Degrees of Assertion

Degrees of Assertion Thesis submitted in partial fulfillment of the requirements for the degree of “DOCTOR OF PHILOSOPHY” by Lavi Wolf Submitted to the Senate of Ben-Gurion University of the Negev Approved by the advisor: Professor Ariel Cohen Approved by the Dean of the Kreitman School of Advanced Graduate Studies Beer-Sheva This work was carried out under the supervision of: Professor Ariel Cohen In the Department of Foreign Literatures and Linguistics Faculty of Humanities and Social Sciences Abstract On the face of it, assertion is a very basic, very simple type of speech act – the speaker has some information that needs to be conveyed, therefore the speaker performs an act that coveys this information. Stalnaker's (1978) account of assertion is a classic in its ability to represent the simplicity of this speech act in an equally simple and elegant manner. The speaker asserts a proposition and the hearer chooses whether to accept or reject this proposition. If it's accepted, the proposition is added to a discourse entity called the common ground that hosts propositions, which are presuppositions that are mutually shared by discourse participants. The theory presented in this dissertation is based on Stalnaker's original insight, and applies it to deal with conversational phenomena that require more fine-grainedness. In order to do that, there is a need for an expanded conversational theory that incorporates other conversational mechanisms besides the common ground, which is discussed in chapter 1. In proposing an expanded conversational theory, this dissertation joins a long line of theories, discussed in chapter 1. Two of which, Ginzburg (1996a) and Farkas & Bruce (2010), propose discourse mechanisms that store conversational items in need to be resolved before a context update may occur, Questions Under Discussion in Ginzburg (1996a) and the Table in Farkas & Bruce (2010), further discussed in the chapter. The discourse mechanism proposed in this dissertation, the Negotiation Zone provides another way to deal with items under negotiation, which differs from previous ones in that it stores Degrees of Assertion, i.e. degrees of strength by which the assertion is performed. In that, this theory joins notable speech-act theories such as Austin (1962), Searle (1969) and Searle & Vanderveken (1985), which distinguish between the propositional content of the speech act and various other factors that contribute to the force of the speech act, 1 further discussed in chapter 1. After discussing negotiation in conversation, the chapter presents the formal theory that serves as the basis for the next chapters. Each one of the next chapters discusses a conversational phenomenon that is best explained by the proposed theory. Chapter 2 deals with modal adverbs and modal adjectives. While the standard conception of these discourse particles is as variants of each other, the chapter discusses data presented in Bellert (1977) and explored in Nilsen (2004) and Piñón (2006, 2009) that shows otherwise. The chapter discusses these accounts and the data, and shows that it is best accounted for in terms of the theory proposed here. A distinction between modal adverbs and modal adjectives is provided, in terms of expressive vs. descriptive epistemic modals (EM) respectively. The former express the degree of belief in the asserted propositional content while the latter describe the degree of evidential support of the asserted claim. While expressive EM modify the speech act, descriptive EM are propositional, which accounts for the properties of modal adverbs and modal adjectives presented in the data. Both modal adverbs and modal adjectives are represented in a unified manner, as the same lexical item only with different scopes. Modal adverbs have wide scope over the asserted proposition, being speech-act modifiers, and modal adjectives have narrow scope being propositional modifiers. The conversational theory proposed in this dissertation shows how the differences between modal adverbs and modal adjectives manifest in terms of different context update effects. The chapter concludes with a short section on the phenomenon of modal concord and how it can be explained in light of the theory discussed in the chapter. Chapter 3 deals with epistemic modal auxiliaries. This chapter adds a syntactic dimension to the semantic one presented in chapter 2, and discusses the root-epistemic distinction. After discussing theories of the root-epistemic distinction, the chapter focuses on Hacquard's (2006) account, based on the classical Kratzerian (1981, 1991) theory of modality combined with Davidsonian (1967) event semantics, which shows the different syntactic locations of EM and root modals. The account of chapter 3 draws from Hacquard's theory which shows that root modals are located right above VP and epistemic modals are located right above T, and proposes a third location for expressive EM – ForceP (cf. Rizzi, 1997) at the left periphery, where speech-act modification 2 occurs. In this manner, the theory preserves Kratzer's insight regarding the uniform nature of modals via different modal flavors and Hacquard's insight regarding the uniform nature of modals via different syntactic scopes, and adds another aspect, the expressive one, to EM theory. The chapter discusses data that shows how the expressivedescriptive distinction manifests with auxiliary EM, and theories that attempt to account for this distinction. The final sections present formal representations for expressive and descriptive epistemic modals, and discuss the phenomenon of A-movement reconstruction and how it can be explained in light of the theory discussed in the chapter. Chapter 4 deals with clarity and hopefully attains it. This chapter is based on Wolf & Cohen (2011), but deviates from it in various ways, for example by claiming that the distinction between adjectival clear and adverbial clearly stems from the latter being expressive and the former descriptive. The chapter focuses on Barker and Taranto (2003) and Barker's (2009) theories of clarity, discusses the theories and proposes an alternative which is based on the notion of the mixture model introduced in previous chapter. A mixture model combines individual probability assignments into a single probability value, thus taking a subjective judgment or belief and 'objectivizing' it. This is a metaprocess utilizing an already established (in chapter 1) cognitive means to process various sources of evidence into a single judgment, which is how individuals attain beliefs with various degrees about propositions, and applying it to process beliefs of various conversational participants (actual and hypothetical) into a single objectivized belief. The chapter distinguishes between adverbial uses (clearly), adjectival uses (clear) and personal uses (clear to x) of clarity, and accounts for each one in terms of the theory presented in previous chapters. Chapter 4 concludes with a discussion of clarity as a private case of the more general phenomenon of vagueness, and proposes an account of vagueness based on the mixture model. Chapter 5 deals with another private case of vagueness, predicates of personal taste. The chapter begins with a discussion of faultless disagreement, a phenomenon that occurs when disputes about personal taste take place, in which on the one hand dispute participants disagree with each other but on the other hand none of them is at fault for saying something wrong. The chapter focuses on Lasersohn's (2005) theory of personal 3 taste predicates, discusses problems of this theory, and proposes an alternative based on Cohen (2011). This alternative utilizes the mixture-model used in chapter 4 to account for clarity, and as proposed in the closing section of chapter 4, applies it to another class of vague predicates. The dissertation concludes with the conclusion, discussing the implications of the theory proposed here and briefly outlining how it can be extended to other areas of research. 4 Acknowledgments "It was starting to end, after what seemed most of eternity to me", the first line of the first book about possible worlds I've ever read (Roger Zelazny's "Nine Princes in Amber"), is a fitting way to begin this section and to finish this project which has occupied my mind for oh so many days and nights. That is not to say that the job started here is completely or even close to being done, as many questions keep arising every day. And if it were up to me, this work would have probably never ended. I, therefore, owe a large debt of gratitude to the people who, in many different ways, were involved in this journey and more importantly, brought it to a stop. First and foremost, a very big thanks to my advisor Ariel Cohen. For believing in me, for many moments of laughter and moments of seriousness, for the ability to give space when needed and to fill the void when required, the immense wisdom and experience both linguistic and otherwise shared with me and the flexibly and talent to steer the dissertation ship through uncertain and ever changing waters. And for the friendship and the good heart which are a rare commodity in this world. I am very grateful for wonderful teachers I had at Ben-Gurion University of the Negev, who helped me transition from Psychology into Linguistics while retaining many of my earlier modes of thought: Nomi Erteschik-Shir, Idan Landau, Tova Rapoport and Dorit Ben Shalom. You've been a source of inspiration and helped me to become a better linguist. And to Suzie Ganot and Merav Saroosi-Peretz, for the much needed assistance and goodwill. A special thanks to Prof. Yishai Tobin, for the support during good times and bad, for the humor, wit and conversation skills. And for showing me that there is much more to linguistics and to life. To my Ben-Gurion classmates and friends: Zehavit Segal, Adi Segal, Aviya Hacohen, Lena Ibnbari, Sharon Taube, Tanya Philippova, Pavel Riazanov, Avraham Niv, Dan Epelman and Yair Dor-Ziderman among many more - it's been a gas. I am also very thankful for the talented and accommodating people of The Institute for Logic, Language and Computation (ILLC) in Amsterdam, in which I've spent the first year of my doctorate: Frank Veltman, Henk Zeevat, Paul Dekker, Maria Aloni, Raquel Fernández Rovira, Floris Roelofsen and Fred Landman's much appreciated guest appearances on 'the vagueness circle'. My time with you has influenced me in many ways. 5 My biggest thanks go to Robert van Rooij who has struggled to structure my mind during a time in which it went all directions at once, and taught me much during this period. Also to Galit Sassoon for the advices and for providing shelter and support as well as professional and personal companionship. To Lucian Zagan, Angelika Port, and Hadil Karawani thanks for helping making the stay in Amsterdam less cold and Dan Zeman for memorable moments during visits. To Ilona Spector Shirtz with whom I had the pleasure to collaborate, thanks for the syntactic savvy and the ability to work very close to deadlines which is very much appreciated. And for having my back. A great big thanks to Christopher Piñón for providing the spark that started the fire – a lecture in The Hebrew University of Jerusalem entitled "Modal adverbs again" at a point in which I was looking for a direction. And for the good conversations and advices. To Hotze Rullmann who was a reviewer for one of my conferenes, for bringing up the idea to look into A-movement reconstruction, thus inspiring my exploration into the syntax-semantics interface. To the people who helped me with beneficial discussions and helpful insights along the way: Hans-Christian Schmitz, Daniel Gutzmann, Tikitu de Jager, Sophia Doering and Magdalena Schwager during my work on chapter 2 as well as the participants of the 'Formal and experimental approaches to discourse particles and modal adverbs' workshop in ESSLLI Hamburg. Edit Doron, Ivy Sichel, Angelika Kratzer, Sabine Iatridou, Kai von Fintel, Tyler Peterson, Lisa Matthewson and Shih-Yueh Jeff Lin during the work on chapter 3 as well as the participants of the 'Modality @ Ottawa U' conference in Ottawa. Chris Barker, Daniel Lassiter, Paul Egré and Nathan Klinedinst during the work on chapter 4 as well as the participants of the 'Vagueness and Language Use' conference in Paris. François Recanati, Isidora Stojanovic, Tamina Stephenson, Hanoch Ben-Yami, Robert Michels, Ivan Kasa, Yannis Kostopoulos, Sanna Hirvonen, Hazel Pearson and Benjamin Lennertz during the work on chapter 5 as well as the participants of the 'Cerisy Context Conference' in Cerisy Château. And many more along the conferences. To my family – my parents Sara and Shmuel and my sisters Chen and Shani whose unconditional love and support always gives me strength. To Gali, my Rock Lobster. Everything I wrote and may write pales in comparison to the ways I can never put down in words, in which you've influenced my life. Keep doing it and you're up for a promotion! 6 Contents Abstract ...........................................................................................................................................1 Acknowledgments .........................................................................................................................5 Contents ..........................................................................................................................................7 Chapter 1: Assertion and negotiation .......................................................................................12 1.1 Assertion .............................................................................................................................13 1.2 Negotiation .........................................................................................................................18 1.2.1 Problems with the Stalnakerian picture ..........................................................18 1.2.2 Expanding the Stalnakerian framework .........................................................20 1.3 Probabilistic common ground ..........................................................................................22 1.3.1 Why probability? ...............................................................................................22 1.3.2 What is probability? .........................................................................................24 1.3.3 Probabilities of probabilities ............................................................................25 1.3.4 Probability spaces ..............................................................................................26 1.3.5 Probabilistic context update .............................................................................28 1.3.6 Assertion operator..............................................................................................29 1.3.7 Mixture model ....................................................................................................33 1.3.8 Negotiation Zone ...............................................................................................35 1.3.9 The aim of assertion .........................................................................................41 Chapter 2: Modal adverbs and modal adjectives .....................................................................43 2.1 Modal adverbs and adjectives – data............................................................................43 2.1.1 Embedding ..........................................................................................................45 2.1.2 Agreements and disagreements .......................................................................48 2.1.3 Nilsen's contrast ................................................................................................50 2.2 Modal adverbs and adjectives – theories .....................................................................51 2.2.1 Modality, the classical theory ..........................................................................52 2.2.2 Bellert (1977) .....................................................................................................54 2.2.3 Nilsen (2004) ......................................................................................................56 2.2.4 Ernst (2009) ........................................................................................................60 7 2.2.5 Piñón (2006, 2009) ............................................................................................63 2.3 Modal adverbs as negotiation chips .............................................................................66 2.3.1 Modal adverbs are expressive ..........................................................................67 2.3.2 Formal account ...................................................................................................73 2.3.3 Explaining the data ............................................................................................78 2.4 Modal adverbs and modal concord ..............................................................................81 Chapter 3: Epistemic modal auxiliaries ....................................................................................86 3.1 The root-epistemic distinction ......................................................................................86 3.2 Epistemic ambiguity .......................................................................................................92 3.2.1 Theoretical ambiguity .......................................................................................92 3.2.2 Empirical ambiguity ..........................................................................................95 3.2.2.1 Embedding .............................................................................................95 3.2.2.2 Agreement-disagreement ..................................................................100 3.2.2.3 Scope ....................................................................................................104 3.2.2.4 Diachronic evidence ...........................................................................107 3.2.2.5 Synchronic evidence ..........................................................................108 3.3 Addressing previous accounts of the expressive-descriptive distinction ..............110 3.3.1 Lyons (1977) ....................................................................................................110 3.3.2 Kratzer (1981) ..................................................................................................112 3.3.3 Papafragou (2006) ...........................................................................................115 3.3.4 Portner (2007; 2009) ......................................................................................117 3.3.5 Anand & Hacquard (2008) ...........................................................................118 3.3.6 von Fintel & Gillies (2011) and Cohen (2013) ............................................121 3.3.7 Huitink (2008) ..................................................................................................123 3.3.8 Tancredi (2007) ................................................................................................124 3.4 Expressivity ...................................................................................................................125 3.5 Formal representations .................................................................................................129 3.5.1 Expressive and descriptive epistemic modals ..............................................129 3.6 Syntactic implications ..................................................................................................130 3.6.1 A-movement reconstruction ...........................................................................130 8 3.6.2 The proposed account .....................................................................................133 Chapter 4: Clarity.......................................................................................................................137 4.1 Clarity and belief ..........................................................................................................137 4.2 Clarity and justification ...............................................................................................138 4.3 Back to belief? ..............................................................................................................140 4.3.1 Personal clarity ................................................................................................141 4.3.2 Adverbial vs. adjectival clarity .....................................................................144 4.3.3 Desiderata ........................................................................................................145 4.4 Belief and probability ..................................................................................................146 4.4.1 The Lottery Paradox .......................................................................................146 4.4.2 Probability and normality ..............................................................................148 4.4.3 Probability and logic ......................................................................................150 4.5 The interpretation of clarity ........................................................................................151 4.5.1 Simple clarity ..................................................................................................151 4.5.1.1 A second mixture model ...................................................................151 4.5.1.2 Clear ....................................................................................................154 4.5.1.3 Publicly available evidence ..............................................................156 4.5.1.4 Clearly .................................................................................................157 4.5.2 Returning to personal clarity .........................................................................159 4.6 The origins of the mixture model ..............................................................................162 Chapter 5: Predicates of personal taste ...................................................................................164 5.1 Faultless disagreement ...............................................................................................164 5.1.1 Basic intuitions ................................................................................................164 5.1.2 Lasersohn 2005 ...............................................................................................165 5.1.3 Problems of Lasersohn 2005 .........................................................................167 5.1.3.1 The pragmatic problem .....................................................................167 5.1.3.2 The semantic problem .......................................................................169 5.1.3.3 The logical problem ...........................................................................170 5.1.4 Recanati (2007) ...............................................................................................171 5.2 Objectivized predicates ...............................................................................................173 9 5.2.1 Cohen (2011) – how to "objectivize" a subjective judgment ....................173 5.2.2 A formal account of objectivized predicates ...............................................175 5.3 Accounting for puzzles and problems .......................................................................176 5.3.1 Faultless disagreement ...................................................................................176 5.3.2 Logical inference ............................................................................................177 5.3.3 Recanati's problem ..........................................................................................178 Conclusion ..................................................................................................................................180 References ...................................................................................................................................184 10 There are times I almost think I am not sure of what I absolutely know. Very often find confusion in conclusion I concluded long ago. In my head are many facts, that, as a student I have studied to procure. In my head are many facts, of which I wish I was more certain I was sure. (Rodgers & Hammerstein: The King and I, A puzzlement) 11 Chapter 1 Assertion and negotiation This chapter is divided into 3 parts. The first deals with the speech act of assertion, opening with Stalnaker's (1978) Assertion and continuing with the speech act theories of Austin (1962) and Searle (1969) and illocutionary logic of Searle & Vanderveken (1985). This part introduces the distinction between propositional content and illocutionary force which serves as the basis for the theory propsed in this dissertation, which distinguishes between the propositional content of assertion and the degree of strangth by which the assertion is performed. Another element of importance is Stalnaker's notion of context update. The second part of this chapter deals with negotiation in conversation. This section discusses conversational elements which are not addressed in Stalnaker's theory, such as the state of assertions that are not accepted into the common ground, yet not rejected either. In order to deal with negotiation in conversation, this section proposes a conversational theory in which such assertions are placed within a Negotiation Zone, a conversational element complimenting the common ground in which assertions reside from the moment they are performed until the moment they are either accepted or rejected. The third part presents the framework of the expanded conversational theory proposed in the second part, via a probabilistic expansion of the common ground and the introduction of the Assertion Operator. The assertion operator is the means by which assertion are represented and inserted into the negotiation zone. It distinguishes between the propositional content of the assertion and the degree of strength by which the assertion is performed. In default conditions this degree of strength is the degree of strength of the sincerity condition of assertion which is belief, but there are conditions, e.g. asserting clarity (cf. chapter 4) in which this degree of strength is more general and represents objectivized belief. This section also deals with the nature of probability and distinguishes 12 between expressive and descriptive probability. The former relates to the personal probability assigned by the speaker to various events and is based on belief, and the latter relates to a more objective notion of probability, relating to the manner by which the available evidence affects the likelihood of events. The chapter concludes with a discussion about the aim of assertion and a novel proposal. 1.1 Assertion What's a better place to start a discussion on assertion than Stalnaker's (1978) seminal paper Assertion? Stalnaker's paper opens with the following four truisms: "First, assertions have content; an act of assertion is, among other things, the expression of a proposition – something that represents the world as being a certain way. Second, assertions are made in a context – a situation that includes a speaker with certain beliefs and intentions, and some people with their own beliefs and intentions to whom the assertion is addressed. Third, sometimes the content of the assertion is dependent on the context in which it is made, for example, on who is speaking or when the act of assertion takes place. Fourth, acts of assertion affect, and are intended to affect, the context, in particular the attitudes of the participants in the situation; how the assertion affects the context will depend on the content." Starting with the first truism, assertion is an act. More specifically, it is a speech act (cf. Austin, 1962; Searle, 1969). To paraphrase Austin, speakers do not only say things but also do things with words. This act has an effect on hearers, i.e. how they perceive the world and furthermore how they act themselves. Thus speech acts affect not only conversational participants but also the world that surrounds them. The type of effect speech acts have is called speech act or illocutionary force (cf. Austin, 1962; Frege, 1956). Questions, for example, have interrogative force, commands have commanding force and assertions assertoric force. The force is an important component of performing a speech act since it distinguishes between different ways that the same proposition can be used: 13 (1) a. The dog is on the lawn. b. Is the dog on the lawn? c. The dog – to the lawn! The first example, an assertion, adds assertoric force to the proposition 'the dog is on the lawn' thereby guaranteeing to the hearer that this proposition is true. The second example, a question, adds interrogative force to the proposition 'the dog is on the lawn' thereby indicating to the hearer that she1is under obligation to state whether the proposition is true or not and the third example, a command with the corresponding force, obliges the hearer to perform an act that will make this proposition true2. The proposition itself, without speech act force, is inert. This is the reason that early speech-act theorists (e.g. Stenius, 1967) use Wittgenstein's analogy in which an utterance's propositional content is like a chemical radical which does not have independent existence and the force or mood by which the utterance is performed is like the functional group that gives the radical substance. The distinction between propositional content and illocutionary force plays a large role in the theory presented in this dissertation, as will be discussed ahead3. Assertion is a most basic speech act, being simply a claim about how things are in the world. When performing an assertion, the speaker provides information about the world which passes on to the hearer. This information has to be accurate in order for an assertion to be successful, i.e. the words spoken have to fit the world. This is the 'words to world' direction of fit (cf. Anscombe, 1957; Searle, 1976) demonstrated by the following scenario: "Let us consider a man going round a town with a shopping list in his hand. Now it is clear that the relation of this list to the things he actually buys is one and the same 1 In this dissertation referring pronouns alternate randomly between masculine and feminine. This dissertation adopts a 'normative theory' of speech acts (cf. Cohen & Krifka, 2011; Searle, 1969) according which the performance of speech acts relates to commitments speaker take upon themselves or impose upon others. 3 This distinction is further divided within illocutionary force into the degree of strength of this force, specifically of the sincerity condition of assertion, which is belief. 2 14 whether his wife gave him the list or it is his own list; and that there is a different relation where a list is made by a detective following him about. If he made the list itself, it was an expression of intention; if his wife gave it to him, it has the role of an order. What then is the identical relation to what happens, in the order and the intention, which is not shared by the record? It is precisely this: if the list and the things that the man actually buys do not agree, and if this and this alone constitutes a mistake, then the mistake is not in the list but in the man's performance (if his wife were to say: “Look, it says butter and you have bought margarine”, he would hardly reply: “What a mistake! we must put that right” and alter the word on the list to 'margarine'); whereas if the detective's record and what the man actually buys do not agree, then the mistake is in the record."(Anscombe, 1957) The content of an assertion, the proposition expressed by virtue of performing the speech act of assertion, is like the list that the detective holds. The list has to fit the items that the man bought in order to be a successful account of how things are. This differentiates assertion from other speech-acts such as requesting or promising which have a 'world to words' direction of fit. Using Searle's (1969) taxonomy, requesting is of the directive speech-act types. It is performed by the speaker in order to get the addressee to do something, and promising, of the commissive type, is performed by the speaker in order to commit to do something. Both requests and promises fit the list that the man holds - a request made by the man's wife will lead him to search for the items on the list and to make the world fit the words written on it, as will a promise performed by the man to bring these items back home. While assertions need to fit the state of the world, the direction of fit between assertions and the state of discourse is not so clear. On the one hand, the propositional content conveyed by an assertion has to be consistent with the state of discourse, but on the other hand successful assertions change the state of discourse by virtue of their assertive force. This is where Stalnaker's second truism comes in: assertions are made in a context, which is a situation that includes a speaker with certain beliefs and intentions, and hearers with their own beliefs and intentions. Stalnaker represents this conversational context in terms of mutually shared presuppositions held by conversational participants to be true. As 15 stated in the truism, each conversational participant has certain beliefs. Beliefs are standardly represented by propositions, when a proposition corresponds to a set of possible worlds in which it is true i.e. a function from possible worlds to truth values. Thus, each conversational participant has a set of proposition she holds to be true. A subset of this set of beliefs is the set of beliefs that each conversational participant holds to be true and to be shared with all the other conversational participants. This set constitutes the speaker presupposition. As Stalnaker puts it: "A proposition is presupposed if the speaker is disposed to act as if he assumes or believes that the proposition is true, and as if he assumes or believes that his audience assumes or believes that it is true as well". (Stalnaker, 1978, pp. 321). These presuppositions constitute the Common Ground, the set of propositions all conversational participants (ideally) share with one another. Being a set of propositions, the common ground is a set of sets of possible worlds. The intersection of this set, therefore, provides the set of possible worlds in which all the propositions of the common ground are true– the Context Set. Each world in the context set is thus considered by the conversational participants to be a “live option” relevant to the conversation. That is – each world in the context set can, for all the conversational participants know, be the actual world. This leads us to the fourth truism:4 assertions affect the context, in particular the attitudes of the participants in the situation. In this regard, it is important to distinguish between two notions of context that exist in the literature. The first notion, notably used in Kaplan (1989) is the situation in which discourse occurs, i.e. the time, location, conversational participants etc. The second notion of context is the totality of information that has been accumulated throughout conversation(s) and is treated by all conversational participants as a given. This second sense is the one used by Stalnaker, and the one that is referred to in this dissertation. 4 The third truism deals with indexicals and therefore doesn't concern the purposes of this chapter, but cf. chapter 5 on predicates of personal taste for a discussion of Kaplan's (1989) theory of indexicality. 16 The fourth truism deals with the effect of assertion on the context, i.e. context update. The force of assertion is its main effect on the context, which according to Stalnaker is achieved by changing the attitudes of conversational participants towards the propositional content of the assertion, from an attitude of disbelief or indifference to an attitude of belief –this is when the assertion is said to be accepted. And, as recalled, the common ground is composed of the presuppositions of conversational participants, i.e. the set of propositions all conversational participants hold to be true. Therefore, if an assertion is accepted, the propositional content of this assertion is added to the set of propositions in the common ground and since the context set is the intersection of the common ground, adding a proposition to the common ground causes worlds to be eliminated from the context set. This is the process of context update which is the main effect and the force of assertion. Using Wittgenstein's (1953) idea of a language game, Stalnaker provides the following description of context update: "One may think of a non-defective conversation as a game where the common set is the playing field and the moves are either attempts to reduce the size of the set in certain ways or rejections of such moves by others". (Stalnaker, 1978:325). This concept is also found in Lewis' (1979) view of assertion as a way to change the shared 'conversational scoreboard', a common register of the progress of discourse. Stalnakerian context update can be represented in the following manner: Figure 1.1: Stalnakerian context update 17 As seen in the figure above, the initial context set is CS0, and it contains both  and  worlds. An addition of  into the common ground eliminates  from the context set by intersecting CS0 with the proposition  i.e. the set of  worlds. The next section discusses problems with the Stalnakerian picture and proposes a way to expand this picture while preserving its qualities. 1.2 Negotiation 1.2.1 Problems with the Stalnakerian framework There are two ways to look at the update process – one is cooperative, the other is dominative. The cooperative notion of context update is that conversational participants work together in order to pass meaning from one to the other. This is conveyed by Grice's well known cooperative principle (Grice, 1975): "Make your contribution such as it is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange in which you are engaged". If everyone adheres to this principle, communication will be successful. This is, of course, so long as the mutual goal of conversational participants is to exchange information about the world in the best possible way. However, this is not the only conceivable goal – conversational participants might also strive to establish their view of the world regardless of whether this view is considered true or not by other interlocutors. This is the dominative notion of context update, which is that every speaker aims to reach a conversational state in which his beliefs are shared by all other conversation participants. The cooperative and the dominative principles exist in every conversation to various degrees, and therefore both need to be represented. This representation, however, is not easily done in a Stalnakerian framework because of the nature of the common ground. As defined, the common ground requires that conversational participants share beliefs and have mutual expectations, but this is not the case if every conversational participant has a different agenda that possibly has nothing to do with beliefs, which she wishes to be 18 accepted by the others. Thus, the Stalnakerian framework as it is can't adequately represent the dominative principle at work. Another problem stems from the limited number of moves in the 'language game' proposed by Stalnaker – either attempts to reduce the size of the set in certain ways or rejections of such moves by others. Thus, this system allows for a speaker to assert a proposition like 'the dog is on the lawn' and for the hearer to either accept or reject this proposition. But what happens if things are not so clear, as is usually the case? When the hearer does not have enough information at her disposal to either rule out or accept the proposition, Stalnaker's system doesn't have a way to deal with it. This relates to the third problem, which stems from the definition of the context set as “the set of possible worlds recognized by the speaker to be the ‘live options’ relevant to the conversation”(Stalnaker, 1978; pp. 321–2). This is a good way to view the possible worlds in the context set, and it should not be abandoned, but in everyday conversation there are always propositions which are asserted and are very much 'live options' even though they are not accepted and thus not part of the context set. Consider the following example of a dialog from Ginzburg (1996: 3): (2) A: Bill left. B: Are you sure? A: I saw his car drive away. B: That's impossible: I hear his voice upstairs. A: Look, his secretary just told me he's left. The last step in our motivation to expand the Stalnakerian picture comes from the proposal nature of assertion, which, as argued in Farkas & Bruce (2010),is not much dealt with in the Stalnakerian account. Although assertion according to Stalnaker is a proposal to change the common ground, the Stalnakerian system does not really specify what it means for an assertion to be a proposal and how this proposal status manifests. Farkas & Bruce argue that theories of context update should not disregard this proposal 19 nature since asserted propositions affect the context in significant ways even if they are not accepted, but only proposed. All of the abovementioned problems seem to stem from the same underlying cause, which is that the Stalnakerian system tells us how assertions affect the context, and what happens when assertions are proposed and then accepted or rejected. But it does not tell us what happens to an assertion from the moment it is performed until the moment it is accepted or rejected. This is an important part of conversation; it may last for a long time during which the status of the assertion in the Stalnakerian system is unclear. However, it is clear that the force of assertion is still in effect, i.e. the assertion is not inert, even when it is in this 'limbo' state. Thus, there is a need to expand the Stalnakerian framework, which is the topic of the next section. 1.2.2 Expanding the Stalnakerian framework Expansions and revisions to the Stalnakerian framework are suggested by many theories. For example, dynamic theories of meaning (inter alia Dekker, 1993; Groenendijk & Stokhof, 1990; Heim, 1983; Kamp, 1981; Landman, 1984; Veltman, 1996) develop the Stalnakerian idea of context update into a semantic system that deals with the non-static development of discourse information as a consequence of the addition of new information by discourse participants, with the underlying conception of meaning as context change potential. Reinhart (1981) discusses a structure paralleling the common ground in terms of discourse topics, Carlson (1982) also talks about discourse topics in an account of dialogue games, and Erteschik-Shir (1997) views the common ground as a set of file cards representing discourse topics. Roberts (1996) and Ginzburg (1996a) offer a new discourse entity, Questions Under Discussion, Gunlogson (2013) offers the new discourse entity public commitments, and Farkas & Bruce (2010), propose the table. This dissertation combines features from the abovementioned theories, with a focus on Ginzburg (1996a), and Farkas and Bruce (2010). 