Modality in Modern Greek [1 ed.] 9781443864077, 9781443816960

Citation preview

Modality in Modern Greek

Modality in Modern Greek By

Eleni Staraki

Modality in Modern Greek By Eleni Staraki This book first published 2017 Cambridge Scholars Publishing Lady Stephenson Library, Newcastle upon Tyne, NE6 2PA, UK British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library Copyright © 2017 by Eleni Staraki All rights for this book reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior permission of the copyright owner. ISBN (10): 1-4438-1696-5 ISBN (13): 978-1-4438-1696-0

To my parents, Constantinos and Evangelia.

TABLE OF CONTENTS

List of Diagrams ......................................................................................... xi List of Tables ............................................................................................. xii Abbreviations ........................................................................................... xiii Chapter One ................................................................................................. 1 Introduction 1 Introduction......................................................................................... 1 1.1 Modal Expressions in Greek ...................................................... 1 2 Linguistic Theories of Modality ......................................................... 3 2.1 Types of Modality ...................................................................... 4 2.2 Kratzer’s framework (1977, 1981, 1991b) ................................. 5 2.3 Portner (1997, 2007a, 2009) ..................................................... 11 3 Conclusion ........................................................................................ 14 Chapter Two .............................................................................................. 15 Modality in Embedded Clauses 1 Introduction ....................................................................................... 15 2 Tense and Aspect in Greek ............................................................... 16 2.1 Informal Approaches to Greek Tense and Aspect .................... 16 2.2 Aspect in Main Clauses ............................................................ 20 2.3 Aspect in Embedded Clauses ................................................... 24 3 Mood Choice..................................................................................... 25 3.1 Mood Choice in Complement Clauses ..................................... 27 3.2 Complementizer Selection ....................................................... 29 3.3 Complements of Modal Verbs ................................................. 30 4 A Framework for Greek Temporal System ....................................... 33 5 Conclusion ........................................................................................ 38 Chapter Three ............................................................................................ 40 Epistemic Modality 1 Introduction ....................................................................................... 40 2 Issues in Epistemic Modality ............................................................ 40 3 Modality and Speech Acts ................................................................ 41 3.1 Modality and Truth Content ..................................................... 44

viii

Table of Contents

3.2 Modality and Evidentiality ....................................................... 54 3.3 Truth Commitment ................................................................... 56 4 Types of Epistemic Modality in Greek ............................................. 57 4.1 Epistemic Possibility Verbs and Adverbs ................................ 58 5 Epistemic Necessity Operators: prepi and tha .................................. 60 6 Embeddability of the Greek Modals ................................................. 62 6.1 The Known Falsity Test ........................................................... 63 6.2 The Assent/Dissent Test ........................................................... 63 6.3 The Yes/No Interrogatives Test................................................ 65 6.4 The Embeddability Test ........................................................... 66 6.5 The Scope Test ......................................................................... 67 7 Complement Clauses ........................................................................ 69 8 The Protasis of the Conditional Constructions.................................. 71 9 Evidentiality in Greek Epistemic Modals ......................................... 72 9.1 The Evidential Component ....................................................... 73 9.2 The Types of Evidence ............................................................. 75 9.3 The Epistemic Commitment ..................................................... 83 10 Conclusions ..................................................................................... 86 Chapter Four .............................................................................................. 87 Modality and Time 1 Introduction ....................................................................................... 87 2 Current Theories ............................................................................... 93 2.1 Future: A Modal with a Tense-like Component ....................... 93 2.2 Future: Aspectual Distinctions ................................................. 95 2.3 Future: An Epistemic Necessity Modal .................................... 96 3 The Characteristics of the Future Reference ..................................... 99 3.1 Future Reference in Greek ..................................................... 100 3.2 Parameters of the Modal Interpretation .................................. 101 3.2.1 Modal Force ................................................................... 101 3.2.2 Modal Base .................................................................... 103 3.2.3 Ordering Source ............................................................. 106 4 The Branching Time Model ........................................................... 107 4.1 Gradience within Possible Worlds ......................................... 109 4.2 Pruning the Set of Possible Worlds ........................................ 110 5 Analysis .......................................................................................... 115 5.1 The Logical Structure of Future Reference ............................ 115 5.2 Determining the Universal Quantificational Domain ............. 116 5.3 Deriving the Readings of Future Reference ........................... 124 5.3.1 The Epistemic Future ..................................................... 124 5.3.2 The Temporal (predictive) Future.................................. 125

Modality in Modern Greek

ix

5.3.3 The Deontic Future ........................................................ 125 5.3.4 The Abilitative Future.................................................... 126 6 Conclusions..................................................................................... 126 Chapter Five ............................................................................................ 128 Priority Modality 1 Introduction ..................................................................................... 128 2 Issues in Priority Modality .............................................................. 129 3 Types of Priority Modality in Greek .............................................. 135 4 Functional and Formal Variation of Priority Modality ................... 136 4.1 Priority Modality with BORI / AS ......................................... 136 4.2 Priority Modality with PREPI / NA ....................................... 137 5 Imperatives...................................................................................... 138 6 Descriptive Priority Modality in Greek........................................... 139 7 The Argument Structure ................................................................. 141 7.1 The Syntax of Modal Verbs ................................................... 141 7.2 Modal Verbs: Control or Raising? ......................................... 143 7.3 Argument Structure of Modality: The Subject ....................... 150 7.4 Incorporating Causality ......................................................... 152 7.5 Modality as causation, force and action ................................. 155 8 Performativity and Priority Modality .............................................. 159 8.1 Imperatives and Priority Modality.......................................... 161 9 Capturing the Meaning of Priority Modality .................................. 163 9.1 Priority Modality as Updating TDL ....................................... 163 9.2 Priority Modality as Modalized Propositions ......................... 170 9.3 An Analysis of the Priority Modality ..................................... 173 9.4 Unembedded and Embedded Occurrences of Performativity ... 175 9.5 The role of Time ..................................................................... 177 9.6 The Allocation of Choice ....................................................... 178 10 Conclusions ................................................................................... 182 Chapter Six .............................................................................................. 184 Dynamic Modality 1 Introduction ..................................................................................... 184 2 Issues in Dynamic Modality ........................................................... 185 2.1 The Ability Puzzle: veridicality and effort ............................. 188 2.1.1 Actuality Entailment and Veridicality .......................... 188 2.1.2 Mood selection patterns ................................................ 189 2.1.3 Inference of Effort ........................................................ 190

x

Table of Contents

2.2 Aspectual Analyses of Actuality Entailments ........................ 191 2.2.1 Perfective and Veridicality ............................................ 191 2.2.2 Scope Ambiguity and Actuality Entailments ................ 192 2.3 Arguments against the Aspectual Hypothesis of AE.............. 195 2.3.1 Actuality Entailments ................................................... 195 2.3.2 Actuality Entailments: Not just Perfective Past ............ 196 2.4 Causative Co-ordinate Frames ............................................... 198 2.4.1 Co-ordinate Frames and Actuality Entailments ............ 198 2.4.2 Complementizer choice and Meaning Shifts ................ 201 3 Ability as modality and ability as force .......................................... 202 3.1 Ability as Modality................................................................. 202 3.2 Ability as Force ...................................................................... 204 4 Conclusions..................................................................................... 209

LIST OF DIAGRAMS

Diagram 1 Modal base and possible worlds ................................................ 7 Diagram 2 The Branching time model according to Thomason (1984) .... 111 Diagram 3 The Epistemic Tree ................................................................ 112 Diagram 4 The Temporal Tree ................................................................ 113 Diagram 5 The Deontic Tree ................................................................... 114 Diagram 6 The Ability Tree .................................................................... 115 Diagram 7 Epistemic and Predictive reading in quantitative terms ......... 120 Diagram 8 False in quantitative terms ..................................................... 121

LIST OF TABLES

Table 1 Greek Tense System: The traditional approach ........................... 17 Table 2 The number of tenses in Greek ..................................................... 20 Table 3 Past and Non-past opposition in the Greek tense system ............. 34 Table 4 Epistemic Modal Expressions in Greek ........................................ 57 Table 5 Truth Conditional Content in Main Sentences ............................. 66 Table 6 Truth Conditional Content in Embedded Sentences ..................... 69 Table 7 Truth Conditional Content in Embedded Sentences ..................... 71 Table 8 Types of Evidence in Greek ......................................................... 78

ABBREVIATIONS

INP PNP IP PP FUT SUBJ OPT PASS IMP NOM GEN PRF sg pl

imperfective non past perfective non past imperfective past perfective past future subjunctive optative passive imperative nominative genitive perfect singular plural

CHAPTER ONE INTRODUCTION

1 Introduction The goal of this book is to provide a formal analysis of Greek modal expressions of epistemic, priority (deontic) and dynamic sentential modality1 main or embedded. The study consists of two components: (1) an empirical investigation of modal expressions in Greek such as prepi ‘must’, bori ‘may’ and particles such as na ‘subjunctive’, tha ‘will’, and (2) a formal analysis of the syntactic and the semantic properties of these constructions with emphasis on comparison of different types of modal structures, internal to Greek as well as cross-linguistically.

1.1 Modal Expressions in Greek The first modal structure I investigate is the set of Greek modal particles, including na (subjunctive), and tha (future) which combine with verbs (see 2.34) and produce modal readings. For example: (1.1) a. Na / as efevgha SUBJ/OPT was-leaving.1sg I wish I should / could go Deontic wish: I consider my wish necessary b. tha efevgha .. FUT was-leaving.1sg I would leave .. Epistemic possibility: As far as I know, it was possible that I leave 1

Sentential modality conveys modal meaning at the level of the whole sentence, and it encompasses two main modal expressions such as (1) the modal auxiliary verbs including can, could, have to, may, might, must, needn’t, ought, should, and (2) the sentential adverbs including allegedly, apparently, certainly, maybe, perhaps, possibly, probably, supposedly.

Chapter One

2

The second modal structure I examine is the set of impersonal modal verbs, including prepi ‘it must’, bori ‘it is possible’ and the personal boro ‘I am able to’. All three take subjunctive na complements (Mackridge 1987; Holton et al (1997); Clairis & Babiniotis (2005), among many others). The latter two, bori and boro, are impersonal (3rd person singular) and personal variants of the same verb. The impersonal bori is an epistemic modal, whereas the personal boro ‘I am able to’ is abilitative or deontic, never epistemic. The modal verb prepi is a necessity modal. It is, either epistemic or deontic, and always impersonal: (1.2) a. * Ta The

pedhia prepun na trone children must.3pl.INP SUBJ eat.3pl.INP Children must eat fruits

b. Ta pedia The children

prepi na trone must.3pl.INP SUBJ eat.3pl.INP

fruta fruits

fruta fruits

Epistemic necessity: As far as I know, the children must eat fruits Deontic necessity: The requirements for a healthy diet dictate that it is necessary that children eat fruits (1.3) a. Ta pedia bori na fijun The children might.3sg.INP SUBJ leave.3pl.PNP Epistemic possibility: As far as I know, it is possible that children leave b. Ta pedia borun na The children can.3pl.INP SUBJ Ability: Children are able to leave. Deontic: The children are allowed to leave.

fighun leave.3pl.INP

One of the main claims that this study asserts is that modality is determined not only by context (see Kratzer 1977, 1981, 1991b; Portner 2007a, 2009; von Fintel & Gillies 2007b; among many others) but also by the speaker’s epistemic model (see a description of it in Giannakidou 1999). All modal expressions will be shown to be non-veridical, or expressing a non-commitment to truth (Giannakidou 1998). I adopt the thesis that modal verbs are non-veridical operators (a proposition embedding function, see Giannakidou 1999). I provide a modest extension of Giannakidou’s proposal on (non) veridicality and show its descriptive and explanatory power in describing sentential modality in Greek. Furthermore, my goal is to investigate the role of non veridicality and how we can employ it in explaining the dependency of perfective non-past (PNP) on non veridical particles and their embedding

Introduction

3

under non-veridical propositional operators: the modal operators. A detailed description follows in chapter 2. This book is organized as follows. In chapter 1, I describe the general empirical and theoretical background for the research; categories and theories of modality more generally based on English. In chapter 2, I present an investigation of clause structure and modal auxiliary verbs in Greek. More specifically, I provide an overview of (a) the tense and aspect system, (b) the mood (selection pattern), and (c) the clause structure introduced by the particles tha (future), and na (subjunctive), and the formal framework within which I will develop the analysis of Greek sentential modality. In the last section, I introduce modal auxiliary verbs within the clause structure (na – complement, interpretational ambiguity, and the tense and aspect selection pattern), and discuss some questions that determine the analysis of modality, and some that will form the empirical core of this study. I will also discuss how the study of modal structures in Greek in particular impacts general theories of modality. In chapter 3, I examine expressions of epistemic modality in Greek. In this section my main goal is to provide a discussion and formal analysis of epistemic modality, focusing on the following main themes: (a) the types of epistemic modality in Greek, (b) the contribution of epistemic modals to the truth conditional content of the proposition they are in, and (c) their evidential nature and characteristics. In chapter 4, I explore (a) the relation between epistemic modality and future reference. In chapter 5, I investigate the sub-varieties of priority modality: deontic, bouletic, and teleological. This part of the study is mainly dedicated to (a) the types of deontic modality, (b) the study of imperative and optative structures, and (c) the issue of argument structure of priority modals. In chapter 6, I examine the essential features and meaning of dynamic modality. The aim of this chapter is to discuss and analyze the nature of dynamic modality typically exemplified by modals of ability and disposition.

2 Linguistic Theories of Modality In this section, I present the frameworks I use in this study. The first subsection outlines the essential ideas formalized by Kratzer, and the parameters I consider important for the truth validation of modal expressions. Next, I present the work of Portner (1997, 2007a, 2009), who offered an extension of Kratzer’s work. Before I move onto the analysis of the frameworks I adopt, I briefly describe the different types of modality.

4

Chapter One

2.1 Types of Modality Portner (2009) introduced the following classifications of modality: epistemic, priority, and dynamic modality. Epistemic (Greek epistƝmƝ ‘knowledge’ and more general information) modality pertains to the speaker’s knowledge or lack thereof, and expresses the possibility of the necessity for a proposition to be true, given what is already known and based on the available evidence: (1.4) Epistemic Modality a. Mary may come to the party b. John must have a good reason for being late tonight c. We may possibly run into them at the concert

Priority modality includes deontic (Greek deon ‘obligation’), bouletic (Greek boulƝ ‘desire’), and teleological (Greek telos ‘goal’) modals (Portner 2009). It expresses a wide range of interpretations, of which the basic features are obligation, permission, exhortation, and optative, given particular goals and desires. As Portner (2009) put it, the term ‘priority’ indicates a possibility as better than, or as having a higher priority than, others: (1.5) Priority Modality a. You must obey the laws of your country deontic b. You should try this red lipstick bouletic c. You could add Tabasco to your soup for a spicy touch

teleological

Dynamic modality includes both volitional and quantificational modals, and conveys a wide range of meanings relevant to ability, disposition, and opportunity (Portner 2009). On the one hand, volitional modality relates to ways in which certain circumstances have an effect upon the actions of an individual’s deliberate intentions. On the other hand, quantificational modality has to do with universal or existential quantification over individuals: (1.6) Dynamic Modality a. Volitional (i) Mary can swim (ii) Mary can enjoy an excellent view from her terrace (iii) Mary will cry when she finds out her dog died

ability opportunity dispositional

Introduction

5

b. Quantificational (i) A typhoon can occur in the western Pacific (ii) A typhoon will occur in the western Pacific

existential universal

Modals also vary according to the kinds of the interpretation they obtain showing signs of lexicalization. For example, some English modals like might are only sensitive to a set of propositions relative to information and/or evidence: (1.7) a. Paul might be at the back yard b. # Citizens might obey the laws of their country

On the other hand, there are modals that can have both epistemic or priority interpretations. Consider the following examples: (1.8) a. Paul may be at the back yard b. You may use the dining hall after 5pm c. Dinosaurs must have died out suddenly d. You must be home by 11pm

epistemic possibility deontic possibility epistemic necessity deontic necessity

As it has been argued many times in the literature (see for example von Fintel & Gillies 2008a/b; among others), modals contain an evidential component, but this characteristic of modality clashes with the traditional account that argues for quantification over possible worlds, and does not capture the evidential meaning of modality. To capture this distinctive attribute of modality without abandoning a unified account, Portner (2007a, 2009) suggested that modals are not evidentials per se; instead modals provide an extra speech act in addition to their standard truthconditional contribution as quantifiers over possible worlds. Based on the work of Stalnaker (1974, 1978, 1987), Portner (2007a, 2009) suggested that modals are performative if they perform a speech act different from, or in addition to, the usual speech act of assertion. In this study, I adhere to the classification of modality as proposed by Portner (2009).

2.2 Kratzer’s framework (1977, 1981, 1991b) Modal expressions in languages from different families exhibit chameleonic properties allowing them to convey a multiplicity of meanings. Consider, for example, the English modal have to: (1.9) a. He has to be studying. b. Protesters have to evacuate the square. c. Mary, you have to wake up early.

epistemic deontic bouletic

6

Chapter One d. I’m running late. I have to go, bye! e. CTA is on strike. You have to take a cab.

circumstantial teleological

The central claim of semantic theories until the mid-seventies was that modals were ambiguous expressions (see for example, Groenendijk & Stokhof 1975; Kamp 1975). The meaning of modal expressions was modeled as an inherent part of its lexical definition. However, lexical ambiguity would proliferate the lexicon when, in fact, it would be more natural to narrow down the multiplicity of modal meaning as context dependent. This proposal was made by Kratzer in a sequel to her seminal papers. Kratzer (1977, 1981, 1991b) innovated the traditional modal analysis by arguing that modals are context-dependent and developed a system in which modals are defined with respect to three basic parameters, thus making them relative to: (1) the modal force, (2) the modal base, and (3) the ordering source. Kratzer kept the first parameter from the modal logic tradition: possibility modals are existential quantifiers, and necessity modals are universal quantifiers over possible worlds. Kratzer’s novel contributions to the analysis of modality were parameters (2) and (3), which derive the meaning of modal expressions via context and capture the context sensitivity of modals. In what follows, I present the three dimensions of Kratzer’s modality theory, which I adopt in this study. As stated above, modals are quantifiers over possible worlds (Kripke 1963; Lewis 1973; Kratzer 1981, 1991b, a.o.), and distinguish existential quantification from universal quantification. For example: (1.10) a. Children may read at the library w (‘Children read at the library’ is true in w) b. Children must read at the library w (‘Children read at the library’ is true in w)

In example (1.10a), I state, formally, that there is a possible world among those compatible to my knowledge, in which children read at the library. May, as a possibility modal, existentially quantifies over possible worlds. In example (1.10b), the necessity modal, the verb must, quantifies universally over all those possible worlds. This is the modal force, and according to Kratzer, it is lexically determined. Modal base, the second essential parameter, determines the kind of modality, as bare modal force does not capture the modal expression’s true meaning. For Kratzer, the interpretations of the modals depend on two types of conversational backgrounds that are provided by the context. The modals can be specified with an in view of-phrase corresponding to sets of

Introduction

7

propositions of information, rules, permissions, facts, evidence, etc. In this case, the conversational background at work will be one of the following: (1.11) a. [In view of what I know,] children may read at the library Ȝw. Ȝp. p is one of the propositions that I know in w b. [In view of school’s rules,] children must read at the library Ȝw. Ȝp. p is one of the propositions that are commanded in w

The first conversational background gives us the possible worlds of the modal base (MB) (the accessible worlds). According to Kratzer, this type of conversational background consists of many different possibilities (realistic, totally realistic, epistemic, stereotypical, deontic, and empty, to mention a few), and contributes the premises, facts, and ideals, from which conclusions and inferences are drawn. Thus, the conversational background is that entity provided by utterances such as what the law dictates (deontic), what the facts are (realistic), what the case is (totally realistic), what is known (epistemic), what is normal (stereotypical), what is commanded (deontic), what is desirable (bouletic), etc. Conversational backgrounds differ from one possible world to another, and they represent a set of propositions in a possible world. For example, the utterance what the law dictates (deontic) provides a set of propositions in a possible world, and it can be represented as a function from the set of possible worlds W into the power set of the power set of W, which assigns to any world w of W the set of all propositions, which are commanded in w (Kratzer 1981). In other words, conversational background is the function ŀf(w) that assigns sets of propositions to possible worlds (see next diagram 1). Where f = conversational background w = actual world W = set of possible worlds

Diagram 1

8

Chapter One

According to Kratzer (1981), the different kinds of conversational backgrounds fall into two main types, epistemic and deontic, and can be represented formally, as follows (following Kratzer 1981): (1.12) a. Epistemic conversational background: The function f which assigns sets of propositions to members of W such that for all possible worlds w  W: f(w) contains all those propositions p which are established knowledge in w. b. Deontic conversational background: The function f which assigns sets of propositions to members of W such that for any possible worlds w  W: f(w) contains all those propositions p such that is commanded in w that p.

Additionally, consider the contrast between the two modal bases in the following examples: (1.13) a. Children might eat fruits and vegetables b. Children must eat fruits and vegetables

Example (1.13a) is evaluated with respect to an epistemic modal base, in which, according to the speaker’s knowledge, it is possible that children eat fruit and vegetables. In contrast, example (1.13b) is evaluated with respect to a deontic modal base, which identifies all those accessible, possible worlds in which certain circumstances hold, for example, the quality of food, the prevention of life-threatening diseases, etc. The difference between epistemic and circumstantial (deontic) modal bases is also evident in the different truth conditions that yield for the examples in (1.13). If children do not eat fruits and vegetables, example (1.13a) will be false, but example (1.13b) will be true because the circumstances are not affected; it still holds in a possible world that children are required to consume fruits and vegetables. The third determining factor for modal meaning is the ordering source. Remember that an accessibility relation essentially connects a world w to a world w’ if, and only if, every situation p that holds in w is possible in w’. However, this definition does not determine under what proposition the accessibility relation is true in w’. Kratzer (1981) associated the conversational background with accessibility relation, as follows: (1.14) If f is a conversational background, the set of worlds W that are accessible in a world w with respect to f is simply the intersection of all possible worlds ŀf(w). That is the set of worlds where all propositions of f(w) are true.

Introduction

9

Hence, according to the conversational background, Kratzer introduced the ordering source, which is the second conversational background and whose role is to order the worlds of the modal base according to some contextual ranking. The concept of the ordering source dates back to Lewis (1981), who argued that we might think of conversational backgrounds as an ordering of possible worlds. Given the facts, ideals, understandings, and attitudes of a context, some facts have greater import than others; thus, we can say that some possible worlds fit the facts of the ideal set of possible worlds A better than others do. Similarly, some worlds differ so much from A that they should be ignored, and possibly omitted altogether from an ordering source. Kratzer (1981), crediting Lewis (1981), defined the ordering source as follows: (1.15) Ordering Source ”g(w): For w and z such that w, z  W: w ” g(w) z iff {p: p  g(w) and z  p} Ž {p: p  g(w) and w  p} “For all worlds w and z, a world w is at least as close to the ideal set of possible worlds g(w) as a world z iff all propositions of g(w) which are true in z, are true in w as well”

The formulation that Kratzer (1981) gave in definition (1.15), orders possible worlds by invoking an unordered set of propositions: an unordered set of propositions p that are true in z, and an unordered set of propositions p that are true in w. The only condition that should be preserved for an ordering is the comparativeness of truth among possible worlds. If two possible worlds have at least as many propositions true between them (i.e., a similar sum of true propositions) then they are considered to be close to the ideal represented by A. In contrast, to the first type of conversational background, i.e., the modal base, an ordering source remains opaque and has to be recovered from the context. Let us examine an example: (1.16) Children must eat fruits and vegetables.

Example (1.16) refers to an ideal situation in which children consume fruits and vegetables. In this case, the modal base defines a set of worlds, fdeontic(w), and each of these worlds expresses the content of the requirement in the base world w. For example, with no exception, children follow the instructions for a healthy diet and eat fruits and vegetables. As is, the modal quantifies only over worlds where the requirements of a healthy diet are fulfilled. In all of the worlds in which the requirements are fulfilled, children consume fruits and vegetables. Facts and ideals are

Chapter One

10

treated in a similar way. However, our world is far from an ideal one, and, our goal is the modal to quantify over all those worlds where the requirements of a healthy diet are not fulfilled. Imagine a case, as is often true, in which children consume hamburgers instead of fruits and vegetables, and gain weight. Then consider the following requirement: (1.17)

Children must consult a nutritionist.

However, above, we excluded all of those worlds in which there was a violation of the requirement for a healthy diet. Thus, we cannot consider the proposition in example (1.17) because there is no violation of the requirements, even though children eat hamburgers. Now, for the modal to quantify over those worlds where violations of the requirement hold, we should include all those relevant facts on the base world w, and, according to Kratzer, order them with respect to an ideal set of possible worlds g(w) (see definition in 1.15). Thus, the ordering source will order all of those relevant worlds, starting from the best possible world in which children fulfill the requirements of a healthy diet, and end with the possible world in which children violate the requirements and consult a nutritionist. Consider the worlds w and z in which children follow an unhealthy diet, and where children go to a nutritionist in w, but not in z; consider also a deontic ordering source g orders the set of worlds: unhealthy eating results in undesirable consequences and those on an unhealthy eating go to a nutritionist. Worlds w and z violate the first proposition in g(w), but w is closer to the ideal ordering set of propositions induced by the ordering source g(w) than z, for the reason that in w, the children eating an unhealthy diet go to a nutritionist, but not in z. To sum up, Kratzer developed a theory of modality in which the contribution of the context plays an essential role in determining two of the dimensions of a modal: modal base f and ordering source g. Modal force determines whether the quantification is existential or universal. Modal base determines the set of worlds in which a modal quantifies over, and ordering source orders a set of worlds with respect to an ideal set of possible worlds g(w). I adopt the framework suggested by Kratzer (1977, 1981, 1991b), and I build on her assumption of the relativity of modality, and the ordering of worlds. Kratzer suggested that modality is a context-dependent linguistic phenomenon. When we analyze modality, the context plays a crucial role.

Introduction

11

2.3 Portner (1997, 2007a, 2009) Portner, in a series of papers, offered an updated version of Kratzer’s standard theory on modality, and offered explanations for certain puzzling issues that Kratzer’s unifying account overlooked. In this section, I present his theoretical framework (Best operator and modality as speech acts) and the classification system of modality that he introduced. I will consider Portner’s contribution in explaining and formalizing modality as an essential part of Kratzer’s framework on modality, thus following his main ideas and formalizations. In the previous section, I discussed conversational background, modal base, and ordering source, as the key factors that determine the type of modality, and demonstrated how Kratzer explained the various kinds of interpretations without postulating massive lexical ambiguity. In the first type of conversational background, the modal base f contributes the accessible worlds, and in the second type, the ordering source g ranks the accessible worlds with regard to how close they satisfy an ideal given by g (see 1.17). Portner (1998, 2009) argued that there are two puzzling issues in ordering semantics, as offered by Kratzer. The first issue is that a conversational background defines only a partial ordering2 dg(w) on the set of possible worlds g(w), and this implies that not every pair of worlds need be related, and that for some pairs of worlds it may be that neither is better than the other in the set of possible worlds. Let us consider an example, and assume a conversational background g, a world w, and a set of worlds {p, q, r}. The conversational background g induces a partial ordering dg(w) where {p, q, r} are the “best” worlds, {p, q}, {q, r}, and {p, r} worlds are not as good as {p, q, r} and are incomparable to each other, which means that a {p, q} world is not in any ordering relation with the other two sets of worlds. Finally, the worlds {p}, {q}, and {r} are the worst among the worlds. The formalization adapted from Portner (2009: 64) follows: (1.18) For any set of propositions P and any worlds w, v: w dP v iff for all p  S, if v  p, then w  p

2

Partial ordering is a binary relationship “d” over the set P which is reflexive, antisymmetric and transitive: x Reflexivity: ada x Antisymmetry: if a d b and b d a then a = b x Transitivity: if a d b and b d c then a d c

Chapter One

12

The definition in (1.18) states that any proposition in P that is true in v is true in w as well, and because w dS w, the order is reflexive (see footnote 2); w is either highly ranked or equally ranked as v (Portner 2009). Portner (1998, 2009) also noted that the ordering can be more complex when there are multiple sets of “best” worlds in which we have a set of worlds g(w) = {p, q, r} and {p, q} are incompatible. The incompatibility of {p, q} creates two sets of “best” worlds as we can have the ordering wpr dg(w) wp and wqr dg(w) wp but not wpq dg(w) wp. Thus, the partial ordering (a reflexive, antisymmetric, transitive binary relation) that the conversational background induces on a set of propositions complexifies the definitions of modal operators. Another issue in the ordering develops when the set of possible worlds P is infinite, which implies that there is always a better world for every possible world in the set. Assume a scenario in which you always want to succeed in every endeavor, and regardless of how successful you imagine yourself to be, you accept nothing other than success. Then, the bouletic ordering will look like this: (1.19) g(w) = a. p1 = “I want to succeed at least once in my life” b. p2 = “I want to succeed at least twice in my life” c. … d. pn = “I want to succeed at least n times”

According to Portner (1998, 2009) this ordering contains no worlds in which failure is the worst world, but only infinite cases of successful worlds, and he suggested an ordering such as the following: (1.20) … dg(w) wn dg(w) … dg(w) wp2 dg(w) wp1 dg(w) w

In this ordering there are no “best” worlds and this is a case of when we want to order a defined set of worlds. These issues indicate that modals are open to two or even more interpretations that can be paradoxical or trivial, and that the mechanism of the conversational background (ordering g) does not limit the readings to less contradictory ones. Thus, our goal is to find a mechanism with which we can determine the relevant background facts for the interpretation of modals while avoiding vagueness and indeterminacy in what is relevant. To simplify things, Portner (1998, 2009), crediting Stalnaker (1984, 1987), employed the Limit Assumption, a concept that originated in the treatment of counterfactual conditionals (Lewis 1973; Pollock 1976;

Introduction

13

Herzberger 1979; Warmbrǀd 1982). According to the Limit Assumption the following statement is true (adapted from Warmbrǀd 1982): (1.21) For every world w there is at least one accessible world w closest to the ideal.

