converbial morphemes impose semantic restrictions in meaning, requiring the eventuality denoted by the converb to precede the eventuality of the main clause. This semantic selectional requirement is satisfied in (24a), as the main clause eventuality is perfective and therefore unambiguously expresses an initial boundary; given this it can naturally follow the eventuality expressed in the temporally anterior embedded clause. In order to satisfy the semantic selectional requirements of the anteriority morpheme -wszy in (24b), the imperfective main clause eventuality has to undergo a tacit adjustment operation—specifically, adding an initial boundary (inception). This is so because the eventuality denoted by an imperfective verb is unbounded; that is, it lacks an initial boundary. Note that the result obtained in our acceptability rating study for sentences like those in (24b) is challenging. The mean score is here 45.71—that is, it is viewed as neither a strong violation nor completely acceptable. Assuming that tacit repair mechanisms should easily arise, the “repaired” sentences should be fully acceptable. At this point it is not possible to say defi nitively whether we are dealing with a sort of coercion in (24b). We therefore scrutinize these data in our ERP study, reported in section 5.4.

404

(24)

chapter eleven

a. ✓Anna we-szła do pokoju za-pali-wszy Ann pfv-walk.pst to room pfv-smoke-ant.ptcp papierosa w pośpiechu. cigarette in hurry ‘Ann entered the room (after) having smoked/lit a cigarette in a hurry.’ b. ?Anna szła do pokoju za-pali-wszy Ann walk.pst.ipfv to room pfv-smoke-ant.prt papierosa w pośpiechu. cigarette in hurry Lit.: ‘Ann was walking to the room (after) having smoked/lit a cigarette in a hurry.’ Intended: ‘Ann started walking to the room after she had smoked a cigarette in a hurry.’

In (25) the semantic selectional requirements of the simultaneity morpheme - ąc are illustrated. Such converbs require that the main clause eventuality be simultaneous with the converbial eventuality. This requirement is naturally satisfied in (25a), in which an imperfective verb in the main clause is used since (as we have already pointed out) imperfective verbs denote unbounded eventualities and as such can naturally overlap with some other eventuality (the converbial one here). In contrast, in (25b) there is a perfective verb in the main clause. In order to satisfy the semantic needs of the converbial simultaneity morpheme, we should expect that some tacit adjustment mechanism will arise, more precisely that of subtractive coercion: the bounded (perfective) eventuality should be stripped of its initial boundary. Note that sentences with the mismatch at hand are judged as more acceptable (mean score: 67.81) than the clashes in which an inceptive coercion is expected. Moreover, it is very unlikely that any repair mechanism is involved in (25b) since (as pointed out in section 2) perfective verbs, being morphologically and semantically marked, are immune to any adjustments. Rather, what seems to be the case in (25b) is that speakers interpret such sentences in such a way that the bounded (perfective) main clause eventuality is included in the unbounded converbial eventuality.

Aspectual Coercion versus Blocking

405

72.66

ANT morph

100 ANT control

18.68 100

SIM morph

60.85 100

SIM control

32.62 100

Figure 11.2: Morphological mismatches in converbs used in sentences: Acceptability rating on the scale from 0 to 100.

(25)

a. ✓Anna szła do pokoju pal-ąc Ann walk.pst.impf to room smoke.ipfv-sim.ptcp papierosa w pośpiechu. cigarette in hurry ‘Ann was walking to the room (while) smoking a cigarette in a hurry.’ b. ?Anna we-szła do pokoju paląc Ann pfv-walk.pst to room smoke.ipfv-sim.ptcp papierosa w pośpiechu. cigarette in hurry ‘Ann entered the room while smoking a cigarette in a hurry.’