20 Ginzburg (1996a) proposes a Lewis (1979) style conversational scoreboard termed gameboard which contains, in addition to a set of known facts corresponding to the propositions in the Stalnakerian common ground, a set of Questions Under Discussion (QUD). This set contains issued that are brought up during conversation and need to be resolved, in the form of questions. Thus, if a certain proposition p is accepted into the common ground, it gets updated into facts but if this proposition is under dispute then until it is fully resolved it's added into QUD as the question whether p. The process of context update, according to Ginzburg (1996a) is that the speaker asserts p, and the hearer then has the option of accepting p and updating it into the common ground, or to discuss whether p. In the latter option, the hearer adds the question whether p as a topmost element in her QUD, and produces an utterance that is relevant to a discussion of this question. Ginzburg's QUD is a good place to start, but it is unsuitable for the purpose of this dissertation, that deals with degree based assertion-modifying expressions, as such degrees are not represented in Ginzburg's theory. Thus, the discourse element of discussion needs a different representation here. Such a representation is found in Farkas and Bruce (2010), who discuss a transitory discourse stage, the table, upon which discourse items that the speaker publicizes her commitment to (based on Gunlogson's (2013) public commitments) are placed before they are accepted by other participants in conversation. Placing an item on the table is a way of proposing a new addition to the common ground. This action affects the context by projecting a set of the future context state called a projected set, which has the potential of replacing the current context state and becoming the new common ground. Similarly to Farkas and Bruce, the proposed account incorporates a transitory stage in which assertions that are performed and not yet accepted or rejected reside, termed Negotiation Zone (cf. section 1.3.8). While Farkas and Bruce do not discuss the mechanism by which interlocutors decide whether or not to accept propositions that are on the table, the proposed theory offers an account of this process. Moreover, this theory assigns an important role to the degree of strength of the assertion which is an element that is lacking from the abovementioned previous accounts. 21 1.3 Probabilistic common ground 1.3.1 Why probability This dissertation is done within a probabilistic framework. A question that immediately comes to mind (cf. Cohen, 2009 for further discussion of this question): why should we abandon formal logic and replace it with probability? The short answer to this question is that we do not have to. Formal logic is a powerful and helpful tool whose worth is tried and tested and there's no reason to abandon it. The long answer is that while we do not have to completely abandon logic and the principle of bivalence, there is a lot to benefit from adding probability to our theoretical toolkit. The first motivation for probability comes from its use as a tool in many different disciplines. As Hajek (2011) puts it: "Probability is virtually ubiquitous. It plays a role in almost all the sciences. It underpins much of the social sciences - witness the prevalent use of statistical testing, confidence intervals, regression methods, and so on. It finds its way, moreover, into much of philosophy. In epistemology, the philosophy of mind, and cognitive science, we see states of opinion being modeled by subjective probability functions, and learning being modeled by the updating of such functions. Since probability theory is central to decision theory and game theory, it has ramifications for ethics and political philosophy. It figures prominently in such staples of metaphysics as causation and laws of nature. It appears again in the philosophy of science in the analysis of confirmation of theories, scientific explanation, and in the philosophy of specific scientific theories, such as quantum mechanics, statistical mechanics, and genetics. It can even take center stage in the philosophy of logic, the philosophy of language, and the philosophy of religion". But perhaps we should not infer from the successful use of probability in non-linguistic disciplines that it should be applicable to linguistics as well. In that case, there is also plenty of linguistic motivation, starting with the simple fact that probability is used in everyday language. If we wish to represent utterances involving probability such as the 22 following, we need to incorporate probability into our semantics in one way or another: (3) a. There is a low/high/80% probability that the Bulls win the game. b. The chances of Jordan making it as a golfer are slim to none. c. In this time of year, any given day is 80% likely to be rainy. In addition to everyday use, there are many linguistic phenomena that benefit from a probabilistic explanation. Cohen (2009) discusses some of these, e.g. generics and frequency adverbs, vagueness and conditionals. He further argues that the use of probabilities needs not even be too intrusive as to rule out formal logic. In fact, probability can be employed while preserving the Tarskian biconditional. Gradable adjectives are another phenomenon in which the use of probability is very beneficial. The most widely held semantic theory of adjectives currently is based on Kennedy (2007), in which gradable adjectives are represented in terms of degrees on a scale. More specifically, adjectives such as long, cold, smart, deep, fast, good, happy, popular etc. are taken to be measure functions from entities to appropriate degrees. This type of semantics, which is very much at use in standard current semantic theories, motivates the use of degrees in semantics. And, as probability is a type of a measure function, it is not farfetched to think of probabilistic semantics as a representation of human cognition. At the very least it suggests that probability has a role to play in semantics and pragmatics. Moreover, as argued by Yalcin (2010) and Lassiter (2011) among others, possible, probable, certain and likely are gradable adjectives which convey epistemic modality. And, since gradable adjectives are already standardly represented by degrees it is natural to represent epistemic gradable adjectives in probabilistic terms. And, once it is established that certain types of epistemic modals benefit from a probabilistic representation, a unified account of all epistemic modals in terms of probability becomes desirable. 23 1.3.2 What is probability The question to deal with before going forward with a probabilistic account is the nature of probability, i.e. what does it mean to say that a certain event has a probability of 0.7? The answer is not a simple one. The philosophical literature on probability is vast and complex, and there are many different theoretical ways to define what probability is (for an overview, cf. L. J. Cohen, 1989; Gillies, 2000; Hacking, 2001; Hajek, 2011; Mellor, 2005). Not only are there many ways to define probability, but the terms vary as well. Thus, even the task of partitioning the theories is not an easy one. For example, Gillies (2000) categorizes probability theories into classical, logical, subjective, frequency and propensity. Mellor's (2005) categories are chances, epistemic probabilities, and credences, and Hacking (2001) simply divides probabilities into belief-types and frequency-types. I will follow Hajek (2011) and divide probability theories into three main concepts: A. A quasi-logical concept that measures the degree of evidential support a claim about the world has. For example, “in light of the relevant seismological and geological data, it is probable that California will experience a major earthquake this decade”. B. The subjective concept of an agent's degree of belief. For example, “I am not sure that it will rain in Canberra this week, but it probably will.” C. An objective metaphysical concept that applies to how the world is like, independently of what anyone thinks. For example, “the probability that a particular radium atom will decay within 10,000 years is high”. Of these three concepts, the first two are of importance to this dissertation as they deal with different notions of belief, the 'objective' evidence-based belief and the 'subjective' which is based on more than evidence alone. The third concept of probability, corresponding to objective reality, is irrelevant to this dissertation. Assigning the first two probability concepts to two types of belief is not something idiosyncratic to this dissertation. For example, Halpern & Fagin (1992) differentiate between two types of probabilistic belief functions – the first is as a generalized 24 probability function, and the second is as a way of representing evidence via mapping from probability functions to probability functions. They show how several puzzles raised in the literature can be solved once belief functions are not treated as uniform but rather divided into either measures of evidence or generalized probability. The former pertains to the first abovementioned concept of probability and the latter to the second. I will term the first type of probability, which pertains to belief as evidence, descriptive and the second type of probability which pertains to belief as subjective degree of confidence in the truth of a proposition expressive, a terminology which will be clarified ahead, and will used throughout the course of this dissertation5. 1.3.3 Probabilities of probabilities The use of probability is usually done with regards to some event, but interestingly, probabilities can also be applied to other probabilities, as in the next example from (Kyburg, 1988): "I might say that the probability that a coin will yield heads on a certain toss is "almost certainly" a half, i.e. that the probability that the probability is a half is very close to one. In contrast, I might say that the probability that a certain person will choose a blue tie, given that she is wearing a blue suit, is 0.8, but I may be no more than 50% confident of my probability judgment. That is, I might say that the probability that the probability is 0.8 is less than 0.5." (Kyburg, 1988:31) The matter of whether probabilities of probabilities, i.e. higher order probabilities constitute a qualitatively different type than 'lower order' ones is under dispute. de Finetti (1977) for example claims that such a concept as a probability of a probability is meaningless, Pearl, (1987) accepts the notion but claims that there is no need to posit two different probability types since both higher and lower order probabilities can be accounted for by classical probability. Kyburg (1988), on the other hand, finds the notion of probabilities of probabilities having a different type quite natural: 5 Note that this view, i.e. the distinction between expressive and descriptive probability and the two types of belief is not shared by Ariel Cohen. 25 "Higher order probabilities, if they are given a different interpretation from that given to first-order probabilities, make more sense. The most natural view discussed by Brian Skyrms… is that first-order probabilities represent propensities or frequencies-objective facts about the world and second-order probabilities represent degrees of belief. I may have a certain degree of belief that the propensity of this die to land five in the long run is 0.3. The degree of belief (the second-order probability) and the long-run relative frequency (the first-order probability) are clearly distinct." I will use the abovementioned Skyrms (1980) as a stepping stone to the claim that just as probability as propensities or frequencies can be embedded by probability as degrees of belief, so can probability as a measure of evidence6. Thus, probability as a measure of evidence is more subjective than probability as propensity or frequency, but still more objective than probability as degrees of belief. For the technicalities of how probabilities of probabilities are computed, cf. Kyburg (1988). For the purposes of this dissertation, the details of how this is done are less important than the hierarchical order. The hierarchy goes from the most subjective to the most objective types of probability, when the former scope over the latter: (4) Belief-type probability >> evidence-type probability >> objective probability 1.3.4 Probability spaces I adopt Yalcin's (2012) definition of probability spaces as a tuple <W,s,A,f> such that: (5) a. W is a sample space, i.e. the set of all possible worlds. b. s W is a probability domain. c. A is a Boolean algebra of subsets of W including s. d. f is a probability measure7 defined over A. e. P() is a propositional function such that for any set of worlds W, model M, world w and assignment function g: 6 Note that this view is not shared by Ariel Cohen, cf. Cohen (2012). A probability measure is a real-valued function defined on a set of events in a probability space that satisfies the probability axioms. 7 26 [|P()|]M,w,g = f ( {wW | (M,w,g) |=  } ) f. P(s)=1 Each probability space is akin to an information state8, i.e. a cognitive state of some conversational participant containing his beliefs about the world and his representation of discourse including the information states of other discourse participants. As stated in Groenendijk, Stokhof, & Veltman (1995): "Information states contain two kinds of information: information about the world, and discourse information. In the end, it is information about the world that counts, but in acquiring such information through discourse, one also has to store information pertaining to the discourse as such". In order to have a formal account of the context update effects of probabilistic assertions the context set can no longer be simply a set of possible worlds since possible worlds by themselves do not carry probabilistic information. Yalcin's (2012) notion of context probabilism is therefore adopted, i.e. the view that the common ground admits of probabilistic landscaping. The common ground can be viewed in two compatible ways– either as the set of propositions shared by all conversational participants, or as the set of information states in which all the propositions in the common ground have the same probability value. As discussed above, each information state is a probability space, the domain of which is the set of possible worlds determined by the intersection of all the propositions a given conversational participant believes. By this view every proposition is defined in probabilistic terms and probabilistic propositions are similar to non-probabilistic propositions. For example: (6) The dog is on the lawn. (7) The probability that the dog is on the lawn is 0.2. The truth conditions of the proposition in (6) are that it has a probability of 1, hence an assertion of (6) has the following propositional content: 8 This intentionally relates to the dynamic perception of context as a set of information states. 27 (8) [| P (on-the-lawn (the-dog)) = 1 |] The truth conditions of the proposition in (7) are that it has a probability of 0.2, hence an assertion of (7) has the following propositional content: (9) [| P (on-the-lawn (the-dog)) = 0.2 |] The next section discusses the context update effect of probabilistic and non-probabilistic assertions. 1.3.5 Probabilistic context update Recall that in the standard Stalnakerian context update an assertion of a proposition  adds  to the common ground and eliminates all worlds in which  is false from the context set. Following and modifying Yalcin (2012),9 a probabilistic context update is one in which an assertion of a probabilistic proposition such as 'P() = x' adds 'P() = x' to the list of propositions in the common ground. Once this proposition is added, all conversational participants agree that the probability of  is at least x10, therefore all the probability spaces in which P() < x are removed from the common ground. Thus, for a non-probabilistic assertion: (10) The dog is on the lawn. The update will be the set of probability spaces in which: (11) [| P (on-the-lawn (the-dog)) = 1 |] 9 Yalcin's context update process is expressive, i.e. unlike in the account proposed here, the common ground in Yalcin's account is not updated directly. Instead, when a speaker asserts a proposition, she expresses her credal state with regards to this proposition and implicitly proposes that other conversational participants adopt the same state towards it. If this happens, the speaker presuppositions change, thereby updating the common ground. 10 Since every probability space in which 'P() ≥ x' entails 'P() = x'. 28 And worlds in which the probability is lower than 1 are removed from the common ground. As for a probabilistic assertion: (12) The probability that the dog is on the lawn is 0.2. The update will be the set of probability spaces in which: (13) [| P (on-the-lawn (the-dog)) = 0.2 |] And probability spaces in which this probability is lower than 0.2 are removed from the common ground, leaving of course probability spaces in which the probability is higher since it entails a probability of 0.211. Thus, in this system, both probabilistic and non-probabilistic assertions are represented by propositions and updated into the common ground in the same way. The standard Stalnakerian context update is now a special case of probabilistic context update – the case in which the probability of the asserted proposition is 1. However, for the sake of simplicity, whenever the propositional content is P() = 1 the representation will simply be , unless otherwise required. 1.3.6 Assertion operator As previously discussed, the speech act of assertion contains, in addition to propositional content, various other elements. Searle and Vanderveken (1985) divide illocutionary force into 7 components: 1. Illocutionary point: The point and purpose of the speech act, i.e. what speakers aim to do when they perform the speech act. The point of assertion is presumingly to convey information about the nature of the world12. 11 Similarly to an update of numerically quantified propositions such as 'John has 3 children', where all the worlds in which John has less than 3 children are removed. 12 For a discussion about the aim of assertion, see section 1.3.9. 29 2. Degree of strength of the illocutionary point: The same illocutionary point can be achieved with varying degrees of strength. For example, the illocutionary point of getting the addressee to do something can be achieved by requesting or by insisting. The latter has greater illocutionary strength. 3. Mode of achievement: Some speech acts require a special mode by which the illocutionary point can be achieved. For example, the illocutionary point of performing marriage is achieved via the mode of being appropriately licensed to marry people. 4. Propositional content conditions: There are certain restrictions on the propositional contents appropriate for various speech acts. For example, the propositional content of an apology can't contain actions that the speaker is not responsible for, it's inappropriate to apologize for something you had nothing to do with. 5. Preparatory conditions: Restrictions on the conditions that need to be fulfilled before the speech act can be adequately performed. For example, in order to marry two individuals the person performing the speech act of declaring has to have the appropriate license. 6. Sincerity conditions: In order for a speech act to be performed the speaker has to possess the appropriate psychological state corresponding to the speech act. For example, in order to request something the speaker has to want it, otherwise the speech act is not sincere. The sincerity condition for assertion is belief, since in order to perform an assertion sincerely the speaker has to believe in the asserted propositional content. 7. Degree of strength of the sincerity conditions: The same sincerity condition can exist with varying degrees of strength. For example, the speaker can be sincere while either requesting or imploring, i.e. in both cases the speaker wanted the thing she asked for. However, the degree of this sincerity condition is higher in the case of imploring. As for assertion, the speaker can believe in the asserted 30 propositional content with varying degrees of strength, ranging from skepticism to total conviction. The account proposed here minimally divides the assertion into two components13, propositional content and degree of strength. This degree of strength corresponds by default to the seventh abovementioned speech-act component i.e. the degree of strength of the sincerity conditions, which in the case of assertion is belief. However, there are cases in which this degree of strength is generalized and becomes the second abovementioned speech-act component, i.e. the Degree of strength of the illocutionary point. One of these cases is discussed in chapter 4 that deals with clarity, and the other case discussed in section 2.4 of chapter 2 that deals with modal concord. The following assertion operator is proposed: (14) Ax<S,C> The first argument, S, stands for the degree of strength by which the assertion is performed and the second argument C is the assertion's propositional content. The order of arguments stands for relative scope, i.e. the degree of strength scopes over the propositional content. Thus, a shorthand representation14of this assertion operator in probabilistic terms, i.e. the assertion of propositional content  with a degree of strength that is the value v of a probability function P, is: (15) Ax P() = v As recalled, it is important for speech acts to be performed sincerely, and the sincerity condition for assertion is belief. This is the reason why an utterance such as the following seems like a contradiction even though logically it isn’t: (16) #I went to the pictures last Tuesday, but I don't believe that I did.(Moore, 1942) 13 I remain agnostic regarding the other components. They are not used in the current account as they are not needed in order to explain the linguistic phenomena discussed. 