Portner (2009) called these worlds Best (f (w), g (w)). The Best operator selects the most ideal worlds from the set of worlds P, given the ordering provided by g(w). Portner argued that there are two pragmatic relations in the treatment of propositions: the common ground (CG) which is the set of propositions that participants of a conversation mutually agree to treat as true, and the common propositional space (CPS) which is the set of propositions that the participants in a conversation are mutually aware of. The CG is a subset of the CPS: (1.22) CG Ž CPS

For example, consider the following scenario where you and your friend Maria talk about John, a common friend of yours, and a possible visit to him. Then, the CPS set includes the following propositions that the participants, you and Maria are aware of: it is a holiday, the weather is bad, and John usually chooses to spend holidays indoors when the weather is bad. The CG set includes the following modal propositions which are also assertions that the participants mutually agree to treat as true: John might be happy if he spends time with us, we might catch up with John at his house on a bad day like this. Thus, the modal expressions belong to the wider set of the CPS (see 1.22). But there is more to the assertion of a modal expression. The assertion of a modal structure of type M(ij) performs two speech acts. First, the modal proposition M(ij) that is added to the CG contributes truth-conditional content to the proposition it is in. Second, the prejacent proposition ij is added to the CPS and is considered to serve an evidentiallike function (see chapter 3). This function imposes restrictions on a modal’s conversational background and determines the ordering source. This is done via a selection function cg (common ground) (adapted from Portner 2007a, 2009): (1.23) Where \ = unmodalized proposition (the prejacent), and, I = modal proposition, then ¢CPS, cg² + M(ij) = ¢CPS’, cg’², where CPS’ = CPS ‫ۥ\ۤ{ ׫‬c, f, g} ‫ۤ{ ׫‬ij‫ۥ‬c, f, g} cg’ = cg ‫ۤ{ ׫‬ij‫ۥ‬c, f, g}}²}

14

Chapter One

A conversation consists of the pair CPS and the selection function cg, which selects those members of the CPS that are mutually presupposed by the participants. When a modal proposition is contributed to the conversation it updates the CPS and the cg: ¢CPS, cg² + M(ij) = ¢CPS’, cg’². Specifically, the CPS’ is the union of two propositions: the modal proposition {ۤij‫ۥ‬c, f, g} (an assertion added to the CPS) and the unmodalized proposition (the prejacent) {ۤ\‫ۥ‬c, f, g}. The updated cg’ consists of the cg and the modalized proposition {ۤij‫ۥ‬c, f, g}. In other words, there are two propositions that are added to the CPS: an assertion (contributing a speech act) and a modal one. According to Portner (2007a): “An utterance which adds one or more propositions to the CPS is felicitous only if at least one of these propositions is a complete or partial answer to a question under discussion. Such a proposition is Relevant.”As we are going to see in chapter 3 in more detail Portner’s account of modality as contributing speech acts explains the intrinsic features of modality.

3 Conclusions In this section, I presented the modal structures of Greek that I will investigate, and the thesis I adopt. The first modal structure I will investigate is the set of Greek modal particles including na (subjunctive), an (conditional), as (optative), and 6a (future) which combine with verbs and produce modal readings. The second modal structure I examine is the set of impersonal modal verbs such as prepi ‘it must’, bori ‘it is possible’ and boro ‘I am able to’. In this book, I adhere to the thesis that modal verbs are non-veridical operators. The framework in which I develop my thesis is that of Kratzer (1977, 1981, 1991b), and Portner (1997, 2007a, 2009). I analyze modality in Greek, considering that context plays a crucial role in determining its meaning.

CHAPTER TWO MODALITY IN EMBEDDED CLAUSES

1 Introduction The purpose of this chapter is to present the tense and aspect system, and explain essential characteristics of mood and complementizer selection of embedded clauses in Greek. I consider sentential modality, the realization of modal meaning hosted above the level of predicates at IP or CP (Portner 2009). First, I focus on the modal particles na (subjunctive), an (conditional), as (optative), and tha (future) that express modality at the level of MoodP (main and embedded clauses) and Now-TP (main clauses) (see Giannakidou 2009). Second, I examine the semantic and syntactic properties of impersonal modal prepi ‘it must’ and bori ‘it is possible’, and personal boro ‘I am able to’ verbs that select complement clauses introduced by the particle na. I will propose that both cases, modal particles and modal auxiliary verbs in Greek, come under the notion of non-veridicality (Giannakidou 1998, 2009), and that the role of the context, as argued by Kratzer (1977, 1981, 1991b) and Portner (1997, 2007a, 2009), is reinforced by those two cases of Greek modality, providing arguments against an analysis of modality inflated with unnecessary ontology of variables (i.e. tense), and against interpreting modality based on scopal (syntactic) configurations. The chapter is organized as follows. In section 2, I present the temporal and aspectual system of Greek. In section 3, I present the mood selection patterns in main and complement clauses in Greek, and define formally the notion of non-veridicality and how that regulates the mood selection in complement sentences. I introduce the reader to the basics of syntactic complementation and present the set of impersonal and personal modal auxiliary verbs in Greek and the role of non-veridicality in the selection of the complement clause introduced by particle na. In section 4, I discuss the formal assumptions and details of the tense system I adopt for this study. Section 5 concludes the chapter.

Chapter Two

16

2 Tense and Aspect in Greek In order to investigate modal verbs in Greek, it is necessary to understand the specific contribution of tense and aspect in the language. Hence, the purpose of this section is to present the temporal and aspectual system of the language, and in continuation with the previous discussion on moods and complementation, to review the major analytical approaches on tense and aspect.

2.1 Informal Approaches to Greek Tense and Aspect Tense is the grammatical category that places a situation in time, alternatively in a temporal dimension, and indicates when the situation takes place. In Greek, there are two morphologically distinct temporal dimensions: past and non-past (see Giannakidou 2009, among others). According to the past and non-past distinction, we have the following aspectual classes (cf. Mackridge 1985; Holton et al. 1997): (2.1) a. graf -o (INP)1 write.impf-1sg.nonpast ‘I write’ (habitual) ‘I am writing’ (progressive) c. e- graf –a (IP) past-write.impf-1sg.past ‘I used to write’ ‘I was writing’

b. grap -s -o (PNP) write.perf-1sg.nonpast [no English equivalent]

d. e- grap- s- a (PP) past-write.perf-1sg.past ‘I wrote’

Past affixes indicate those eventualities located in a past time, while non-past affixes indicate those eventualities located in a present or future time. Future reference is attributed with a distinct temporal dimension (it refers to a future time). However, morphologically, future tense lacks distinct verbal affixes, and only the combinations of the particle tha ‘will’ yield a futurate meaning. Thus, the temporal dimensions encompassing the Greek tense system in terms of morphology are past and non-past. I will not conflate the term ‘tense’ with qualities such as uncertainty, frequency, completion, duration, possibility, or whether information derives from experience or hearsay (evidentiality). These qualities, although present and sometimes difficult to distinguish morphologically or syntactically from 1 INP = Imperfective nonpast, IP = Imperfective past, PNP = Perfective nonpast, PP = Perfective past.

Modality in Embedded Clauses

17

tense proper, are independent notions and should remain as such. In the next sections, I give a full argumentation of my thesis as described above. There is a consensus for the number of tenses in Greek: they are eight, all traditional grammars (Tzartzanos 1928; Triandafillidis 1938; Tsopanakis 1994) state. Greek tense system consists of the following tenses: Present ghraf-o I write / I am writing

Past Progressive e-ghraf-a I was writing

Future Progressive tha ghraf-o I will be writing

Simple Future tha ghraps-o I will write

Past (Aorist) e-ghraps-a I wrote

Present Perfect eho ghraps-i I have written

Past Perfect iha ghraps-i I had written

Future Perfect tha eho ghraps-i I will have written

Table 1 Greek Tense System: The traditional approach Although it sounds trivial, the number of tenses in Greek may be disputed, because there are several issues that we need to find a satisfactory answer to. The questions that arise include, whether finding the number of tenses is an important matter, what the criteria are in identifying a tense, and, finally, whether we should consider tense, aspect, and mood as distinct or intertwined categories. The first question seeks an answer to whether certain syntactic structures with particles such as tha should be acknowledged as a tense or a modal, or both. The first question, moreover, needs an answer with regard to the status (questions such as: dependent vs. non-dependent, tensed vs. non-tensed) of a much-disputed verb form: the perfective non-past (PNP). The second question defines the terms of answering the first question. The third question is very clear; it tackles the issue of interaction of tense, aspect, and mood. Traditional grammars agree on the number and the naming of the tenses (see Table 1), and do not tackle the above-mentioned questions. The next three approaches on the Greek tense system I review acknowledge the

Chapter Two

18

distinction in past and non-past introduced earlier, and give a different accommodation of tenses. In particular, Mackridge (1987) argued that the Greek verb system could be easily described by two distinctive features: tense and aspect. Tense locates an event in time, and aspect encodes the internal structure of the event. Mackridge (1987) pointed to the importance of the two-fold tense-aspect system and presented a temporal configuration, obtaining the six (6) tenses2 of Greek. (2.2) a. Imperfective non-past b. Perfective non-past c. Perfect non-past d. Imperfective past e. Perfective past f. Perfect past

(INP or present) (PNP or aorist subjunctive) (PRNP or present perfect) (IP or imperfect) (PP or aorist / simple past) (PRP or past perfect / pluperfect)

Mackridge (1987) noticed, however, that the Greek verb system could be easily described without the third parameter, i.e., the perfect aspect (PRF). Since perfect tenses are formed periphrastically, we can do away with the use of PRF, which is the combination of the auxiliary verb eȤo “I have” in imperfective non-past (INP) and the dependent verb form PNP. Although perfect tenses are periphrastic tenses, the tense configuration does not say anything about the Greek future tenses that are also periphrastic. The tense configuration proposed by Mackridge (1987) lacks equality of distribution, as some periphrastic tenses are part of the configuration, and some are not. At the same time, PNP is represented as an independent tense form, which is clearly not the case, as PNP has to combine with some particle in the structure (na, as, tha) in order to render it grammatical. Holton et al. (1997:109) pointed out, consistent with the work of Mackridge (1987), that the Greek verb system only distinguishes the two forms of past and non-past. The reason behind this distinction is that the future and perfect tenses are formed periphrastically using the particle tha ‘will’ and the perfect forms eȤo “I have” and iȤa “I had”, and the perfective non-past form, respectively. According to Holton et al. (1997:224-233), there are eleven tenses3; four are tenses proper (present, 2

The parentheses contain the names of tenses we find in descriptive grammars. Also, with the exception of (2.24b) which is a dependent verb form the rest can be used without a particle or auxiliary verb. 3 The PNP (perfective non-past) does not count as tense in their tense system; however, it is included in the chart.

Modality in Embedded Clauses

19

past continuous, present and past perfect) and seven are tenses with salient modal meanings (simple past, progressive future, future simple, conditional progressive, conditional simple, future perfect, and perfect conditional. Clairis and Babiniotis (2005) argued that grammatical tenses indicate the temporal dimension of an eventuality (past, present, future), but this is only achieved by indicative verb forms, contrary to the modal verb forms, which have a distinct modal meaning (possibility, wish, obligation, etc.) and are formed with the use of a particle (na, as, tha) or with a verbal suffix (imperative). The future, Clairis and Babiniotis (2005) argued, denotes both modality and tense. The argumentation of Tsangalidis (2002a/b) is interesting in terms of its methodological grounds. By establishing a set of criteria, he tried to distinguish which verb forms are temporal and identify the status of PNP. For Tsangalidis (2002a) a verb form is temporal when it features paremphatic morphology. A second requirement is that the verb form should combine with non-temporal particles or clitics in the structure. The third requirement is that the verb form distribution is the same as that of the rest of the tenses. PNP is a paremphatic verb form, as it is accompanied by particles and clitics, and has the same distribution as the rest of the tenses. Therefore, Tsangalidis (2002a/b) concluded that PNP features temporality; however, he did not provide any further details on what exactly is the temporal contribution of PNP. However, Tsangalidis raised more issues that just the temporality of the PNP. Tsangalidis (2002a/b, 2011) supporting his thesis that PNP is not a defective verb (contra Giannakidou 2009) form lacking temporal features, added two more criteria. These criteria target two verb forms: the perfective and the non-past. Let us consider the argumentation examining a verb in its PNP form: aJapiso ‘to love’. Tsangalidis (2002b) argued that PNP is perfective because it consists of the past simple (aorist) stem: aJapis-, which is typically used to describe completed actions. Tsangalidis’ (2002b) second argument for the perfective nature of PNP4 is that this verb form is usually used to express the opposition between imperfective vs. perfective: tha aJapo ‘I will be loving’ vs. tha aJapiso ‘I will love’. The PNP is a perfective form because, and this is the third argument of Tsangalidis, it is not selected by verbs such as arhizo ‘start’ or stamato ‘stop’, which typically select complements in the imperfective aspect. Tsangalidis (2002b) presented more arguments in favor of PNP temporality. First, the PNP aJapiso does not express a past time; on the 4

Giannakidou (2009) argues in favor of the perfective nature of the PNP’s as well.

Chapter Two

20

contrary, it indicates a futurate interval. Second, it employs the same verbal suffixes of the present tense in Greek. Third, PNP never refers to utterance time (for an opposite view, see Giannakidou 2009). Finally, PNP is used to express temporal opposition, for example, an aJapiso ‘if I love vs. an aJapisa ‘if I loved’. For these reasons, Tsangalidis (2002a/b, 2011) suggested that we consider PNP as a proper member of the tense system of Greek, and not a defective and dependent verb form. The tense systems as presented above can be criticized on several methodological grounds. To begin with, the number of tenses varies, as does the working definition of tense. The reason for the first discrepancy is the role of PNP (perfective non-past). Regardless of its representation in the charts with tenses, it is considered as a dependent verb form (lacking temporality) that renders itself functional in a structure only when it combines with a particle. The result is that the number of tenses in Greek varies depending on the status of PNP and the role of particles. The reason for the second discrepancy seems to be the lack of consensus that in Greek, tense, aspect, and modality are conflated categories. The consequences of the methodological discrepancies are summarized in the table 2 below showing the two major approaches to the Greek temporal system. For the first approach, there are only tenses (traditional grammars, Mackridge 1987), and, for the second approach, there are tenses and tenses with inherent modal features (Holton 1997; Clairis & Babiniotis 2005): Theories Traditional Mackridge (1987) Holton et al (1997) Clairis & Babiniotis (2005)

Tenses 8 5 5 5

Tense and Modality

6 18

Table 2 The number of tenses in Greek

2.2 Aspect in Main Clauses In this section, I review the aspectual system of Greek in terms of the imperfective vs. perfective (see Giannakidou 2009; among others). Comrie (1976) argued that, in general, aspects are different ways of viewing the internal temporal constituency of a situation. More specifically, the perfective is the aspectual distinction that indicates that a situation is viewed as a single whole, and the imperfective is the aspectual distinction that refers to the internal structure of a situation, viewing a situation from

Modality in Embedded Clauses

21

within (Comrie 1976; among many others). Aspect does not indicate the temporal dimension of an eventuality, but rather, it gives either an internal view of a situation, how a situation evolves or relates in time (with other eventualities), or an external view of a situation, how a situation is posited in time or in relation to other eventualities. The temporal dimension belongs to the grammatical category of tense. Temporal relations or views belong to the grammatical category of aspect. The Greek aspects, perfective and imperfective, are marked by the absence or presence of distinct verbal infixes. The perfective aspect is recognized by the presence of the infixed -s- (sigmatic perfective) in the verb, along with a sort of phonological change with regard to Greek language phonology, while the imperfective aspect is recognized by the absence thereof. For example: (2.3) a. krivo o krip – s – o b. ghrafo o ghrap – s – o

However, the infixation pattern is not systematic and a few other morphological and phonological changes usually occur as a means of distinguishing perfective and imperfective. These morphophonemic alternations include the following: (2.4) a. Sigmatic perfective (default) krivo ‘to hide’ o krip – s – o b. Asigmatic perfective pleno ‘to wash’ o pl – i – no c. Suppletion troo ‘to eat’ o efagha

The examples (not an exhaustive list) provided above strongly indicate that the stems come inherently specified for the perfective and that it is not only the infixation of -s- that contributes to the aspectual distinction (see Tsimpli 1992; Xydopoulos 1996; and against this view Rivero 1990, 1992a). Holton et al. (1997) along with some traditional grammars (Triandafillidis 1941; Tzartzanos 1946), acknowledged three grammatical aspects: perfective, imperfective, and perfect. As demonstrated earlier, the first two constitute a contrastive pair with designated verbal infixes or grammar processes (suppletion). The contrastive distinction, perfective vs. imperfective, is present in all Greek tenses. The perfect is formed periphrastically with the use of the auxiliary verb eȤo ‘I have’ / iȤa ‘I had’ and the perfective non-past (PNP) verb form.

Chapter Two

22

A characteristic of perfective (P) in Greek is that it does not indicate specifically momentary or punctual situations; instead, it reduces the eventuality into a single point: (2.5) Eklisa ta parathira ke klidosa tis Closed.1sg.PP the windows and locked.1sg.PP the I closed the windows and locked the doors

portes doors

It cannot possibly be that the situations denoted were two momentary and punctual situations. Instead, we can view the situations as a single point (see Comrie 1976; and for Greek Mackridge 1987, Clairis & Babiniotis 2005). The closing of windows and the locking of doors are situations with internal complexity, and phases of identical eventualities; however, we can view them as having no internal structure. Comrie (1976) argued that this could denote the (possible) onset of a situation. This claim holds in Greek, as shown by the following examples: (2.6) a. Agapise ton Pavlo prin djo Loved.3sg.PP the Pavlo ago two She fell in love with Pavlos two years ago b. Tha aghapisi ton Pavlo FUT love.3sgPNP the Pavlo She will fall in love with Pavlos

Ȥronia years

Example (2.6a) focuses on the single point of the beginning of someone’s love towards Pavlos. Example (2.6b) focuses on the single point of the beginning of someone’s love that has not started yet, but is probably anticipated. Thus, I agree with Comrie (1976) that the inchoative verb meaning depends on the context and not on the inherent meaning of the verb. Comrie (1976) argued that there are two essential functions of the imperfective: habituality and continuity. Continuity is further distinguished between two functions: non-continuity and progressiveness. This aspectual generalization is also found in the Greek imperfective, and context provides information determining the distinction between imperfective functions (habitual, continuous, generalization, or progressive eventuality), as traditional grammars also attest (Mackridge 1987; Holton et al 1997; Clairis & Babiniotis 2005). For example (cited from Holton et al 1997: 219): (2.7) a. Ti kani What do.3sg.INP What is Nikos doing?

o the

Nikos; Nikos

Modality in Embedded Clauses

23

b. Troi eat.3sg.INP He is eating. c. Dhidhaski teach.3sg.INP He is teaching. d. Dhidhaski ghlosologia teaches.3sg.INP linguistics He teaches linguistics.

The answer to the question in example (2.7b) means that Nikos is eating (now), and as Holton et al. (1997:219) noted, the progressive interpretation is the default one, since we cannot interpret eating as an automatic pattern of behavior that might have been inherited or acquired through frequent repetition (habitual). Although there is ambiguity in example (2.7c) the answer can either mean that Nikos is teaching (now) or that he is a professor. The answer in example (2.7d) is unambiguous, because it indicates the recurring event of teaching linguistics. Therefore, Greek does not distinguish between habituality and continuity with explicit morphology. The concept of habituality involves a distinguishing quality that extends over a period of time. As Comrie (1976) explained “…the situation referred to is viewed not as an incidental property of the moment but, precisely, as a characteristic feature of a whole period”. In Greek, adverbs expressing the stretching of a situation over time are employed to determine the nature of a situation that extends over time and is not mere repetition. For example: (2.8) a. Djavazi poli sihna / kathimerina Read.3sg.INP very often / daily He reads very often/daily b. Mu ghrafi email kathe mera/kathe vdhomadha me write.3sg.INP email every day/every week He writes emails to me every day/every week

Imperfective can be used to indicate a state of affairs or a characteristic or the lack thereof. For example: (2.9) a. Afto to pedi ine This the kid is.3sg.INP This kid has a heart of gold b. To molivi dhe ghrafi the pencil not write.3sg.INP The pencil does not write

malama gold

24

Chapter Two

Therefore, progressive distinction is not indicated by explicit morphology in Greek (see 2.8a-b).

2.3 Aspect in Embedded Clauses Aspect in embedded clauses can be either perfective or imperfective, denoting a momentary and/or a punctual situation (2.10a), a habitual, continuous generalization or a progressive eventuality (2.10b): (2.10) a. Thelo na want.1sg.INP SUBJ I want to talk to you b. Thelo na want. 1sg.INP SUBJ I want to be talking to you

su you

miliso talk.1sg.PNP

su you

milao talk.1sg.INP

However, there are certain restrictions with verbs that denote the onset or the result of an eventuality. Specifically, the imperfective aspect is required after verbs that denote the onset or the end of a situation, and those of perception. For example: (2.11) a. Arhise na dhulevi / * na dhulepse He started reading / * He started to worked b. Stamatise na dhulevi / * na dhulepse He stopped working / *He stopped to worked (2.12) a. Akusa ti Maria na traghudhai / * na traghudhise I heard Mary singing / * I heard Mary to sung b. Idha ton Jani na erhete / * na irthe I saw John coming / * I saw John to came

The perfective is required by verbs that denote the end of a situation. For example: (2.13) a. Kontevi na teliosi Near.3sg.INP SUBJ finish.3sg.PNP The movie is about to finish

to the

ergho movie

For the syntactical representation of clause, I follow the standard assumptions contributed by Philippaki-Warburton (1993, 1994) and Giannakidou (1998) which inform us that the CP domain of a Greek clause and a subject-verb-object (SVO) order should be as follows:

Modality in Embedded Clauses (2.14) [C oti/‡ [Topic/Focus [MoodP [T/AgrP [VP]]]]]]

na/imperative/‡

25 [NegP

dhen/min

[Now-TP

tha

Adopting the reasoning of Philippaki – Warburton (1993, 1994) and Giannakidou (1998), I assume that since there is no overt counterpart to the na particle for the indicative, Mood0 hosts a zero morpheme in indicative clauses, the verb stays in T/Agr0, and clitics are adjoined to T/AgrP (proclisis, see example). Imperative is morphologically instantiated on the verb; thus, the verb moves to Mood0. The null element that is hosted under C0 is justified in cases where the na particle is preceded by overt realizations of complementizers compatible with na, for example, ja na (purpose clauses), prin na (temporal clauses), etc. (see Giannakidou 1998).

3 Mood Choice According to Portner (2009), among others, languages tend to distinguish between three types of moods: verbal, notional, and sentence, and roughly, three categories: indicative, subjunctive, and imperative. Greek employs a wide range of morphemes to modify words and express different grammatical categories such as tense, aspect, person, number, gender, and case. However, in Greek, morphology has a limited contribution to modality, and specifically to moods. Greek distinguishes three moods: indicative, subjunctive, and imperative. The subjunctive mood is marked by the presence of particles such as na, as, and not by inflection. Particles precede the inflected verb and the clitic pronouns. Verbal mood is distinguished from the other two types (notional and sentence mood) because verbal inflections are used to express how an action or a state is conceived by a speaker. The next example displays the imperative marked with the morpheme for the 2nd singular person –u in Greek (not an exhaustive list of the imperative verbal suffixes): (2.15) a. Eksafanis-u apo ta matja mu! Get out of my sight! b. Mathe to Know this!

With imperative verb forms enclisis arises when it combines with clitics (2.15b). Notional mood deviates from the standard definition in terms of form. This means that it still expresses an attitude (desire, wish, requirement, etc.) towards what is said, but in a different form (infinitives, modal

Chapter Two

26

concord, and modal subordination; see Portner 2009, among others). For example: (2.16) a. I want to travel around the world expressing a desire or a wish b. Thelo na taksidhepso ton kozmo Want.1sg.INP SUBJ travel.1sg.PNP the world I want to travel around the world

In English the notional modality manifests in the infinitive, but in Greek, which lacks uninflected verb forms, this is substituted with what we conventionally call the subjunctive: a periphrastic verb form consisting of the particle na (subjunctive), as (optative) (see Mackridge 1987; Holton et al. 1997, among others). The as (optative) particle is used only in main clauses, while na (subjunctive) can be used in both main and embedded clauses as follows: (2.17) Na/as to SUBJ/OPT the Let him/her know it

mathi know.3sg.PNP

With the indicative, clitics appear before the verb. For example: (2.18) To emathe It learn.3sg.PP She/he learned it

Sentence mood, in turn, is the communicative role played by a sentence’s propositional content in discourse (illocutionary force), as expressed through linguistic means. For example: (2.19) a. I wonder if you can close the window b. Anarotjeme an boris na klisis to parathiro Wonder.1sg.INP if can.2sg.INP SUBJ close.3sg.PNP the window I wonder if you can close the window.

The sentences in (2.19) are both assertions; however, their communicative role is that of a request. Thus, from what has been said thus far, mood is one of the functional categories of syntax, the requirement of which is supported by the fact that Greek relies more on syntactical distinctions between indicative, subjunctive and imperative moods rather than morphological ones. Greek utilizes a limited morphology and a wide set of syntactic means in

Modality in Embedded Clauses

27

compensating the loss of mood morphemes from Ancient Greek (AG) (see Veloudis & Philippaki 1983; Holton et al. 1997; Tsangalidis 2001a/b, 2003, among many others). Syntax has amended the loss by distinguishing between the indicative and subjunctive with the use of the subjunctive particle na (see more detailed discussion in the next section), which appears in a preverbal position unless mediated by an object clitic pronoun, while the imperative retains its morphological status, as demonstrated by the following examples: (2.20) a. Thelo i Maria na potisi ton kipo Want.1sg.INP the Maria SUBJ water.3sg.PNP the garden I want Maria to water the garden b. Thelo i Maria na ton potisi Want.1sg.INP the Maria SUBJ the.acc water.3sg.PNP I want Maria to water the garden c. * Thelo na i Maria potisi ton kipo want.1sg SUBJ the Maria water.3sg the garden

3.1 Mood Choice in Complement Clauses Mood choice in embedded clauses in Greek is realized in complementizer position C (see Giannakidou 2009), and is regulated by non-veridicality (see Giannakidou 1995, 1998, 1999, 2010), a concept that originated in Montague (1969) and his influential paper5 on perception verbs and the existence of dubious epistemological, metaphysical, and ethical entities such as pains, tasks, events, and obligations. The notion of veridicality and nonveridicality has been employed in a number of studies most notably in those by Giannakidou (1995, 1997, 1998, 1999) and Zwarts (1995), who capitalized on the related concept6. The key point here is that certain entities would come with two senses: veridical and non-veridical. The distinguishing feature of (non) veridicality is the commitment to truth in an epistemic model of an individual formally represented as follows: 5

Montague (1969) determined the essential qualities of veridicality in terms of existence. For example, the perception verb see is veridical because, if John sees a unicorn is true, then it must be true that a unicorn exists, and that is a logical consequence. However, in the sentence John seeks a unicorn the verb seek is nonveridical, because, if John looks for a unicorn, it is not necessary true that a unicorn exists. 6 Giannakidou (1995) and Zwarts (1995) explained a variety of unrelated phenomena, i.e., NPIs, PIs, mood choice, temporal connectives, to mention a few.