Morphological mismatches. As for the sentences with morphological mismatches, the results are shown in figure 11.2. Interestingly, there is a rather big contrast in the mean acceptability scores obtained for sentences with simultaneous converbs in “no morphological mismatch” contexts, such as those in (26a) (mean score: 72.66), and for simultaneous converbs in “morphological mismatch” contexts, such as those in (26b) (mean score: 18.68). In contrast, there is a rather small difference in the mean acceptability scores obtained for sentences with anterior converbs in “no morphological mismatch” contexts, such as those in (27a) (mean score: 60.85), and for sentences with anterior

406

chapter eleven

converbs in “morphological mismatch” contexts, such as those in (27b) (mean score: 32.62). (26)

a. ✓Anna szła do pokoju pal-ąc Ann walk.pst.ipfv to room smoke.ipfv-sim.ptcp papierosa w pośpiechu. cigarette in hurry ‘Ann was walking to the room (while) smoking a cigarette in a hurry.’ b. *Anna szła do pokoju za-pal-ąc Ann walk.pst.ipfv to room pfv-smoke-sim.ptcp papierosa w pośpiechu. cigarette in hurry ‘*Ann was walking to the room (while) having smoking a cigarette in a hurry.’

(27)

a. ✓Anna we-szła do pokoju za-pali-wszy Ann pfv-walk.pst to room pfv-smoke-ant.ptcp papierosa w pośpiechu. cigarette in hurry ‘Ann entered the room (after) having smoked/lit a cigarette in a hurry.’ b. *Anna we-szła do pokoju pali-wszy Ann pfv-walk.pst to room smoke.ipfv-ant.ptcp papierosa w pośpiechu. cigarette in hurry Lit. ‘Ann entered the room (while) having smoked a cigarette in a hurry.

At fi rst glance, these results are unexpected because in both cases the same kind of violation is involved, namely a morphological mismatch between a (SIM or ANT) converbial morpheme and a verbal stem marked as either perfective or imperfective. Despite the same kind of violation being involved, morphologically ill-formed anterior converbs are judged to be more acceptable in general than the ill-formed simultaneous con-

Aspectual Coercion versus Blocking

407

verbs. On closer inspection, the obtained results are in fact compatible with our expectation, formulated in section  4.3, that imperfective aspect in mismatch contexts should leave some space for tacit repair mechanisms (coercion). Note that ill-formed anterior converbs involve combinations of imperfective verbal stems and the converbial anteriority morpheme -wszy. In contrast, in the case of ill-formed simultaneous converbs there are combinations of perfective verbal stems and the converbial simultaneity morpheme - ąc. It should be emphasized that if some tacit repair mechanism were indeed involved, such forms should be as acceptable as their well-formed counter parts, contrary to fact. Our fi ndings suggest that no coercion can occur in these morphological mismatch contexts and that we should treat such cases as a violation. A similar pattern of results was obtained in our second online acceptability rating questionnaire presented in section 5.2.2. 5.2.2 ACCEPTABILITY OF CONVERBS IN ISOLATION. Material, Participants, and Experimental Task: In the second questionnaire study we tested potentially grammatical and potentially ungrammatical (but phonologically permissible) simultaneous and anterior converbial forms used in isolation. The stimuli were formed from real and jabberwocky verbal stems. The material consisted of 80 stimuli. Each participant saw altogether 100 words: 10 stimulus items x 8 conditions = 80 + 20 fi ller items (nominalizations). The conditions were the following: grammatical simultaneous converb, ungrammatical simultaneous converb, “potentially correct” nonsense simultaneous converb, “potentially incorrect” nonsense simultaneous converb, grammatical anterior converb, ungrammatical anterior converb, “potentially correct” nonsense anterior converb, and “potentially incorrect” nonsense anterior converb. Some examples of the material used in the second questionnaire study are given in (28) and (29).