14 Given that the asserted propositional content is the same as the propositional content that the degree of strength is applied to. 31 Of course, it is not a logical contradiction that it is true that the speaker went to the pictures and that it is true that the speaker does not believe it. But it is inappropriate to assert such a thing, and speech act theory has a straightforward explanation in terms of a violation of the sincerity condition - the speaker has to believe in what she asserts. And, since the theory proposed in this dissertation deals with degrees of belief, a belief norm of assertion15 is adopted and modified to a degree based account: (17) One must assert that p only if one believes to a high degree that one knows that p. Adopting such a norm for assertion is in line with standard pragmatic theory e.g. Grice's maxim of Quality (Grice, 1975): "Contribute only what you know to be true. Do not say false things. Do not say things for which you lack evidence". As discussed in Davis, Potts, & Speas (2007), Gricean quality has a parallel in Lewis (1976): Lewisian quality: “The truthful speaker wants not to assert falsehoods, wherefore he is willing to assert only what he takes to be very probably true. He deems it permissible to assert that A only if P(A) is sufficiently close to 1, where P is the probability function that represents his system of degrees of belief at the time. Assertability goes by subjective probability.” (Lewis 1976:133) Davis, Potts, & Speas (2007) represent quality by the speaker's degree of belief and propose that contexts come with a quality threshold which is a numerical value in the real interval [0,1], i.e. a probability value. In the same vein, this dissertation represents unmodified assertions as performed with a default degree of strength equal to or greater than high, when high stands for some numerical degree close to 1. Thus, an assertion of: (18) The dog is on the lawn. 15 cf. Lackey (2007) for a survey of norms of assertion. 32 Is represented by the following formula: (19) Ax P(on-the-lawn(the-dog)) ≥ high The speaker x asserts the propositional content 'the dog is on the lawn' with the default degree of strength for assertion which is equal to or greater than high. And an assertion of: (20) The probability that the dog is on the lawn is 0.2. Is represented by the following formula: (21) Ax P( P (on-the-lawn(the-dog)) = 0.2) ≥ high The speaker x asserts the propositional content 'the probability that the dog is on the lawn is 0.2' with the default degree of strength for assertion which is equal to or greater than high. As can be seen, this is a case in which expressive probability i.e. the degree of belief of the speaker, has wide scope over descriptive probability i.e. the statement that the world is such that there is a 0.2 probability that the dog is on the lawn. 1.3.7 Mixture model As recalled, the context update process involves a speaker who performs an assertion and a hearer who has to decide whether to accept, reject, further discuss, or agree to disagree on it. The hearer's decision process takes into account various sources of evidence at the hearer's disposal. These sources may include direct knowledge (e.g. perception), deductive processes (e.g. inference), or reported information (e.g. hearsay).16 The evidence provided by each source is not, in general, conclusive, for two reasons. One is that some sources are more reliable than others. For example, the accuracy of perception will depend on the reliability of the sense organ (e.g. eyes) or sensing device (e.g. camera); the weight of hearsay evidence will depend on the reliability of the 16 If this classification of knowledge sources brings evidentials to the mind of the reader, this is not a coincidence. 33 reporter, etc. The second reason why the evidence is not conclusive is that the evidence provided by the source is generally, by its nature, graded and not certain. For example, one possible source of evidence is inference—but inference is often probabilistic. Even if the calculation of probabilities is entirely accurate, the result is still only a probability measure, not a definite result. What is less often acknowledged is the fact that reported evidence is also graded: the speaker may believe what she says to various degrees, and, consequently, make statements with varying illocutionary strengths. A speaker who utters (22) expresses more confidence than a speaker who utters (23) who, in turn, expresses more confidence than a speaker who utters (24)17: (22) The dog is certainly on the lawn. (23) The dog is on the lawn. (24) The dog is possibly on the lawn. The different illocutionary strength of the utterances must be taken into account by the hearer in deciding whether or not to accept their propositional content. In order to formalize the two factors that affect the relative strengths of different sources of evidence, we assign two values to each source of evidence i: a probability measure, Pi(), indicating the strength of the evidence for  according to source i, and a weight wi, indicating the reliability of the source. Of particular interest here is a specific type of reported information: an unmodalized assertion made by the speaker. Thus, if the speaker asserts (23), the probability value will be high, and the weight will reflect how reliable the speaker is considered to be. The probability that the hearer assigns to a proposition is a weighted sum of the probabilities assigned to it by the various sources. The sum of the weights is equal to 1, so that the result can be easily shown to be a probability measure itself: (25) P(  ) =  wiPi(  ) n i=1 17 At this stage, the claim is based on the reader's intuition, but see chapter 2 for a formal explanation. 34 The formula states that the probability of  is the sum of weighted probabilities assigned to  by all available sources of evidence pertaining to . This type of probability measure is often referred to as a mixture model (cf. McLachlan & Peel, 2000). Note that because the sum of the weights is 1, the value of the probability does not necessarily increase if the number of sources increases. If P() exceeds the hearer’s threshold of acceptance18, the proposition is accepted; if P() exceeds this threshold, the proposition is rejected. Otherwise it is left in the negotiation zone – which is discussed in the next section. 18 The assumption is that, by default, the threshold of acceptance is the same degree as the default for standard assertion, i.e. high. However, this threshold can change, cf. Davis, Potts, & Speas, (2007) and Davis (2009). 35 1.3.8 Negotiation Zone The Negotiation Zone (NZ) is a discourse entity similar to Farkas and Bruce's table, but unlike the table, the NZ hosts propositions coupled with degrees of strength, the first argument of the assertion operator. The NZ is a set of assertions, not propositions. This means that its members have propositional content and degrees of strength as well as a record of the assertion performer i.e. a set of assertion operators. Whenever a conversational participant performs an assertion, the appropriate assertion operator is added to the NZ. For example, when the speaker performs the following unmodified assertion: (26) The dog is on the lawn. This assertion has a degree of strength based on the speaker's information state, i.e. the speaker's information state is such that the asserted proposition has a degree of probability of equal to or greater than high on it. Now the hearer has to take this proposition under consideration. She inspects her own information state, which contains various evidential sources available to her which pertain to the asserted proposition. i.e. the hearer activates a mixture model. While this happens the speaker's assertion is neither accepted nor rejected. Both the speaker and hearer have a representation of the speaker's assertion,19 but even though this representation is part of the common ground, the propositional content of this representation can't be as long as it is not accepted and updated. What happens to it? As discussed in example (2), repeated here, even in this state of 'limbo' assertions still have effect on the discourse Ginzburg (1996: 3): (27) A: Bill left. B: Are you sure? 19 The common ground and the information states within it contain a register of discourse that includes details of assertions performed during discourse (cf. Groenendijk et al., 1995; Stalnaker, 1978). 36 A: I saw his car drive away. B: That's impossible: I hear his voice upstairs. A: Look his secretary just told me he's left. Moreover, assertions can stay in this indefinite state for a very long period of time (Ginzburg, 1996b): (28) B(1): after all your father's a generation younger than my father isn't he, basically a(2): well I should think so B(3): cos your father's now SEVENTY is he a(4): seventy two or (simultaneous with next turn:) seventy three B(5): SEVENTY TWO. yes. well father would have been seventy eight I suppose. if he had been alive still C(6): good lord a(7): that all ?(8): goodness C(9): my father would have been a hundred and twenty seven B(10): no, not seventy eight, eighty eight a(11): (coughs) B(12): yes. (simultaneous with next turn:) no. yes a(13): as bad as Charlotte's B(14): no, not seventy eight. yes, eighty eight. no, I'm sorry, sixteen years, yes eighty eight, that sort of thing 37 C(15): a half generation then The issue raised in step (1), i.e. the proposition (or question whether) 'your father's a generation younger than my father', is resolved only in step (15). Of course, it is still very much relevant to conversation, and thus must be stored in a form that can be examined by other discourse participants in order to reach a conclusion about it. The form is the assertion operator, the place of storage is the NZ and the examination is done via the mixture model. After the activation of the mixture model there are three possible results – a) A probability value of greater than high – the assertion is accepted by the hearer. b) A value of lower than low – activation of a mixture model to check if the negation exceeds high, and if it does the assertion is rejected by the hearer. c) Any other value – the assertion is under negotiation, i.e. neither accepted nor rejected by the hearer. If the result is a probability value of greater than high, the hearer utters an acceptance phrase that enables the speaker to understand that the hearer accepts the assertion. The speaker than incorporates the hearer's response as a new source of evidence and activates a new mixture model that includes the hearer, her weight and her judgment. If the result is greater than high the speaker utters some acceptance phrase back and the proposition is updated into the common ground, conversational participants update their beliefs in the asserted proposition to a full degree of 1, and the proposition is removed from the NZ. Note that even if a proposition is accepted and as a result updated into the common ground, belief is corrigible, i.e. conversational participants are always prepared for the possibility that their beliefs may turn out to be false. In this regard, following Cohen (1996), I adopt an action analysis of belief (cf. Braithwaite, 1932) according which a belief that p is a disposition to act as if p were true, with respect to some practical objective. This accords well with Stalnaker's (1978) claim that: "A speaker presupposes that P at a given moment in a conversation just in case he is disposed to act, in his 38 linguistic behavior, as if he takes the truth of P for granted, and as if he assumes that his audience recognizes that he is doing so". It is also important to note the difference between probability update and probability revision. As discussed in Wang (1993), probability update occurs when a previous degree of probability assigned to a proposition p is assigned new value. This is what happens when a proposition is added to the common ground. If the proposition is nonprobabilistic, e.g. 'the dog is on the lawn' then the update is from any previous degree of probability to 1. If the proposition is probabilistic, e.g. 'there is 20% probability that the dog is on the lawn' then the update is from any previous degree of probability to 0.2. Probability revision, on the other hand, is when a previous degree of probability assigned to a proposition p is affected by various factors and changes to a new value accordingly. This is what happens when the mixture model is activated – a previous degree of belief is affected by various evidential sources and information states with varying degrees of belief with regards to the same proposition as will be discussed shortly. In the second option, in which the result is a probability value of 0, the hearer utters a rejection phrase that enables the speaker to understand that the hearer rejects the assertion. The speaker than incorporates the hearer's response as a new source of evidence and activates a new mixture model that includes the hearer, her weight and her judgment. If the new result is lower than high the speaker can retract the assertion, and the item stays in the NZ pending new evidence. Let's go through a scenario of assertion and acceptance that will help understand the process: Tom and Jerry go out to an evening in Tel-Aviv and have to return back to Beer-Sheva before the last train leaves. Tom remembers looking at a train schedule according which the last train leaves at 23:00. He therefore asserts to Jerry: 'the last train leaves at 23:00'. In order to sincerely and felicitously perform this assertion, Tom's degree of belief in the asserted proposition has to be at least equal to or greater than high. Jerry hears Tom's assertion, activates his mixture model, which results in a probability which is also greater than high. He therefore accepts Tom's assertion, uttering 'OK', 39 indicating to Tom that the result value of Jerry's mixture model is greater than high. Jerry's acceptance phrase is also, of course, an assertion, and every assertion places the asserted item into the NZ and triggers an activation of the mixture model by the hearer in order to revise probabilities. Thus, Tom activates the mixture model and revises the probability he assigns to his previous assertion in light of Jerry's acceptance. This is actually a redundant step in this case, since Tom's mixture model has already surpassed the threshold of acceptance, but there are cases in which this step is not redundant. The next step is an utterance of some acceptance phrase by Tom. The last step is context update – both conversational participants update their belief in the asserted proposition to a full degree of 1, and the proposition is added to the common ground20. The following is a discourse table of the steps described above: Discourse steps Speaker: Tom Negotiation Zone Hearer: Jerry Step 1: Assertion Tom1 = {PTom()= NZ1 = NZ0  Jerry1 MIX1Tom() ≥ high} {ATomP() ≥high} Tom2 = Tom1 NZ2 = NZ1 Step 2: Mixture-Jerry = {PJerry() =MIX1Jerry() = undefined } Jerry2 = { PJerry() = MIX2Jerry()≥high} Step 3: Acceptance- NZ3 = NZ2  Tom3 = Tom2 Jerry Jerry3 = Jerry2 {AJerry P() ≥high} Step 4: Mixture-Tom Tom4 = {PTom() = NZ4 = NZ3 Jerry4 = Jerry3 NZ5 = NZ4  Jerry5 = Jerry4 MIXTom()≥ high} (redundant) Step 5: Acceptance- Tom5 = Tom4 Tom {ATom P() ≥high} (redundant) Step update 6: Context Tom6 = {PTom() = NZ6 = NZ5 - MIXTom() = 1} Jerry6 = { PJerry()= MIXJerry()=1} 20 This model is perhaps somewhat idealized in assuming that once a proposition is accepted all doubts disappear and it gets a value of 1. Whether this is in fact the case is a matter of further empirical study. 40 {Ax <S,C>| C = } Result: CG' = CG  Figure 1.2: Discourse table 1.3.9 The aim of assertion As recalled from section 1.3.6, the speech act of assertion is composed of propositional content and illocutionary force, when the latter is divided into seven components. The first of which, the Illocutionary point stands for the aim of assertion. But what is the aim of assertion? The literature has several answers for this question. Macfarlane (2010), surveying the philosophical literature divides the answers into four main categories: (29) a. To assert is to express an attitude. b. To assert is to make a move defined by its constitutive rules. c. To assert is to propose to add a proposition to the common ground. d. To assert is to undertake a commitment to the truth of the asserted proposition. The theory proposed in this dissertation suggests a fifth option – to assert is to 'refresh' the discourse state. The speaker who performs the assertion introduces a new piece of evidence with regards to the asserted proposition. This, in turn, brings about an activation of a mixture model by every conversational participant, which causes a revision of previously assigned probabilities. As a consequence, this can bring about an update of the asserted proposition to a new probability value by every conversational participant and the addition of the proposition into the common ground, but it doesn’t have to. The essential effect of assertion is, thus, not the addition of a proposition into the common 41 ground, but rather a feasibility check of a proposition in light of new information. This effect is achieved by virtue of expressing an attitude, as in (29), with the addition of gradability, i.e. different degrees this attitude may be expressed by. 42 Chapter 2 Modal adverbs and modal adjectives 21 This chapter deals with a subclass of epistemic modals called 'modal adverbs' i.e. the adverbs possibly, probably, certainly and their relation to the corresponding modal adjectives possible, probable, and certain22. Sentences containing modal adverbs seem on a first glance to be equivalent to ones containing modal adjectives: (1) a. It's possible that the dog is on the lawn. b. The dog is possibly on the lawn. The immediate impression of both sentences is that one is a variant of the other, both convey the possibility that the dog is on the lawn. And indeed, the literature by and large does not distinguish between modal adverbs and modal adjectives as well (cf. Ernst, 2009; Jackendoff, 1972; Jacobson, 1978; Kratzer, 1981; Perkins, 1983). However, what seems at first glance to be the case turns out to be more complicated than expected. The chapter opens with data on modal adverbs and adjectives, continues with theories that deal with modality in general and with the modal adverbs and modal adjectives in particular, and proposes a novel formal account of adverbs as negotiation chips. 2.1 Modal adverbs and adjectives - data The first and most recognizable feature of modal adverbs that distinguishes them from modal adjectives is their property Speaker orientedness (cf. Jackendoff, 1972) i.e. 21 This chapter is based on Wolf & Cohen (2009). However, note that the view expressed in this dissertation according which modal adverbs and modals adjectives are both epistemic is different than the view expressed in Wolf & Cohen (2009). 22 This chapter does not deal with non-epistemic interpretations of modal adjectives. 43 utterances containing modal adverbs convey that the person whose judgment is reflected by these adverbs is the speaker. This feature does not standardly manifest in utterances containing modal adjectives, as can be seen in the following example (Nuyts, 2001a): (2) A: It is probable that they have run out of fuel. B: Who says so? (3) A: Probably they have run out of fuel. B: #Who says so? While the question (2) is quite natural, i.e. the hearer wants to know whose judgment is conveyed by the speaker's assertion (while it may be the speaker's, it doesn’t have to be), the question in (3) doesn’t sound quite right. The feeling arises since in the modal adverb case it seems clear23 whose judgment it is. The Kratzerian theory of modality, discussed previously in this chapter, offers a unified account of modality, which is one of its greater strengths. In accord with this unified account, the common assumption regarding modal adverbs and modal adjectives (e.g. Ernst, 2009; Jackendoff, 1972; Jacobson, 1978; Perkins, 1983) is that they are different grammatical variations on the same semantic meaning. However, as raised by Bellert (1977) and explored in Nilsen (2004) and Piñón (2006, 2009) modal adverbs are not semantically synonymous with modal adjectives, a fact which is manifested by different behavior patterns in various linguistic constructions. The following subsections discuss the various ways in which modal adverbs differ from modal adjectives. 23 The intuition arises since modal adverbs have a strong tendency to be speaker oriented, but it's important to note that this tendency is contextual, as for all other modals. Thus, in this example as in others, judgments should be performed on a scale, i.e. comparing between the examples and estimating the salient reading in each accordingly. 44 2.1.1 Embedding While modal adjectives are easily embedded under negation, modal adverbs are very difficult (Bellert, 1977)24: It's impossible/not possible that It's improbable/not probable that (4) a. ,John has/will come. It's uncertain/not certain that #Impossibly/Not possibly #Improbably/Not probably #Uncertainly/Not certainly b. ,John has/will come. While modal adjectives are easily embedded under questions, modal adverbs are very difficult (Bellert, 1977): possible probable (5) a. Is it certain that John has/will come? possibly probably b. #Has/Will John certainly come?25 24 Bellert also uses the pair evident/evidently. This is replaced by certain/certainly here, in order to keep the evidential component away. 45 The third embedding difficulty of modal adverbs, noted in Piñón (2006), is in the antecedent of conditionals: possible probable (6) a. If it is certain that the socialists will win the elections, the rich will worry about a luxury tax. possibly probably b. #If the socialists certainly win the elections, the rich will worry about a luxury tax. However, modal adverbs are good in the consequent of conditionals: possibly probably (7) If the socialists win the elections, the rich will certainly worry about a luxury tax. 25 It's important to note that stressing the modal adverb in these examples results in higher acceptability. While I shall not deal with focus in this dissertation (but cf. Döring (2012) for an account of focus sensitivity in modal adverbs), I suggest that the reason for the higher acceptability stems from the main effect of focus which is bringing a discourse item from the background to the foreground of the assertion, i.e. from non-at-issueness to at-issueness. 