28

Chapter Two (2.21) Epistemic Model of an individual (cited from Giannakidou 1998) An epistemic model of an individual x, ME(x), is a set of worlds w’ accessible from a world w, compatible with x’s beliefs in w. (2.22) Truth in a Model (adapted from Giannakidou 1999) A proposition p is true in an epistemic model ME(x) if ME(x) Ž p w [w  ME(x) o w  Ow’.p (w’)] A proposition p is false in an epistemic model ME(x) if ME(x) ‫م‬p w [w  ME(x) ե w  Ow’.p (w’)]

Following Giannakidou (1999), this study assumes that propositions are true or false in a context c with respect to an individual’s epistemic state (e.g., 2.21). A necessary condition for a proposition to be true is for the individual’s epistemic model to be a subset of the worlds where p is true: ME(x) Ž p (e.g., 2.22). In the case of embedded propositions, p may be evaluated with regard to the model associated with the matrix subject or with the speaker (Giannakidou 1999). A veridical propositional operator conveys the commitment and the certainty of an individual to the truth of a proposition. On the contrary, a non-veridical operator expresses the non-commitment and uncertainty of an individual to the truth of a proposition. The difference of veridical and non-veridical propositional operators is expressed formally in the following definition (adapted from Giannakidou (2009)): (2.23) Definition of non-veridicality for propositional operators: a. A propositional operator F is veridical iff Fp entails or presupposes that p is true in some individual’s epistemic model ME(x); otherwise F is nonveridical. b. A propositional operator F is antiveridical iff Fp entails that not p in some individual’s epistemic (or deontic) model: Fp ĺ ™ p in some ME(x).

The definition of non-veridicality, as Giannakidou (1999) pointed out, requires truth entailment, not existence. Specifically, the definition of veridicality in example (2.23a) involves the notion of existence in an individual’s epistemic model. A propositional operator F is veridical iff Fp entails p meaning, whenever Fp p is true, then p is also true. In example (2.23b) the propositional operator Fp p is non-veridical if p does not hold. This means that we are not committed to the truth of p; p may or may not be true. In essence, non-veridicality does not entail the falsity of p. As I mentioned earlier, I assume that modal auxiliary verbs are non-veridical, and this implies that we are not committed to the truth of the modal’s argument, i.e., the truth of a CP.

Modality in Embedded Clauses

29

3.2 Complementizer Selection The type of complementizer is determined by the type of predicate in the main clause. If the main verb conveys a commitment to the truth of the complement clause, then it is veridical and selects the indicative. Indicative clauses have three varieties depending on the type of complementizer: (a) following verbs of saying, believing, or thinking (semi-factives) they are introduced by oti or pos ‘that’ (see 2.24a), (b) following verbs expressing psychological states (factive verbs) (joy, sorrow, regret, etc.), they are introduced by pu ‘that’ (see 2.24b), which has been characterized as the factive complementizer (see Christides 1981; Varlokosta 1994; Roussou 1994), and (c) following verbs expressing fear they are introduced by mipos ‘that’ (see 2.24c). For example: (2.24)Indicative Complements a. Oli kserun oti / pos i Maria ine engios everybody know.3pl.INP that the Maria is.3pl.INP pregnant Everybody knows that Maria is pregnant b. I Maria metaniose pu ipe tin alithia the Maria regret.3sg.PP that tell.3sg.PP thetruth Maria regretted for telling the truth c. O Janis anisihi mipos i Maria den erthi The John worry.3sg.INP that the Maria not come.3sg.PNP John worries about Maria not coming

If the verb in the main clause does not express any commitment to the truth of the complement clause, it is non-veridical and selects the subjunctive. Subjunctive complements come after verbs of desire, wishing, hoping, planning, promising, requesting, and modal verbs, and they are introduced by na, which can be translated as either ‘that’ or ‘to’. Subjunctive clauses also follow verbs expressing perception, and in that case, they correspond to the – ing suffix of English. For example: (2.25) Subjunctive Complements a. Thelo na su want.1sg.INP SUBJ you I want to talk to you b. Iposhome na promise. 1sg.INPSUBJ not I promise that I won’t be late c. Pedja, prepi na Kids must.3sg.INP SUBJ Kids, you must sleep.

miliso talk.1sg.PNP min arghiso be-late.1sg.PNP kimithite sleep.2pl.PNP

30

Chapter Two d. Ton idha him see.1sg.PP SUBJ I saw him running.

na trehi run.3sg.PNP

Some verbs that select oti ‘that’ can be introduced by a na complement clause: (2.26) a. Pistevo oti tha erthi o Janis Believe.1sg.INP that FUT come.3sg.PNP the I believe that John will come with us b. Pistevo na erthi o Janis Believe.1sg.INP SUBJ come.3sg.PNP the I believe that John will come with us

mazi mas John with us mazi mas John with us

In this pair, the difference is that oti (2.26a) expresses an objective, while na (2.26b) expresses a subjective judgment. I will not go into further details here on the complementizer selection, but for more information on this topic in Greek and the meaning shift, see Giannakidou (2010) and Giannakidou and Staraki (2013).

3.3 Complements of Modal Verbs Greek also employs a set of impersonal and personal verbs that select a complement clause introduced by the particle na. The purpose of this section is to present an overview of the category of modal auxiliary verbs, and their distinguishing properties. The impersonal modal verbs I examine are prepi ‘it must’, bori ‘it is possible’ and boro ‘I am able to’. All three take subjunctive na complements (Mackridge 1987; Holton et al (1997); Clairis & Babiniotis (2005), among many others). The latter two, bori and boro, are impersonal (3rd person singular) and personal variants of the same verb. The impersonal bori is an epistemic modal, whereas the personal is abilitative or deontic, never epistemic. The modal verb prepi is a necessity modal, epistemic or deontic, and is always impersonal, as in the following example: (2.27) a. * Ta pedhia prepun na trone fruta The children must.3pl.INP SUBJ eat.3pl.INP fruits Children must eat fruits b. Ta pedia prepi na trone fruta The children must.3sg.INP SUBJ eat.3pl.INP fruits Epistemic necessity: As far as I know, children must eat fruits Deontic necessity: The requirements for a healthy diet dictate that it is necessary that children eat fruits

Modality in Embedded Clauses

31

(2.28) a. Ta pedia bori na fighun The children might.3sg.INP SUBJ leave.3pl.PNP Epistemic possibility: As far as I know, it is possible that children leave b. Ta pedia borun na fighun The children can.3pl.INP SUBJ leave.3pl.INP Ability: Children are able to leave. Deontic: Children are allowed to leave.

The na complement of both impersonal and personal modal verbs comes in the following temporal and aspectual permutations: (2.29) Na – complement a. Non-Past Perfective b. Non-Past Imperfective c. Past Perfective d. Past Imperfective

(PNP) (INP) (PP) (IP)

With the exception of Giannakidou (1999) that suggested the nonveridicality of modals, and Iakovou (1999) and Iatridou (1990) who offered a descriptive account of Greek modality, there are no other previous studies on the Greek modal auxiliary system. Giannakidou (1999) argued that (non)veridical functions are typically proposition-embedding functions, for example, modal operators that take a clause as their complement, and she presents for consideration modals with non-veridical properties. However, Giannakidou (1999) did not deliver an account of modals and how non-veridicality regulates their interpretation in embedded contexts. Thus, one purpose of this study is to provide a modest extension of her proposal on non-veridicality and show its descriptive and explanatory power in describing sentential modality in Greek. Furthermore, my goal is to investigate the role of non-veridicality and how we can employ it in explaining the dependency of PNP on non-veridical particles and its embedding under non-veridical propositional operators: the modal operators. Therefore, I adopt the thesis that modal verbs are non-veridical operators (adapted from Giannakidou (1999)): (2.30) Nonveridicality of modals a. ۤ„‘”‹Ԅ‫ۥ‬c, ME(s) = 1 ե ۤԄ‫ۥ‬MB(s) = 1 b. ۤ’”‡’‹Ԅ‫ ۥ‬c, ME(x) = 1 ե ۤԄ‫ ۥ‬MB(s) = 1

In Giannakidou’s (1999) account a modal’s nonveridical properties are determined based on an epistemic model of a speaker ME(x) (see 2.21). In addition, I assume that particles like na (subjunctive), an (conditional), as

32

Chapter Two

(optative) and tha (future) are also non-veridical operators that do not entail commitment to truth. Their semantic definition follows: (2.31)ࠣna / an / as / tha Pࠤc, ME(s) = 1 ե ۤ߶‫ ۥ‬ME(x) = 1

I assume that the modal base of a modal is also related to a speaker’s epistemic model, thus, relating the truth of a modal not only to contextual information (Kratzer 1977, 1981, 1991b) but to a speaker’s epistemic status (indirect evidence), as well. Specifically, the novel thing that I argue for is the modeling of the speaker’s epistemic status, which is being determined by non-veridicality (Giannakidou 1998), and not only context (Kratzer 1977, 1981, 1991b; Portner 2007a, 2009; and von Fintel & Gillies 2007; among others). By modeling a speaker’s epistemic model via non-veridicality I aim to show two things. First, I will demonstrate that a speaker can also determine the meaning of a modal by providing a body of evidence (information, knowledge, laws, desires, etc.) leading to an indirect inference or deduction. Second, I will demonstrate that a speaker’s epistemic model provides a speech act that updates the CG among participants in a conversation, an idea that is compatible with what Portner (2007a, 2009) and von Fintel & Gillies (2007) suggested as being an essential part of a modal’s meaning. I would like to extend this assertion in a modest manner and make it more profound. Later in the analysis of Greek modality we will see in greater detail how a speaker’s epistemic model hypothesis influences the related background information (context, speaker, participants, shared knowledge, etc.) in determining a modal’s meaning. The complement of an embedding verb in Greek can be one of either of the two following types of complement clauses: the indicative or the subjunctive. Considering what we have said thus far about nonveridicality, we assume that the divide within the class of verbs depends on the availability of a truth inference meaning, and whether an epistemic agent (speaker, subject of the main clause, etc.) is committed to the truth of the complement clause (see Giannakidou 2009). Following standard assumptions (Philippaki – Warburton 1985, 1993; Tsimpli 1990; Giannakidou 1998, 2009), I adopt the following structure when representing modal verbs: (2.32) Modal Auxiliary Structure [C oti/‡ [MoodP na [NegP dhen/min [TP [T/AgrP[VP]]]]]]

Modality in Embedded Clauses

33

4 A Framework for Greek Temporal System Focusing on dependency, Giannakidou (2009) presented a syntacticsemantic account of the PNP and particles (tha, na, as), arguing that the function of the particles is the introduction of the variable now (n)7 into the syntax. The introduction of the variable now is needed, because the present tense in the PNP8 cannot introduce n. Furthermore, Giannakidou (2009) argued that the Greek PNP must be preceded by the subjunctive and nonveridical particles9 in order to be grammatical. These particles have the function of present (PRES): they introduce n into the syntax and function as PRES because the Greek non-past cannot introduce n. Thus, Giannakidou (2009) in contrast to Tsangalidis (2002a/b), argued that the PNP is a deictic, temporal interval, when it combines with one of the particles. Non-past contains a temporal variable t that cannot be interpreted deictically (i.e., as a free variable picking up the utterance time). Non-past is formally represented in the following definition: (2.33)ࠣnonpastࠤ= Ȝȇ Ȝt ȇ((t, f)) Giannakidou (2009)

Giannakidou (2009) observed that the reason PNP is dependent relates to aspect. The imperfective aspect is used for generic statements (expressing habituality) and as such, the dependent interval (t,f) in INP can be identied with an adverbial time because the imperfective aspect does not make direct reference to the utterance time. This is not the case with the perfective aspect, as Giannakidou (2009) said, prohibiting the 7

The now (n / Now) operator has been introduced by Kamp (1971) in order to accommodate the differences between the set of the following sentences that were unexplained in Prior’s tense system:

(a) (b)

A child was born that would become ruler of the world. P[x[child(x) š born(x) š F(become-ruler-of-the-world(x))]) A child was born that will become ruler of the world. P[x[child(x) š born(x) š NF(become-ruler-of-the-world(x))])

The n operator relates F (Future) to the utterance time (b) rather than to the P (Past) operator in the scope of which F would have to be without n operator (a). 8 In Giannakidou (2009), and in the traditional grammars (see Mackridge 1987; Holton et al. 1997), PNP refers to a verb form that is perfective non-past. 9 This set of particles includes na (subjunctive), an (conditional), as (optative) and tha (future).

Chapter Two

34

introduction of n into the structure. Particles such as na and tha having the function of PRES10 solve the problem of introducing n into the syntax. Note that the function PRES indicates simultaneity and is identified with the utterance time. For example: (2.34) a. Na erthis an thes (tora/avrio) SUBJ come.2sg.PNP if want.2sg.INP (now/tomorrow) You should come, if you want (now/tomorrow). b. Tha erthis an thes (tora / avrio) FUT come.2sg.PNP if want.2sg.INP (now / tomorrow) You will come, if you want(now / tomorrow).

I adopt Giannakidou’s (2002, 2003, and especially 2009) arguments with regard to the past vs. non-past opposition, and the aspectual and temporal distinctions for the Greek verbal system: PERFECTIVE

PAST IMPERFECTIVE

NON-PAST PERFECTIVE IMPERFECTIVE

Table 3 Past and Non-past opposition in the Greek tense system PNP is the proper opposition for INP in terms of temporal perspective only when it is combined with an auxiliary particle. As mentioned, the basic morphological distinction for tense in Greek is past and nonpast11, and for aspect perfective and imperfective. Greek language lacks present tense (Giannakidou 2009) in the sense that the verb form corresponding to the misnomer present qualifies both in rendering progressive and/or habitual acts. Particles such as na (subjunctive), an (conditional), as (optative) and tha (Now-TP) introduce the parameter n, which is identified with either the utterance time, in main clauses, or a relative now, in embedded clauses (Giannakidou 2009). Hence, non-past interval instantiates a property that lacks temporal orientation and does not give the utterance time as its value. The definition of the non-past tense is

10

These particles function as PRES because the Greek non-past is not PRES and cannot introduce n (Giannakidou 2009). 11 The idea of a binary distinction of tenses (past vs. nonpast) is not novel, and it has been suggested for English as well. In contrast to traditional accounts, Ogihara (1986) and Enç (1996) argue that there is only past tense in English. We understand their proposal to mean that there is a binary distinction: past vs. nonpast.

Modality in Embedded Clauses

35

provided in (2.35a). The past tense signals anteriority, referring to a time prior to the utterance time, as demonstrated in (2.35b): (2.35) a.ህnonpastሆ= Ȝȇ Ȝt ȇ((t, f)) b.ህPASTሆg, c if g(t) < n, in which case ህPASTሆg, c = g(t)

Particles na (subjunctive), an (conditional), as (optative) and tha (future) introduce the n time parameter (or, utterance time) independently since the PNP cannot introduce it. The utterance time then serves to anchor the left boundary of the PNP, thus yielding a future time reference. For example: (2.36) *Milisi i Maria Speak.3sg.PNP the Maria Intended: Maria should speak / Let Maria speak

(2.37) a. The event of speaking happens at the utterance time b. The event of speaking happens at a future time TP: t e [speak (m,e) š e Ž t š t  [t’, f]]

The PNP has to be located in another time n because of its anaphoric nature. Nevertheless, according to the derivation in (2.37), there is not a higher node offering the utterance time (n) that the PNP requires in order to be anchored. The PNP is ungrammatical without the presence of a marker and in our case without the na marker. We can see tha the problem

36

Chapter Two

with the derivation in example (2.37) is that the t’ is a dependent variable. The t’ and n cannot overlap. The non-past tense that is expressed by the PNP has to map the time t to an open interval that starts at some other time t’ and stretches through to the future (Giannakidou 2009). In other words, t’ is purely dependent, and the utterance time has to be provided externally to the PNP in order to render the sentence grammatical. Recall the denotation of non-past from Giannakidou (2009): (2.38) [[non-past]] = O5s,t Ot P(t) š t  [t’, f] t’ n: The t’ is not the utterance time

Example (2.36) acquires temporal orientation when the particle na is introduced into the clause. For example: (2.39) Na milisi i Maria SUBJ speak.3sg.PNP the Maria Maria should speak / Let Maria speak MoodP: e [speak (m, e) š e Ž t š t  [ n, f)]

Notice in this derivation that adding na disambiguates the interpretation of the sentence. Now, the event of speaking starts at the time of utterance and stretches through to infinity. Notice the following difference between the derivations in (2.37) and (2.39): (2.40) a. LF Derivation in (2.37) provides: [t’, f] = the event starts at a time, which is non-past. b. LF Derivation in (2.39) provides: [n, f] = the event starts at the utterance time and stretches to the future.

When MoodP is not projected n will be provided by the next lowest head Now-TP.

Modality in Embedded Clauses

37

(2.41) Now – TP: e [speak (m, e) š e Ž t š t  [n, f)]

Now-T: tha: n

TP: Ot e [speak (m, e) š e Ž [ t’, f)]

Milisi i Maria ‘Maria speaks’

Therefore, particles na (subjunctive), an (conditional), as (optative) and tha (future) independently introduce the n time parameter (or, utterance time) into the structure. Giannakidou’s (2009) account is efficient for two reasons: (a) it pointed to a unifying account of a grammatical category (particles) which is characterized by semantic and syntactic dissimilarity, and (b) it offered an account of mood choice, which in Greek is both morphologically and syntactically defined, based on the notion of non-veridicality. For the analysis of Greek aspects, I adhere to standard assumptions (Kamp and Reyle 1993; Klein 1994; von Stechow 2002, for Greek Giannakidou 2003, 2009) and incorporate into the analysis the functions employed for the perfective and the imperfective by Giannakidou (2009). The perfective aspect is a modifier function that takes a VP meaning, P as its input, and returns a predicate of times such that an event characterized by P is included in those times: (2.42)ህPFTሆ= OP Ot e [ P(e) š e Ž t]

A common characteristic in languages is aspectual shifts (Zucchi 1998), and Greek does not deviate in these terms (Giannakidou 2003). Statives can be modified by perfective, resulting in an eventive reading that produces an inchoative reading, for example: (2.43) I Irini agapise The Irene love.3sg.PP Irene fell in love with Alex. e [Irene love (e) š e Ž t]

ton the

Alexi Alex

The example in (2.43) illustrates the aspectual shift of a stative verb after the application of perfective triggering an eventive reading: i.e., the start of a falling in love event. In this respect, Greek differs from languages like French, where statives, in combination with the perfective aspect, designate the culminative and not the inchoative reading (Hacquard

38

Chapter Two

2006). The Greek verb does not contain eventuality information and is underspecified for lexical aspect (Giannakidou 2009). Therefore, if we consider that aspectual shift is a common characteristic in Greek, and the verb does not have a prespecified lexical aspect, then, as it will become clear later, we cannot argue for a temporal interpretation of a modal based on the kind of its embedded predicate, namely, the lexical aspect. Thus, as it will become evident in the next sections, I contrast the theories of studies such as those of Condoravdi (2002)12, which considered the lexical content (lexical aspect) of an embedded predicate as having a direct relation in the temporal interpretation of a modal, or correlating to any subsegment of time, for example, the utterance time (for contextual dependence vs. lexical ambiguity, see Kratzer 1977; Werner 2003, 2006). The imperfective aspect in Greek is used for either habitual and generic statements, as well as progressive events. For analyzing the Greek imperfective aspect with a function PROG, I adopt the following two standard functions provided by Giannakidou (2009): (2.44) Imperfective Progressive PROG PROG = OP Oi s t [(t  C  š t Ž i ) o P (s, t) š s o i ]

When the imperfective aspect is used for generic statements, it contributes the function GEN: (2.45) GEN = OP Oi GENt [(t  C š t Ž i; P(t)]

The imperfective aspect does not encode any eventuality information, and expresses a generalization over times creating an interval for the VP property to map on. As Giannakidou (2009) pointed out, Greek verbs are eventive when in perfective, and denote progressive or generic intervals when in imperfective.

5 Conclusion In this chapter, I presented the temporal and aspectual system of Greek. Greek inflects for both tense and aspect (Mackridge 1985, Holton et al.

12 The position that different modal readings are due to the context and not to lexical ambiguity (Kratzer 1977; Werner 2003, 2006) contrasts not only to the account of Condoravdi (2002). For example, Ross (1969) and Lightfoot (1979) discussed about the syntactic differences among modal verbs, and Cinque (1999) discussed about the scopal relations that affect the interpretation of modals.

Modality in Embedded Clauses

39

1997, among others) and employs three morphological distinctions regarding the mood, the tense and the aspect in a binary mode, as follows: (2.46)

Mood distinction: ± imperative Tense distinction: ± past Aspect distinction: ± perfective

I discussed the essential role of non-veridicality in complementizer selection. If the main verb conveys a commitment to the truth of the complement clause, then it is a veridical operator and selects the indicative. If the verb in the main clause does not express any commitment to the truth of the complement clause, it is a non-veridical operator and selects the subjunctive. Non-veridicality determines the complement of the set of modal auxiliary verbs in Greek such as prepi ‘it must’, bori ‘it is possible’ and boro ‘I am able to’ which take subjunctive na complements, as modals are considered non-veridical operators. In the following chapters, I elaborate on non-veridicality and show its descriptive power of describing sentential modality in Greek. The goal is to investigate the role of non-veridicality, to explain the dependency of the PNP on non-veridical particles and to investigate the embedding of the PNP under non-veridical propositional operators: the modal operators.

CHAPTER THREE EPISTEMIC MODALITY

1 Introduction In this chapter, I discuss the puzzles related to the interpretation of epistemic modality and consider whether the epistemic modality in Greek complies with the frameworks and analyses suggested in the relevant literature. We will see that the epistemic modality in Greek is not a genuine evidential operator; rather, it incorporates an evidential component in its semantic set up as a presupposition (pace von Fintel & Gillies 2010; for Greek see Staraki 2012).

2 Issues in Epistemic Modality It is known that modal expressions convey either an epistemic or a root (priority and dynamic) modality. It is also true that modality is contextsensitive. Kratzer’s work (1977, 1981, 1991) offered a way to interpret modals by assuming that they are context-dependent and defined according to three basic parameters: (a) the modal force, (b) the modal base, and (c) the ordering source (see chapter 1). However, a number of puzzles challenge the standard theoretical framework that has been suggested for the treatment of epistemic modality. The first puzzle has to do with the truth-conditional content of epistemic modal expressions. On the one hand, it has been proposed (see for example Lyons 1977) that epistemic modals do not contribute to the truth-conditional content of the proposition that they are part of, and that epistemic modals are only related to the speaker’s evaluation. Epistemic modals are extra-linguistic operators with a purely evidential contribution that affects the type of speech act performed by a sentence (see Portner 2009). On the other hand, epistemic modality is not only considered as contributing to the truth content of a proposition, but also as offering a speech act (see Swanson 2006a, 2006b, 2007; Portner 2007a, 2009; von Fintel and Gillies 2007; Yalcin 2007a/b; among others). Evidence in favor

Epistemic Modality

41

of a non-truth conditional analysis of epistemic modality comes mainly from two empirical facts: first, the lack of embeddability which is the difficulty of embedding epistemic modality under other propositional operators (Jackendoff 1972; Lyons 1977; Palmer 1986; Drubig 2001, Papafragou 2006; Portner 2009; von Fintel & Gillies 2008b), and second, the interpretation of epistemic modals in assent/dissent contexts. I will examine how strong these arguments are in the case of Greek epistemic modals later in this discussion. A second puzzle has to do with the relationship between epistemic modality and evidentiality. The question here is two-fold and related to the first puzzle: whether epistemic modality is part of a greater category, evidentiality (see Westmoreland 1995, 1998; Drubig 2001; Nuyts 2001a; Faller 2002; Aikhenvald 2005; von Fintel & Gillies 2007b; among others), or, if evidentiality is an essential part of epistemic modality’s meaning (Matthewson et al. 2006; McCready & Ogata 2007; Matthewson, Rullman & Davis 2006, 2008). The supporting evidence of the view that epistemic modality is part of the evidentiality category is the fact that epistemic modality does not contribute truth-conditional content, and that epistemic modality has a different argument structure than other modals, taking wide scope over other propositional operators such as negation or tense (Drubig 2001; Condoravdi 2002; Groenendijk & Stokhof 1975; Enç 1996; Abusch 1997; among others). In this study, I argue that epistemic modality in Greek is part of the proposition expressed by the sentence it appears in (thus involving quantification over possible worlds), and that it does not contribute an extra truth-conditional like evidential markers. I also propose that the interpretation of a modal sentence depends on contextual parameters (Kratzer 1977, 1981, 1991) to resolve the indeterminacy that does not uniquely involve speaker’s knowledge and an admissible context (von Fintel & Gillies 2006 and sequel), the truth of a modal proposition (Portner 2006, 2007a), or scopal relations (contra Jackendoff 1972; Zubizaretta 1982; Picallo 1990; Brennan 1993; Cinque 1999; Drubig 2001; Butler 2003; Iatridou & von Fintel 2003; Hacquard 2006, among others).

3 Modality and Speech Acts A modal expression can be viewed as a performative one; a speech act that is performed in addition to that of the assertion. For example, subjective epistemic modals, according to Lyons (1977), modify the speech act of a certain sentence, contributing a weaker speech act than that of the

42

Chapter Three

assertion. In other words, a modal expression performs two speech acts that of an assertion and a speech act of another kind. The view that modality performs a speech act (see also Halliday 1970; Palmer 1986; Bybee and Fleschman 1995b; among others) has also been supported by many linguists. Swanson (2006a, 2006b, 2007) argued that a modal expression with ‘might’, for example, performs two more speech acts besides the assertion. For example: (3.1) a. The keys are on the desk. b. The keys might be on the desk.

According to Swanson (2006a, 2006b, 2007), in example (3.1b) the addressee is advised to modify her belief about her knowledge of the world (current situation) and at the same time, the modal expression raises the possibility of something that the addressee might not have been aware of, i.e., the possibility that the keys are on the desk. Both examples in (3.1) are assertions but they differ in terms of convincing the addressee to believe something. The assertion in example (3.1a) gets the addressee to fully believe p; the probability that is assigned is 1. However, the assertion with the modal expression introduces a hedged recommendation that the addressee not rule out the possibility that p (Swanson 2006b); probability assigned 0.5. The difference of those two assertions is the assignment in terms of probabilities, and for Swanson, assigning probabilities is an indication that an epistemic modal, such as ‘might’, is a speech act modifier. Raising a possibility that the addressee should not ignore is the second kind of speech act according to Swanson. This is best represented in terms of the addressee’s belief state. When the addressee is not aware of the possibility that the keys are on the desk, the coarse-grained belief state does not assign a probability to it. However, when the speaker employs a modal expression she raises the possibility, thus affecting the addressee’s belief state and making him assign a probability. Likewise, Yalcin (2007a/b) argued that epistemic modal expressions perform a speech act that is evaluated within a context for rationality R and advisability A and not for its truth condition T. Thus, according to Yalcin, R evaluates the consistency within a context of the speech act provided by the modal expression. A, on the other hand, evaluates the worthiness of the speech act being recommended or suggested. The evaluation of a modal proposition takes place via modal resolution, which represents an agent’s partition of the set of possible worlds and the ability to track the possibilities he is aware of regarding which are open or more likely than others.

Epistemic Modality

43

Thus, epistemic modal expressions are not considered as assertions of a declarative proposition. Von Fintel and Gillies (2007b: 43) argued that epistemic modality has a dynamic contribution to the context that it is part of, and is an evidential comment on the prejacent proposition. In the framework of dynamic semantics, a proposition is defined in terms of context change potential (CCP), meaning that a proposition is not defined in terms of its truth conditions, but rather in terms of updating the context in various ways. Von Fintel (2003), von Fintel & Gillies (2006, 2007b, 20008a, 2008b) in a reconsideration of the role and content of epistemic modality argued that epistemic modals are evidential markers (e.g., 3.2a), signaling the presence of an (indirect) inference or deduction, and thus involving a non truth-conditional comment on their prejacent. Von Fintel and Gillies demonstrated that the argument that epistemic modality cannot be embedded is misleading, as examples (3.2b-g) illustrate (negation, tense, quantifiers over modals): (3.2) a. It must be raining b. There can’t have been a mistake. Negation over modal c. Bill thinks that there might have been a mistake. Attitude predicate over modal d. Where might you have put the keys? Question over modal e. The keys might have been in the drawer. Past over modal f. The editor reread the manuscript because there might have been a mistake. Causal operator over modal g. The detective interviewed every resident who (based on the time of the accident) might have seen the accident. Quantifier over modal

In their approach, epistemic modals serve as a comment: that there is a world in the relevant context in which the prejacent is true. Epistemic modal sentences contribute two speech acts (crediting Simons 2006 work on perentheticals), as follows (example taken from von Fintel & Gillies 2007b: 44): (3.3) There might have been a mistake (in the calculation)

According to von Fintel & Gillies (2007b), an epistemic modal sentence in example (3.3) functions as an update of the relevant context it is part of by changing the information state. Modalized sentences are CCPS. The modalized sentence in example (3.3) makes two speech acts: (a) an assertion compatible with the available evidence, and (b) a proffering that there has been a mistake. As von Fintel & Gillies (2007b: 54) put it:

Chapter Three

44

(3.4) a.