(28)

a. real stems: grammatical ✓pal-ąc smoke.ipfv-sim.ptcp ‘smoking’ ✓za-pali-wszy pfv-smoke-ant.ptcp ‘having smoked’

chapter eleven

408

b. real stems: ungrammatical *za-pal-ąc pfv-smoke-sim.ptcp *pali-wszy smoke.ipfv-ant.ptcp (29)

a. jabberwocky stems: potentially grammatical ?brzdęśni-ąc V.ipfv-sim.ptcp ?z-brzdęśni-wszy pfv-V-ant.ptcp b. jabberwocky stems: potentially ungrammatical *z-brzdęśni-ąc pfv-V-sim.ptcp *brzdęśni-wszy V.ipfv-ant.ptcp

Thirty-two first- and second-year students from the University of Wrocław, all of them native speakers of Polish, took part in this online questionnaire study. The participants were asked to evaluate the acceptability of presented words on a sliding scale ranging from 0 (absolutely unacceptable) to 100 (perfectly acceptable). Only the minimal and maximal numerical values (0 and 100) were presented. Results and Discussion: The results of the second questionnaire study on real converbs tested in isolation are shown in Figure 11.3. In this questionnaire we obtained a similar pattern of acceptability ratings for real converbs as in the first questionnaire; that is, anterior converbs with morphological violations were judged more acceptable than their simultaneous ungrammatical counterparts. However, all jabberwocky converbs, regardless of whether they were potentially grammatical, were judged equally unacceptable (see Figure 11.4). To sum up, we found that, among native speakers, there is a considerable difference in the acceptability rating of well-formed simultaneous converbs and well-formed anterior converbs: the latter were evaluated as generally less acceptable than the former. Conversely, ill-formed simultaneous converbs turned out to be less acceptable than ill-formed anteriority converbs. Why should ungrammatical anterior converbs be judged more acceptable than ungrammatical simultaneous converbs?

Aspectual Coercion versus Blocking

409

93.3

ANT morph

100 ANT control

17.2 100

SIM morph

80.56 100

SIM control

36.4 100

Figure 11.3: Morphological mismatches in converbs used in isolation: Acceptability rating on the scale from 0 to 100.

12

ANT morph

100 ANT control

8.03 100

SIM morph

6.34 100

SIM control

8.1 100

Figure 11.4: Morphological mismatches in jabberwocky converbs used in isolation: Acceptability rating on the scale from 0 to 100.

From the theoretical point of view, these fi ndings are totally unexpected because both of them instantiate the same violation of morphological selectional restrictions imposed by converbial morphemes. A clue to this puzzle comes from the observation that all jabberwocky converbs, both those potentially grammatical and those potentially ungrammatical, are judged to be equally bad. It seems that someone who does not understand a lexical item (because it does not belong to the lexicon of the language under discussion) will fi nd it impossible to judge its acceptability. To test to what extent the frequency of simultaneous and

chapter eleven

410

9.54%

90.46%

Figure 11.5: Frequency of the occurrences of stimulus and anterior converbs in the IPI PAN corpus.

anterior converbs influences their acceptability when they are ill-formed we conducted a corpus-based analysis of the frequency of both kinds of converbs.

5.3

A Corpus-Based Analysis

The main fi nding of our corpus-based analysis is that anterior converbs are less frequent than simultaneous converbs. For instance, if you look at the frequency of occurrences of both kinds of converbs used in our experiment in the Polish National Corpus IPI PAN (2nd edition, 250 M segments; see Przepiórkowski et al. 2012), there were 12,990 occurrences altogether, of which the simultaneous converbs make up 90.46  percent (11,750 occurrences) and anterior converbs only 9.54  percent (1,240 occurrences). These fi ndings are displayed in Figure 11.5. The results from both online acceptability rating studies, together with our corpus-based analysis of the frequency of both kinds of converbs, point to the conclusion that people are less sensitive to morphological violations encountered in less frequently used (and hence less familiar) forms. This observation plays an important role in the interpretation of the results obtained in our ERP study.