46 These data are strengthened by an observation of Papafragou (2006), originally made with regards to modal auxiliaries but very much applicable to our case as well – while modal adjectives are easily embeddable under factives, modal adverbs are not so easy: possible probable (8) a. It is surprising that it is certain the socialists will win the elections. possibly probably b. #It is surprising that the socialists will certainly win the elections. Another embedding difficulty for modal adverbs is under tense: possible probable (9) a. It was that the socialists won the elections. certain possibly probably b. The socialists won the elections. certainly In the first example, the modal adjectives can be embedded under tense, hence the reading is that in the past there was a possibility/probability/certainty that the socialists 47 won. This option is not open for modal adverbs, hence the reading is that at present there is a possibility/probability/certainty that in the past the socialists won the elections26. 2.1.2 Agreements and disagreements Agreement and disagreement cases offer a good diagnostic of the different discourse effects of modal adverbs and adjectives, since agreements and disagreements can target the whole asserted proposition and not just parts of it. Papafragou (2006) uses an assent-dissent diagnostic on modal auxiliaries. This test can easily be converted to show the differences between modal adverbs and modal adjectives. The tests show a tendency for assents and dissents of utterances containing modal adverbs to target the prejacent27 and utterances containing modal adjectives to target the entire proposition including the modal adjective: (10) A: It's possible/probable/certain that John is at home. B: That's not true/I agree. = It's not true/The hearer agrees that it's possible/probable/certain John is at home. ≠ It's not true/The hearer agrees that John is at home. (11) A: John is possibly/probably/certainly at home. B: That's not true/I agree. ≠ It's not true/The hearer agrees that John is possibly/probably/certainly at home. = It's not true/The hearer agrees that John is at home. 26 Of course, the utterance in this example is only acceptable if the facts about the past are not mutually known. 27 A term adopted from Horn (1996) and von Fintel (1997), originated in medieval times, meaning the asserted proposition precluding said adverbs/adjectives. 48 An example which demonstrates this point can also be seen in MacFarlane (2011), originally intended to show the difference between the modal auxiliary might and the epistemic hedge 'for all I know'. Here, it is modified to show that a similar difference holds between the modal adverb possibly and the modal adjective possible: (12) First case: You overhear George and Sally talking in the coffee line. Sally says, “It's possible that Joe is in Boston right now”. You think to yourself: I know that Joe isn’t in Boston, because I just saw him an hour ago here in Berkeley. Question: Did Sally speak falsely? Second case: Scene as before. Sally says, “Joe is possibly in Boston right now”. You think to yourself: Joe can’t be in Boston; I just saw him an hour ago here in Berkeley. Question: Did Sally speak falsely? Intuitively, Sally is more committed to the claim that John is in Boston in the second case, while her statement in the first case seems to only raise a possibility. Thus, the second case is more open to dissent than the first. If the intuition does not seem so sharp here, it is due to the weak effect of the possibility modal as conveying a low degree of certainty. Replacing possibly with probably (and certainly certainly) enhances the effect. To these conversational examples I add another – the 'reason to assert' test: (13) Scenario: A reality show in which 15 participants are competing for the role of 'Israel's next top barista'. One of the participants got voted out, leaving 14. Option A: Spectator 1: It's possible that John will be Israel's next top barista. Spectator2: Why do you say that? 49 Spectator1: Well, Shirley was just voted out. Option B: Spectator 1: John will possibly be Israel's next top barista. Spectator2: Why do you say that? Spectator1: #Well, Shirley was just voted out. The fact that one participant got voted out seems reason enough to assert that it's possible that one of the other participants will win, but not that one of the participants will possibly win. 2.1.3 Nilsen's contrast The following contrast is presented in Nilsen (2004): (14) a. It's possible that Le Pen will win even though he certainly won’t. b. # Le Pen will possibly win even though he certainly won’t.28 If modal adjectives and modal adverbs are grammatical variants of the same semantic meaning, i.e. if there is a single scale running from possibility to certainty and both modal adverbs and modal adjectives occupy it, then both sentences should be equally bad since a certainty that something won’t happen means there is no possibility it will happen. However, since the conjunction of the modal adjective possible with the modal adverb certainly not is better than the conjunction of the modal adverb possibly with the same modal adverb as in the first example, this is clearly not the case. 28 Note that the difference is much stronger if the sentence is turned to the past: 'It was possible for Le Pen to win, even though he certainly didn’t' vs. #'Le-Pen possibly won even though he certainly didn’t'. 50 Note that when the sentential order of modal adjective and modal adverb is reversed the utterance becomes worse: (15) #Le Pen will possibly win even though it is certain that he won’t. This is another data item that needs to be explained. The next section discusses theories that attempt to account for these data and problems with these theories. 2.2 Modal adverbs and adjectives – theories As stated in the beginning of this chapter, modal adverbs and modal adjectives are a subclass of epistemic modals. The term 'epistemic modality' is derived from the Greek episteme which stands for ‘knowledge’. However, as stated in Kratzer (2012): "In spite of their name, epistemic modals do not have to relate to knowledge. The modals in the following examples show the same morphological and syntactic behavior as bona fide instances of what are commonly called “epistemic modals” in languages like English or German. Yet none of those modals necessarily relates to knowledge in a technical sense. Unlike knowledge, newspaper reports can be false. The term “epistemic modal” is a misnomer." Kratzer uses the following examples: (16) According to the newspaper reports, the thief must have entered through the kitchen window. (17) According to the testimony we heard, the defendant must be suffering from severe emotional distress. 51 Thus, it is more appropriate to view epistemic modality as belief, when belief can be viewed either as personal judgment or a measure of the available evidence 29 (cf. chapter 1). The next subsections discuss the classical Kratzerian theory of modality, and the various accounts of modal adverbs and adjectives. 2.2.1 Modality, the classical theory The classical theory of modality is due to Angelika Kratzer (inter alia 1977, 1981, 2012). It is a uniform account of all types of modality. This account treats all modals as truthconditional quantifiers over possible worlds (cf. Hintikka, 1961; Kripke, 1963). Possibility modals are existential quantifiers and necessity modals are universal. The domain of quantification is determined by two contextually supplied conversational backgrounds, the modal base and the ordering source. A conversational background is a function from worlds to sets of propositions. Once conversational backgrounds are established, a set of propositions corresponding to these conversational backgrounds are defined. This set determines an accessibility relation from each possible world to a set of possible worlds which are the domain of quantification for the modal. For example, with regards to the epistemically modalized utterance: (18) The dog might be on the lawn. The established modal base is epistemic, hence the contextually determined set of propositions stand for what is known in the world in which the speaker uttered the sentence. These propositions can include, for example: (19) a. The dog tends to be on the lawn. b. The dog wasn't seen in the house. c. The kids like to play with the dog. 29 It is important to note that the classic account of epistemic modality, as expressed e.g. in Kratzer (2012) and von Fintel & Gillies (2011) views EM as based on evidence rather than belief. But cf. Anand & Hacquard (2008), discussed in chapter 3, for a belief-based account of EM. 52 d. The kids went outside. Applying an intersection on this set results in a set of worlds in which all these propositions are true, and these are the epistemically accessible worlds which provide the domain of quantification. And, since the utterance contains a possibility epistemic modal, the utterance is true iff there is at least one possible world within the modal base in which the dog is on the lawn. But this is not enough. There are many possible worlds in which all the propositions of the modal base are true, and some are more feasible than others. For example a world in which dogs and kids have stable behavioral patterns is more likely than a world in which the dog changes tendencies regularly and the kids suddenly hate to play with the dog. Thus, there should be some way to represent the more likely worlds within the modal base30. Kratzer therefore employs a second conversational background, the ordering source, which imposes an ordering on the modal base, i.e. some worlds are better than others. In our example, the ordering source can be stereotypical, which means the set of propositions that describe what is normally expected to happen in the world. These propositions can include, for example: (20) a. Dogs have stable non-changing tendencies. b. Kids have stable non-changing animal preferences. Formally, a possibility epistemically-modalized sentence is represented as: (21) w  [|  |]f,g iff there is a w'  BEST( f (w), g(w)) s.t. w'  [|  |]f,g In words, 'might ' is true in a world w with respect to a modal base f and an ordering source gif and only if there is an accessible world w' which is a member of the best epistemically accessible possible worlds in which  is true31. For the purposes of the rest of the chapter, the most important aspects of Kratzer's theory are its uniform nature, i.e. that all modals have the same logical form, and that all modals 30 Of course, another important motivation for the dual conversational backgrounds is in order to account for cases of deontic modality, but this type of modality is not discussed here. 31 Assuming that there is always such a set, by adopting the limit assumption (Stalnaker, 1968). 53 are truth conditional, i.e. part of the propositional content of utterances. These aspects, which are a clear advantage, can also be a disadvantage in the case of modal adverbs and adjectives, which do not seem to fit the classical template. And indeed, as of yet there is no explicit contextualist account for these particles. 2.2.2 Bellert (1977) Bellert (1977) accounts for the differences between modal adverbs and modal adjectives in terms of different relations to the prejacent. Modal adverbs are predicates over the truth of the prejacent while modal adjectives are predicates over facts/events/state of affairs within the proposition. Thus, sentences with modal adjectives convey one proposition while utterances with modal adverbs convey two propositions – the prejacent and a metalinguistic proposition about the truth of the prejacent. According to Bellert this is the reason why modal adjectives can be negated and participate in questions while modal adverbs cannot. Repeating (5): possible probable (22) Is it certain that John has/will come? possibly probably (23) #Has/Will John certainly come? The pronoun it in (22) refers to the event of John's coming and the question is about the possibility/probability/certainty of this event occurring. It doesn’t make sense to ask about the truth of a proposition and qualify this truth at the same time. By the same token, the next example is infelicitous since it isn’t possible to qualify the truth of a negated proposition: 54 #Impossibly/Not possibly #Improbably/Not probably #Uncertainly/Not certainly ,John has/will come. (24) As an indication for the existence of two propositions in the case of modal adverbs (the prejacent and a metalinguistic proposition about the truth of the prejacent), Bellert claims that it's possible to turn sentences with modal adverbs into two separate statements when the first of which, containing the adverb, is an explicit statement about the truth of the other, while this can't be done with modal adjectives: (25) It is possibly/probably/certainly true that John has/will come. (26) #The truth that John has come is possible/probable/certain. Bellert's account has several problems, discussed in Piñón (2006, 2009). The first of which is that (27) is more acceptable than (25) even though both are sentences containing modal adjectives in which there are two statements, the first of which is a statement about the truth of the other: (27) It is a possible/probable/certain truth that John has come. The second is a problem about Bellert's requirement that adverbs modify the truth of a proposition while modal adjectives modify the facts/events/state of affairs within the proposition (cf. Bellert (1977: 343). However, (28) is still not an acceptable sentence: (28) # The fact/event/state of affairs that John has come is probable/possible/certain. 55 2.2.3 Nilsen (2004) Nilsen (2004) notes that modal adverbs are barred from the same linguistic environments that Negative Polarity Items (NPI) such as yet, ever, and any are acceptable in, i.e. downward entailing (DE)32. This holds for the previously discussed examples in the section on embedding. Negation is, of course, DE. So are antecedents of conditionals (cf. Kadmon & Landman, 1993; Linebarger, 1987), questions (cf. van Rooy, 2003) and factive presuppositions (Kadmon & Landman, 1993). Nilsen provides another example for a DE operator that can embed modal adjectives but not modal adverbs, never: (29) a. #Stanley never possibly/probably/certainly ate his wheaties. b. It was never possible/probable/certain that Stanley ate his wheaties. Following these data, Nilsen posits that modal adverbs are in complementary distribution with NPI and are thus Positive Polarity Items (PPI). Nilsen explains the differences between modal adverbs and modal adjectives in terms of levels of plausibility (cf. Chierchia, 2004), i.e. the degree by which a proposition p is accepted in an information state. For example, the degree of plausibility for possible is: (30) [|possible|] = p. f(p) low (Nilsen, 2004:830-831) When f stands for an entrenchment function (cf. van Rooy, 2006) that provides the degree of plausibility of a proposition and low stands for a contextually-determined numerical value. As a constraint on these degrees, if the plausibility degree of a proposition is low then the plausibility degree of its negation is high and vice versa. Let's take a moment to dwell on this constraint since it will be important later on. Say that the scale of degrees runs from 0 to 1, and that the degree of low is 0.1 and high is 0.9. Then the constraint on 32 A linguistic environment is DE when it reverses the semantic strength relation between expressions that are placed in it. For example, negation is DE because the semantic relation between the expressions 'The dog barked' and 'The dog barked loudly' (the latter is semantically stronger than the former since it entails the former) is reversed under negation – 'The dog didn't bark' entails 'The dog didn't bark loudly'. 56 plausibility states that if a person has a low belief (i.e. 0.1) that a proposition p is true, he has a high belief (i.e. 0.9) that p is false, which is consistent. In accordance with this constraint, Nilsen's degree of plausibility for certainly is as follows33: (31) [|certainly|] = p. f(p) high Modal adverbs are PPI, therefore the domain upon which possibly is defined shrinks, which in turn strengthens the degree of plausibility: (32) [|possibly|] = p. f(p) >low Thus, Nilsen accounts for the contrast in (14), repeated here: (33) It's possible that Le Pen will win even though he certainly won’t. (34) # Le Pen will possibly win even though he certainly won’t. (33) constitutes an assertion that Le Pen will win with a plausibility of equal to or greater than low and Le Pen will not win with a plausibility of equal to or greater than high. This is consistent due to the abovementioned constraint that if the plausibility of a proposition in an information state is low then the plausibility of its negation in the same information state is high. (34) on the other hand, is inconsistent, since it constitutes an assertion that Le Pen will win with a plausibility of strictly greater than low and Le Pen will not win with a plausibility of equal to or greater than high. Nilsen's account has several problems. First, his account of certainly is problematic. Modal adverbs should be stronger than modal adjectives, thus the degree of certainly in Nilsen should not be as stated in (31) repeated here as the first example, but as (36)34: (35) [|certainly|] = p. f(p) high (36) [|certainly|] = p. f(p) >high35 33 Nilsen does not provide an account for certain or for the difference between this adjective and the modal adverb certainly, as is further discussed ahead. 34 This is in fact Ernst's (2009: 529) representation of Nilsen's account of certainly. 35 Or, more plausibly, very-high. 57 However, this does not account for the contrast itself. Recall the constraint on degrees of plausibility - if the plausibility degree of a proposition is low then the plausibility degree of its negation is high and vice versa. However, using the same abovementioned numerical degrees, now it turns out that the representation of (33) is that Le Pen will win with a plausibility of equal to or greater than 0.1 and Le Pen will not win with a plausibility of strictly greater than 0.9, which is inconsistent. The second problem is that Nilsen’s account crucially depends on the claim that modal adverbs are semantically stronger than the corresponding modal adjectives. This claim is challenged in Piñón (2006, 2009): (37) a. It’s possible that the socialists won. Indeed, it’s even certain that they won. b. The socialists possibly won. Indeed, they even certainly won. c. It’s possible that the socialists won. Indeed, they even certainly won. d. #It’s possible that the socialists won. Indeed, they even possibly won. e. #It’s certain that the socialists won. Indeed, they even certainly won. The above sentences all have two parts, when the second part in each sentence is marked for greater semantic strength by the indeed…even phrase. According to Nilsen's account, all the above sentences, including the ones that contain a pair of a modal adjective and the corresponding modal adverb, should be felicitous since the latter are stronger than the former. This is clearly not the case. A similar argument is found in Ernst (2009): (38) a. She didn’t buy a book at Borders—I’d go so far as to say she didn’t buy any book (at Borders). b. (With her income,) She wouldn’t buy a car, wouldn’t buy a bicycle—I’d go so far as to say she wouldn’t even buy a little plastic scooter. c. For her vacation, it’s possible that she’d go to Albany, more possible that she’d go to Boston—and I’d go so far as to say that she’d possibly go to Paris. 58 Ernst's claim is slightly different than Piñón's. The contrast is not between felicitous and infelicitous utterances, but rather between the first two utterances which exhibit a strengthening pattern and the last utterance which does not. This lack of strengthening indicates that modal adverbs are not semantically stronger than modal adjectives. The third problem is empirical. Nilsen expects the upward entailing often and always to allow modal adverbs within their scope, and the downward entailing rarely and never to block them. He provides the following examples (Nilsen's [15] and [16]) to attest to that: (39) a. His retaliations killed or endangered innocents and often possibly had little effect in locating terrorists. b. ??His retaliations killed or endangered innocents and rarely possibly hadan effect in locating terrorists. (40) a. This is a fun, free game where you’re always possibly a click away from winning $1000! b. ??This is a fun, free game where you’re never possibly further than aclick away from winning $1000! However, performing a simple Google search turns out the following results: (41) a. "often possibly" = 292,000 b. "always possibly" = 172,000 b. "rarely possibly" = 25,900 c. "never possibly" = 803,000 While often possibly is more frequent than always possibly, the pattern is the opposite for always possibly and never possibly, which means we can't draw conclusive empirical results. The following are online examples with the DE operators outscoping the modal adverb: 59 (42) a. "The fungus forms an underground sphere, an ascocarp that breaks the surface to form a crown-shaped cup, pale on the inside with a roughly hairy dark brown outside layer. The cups are 5-7 cm across and up to 5cm tall. Like most cup fungi the flesh is brittle. It is a vernal species associated with cedar trees and very rarely possibly with yew." (Retrieved from: http://www.wbrc.org.uk/WORCRECD/32/Westwood_Brett_Bingham_John-Two_uncommo.html) b. "I can never possibly express my gratitude, so let me say simply, thank you. Sincerely, Hillary Rodham Clinton." (Retrieved from: http://www.boston.com/news/politics/politicalintelligence/2008/06/clinton_thanks_1.html) It's important to note, though, that other factors may affect the acceptability of modal adverbs under DE operators, for example if these operators are in turn also embedded under a modal operator such as can. This issue is not explored here and needs to be further examined. 