•‹ˆ•ۤɔ‫׎ ്  ۥ‬ ‫‡•‹™”‡Š–‘׎‬ b.•ۤ—•–ij‫ ۥ‬ccp ൌ{w ‫ۤ• ׷ • א‬ij‫ ۥ‬ccp = s } sۤ‹‰Š–ɔ‫ =ۥ‬൜

The definitions in (3.4a-b) say that the sentence containing the epistemic modal operator might or must takes the information state s (contextually relevant body of information), which is carried by the prejacent M, and returns all or none of it depending on whether the information that M carries is compatible with the information state s. In other words, epistemic modality is an update operator with evidential properties (evidential non -truth component) and pragmatic contribution (speech acts).

3.1 Modality and Truth Content The view that modality performs some kind of a speech act closely relates to the question about the truth-conditional content of epistemic modality.It has been argued that epistemic modality does not contribute to the truthconditional content of the sentence it appears in, but rather, it functions as an extra-linguistic operator that denotes a comment and/or the commitment on the part of the speaker to a statement (Halliday 1970; Lyons 1977; Palmer 1986, 2001; among others), thus taking scope over all other propositional operators (Drubig 2001). This argument implies a close relationship between epistemic modality and evidentiality, and triggers two main theoretical approaches. On the one hand, epistemic modality is analyzed in terms of evidentiality (Westmoreland 1998) which lacks by virtue of its features (scoping over all propositional operators) any truthconditional content (Drubig 2001), and shows the speaker’s rational motive in uttering a sentence (Faller 2002) or performing a speech act (see ‘doxastic advice’ in Swanson 2006). On the other hand, evidentiality is considered an essential component of epistemic modality, thus contributing to the truth-conditional content of the proposition it is in (for example, McCready & Ogata 2007; Matthewson, Rullman & Davis 2006, 2008; von Fintel & Gillies 2007; among others). In Greek, the paradigm of epistemic modality looks like the examples (3.5), and the English translational counterparts demonstrate the same phenomenon: (3.5) a. I Maria bori na ine The Maria might SUBJ Mary might be at the mall

stin be

aghora at-the

epistemic mall

Epistemic Modality b. I Maria ine The Maria is Mary is at the mall

stin at-the

aghora mall

45 declarative

Epistemic modality pertains to the speaker’s knowledge or lack thereof, and expresses the possibility or the necessity of a proposition’s truth, given what is already known and based on the available evidence. Thus, epistemic modality evaluates the evidence and indicates the degree of commitment on the part of a speaker to a proposition. The interpretation of example (3.5a) then, involves an epistemic conversational background that contributes to the proposition expressed by the utterance; in light of what is known, Mary may be at the mall. In contrast, in example (3.5b), a declarative sentence states how the world is, and not how the world might be. Declarative sentences are evaluated relative to truth conditions: Mary is at the mall is either true or false, based on a given situation. On the contrary, Mary might be at the mall is only evaluated as relative, appropriate or inappropriate to a body of evidence that might hold or not, and thus cannot be evaluated as true or false. However, the above theoretical approaches contribute to the misconception of the actual roles of evidentiality and epistemic modality in a proposition. In section 3.3, I will show that epistemic modality contributes to the truth-conditional content (see Papafragou 2006; Portner 2009; among others) of a proposition. Furthermore, I will show that in Greek evidentiality is a component of epistemic modality. Much of the argumentation will be based on epistemic modal expressions that contribute to the update of a context via a speech act. The first goal is to understand how an epistemic and a declarative proposition are different. The ultimate goal will be an account that relates a function to a concept of epistemic modality as a context change potential (see von Fintel & Gillies 2007b). First, let us survey the literature on the (non) truth-conditional analysis. A number of semantic and syntactic puzzles related to the nature and the formalization of knowledge, call for an alternative proposal1 in which the semantic treatment of epistemic modals is largely relevant to evidential markers/operators. Halliday (1970: 349), within a functional approach,

1

The literature on non truth-conditional analysis is vast, see Leech 1971; Jackendoff 1972; Lyons 1977; Palmer 1986, 1990, 2001; McDowell 1987; Nuyts 1993, 2001b; Westmoreland 1995; Garrett 2000;Drubig 2001; Faller 2002, 2006b; von Fintel 2003; Yalcin2007a,b; Stephenson 2007; Portner 2007a; von Fintel and Gillies 2007b.

Chapter Three

46

argued that modality2 represents the speaker’s assessment of probability and predictability, and a part of the attitude taken up by the speaker is thus external to the truth content of the proposition. It has also been suggested in the linguistic literature that evidentiality is in fact part of epistemic modality and that there are two types (subjective vs. objective) of epistemic modality that differ in terms of their scopal position within a proposition and in their interpretation. This analysis originates with Lyons (1977) who supported the view of subjective and objective epistemic modality. For example: (3.6) It may snow tomorrow.

The interpretational differences between subjective and objective modality are captured in terms of scope (see Lyon 1977). The hypothesis is that subjective modals are illocutionary force operators and as such they are above the propositional level. Thus, they function as performatives and express different degrees of commitment to factuality (see Lyons 1977: 805). There is a difference, for example, between example (3.6) when it is uttered by a layperson and when it is uttered by a meteorologist. On its subjective reading, example (3.6) conveys one’s personal belief or commitment state, while on its objective reading, it relays a conclusion based on reliable facts. According to Lyons (1977), each subtype of epistemic modality features its own characteristics and contributes to a different propositional level; subjective modality contributes to the illocutionary force of the proposition, while objective modality contributes to its truth-conditional content (see Nuyts 2001b who describes such cases as intersubjective). Thus, by virtue of its distinct contribution, subjective epistemic modality does not contribute to the truth-conditional semantics of sentences containing it. The main argument of non truth-conditional analyses is that epistemic modals are difficult to embed (Lyons 1977). For example: (3.7) Subjective modality a. ? If John must/may be in Paris, his friends will be very happy. b. ? It is surprising that John must/may be in Paris. c. ? Mary told me that John must be in Paris.

2

Halliday (1970) uses the term modality throughout his discussion without specifying whether he refers to the epistemic or deontic modality. I will assume that he talks about epistemic modality, and that deontic modality cannot relate to evidentiality.

Epistemic Modality

47

(3.8) Objective modality a. If it may snow tomorrow, snow ploughs will clean designated streets. b. It is surprising that it may snow tomorrow. c. The weatherman forecast told viewers that it may snow tomorrow.

Subjective modality is not embeddable either in the antecedent of a conditional, as in example (3.7a) or in the complement of a factive, as in example (3.7b), or in a propositional attitude expression, as in example (3.7c). In contrast, objective modality can be embedded in these contexts (e.g., example 3.8). Another diagnostic that provides evidence in favor of subjective vs. objective modality and their distinct semantic contribution to a proposition is that subjective modality cannot be assented or dissented to, in contrast to objective modality. For example: (3.9) This professor must/may be smart. a. ? Is that so? b. ? I don’t believe it. c. ? I agree. d. ? That’s not true. (3.10) The weatherman forecast told viewers that it may snow tomorrow. a. Is that so? b. I don’t believe it. c. I agree. d. That’s not true.

The examples in (3.9) show that subjective epistemic modal verbs cannot be the topic of agreement, disagreement, or doubt, because the replies in examples (3.9a-d) target the proposition that is embedded under the scope of the epistemic modal. On the contrary, the modal is the target by being the topic of agreement, disagreement, or doubt in examples (3.10a-d). Lyon’s (1977) claimed the following: “… subjective modality conveys solipsistic claims dependent on a speaker’s evidence, while objective modality conveys claims dependent on public scrutiny; hence, epistemic modal operators do not contribute to the truth-conditional content of a proposition, but rather indicate the speaker’s degree of commitment to the truth of the prejacent proposition (Halliday 1970; Palmer 1986; Bybee & Fleischman 1995; among others).” Palmer (1986: 54-55, 2001) considered evidentiality and epistemic modality as similar categories. Both indicate the source and/or the degree of the speaker’s commitment to a statement in the scope of an operator3. 3

I cite the following excerpt from Palmer (1986: 51):

48

Chapter Three

Another argument in favor of the relation between epistemic modality and evidentiality is the way of acquiring knowledge. Frajzyngier (1985, 1987), for example, maintained a similar approach to Palmer (1986) and suggested that the many ways of acquiring knowledge correspond to different degrees of certainty about the truth of a proposition, thus suggesting a link between evidentiality (more credible evidence) and epistemic modality (less credible evidence) (see de Haan 1997 for an outline of evidential hierarchy: direct (visual, auditory, non – visual: more credible) and indirect (inference, quotative: less credible) evidence). Although there is no definitive understanding of which part of the modal system is in fact evidential (direct evidence) and which part is not, Willett (1988) proposed that evidentiality is part of the marking of epistemic modality, a view which later develops into viewing evidentiality as the essential component of epistemic modality (von Fintel & Gillies 2007, 2008b; among others). The relation between the two grammatical categories was further strengthened when Bybee, Perkins, and Pagliuca (1994) and Bybee and Fleischman (1995) proposed that epistemic modality is a clausal scope operator that feed upon indirect evidence, thus implying the lack of commitment on the part of the speaker to the truth of the proposition under the scope of the operator. The suggested proposals of epistemic modality as evidentiality can be narrowed down to the point that Westmoreland (1998), Drubig (2001), and von Fintel & Iatridou (2002) suggested that epistemic modality lacks truth value altogether and, as such, belongs to the extra-propositional layer of a clause that discloses a source of information. However, there are counterexamples (Papafragou 2006; Portner 2007a, 2009) to the embedding and assent/dissent diagnostics. For example: (3.11) a. It is surprising that it may snow in Sahara, since there are no records indicating it had snowed there before or since 1979. b. Ann: The weather forecast told viewers that it may snow in Sahara Bill: That is not true. (= It is not true [that it may snow in Sahara])

“ There are at least four ways in which a speaker may indicate that he is not presenting what he is saying as a fact, but rather: (a) that he is speculating about it (b) that he is presenting it as a deduction (c) that he has been told about it (d) that it is a matter only of appearance, based on the evidence of (possibly fallible) senses. ... All four types are concerned with the indication by the speaker of his (lack of) commitment to the truth of the proposition being expressed”.

Epistemic Modality

49

Example in (3.11a) shows that epistemic modality can embed, and example (3.11b) illustrates a case where the truth evaluation that is not true targets the meaning involving the modal. Thus, we assume that a purely descriptive approach cannot capture the special properties of epistemic modality. Another dimension of the problem of epistemic analysis relates to the indeterminacy of source among different types of knowledge. Papafragou (2006) in a reexamination of epistemic modality provided an alternative account, in which she presented counterarguments to the literature on non truth-conditional analysis. Papafragou’s (2006) main suggestion was that epistemic modality does contribute to the truth-conditional content; otherwise, modalized and unmodalized propositions would be interpreted in the same way albeit with different degrees of speaker commitment4. For example (taken from Papafragou 2006: 1693): (3.12) a. My grandfather must be sick. b. My grandfather may be sick. c. My grandfather is sick.

Examples (3.12) do not express the same proposition. Papafragou (2006) argued that, otherwise in the scenario where one’s grandfather is sick, examples (3.12a-b) would also be false as in example (3.12c), which actually is false. In examples (3.12a-b), the speaker has just said that in view of his/her available information, it is epistemically necessary (3.12a) or possible (3.12b) that his/her grandfather is sick. Papafragou (2006) also argued that the exclusion of a subjective epistemic reading returns counterintuitive results. For example: (3.13) Mary: Paul:

Are you going to the Sunday mass? Maybe/I might.

Paul in example (3.13) has not said anything truth-evaluable: he has merely expressed a certain degree of commitment to a proposition. However, Paul has said that there is a (subjective) possibility that he will go to Sunday mass. Likewise, if epistemic modality is not considered as part of the propositional content, then contradiction yields in example (3.14b):

4

Unmodalized sentences convey a stronger claim compared to the epistemically modalized sentences (see Kartunnen 1972; Lyons 1977; Perkins 1983; Kratzer 1991; Stone 1994; Giannakidou 1999; among others)

50

Chapter Three (3.14) a. Pauli may visit us, and hei may not visit us. b. p & (ap) c. ¡ p & (¡ ap)

However, example (3.14c) conveys that the speaker’s beliefs are compatible with the proposition that Paul visits us, and with the proposition that he does not visit us, as well. Thus, the key observation made by Papafragou (2006) was that, first, the distinction of subjective vs. objective modality offered by Lyons (1977), does not square with the distinction between a non- truth vs. a truth-conditional analysis of epistemic modality. Second, Papafragou observed that both subjective and objective interpretations contribute to the propositional content. Still, this observation does not provide solid evidence for epistemic modality’s contribution to the propositional content. To support her claims, Papafragou (2006) proposed that the key difference between subjective and objective modality is about the type of epistemic agent providing the background assumptions for epistemic modality (Papafragou 2006: 1700). Papafragou, crediting von Fintel & Gillies (2004), pointed to (a) indeterminacy of the relevant community group parameter Gc, and (b) indexicality, in explaining why subjective epistemic modality fails in the diagnostics (embeddability, factives, and assent/dissent patterns). On the one hand, objective epistemic modality allows the epistemic agent to vary (speaker, speaker and hearer, or some other epistemic subset holding the relevant knowledge), while subjective epistemic modality allows only one member, the speaker, to convey epistemic claims. Thus, Papafragou (2006: 1694) offered the reformulated semantics for the epistemic may, as follows (where parameter Gc is the contextually supplied group): (3.15) [[may w]]c, i = 1 iff x  Gc: w’  fx(i): [[w]]c, = 1 Gc = {speaker} / {speaker + hearer} / {speaker + community}

The statement in (3.15) tells us that knowledge bases vary among the individual members of Gc, and, according to Papafragou (2006), captures the essential assertion that an epistemic interpretation indicates a speaker and a comment expressing personal opinion, belief, or information, while at the same time, allowing the speaker to adopt a flexible interpretation. On the other hand, subjective epistemic modality is indexical in nature. In terms of Papafragou’s (2006) analysis, this means that subjective epistemics are tied to the present moment of a conversational exchange, and restricted to what the current speaker knows at the time of utterance (UT) (Papafragou 2006: 1695). These essential characteristics distinguish them from objective modals, which do not bear any temporal or other

Epistemic Modality

51

commitments. Considering that a subjective epistemic modal is indexical, the proposition embedded under the modal operator cannot provide the environment in which a speaker will be evaluating a proposition with regard to his/her beliefs. For example (from Papafragou 2006: 1696): (3.16) ? If I conclude that the Earth is flat, then I’m in trouble.

Hence, the conclusions that Papafragou (2006) reached (modal being relativized to the relevant epistemic agent that the context supplies) offered a way of explaining the subjective vs. objective split hypothesis without proliferating the lexicon with separate lexical entries of modals. Moreover, she maintained a unified account in which the subjective and objective split is in fact a difference in the ordering source: stereotypical, involving superstitions, and objective, involving propositions defensible in an educated society (Kratzer 1981)5. For example: (3.17) a. Subjective: b: Objective:

Ordering Source Ž Speaker’s knowledge/beliefs Ordering Source Ž Society’s knowledge/beliefs

Papafragou’s (2006) analysis offers counterarguments to the traditional view that epistemic modals lack truth-conditions, and only relate to the speaker’s evaluation. However, Papafragou’s account (2006) singles out the interpretation parameters that are identified with some cluster of semantic or morpho-syntactic properties and a set of diagnostics, and captures those features of epistemic modality that depend on formal criteria only without incorporating the pragmatic contribution of epistemic expressions, for example, speech acts. Empirical evidence from Greek, a language where the set of diagnostics returns mixed results, suggests that a purely formal treatment of epistemic modality cannot fully capture its not so ordinary properties, which involve a relation between the output of a semantic interpretation and the use of a linguistic expression that is made in a context. Portner (2007a, 2009) in the same line of thinking as Papafragou (2006), pointed to the weaknesses of a non-truth conditional account of epistemic modality. Portner’s preliminary supporting evidence is the fact that subjective epistemic modals can be embedded in other contexts (i.e., propositional attitude verbs). For example: 5

What is considered to be an educated society and what is considered educated or part of the common knowledge of that society is subject to argument, but I am not going into this discussion here.

52

Chapter Three (3.18) a. Paul thinks that John must be lonely. b. There can’t have been a mistake.

(negation scoping over modal, in Portner (2009) crediting von Fintel & Gillies 2007b: 12) Portner (2009) argued that this might be due to the number of the matrix subject (instead of we) especially in example (3. 18a), but he also noticed that this is not a very strong argument, since we can have the following cases. For example: (3.19) ? Paul doubts/hopes that a large wave must hit the east coast.

Portner (2009) observed that in the case of other matrix verbs, the embedded modal is perfectly acceptable. For instance: (3.20) a. Paul thinks that it may rain. b. It is compatible with Paul’s beliefs that it will snow. c. Paul thinks that it will snow (but he is not sure).

Thus, subjective epistemic modals do embed, and the restricted distribution of subjective epistemic modality is not due to the number of the kind of the verbs in the matrix clause. Another weakness of non truth-conditional analysis that both Papafragou (2006) and Portner (2009) observed was the assent, dissent and questioning patterns. For example: (3.21) Subjective Modality Ann: Paul must be at home. Bill: That’s not true / I agree / Are you sure? (3.22) Objective Modality Ann: The snow precipitation this month must have been below average. Bill: That’s not true/I agree/Are you sure?

The difference according to the standard non truth-analysis is a difference in what the target of the assent, dissent, or question is. In the case of subjective modality, Bill’s response in example (3.21) targets the proposition Paul is at home, while in the case of objective modality, Bill’s response in example (3.22) targets either the proposition the snow fall was below average or that the evidence supports the conclusion that the snow fall was below average. Portner (2009) disagreed with this distinction, arguing that both participants in the conversation make modal claims, but each participant makes his/her own claims. This argument is similar to that

Epistemic Modality

53

of Papafragou (2006) who proposed that subjective modality differs in that it is externally inscrutable. This means that participants in a conversation do not have an infallible and full access to each other’s epistemic grounds. The same is true for questions. Consider the following examples: (3.23) Questions with modals: a. Must Paul be a fraud? b. May Paul be a fraud?

The questions in (3.23) are grammatical as long as the speaker is addressing the modal inquiry to him/herself. Thus, according to Papafragou (2006: 1698), questions bypass the inscrutability restriction; otherwise, these questions cannot be answered by one conversation participant to another in order to examine them for the accuracy of his/her modal claims. We observe a few things that are characteristic of epistemic modality and of the distinction between subjective and objective epistemic modality. First, there is empirical evidence for the subjective vs. objective epistemic modality distinction, which is provided from the difference in the interpretation of epistemic modals. Second, the subjectivity of epistemic modals is not related only to the knowledge/information/beliefs of a speaker, but to that of other epistemic agent(s) as well. Third, the epistemic subjective vs. objective modality distinction is not compatible with the argument that epistemic modality does not provide for the truthconditions of the proposition that contains it. According to Portner (2007a), these special characteristics are better accommodated in a framework that extends beyond the semantic content, and involves the pragmatic contribution of epistemic modality within a general theory of discourse. Epistemic modals contribute to the truthconditional content of the proposition they are in, and also have an evidential-like effect at the speech act level. In formulating this proposal, Portner employed an extended version of Stalnaker’s (1974, 1978, 1987) discourse framework, and argued that there are two pragmatic relations in the treatment of propositions: the common ground (CG), which is the set of propositions that participants of a conversation mutually agree to treat as true, and the common propositional space (CPS), which is the set of propositions that participants in a conversation are mutually aware of; CG is a subset of the CPS: (3.24) CG Ž CPS

54

Chapter Three

The assertion of a modal structure M(ij) performs two speech acts (from Portner 2007a): (3.25) a. Assertion: M(ij) added to the Common Ground b. Evidential-like function: ij added to Common Ground

First, the modal proposition M(ij) is added to the CG, and contributes truth-conditional content to the proposition it is in. Second, the prejacent proposition ij is added to the CPS and is considered to serve an evidentiallike function. This function imposes restrictions on a modal’s conversational background and determines the ordering source. This is done via a selection function cg (common ground) (adapted from Portner 2007a, 2009): (3.26) ¢CPS, cg² + M(ij) = ¢CPS’, cg’², where CPS’ = CPS ‫ۥ\ۤ{ ׫‬c, f, g} ‫ۤ{ ׫‬ij‫ۥ‬c, f, g} cg’ = cg ‫ۤ{ ׫‬ij‫ۥ‬c, f, g}}²}

A conversation consists of the pair: CPS and the selection function cg that selects those members of CPS which are mutually presupposed by the participants. When a modal proposition is contributed to the conversation, it updates the CPS and the cg: ¢CPS, cg² + M(ij) = ¢CPS’, cg’². Specifically, CPS’ is the union of two propositions: the modal proposition {ۤij‫ۥ‬c, f, g} (assertion added to the CPS) and the unmodalized proposition (the prejacent) {ۤ\‫ۥ‬c, f, g}. The updated cg’ consists of the cg and the modalized proposition {ۤij‫ۥ‬c, f, g}. In other words, there are two propositions that are added to CPS, an assertion (contributing a speech act) and a modal one.

3.2 Modality and Evidentiality It has been argued that evidentiality and modality are intertwined. The first approach argues that epistemic modality is just an evidential operator (Westmoreland 1995, 1998; de Haan 1999; Aikhenvald 2005), the second that evidentials are epistemic modals (Matthewson et al. 2006; McCready & Ogata 2007; Matthewson 2009), the third that epistemic modality is a subset of evidentials (Nuyts 2001a; von Fintel and Gillies 2007b; 2010), and the fourth that epistemic modality and evidentiality sometimes coincide partially or wholly (Faller 2002, 2010; Davis et al. 2007a/b; Matthewson et al. 2007). The first position that epistemic modality is an evidential operator, has both a weak and strong version (see Portner 2009 for a thorough presentation). The weak version argues that epistemic

Epistemic Modality

55

modality incorporates an evidential component as part of its meaning (Nuyts 2001a; von Fintel & Gillies 2007b, 2010; among others), while the strong version argues that epistemic modal verbs are genuine evidential operators (Westmoreland 1995, 1998, Drubig 2001; among others). The hypothesis that epistemic modality is intertwined with evidentiality is based on the following three arguments. First, epistemic modality does not contribute to the truth-conditional content of a proposition. Since the relevant data for this discussion are only from English, I provide in the next sections a presentation and analysis of Greek that indicates the opposite: epistemic modality contributes to the truthconditional content of the proposition that it is part of. The second argument concerns scope properties. Greek data indicate, to the contrary, that propositional operators in Greek such as tense (see Staraki 2010), negation, and quantifiers can take a wide scope over epistemic modality, and thus we can have an epistemic modal assessment. Consider negation, which does not alter the epistemic interpretation in the following example: (3.27) a. De bori/prepi na ine o Not might/must SUBJ is the John might/must not be at home. b. Bori/prepi na min ine o might/must SUBJ not is the John might/must not be at home.

Janis sto spiti John at home Janis sto spiti John at home

The counterexamples from Greek show that quantifiers can be interpreted as taking a wide scope over epistemic modality. In fact, in Greek, a quantifier phrase (QP) can be realized in a position before or after epistemic modal verbs at the surface without affecting an epistemic interpretation. For example: (3.28) Kathe kinisi su bori na parakoluthite (kathe kinisi su), alla dhen parakoluthite (ke) kathe su kinisi Every move of yours may be surveilled a. kathe kinisi su x (bori x na parakoluthite) every move yours x (might x SUBJ be-surveilled) b. bori (kathe kinisi su na parakoluthite) might (every move yours SUBJ be-surveilled)

The third argument involves the associated relevant concepts that epistemic modality and evidentiality share, for example, inference and sensitivity to indirect evidence. In Greek, as we will discuss later in this chapter, epistemic modality is not an evidential operator but rather it

Chapter Three

56

incorporates an evidential component (pace von Fintel & Gillies 2007b, 2010). For example: (3.29)

Prepi na hionizi. Must SUBJ snowing It must be snowing.

As in English, epistemic modal expressions in Greek are felicitous when the speaker has inferred that it is snowing, and not when the speaker has directly observed it. Of course, in other languages, such as Dutch, for example, a modal can distinguish between modal and evidential uses depending on the context, but Greek lexically distinguishes epistemic modality with an evidential component from evidential operators. We will tackle these issues in section 3.5.

3.3 Truth Commitment Truth commitment, with respect to the speaker’s epistemic model, regulates modality (see Giannakidou 1998). Previous treatments of modality, as we saw in the above sections, analyze it as either a speech act (performative) or a pure evidential operator (indicating the source of information). Both approaches are based on the misleading premise that modality does not contribute to the truth-conditional content of the proposition it is part of. These analyses cannot be extended to Greek, where modality and the distribution of mood (indicative vs. subjunctive) in the proposition in the scope of modals are determined by the commitment to truth of an individual’s epistemic model (see Giannakidou 1998; 1999; Staraki 2010a). If the main verb conveys a commitment to the truth or falsity of the complement clause, then it is veridical and selects the indicative. If the verb in the main clause expresses no or a weaker epistemic commitment to the truth of the complement clause, it is nonveridical and selects the subjunctive, i.e., modal verbs and modal expressions (see chapter 2). Contextual information is represented in the selectional restrictions of a (non-)veridical context as with modal verbs. This is a statement that holds for both Greek and English. For example: (3.30) a. Prepi na hionizi Must SUBJ snowing It must be snowing. b. * Prepi oti erhete Must that.IND coming c. He must be coming d. * He must that coming

Epistemic Modality

57

In the following section, I show two things. First, modality is regulated by the truth commitment in a speaker’s epistemic model. Second, modality in Greek incorporates an evidential component that is sensitive to contextual information and is modeled in a speaker’s epistemic model.

4 Types of Epistemic Modality in Greek In this section, I discuss the two main types of sentential modality I consider as the central case of my study research: (a) the modal particles na (subjunctive), and tha (future), and (b) the set of impersonal prepi ‘it must’ and bori ‘it is possible’ and personal boro ‘I am able to’ verbs that select a complement clause introduced by the particle na. In Greek, like other languages, modality expresses a great variety of modal meanings, conveying the speaker’s evaluation or judgment of a statement. Epistemic modality refers to the possibility or necessity of a certain mode of knowledge or information (doubt, certainty, etc.). In other words, epistemic modality conveys the degree of certainty with respect to the truth of a statement or a situation. Modality in Greek is expressed either through grammatical or lexical means. Here, I introduce the grammatical and lexical means used when conveying epistemic modality. The grammatical means of expression include various types of verbs, while the lexical means of expression include impersonal verbs6, adverbs, and adjectives. The following table shows the means that Greek language uses in expressing epistemic modality: EPISTEMIC MODALITY MODAL VERBS (impersonal) Prepi na ‘must’ Tha prepi na ‘should’ Bori na ‘may/might’ Endehete na ‘it is probable’ Ine pithanon na ‘it is possible’ etc.

ADVERBS

ADJECTIVES

Isos (na) ‘perhaps/maybe’ Mipos (na) ‘maybe’ Pithanon na ‘possibly’ etc.

Pithanos ‘possible’ Endehomenos ‘probable’ etc.

Table 4 Epistemic Modal Expressions in Greek 6

Na particle is used both in impersonal modal verbs and modal adverbs. However, the second case is rare. Its role is to introduce a non-veridical proposition expressing non-commitment to truth (see Giannakidou 1998, 2009).

Chapter Three

58

Consider the following examples of sentential epistemic modality: (3.31) Speaker A:

Speaker B:

Kapjos erhete Someone coming.3sg.INP Someone is coming. a. Prepi na ine o Pavlos Must.3sg.INP SUBJ is.3sg.INP the Pavlos It must be Pavlos. b. Bori na ine o Pavlos Might.3sg.INP SUBJ is.3sg.INP the Pavlos It might be Pavlos. c. Tha ine o Pavlos FUT is.3sg.INP the Pavlos It should be Pavlos. d. Tha prepi na ine o Pavlos FUT must.3sg.INP SUBJ is.3sg.INP the Pavlos It should be Pavlos.