Aspectual Coercion versus Blocking

5.4

411

ERP Study

5.4.1 MATERIAL AND EXPERIMENTAL CONDITIONS. The experimental conditions, together with examples illustrating them, are provided in section  5.1.6 As far as the experimental material is concerned, 300 stimulus items were constructed. All of them consisted of a main clause followed by a converbial clause. The structure of the main clauses was: subject, fi nite past tense verb, followed by either an object or an adverbial. Converbial clauses consisted of the following sequence: a simultaneous or an anterior adverbial participle derived from a transitive verb, followed by a direct object and an adverbial. (The latter was included to prevent spillover effects.) Two stimulus lists were created using these 300 tested sentences. Each list containing 150 sentences (50 per condition) was supplemented by 150 fi ller stimuli, of which 75 were well formed and plausible and 75 were illformed. The fi ller stimuli were complex sentences: similar to tested sentences, they consisted of a fi nite main clause and an adverbial clause. The incorrect fi ller sentences mainly contained different kinds of aspectual mismatches. All the stimuli were pseudo-randomly ordered. The critical word for us was the converb, but triggers were marked both on the critical word and the word following it. The length of the critical words (the converbs) was controlled for. This was important because of the peculiarities of Polish aspectual morphology. Briefly, in Polish there are bare imperfectives (nonderived forms), such as pisać impf ‘to write’ and secondary imperfectives, such as [[podpisperf]+ywać]impf ‘to sign’. Because secondary imperfectives contain the imperfectivizing suffi x -ywa, they are usually longer than bare imperfectives. To control the length of imperfective forms of the critical words, 50 percent of them had a bare imperfective form and the other 50  percent of the critical words had a secondary imperfective form. What about perfectives in Polish? As we pointed out in section 2, the most productive way of creating perfectives is to add a prefi x to a bare imperfective form: for example, pisać impf ‘to write’—na-pisać perf ‘to fi nish writing’. This perfectivizing strategy makes a derived perfective form longer than its bare imperfective base. The second way to form a perfective form is to use a specific (perfectivizing) semelfactive morphology: for example, kichać impf ‘to sneeze (several times)’—kich- ną- ć perf ‘to sneeze once’. The third way is to alternate a stem of the secondary imperfective form, which usually contains a lexical prefi x: for example,

412

chapter eleven

[[podpisperf]+ywa- ć]impf ‘to sign’ (either a continuous reading or a habitual reading)—[pod-[pisać impf]]perf ‘to sign’. Using the latter two ways of forming perfective aspectual forms guarantees that the perfective members of the respective aspectual pairs are shorter or at least as long as their imperfective counter parts. To control the length of perfective forms of the critical words, 50  percent of them were formed by adding a prefi x to a bare imperfective stem, and the other 50  percent of the critical words were formed by alternating a stem of the corresponding nonsemelfactive or secondary imperfective form. AND EXPERIMENTAL PROCEDURES. For our ERP, study 43 Polish native speakers (29 females, mean age 23.9, 18–37 years of age) were recruited at the University of Wrocław and received partial course credit for their participation in the experiment. All of them were right-handed according to the Edinburgh Handedness Inventory (Oldfield 1971) and had normal or corrected vision. None had neurological or psychiatric disorders or reported neurological traumas. Participants were tested individually in one session. The whole experiment (including the application of electrodes) took approximately ninety minutes. Following the application of the EEG electrodes, subjects were seated in front of a Samsung twenty-two-inch LCD computer screen approximately one meter from the screen. All stimuli were presented in a white courier font, size 48, on a black background using Presentation software. The experimental session was preceded by instructions and a trial session. As part of the instructions the participants were asked not to move or blink while a sentence was displayed. They were informed that the sentences would be presented segment by segment and that each sentence would be followed by an acceptability judgment question. The participants were instructed to provide their judgments as quickly as possible. They were also told which button on the Razor keyboard corresponded to which answer and which index fi nger they should use. To avoid the effects of lateralized readiness potential, half of the participants performed the task with the right hand, and the other half with the left hand. After reading a written instruction the participants received a practice block with several sentences related to the experiment to familiarize each subject with the task. After practicing, the participants received explicit feedback about the errors they had made. The trial session was fol-