2.2.4 Ernst (2009) Ernst's account agrees with the basic idea in Nilsen (2004) i.e. that modal adverbs (as a subset of speaker oriented adverbs, henceforth SpOAs) are positive polarity items. However, it differs with regards to the formal treatment of polarity. Adopting Giannakidou (1999) (non)veridicality approach, Ernst claims that modal adverbs, and PPI in general, encode the speaker's commitment to the truth of the asserted proposition as part of their lexical meaning. While the empirical goals of the paper are less to account for the differences between modal adverbs and modal adjectives, and more to account for the relative order of different SpOAs and their relation to negation, it raises points which are important to discuss. Ernst explains the embedding data, i.e. the reason why modal adverbs are bad under negation, conditionals and questions, by the following licensing conditions: 60 (43) a. A positive polarity item A is blocked in the local scope of a nonveridical operator. b. In certain cases, A may be licensed indirectly despite being in the local scope of a nonveridical operator in a sentence S, iff S gives rise to a positive implicature φ. Nonveridicality is formally defined in the following manner (Giannakidou 1999, in Ernst, 2009: 509 ) : (44) Let c = <cg(c),W(c),M, s, h,wo, f, . . .> be a context. (i) A propositional operator Op is veridical if it holds that: [Op p]c= 1→[p]=1 in some epistemic model M(x) ∈c; otherwise Op is nonveridical. (ii) A nonveridical operator Op is antiveridical iff it holds that: [Op p]c= 1→[p] = 0 in some epistemic model M(x) ∈c. In prose, an operator is veridical if, when combined with a proposition this combination yields truth in some epistemic model (i.e. the beliefs of the speaker), otherwise nonveridical. An operator is antiveridical if, when combined with a proposition this combination yields falsehood in an epistemic model. Thus, according to these definitions, negation is both nonveridical and antiveridical. Conditionals are therefore nonveridical since if a conditional is true the antecedent or consequent don’t have to be, and questions are nonveridical since questioning a proposition p does not entailp. Thus the embedding facts are explained. Another good point raised in the paper, which is related to the discussion in chapter 3, is the subjective-objective distinction. As recalled, Ernst's claim is that modal adverbs encode the speaker's commitment to the truth of the asserted proposition. An important aspect of this commitment is the property of subjectivity. Ernst follows Lyons (1977) Nuyts (2001) and Papafragou (2006) among others and claims that epistemic modal adverbs (and epistemic modality in general) can be used either subjectively or objectively. However there is a tendency: 61 "epistemic modal adverbs like possibly, as well as modals such as may/might, tend to be subjective, while the corresponding adjectives and nouns are objective ((there be) a possibility of/that and possible (that))." (Ernst, 2009: 522) Ernst doesn’t explain why this tendency exists, and speculates that it is lexical: "The lexical nature of polarity licensing also means that there can be some arbitrariness and variation, such as …the difference between the PPI possibly and its non-PPI adjectival counterpart possible. In [this] case there may be an effect of adverb-properties vs. adjective-properties, if it is generally true that adverbs tend to be more backgrounded and subjective, and thus more likely to be PPIs; this is “lexical” in that category specification is part of lexical entries". (Ernst, 2009: 527) While I'm in agreement with most of the points raised in Ernst (2009), the veridicality approach to speaker oriented adverbs is at times too strong and at other times too weak. Too strong, because of the requirement for nonveridical operators to block speaker oriented adverbs. Actually there are nonveridical operators that allow modal adverbs – for example can't and might. As recalled, a nonveridical operator is an operator that if its combination with a proposition yields truth, then the bare proposition is not necessarily true. This is the case with both can't and might, since if the dog can't - or might - be on the lawn it's not necessarily true that the dog is on the lawn. Combined with possibly: (45) a. The dog can't possibly be on the lawn. b. The dog might possibly be on the lawn. The account is too weak because it does not require veridical operators to block modal adverbs. Yet, some veridical operators, such as certain and certainly do block modal adverbs (if it is true that it is certain that – or certainly - the dog is on the lawn, then it's true that the dog is on the lawn). Combined with possibly: (46) #Certainly, the dog is possibly on the lawn. b. #It is certain that the dog is possibly on the lawn. 62 2.2.5 Piñón (2006, 2009) Piñón (2006) offers an analysis in which modal adjectives modify propositional content while modal adverbs modify assertions. As initial evidence for the assertion-modification aspect of modal adverbs, Piñón mentions their ability to replace affirmative answers to yes/no questions: (47) Did the socialists win the elections? a. Yes. b. Possibly/Probably/Certainly. c. ?It's possible/#It's probable/#It's certain. The naturalness of using modal adverbs without any other sentential material, as opposed to modal adjectives, suggests that modal adverbs have a component of assertive force that modal adjectives lack. As for the manner by which modal adverbs modify assertions, Piñón draws from illocutionary logic (cf. Searle & Vanderveken, 1985; Vanderveken, 1990, 1991) and proposes that modal adverbs modify the sincerity condition of assertions, i.e. how sincere the speaker is in terms of believing in the propositional content she asserts. Formally: (48) a. Ax (p) ‘x asserts p’ A is the speech act i.e. assertion, x is the speaker, p is the proposition b. sincerity (p, p', x,A ) ‘p is a sincerity condition of an assertion by x of p' ’ Thus, the representation of standard non-modalized assertions is: (49) assert ~>p [ Ax(p) /\ sincerity (d [believe(d,x, p) /\ d≥ high], p,x,A)] When x is the person who performs the assertion, p is the asserted proposition, and d is the degree of belief of x in p. The following is an example of an assertion and its representation: 63 (50) a. John: The socialists will win. b. s-w: the-socialists-will-win assert(s-w) ~>AJohn(s-w) /\ sincerity (d [believe(d, John, s-w) /\d≥high], s-w, John ,A) The formula states that John asserts the proposition 'the socialists will win', with a sincerity condition that has a degree of belief which is equal to or greater than high, when high stands for some numerical value and is assumed to be the default degree of belief for assertion. Modal adverbs modify assertions: (51) a. John: The socialists will possibly win. b. s-w: the-socialists-will-win possibly(s-w) ~>AJohn(s-w) /\ sincerity (d [believe(d, John ,s-w) /\d≥low], s-w, John ,A) The formula states that John weakly asserts (see ahead) the proposition 'the socialists will win', with a sincerity condition that has a degree of belief which is equal to or greater than low. Note that the modal adverb directly modify the speech act, i.e. instead of assert(p) there is possibly(p)36, so we are not dealing with a full assertion but some weaker for of one. The second thing to note is that the modal adverb is not part of the propositional content, i.e. the proposition that the speaker (weakly) asserts is the same proposition that is asserted in the case of the standard non-modalized utterance. Thus the only difference between standard assertions and modal adverb assertions lie in the degree of belief of the speaker. For standard assertions it is high, for possibility modal adverbs it is low, for necessity modal adverbs (i.e. certainly) it is very high. Piñón doesn’t state the 36 It is not clear whether Piñón considers this modification to be an assertion or a different type of speech act. 64 degree for probably but we can reasonably assume some degree between 0 and 1, i.e. above 0.5 (cf. Lassiter, 2011; Yalcin, 2010)37. The difference between modal adverbs and modal adjective is represented at the content level: (52) a. John: It's possible that the socialists will win. b. s-w: the-socialists-will-win assert(It's possible that the socialists will win) ~> AJohn( s-w) /\ sincerity (d [believe(d, John , s-w) /\d≥ high], s-w, John ,A) The formula states that John asserts the proposition 'it is possible that the socialists will win', with a sincerity condition that has a degree of belief which is equal to or greater than high. Note that the modal adjective is part of the propositional content, i.e. the proposition that the speaker (fully) asserts is different from the one asserted using the corresponding modal adverb. Piñón (2006) accounts for Nilsen's contrast in the following manner: (53) a. It's possible that the socialists will win the elections, even though they certainly won't. b. Ax(◊s-w) /\ sincerity (d [believe(d,x, ◊s-w)/\ d≥ high], ◊s-w,x,A) /\ Ax( s-w) /\ sincerity (d [believe(d,x, s-w)/\ d≥ very-high],  s-w,x,A) (54) a. #The socialists will possibly win the elections, even though they certainly won't. b. Ax(s-w) /\ sincerity (d [believe(d,x,s-w)/\ d≥ low], s-w,x,A) /\ Ax( s-w) /\ sincerity (d [believe(d,x, s-w)/\ d≥ very-high],  s-w,x,A) 37 Both Lassiter (2011) and Yalcin (2010) provide a parallel account of probable and probably, without addressing the differences between the modal adverb and the modal adjective. 65 The formula in (53) states that the speaker asserts the proposition it's possible that the socialists will win, with the degree of belief, 'greater than high', and asserts the proposition the socialists will not win, with the degree of belief 'greater than very high'. Since the first conjunct's propositional content refers to the existence of a possible world where the socialists win, and the second conjunct's content is that in the actual world the socialists do not win, there is no contradiction between the two propositional contents, and the sentence is felicitous. The formula in (54) states that the speaker asserts the proposition the socialists will win, with a degree of belief 'greater than low', and asserts the proposition the socialists will not win, with a degree of belief 'greater than very high'. This results in a clash between a proposition and its negation. The speaker cannot be considered to be sincere if she asserts both a proposition and its negation. Thus, the corresponding sentence is infelicitous. I believe Piñón's account is on the right track and provides an elegant explanation for the differences between modal adjectives and modal adverbs. However, it does little to account for their similarities. Consider (51) and (52) again, in which possibility is expressed in completely different manners. In the former, possibility is expressed by a modal operator, while in the latter it is expressed by a degree of belief. Yet, surely, it is a very similar notion in both, as evidenced by the same lexical root, shared by possible and possibly. Thus, Piñón's theory leaves a very fundamental morphological fact completely coincidental. Hence, there is a need for a theory that maintains Piñón's insight while also preserving the big advantage of the classical account, i.e. the ability to provide a unified formal explanation for modality rather than split epistemic modality into different lexical items. 2.3 Modal adverbs as negotiation chips The account proposed in this dissertation is that modal adverbs are an example of 'negotiation chips' (other examples are discussed in the following chapters) – discourse items that affect the likelihood that an utterance will be accepted by other conversational 66 participants and updated into the common ground. The manner by which modal adverbs achieve this goal is modifying the sincerity condition of assertion by expressing the speaker's degree of belief in the asserted proposition. 2.3.1 Modal adverbs are expressive An important feature of the proposed account is that modal adverbs are expressed rather than proposed propositionally, i.e. they belong to a dimension of meaning separated from the one usually dealt with by truth-conditional formal semantics. The two different modes of expression are termed descriptive vs. expressive (Kaplan, 1999; Potts, 2007) at-issue content vs. conventional implicature (CI) content (Potts, 2005) or truth-conditional vs. use-conditional (Gutzmann, 2012). Expressives have six properties, discussed in Potts (2007). This section will discuss these properties and show that modal adverbs share them and modal adjectives do not. Potts’s properties of expressives (Potts, 2007: 166): A. Independence: Expressive content contributes a dimension of meaning that is separate from the regular descriptive content. B. Nondisplaceability: Expressives predicate something of the utterance situation. C. Perspective dependence: Expressive content is evaluated from a particular perspective. In general, the perspective is the speaker’s, but there can be deviations if conditions are right. D. Descriptive ineffability: Speakers are never fully satisfied when they paraphrase expressive content using descriptive, i.e., nonexpressive, terms. E. Immediacy: Like performatives, expressives achieve their intended act simply by being uttered; they do not offer content so much as inflict it. 67 F. Repeatability: If a speaker repeatedly uses an expressive item, the effect is generally one of strengthening the emotive content, rather than one of redundancy. A. Independence The first property of expressives is the most basic one, from which all the other properties are derived. Expressive meaning resides in a dimension separate from regular truth conditional meaning (but cf. Gutzmann (2012) on mixed expressive-descriptive expressions which contribute both truth-conditional and expressive content). Thus, assertions containing expressive content such as (55) operate on two levels – the descriptive truth-conditional level in which the propositional content is the prejacent of the epithet 'that bastard', and the expressive non-truth-conditional level in which the expressive content is the negative stance expressed by the speaker towards Kresge (Potts, 2007: 168): (55) That bastard Kresge is famous. Descriptive: Kresge is famous. Expressive: Negative stance - Kresge is a {bastard/bad in the speaker’s opinion}. As a result of this separation if the hearer agrees with (55), this agreement is perceived as targeting solely the descriptive content. In other words, this type of agreement incurs no commitment for an expression of a negative stance towards Kresge by the hearer. This example brings to mind Papafragou's (2006) assent/dissent diagnostic (cf. section 2.1.2). And indeed, the next examples show that modal adverbs pattern with expressives since it is felicitous for the hearer to assent to or dissent from the speaker's assertion without assenting to or dissenting from the modal adverb, while this is not the case when modal adjectives are involved: (56) Sue: John will probably be the next chairman. Dave(a): That's true/I agree, he will be. 68 Dave(b): That's not true, he won't! Dave(c): #That's true/I agree but he won't be the next chairman. (57) Sue: It's probable that John will be the next chairman. Dave(a): #That's true/I agree, he will be. Dave(b): #That's not true, he won't! Dave(c): That's true/I agree but he won't be the next chairman. B. Nondisplaceability. Expressives can't be displaced from their immediate linguistic environment (thus, this property is connected to immediacy). This is due to the fleeting temporal nature of emotive expression as opposed to descriptive statements. Therefore, trying to linguistically displace expressives using modals, conditionals or reports of past events, is bound to fail, i.e. leads to infelicity. Potts' examples (2007: 170): (58) a. That bastard Kresge isn’t late for work. (#He’s a good guy.) b. It’s just false that that bastard Kresge is late for work. (#He’s a good guy.) c. #If that bastard Kresge arrives on time, he should be fired for being so mean. d. Maybe that bastard Kresge will be late again. (#Then again, maybe he’s not a bastard.) Applying similar examples to modal adverbs and adjectives, shows that while modal adjectives can be displaced, there are cases in which modal adverbs exhibit an even 69 stronger resistance to displaceability than the expressives in the previous example. This is in line with Potts' claim: "For some classes of lexical item, nondisplaceability is so strong that even syntactic embedding is impossible" (Potts 2007: 169). (59) a. #John won't probably be the next chairman.38 (#He will possibly be the next chairman.) b. It’s just false that John will probably be the next chairman. (?He will possibly be the next chairman.) c. It's just false that it is probable that John will be the next chairman. (It is possible). d. #If John will probably be the next chairman, I will be nice to him. e. If it's probable that John will be the next chairman, I will be nice to him. f. #Maybe John will probably be the next chairman39. (#Then again, maybe only possibly.) g. Maybe it's probable that John will be the next chairman. (Then again, maybe it's only possible.) C. Perspective dependence Under default conditions, expressives are bound to the perspective of the speaker. Thus, it is infelicitous to follow an assertion containing expressives with a remark that the speaker doesn’t share the point of view expressed: (60) a. #That bastard Kresge is famous, but I personally think that he's a good guy. b. #John will probably be the next chairman, but I personally think that he won't. 38 As stated in previous footnote, stressing the modal adverb brings it from not-at-issueness to at-issueness and turns the utterance felicitous. The intended reading is unstressed. 39 See section 2.4 on modal concord for interactions between auxiliary modals and modal adverbs. 70 c. It's probable that John will be the next chairman, but I personally think that he won't. D. Descriptive ineffability Native speakers find it very difficult to translate expressive terms into descriptive ones, for example trying to explain what 'damn' means in words is pretty hard. Gutzmann (2012) states that the theoretical status of descriptive ineffability as a property of expressive content is not sound. It has been criticized by Geurts (2007) since descriptive ineffability can be found all over the lexicon, thus not unique to expressives. That being said, perhaps it is not a coincidence that the meaning of modal adjectives is not under dispute in the literature while the meaning of modal adverbs is, i.e. the controversy regarding the nature of modal adverbs in the literature is an indication to the difficulty characterizing modal adverbs in descriptive theoretical terms. E. Immediacy Expressives have immediate effect and are thus bound to the utterance situation: (61) a. That bastard Kresge was late for work yesterday (= the speaker believes today that Kresge is a bastard) #but he’s no bastard today, because today he was on time. Potts (2007: 180) b. John was possibly late yesterday (= the speaker believes today that John was possibly late yesterday) #but today we found out that he had come on time but had forgotten to punch his card. c. It was possible that John was late yesterday (= the speaker believed yesterday that there was a possibility John was late) but today we found out that he had come on time but had forgotten to punch his card. 71 F. Repeatability Repeatability: If a speaker repeatedly uses an expressive item, the effect is generally one of strengthening the emotive content, rather than one of redundancy… Potts (2007: 182): (62) a. Damn, I left my keys in the car. b. Damn, I left my damn keys in the car. c. Damn, I left my damn keys in the damn car. It is hard to insert the modal adverb many times in the same sentence. However, stacking different lexical items with the same modal meaning is possible, and in this case the effect is indeed strengthening: this should not be the case if modals behave compositionality since stacking possibility modals results in a remote possibility40which results in a weakened effect rather than strengthened: (63) a. The dog is possibly on the lawn. b. Maybe the dog is possibly on the lawn. c. Maybe the dog might possibly be on the lawn. Modal adjectives, on the other hand, can be stacked, and the effect is compositional: (64) a. It's possible that the dog is on the lawn. b. It's possible that it's possible the dog is on the lawn. c. It's possible that it's possible that it's possible that the dog is on the lawn. The next section discusses the formal account of modal adverbs as 'negotiation chips', i.e. modifiers of the degree of strength of assertion. Treating assertion as a speech act with different levels of meaning bears similarities to Potts' (2007) account of the expressive dimension and Gutzmann's (2012) account of use-conditional meaning. It is in principle  stands for 'there is a possibility that ' while  stands for 'there is a possibility that there is a possibility that ' etc. 40 72 possible to have an expressive formal account of epistemic modals along the lines of Potts and Gutzmann. I have chosen a different path which seems better suitable to show the similarities as well as the differences between expressive and descriptive modality in terms of different scopes for the same lexical items. It is unclear how to account for modal adjectives and descriptive epistemic modals in the same formal manner as modal adverbs and expressive epistemic modals via the expressive theories of Potts and Gutzmann. 