The responses speaker B may provide to speaker A’s utterance can vary from impersonal modal verbs (e.g., 3.31a-b), introducing a subjunctive na-complement to modal particles (e.g., 3.31c) introducing a main clause. Examples provided in (3.31) are all considered cases of epistemic modality, since they all express an opinion, that the person that is coming is Pavlos, based on incomplete information/evidence and/or a reasoning that involves the formation of conclusions from incomplete information/knowledge. The same set of examples illustrates the difference in the degree of certainty. In particular, examples (3.31a) and (3.31c) relay the highest degree of certainty of the prejacent predicate when compared with the rest of the modal expressions of the set.

4.1 Epistemic Possibility Verbs and Adverbs This section, in particular, discusses the modal verbs Greek employs to express epistemic possibility. The Greek language uses bori ‘may/might’ and verbs denoting mental states, like fantazome na ‘I imagine that’, nomizo pos ‘I think that’, pistevo na ‘I believe that’, and impersonal verb phrases like ine pithanon na ‘it is possible that’, endehete na ‘it is probable that’, etc. The epistemic verb bori ‘may/might’ conveys possibility, and it can also occur in questions, for example:

Epistemic Modality (3.32) a. Bori na.SUBJ ine It might be b. Bori na ine o Pavlos ; Could it be Pavlos?

o

59 Pavlos Pavlos

The na-complement of bori ‘may/might’ can indicate a past, present, or future event, depending on the tense being marked on the predicate, while the modal verb itself can also be in past, present, or future tense (contra Palmer 1993). For example: (3.33) a. Bori /boruse / tha boruse na ine o Pavlos Might.3sg.INP/ IP /FUT might.3sg.IP SUBJ is.3sg.INP the Pavlos It might be/might have been/could have been Pavlos b. Bori na ine / itan o Pavlos Might.3sg.INP SUBJ is.3sg.INP / have been.3sg.IP the Pavlos It might be/have been Pavlos.

Past tense of the modal verb boruse ‘might have’ indicates a possible situation in the past, and future tense tha boruse ‘might have / could have’ conveys a counterfactual statement. Other means expressing epistemic possibility are the following mental state verbs: Lexical Expressions Verbs (3.34)

Fantazome pos/nomizo pos I guess/think

Other Means Periphrastic impersonal modality (3.35) Adverbs (3.36)

Ine pithanon/ine dhinaton It is possible/probable to Isos/Pithanon Perhaps/Possibly

Evidential Adverbs (3.37) Profanos Apparently

Both epistemic possibility and necessity modals in Greek are not different from their English counterparts, as they are also interpreted on the basis of a body of evidence, information, or partial knowledge. This

Chapter Three

60

becomes more evident when we compare modalized and unmodalized sentences. Unmodalized sentences convey a stronger claim compared to epistemically modalized sentences (see Kartunnen 1972; Lyons 1977; Perkins 1983; Kratzer 1991; Stone 1994; Giannakidou 1999; among others), as in the following examples: (3.38) Context: Imagine that you are standing by the window and looking outside where it is snowing. You utter: a. Hionizi! b. # Bori na hionizi It might be snowing. c. # Prepi na hionizi It must be snowing. d. # Tha prepi na hionizi It should be snowing.

Modalized sentences are sensitive to a body of evidence or information. The personal paradigm of the verb boro ‘can/ able to’ also conveys permission, a suggestion that some obstacles be avoided, the approval to do something, and the abilities engaging agents’ particular quality that permits or facilitates achievements or accomplishments. For the purpose of this chapter, I shall be concerned only with the possibility meaning of bori as both the permission and ability meanings express priority modality (see chapters 5 and 6 respectively).

5 Epistemic Necessity Operators: prepi and tha Greek language uses prepi ‘must’, tha ‘will’, and tha prepi ‘should/ought’ to express epistemic necessity. For example: (3.39) a.

b.

c.

Prepi na ine must SUBJ is It must be Pavlos. Tha ine o FUT is the It should be Pavlos. Tha prepi na FUT must SUBJ It should be Pavlos.

o the

Pavlos Pavlos

Pavlos Pavlos ine is

o the

Pavlos Pavlos

Epistemic Modality

61

The epistemic necessity verb prepi ‘must’ conveys necessity, and it can also occur in questions, for example: (3.40) a.

b.

Prepi na ine must SUBJ is It must be Pavlos. Prepi na ine must SUBJ is Could it be Pavlos?

o the

Pavlos Pavlos

o the

Pavlos? Pavlos

The na-complement of prepi ‘must’, and tha prepi ‘should/ought’ indicate a past, present or future event depending on the tense being marked on the predicate, while the modal verb itself can also be in past, present or future tense, for example: (3.41) a. b.

Prepi /eprepe / tha eprepe na ine o Pavlos It must be/must have been Pavlos. Prepi na ine / itan o Pavlos It must be/have been Pavlos.

Past tense on the modal verb eprepe ‘must have’ indicates a necessary situation in the past, and future tense tha eprepe ‘must have / could have’ conveys a counterfactual statement. Other means expressing epistemic necessity are the following mental state verbs: Lexical Expressions Verbs (3.42) Anangazome I am being forced Other Means Periphrastic impersonal modality (3.43) Ine anangeo It is necessary to Inherent modality (3.44) a. b.

Kanenas dhen ine ikanos Nobody is able to To sxedio dhiasosis tis elinikis ikonomias oloklironete se pende xronia The bail-out plan of Greek economy will reach its goals in five years.

62

Chapter Three

Phrases in (3.44a) and (3.44b) have modalized readings as (3.45a) and (3.45b) show: (3.45) a. Nobody is able to go to Heraklio in half hour. b. The bail-out plan can reach its goals in five years.

Sentences in (3.45a) and (3.45b) explicitly feature the modal element which seems to be implied in the verb in the sentences in (3.44a) and (3.44b) conveying epistemic necessity.

6 Embeddability of the Greek Modals The argument that epistemic modality does not contribute to the truthconditional content of a proposition has been put to the test, and a set of diagnostics has been suggested (see Leech 1971; Jackendoff 1972; Lyons 1977; McDowell 1987; Palmer 1990; Nuyts 1993, 2001; Adger & Quer 1997; Westmoreland 1995; Cinque 1999; Garrett 2000; Drubig 2001; Faller 2002; von Fintel 2003; among others): (a) scope (see Cohen, 1971; Wilson, 1975), (b) factives, and (c) assent/dissent patterns. Drubig (2001) also suggests a set of additional diagnostics such as: (a) negation, (b) questions (non -assertive contexts), (c) tense/aspect, (d) focus phenomena, which demonstrate evidence in favor of the non – truth conditional approach of epistemic modality. In this section, I will be concerned with epistemic modal expressions in Greek, and raise the obvious question: are epistemic modal expressions in Greek non truth-conditional, as many researchers have claimed? In Greek modality is regulated by the truth commitment in a speaker’s epistemic model. In support of this claim, I employ two diagnostics: (a) the truth value diagnostic, and (b) the scope and embeddability diagnostic. Each diagnostic comprises of the following tests respectively: (a) Known Falsity (Faller 2002), and Assent/Dissent (Papafragou 2000, 2006; Faller 2002, 2006; Matthewson et al. 2007), (b) Embeddability (Papafragou 2006; Portner 2007a, 2009), and Scope (Papafragou 2006; Portner 2007a, 2009). The first two tests7 check whether the meaning of epistemic modals can be questioned, rejected, or, (dis)agreed with. If yes, then we conclude that epistemic modal expressions contribute to the truth conditions of the proposition they are part of.

7

English examples are also grammatical for some English speakers.

Epistemic Modality

63

6.1 The Known Falsity Test Epistemic modal expressions cannot be used when the truth value of the prejacent is known, thus indicating that epistemics do not just contribute a comment on the prejacent, but rather to the truth content of the proposition as a whole. Consider the following contexts and possible epistemic expressions: (3.46) a.

b.

c.

Context: It’s snowing. # Bori na Might SUBJ # It might be snowing Context: It’s not snowing # Bori na Might SUBJ It might not be snowing Context: It’s not snowing Bori na Might SUBJ It might snow.

hionizi snowing.

mi NOT

hionizi snowing.

hionisi snow

6.2 The Assent/Dissent Test According to Lyons (1977), epistemic modality markers cannot be questioned, doubted, rejected or (dis)agreed with. Thus, the replies provided in examples (3.47a-d) show that the epistemic modal must does not contribute to the propositional meaning, especially when it is subjectively interpreted. For example: (3.47) John must be coming late to work. a. ? Is that so? (= Is it the case that this professor must be smart?) b. ? I don’t believe it. (= I don’t believe that this professor must be smart) c. ? That’s not true. (= It is not true that this professor must be smart) d. ? I agree. (= I agree that this professor must be smart)

However, in Greek, it is possible to doubt, agree, or disagree with the speaker’s subjective evaluation of epistemic possibility or necessity. The grammatical results in examples (3.48a-b) indicate that, contrary to Papafragou’s (2006) claims, subjective epistemic modality can be challenged, and is not externally inscrutable. For example:

Chapter Three

64

(3.48) O Janis prepi/bori na erhete arga sti dulia The John must/might SUBJ be coming late to work John must/might be coming late to work. a. Alithia; Really? b. Den to pistevo NOT it believe.1sg.NP I do not believe it! (3.49)

Afti i dhulia tha prepi/bori na jini avrio This job should / must / might be done tomorrow a. Alithia? Really? Ishii oti afti i dhulia tha prepi/bori na jini avrio? Is it true that this job should/must/might be done tomorrow? b. Dhen to pistevo Not it believe.1sg.INP I do not believe it. Den pistevo oti afto tha prepi/bori na jinni avrio Not believe.1sg.INP that this should/must/might SUBJ done.3sg.PNP tomorrrow I do not believe that this should/must/might be done tomorrow c. Afto dhen ine alithia That’s not true Dhen ine alithia oti afti i dhulia tha prepi/bori na jini avrio It is not true that this job should/must/might be done tomorrow d. Simfono I agree Simfono oti afti i dhulia tha prepi/bori na jini avrio I agree that this job should/must/might be done tomorrow

Examples in (3.49a-c) illustrate that what is being negated, questioned, and disagreed with is the proposition John must/might be coming late to work, and not the proposition John comes late to work, otherwise the utterer of the unmodalized expression would know that the statement John comes late to work is true. Recall the Known Falsity Test (KFT), in which if the truth value is known, then an epistemic claim clashes with the meaning of the prejacent. The two tests in conjunction show that epistemic modal operators target the propositional meaning, and contribute truth

Epistemic Modality

65

content. The same is true for the epistemic modal expression tha prepi/bori.

6.3 The Yes/No Interrogatives Test Epistemic interpretations of modals cannot take place sentence initially in yes/no interrogatives (Papafragou 2006). However, in Greek, this claim does not apply. For example: (3.50) a. Bori/Tha boruse aftos o nearos na ekane to egklima? b. ? Might/Could this young boy commit the crime? c. Yes/no d. Ne/Ohi (3.51) a. Na irthe o Janis? Could it be that John came? b. Ne/Ohi Yes/no

The yes/no interrogatives test displayed in examples (3.50) and (3.51) show that epistemic modal operators target the propositional meaning, as they ask about the possibility of a specific young man committing a crime or being present, and not if the young man actually committed the crime or was in fact present. Consider a better example in which the truth or validity of epistemic possibility is questioned: (3.52) Context: Two friends, Ann and Bill are talking about their friend, Paul, and his possible whereabouts on the night of the crime. Ann: O Pavlos (tha) prepi/bori na ine sto spiti Pavlos should/must/might be at home Bill: Ohi, den ine etsi. Akusa oti ehi mia dulia. O Pavlos den ehi epistrepsi akoma. No, that’s not true. I heard he had to run an errand. Pavlos hasn’t returned yet.

Contrary to the claims made in the literature, in Greek, the hearer can assess the data presented and evaluate the truth-conditional content of epistemic possibility or necessity. Tests have shown that epistemic modal expressions in Greek contribute to the truth-conditional content of the proposition they are in. In other words, epistemic modal expressions have truth-conditional content. A summary of the results follows:

Chapter Three

66

KNOWN FALSITY ASSENT/DISSENT YES/NO INTERROGATIVES

Non-Propositional No No

Propositional Yes Yes

No

Yes

Table 5 Truth Conditional Content in Main Sentences Let us now consider the second group of tests which involve embeddability and scope.

6.4 The Embeddability Test Epistemic modal expressions cannot be considered part of the propositional content of an embedded clause. Epistemic modals cannot be the complements of factive and/or telling predicates, as demonstrated in the following examples (taken from Papafragou 2006: 1690 – 1691): (3.53) Superman must be jealous of Lois. a. ? It is surprising that Superman must be jealous of Lois. b. ? Spiderman told me that Superman must be jealous of Lois. c. Spiderman told me he thought Superman must be jealous of Lois. d. Spiderman expressed the opinion that Superman must be jealous of Lois.

The infelicitous examples in (3.53a-b) are attributed to the fact that epistemic modality cannot carry a factive certainty or be reported as a statement of fact because it involves the speaker’s assessment of the proposition expressed by the utterance. However, examples (3.53c-d) illustrate a felicitous report contrary to what is expected by the embeddability test. The embeddability test of epistemic modality returns inconclusive results. Consider the following examples: (3.54) O Janis prepi/bori na apolithi apo tin eteria. John must/may be laid off from the company. Telling Predicates a. ?? I Dalia mu ipe oti o Janis prepi/bori na apolithi(-ke) apo tin eteria Dalia told me that John must/may be laid off from the company. Factive Predicates b. O Janis siniditopiise oti prepi/? bori na apolithi apo tin eteria John realized that he must/may be laid off from the company. c. O Pavlos ghnorize oti o Janis tha eprepe/? tha boruse na apolithi

Epistemic Modality

67

Pavlos knew that John must have been / might have been laid off from the company. Multiple-embedding d. I Dalia mu ipe oti nomizi oti o Janis ? bori / prepi na apolithi(-ke) apo tin eteria Dalia told me that she thinks that John may / must be laid off from the company e. I ghnomi tis Dalias ine oti o Janis bori / prepi na apolithi(-ke) apo tin eteria Dalia’s opinion is that John may / must be laid off from the company

In Greek, the embedding of epistemic modal auxiliary verbs and modal operators is possible, as example in (3.54) illustrates. However, when the impersonal verb prepi ‘it must’ is embedded it cannot convey epistemic necessity, only deontic necessity. On the contrary, its impersonal modal counterpart bori ‘it may’ conveys epistemic possibility, although not for all speakers, both in unembedded and embedded contexts such factive predicates. As for embedding under telling predicates, as is the case in example (3.54a), the modal expression with prepi ‘it must’ is felicitous and conveys an epistemic necessity meaning when the prejacent is in past perfective (PP), such as apolithike ‘he was fired’. The same observation is true for the examples in multi-embedding.

6.5 The Scope Test Epistemic modals cannot fall under the scope of a conditional. If they do, then epistemic modals contribute truth conditions (Cohen 1971; Wilson 1975; Papafragou 2006; Portner 2009). Let us explain the mechanics of this test, which are best illustrated with the following example: (3. 55) If Paul went to bed but John didn’t, we won’t be sleeping tonight!

Under which circumstances won’t we be sleeping tonight? (a) If Paul went to bed; (b) If John didn’t go to bed, or (c) If there is a contrast between the fact that Paul went to bed and the fact that John didn’t. The third circumstance does not contribute to truth-conditions; thus, the contrastive but is not truth-conditional. As Papafragou (2006) argued the environment inside the antecedent of a conditional cannot be an environment in which the speaker performs a mental evaluation of a proposition with respect to his/her belief-set. For example (adapted from Papafragou 2006: 1690):

Chapter Three

68

(3.56) a. ? If Max must / may be lonely, his wife will be worried.

subjective

b. If Paul may get drunk, I am not coming to the party. c. ? If Paul may get drunk, I will be mad at him.

objective Subjective

The test returns the same results for Greek epistemic modality, as well. For example: (3.57) a. ? An o Janis prepi/bori na fiji, i jineka tu tha stenahorithi If the John must/might SUBJ go, the wife his FUT sorry If John has to /might go, his wife will feel sorry. b. ? An o Janis prepi/bori na fiji, i jineka tu tha proaxthi If the John must/might SUBJ go, the wife his FUT be.promoted If John has to/might go, his wife will be promoted.

However, it is possible to find cases in which the truth of an epistemic modal is evaluated within the scope of a conditional. This is done with the use of the right context. Consider the following examples in English and Greek: (3.58) Ann: Bill:

If it is snowing in Sahara, Bechar must be buried in snow. That's not true. In 1979 it snowed on Sahara, but the snow melted within hours. So, it could be that Bechar is not buried in snow.

(3.59) Ann:

An hionizi sti Sahara, tote to Bechar (tha) prepi/bori na ehi thafti sto hioni Afto dhen ine alithia. To 1979 hionise sti Sahara, ala to hioni eliose mesa se ores. Ara, bori to Bechar na min ehi thafti sto hioni.

Bill:

In this dialogue, Bill, the hearer, evaluates the premises on which Ann, the speaker, has based her epistemic modal claims. Bill concludes that the premises are not sound, thus, Ann’s epistemic modal claim is false. Bill did not target the truth of the proposition Bechar is buried in snow but the epistemic necessity claim. Then Bill offers another epistemic claim with a lower degree of commitment It could be that Bechar is not buried in snow. Therefore, an epistemic modal expression in Greek, but in English as well, can be evaluated by the hearer within the scope of a conditional. A summary of the results follows:

Epistemic Modality

EMBEDDABILITY SCOPE

69

Non-Propositional

Propositional

No Yes

Yes No

Table 6 Truth Conditional Content in Embedded Sentences Thus far, I have discussed the embeddability of epistemic modals and found that embedding of epistemic modal auxiliary verbs and modal operators alike in Greek is possible. On the one hand, the impersonal verb prepi ‘it must’ is embedded is cannot convey epistemic necessity meaning but only deontic necessity. On the other hand, its impersonal modal counterpart bori ‘it may’ conveys epistemic possibility both in unembedded and embedded contexts such factive predicates, and the same holds for multi-embedding (e.g., section 3.3.4).

7 Complement Clauses Recall from chapter 2 (see example 2.24) that Greek distinguishes two types of complement clauses, indicative and subjunctive, in which the complementizer depends on the main verb. If the main verb conveys a commitment to the truth of the complement clause, then it is veridical and selects the indicative. Modal verbs are non-veridical operators; thus, we predict incompatibility and the lack of an epistemic reading (possibility or necessity), when they embedded under an indicative complement clause. For example: (3.60) Indicative Complements a. * Oli kserun oti/pos prepi/ bori i Maria na ine engios everybody know.3pl that must/might the Maria na.SUBJ is pregnant Everybody knows that Maria must/might be pregnant (after the b. * I Maria metaniose pu prepi/bori na ipe tin alithia the Maria regretted.3sg that must/might SUBJ told.3sg the truth Maria regretted that she must/might have told the truth. c. * O Janis anisihi mipos i Maria den prepi/bori na erthi The John worries.3sg that the Maria not must/might SUBJ comes.3sg John is concerned that Maria must/might not come.

Modal verbs cannot be embedded in indicative clauses, and neither can epistemic possibility nor necessity meaning when is being produced. Assertive contexts do not allow embedding of modality.

Chapter Three

70

Let us now consider the second embedding environment. If the verb in the main clause does not express commitment to the truth of the complement clause, it is non-veridical and selects the subjunctive. Subjunctive complements come after verbs of desire, wishing, hoping, planning, promising, requesting, modal verbs and they are introduced by na ‘that’ or ‘to’. Subjunctive clauses also follow verbs expressing perception, in which case they correspond to the –ing suffix in English. Let us examine the embedding of modals in these kinds of contexts with the following examples: (3.61) Subjunctive Complements a. * Elpizo na min prepi/bori na erthi o Janis * I hope that John must/might not come b. * Idha na prepi/bori na perpatai viastikos * I saw that he must/might walk hurriedly c. * Pistevo na prepi/bori na erthun tin epomeni evdomada * I believe that they must/might come next week

Embedding under verbs of perception is also ungrammatical, although for some speakers examples like the one in (3.61a) still conveys an epistemic possibility assessment, for example: (3.62) a. b.

Akusa oti prepi/bori na pulise to ktima I heard that he must/might have sold his property. ? Vlepo oti prepi/bori na erhete avrio ? I see that he must/might be coming tomorrow.

Embedding under verbs the verbs ksero ‘to know’ and matheno ‘to find out / learn’ also returns ungrammatical results, for example: (3.62) a. b.

* Ksero (oti/na) (prepi/bori) na pulise to ktima tu * I know that he must/might have sold his property. * Ematha (oti/na) (prepi/bori) na erhete avrio * I found out that he must/might be coming tomorrow.

We understand from the examples above that some non-veridical verbs can embed a modal verb such as akuo ‘to listen’, and vlepo ‘to realize’ while some cannot such as elpizo ‘to hope’, idha ‘to see’ and pistevo ‘to believe’. The reason is that embedding under non-veridical verbs provides contrastive examples (e.g., 3.61 vs. 3.62), as non-veridical verbs have different degrees of sensitivity to knowledge and belief. For example, idha ‘I saw’ in example (3.61b) carries a greater degree of commitment to truth than vlepo ‘I realize/ I conclude’ in example (3.62b). Finally, these

Epistemic Modality

71

examples show that even within the class of non-veridical complements, there are degrees of commitment to truth, which is evidence that nonveridicality is gradient (see Staraki, to appear). To conclude, modal verbs cannot be embedded in subjunctive clauses, and neither epistemic possibility nor necessity meaning can be produced. Contexts expressing desire, wishing, hoping, planning, promising, requesting, and perception do not allow embedding of epistemic modality. Rather, these contexts convey a deontic or ability reading.

8 The Protasis of the Conditional Constructions Conditional clauses are introduced by an, ean, ama ‘if’, ite .. ite ‘whether … or …’, efoson ‘provided that’, etsi ke ‘if so much as’, and se periptosi pu ‘in the event that …’, occasionally the subjunctive particle na can introduce a protasis of a conditional construction. Here I examine whether epistemic modals can be embedded in the protasis of conditional constructions. The purpose of this test, as I showed earlier (see the scope test) for the apodosis of a conditional construction, is to find out whether epistemic modals contribute to truth-conditions (see Cohen 1971; Wilson 1975; Papafragou 2006; Portner 2009). Here is an example: (3.63) a. * An (tha) prepi/bori na sinandithikan, tha tis edhose ta luludia * If they should/must/might meet, he should have given her the flowers. b. * An (tha) eprepe/boruse na sinandithikan, tha tis edhose ta luludia * If they should/must/might met, he should have given her the flowers.

Although in section 3.3 we found cases where the truth of epistemic modality was evaluated within the scope of the conditional apodosis, the examples with an epistemic modal embedded in the scope of a protasis are all ungrammatical. A summary follows:

EMBEDDABILITY SCOPE

Non-Propositional No

Propositional No

No

No

Table 7 Truth Conditional Content in Embedded Sentences

72

Chapter Three

9 Evidentiality in Greek Epistemic Modals In the previous sections, I provided evidence from both main and embedded environments that epistemic modal verbs and particles in Greek are not evidential operators because, first, the results from tests are inconclusive, and second, there are clear-cut cases in which modal verbs can embed under other propositional operators. However, we cannot say that epistemic modal expressions relate to evidentiality because they share similar concepts. As Nuyts (2001b) argued evidentials and epistemic modals suggest inference, indirect, or direct evidence, hearsay, and more generally, both indicate the quality of the nature of evidence (information) on which a speaker bases his/her epistemic assessment. In this section, I argue that epistemic modality in Greek incorporates an evidential component as part of its meaning (for English see Nuyts 2001a; von Fintel & Gillies 2007b, 2010; among others). I discuss the internal structure of epistemic modal verbs and particles in Greek, such as prepi ‘it must’, bori ‘it may/might’, tha prepi ‘should’ and tha ‘will’, and show that epistemic modality in Greek involves two components: the evidential component (inference based on indirect evidence) and the nonveridical component (lack of commitment to truth). With regard to the first component, I will show that the variation in the meaning and the distribution of epistemic modals cannot be accounted for unless we assume that epistemic modals are sensitive to different types of indirect evidence, and thus to a wider set of modal base types determining the domain of possibilities that a modal will quantify over. Indirect evidence, as I will prove, is a presupposition (Staraki 2012, to appear) on the modals definition. Thus, indirect evidence seems to be an important component, which is incorporate in the set-up of a modal base (see von Fintel and Gillies 2010) as a presupposition (for Greek see Staraki 2012, to appear). With regard to the second component, I will argue that epistemic modality in Greek employs the grammatical means to mark the lack of epistemic commitment: the embedding of a CP introduced by the nonveridical particle na, which typically introduces embedded complements with the subjunctive. Extending in a modest way Giannakidou’s (2009) Temporal Polarity analysis (TPA henceforth), I will explain the characteristics of epistemic modality in Greek and show that it is regulated by non-veridicality (Giannakidou 1998).

Epistemic Modality

73

9.1 The Evidential Component Before proceeding with the investigation of the internal structure of epistemic modality in Greek, it is necessary to specify how the concepts of epistemic modality and evidentiality are understood in this study. I consider the standard definitions for both concepts (see chapter 1, and Kratzer 1977, 1981, 1991; Portner 2009; among others). Epistemic modality refers to the evaluation of a proposition on the basis of an epistemic modal base or accessibility relation. For example: (3.64) a. John must/may [be the murderer] b. O Janis (tha) prepi/bori [na ine o įolofonos] Epistemic necessity: (THA) PREPI ([Op. w’ BEST (f, g, w): ‘John is the murderer in w]) Epistemic possibility: BORI ([Op. w’ BEST (f, g, w): ‘John is the murderer in w])

On the one hand, epistemic necessity conveys that all the evidence we have leads to the inescapable conclusion that the statement 'John is the murderer' is true iff for every epistemically accessible possible world 'John is the murderer' is true in w. On the other hand, epistemic possibility conveys that there is evidence that the statement 'John is the murderer' is true iff there is some epistemically accessible possible world such that 'John is the murderer' is true in w. For evidentiality, I assume that it is the marking of the nature of evidence (see Friedman 1986, 2003; Joseph 2003; Aikhenvald 2005; among many others) and show that, although Greek does not have a designated grammatical category to express evidentiality (Joseph 2003), other lexical means are used that indicate the mode of acquiring the evidence (e.g. inference). Examples of evidentiality in example (3.65) assert the reliability and the nature of information; some sort of information is being discussed about John being the murderer. Evidentiality distinguishes two types: (a) direct (based on visual and auditory evidence), and (b) indirect (based on nonvisual, inference, quotative, reportive evidence): (3.65) a. It is said b. Ipan

[that [oti

John is o Janis

the ine

murderer] o įolofonos]

74

Chapter Three

Giannakidou (2011) argued that the particle tha ‘will’ is both a modal and an evidential. The particle tha ‘will’ is an epistemic modal operator with present (n) perspective, with the time of assessment being now (n) (see chapter 2). Furthermore, it has the force of a necessity modal such as must, and like must, it also has evidential use (with the perfective past). For example: (3.66) a. O Pavlos malon tha ine The Pavlos probably FUT is Pavlos is probably home this time. b. O Pavlos isos {*tha/na} ine The Pavlos perhaps {*FUT/SUBJ} Perhaps Pavlos is home this time. c. O Pavlos malon (tha) prepi na The Pavlos probably (FUT) must SUBJ Probably Pavlos should be home this time. d. O Pavlos isos (tha) prepi na The Pavlos perhaps (FUT) must SUBJ Perhaps Pavlos should be home this time.

sto spiti tetia ora the house this time sto spiti tetia ora is the house this time ine sto spiti tetia ora is the house this time ine sto spiti tetia ora is the house this time

The modal base ŀf(w) for the particle tha ‘will’ selects all those worlds from a set of evidence that are then ordered by the ordering source g(w). In the case of example (3.67) the modal base is epistemic w  W: w  ŀf(w) (see Kratzer 1981), and includes all those worlds in which we have evidence about Pavlos being home at this time of day, ordered by the ordering source g(w). For example: (3.67) O Pavlos tha ine sto spiti tetia ora The Pavlos FUT is the house this time Pavlos must be home this time. THA ([Op. w’ BEST (f, g, w): ‘Pavlos be at home in w])

Modal particle tha ‘will’ quantifies universally over a subset of worlds in the modal base (Giannakidou 2011) and over those worlds that we have some sort of evidence for (see Staraki 2012), thus, the particle tha ‘will’ is non-veridical (see Staraki 2010a on the non-veridicality of particle tha ‘will’ and Giannakidou 1998, 1999, 2011 on the non-veridicality of modal particles and tha ‘will’). The particle tha ‘will’ is also an evidential. As an evidential, Giannakidou (2011) argued, tha ‘will’ signals an inference based on indirect evidence. For example, assume the following:

Epistemic Modality

75

(3.68) Context: You know that Melina is an excellent student. She aces every test she takes and she always has the highest grades in class. Today she had to take another test. You utter: a. I Melina tha ehi ghrapsi arista The Melina FUT has written excellent Melina should have aced the test THA ([Op. w’ BEST (f, g, w): ‘Melina aced the test at time t < n’ = 1])

The epistemic necessity operator tha ‘will’ conveys that all the evidence we have about Melina taking the test leads to the inescapable conclusion that Melina aced the test. Thus, the modal base should be determined by evidence or some sort of information, for example, evidence about high grades and excellent results on tests that Melina has taken.