5.4.2 PARTICIPANTS

Aspectual Coercion versus Blocking

413

lowed by six experimental blocks containing fi fty sentences each. After each block there was a pause to give subjects the opportunity to relax. Each trial consisted of the following events. A fi xation cross appeared in the center of the screen for 1,000 milliseconds (ms), after which a stimulus sentence was presented in a word-by-word or segment-by-segment (in the case of prepositional phrases) manner. Each word or segment appeared in the center of the screen for 550 ms, followed by a short interval in which the screen was blank. Sentence- ending words appeared with a full stop. Every sentence was accompanied by an acceptability judgment question. The possible answers were: ‘acceptable’, ‘unacceptable’, and ‘I don’t know’. After 4,000 ms the next trial started automatically. The questions were used to control the level of attention. In Condition 1 (control condition) in sets A and B, the expected answer was ‘acceptable’. In Condition 2 (morphological mismatch condition), the expected answer was ‘unacceptable’. In Condition 3 (pragmatic/semantic mismatch condition), the judgments were more delicate, so we did not a priori assume any correct answer. Thus only in Conditions 1 and 2 could a high number of incorrect answers be a reason to throw out the data from the analysis, since it would indicate that a subject had paid no or little attention. In contrast, in Condition 3 all the answers were considered possible and relevant for further analysis. In the event, no participant was rejected because of a high number of incorrect answers, indicating that they were indeed paying attention above the chance level. 5.4.3 EEG RECORDINGS. The EEG activity was measured with 24 Ag/ AgCl- electrodes, which were attached to the scalp using the Easycap system at Fz, FCz, Cz, CPz, Pz, POz, FC1, F3, C3, P3, O1, FC5, CP5, F7, P7, FC2, F4, C4, P4, O2, FC6, CP6, F8, P8. The ground electrode was positioned at AFz. Electrode positions were chosen in accordance with the international 10/20 system (Jasper 1958). Signals were referenced to the A1 electrode (left mastoid) and later re-referenced to the average of the left (A1) and right (A2) mastoid. Horizontal eye activity was measured by placing two electrodes 2 cm laterally to the right (EOGR) and the left (EOGL) canthus. Vertical eye activity was measured by placing two electrodes 3  cm above (EOGU) and below (EOGD) the pupil of the right eye. Electrode impedances were kept below 5 kΩ. All electrophysiological signals were digitized with a frequency of 250 Hz. A high cut- off fi lter of 30 Hz was used. The ERPs were fi ltered offl ine with a 10 Hz low pass for the plots, but all statistical analyses were computed using nonfi ltered

414

chapter eleven

data. During the visual inspection of the quality of the recorded data obtained for each participant, we decided to exclude the data of four participants because of the high number of artifacts. The grammatical sentences in Condition 1/Set A with imperfective main verbs and simultaneous converbs were correctly judged as ACCEPTABLE by 74.3  percent of the participants, and the ungrammatical sentences in Condition 1/Set A with imperfective main verbs and incorrectly formed simultaneous converbs were correctly judged as unacceptable by 73.5  percent. Interestingly, in the case of anterior converbs there were more erroneous answers. The grammatical sentences in Condition 1/Set B with perfective main verbs and anterior converbs were correctly judged as acceptable by 71.7 percent of the participants, and the ungrammatical sentences in Condition 2/Set B with incorrectly formed anterior converbs were correctly judged as unacceptable by 64.9  percent. It thus seems that the grammatical status of sentences with anterior converbs is less obvious to native speakers, a result that has been confi rmed by an additional acceptability rating study (as reported in section 5.2). As for the sentences with pragmatic/semantic mismatches, the sentences with imperfective main verbs and correctly formed anterior converbs in Condition 3/Set A (the inception condition) were judged as acceptable by 62.32 percent of the participants, while the sentences with perfective main verbs and correctly formed simultaneous converbs in Condition 3/Set B (the inclusion condition) were judged as acceptable by 83.13 percent. Thus it seems that the inclusion condition is more acceptable than the inception condition, a result that has been confirmed by an additional acceptability rating study (see section 5.2.). Note that the grammaticality status of sentences with an expected inceptive coercion seems to be less obvious to the native speakers (9.51  percent of whom responded “I don’t know”) than the grammaticality of sentences with inclusion (4.71  percent of whom responded “I don’t know”). We will return to these findings in our discussion of the ERP results in section 5.4.6. The behavioral results are summarized in Table 11.5.