2.3.2 Formal account As discussed in chapter 1, this dissertation proposes a theory of conversation by which, in addition to the common ground (represented in probabilistic terms, i.e. a set of information states), the conversational context registers assertions which were previously performed and are under negotiation. Each of these assertions is an expression of an information state of some conversational participant with regards to a proposition, and it is represented by the following assertion operator, repeated here: (65) Ax<S,C> The speaker x asserts propositional content C with a degree of strength S. The assertion operator itself is represented by the following probability function: (66) Ax P() = v P is the probability function which yields some probability value v when applied to the propositional content. This value is defined on the speaker's information state i.e. the probability space of the speaker x and stands for the degree of belief of x in , which is the degree of strength for the sincerity condition of assertion. This degree of strength serves as the degree of strength S of the assertion. While an assertion is under negotiation, i.e. from the moment the assertion is performed and until it is accepted or rejected by conversational participants, it resides in the 73 Negotiation Zone (NZ). The NZ is thus a set of assertion operators pertaining to various propositions. The manner by which conversational participants decide whether to accept or reject assertions is through an activation of a mixture model, repeated here: (67) P(  ) =  wiPi(  ) n i=1 If the value of the mixture model surpasses some contextual threshold of acceptance, by assumption high, then the assertion is accepted and  is updated into the common ground, i.e. the probability value assigned to  becomes 1, and all information states in which the probability of  is less than 1 are removed from the common ground. The difference between the standard Stalnakerian context update and the probabilistic context update proposed here, is as follows – figure 2.1 depicts a standard context update in which proposition  is added to the common ground thereby intersecting with the initial context set: Figure 2.1: Stalnakerian context update The figure on the left depicts the initial context set as a set of worlds. Since nothing is known about , some of the worlds in the context set  worlds and others are  worlds. The figure on the right depicts the new state of discourse once  is accepted into the 74 common ground – the proposition , which is the set of  worlds, is intersected with the initial context set to yield the new one. The initial context set in this dissertation is richer - it is composed of probability spaces, hence a more accurate representation of the initial probabilistic context set is: Figure 2.2: Probabilistic context set As can be seen, the probabilistic context set contains probability spaces rather than possible worlds, when each probability space is an information state IS. Thus, when a speaker asserts a standard non-modified assertion the representation is: (68) The dog is on the lawn Ax P (on-the-lawn(the-dog)) ≥ high In prose, the speaker asserts the propositional content 'the dog is on the lawn' with a degree of strength which is equal to or greater than high. If this assertion is accepted and updated, this assertion's Context Update Effect (henceforth CUE) will be removing all information states except from IS1 and IS3 from the context set, since those are the information states in which the propositional content has a probability of 1. Now for an assertion containing the modal adjective possible: 75 (69) It's possible that the dog is on the lawn. Ax P (P(on-the-lawn(the-dog)) > 0) ≥ high The speaker asserts the propositional content 'it's possible that the dog is on the lawn', represented in probabilistic terms, i.e. 'the probability that the dog is on the lawn is greater than 0', with a degree of strength which is equal to or greater than high. Note that the degree of strength for a modal adjective-modified assertion is the same as the degree of strength for non-modified assertion, thus the chances of this assertion to be accepted by the hearer(s), everything else being equal, are the same as the chances of any nonmodified assertion. Also note that the type of modal adjective does not affect the degree of strength but only the propositional content. If this assertion is accepted and updated, this assertion's CUE will be removing IS4 from the context set, since this is the only information state in which the propositional content has a probability of 0. The following are representations of the other modal adjectives-modified utterances: (70) It's probable that the dog is on the lawn. Ax P (P(on-the-lawn(the-dog)) > 0.5) ≥ high (71) It's certain that the dog is on the lawn. Ax P (P(on-the-lawn(the-dog)) = 1) ≥ high As discussed previously, I assume following Yalcin (2010) and Lassiter (2011) that the degree assigned to probable is greater than 0.5. I also assume that the degree assigned to certain is the same as the degree assigned to necessity modals, i.e. 1 which corresponds to full certainty (but see chapter 1 section 1.3.8 on the corrigibility of belief). The CUE of (70) is to remove all information states except from IS1, IS3, and IS5 from the context set, and the CUE of (71) is to remove all information states except from IS1, and IS3 from the context set. Note that the CUE of (71) and (68) is the same, but the conversational impact is different. In (68) the speaker proposes to make the propositional 76 content common ground, and in (71) the speaker claims that this propositional content is already common ground. Also note the entailment patterns – the stronger claims (71) and (68) illocutionary entail the weaker ones (70) and (69) since the CUE of the former is a subset of the CUE of the latter. The representations of modal adverbs-modified assertions are: (72) The dog is possibly on the lawn. Ax P (on-the-lawn(the-dog)) > 0 (73) The dog is probably on the lawn. Ax P (on-the-lawn(the-dog)) > 0.5 (74) The dog is certainly on the lawn. Ax P (on-the-lawn(the-dog)) = 1 The formulas state that the speaker asserts the propositional content 'the dog is on the lawn' with the degrees of strength equal to or greater than 0 (for possibly), to 0.5 (for probably) and equal to 1 (for certainly). Note the difference between modal adjectives and modal adverbs – the former modify the propositional content while the latter modify the degree of assertion. Note the similarity between modal adjectives and modal adverbs – both are represented by the same degrees of probability, since the lexical root of both is the same. The difference is a matter of scope – modal adverbs modify the whole speech act and therefore scope over the propositional content, and modal adjectives modify the propositional content and therefore have narrow scope. There are differences in terms of conversational effects as well. If accepted, the CUE of all of the modal adverbs-modified assertions is the same as the CUE of non-modalized assertions, i.e. removing all information states except from IS1, and IS3 from the context set. However, asserting a modal adverb modified utternce without the coresponding degree of belief is insincere. Thus, only IS1, and IS3 can sincerely assert (74), only IS1, IS3 and IS5 can sincerely assert (73) and only IS1, IS3, IS5 and IS2 can sincerely assert (72). 77 2.3.3 Explaining the data Embeddings: Modal adverbs are speech-act modifiers, and as such operate on a realm beyond the propositional one dealt with in standard semantic theories. In Wittgensteinian terms, they are not part of the utterance radical. Thus, following the classical literature on speech acts, it is not surprising that these instances of speech act modification do not interact with truth-conditional propositional material. As Lewis (1970) put it: "…the entire apparatus of referential semantics …pertains to sentence radicals and constituents thereof. The semantics of mood is something entirely different." Or recently, in Potts (2005: 42): "CIs [conventional implicatures] never appear in the scope of other operators. This imparts the sense that they always have widest scope. But, as with the indexicals of Kaplan (1989), it seems more accurate to think of them as scopeless." The account proposed in this dissertation as well as the data show that in some cases an interaction between modal adverbs and propositional operators such as negation is not possible, but when this interaction is possible modal adverbs have wide scope. In terms of the formal account, modal adverbs have wide scope over the propositional content and therefore can't be embedded under propositional modifiers. Note that Krifka (2001) discusses several cases in which speech act modifiers are embeddable. Thus, the claim in this dissertation is not that modal adverbs are entirely non-embeddable, but that they are as hard to embed as other speech act modifiers. Consequents of conditionals: While modal adverbs are hard to embed under antecedents of conditionals, they are ok in consequents. Repeating (7): 78 possibly probably (75) If the socialists win the elections, the rich will certainly worry about a luxury tax. In order to account for that, I adopt Cohen's (2010) theory of conditionals as illocutionary operators, in which conditionals modify the strength of the sincerity condition of the speech act. I.e. when a speaker asserts the conditional > , she asserts with a sincerity condition that the belief in the probability P( | ) is high. Thus, (75) can be represented in terms of the assertion operator, but since the first part of the assertion operator i.e. the asserted content is different from the second part of the assertion operator i.e. the degree of strength, we need to drop the shorthand used so far and return to the original formulation of the assertion operator: (76) Ax<S,C> First, a representation of a standard non-modified conditional: (77) If the socialists win the elections the rich will worry about a luxury tax. Ax< P (worry(the-rich) | win(the-socialists)) ≥ high, worry(the-rich) > Now, a representation of the conditional consequent modified by a modal adverb: (78) If the socialists win the elections the rich will possibly worry about a luxury tax. Ax< P (worry(the-rich) | win(the-socialists)) > 0 , worry(the-rich) > The formula states that the speaker asserts the propositional content 'the rich will worry about a luxury tax' based on the degree of belief that the probability that the rich will worry about the luxury tax given that the socialists win is low, specifically equal to or greater than 0. 79 According to this account, a modal adverb can't modify the antecedent of a conditional since the antecedent, unlike the consequent, is not asserted and therefore can't be modified by an illocutionary operator. Agreements and disagreements: Agreements and disagreements concern the asserted propositional content and not the expressive degree of strength of this assertion. Therefore, agreements and disagreements target the prejacent of modal adverb assertions while they target the entire modalized utterance when modal adjectives are concerned. Scope over tense In order to explain why modal adverbs scope above tense, there is a need for a syntactic account in which the position of speech act modifiers is shown to be higher than TP. See section 3.6 in the next chapter. Nilsen's contrast: Repeating (14) - (15): (79) It's possible that Le Pen will win even though he certainly won’t. Ax P (P(win(Le-Pen))> 0) ≥ high /\ Ax P (win(Le-Pen)) = 1 (80) # Le Pen will possibly win even though he certainly won’t. Ax P(win(Le-Pen))> 0 /\ Ax P (win(Le-Pen)) = 1 The difference between the two assertions stems from the interplay between modification of the propositional content and modification of the speech act. In (79) the speaker asserts the propositional content 'it's possible that Le Pen will win' with the default degree of 80 assertion high, and asserts the propositional content 'Le Pen will not win' with the degree of assertion 1 which corresponds to certainty. In (80) the speaker asserts the propositional content 'Le Pen will win' with the default degree of assertion high, and asserts the propositional content 'Le Pen will not win' with the degree of assertion 1 which corresponds to certainty. The intuition to explain is that (79) is not perfect but still acceptable while (80) is downright infelicitous. The reason for the infelicity of the latter is that it is unassertable, i.e. the speaker can't sincerely assert a proposition with a degree of belief of greater than 0 and its negation with a degree of belief of 1. It follows that a speaker can assert contradictory propositional contents if the degrees of belief do not contradict themselves, as in the following: (81) Le Pen will possibly win even though he probably won’t. This prediction is borne out since (81) is better than (80). As for (79), this is an assertable utterance since the speaker may have a full personal belief that a proposition is false, while acknowledging that other conversational participants, i.e. other information states within the context set may believe to at least a low degree of certainty that this proposition is false. Note, however, that this is consistent only on the level of assertability, since the CUE of both propositional contents is contradictory and hence both cannot be updated at the same time. This is why even though (79) is better than (80), it is still not a completely good utterance. 2.4 Modal adverbs and modal concord This short section discusses modal concord and proposes a preliminary research direction to be explored in the future, based on the theory presented in this chapter. Modal concord is a phenomenon in which a combination of a modal adverb with an auxiliary verb yields a seemingly non-compositional result: 81 (82) a. This gazebo may have been built by Sir Christopher Wren. b.Possibly this gazebo was built by Sir Christopher Wren. c. Possibly this gazebo may have been built by Sir Christopher Wren. (Halliday, 1970) As noted in Halliday (1970) and Lyons (1977), it seems that all the above sentences convey the same meaning. This is puzzling, of course, if all epistemic modals are to act compositionally, since in that case we would expect to have double modality, as in the following interaction between two modal adjectives or even between a modal auxiliary and a modal adjective: (83) It's possible that it's possible that Sir Christopher Wren built this gazebo. b. It might be possible that Sir Christopher Wren built this gazebo. The question that arises is, why do modal adverbs give rise to such seemingly noncompositional readings when combined with modal auxiliaries while modal adjectives do not? Since this section is located at the end of this chapter, the answer to this puzzle may be quite obvious at this point. But before giving away the answer, a short theoretical survey is in order. Geurts & Huitink (2006) state the following conditions for modal concord to arise, which need to be accounted for: A. Matching flavor: (84) a. John might possibly work on Sunday. (concord) b. John might have to work on Sunday. (no concord) B. Matching force: (85) a. There may possibly have been weapons of mass destruction in Iraq. (concord) 82 b. There may certainly have been weapons of mass destruction in Iraq. (no concord) Geurts and Huitink (2006) explain modal concord by claiming that modal adverbs have a functional meaning which is to perform a type check on modal auxiliaries. Thus, if modal adverbs combine with modal auxiliaries and the two modals are of the same sort, the auxiliary is deleted. Zeijlstra (2007) explains modal concord by claiming that modal auxiliaries are expletives that carry uninterpretable features, and modal adverbs are probing operators that check these features under syntactic agreement. Huitink (2012) explains modal concord by claiming that modal adverbs impose constraints on the ordering source of the modal auxiliaries they interact with, thus assuring the same flavor, plus there is an agreement relation between the uninterpretable features of adverbs and interpretable features of auxiliaries that accounts for force .Anand & Brasoveanu (2010) also present a theory of modal concord; however this theory deals only with deontic/circumstantial modal concord and therefore will not be discussed here. There are some problems that these theories can't account for. For example, as stated in Anand & Brasoveanu (2010) and Zeijlstra (2007), modal adverbs affect a type of strengthening of the auxiliaries’ force: (86) a. John must be at home. b. John must certainly be at home. Modal adverbs also affect a weakening of the auxiliary force: (87) a. John might be at home. b. John might possibly be at home. This is a problem for the accounts of Geurts & Huitink (2006) and Huitink (2012). Zeijlstra (2007) can account for this, since the strengthening adverb is the semantically active modal element in Zeijlstra's theory. However, a problem that Zeijlstra's theory can't account for is that modal auxiliaries can be compositional. For example, a scenario 83 in which two scientists argue whether it is possible that a asteroid will crash on earth. A third scientist says: (88) Perhaps the asteroid might crash and perhaps it can’t. (Ariel Cohen, p.c.) By Zeijlstra (2007), the auxiliary should have its feature checked by the adverb and thus should not be interpretable compositionally. The theory proposed in this chapter can account for modal concord and the abovementioned problems in the following manner – modal adverbs expressively modify the speech act while modal auxiliaries are ambiguous between descriptive and expressive readings (cf. next chapter) and therefore may modify either the proposition or the speech act. In the presence of modal adverbs, auxiliaries can’t modify the speech act and are thus descriptive, and the resulting meaning is a descriptive EM with either a weakened (by possibility modal adverbs) or strengthened (by necessity modal adverb) assertion strength. In order for the differences between expressive and descriptive EM and their interaction to be clear, the following utterances are presented together with formal representations. These include a non-modalized utterance, an utterance containing a modal auxiliary, an utterance containing a modal adverb, and finally the utterance presented in the beginning of the section and repeated here, with the modal concord: (89) a. This gazebo was built by Sir Christopher Wren. b. AxP (build(gazebo,SCW)) ≥ high Speaker x asserts the propositional content ‘this gazebo was built by SCW’ with the default degree of strength which is equal to or greater than high. (90) a. This gazebo may have been built by Sir Christopher Wren. b. AxP( P(build(gazebo,SCW)) > 0)≥ high Speaker x asserts the content: ‘it’s possible this gazebo was built by SCW’ with the default degree of strength ≥high. 84 (91) a. This gazebo was possibly built by Sir Christopher Wren. b. AxP (build(gazebo,SCW)) > 0 Speaker x asserts: ‘this gazebo was built by SCW’ with a lower than default degree of strength >0. (92) a. Possibly this gazebo may have been built by Sir Christopher Wren. b. AxP( P(build(gazebo,SCW)) > 0)> 0 Speaker x asserts the content: ‘it’s possible this gazebo was built by SCW’ with the low degree of strength >0. Thus, modal concord is accounted for with the surprising explanation that (92) is not really a case of modal concord, since both the modal auxiliary and the modal adverb have an effect, and this effect is compositional. The reason that it seems like a modal concord is because the expressive effect is hard to intuitively grasp (cf. Potts' (2007) property of descriptive ineffability for expressives, discussed in section 2.3.1), especially when combined with the descriptive. As for the problems posed to previous theories, strengthening/weakening is accounted for by the modal adverb property of modifying the strength of the speech act. Thus, possibility EM weaken the speech act and necessity EM strengthen it. The compositional readings of modal auxiliaries are accounted for by modal auxiliaries' property of being part of the propositional content when interacting with modal adverbs, thus compositional. 85 Chapter 3 Epistemic modal auxiliaries The previous chapter dealt with the differences between epistemic modal adverbs and modal adjectives, has shown how these differences are linguistically manifested, and explained these differences via an ambiguity between expressive and descriptive EM. This chapter discusses epistemic modal auxiliaries and shows that the account of modal adverbs and modal adjectives applies to all types of epistemic modals. While Chapter 2 has focused on a semantic account of modality, this chapter will incorporate syntactic aspects as well. 3.1 The root-epistemic distinction The literature on modal auxiliaries e.g. can/could, may/might, must, etc. distinguishes between two basic types, epistemic modals and root modals (cf. Jackendoff, 1972; Perlmutter, 1971; Ross, 1969). The former consist of the type of modality discussed in the previous chapter, the latter contains deontic, circumstantial, ability, etc. Basic modal auxiliaries such as may, may convey different types of modality but there are ways to disambiguate, by making the implicit conversational backgrounds explicit, using an 'according to' type of phrase (cf. Kratzer, 1981, 1991). The first of the following sentences is ambiguous, the next two contain root modals and the last is epistemic: (1) The dog may be on the lawn. (ambiguous) (2) According to the house rules, the dog may be on the lawn. (deontic) (3) In light of the dog's health, he may run on the lawn for up to 5 minutes. (ability) (4) According to what we know the dog may be on the lawn. (epistemic) An early account of the root-epistemic distinction, Ross (1969) is that epistemic modals are one place predicates while root modals are two place relation predicates. The former 86 has the asserted proposition as an argument and the latter depicts the relation between the subject and the rest of the proposition. Thus, in (4), the argument is 'the dog is on the lawn' while in (2) the arguments are the dog and on the lawn. A different explanation (cf. Hofmann, 1966) is that the syntactic composition of root modals is done via a control structure while epistemic modals have a raising structure. Brennan's(1993) account combines the two abovementioned theories. In this account root modals are merged at the VP level, are two place predicates, and have a control structure while epistemic modals are IP level, one place predicates, with a raising structure. However, the control-raising distinction for epistemic-root modality is problematic. As is argued in Bhatt (1998) and Wurmbrand (1999) many deontic modals have a raising structure and therefore the claim that epistemic modals are raising verbs while root modals are control, cannot hold. A recent and promising account of the root-epistemic distinction is discussed in Hacquard (2006). This account combines the classic Kratzerian theory with Davidsonian (1967) semantics. In this account, rather than being relative to possible worlds, modals are relative to events, which are divided into three types – VP events, speech events and attitude events. Each event is part of the object language and is represented in the syntax as an event pronoun e that is bound locally by aspect or the speech event. There are two possible syntactic locations (determined by the type of the modal) for modals to appear in – right above VP or right above T. The former location is bound by the aspect that quantifies over the event of circumstances of context, i.e. time, location, participants. The latter location is either bound by speech event (for modals in the main clause) or the aspect that quantifies over an embedded attitude event, i.e. propositional attitude holders (for modals in a complement clause). Thus, different syntactic locations have different sets of accessible possible worlds, i.e. different modal bases, because of the different events that appear in these locations. The different positions for root and epistemic modals in Hacquard (2006) are supported by scope data. For example, epistemic modals tend to have wide scope over quantifiers while root modals have narrow scope: (5) a. Every radio might get Chicago stations and no radio might get Chicago stations. 