9.2 The Types of Evidence Giannakidou’s (2011) conclusions about tha ‘will’ can be extended, with slight changes, over the set of epistemic modal operators in Greek. The first change I suggest is that tha ‘will’ is not an evidential as Giannakidou (2011) suggested. Tha ‘will’ is a modal operator that incorporates in its semantic set up an evidential component. In other words, it incorporates an evidential signal that affects the modal base apparatus (pace von Fintel & Gillies 2010). The suggestion made here has the theoretical advantage of acknowledging the truth-conditional contribution of epistemic modality and at the same time, taking into account the evidential characteristics of epistemic modality in Greek, as we will see later. I will capitalize on this conclusion later in this discussion when I argue that a set of presuppositions contributes to the semantic set-up of a conversational background for epistemic modals in Greek (see Staraki 2012). Epistemic modal operators are not equally sensitive to all types of evidence (see Staraki 2012). This is a new observation for Greek. A first observation is that epistemic modal operators are sensitive only to indirect evidence, and not to direct evidence. A second observation is that epistemic operators differ in the degree of sensitivity to indirect evidence. In the discussion that follows, I distinguish the type of information and the degree of commitment that the set of epistemic modal operators exhibit, thus making clear what the evidential signal of epistemic modality in Greek is. Formalization follows. The type of reasoning that is illustrated in example (3.69) is based on evidence that can be characterized as circumstantial reasoning based on

76

Chapter Three

external evidence. Imagine you witnessed a car accident in which a pedestrian was hit. In this case and based on external sensory evidence, you can only use the prepi ‘must’ and tha prepi ‘should’ modal operators. For example: Circumstantial Inference (3.69) a. Prepi na ine zontanos; ton idha na Must SUBJ is.3sg alive; him saw.1sg.PP SUBJ He must be alive; I saw him moving. b. tha prepi na ine zontanos; ton idha na FUT must SUBJ is.3sg alive; him saw.1sg.PP SUBJ He should be alive; I saw him moving.

kinite! moving.3sg.INP kinite! moving.3sg.INP

c. ? tha ine zontanos; ton idha na kinite! FUT is.3sg alive; him saw.1sg.PP SUBJ moving.3sg.INP He should be alive; I saw him moving. d. ? Bori na ine zontanos; ton idha na kinite! Might SUBJ is.3sg alive; him saw.1sg.PP SUBJ moving.3sg.INP He might be alive; I saw him moving. Inferential Interpretation: Modal Base (ŀf(w)): { The pedestrian that was hit is moving} Ordering Source (g(w)): { When someone moves is alive}

The modal operators prepi ‘must’, tha prepi ‘should’, and tha ‘will’, can also express an inferential process in which any external, observable evidence is lacking and the speaker bases his/her reasoning on previous experience or general world knowledge (see also Mari 2009 for Italian future), i.e., in contexts that can be characterized as generic. Imagine, for example, you are picked up every morning at 7am by Maria; around that time the bell rings. Since you know that Maria arrives around that time you utter the following statement: Generic Inference (3.70) a. Prepi na ine i Maria; afti me pijeni sti dulia Must SUBJ is.3sg the Maria; she me drop.off.3sg to.the work It must be Maria; she drops me off at work. b. tha prepi na ine i Maria; afti me pijeni sti dulia FUT must SUBJ is.3sg the Maria; she me drop.off.3sg to.the work It should be Maria; she drops me off at work. c. tha ine i Maria; afti me pijeni sti dulia FUT is.3sg the Maria; she me drop.off.3sg to.the work It should be Maria; she drops me off at work. d. ? Bori na ine i Maria; afti me pijeni sti dulia Might SUBJ is.3sg the Maria; she me drop.off.3sg to.the work

Epistemic Modality

77

It might be Maria; she drops me off at work. Inferential Interpretation: Modal Base (ŀf(w)): { Bell rang around 7am every morning} Ordering Source (g(w)): { Maria picks me up at 7am every morning}

The epistemic modal operator bori ‘it might/may’ is grammatical in the context that can be characterized as conjectural. In this case, evidence is lacking. Imagine that it is midnight and you are not expecting anyone to call you; in this case, the modal operators prepi ‘must’, tha prepi ‘should’, and tha ‘will’ are not grammatical. Consider the following examples: Conjectural Inference (3.71) a. # Prepi na ine i Maria, an ke dhen perimeno kanena. Must SUBJ is.3sg the Maria, if and not expect anyone It must be Maria, though I don’t expect anyone. b. # tha prepi na ine i Maria, an ke dhen perimeno kanena. FUT must SUBJ is.3sg the Maria, if and not expect anyone It should be Maria, though I don’t expect anyone. c. # tha ine i Maria, an ke dhen perimeno kanena. FUT is.3sg the Maria, if and not expect anyone It should be Maria, though I don’t expect anyone. d. Bori na ine i Maria, an ke dhen perimeno kanena. Might SUBJ is.3sg the Maria, if and not expect anyone It might be Maria, though I don’t expect anyone. Inferential Interpretation: Modal Base (ŀf (w)): {Phone rang in an unexpected time} Ordering Source (g (w)): { ‡ }

Epistemic modal verbs are not used when we have direct evidence; for example, assume that it is snowing outside and you are looking out the window. You cannot use the modal verbs prepi ‘must’ and bori ‘may’ or their variants with the particle tha ‘will’, since all of them, as we saw, correspond only to indirect inference. Consider the following examples: Direct Evidence (3.72) a. # Prepi na hionizi, to vlepo Must SUBJ snowing.3sg, it see.1sg It must be snowing, I see. b. # tha prepi na hionizi, to vlepo FUT must SUBJ snowing.3sg, it see.1sg It should be snowing, I see. c. # tha hionizi, to vlepo FUT snowing.3sg, it see.1sg It should be snowing, I see.

Chapter Three

78

d. # Bori na hionizi, to vlepo Might SUBJ snowing.3sg, it see.1sg It might be snowing, I see.

From the previous examples, I conclude that Greek modals encode distinctions of information source (circumstantial, generic, and conjectural) and impose certain constraints that correspond to the type of evidence. The results follow: Evidence Modals Prepi ‘must’ Tha prepi ‘should’ Tha ‘will’ Bori ‘might / may’

Circumstantial

Generic

Conjectural

Direct

9

9

#

#

9

9

#

#

?

9

#

#

?

?

9

#

Table 8 Types of Evidence in Greek

I assume that there are two constraints imposed on the conversational background (modal base) of epistemic modals in Greek, and I treat them as presuppositions. Their satisfaction determines the conversational background; hence, the assignment of truth values is relative to context c, world w and time t (see Staraki 2012, to appear). The first presupposition strictly involves the kind of evidence that a modal is sensitive to. Epistemic modals in Greek, as we saw, express the usual quantificational meaning relative to a modal base that is determined by context (see Kratzer 1977, 1981, 1991). Evidence (indirect evidence specifically), provided by context, determines the accessible worlds. Direct evidence nullifies quantification over possible worlds. Thus, the presupposition (see Staraki 2012) of indirect evidence has to be part of an epistemic modal’s semantic denotation in Greek (pace von Fintel & Gillies 2010). The distinguishing characteristic between direct and indirect evidence is settledness (for example, facts established or decided beyond doubt), and one way to define the concept of direct settledness is the following generalization by von Fintel & Gillies (2010: 374):

Epistemic Modality

79

(3.73) A kernel K directly settles whether P iff either X Ž P or X ŀ P = ‡ for some XK Precondition (von Fintel & Gillies 2010: Definition 4): K is a kernel for BK, BK is determined by the kernel K, only if: i. K is a set of propositions (if PK then PŽW) ii. BK = ŀK where BK = modal base

I propose that statement in (3.73) is a presupposition on the indirectness requirement of epistemic modals in Greek. This states that a modal base ŀf(w) directly settles whether the prejacent p is true iff either there is a proposition in ŀf(w) that entails p (X Ž P) or there is a proposition in ŀf(w) that contradicts p (X ŀ P = ‡). Thus, the ungrammaticallity of examples in (3.72) that I repeat next can be explained as follows: Direct Evidence (3.74) a. # Prepi na hionizi, to vlepo Must SUBJ snowing.3sg, it see.1sg It must be snowing, I see. b. # tha prepi na hionizi, to vlepo FUT must SUBJ snowing.3sg, it see.1sg It should be snowing, I see. c. # tha hionizi, to vlepo FUT snowing.3sg, it see.1sg It should be snowing, I see. d. # Bori na hionizi, to vlepo Might SUBJ snowing.3sg, it see.1sg It might be snowing, I see. Presupposition on Settledness: K = {It is snowing, I see it is snowing} Modal Base (ŀf (w)): {It is snowing} XŽP

K (Kernel) contains the set of propositions {It is snowing, I see it is snowing} that entail {It is snowing}, thus, K (Kernel) directly settled the modal base, and the presupposition on indirectness of epistemic modals in (3.74) was not met. Therefore, the following constraint holds for epistemic modals in Greek (see Staraki 2012): (3.75) ۤሺIሻ‫ۥ‬c, f, g, w = 1 iff K does not directly settle ۤI‫ۥ‬c, f, g, w

80

Chapter Three

An epistemic modal, according to the condition (3.75), is directly settled, when the kernel of the modal base entails or contradicts the propositions in the modal base. The second presupposition, I propose, involves the conversational background as well: the type of evidence that each epistemic modal is sensitive to, which reveals the degree between a stronger and weaker epistemic commitment (Staraki 2012), and the speaker's most objective reasoning (circumstantial inference) to most subjective reasoning (conjecture). For a more detailed analysis see Staraki (2012). I assume (see Kratzer 1977, 1981, 1991; Giannakidou 1998) that indirect evidence will be decomposed into epistemic anchors D, a free variable defined by the context which we will see in more detail below, and a domain fixing function f(D). Anchors represent individuals or situations, but I will also assume that they can represent evidence, and domain fixing functions will map anchors to ordered sets of possibilities. To this, I add necessities as well. Thus, I propose that the lexical entries for epistemic modals of possibility and necessity are the ones in definitions (3.76), as follows: (3.76) Epistemic Possibility Operator: ۤሺIሻ‫ۥ‬c, f, g, w = Oq.OD.w’ BEST(f(D), g, w)ŀۤሺIሻ‫ۥ‬c, f, g, w z ‡ Epistemic Necessity Operator: ۤሺIሻ‫ۥ‬c, f, g, w = Oq.OD.w’ BEST(f(D), g, w)ŀۤሺIሻ‫ۥ‬c, f, g, w z ‡

Some representative examples follow: (3.77) a. Prepi na ine i Maria; afti me pijeni sti dulia Must SUBJ is.3sg the Maria; she me drop.off.3sg to.the work It must be Maria, she drops me off at work. PREPI ([Oq.OD.w’ BEST(f(D), g, w): ‘It is Maria at time t’ = 1]) b. Bori na ine i Maria, an ke dhen perimeno kanena. Might SUBJ is.3sg the Maria, if and not expect anyone It might be Maria, though I don’t expect anyone, BORI ([Oq.OD.w’ BEST(f(D), g, w): ‘It is Maria at time t’ = 1])

In the examples above, the epistemic anchor D dictates the kind of information that will be supplied to the domain fixing function f(D). However, as we saw earlier, epistemic modals in Greek are not equally sensitive to evidence, and the epistemic anchor is indeterminate the way it is. Formal representations, as shown in (3.77), although they correctly include an epistemic anchor D sensitive to indirect evidence, do not predict the particular sensitivity of each of the epistemic modal operators. Thus, I

Epistemic Modality

81

propose a set of presuppositions for epistemic modal operators in Greek that regulate the propositions that an epistemic anchor will pick up. However, at this point, one might ask why and whether the evidential component of epistemic modals is a presupposition or can be treated as such. The motivation behind this choice is that, as I showed in section 3.3, epistemic modality contributes to the truth-conditional content of the sentence that it is part of. In other words, whatever the evidential meaning is within epistemic modality’s semantic content, it cannot be targeted with tests (embeddability, yes/no interrogatives, etc.) questioning the truth or validity of an epistemic modal expression. Thus, I think that the evidential component of epistemic modality, at least in Greek, acts like a presupposition8. The quantificational force of epistemic modals in Greek is determined by context, except in the case of tha ‘will’, which has only a universal force and is lexically determined (see Giannakidou 2011). Indirect evidence imposes restrictions on the conversational background of epistemic modals as we saw in examples (3.69 – 3.70). I treat these restrictions as presuppositions that constrain the accessible worlds of an epistemic modal. The presupposition constraint is part of the denotation of an epistemic modal and is sensitive strictly to indirect evidence that is determined by the settledness condition displayed in (3.73). I start with the semantics of necessity operators prepi ‘must’ and tha prepi ‘should’, for example: (3.78) Presuppositions on PREPI / THA PREPI (epistemic): The epistemic anchor D of ۤሺ‫݄ܽݐ‬ሻ‫ ۥ݅݌݁ݎ݌‬c, f, g, w is defined if ŀf(w) is circumstantial. A circumstantial background is a function f such that for all worlds w in the domain of f, f(w) contains a body of external perceptual evidence that surrounds a situation or an event and has a counterpart in all w’ ŀf(w), and a doxastic ordering source gd(w). If defined, then: ۤ‫܀۾‬۳‫۾‬۷Ȁ‫܂‬۶‫܀۾ۯ‬۳‫۾‬۷ሺIሻ‫ۥ‬c, f, g, w = Oq.OD.w’ BEST(f(D), gd, w)ŀۤሺIሻ‫ۥ‬c, f, g, w z ‡

This definition of presupposition on the epistemic necessity operators prepi ‘must’ and tha prepi ‘should’ incorporates previous claims I have made in this study and elsewhere (e.g., Staraki 2010a, 2012). Now, we have an epistemic necessity operator that is sensitive to circumstantial

8

This is a subject for future research.

82

Chapter Three

evidence, and an epistemic anchor that picks up the set of circumstantial propositions plugging them into the domain fixing functions f(D). The presuppositions of prepi ‘must’, tha prepi ‘should’, and tha ‘will’ are the following: (3.79) Presuppositions on PREPI/THA PREPI/THA (epistemic): The epistemic anchor D of ۤ‫݅݌݁ݎ݌‬Ȁ‫݅݌݁ݎ݌݄ܽݐ‬Ȁ‫ ۥ݄ܽݐ‬c, f, g, w is defined ŀf(w) is generic. A generic background is a function f such that for all worlds w in the domain of f, f(w) contains a body of previous personal experience or general world knowledge in w, and a doxastic ordering source gd(w). If defined, then: ۤ‫܀۾‬۳‫۾‬۷Ȁ‫܂‬۶‫܀۾ۯ‬۳‫۾‬۷Ȁ‫܂‬۶‫ۯ‬ሺIሻ‫ۥ‬c, f, g, w = Oq.OD.w’ BEST(f(D), gd, w)ŀۤሺIሻ‫ۥ‬c, f, g, w z ‡

Finally, the presupposition on the epistemic anchor of possibility operator is the following: (3.80) Presuppositions on BORI (epistemic): The epistemic anchor D of ܾۤ‫ ۥ݅ݎ݋‬c, f, g, w is defined if ŀf(w) is conjectural. A conjectural background is a function f such that for some worlds w in the domain of f, f(w) contains a body of incomplete evidence in w, and an empty ordering source g‡(w). If defined, then: ۤ۰‫܀۽‬۷ሺIሻ‫ۥ‬c, f, g, w = Oq.OD.w’ BEST(f(D), g‡, w)ŀۤሺIሻ‫ۥ‬c, f, g, w z ‡

Epistemic modals in Greek encode both the source and the type of evidence and force (existential / universal) (contra Matthewson et al 2007 who argue that modals should choose between either encoding source (St’át’imcets evidentials) or force (English modals)). I have also argued that the conversational background (modal base) of epistemic modal verbs in Greek incorporates an evidential component that is further constrainted by a set of presuppositions. The first presupposition distinguishes between indirect and direct propositions, and the second presupposition further delimits the set of propositions that a modal is sensitive to. The propositions determined by those two constraints on the modal base feed the epistemic anchor, which represents the context sensitivity of epistemic modality. Finally, those propositions picked out by the epistemic anchor will be mapped to an ordered set of possibilities. In this section, I showed that the main feature of epistemic modality is the lack of epistemic commitment, and that the lack of commitment to truth of the prejacent is regulated by non-veridicality (see Giannakidou

Epistemic Modality

83

1998 and sequel; Staraki 2010a, 2012). Hence, non-veridicality regulates epistemic modality. In addition, I argued that since modal operators are non-veridical operators, they select temporal polarity items such as naclauses: linguistic items which do not commit an agent to the truth of p.

9.3 The Epistemic Commitment According to Palmer (1986: 53 – 54): “… evidentials are not indications of some objective modality, but are subjective in that they indicate the status of the proposition in terms of the speaker’s commitment to it”. In the previous sections, we considered the interpretation of modal operators as sensitive to inferential evidence that indicate the type of information: circumstantial, generic, and conjectural. The next component of epistemic modality I discuss is the degree of speaker’s epistemic commitment and its relation to evidential and modal meaning. Epistemic modality in Greek incorporates an evidential component sensitive to indirect evidence. That implies lack of commitment to the truth of the prejacent proposition on speaker’s part. Consider the following examples (3.81a – d) that contrast to (3.81e) in terms of commitment to truth: (3.81) a. # Prepi na ine kali i tenia, ime sighuros Must SUBJ is.3sg good the movie, I.am sure. The movie must be good, I am sure. b. # tha prepi na ine kali i tenia, ime sighuros FUT must SUBJ is.3sg good movie, I.am sure. The movie should be good, I am sure. c. ?? tha ine kali i tenia, ime sighuros FUT is.3sg good the movie, I.am sure. It should be a good movie, I am sure. d. # Bori na ine kali i tenia, ime sighuros Might SUBJ is.3sg good movie, I am sure. It might be a good movie, I am sure. e. Ine kali i tenia, ime sighuros is.3sg good movie, I am sure. It’s a good movie, I am sure.

We observe a difference between examples (3.81a-d), which encode inference based on indirect evidence (thus lack of commitment to the truth of the embedded) na-clauses, and example (3.81e), which makes a claim about the positive qualities of a movie, implying a commitment to the truth of what the speaker’s statement. The set of examples in (3.81a-d) are contrastive to the continuation egho ime sighuros 'but I am sure'. There is

84

Chapter Three

no contrast in the last example (3.81e). The only exception is the modal particle tha ‘will’ which can be considered a weak epistemic modal, which does not contradict the continuation egho ime sighuros 'but I am sure'. The lack of commitment on the actualization of a situation can be formally represented in terms of non-veridicality. Non-veridicality implies non-commitment (see chapter 2). It is expressed formally in the following definition (from Giannakidou 2009) repeated here in definition (3.82): (3.82) non-veridicality for propositional operators (Giannakidou 2009): i. A propositional operator F is veridical iff Fp entails or presupposes that p is true in some individual’s model M(x); p is true in M(x), if M(x)  p. ii. If (i) is not the case, F is nonveridical. iii. A nonveridical operator F is antiveridical iff Fp entails not p in some individual’s model. When this is the case, M(x) ŀp = ‡ (3.83) Epistemic model of an individual (Giannakidou 1998): An epistemic model of an individual x, ME(x), is a set of worlds w’ accessible from a world w, compatible with x’s beliefs in w.

The definition of veridicality involves the notion of truth entailment in an individual’s epistemic model. A propositional operator F is veridical iff Fp entails p meaning, whenever Fp p is true, then p is true. In a propositional operator Fp p is nonveridical if p does not hold. This means that we are not committed to the truth entailment of p; p may or may not be true. In essence, nonveridicality does not entail the falsity of p. In brief, a proposition is veridical when it presupposes or entails that p is true in an individual’s epistemic model, as in statements (3.82i and 3.83), and a proposition is non-veridical, if the proposition entails that not p in an individual’s epistemic model, as in statements (3.82ii and 3.83). As I mentioned earlier (see chapter 2), I assume that modal auxiliary verbs (Giannakidou 1998) and the particle tha ‘will’ are non-veridical (see Staraki 2010a, Giannakidou 2011). I further assume that an epistemic model is a set of worlds anchored to an individual (the individual anchor; Farkas 1992a, b; Giannakidou 2009) that represents worlds compatible with what an individual believes. Extending her assumptions on epistemic models, I assume that, when a na – clause embeds under an epistemic modal operator, the concept of the epistemic model will be relevant. I propose the following generalization: (3.84) Nonveridicality of Modals (first approach): ۤሺԄሻ‫ ۥ‬c, ME(x) = 1 ե ۤԄ‫ ۥ‬ME(x) = 1

Epistemic Modality

85

The definition (3.84) states that at least one epistemic agent (the speaker or the subject of the modal verb) is not committed to the truth of the complement na-clause. However, the statement is quite general and does not explain the non-veridicality that is relevant to epistemic modal verbs. Epistemic modal verbs are sensitive to evidence and information in determining the set of possible worlds they will quantify over, and represent possible states of affairs: alternative ways an actual world might be or could have been. Each epistemic modal verb quantifies over a set of alternatives according to what an agent believes is epistemically possible or necessary to be realized. To capture this, I will employ an agent’s epistemic model and propose (see Giannakidou 2009), that it includes the set of possible worlds MPW(x) (see Staraki 2010a, to appear). Evidence partitions an agent’s epistemic model MPW(x) into two subsets, distinguishing those possible worlds W1 compatible with the evidence relevant to each epistemic modal verb, from those that are not compatible with the evidence in W2. The worlds W1 that are compatible with the relevant evidence are the accessible worlds. According to these premises, I propose the following formalization of partition in an agent’s epistemic model: (3.85) Epistemic Necessity in an Agent’s Epistemic Model MPW(x): a. MUST / (THA) PREPI ([Op. w’ W1 & W1  MPW(x): ۤ’‫ۥ‬c, f, g, w = 1 in w’] o W1 Ž p) For all worlds w’ that are members of W1 and W1 is a subset of MPW(x) s.t. the complement of W1 is true in w’ then W1 is a subset of p b. MUST / (THA) PREPI ([Op. w’’ W2 & W2  MPW(x): ۤ’‫ۥ‬c, f, g, w = ‡ in w’’] o W2 ŀ p =‡) For all worlds w’’ that are members of W2 and W2 is a subset of MPW(x) s.t. the complement of W2 is false in w’’ then the intersection of W2 with p is an empty set. (3.86) Epistemic Possibility in an Agent’s Epistemic Model MPW(x): a. MAY / BORI ([Op. w’ W1 & W1  MPW(x): ۤ’‫ۥ‬c, f, g, w = 1 in w’] o W1 Ž p) For some worlds w’ that are members of W1 and W1 is a subset of MPW(x) s.t. the complement of W1 is true in w’ then W1 is a subset of p. b. MAY / BORI ([Op. w’’ W2 & W2  MPW(x): ۤ’‫ۥ‬c, f, g, w = ‡ in w’’] o W2 ŀ p =‡) For some worlds w’’ that are members of W2 and W2 is a subset of MPW(x) s.t. the complement of W2 is false in w’’ then the intersection of W2 with p is an empty set.

86

Chapter Three

One the one hand, worlds that are not compatible W2 ŀ p =‡ with the evidence in an agent’s epistemic model MPW(x) in the actual world are not included in the set of propositions p. On the other hand, worlds that are compatible W1 Ž p with the evidence in an agent’s epistemic model MPW(x) in the actual world are included in the set of propositions p. Therefore, we accomplished the following two: first, we captured nonveridicality of epistemic modal verbs in an agent’s epistemic model, and second, we predict that modal verbs in general select complements that are non-veridical such as na-clause (subjunctive). Subjunctive na-clauses are non-veridical polarity items meaning that they will embed under operators that will have same characteristics. Here, I showed that modal verbs are nonveridical operators, thus they will select for subjunctive na-clauses, and not let’s say indicative oti-clauses (see chapter 2).

10 Conclusions I summarize the results of this chapter as follows. First, epistemic modals in main contexts contribute to the truth conditional content of the proposition that they are in. In other words, epistemic modal expressions have truth-conditional content. Second, epistemic modals embedded under factive and telling predicates, or multiple embedding return inconclusive results. In general, embedding of epistemic modal auxiliary verbs and modal operators is possible in Greek. We can also find cases in which the truth of epistemic modality is evaluated within the scope of a conditional antecedent. Third, embedding of modals under complement clauses and protasis of a conditional construction is impossible. Fourth, epistemic modality in Greek is not an evidential operator, but rather it incorporates an evidential component that is based on indirect evidence and is formally represented in the semantic set-up of an epistemic modal as a presupposition.