5.4.4 BEHAVIORAL RESULTS.

For the statistical analysis we conducted an analysis of variance (ANOVA) for regions of interest (ROIs). Given the shortcomings of the repeated-measures ANOVA, we also used generalized additive modeling (GAM) as implemented by the MGCV package (see

5.4.5 ERP RESULTS.

Aspectual Coercion versus Blocking

415

table 11.5 Summary of the Behavioral Results Answers Obtained (Percent) Set

Condition

A

B

Acceptable

Unacceptable

I do not know

1. Control ✓ SIM converb

74.3

17.5

8.2

2. Morphological mismatch *SIM converb

18.7

73.5

7.8

3. Semantic or pragmatic mismatch inception imperfective + ✓ ANT converb

62.32

28.18

9.51

1. Control ✓ ANT converb

71.7

24.6

3.7

2. Morphological mismatch *ANT converb

28.9

64.9

6.2

3. Semantic or pragmatic mismatch inclusion perfective + ✓ SIM converb

83.13

12.7

4.17

Note: SIM = simultaneity; ANT = anteriority; ✓ = grammatical; * = ungrammatical

Wood 2011 for the GAM; Wood 2004 for the strictly additive GCV-based model method and basics of GAM; Wood 2006 for an overview; Wood 2003 for thin plate regression splines).7 For space reasons, we report here only the most relevant statistical results. 8 Regions of interest (ROIs). We defi ned the following ROIs: left anterior (F3, F7, FC1, FC5), central anterior (C3, C4, Cz, FCz, Fz), right anterior (F4, F8, FC2, FC6), left posterior (CP5, O1, P3, P7), central posterior (CPz, O2, POz, Pz), right posterior (CP6, O2, P4, P8), and midline (CPz, Cz, FCz, Fz, POz, Pz). The midline ROI was not relevant for further analysis. Mean voltages for single ROIs were calculated from the averages of all participants. The time windows for our analysis of the two morphological and two semantic/pragmatic (“inclusion” and “inception”) mismatches were selected on the basis of visual inspection of average wave forms. For each chosen time window, we performed a paired t-test of the mean voltages in all six ROIs. Results of Comparison 1 (Condition 2/Set A (morphological violation) vs. Condition 1/Set A (control)). Figure  11.6 shows a very strong positive-going component between 600 and 1,000 ms following word onset in all locations, with a peak at 800 ms. The effect is the strongest in the left, central, and right posterior ROIs. Given the positive deflection of the reported ERP waves and their characteristic posterior distribution,

416

chapter eleven

Figure 11.6

as well as their latency, the observed ERP pattern would correspond to the P600 component. There was a main effect of condition (F(1,18) = 29.25, p < 0.001) and ROI (F(5, 90) = 13.19, p < 0.001), and an interaction between condition and ROI (F(5, 90) = 16.39, p < 0.001). The statistical analysis of the P600 effect in the time window 600–1200 ms revealed an effect of condition in all ROIs: In the central anterior ROI (t(18) = –4.59, p < 0.001), in the right anterior ROI (t(18) = –3.99, p < 0.001), in the left posterior ROI (t(18) = –5.81, p < 0.001), in the central posterior ROI (t(18) = –7.32, p < 0.001), and in the right posterior ROI (t(18) = –6.49, p < 0.001). Results of Comparison 2 (Condition 2/Set B (morphological violation) vs. Condition 1/Set B (control)). In Figure  11.7 a positive-going component (P600) was present between 700 and  900 ms (post- onset) following word onset in the central posterior and right posterior ROIs. However, it was preceded by a negative component within 300–500 ms (post- onset). The negative component was the strongest in the central anterior and left anterior ROIs, though the effect was statistically significant also in the right anterior and left posterior ROIs. Given the negative deflection of the reported ERP waves peaking between 300 and 500 ms, as well as their characteristic bilateral distribution, the observed ERP pattern would most probably correspond to the N400 component. Regarding the time window 300–500 ms, there was a main effect of condition (F(1,19) = 7.53, p