CHAPTER FOUR MODALITY AND TIME

1 Introduction In this section, I lay out the essential questions that concern the study of future reference. My aim is to present all the elements that affect the representation, interpretation, and eventually the modeling of what the future reference is and how a unified formal analysis of future morphemes such as the Greek tha and the English will is motivated. Future reference in both English and Greek standardly conveys both temporal and modal interpretation: (4.1) a. O Pavlos tha fighi avrio The Paul FUT leave.3sg tomorrow b. Paul will leave tomorrow

temporal

(4.2) a. O Pavlos tha efighe The Paul FUT left.3sg b. Paul will have left already

epistemic

idhi already

Greek and English use what we call future morphemes (FUT) that precede the tensed verb. FUT morphemes are followed by the non-past or past tense and yield temporal and epistemic readings. The very first question regarding the FUT morpheme is whether it is ambiguous between a pure tense (shifting the event time forward) and a modal (conveying a probability evaluation) interpretation, or whether the examples are instances of modal FUT morphemes that derive a temporal reading. Quite generally, the answers to this question fall under either (1) the approach that treats future and modal uses as distinct semantic objects, or (2) the approach that merges future and modal uses as a single realization of the future reference. I adopt the thesis that FUT morphemes are modals and their various readings derive via a set of conditions and a gradient quantification, as I discuss in detail in section 4. Intuitively, the future reference is associated with the unknown and the human propensity of predicting, guessing, and discovering cues and hints

88

Chapter Four

that will determine, as much as possible, the unknown. This process of moving from the unknown to the closest possible known is thought to have an inherent epistemic nature, meaning that, if I do not know or believe in something, and if I do not have premises, more formally, on which to base my reasoning, I cannot produce judgments, express claims or draw conclusions, and predict or evaluate the future, the unknown. This theoretical precondition naturally has its drawbacks and gives birth to both approaches mentioned above. Therefore, for some authors (see Kissine 2008), a purely temporal approach excludes modal readings altogether because it is impossible to ever have solid premises (epistemic) to predict the future or to equalize a deontic-factive reading (see Ziegeler 2006; Kissine 2008) with the future which is actually non-factive. The only thing we can do is plan the future; thus, future reference is simply a tense. Other readings can be derived via pragmatics (Kissine 2008). However, the logical structure of future reference as I propose here can, in fact, be based on mere observations. Future reference features the following reasoning pattern: provided a proposition p and the knowledge or the belief that q o p, concludes FUT (p). The FUT morpheme is, as I will show, the grammatical indication of being uncertain, as something else could be the actual cause of q as well. A modal approach (Jespersen 1924; Prior 1967; Bertinetto 1979; Smith 1978; Haegeman 1983; Thomason 1984; Comrie 1985; Palmer 1987; 1986; Enç 1996; Sarkar 1998; Copley 2002; Condoravdi 2002; Squartini 2004; Kaufmann 2005; Jaszczolt 2006, among many others) on the contrary, accepts that knowledge is our theoretical tool to carve out the future; thus, we become able to determine what is possible and what can be predicted. Mari (2009) argued that the future is incompatible with noninferential evidence (see also Giannakidou & Mari 2013a/b, 2014, for a development of this claim), which means that the future is inherently epistemic. This theoretical stance, however, represents a rather narrow perspective, especially if we take into account the examples (4.3-4.5) that follow. A unified approach is the objective, but how can one bring the interpretational variety of future reference under one concept? Yes, the future reference is non-factual but that does not mean that it is only epistemic or predictive (temporal) by default. The future reference involves a wide range of interpretations other than the epistemic and the predictive. For example: (4.3) a. O Pavlos tha fighi tora amesos The Paul FUT leave.3sg right now b. Paul will leave right now

directive

Modality and Time

89

(4.4) a. O Pavlos tha pji ke trianta bires The Paul FUT drink.3sg and thirty beers b. Paul will be able to drink 30 beers

ability

(4.5) a. Polemi tha Wars FUT b. Wars will be waged

generic

dhieksaghonde waged.3pl

Therefore, my goal in this paper is to show that the variety of interpretations can indeed be part of a general theory that brings together the different realizations of future reference. This claim is based on the hypothesis that future reference underlies the concept of normative conditions that generally conforms to the following reasoning template: if such and such conditions hold then it is appropriate or it is associated with such and such result. The result, it can be argued, is the criterion of how close to the truth a proposition p comes. The concept of proximity to the truth will prove very important when FUT morphemes are analyzed within a gradience model. The interpretational questions about the future involve the universal quantification as well. FUT morphemes like tha and will are universal quantifiers that convey necessity. Practically, this means that they quantify the entire set of possible worlds and provide interpretations similar to those of must (see Copley 2002; Kaufmann 2005; von Fintel & Gillies 2010; Giannakidou & Mari 2014; among others), as in the following examples: (4.6) a. O Pavlos tha ine The Paul FUT be.3g b. Paul must be / **will be sick

epistemic

arostos sick

Notice that the English counterpart, will, of the Greek FUT, tha, cannot be used in example (4.6), and instead must is used. The interpretation, thus, is equivalent to the epistemic necessity modal must. However, as I will show in detail later in the discussion, this is not the case. The main reason that we cannot follow an account that treats tha and will as akin to must is the following. FUT morphemes exhibit a wide range of interpretations, ranging from epistemic to ability and temporal, readings that must cannot convey. Observe the disparity in interpretations in the following examples: (4.7) a. Paul will get mad b. Paul must get mad (4.8) a. John will leave tomorrow b. John must leave tomorrow

dynamic teleological temporal deontic

90

Chapter Four

Examples (4.7) and (4.8) are indications that there is no similarity between must and will other than the epistemic. As a consequence, this implies that there are differences between must and will in terms of gradience. Must and will are distinguishable because they represent different degrees in the scale of certainty, as I show in the following section. Another important fact about FUT morphemes relates to certainty about the truth of p, meaning that, no matter what is expressed by a structure such as the [FUT p] proposition, the speaker is not committed to the realization of p. Naturally, non-commitment is a well-known property of FUT morphemes (Werner 2006; Staraki 2009, 2010, 2013; Giannakidou & Mari 2013a/b, 2014; among others). However, the issue of commitment has remained partially unresolved with no satisfactory answer because in the literature it is either mentioned with no further formal modeling or its analysis is directly associated with specific grammar components (such as adverbs). In fact, this is the approach Giannakidou and Mari (2014) proposed; that there are three types of commitment: full, partial and trivial. Moreover, they modeled FUT morphemes as non-veridical (see Staraki 2009, 2010; 2013 for an account on FUT in Greek based on nonveridicality), which draws a connecting line between epistemic adverbs and the commitment on the part of the speaker. This approach seems to be accurate, relating the scalability of adverbs with the variance of a speaker’s commitment. However, there are two shortcomings of this approach. First, it measures non-commitment through a very specific grammar component: epistemic adverbs. In this study, I provide an underspecified scheme that feeds upon contextual and non-contextual information when modeling non-commitment. This scheme is a scale that offers a set of degrees of certainty. Second, commitment, as I argue (contra Giannakidou & Mari 2014) here, has to be relativized to some value. Definitely, we concentrate on the speaker when we talk about certainty and commitment, but the goal of a formal analysis of a proposition is to show how close we, the speakers, approach the truth, when we use a proposition of the structure [FUT p], how certain and committed we feel, and what degree of certainty and commitment we want to relay to the world. Thus, the key value and goal of our scale of certainty will be the truth of a proposition p via a set of assigned symbolic probabilities. A third fact about FUT is how the predictive (i.e., temporal) reading is borne out. Even if we accept that tha in Greek and will in English are modals, we still have to explain how a pure temporal reference derives from a modal morpheme. There are four approaches currently in the

Modality and Time

91

literature that are related by a common characteristic: they derive temporality through a specific feature of a FUT modal. Enç (1996), for example, derived temporality via the intentional meaning that a modal inherently has; it is this property of planning out that renders the temporal reading. This approach falls short when we consider that all interpretations of FUT do not feature intentionality, for example, the ability future. Condoravdi (2002), on the other hand, pointed to the grammatical aspect (statives vs. eventives) differences that the verb following the FUT morpheme has; only eventives have that quality of shifting the eventuality time forward, thus yielding the future reading. This distinction might hold true for English but it does not for Greek (see Staraki 2013), and eventually, it fails to have the generality we seek when we model linguistic phenomena. Another equally important approach is that of Copley (2002) who argued that the future is the causal result (net force, in Copley’s terms) of some situation. Thus, the future is inherently causal; a structure that modals do not feature. Copley (2002) also employed a set of ordering sources, inertial and bouletic, to derive the temporal and modal readings respectively, but she did not manage to capture the various readings of future reference. Considering FUT morphemes as inherently causal is again a piece of the whole picture. Finally, Giannakidou & Mari (2013a/b, 2014) argued that the predictive future (pure temporal reading of FUT) derives from a choice between modal bases: if the modal base is metaphysical then we have a temporal reading. Otherwise, we have a modal reading. However, this approach does not provide an answer for how and why we eventually have these differentiated modal bases. The variety of answers, however, does not offer a satisfactory and generalizable account. The reason for this is, as I will show in detail later in this chapter, that temporality cannot be derived from the modality of just one feature, regardless of whether we call it prediction, intention, or cause. By postulating the difference we cannot explain it. In other words, all modal auxiliary verbs, epistemic, deontic, etc., contribute some sort of temporal meaning (e.g., Condoravdi 2002). In this sense, however, will and tha are not unique. What is the source, then, of the pure temporal reading and specifically the future meaning? I propose that it is the correlation among worlds in the modal base and the degree of information required that derives a pure temporal reading; when the degree of certainty increases, we have a pure temporal reading. The temporal (predictive) reading is an epiphenomenon of an individual’s degree of certainty. This is the idea I model formally in this section. As might be expected, considering knowledge as the decisive factor that helps us to determine what a modal reading is and what a pure

92

Chapter Four

temporal reading is, is not new. Knowledge and belief have been the cornerstones of future reference theory, which can be deduced to the following statement: the more I know, the better the epistemic claims I can make for the future. In fact, in many theories, knowledge acts as an ordering source (Copley 2002; Werner 2003; Mari 2009, 2013, 2014; Giannakidou & Mari (2013a/b, 2014)), thus determining a set of metaphysical alternatives in which a proposition p is true in the future. However, the usage of terms such as knowledge and belief is more often than not vaguely defined; no theories have yet proposed a way to model the so called enough knowledge that derives the distinction between a modal reading and a temporal reading. In this study, I propose that the interpretational variation of FUT morphemes is due to the different orderings that the worlds in the modal base can have. Specifically, I show that the modal base is of one nature, meaning that it features information (for a similar approach see Yalcin 2007; Hacquard 2006, 2010). I am not making the distinction between knowledge and belief because, as we will see later, an epistemic claim can in fact be made even in the absence of knowledge and belief (see Staraki 2013, chapter 3). Gradience is another key property of FUT morphemes, as I preliminarily showed, and this I analyze as a scale of certainty (see Staraki, ms.) that captures the various degrees of commitment within the FUT category and explain the different orderings in the modal base. In order to connect modal semantics (Kratzer 1977, 1981, 1991) with scale semantics (Kennedy 2007) I introduce a set of conditions that underlie the FUT structure, and regulate the relation among worlds in the modal base. In section 2, I discuss the disadvantages of current theories and present arguments that challenge them. In section 3, I show that, although FUT morphemes are considered universal necessity operators, they are characterized by a quantificational gradience (the size of the universal quantificational domain, as we will see it later) that leads to a wide range of interpretations. In section 4, I discuss the disadvantages of the branching-time model. In section 5, I present a formal analysis of FUT. The formal model I present here features degrees of certainty on the part of a rational individual based on a set of conditions that regulate the levels of certainty. While FUT morphemes look like regular modals, they in fact feature relatively unexplored and non-modeled underlying logical properties such as (un)certainty, conditions on the modal base worlds set, commitment, and gradience on the level of the quantificational domain. Section 6 summarizes the interesting ramifications in the analysis of FUT via a set of conditions and gradience on certainty, commitment, and quantification.

Modality and Time

93

2 Current Theories In this section, I discuss the shortcomings of current theories on the future reference, and offer a preliminary discussion on the formal analysis I propose in this study. There are four main approaches in the contemporary semantic literature for future reference although they differ in their details. In the first approach, the English auxiliary will incorporates a tense-like semantics into its modal definition (Jespersen 1924; Smith 1978; Haegeman 1983; Palmer 1987; 1986; Enç 1996; Sarkar 1998; Copley 2002; Condoravdi 2002; Jaszczolt 2006; Mari 2011 for Italian). In the second approach, the English auxiliary will is ambiguous between a modal and a non-modal interpretation (Comrie 1985; Hornstein 1977, 1990; Kamp and Reyle 1993). In the third approach, the morpheme will is a tense (Kissine 2008). Finally, in the fourth approach the morpheme will is an epistemic necessity modal that expresses either a predictive or a nonpredictive reading (Kaufmann 2005; Giannakidou 2012, 2012a; Giannakidou & Mari 2013a/b, 2014).

2.1 Future: A Modal with a Tense-like Component Enç (1996) argued that the auxiliary will incorporates a tense-like component in its meaning; moreover she attributed future reference to the modal auxiliary. For example, (Enç 1996: 33, notation adapted): (4.9) WILL[S] is true at ¢w, i² iff in every world w’ accessible to w there is an interval i’ such that i < i’ and S is true at ¢w’, i’².

Within the framework of possible world semantics, Enç (1996) assumed that the temporal contribution of the necessity modal will is represented over world-time pairs. Specifically, Enç (1996: 354) pointed out that the variable i is the time of evaluation or the UT, which is replaced by a future time, and the sentence in the scope of the modal is evaluated with respect to the new time. Will is treated not as a tense but as a necessity modal operator quantifying universally over possible worlds consistent with predictions. Futurity yields to the ordering of the worlds as histories. In other words, the predictive reading of will derives from an evaluation of a current world-time pair ¢w, i² and a successive world-time pair ¢w’, i’². In this approach, a proposition like Paul will dance can indeed be predicted by the rule in definition (9), as it denotes a world w’ accessible from w and an interval i < i’. Since the epistemic modality of will is not evaluated now, then a proposition such as Paul will be in school

94

Chapter Four

now corresponds to a world w’ accessible from w and an interval i = i’, since there is no predictive reading. However, consider a proposition like Paul would have danced yesterday where the epistemic claim has a past temporal orientation; accordingly we would have to consider a world w’ accessible from w and an interval i’ > i. This approach, however, does not offer much other than just positing a set of models to derive all possible temporal interpretations of will. This approach cannot be generalized. Another drawback of the approach suggested by Enç (1996) is that the UT (the so called now) does not offer anything when deriving the predictive reading. The FUT modals will and tha do not always feature UT as their reference point, in relation to which they are shifting their time forward. Hence, it would be misleading if we accepted that a predictive reading is the result of shifting the time forward with respect to the UT (read now). Let us see some examples of the English will and Greek tha that express future reference without the UT as their reference point. For example: (4.10) a. Otan erthis sto spiti mu tha su dhikso ti aghorasa When come.2sg into house of-mine FUT you show what bought.1sg When you come into my house I will show you what I bought b. O Pavlos tha fighi mono an dhiavasi ta mathimata tu The Paul FUT go.3sg only if study.3sg the homework his Paul will leave only if he does his homework

Examples in (4.10) show a future interval succeeding another interval sometime in the near or distant future. None of the above examples makes a reference to the UT. A second reason is that the inherent tense-like component of will proves illusive, as we can derive a pure temporal reading from a modal reading just by context shift, as in the following example: (4.11) a. It will snow for hours in Chicago b. Paul will have bought a gift already c. Paul will take his pills now

generic epistemic deontic

(4.12) a. It will snow for hours tomorrow in Chicago b. Paul will have bought a gift by tomorrow evening c. Paul will take his pills tomorrow

temporal temporal temporal

If there is a tense-like component, then we are ignorant as to what special mechanisms would be needed to silence it and under what conditions. One might suggest that it is the context itself that turns a modal into a temporal reading and point to the choice of adverbs. However, as we

Modality and Time

95

will later see when I discuss the work of Giannakidou and Mari (2013a/b, 2014) adverbs indicate the status of the context meaning that, they are used to indicate specifically what a speaker intends to convey. Adverbs do not shift the context, but rather, they point to a shift in context in examples (4.11) and (4.12) the adverbs indicate either a non-settled situation (i.e., a modal reading) or a settled situation (i.e., a temporal reading). In this study, I am not identifying adverbs as those essential factors that give rise to the different readings of the FUT morphemes will and tha. In fact, I also omit the UT as the essential and sufficient cause that generates the temporal and modal readings. Instead, I propose that the FUT various readings of the FUT morphemes are the result of gradient quantification and the degree of commitment that a speaker expresses.

2.2 Future: Aspectual Distinctions Condoravdi (2002), on the other hand, suggested that one way of resolving the ambiguity between temporal and modal readings is that the eventive predicate, but not the stative forward-shifts the eventuality description’s time. This hypothesis is developed in a branching-time model (Thomason 1984) in which time branches out towards the future, resulting to multiplication of worlds: (4.13) a. John will dance (probably) b. John will be tired (tomorrow)

If the predicate is eventive, as is the case in example (4.13a), the time will be forward-shifted t0 ” t1, branching out towards the future, and thus deriving a temporal reading of the FUT morpheme. On the contrary, if the predicate is stative, as is the case in example (4.13b), the time will not branch out but overlap t0 R t1, thus leaving the metaphysical possibilities closed and allowing only the modal (epistemic) reading. In this treatment, Condoravdi assumed the following generalization: (4.14) Eventive predicates obligatorily forward-shift the eventuality w.r.t. UT; whereas stative predicates do not: AT (t, w, P) = a. e[P(w, e) š W (w, e) Ž t] if P is eventive For an eventive predicate to be AT a time that event must be included within the time b. e[P(w, e) š W (w, e) R t] if P is stative For a stative predicate to be AT a time that state must overlap time

96

Chapter Four

However, contrary to the prediction suggested by Condoravdi (2002), the English will and the Greek FUT morpheme tha allow temporal and modal readings with both eventive and stative predicates. For example: (4.15) a. To vivliopolio tha klisi stis 5pm (malon /avrio) The bookstore FUT close.3sg at-the5pm (probably/tomorrow) The bookstore will close at five (probably/tomorrow) b. O Janis tha ine sto sholio (malon /avrio) The John FUT be.3sgat-the school (probably/tomorrow) John will be at the school (probably/tomorrow)

Both examples in Greek and English show that an account based on the aspectual distinctions cannot be generalized, because both eventives and statives can be used to express either a pure temporal or modal reading. The adverbs are used to indicate the speaker’s intended interpretation; otherwise, with no contextual information, both examples (4.15a) and (4.15b) can be interpreted as either events that refer to a future interval or as modal evaluation claims. The work of Condoravdi (2002) advances the assertions of Enç (1996) as she also pointed to the importance of the UT as the characteristic that distinguishes a modal interpretation from a temporal interpretation. This approach, however, I consider to be ineffective as I discussed earlier. However, there is one additional factor to be considered. It is true that the aspectual distinctions contribute to the structure of the notion of future orientation, but orientation just predisposes us to the future, and in fact, it is quite different from when we discuss temporal reference, which expresses intervals, (i.e., distinctions of time). Thus, I assert that the argument of grammatical aspect is insufficient when it is used to explain the interpretational variation of FUT morphemes. Therefore, I focus on what the future reference means in terms of degree of commitment in relation to certainty, based on the levels of information that are expressed via FUT morphemes. What I will model is not the relation of the UT to a subsequent interval, but the notion of certainty as being inversely proportional to the degree of the speaker’s commitment to his/her assertion. In other words, the pure temporal reading derives from a high degree of certainty, rather than aspectual distinctions of the verb.

2.3 Future: An Epistemic Necessity Modal Giannakidou and Mari (2013a/b, 2014) argued that the speaker’s knowledge at the UT partitions the set of metaphysical branches into two subsets (see Mari 2013): (a) the reasonable worlds in which a [FUT p] is

Modality and Time

97

true, and (b) the unreasonable worlds in which [FUT p] is not true. The future, hence, is non-veridical (see Giannakidou 1998, 2013) and universal quantification takes place in the subset of the reasonable worlds considered to be privileged worlds, thus, deriving a partial commitment that licenses an evaluative layer of positive bias. The speaker’s judgment about the likelihood of the actual world to be a reasonable or privileged one is high. Adverbs indicate this bias and the speaker’s confidence. This account’s new argument is that the future has an additional epistemic and evaluative layer, which explains the systematic, cross linguistic ambiguity of the future reference between predictive and epistemic readings. For the predictive (temporal) reading, the necessity modal tha features a metaphysical modal base, an epistemic ordering source, and the universal quantification takes place in reasonable worlds: worlds where nothing unexpected happens and everything develops as planned. Giannakidou and Mari (2013a/b, 2014) explained that the temporal reading of tha involves the forward-shift of an interval i’ in relation to the UT, which they notated as [tu, f). This, Giannakidou and Mari (2013a/b, 2014) put formally as follows (from Giannakidou & Mari 2014): (4.16) FUT (NON-PAST (p)) = 1 w.r.t i iff: a. w’ Besti(w0) : t’[tu, f) š p(w’, t’) b. ‫ת‬f(w0) = Best ‫ ׫‬CBest (non-veridicality)

The epistemic reading, on the other hand, is the result of an epistemic modal base, an epistemic ordering source, and the universal quantification takes place in the privileged or BEST (a term coined by Portner 1998) worlds, meaning that the universal quantifier quantifies over a restricted, unique set of alternatives (normal or bouletic), formally presented as the following (from Giannakidou & Mari 2014): (4.17) FUT (PAST (p)) = 1 w.r.t i iff: a. w’ Besti(w0) : t’‫ ط‬tu š p(w’, t’) b. ‫ת‬f(w0) = Best ‫ ׫‬CBest (non-veridicality)

Giannakidou and Mari’s (2013a/b, 2014) account seems promising, as it provides an explanation of ambiguity via the speaker’s knowledge and a positive bias toward certain metaphysical alternatives. However, I would like to point out a couple of ontological and empirical weaknesses that previous theories have also suffered from. It is a shared assumption that an epistemic claim is based on the speaker’s knowledge, however, as it has been shown by Staraki (2013) this is far from being the only truth since, first, epistemic claims can also be issued in the absence of knowledge, and second, epistemic modal verbs

98

Chapter Four

and particles exhibit different degrees of sensitivity to different types and degrees of knowledge. Giannakidou and Mari’s (2013a/b, 2014) work on the contrary, considered FUT morphemes to be akin to must without taking into account that FUT morphemes express degrees of commitment and certainty. In addition, another essential aspect that Giannakidou and Mari disregarded is that FUT morphemes are far from being uniquely epistemic. As shown here, FUT morphemes demonstrate a variety of interpretations ranging from epistemic to deontic. Thus, Giannakidou and Mari’s (2013a/b, 2014) arguments about the epistemic and evaluative layer of the future should be reassessed. According to Giannakidou and Mari (2013a/b, 2014) the set of privileged worlds contains these metaphysical branches, in which the laws at work in w0 at the UT also hold. The speaker is aware of these laws, and nothing unexpected happens. The privileged worlds are not doxastic alternatives and the actual world is part of the set of privileged worlds (Giannakidou and Mari 2014: 20). A proposition [FUT p] that belongs to the privileged set of worlds has a predictive reading and everything else is an epistemic claim. The argument of privileged worlds serves a dual purpose as it accommodates both the characteristic uncertainty of an epistemic claim and the feature of planning (everything goes as expected) of predictive reading. However, this theoretical approach incorrectly implies that a privileged set of future courses of events is meant to be realized because the set of privileged worlds contains the actual world. I find this argument leads us to the wrong conclusions because it implies that, among the many metaphysical branches (the future courses of events), there is an actual/privileged future. The speaker might have a strong and positive bias toward a certain course of events but that differs from actually incorporating the real world into the set of privileged worlds, as it implicates some sort of pre-commitment to actualization. No matter how high our expectations are for a course of events in the future, the future is still indeterministic, meaning that, there is always the possibility of the unexpected and unknown. With the argument of privileged worlds, we have to accept that p is already considered as true at the UT, in the actual world. Instead, I argue that in order to single out the predictive reading, we need to represent the relationship that holds between worlds in the modal base and the available information. We must seek a model in which the meaning of the future is yielded from a proportional relation of what we consider highly possible to what we consider less possible: a relation of degrees.

Modality and Time

99

Giannakidou and Mari’s (2013a/b, 2014) argument about the privileged set of worlds cannot explain a proposition in which FUT morphemes are multiply embedded as in the following example: (4.18) John will marry when he will graduate from the university when he will find a job when he will finish the construction of his house.

John’s future in the proposition above is not evaluated at UT in the actual world, but rather at a future point in time that is followed by other future intervals as preconditions. It is difficult to understand then what the actual world at the time of the utterance will favor and predetermine, as Giannakidou and Mari’s (2013a/b, 2014) argued. Recall that in Giannakidou and Mari (2014: 20) account, the set of privileged worlds contains the actual world. This formalization consequently means that privileged worlds and the actual world are considered as comparable, meaning that, ReasFut(w0) [actual world/now] d ReasFut(w1) d ReasFut(w2) d ReasFut(w3) d ReasFut(w4) and w’ReasFut(tu). This formally means that w2ReasFut(tu) and w3ReasFut(tu) and so on, and that w1  ReasFut(w3) which is inconsistent. How can UT determine the future and at the same time be the same with its development? The privileged set of worlds also means that w0, w1, w2, w3, w4 are considered the actual future, since there are comparable. However, w1, w2, w3, w4 have not happened yet. Gradience is not represented. A formal analysis of FUT morphemes must allow the evaluation to take place at any given interval, as some sort of reassessment in relation to new information acquired, not only at the UT and in the actual world. The pursuit of the future reference involves the representation of dynamic and evolving meaning. Hence, the ontology of the set of privileged worlds cannot be considered a tool to single out unique, metaphysical branches in the set of possible worlds. The predictive reading, as I will show, is the result of measuring the relation between knowledge and the informational correlation of propositions, which raises certainty on the part of the speaker.

3 The Characteristics of the Future Reference In this section, I present data from Greek and English showing the essential characteristics of the category future. There are many important questions that are disputed when dealing with the future reference, but many times these are presented in a way often leading to confusion. I focus on the following points. First, the factors of the modal

100

Chapter Four

interpretation: in an updated version of Kratzer (1977, 1981, 1991) that was first argued in Staraki (in press), I show that the modal force, modal base, and ordering source should be adjusted to represent gradience and uncertainty, key features of future reference. Second the branching time and its relation to the actual time and world: I argue that a branching time model (Thomason 1984) is rigid and cannot represent the degrees of commitment and uncertainty. Third, the issue of underlying logical structure which involves the quality and quantity of knowledge; I show knowledge is not the only precondition licensing expressions of future reference and present arguments in favor of an underlying reasoning pattern1 for future reference in general: provided that proposition p and the knowledge or the belief that q o p, then conclude FUT (p). The FUT morpheme as the grammatical indication of being uncertain of something else might be the actual cause of q. A formal analysis is presented fully in section 4. First, I will provide a relative background on the future in Greek and English.

3.1 Future Reference in Greek Particle tha ‘will’ combines with all verb forms inflected for tense, aspect, person and voice, and it is used to express both future temporal reference and modal assessments (see Tsangalidis 1999; also Mackridge 1987; Joseph & Philippaki-Warburton 1987; Xydopoulos 1996; Roberts & Roussou 2003; Roussou 2000a; Roussou & Tsangalidis 2010; Giannakidou 2009, 2011; among others). With the exception of example (4.19d) that expresses an epistemic assessment2, all combinations of particle tha are used to express either temporal or modal readings:

1

Definitely this reasoning pattern might prove to be invalid, but this is because q might not always cause p, or because q might imply p by default. However, these are the characteristics of future reference: the uncertainty and the lack of commitment. In addition, these are the features of FUT morphemes both in Greek and English in the analysis I offer here. Therefore, the possible invalidity of the reasoning procedure is not an argument against analyzing future reference within an abductive reasoning frame. 2 Mackridge (1987:385) also considered (4.19d) as an expression of a temporal relation and not only an expression of modality, as it is usually presumed in the literature (see Holton et al. 1997; Tsangalidis 1999, among others). I agree with Mackridge (1987).

Modality and Time (4.19) a. Non-Past Imperfective Tha milai She will / must speak

101

b. Non-Past Perfective Tha milisi She will / must be speaking

c. Past Imperfective Tha miluse She would (be) speak(ing) She would have spoken She was about to speak She must have been speaking

d. Past Perfective Tha milise She must have spoken She would have been speaking She must have spoken

As we notice in example (4.19c) the combination of tha with imperfective has the widest range of interpretations. For example, it can be the apodosis in conditional sentences, polite request, ‘future of the past’, and inference about past events (Tsangalidis 1999 crediting Ben-Mayor 1980:73). The first characteristic of the examples (4.19) is the speaker’s lack of factual information and commitment on what is consistent with fact or reality. In this sense, Greek fits the general picture where the future reference is speculative and open to change (Comrie 1985). The second characteristic is that aspect is not solid criterion in order to disambiguate temporal and modal readings.

3.2 Parameters of the Modal Interpretation When we talk about parameters of modal interpretation, we refer to modal force, modal base, and ordering source (see Kratzer 1977, 1981, 1991). Literature relative to modal interpretation of the future reference (see Copley 2002; Condoravdi 2002; Mari 2009; Giannakidou & Mari 2013a/b, 2014) focuses either to one or two of those interpretation parameters, by assigning some default content to each in order to explain the future reference’s intrinsic properties (i.e., epistemic or predictive (temporal) meaning). Here, I argue that these parameters should be adjusted to represent the gradience that future reference exhibits. 3.2.1 Modal Force Let us see a few characteristic examples: (4.20) a. Ta skilia tha The dogs FUT The dogs will starve

pinasun starve.3pl

102

Chapter Four b. O Janis tha epistrepsi me the John FUT return.3sg with John will come back with his girlfriend c. Tha nikisis se afton to FUT win.2sg in this the You will win this contest! d. O Janis tha pji ke 10 bires The John FUT drink.3sg 10 beers John will even drink ten beers

ti fili the girlfriend

tu his

dhiagonizmo contest

FUT morphemes are considered necessity modals in all relevant theories with the exception of Kissine’s (2008). However, the set of examples in (4.20) illustrate the case of what I call variant modal force (see also Staraki, in press). While all modal statements express that the propositions contained are necessarily true (it is not possible to be false in any circumstances), nevertheless each of them conveys a different degree of certainty3 that the contained propositions are necessarily true. For example, the proposition in example (4.20a) the dogs will starve which is an epistemic claim, and the proposition in example (4.20b) John will get back with his girlfriend, which is a predictive (temporal) statement, are necessarily true as they both contain the necessity modal tha for Greek and will for English. However, example (4.20a) when compared to example (4.20b) is further from approaching the ideal (of being actually true) than (4.20b). The proposition in (4.20b) comes closer to being true than (4.20a). Similarly, example (4.20c), which is a proposition with a deontic (specifically, teleological) content, conveys a necessity closer to being true (more like a requirement, see Staraki, in press) than the proposition in example (4.20d), which has an ability reading. The degree of differentiation between an epistemic claim and a predictive reading is based on how close a proposition p comes to actually being true, as previously stated. Moreover, this characteristic property, as we will see in the next section, depends on the entire aggregation of propositions from which conclusions can be drawn, the modal base. The FUT morphemes feature a variant modal force of necessity, which is an indicator that the modality of the future reference is characterized by a gradience of uncertainty4. Thus, the result of the modal valuation is degrees of necessity.

3

I use terms such as certainty and uncertainty interchangeably. The relation of the variant modal force of necessity and uncertainty will become clearer in the following section about modal base. 4

Modality and Time

103

3.2.2 Modal Base Modal base contains all those worlds that form the basis of a modal claim; this is the traditional definition that I also adopt. Giannakidou and Mari (2014) made an interesting claim: the modal base of FUT morphemes is non-veridical (for non-veridicality, see Giannakidou 1998, Staraki 2013), which means that the set of worlds it consists of contains both p and not-p worlds. Consider, though, the case in example (4.20a). If the modal base contains worlds that involve past knowledge or information that is veridical (thus we know if the worlds on which we have based our claim are true or not), then the modal base contains more worlds where p is true than worlds where p is false. For example, I know from past experience that Paul forgets to feed the dogs, that there is no one else there to feed the dogs besides Paul, and that the dogs have starved in the past, then the modal base is veridical, not non-veridical. But also consider a different context for the same example in example (4.20a). Paul just announced to me he is going to Rome for two days. My claim in example (4.20a) will be based on a nonveridical set of propositions because I guess or assume that Paul forgets to feed the dogs and that There is no one else there to feed the dogs besides Paul and that the dogs have starved in the past. Therefore, modal base cannot be strictly non-veridical, as Giannakidou and Mari (2014) argued. In contrast, I argue that the modal base of the future reference is a nonhomogenous space in which veridical and non-veridical subsets of worlds are proportioned in relation to the whole set of the worlds in the modal base. This property of proportional non-homogeneity identifies more accurately the variant degree of uncertainty that the future reference exhibits. Think of it this way. If the modal base contains a higher percentage of non-veridical propositions5, then the expression with the FUT morpheme will or tha will be further from actually being true, but if the modal base contains a higher amount of veridical propositions then the chances are that an expression with the FUT morpheme will or tha will be closer to actually being true. The status within the modal base I present here solves the issue of the missing premise6 to which Giannakidou and Mari 5

Technically speaking, the modal base is a function from possible worlds to propositions. However, I use the term "proposition" to emphasize the essential part of my argumentation. 6 The argument of the missing premise is based on non solid facts. When we lack something, we usually know what we lack. However, in future reference expressions, we do not know what we lack since what we lack never actually

104

Chapter Four

(2013a/b) ascribed the uncertainty of an epistemic future. It is not that we lack the premises on which to base our evaluations, but it is the veridicality and/or non-veridicality proportions within the modal base that determine the uncertainty an individual has. Modal base is also non-homogenous for another reason. Every modal base is characterized by a relation that holds among the worlds it contains. To put it differently, there are correlations with regard to information among worlds in the modal base, such that these correlations reflect the overall degree of uncertainty. For example, consider the case of ability in example (4.20d) and what the modal base for this particular case should contain. How sure can we be that John will even drink ten beers? Well, it depends not only on what kind of propositions the modal has (veridical vs. non-veridical) but also on what correlations hold among the worlds of the modal base. A proposition might increase the degree of certainty we have for another one, for example, John used to drink 10 beers when he was a teenager and on vacation away from his parents makes my claim John will even drink ten beers more credible. However, if I also consider that John is not a teenager anymore, John is too responsible of a person to act in such a way, and John is with his parents now, these propositions decrease the first proposition’s quality of being close to the truth. Thus, we observe that the modal base features relations among worlds and that these worlds are in fact competing with each other to accurately explain the actual world. This relation among propositions in the modal base differs depending on the type of future reference. A temporal interpretation of future reference, for example, features a greater number of worlds in the modal base that induce or justify a predictive reading. Consider the example (4.20b) John will get back with his girlfriend. If I know that John scheduled and promised his return and that John always keeps his promises, etc., then those worlds in the modal base grant us a higher degree of certainty toward7 the truth of the proposition John will come back with his girlfriend. As it has been argued the quantificational domain for a modal in general is the set of BEST worlds (see Portner 2009) that contains the best ordering (via the ordering source), and in which the unique set of possible worlds selected represents normal worlds (strange and unpredictable things do not happen; this basically excludes any notion of uncertainty and occurred. Therefore, the missing premise confuses rather than solves problems of future reference meaning. 7 I use the expression toward the truth because expressions are evaluated based on a scale of degrees of certainty and not on just two values true or false.

Modality and Time

105

gradience). By definition, this is a very constricting statement, if not a misleading one, especially when it involves modals. The reason is that modal expressions and especially expressions of future reference, as I showed by pointing to the relations among worlds in the modal base (see Staraki 2013; in press), inherently carry the property of uncertainty about how things are going to evolve and the lack of commitment to whether things as described in a proposition are actually going to be realized. Uncertainty and lack of commitment are related to gradience and, this is why the formal representation of the future in general should be able to express and predict the gradience that natural language exhibits. Therefore, I think that there is not actually a unique set of BEST8 worlds, but a set of competing worlds within the modal base and a set of competing worlds that are selected for the universal quantification by the procedure I call BEST. BEST in this study determines the size of the intersection between the modal base and the set of possible worlds (see definition in 4.25). The motivation behind this theoretical approach is the incorporation of the gradience and uncertainty inherent in the FUT morphemes. In order to achieve this goal, we have to consider the comparative relations that hold between competing possible worlds (i.e., those in the modal base vs. those in the intersection). Otherwise, every possible world is equal, and there is no gradience of certainty as the natural language examples indicate. For example, an epistemic claim such as the one in example (4.20a) features a smaller intersection, in which p the dogs will starve is true. Now, if we compare example (4.20a) with example (4.20b), the predictive statement features a bigger intersection in which p John will get back with his girlfriend is true. Specifically, the epistemic reading of the expression Mary will be home at this time is based on a set of worlds for its evaluation: the modal base. If I assume that usually Mary is home at this time of day, and that Mary sometimes changes her usual schedule, and furthermore, that Mary changes her route almost every day then there is a smaller number of BEST competing worlds representing the possibility of Mary being home at this time. In this case, the intersection is small. In 8

BEST I use here might be confused with the BEST operator introduced by Portner (1998, 2009). However, in this study, BEST is a heuristic bound and not an operator which greatly differs from Portner’s as we will see in the following discussion. BEST, as I argue, is conditioned by principles influencing the process of the universal quantification. Also, I do not use attributive terms such as Privileged, (Non) Reasonable as these terms prove inadequate to describe and formalize the comparative and quantitative relations holding among possible worlds.

106

Chapter Four

contrast, in the predictive reading of a FUT statement such as Paul will arrive at 5p.m. the set of BEST competing worlds in which this proposition is true is bigger, because I know that Paul keeps his promises, Paul has never failed me, and Paul always is on time. Of course, there is always a number of competing worlds in the temporal (predictive) reading even at a minimum quantity in which Paul will arrive at 5p.m. is false. This is because, as humans, we are not omniscient and have no control over the future. In the same line of thought, example (4.20c) features a bigger intersection, while (4.20d) features a smaller intersection. In other words, the essential distinguishing property between the readings of expressions with FUT morphemes is the quantitative relation that holds between worlds where p is true and where p is false. Hence, the modal base is a set of competing worlds from which some, all, or great part of them can become part of the quantificational domain: the set of BEST competing worlds. The suggestion made here differs from the relevant accounts (Portner 1998, 2009; Condoravdi 2002; Mari 2009, 2013; Giannakidou & Mari 2013a/b, 2014), as it describes and models the gradience and uncertainty of the future reference in quantitative terms. Details of the actual formalization follow in section 5. 3.2.3 Ordering Source Ordering source orders the worlds in the modal base (Kratzer 1977, 1981, 1991) and, as I argue here, it reflects either an individual’s (see Staraki, in press) or a contextual (see Kratzer 1981) ordering. I find that Kratzer’s ordering source contains the seed of gradience, otherwise, if properly extended, as I propose here, it could be considered to be part of a scale of certainty, based on different ordering permutations among possible states. Let us consider the epistemic claim in example (4.20a) in comparison to examples (4.20b), (4.20c) and (4.20d) in regard to the degree of certainty that the proposition in each one of the cases has the possibility of being true. At first glance, all the sentences express a possible state in the future. However, the possible states are not of the same degree: expressions of future reference with tha and will have variant modal force, as I showed in section 3.2.1 Staraki (in press) argued that different permutations on the worlds in the modal base derive different readings. I assume that this is due to variant orderings of the worlds within the set of the BEST competing worlds. For example, example (4.20a) is an epistemic evaluation and, although it is considered necessary when compared to (4.20b), is less necessary than (4.20b).

Modality and Time

107

There are three types of individual’s ordering source, I argue, that represent in a formal way the ordering of competing worlds. A strict ordering is defined when an individual believes world1 to a world2 but not world2 to world1 (i.e., 4.20b). This case of ordering matches the temporal reading of an expression with FUT morphemes such as tha and will. The ordering is strict because it does not leave many alternative options alive, as we consider it almost a necessity that has a high degree of certainty to become the actual state. A weak ordering, on the contrary, is found when there is an individual or contextual preference of a possible statex (see example in 20a). This case of ordering matches the epistemic reading of a future expression. A weak ordering of the worlds complies with the fact that an individual is less sure about the claims he makes, and it also demonstrates the idea that there is a lower degree of certainty assigned whenever we utter an epistemic claim. Before I conclude this section, I add one more type of ordering source. Consider expressions such as wars will happen or human beings will have two lungs, which convey a truth not bound by time or place. In these cases, the ordering source is invariant, meaning that the ordering is not affected by an individual’s preferences or contextual adjustments, as they represent generic statements and their truth is very close to the ideal, if not overlapping with it. To sum, I consider expressions of future reference with tha and will as featuring either an individual’s or a contextual ordering source which derives a scale of certainty which determines the type of interpretation that an expression of future reference acquires.

4 The Branching Time Model Thomason (1984) offered a world-time model to describe a settled past and an indeterministic future. Essentially, a frame of W ൈ T forward branching time is used, times are represented in linear structure and there is a 3-place relation on T ൈ W ൈ W , ؄, such that (1) for all t, ؄t, is an equivalence relation, and (2) for any w, w’  W and t, t’  T, if w’ ؄t’ w and t’ ‫ ط‬t, then w’ ؄t w. Since then, the idea of the ontology of branching time has permeated the field of linguistics. For quite some time now, linguists in the modality field (Enç 1996; Condoravdi 2002; Werner 2003, 2006; Giannakidou & Mari 2013a/b, 2014; among so many others) have propounded the idea that the underlying structure of possibilities and necessities is based on a fixed (Condoravdi 2002) and veridical past (Giannakidou & Mari 2014), as well as, an indeterministic (Condoravdi

108

Chapter Four

2002) and non-veridical future (Giannakidou & Mari 2014). The idea is that world-time pairs are complete histories developed over time and have multiple copies of those worlds with an identical past and an indeterministic future. However, this ontology is not adequate to accurately represent concepts such as uncertainty, and gradience, or the dynamic relationships between possible worlds (propositions), which I pointed to as the essential characteristics of the future reference. Let us recall what we have established in the previous sections for the parameters of the modal interpretation of the future reference. We have seen that the future reference is characterized by a variant modal force of necessity, a modal base which is partitioned to veridical and nonveridical sets of propositions proportional to the whole set of propositions, and that the ordering source is distinguished between strict, weak, and invariant orderings. In brief, the different interpretations of the future reference indicate the existence of gradient necessity. This greatly differs from the linear representation of world-time pairs and the essential characteristic of the branching-time model that is the equivalent of world-time pairs conditioned by monotonicity. This concept is rather rigid to handle the future’s dynamic properties, since it presupposes that a FUT proposition is non-defeasible. Moreover, this is far from being a characteristic of modality, which is essentially non-veridical (see Staraki 2013). The way the branching-time model is constructed cannot effectively represent the differences between the various readings of the future reference. For example: (4.21) a. Tha ine i adherfi tou Jani FUT is the sister the It will be John’s sister b. O Janis tha epistrepsi stis 9mm the John FUT return.3sg John will come back at 9pm c. Tha pjis to farmako su tora! FUT have.2sg the medication your You will have your medication now! d. O Janis tha kolimbisi ke 2 milia The John FUT swim.2sg and John will be able to swim 2 miles

epistemic John temporal at 9pm deontic now! ability 2 miles

In all illustrated examples, the aim is to express the necessity of various degrees and communicate an intension of a rational individual, i.e., the proximity of a proposition p to the truth. Thus, the underlying logical structure of modality, and in our case of necessity, must be based on a modal base that contains veridical and non-veridical propositions and a

Modality and Time

109

quantificational domain consisting of the set of worlds in which p is true and the complement set of alternative worlds in which p is not true. These dynamic changes cannot be handled monotonically but only nonmonotonically. Thus, the model I develop is based on the comparison among possible worlds, as these are not equal. In sum, the employment of the branching-time model (from Thomason 1984 and the subsequent improvements by Condoravdi 2002 and Giannakidou & Mari 2014) cannot depict these features of future reference. Let us explore this further.

4.1 Gradience within Possible Worlds First, all world-time pairs do not belong in an equivalent set of worlds, and the reason is that future reference with tha and will exhibits degrees of necessity. Worlds, as I argue, do not have identical pasts and different futures (equivalence class depending on time), and the reason is that the evaluation of worlds does not depend on time (for example, the possibilities decrease with time, as argued in Condoravdi 2002 and Giannakidou & Mari 2014), as I also show in the next section, but on quality and quantity of information relative to a time. This means that the open possibilities are inversely proportional to the quantity and quality of information available (see Staraki 2009, 2010, 2013). In other words, more sources and a better quality of information are the factors that decrease the set of possibilities. Possibilities do not decrease in a deterministic and monotonic way, meaning that they decrease because they are totally dependent on what happened in the past (as argued in Condoravdi 2002), or because we distinguish between privileged and reasonable worlds (Giannakidou & Mari 2013a/b). On the contrary, possible worlds are of distinct value based on which one has higher potential to actually evolve into the continuation of our actual world. This characteristic is present in all cases of future reference. Consider the proposition it will be John’s sister and the modal base containing the veridical (v henceforth) and nonveridical (nv henceforth) propositions such as I do not know although I have heard about John’s sister (v), I believe John’s sister has no reason to call me (nv), caller ID indicates the city where John’s sister leaves (v), the calls are repeated (v), these days phone hacking happens often (v). The modal base consists of what the worlds w1 it is John’s sister and w2 It is not John’s sister have in common, but w1 and w2 do not belong in an equivalence relation because it would mean that w1 and w2 both are close to developing into the actual world. The ordering source as argued by Staraki (in press) reflects the preferences or the degree of uncertainty of an individual, and here the

110

Chapter Four

ordering source represents a weak ordering, such as something to the effect of w1 z w2 š w2 d w1 reflecting a higher degree of certainty for at least one world. We see that although possible worlds share a modal base there is still some predisposition for at least one world, although we leave alternative worlds as live options of turning w2 into the actual world, but with a higher degree of certainty. Another case against equivalence relations is the temporal reading of the future expression John will come back at 9p.m. The modal base contains worlds such as John’s arrival is scheduled (v), John is always on time (v), John always keeps his promises (v), and John will not show up unless he has an accident (v). The modal base is what world w1 John arrived on time as scheduled and w2 John did not arrive on time due to an accident have in common; however, w1 and w2 do not belong in an equivalence relation because it would mean that both w1 and w2 are close to developing into the actual world. The ordering source here is strict, as we want to represent a higher degree of certainty assigned and a predisposition like w1 z w2 Ÿ w1 > w2 š w2 < w1. Once again we are not dealing with an equivalence class of possible worlds. The set of worlds in the modal base this time favors w1 instead of w2. This happens without the cost of the assumption that a future statement is true only if there is something at the present to make it true, just because it is part of the present (contra Condoravdi 2002; Giannakidou & Mari 2014). Therefore, the interpretation for the various future expressions we saw depends on the quantity and quality of the worlds in the modal base, as not all possible worlds share the same property of being close to the truth. We want to exclude worlds where both p and not p are true, and avoid treating worlds as multiple copies of each other (Condoravdi 2002) with identical pasts which implies worlds have indistinguishable paths of development. Here we managed to represent the worlds as having degrees of necessity and certainty in a gradient relation, and to show that the modal base differentiates in the veridical and non-veridical proportions, which affects the interpretation of expressions with tha and will. Degrees of necessity imply some sort of comparison and competition. Let us move, then, to the next major observation regarding the features of the future reference.

4.2 Pruning the Set of Possible Worlds When a world of the future reference is modeled in the branching-time model, there is only one representation: a single path into the past but multiple open paths to the future. From the discussion up to this point, it

Modality and Time

111

should not come as a surprise when I say that this representation is limited and uninformative (see Diagram 2). In this section, I present evidence indicating that each interpretation of future reference we consider features a different number of possible worlds sets. This means that there is a certain degree of saturation for every proposition of future reference, in which there are no more possible worlds to consider. In other words, the likelihood in relation to the p worlds of the set is that p is not true. W1 W2

W0

W3

W4 W5

Diagram 2 The Branching time model according to Thomason (1984)

Let us consider the case of an epistemic claim. When we make an epistemic claim, we actually evaluate possible scenarios in a subjective manner in regard to what is going to evolve in the continuation of our actual world. There might be more than one possible scenario that we consider true, and a possible continuation of the actual world (see Diagram 3). An epistemic evaluation conveys a lower degree of certainty because there are other properly or sufficiently qualified possible scenarios in parallel with the one possible scenario that ranks above the rest. It is possible, for example, that not only w1 but also w2 will happen, although w1 is assigned a higher degree of certainty when compared to w2. When I utter Mary will be killed I evaluate the possibility of Mary being killed because she is on a roller coaster at an amusement park. However, it is not only one scenario that I am evaluating. There is one scenario, Mary’s death, to which I assign a higher degree of credibility and certainty, but there might

112

Chapter Four

be other competing scenarios to the one I consider as highly likely9. For example, there is the possible scenario that Mary will be just severely injured or Mary will come down safely. An epistemic claim is considered as such because there are a number of parallel scenarios competing for credibility with the one projected by an individual. Those competing possible scenarios, which depend on slightly different permutations by the ordering source, are as many as quantification of the modal base can generate, depending on the quality and quantity of information we have. If, an hour later, I find out that this amusement park has been approved as safe or on the contrary, that the amusement park has had a high rate of accidents, then accordingly, my epistemic claims will feature a different set of possible competing scenarios that may also turn into a predictive reading. This I will explain in greater deal in the following paragraphs.

Diagram 3 The Epistemic Tree

Consider the same example of Mary will be killed. What turns an epistemic reading into a predictive one? It is the notion that there are not many possible scenarios competing with each other (see Diagram 3). 9

Notice that the trees represent symbolic probabilities as these represent the best way the subjective and non-numeric estimates of a rational individual on how things are going to evolve. Future reference is a form of subjective inference expressing an individual’s personal ability to determine a possible evolution for a situation at hand.

Modality and Time

113

When I utter Mary will be killed I am abstractly pruning the set of possible worlds. In fact, I only consider one branch assigned to have a very high percentage of certainty. It is possible that only w1 will be realized and to that I assign a high degree of certainty. Of course, the future is not something we can control fully and know explicitly. However, in our probabilistic way of thinking and evaluating we, in fact, consider one possible world as the single development of our actual world. I am sure of Mary’s death because it is someone’s plan or because her actions will inevitably lead her to her own death. Does the statement Mary will be sick have a predictive reading? The answer is yes. Imagine that you hold a vial containing a deadly virus, which you release in her hometown. What else could happen other than that Mary becomes sick? However, even in this planned and, in a way, deterministic process, we still have to consider a small number of possible worlds where things do not go as planned.

Diagram 4 The Temporal Tree

The deontic case differs from the previous examples as well. First, think of what happens when we utter a future reference expression that denotes some sort of imposed direction, for example, the country will continue on its reform path when this is uttered by the Finance Minister of the state. This statement leaves no space for misinterpretation, as there is one path that should be followed in the future, and that is the path of reforms. The deontic case then illustrates the following interpretation: it is required that w1 is realized and not w2 (see Diagram 4). At this point, the deontic case might look like the case of the predictive (temporal) reading of a future expression with a FUT morpheme such as will and tha. However, there is a difference and it is a relational one. While in the temporal reading of a future expression, an individual creates the possibility for an alternate scenario, in the deontic case, the individual

114

Chapter Four

leaves no room for an alternate path. In section 5, we will see that this difference is located at the quantificational domain and how it is restricted.

Diagram 5 The Deontic Tree

Finally, the ability case restricts the set of possible worlds to those where an individual has the abilities mentioned (see Giannakidou & Staraki 2013) and these abilities are known to the participants. Thus, the future reference expression involves these possible worlds (we might call it the ability worlds) that are at least as normal as w and compatible with what an individual is capable of doing at w (see Diagram 5). In other words, the ability future reference is about possible worlds in which an individual has tried and therefore knows that he/she has the ability to do p. This requires that the modal base is defined in a world with respect to an individual, not just abstractly in a world, so the ability modal differs from the epistemic, deontic, and temporal readings of future reference in that it is dependent on an individual. The FUT morphemes will quantify over those worlds in which the ability of an individual is relative, excluding worlds where this is not true.

Modality and Time

115

Diagram 6 The Ability Tree

Therefore, we cannot adopt the branching-time model, at least the way it is, which has proven to be inelastic and suppresses the dynamic permutations possible worlds are involved in. Moreover, it represents propositions as strictly featuring one interpretation (recall the common pasts and the equal paths to the future) which is not true in natural language. A dynamic system has to be open in involving reconsideration and modification; a revision of what we believe and know can only be handled in a non-monotonic way of reasoning. This way, as we will see, we revise possibilities that either precede or follow as natural causes; thus, we need to consider the two cases of prior possibilities and posterior possibilities. This will become more apparent later in the analysis when we consider propositions as dynamic members of the CG, for example, when we want to consider propositions within context. The process of pruning the possible worlds is based on the quality and quantity of information (see previous sections where I mention modal base and ordering source).

5 Analysis 5.1 The Logical Structure of Future Reference Future reference is about inference based upon beliefs, partiality of knowledge, and a lack of complete control over planning. As stated above,

116

Chapter Four

future reference features the following reasoning pattern: provided a proposition p and the knowledge or the belief that q o p, then we can conclude FUT (p). The FUT morpheme as the grammatical indication of being uncertain as something else might be the actual cause of q. Our claims are often drawn due to a lack of evidence to the contrary or as the most likely explanation, and our ignorance has to be retracted or revised when we have a new piece of information that contradicts old beliefs. Future reference with tha and will express a non-homogenous domain that can be revised and alter its truth-values depending on the new information that has to be reasoned out. Therefore, I propose that in order to be able to handle the dynamic characteristics of future reference, we should analyze it in a non-monotonic way. This means that whenever we have a new piece of information that is part of the set of propositions we know and/or believe, that set could be reduced, increased, or unchanged. The underlying logical structure of the future reference involves the following tuple ‫ˆ תۃ‬ሺ™ሻǡ ’ǡ ǡ ǡ ‫ ۄ‬where ŀf(w) represents the future reference’s knowledge base, including and not limited to, information, norms, goals, evidence, stereotypes, and the external world knowledge that a rational individual has, where p is the set of propositions that describes a set of facts and/or beliefs possessed by a rational individual. R is the function that assigns logic credal relations between propositions in the ŀf(w). B(EST) is a process that determines the size of the intersection between the modal base and the possible worlds set, and S is the scale of certainty that consists of ‫ܦۃ‬ǡ ‫ط‬ǡ ߜ‫ ۄ‬and assigns degrees of certainty to those worlds in the BEST intersection.

5.2 Determining the Universal Quantificational Domain I will now examine the conversational background of FUT morphemes: the modal base and the ordering source. Recall from section 4.1 the inferential nature of future reference. FUT particles tha and will feature a number of interpretations such as intentions, orders, promises, and threats, which are relayed via a future expression. A generic, habitual, or eternal truth can also be expressed with tha and will. Hence, my main analytical claim is that the FUT particles tha and will contribute to the structure of an inferential component (see similar claim for Italian in Mari 2010), and that the various readings depend on the type of the modal base and the ordering source involved. In addition, I have shown that modal base is not homogenous as it can contain both veridical (facts) and non-veridical (not committed or certain about their truth value) propositions. We have also seen that a modal base

Modality and Time

117

represents a rational individual’s and the external world’s knowledge and beliefs. Then, future reference expressions with will and tha will be about modeling the beliefs, quite generally, of an ideally rational individual reflecting upon his/her beliefs and knowledge. He following is the definition of the modal base of the future reference: (4.22) MODAL BASE i) A future reference modal base ŀf(w) with regard to a context c and a (group of) rational individual(s) i whose knowledge is relevant in c such that for any possible world w, ŀf(w) is established knowledge and/or belief in w and is known or believed by i. ii) A future reference modal base ŀf(w) comprises a non homogenous w.r.t. their truth value set of propositions (axioms) which represent and which are interpreted as the total set of beliefs or knowledge of a rational individual.

Relation R is a function with regard to the connections between worlds within the modal base, which creates preliminary, mutually consistent subsets (the competing worlds) by mapping inferential links among worlds. This preliminary stage within the modal base is the linking interaction, during which logical connections among worlds are created (see section 3.2.2 for the description of the process) and consistency (accept the worlds that might be sufficient to explain an observation) is established. For example, when I see dark clouds in the sky, it might be highly likely that it will rain, but it is unlikely to rain if the dark clouds are smog clouds. Therefore, I argue that the construction of the modal base is non- monotonic, in the sense that a newly proposition is added reduces (or increases) the set of possible words. To understand this, let us consider the following example. A proposition D might be a sufficient and/or necessary condition toward E, but propositions D š J might not be (for example, P(B|A) = high likelihood but, P(B|A š C) = low likelihood or impossibility). These subsets of logically or causally linked propositions are either mutually exclusive, similar or multiple variations of a subset, and constitute the set of acceptable, start-off, and competing worlds in the modal base in order to evaluate a proposition of the form FUT(p). There are three types of mapping relation (credal links) that R assigns, Rand, Ror and Rspecific and they are formally defined as follows:

118

Chapter Four (4.23) R-Credal Links i) Rand is a set of conjectures which satisfies the relation R(A ‫ ת‬B) = {P: P(A) š P(B)} Ÿ q there is a possibility of A in conjunction with the possibility of B in drawing the inference q. ii) Ror is a set of disjunctions which satisfies the relation R(A ‫ ׫‬B) = {P: P(A) › P(B)} Ÿ q there is a possibility of A in disjunction with the possibility of B in drawing the inference q. iii) Rspecific is a set of propositions which satisfies the relation (P(A), P(B)  p) Ÿ q the possibility of A and/or the possibility of B are part of the possible worlds set that includes the most specific classes.

R-Credal links is a preliminary partial ordering stage that creates the set of consistent worlds competing within the modal base ‫ת‬f(w). These competing worlds are suitable candidates for the universal quantification, and possible matches to an outcome from the possible worlds set (p). A definition of competing worlds is as follows: (4.24) Competing Worlds (cwx): (cwx‫ת‬f(w) [(BEST(cwx) t BEST (cwx’)) š |‫ת‬f(w)|=|‫ת‬f(w)’| o cwx‫ת‬f(w)[( BEST(cwx) > BEST(cwx’)]])

The definition (4.24) says that for all competing worlds in the modal base, there is a set of competing worlds that is a better candidate for participation in universal quantification. The size of the universal quantificational domain of FUT morphemes tha and will depends on the reduction of the domain of those competing worlds that render some p false (see Staraki 2013 for a similar concept). BEST determines the size of the universal quantification domain by decreasing the set of cwset‫ת‬f(w) (competing alternative worlds in the modal base) and increasing the set of cwset‫ת‬p (possible true outcomes). The more the size of the intersection increases the closer a rational agent comes to an ideal match between a proposition and truth. Formally, this is expressed as follows:

Modality and Time

119

(4.25) BEST10 is a heuristic bound that determines the relative size of the intersection between the set(s) of competing worlds in the modal base cwx  ‫ת‬f(w) satisfying the worlds in p: BEST = {cwx | CW Ž ‫ת‬f(w) (p BEST (‫ת‬f(w)) š p  BEST (‫ת‬f(w) ‫{׫‬cwx}))} Where cw = competing world CW = the set of competing worlds Iff there are cwi and cwj which are consistent inferences to a proposition p then cwi is at least as plausible as cwj, cwi dplausible cwj, iff cwi ‫ ك‬cwj, and the most plausible competing world is that cwi s.t. cwj is true that cwj world1 in ‫ת‬f(w):

Modality and Time

123

(4.30) Invariant Ordering Source u, z  W: u z z š (u < z { u > z)

The three types of ordering sources I offer here permute the competing propositions inside the intersection created by BEST and represent the accessible worlds that manifest the speaker’s reflection with regard to his/her knowledge and beliefs. The ordering will be based on the following criterion of plausibility with regard to a rational individual i: (4.31) Plausibility Criterion: If both proposition FUT(p) and proposition FUT(p’) are consistent inferences to a proposition q then proposition p is at least as plausible as p’, p dplausible p’, iff p ‫ ك‬p’, and the most plausible proposition is that p s.t. p’ is true that p’