87 b. #Every radio can get Chicago stations and no radio can get Chicago stations. (Hacquard, 2006: 118) The might modified sentence has an epistemic interpretation and the can modified sentence has circumstantial (root modality)41. The infelicity of the second utterance stems from a narrow scope reading of the modal, i.e. 'Every radio is such that it is possible for it to get Chicago stations and there is no radio for which it is possible to get Chicago stations', which is a contradiction. The first sentence, on the other hand, has a wide scope modal, i.e. 'it is possible that every radio gets Chicago stations and it is possible that no radio gets Chicago stations', which is not contradictory. More evidence regarding modal scopes comes from interactions between root and epistemic modals themselves. Hacquard (2011) discusses such cases, for example: (6) John may have to watch TV. The modals may and have to are both ambiguous, both have epistemic and deontic (root) readings. However, when stacked together as in the above sentence, the only available reading is the one in which epistemic modality has wide scope and deontic narrow, i.e. "in light of what is 'known' it is possible that John has an obligation to watch TV". The other reading, in which the deontic modal has wide scope, should be available as well: "it is permissible that in light of what it 'known' it is necessary that John watches TV". Of course, this reading doesn’t exist, which again supports the high syntactic position of epistemic modals as opposed to the low position of root modals. Hacquard (2011) also discusses a counterexample from von Fintel & Iatridou (2004): (7) “An insurance company will only pay for an expensive test if there is a possibility that the patient may have Alzheimer's”. Such a state of affairs can be reported as follows, with a deontic modal taking scope over an epistemic adjective: 'For the test costs to be reimbursed, it has to (DEONTIC) be possible (EPISTEMIC) that the patient has Alzheimer’s'." However, note that the epistemic modal used in this example is a modal adjective. As discussed in chapter 2, modal adjectives are specialized for descriptive epistemic meaning which is syntactically low, thus has narrow scope when interacting with various 41 Cf. Kratzer's (1991) famous hydrangeas example. 88 operators42. And indeed Hacuard (2011) adds that epistemic modal auxiliaries do not seem to be able to be embedded in such a manner. A third type of scope data comes from interaction with negation. This can easily be observed in English: (8) a. According to the house rules, the dog may not be on the lawn. b. According to what I know, the dog may not be on the lawn. The same modal auxiliary may, has a deontic interpretation in the first sentence and an epistemic interpretation in the second sentence, explicitly provided by the 'according to' phrase. As can be seen, the deontic modal scopes below negation, i.e. the reading is 'it is not permissible for the dog to be on the lawn'. The epistemic modal, on the other hand, scopes above negation, i.e. the reading is 'it is epistemically possible that the dog is not on the lawn'. Hacquard (2011) also discusses cross-linguistic examples such as the Malay examples in Drubig (2001): (9) a. Dia mesti tidak belajar. (epistemic) he must not study b. Dia tidak mesti belajar. (deontic) he not must study This examples shows how the modal Mesti (must), which can manifest either epistemic or deontic interpretation, gets an epistemic interpretation when it is structurally above negation and a deontic interpretation when below, again supporting the modal scopes proposed in Hacquard. The last item of scope data discussed in Hacquard (2011) is tense. For example, in the next sentence have to gets a teleological (root) interpretation: (10) Mary had to take the train to go to Paris. The interpretation of this sentence is: 'in the past Mary's objectives were such that she had a necessity to go to Paris', i.e. the modal is embedded under tense. The next sentence 42 See section 3.6 for a discussion of the syntactic location of descriptive epistemic modality. The account proposed in this dissertation does not deal with deontic modality and therefore cannot state the exact syntactic location of those deontic modals that outscope descriptive epistemic modals. This is left for future research. 89 containing an epistemic modal, on the other hand, cannot have such a narrow scope modal reading: (11) Mary had to be the murderer. Hacquard states that the (epistemic) reading of this sentence is: 'in light of what is known now, there is a necessity that in the past Mary was the murderer', i.e. the epistemic modal has wide scope and it is tense that is embedded. The same holds for future tense: (12) Tomorrow, Marikos may be dead. (Groenendijk & Stokhof, 1975) There is an epistemic possibility today that Marikos will be dead tomorrow. Compare with: (13) Tomorrow, Marikos may be let out of the house. The deontic reading is that tomorrow Marikos will be allowed out of the house, hence again deontic reading is embedded under tense. The account proposed in this dissertation adopts Hacquard's account with regards to the syntactic positions43 of root and epistemic modals, i.e. root modals are located right above VP while epistemic modals are located right above T, in the following manner (Hacquard, 2010: 96): (14) I propose the following addition to Hacquard's account – epistemic modals have two uses, descriptive and expressive. The descriptive use is as depicted in Hacquard and in line with the classical Kratzerian theory of modality, i.e. descriptive epistemic modals are 43 And remains agnostic with regards to the event semantics aspect of Hacquard's work. 90 truth-conditional propositional modifiers. In addition to the descriptive use, there is the expressive, which modifies the entire speech act and not just the proposition. The expressive epistemic modal is located at the left periphery, specifically ForceP (cf. section 3.6). In this manner, the account proposed in this dissertation preserves the traditional hierarchy in which root modals are the more basic type of modality and syntactically located lower than epistemic modality. The account takes one additional step and adds a further distinction between different types of epistemic modals - descriptive epistemic modals are the more basic type of epistemic modality and located lower than expressive epistemic modals. Hence resulting in the following hierarchy: (15) expressive EM >> descriptive EM >> root modals This hierarchy is both syntactic and semantic. Low root modals, VP level, address the ways the objective world can is, e.g. in light of laws of nature and man; intermediate descriptive EM, TP level, address the discourse mutual state of belief, i.e. in light of the common ground, and high expressive EM, left periphery level, address the subjective belief of the speech act performer, i.e. in light of the speaker's belief. This hierarchy cashes in Hacquard's (2006) observation, discussed previously in this chapter, regarding three types of events - VP events, attitude events and speech events. In a similar vein, the proposed theory is compatible with viewing root modals as relating to VP events, descriptive EM as relating to the propositional attitudes of discourse participants, i.e. the discourse event and expressive EM relate to the speech act event. The account proposed here expresses this observation by expanding the syntactic and semantic framework of the classical theory of modality. The following sections discuss epistemic ambiguity both theoretically and empirically. Section 3.2.1 discusses how theories of epistemic modality are divided between truthconditional and non-truth-conditional claims. Section 3.2.2 discusses how the data is also divided, when some of it is best accounted by truth-conditional theories and some of it best accounted by non-truth-conditional theories. The conclusion is that there is no single account that can explain the whole range of data but there is a unified way to capture 91 epistemic ambiguity in a manner that preserves Kratzer's insight regarding the uniform nature of modals via different modal flavors and Hacquard's insight regarding the uniform nature of modals via different syntactic scopes. 3.2 Epistemic ambiguity 3.2.1 Theoretical ambiguity As discussed in Papafragou (2006), Portner (2009) and Hacquard (2011) the literature on epistemic modality is divides between truth-conditional and non-truth-conditional accounts. Truth conditional theories of epistemic modality are based on the classical Kratzerian view, discussed in chapters 2. According to this view, an epistemically modalized utterance contains the modal as part of the proposition. Non-truth-conditional theories, e.g. of the functional/descriptive tradition (cf. Halliday, 1970; Palmer, 1986) account for epistemic modals in terms of the speaker's assessment, evaluation or commitment to the asserted proposition precluding the modal, i.e. the prejacent. Since epistemic modals provide a measure of such assessment, they are not part of the asserted propositional content, but act as a side remark on what has been asserted. Other approaches to epistemic modality also walk the truth conditional divide. For example, dynamic theories of meaning, discussed in chapter 1, offer a non-truthconditional account of epistemic modality. In one of these theories, Veltman (1996) explains might utterances as consistency checks on information states. By this account, when a speaker asserts an utterance of the form 'might ', the hearers examine their information states s to see whether an update with  does not result in an empty set, i.e. if  does not contradict the set of beliefs composing s. Importantly, if indeed no contradiction occurs, no update takes place, i.e. s remains the same, which means that the epistemic modal has no context update effect. As formalized in Portner (2009): (16) Given a model M = <W, V> and an information state s ⊆W, the information change potential []M of  is defined as follows: 92 If is of the form, for any information state s , s []M= a. s , if s []M∅44; b. ∅ otherwise. In prose, an update of information state s with might  results in s as long as an update with the prejacent  does not result in the empty set. Otherwise, no update is possible and the utterance should be rejected. Other non-truth-conditional EM theories include Drubig (2001) and Westmoreland (1998) who claim that epistemic modals are evidentials, i.e. mark the source of evidence a speaker has for making an assertion, which is non-truth-conditional45. Probabilistic theories of EM also walk the truth-conditional divide. For example, Lassiter (2011) differs from contextualism in defining epistemic modals in terms of probability rather than possible worlds, while providing a truth-conditional account of epistemic modality. Yalcin's (2012) probabilistic account, on the other hand, is expressive and can be associated with the non-truth-conditional camp. The common ground in Yalcin's account is not updated directly. Instead, when a speaker asserts a proposition, she expresses her credal state with regards to this proposition and proposes that other conversational participants adopt the same state towards it. If this happens, the speaker presuppositions change, thereby updating the common ground. Cohen (2013), further discussed in section 3.3.6, proposes a probabilistic truth-conditional account of EM along the lines of the modified Kratzerian account of von Fintel & Gillies (2011) in which EM involve not only one but a cloud of propositions which are put into play at the same time. It appears that the truth-conditional debate concerning epistemic modality is a heated one. But one does not necessarily have to choose sides in this debate, as there is a middle way – the ambiguity account proposed in Lyons (1977). In Veltman's terms, instead of the empty set  the representation is 1, the absurd state. This matter is under debate. For example, Matthewson, Rullmann, & Davis, (2008) and McCready & Ogata (2007) claim that evidential do contribute to truth conditions, and Faller (2002) claims that evidentials are speech act force-modifiers while epistemic modals are truth-conditional. The connection between epistemic modality and evidentiality is fascinating, but it is not discussed in this dissertation. 44 45 93 Lyons presents an account of epistemic modality which distinguishes between subjective and objective epistemic modals. The former are force modifiers similarly to the descriptive/functional tradition, and the latter are truth conditional. For example, the next utterance has two readings: (17) Alfred may be unmarried. This is either a subjective epistemically modalized assertion in which the speaker is expressing a low commitment to the proposition 'Alfred is unmarried' or an objective epistemically modified assertion in which the speaker claims (which a high level of commitment) that there is a possibility that Alfred is unmarried. While the first reading is intuitive, the second needs a little explaining. Lyons offers a scenario in which it is known that Alfred is a member of a community of 90 people and that this community contains about 30 married individuals. All this being known, the claim that there is such a possibility (objectively 2/3 chance) can be made with high conviction. Lyons (1977) theory is formalized in Portner (2009:123): (18) a. Alfred is unmarried. b. I-say-so + it-is-so + [[ Alfred is unmarried ]] (19) a. Alfred may be unmarried. b. Subjective: Poss + it-is-so + [[ Alfred is unmarried ]] c. Objective: I-say-so + Poss + [[ Alfred is unmarried ]] Lyons divides speech acts into three parts, the tropic, the neustic, and the phrastic. The tropic, corresponding to the I-say-so section of the formula, represents the performance of the speech-act. The neustic, corresponding to the it-is-so section of the formula, represents the speaker’s level of commitment to the phrastic, which is the asserted propositional content. Subjective epistemic modals modify the tropic while objective epistemic modals modify the neustic. Lyons' account is further discussed in section 3.3.1. 94 3.2.2 Empirical ambiguity The data in this section runs parallel to the data provided in chapter 2, with the aim to show that modal adverbs and modal adjectives are not an idiosyncratic phenomenon, but a part of the larger picture pertaining to epistemic modality. This picture has the potential to be expanded further in order to deal with root modals as well46 and the connection between epistemic modality and evidentiality47. Among the diagnostics in the following subsections, the ones dealing with embeddings, agreements/disagreements and scope (especially negation) are standardly taken to be indicators for non truth conditional content (cf. Papafragou, 2006) and illocutionary modification (cf. Faller, 2002) 3.2.2.1 Embedding Papafragou (2006) discusses the following examples in which expressive (subjective in Papafragou's terms) epistemic modals are hard to embed while descriptive (objective in Papafragou) epistemic modals embed easily48: (20) Antecedents of conditionals: a. ?If Max might/must be lonely, his wife will be worried. b. If it might rain tomorrow people should bring umbrellas. (21) Factives: a. ?It's surprising that Superman might/must be jealous of Lois. b. It’s surprising that it might rain tomorrow. 46 C.f. Tobin (1996) about participant-internal must vs. participant-external have to, Palmer (1986) about subjective vs. objective deontic modals and Kolodny & MacFarlane (2007) about subjective vs. objective ought. 47 C.f. Faller (2002), Matthewson et al. (2008). 48 The choice between expressive and descriptive uses of epistemic modals is of course context-dependent. Papafragou's expressive examples therefore have emotive contexts while the descriptive examples contain neutral contexts such as weather reports. The given contexts are still within the confines of the sentence, though, and therefore limited. The more elaborate a context is the more clear the readings become. See Cohen, Wolf & Taube (unpublished), discussed ahead. 95 (22) Verbs of saying: a. ?Spiderman told me that Superman must be jealous of Lois. b. The weather forecast told viewers that it may rain tomorrow. In addition to the abovementioned sentences, embedding data has been investigated experimentally. Cohen, Wolf and Taube (unpublished) performed an experiment, in which 31 native English speakers filled an online questionnaire containing 10 scenarios. 5 of these scenarios contained a backstory that gave rise to an expressive (subjective in Cohen, Wolf and Taube's terms) context and the other 5 contained a backstory that gave rise to a descriptive (objective in Cohen, Wolf and Taube's terms) context. For each scenario, participants judged the acceptability of 3 sentences containing an epistemic modal embedded under either a conditional, a factive, or a verb of saying or belief. The answers were either 1 for acceptable, or 0 for unacceptable. The following is a Sample of an expressive scenario: (23) Greg and Claire are on a trip to France. Claire heard about a great museum from a friend, and wants to visit it. Greg, wanting to please Claire, asks one of his friends in France, Jacques, about the museum. Jacques vaguely remembers that the museum was closed in late 2007. (24) Conditional embedding: Jacques: If the museum might have closed in late 2007, you shouldn't go there. (25) Factive embedding: Jacques: It is sad that the museum might have closed in late 2007. (26) Verbs of saying/belief: Greg: Jacques said/believes that the museum might have closed in late 2007. 96 The following is a sample of a descriptive scenario: (27) Sophia is an Olympic runner. She also suffers from asthma, and uses an inhaler to help her breathe if she has an attack. One day her mother, Betty, calls her up with some startling news she saw in the paper, that experts believe that a lot of athletes with asthma have been taking the wrong medication. (28) Conditional embedding: Sophia: If many athletes might be taking the wrong medication, I should consult with a doctor. (29) Factive embedding: Betty: It is appalling that many athletes might be taking the wrong medication. (30) Verbs of saying/belief: Sophia (to her friend Henry): My mother said/thinks that many athletes might be taking the wrong medication. The results of the experiment are: (31) Antecedents of conditionals: a. Expressive: 0.26 b. Descriptive: 0.59 c. Significance: p<0:0001 (32) Complements of factives: a. Expressive: 0.52 b. Descriptive: 0.7 c. Significance: p=0.03. (33) Verbs of saying/belief: 97 a. Expressive: 0.81 b. Descriptive: 0.88 c. Not significant The results of the experiment show that descriptive EM are significantly better embedded under conditionals and factives. However, embeddings of both expressive and descriptive epistemic modals under verbs of saying/belief are more or less equally acceptable. It is interesting that expressive epistemic modals turn out to be embeddable under verbs of saying or belief in light of their being speech-act modifiers. Propositional attitudes are not discussed in this dissertation, thus a full account of this embbedability is not discussed here. But two points can be stated in this regard. The first, following Cohen (2013) is that the reason expressive (or subjective in Cohen's work) epistemic modals are felicitous under verbs of saying or belief is that these verbs provide an assertive context. Verbs of saying are reports of assertions, thereby turning the prejacent proposition into an assertion and verbs of belief can be defined as dispositions to assert (Braithwaite, 1932). The second point is that many types of speech act or illocutionary modifiers (cf. Vanderveken, 1990) including modal adverbs can also be embedded under verbs of saying/belief: (34) a. Sophia said/believes that Mary is presumably/certainly a lawyer. The problem of embedding under verbs of saying/belief is therefore one that any account of speech-act modifiers faces. This in itself does not rule out expressive epistemic modals from being such modifiers. Another source of data pertaining to embeddings of epistemic modals is corpora. Hacquard & Wellwood (2012) conducted a corpus study in which might, can, and must are examined in antecedents of conditionals, questions, and complements of attitude predicates. The assumption is that might is an epistemic modal, can is a root modal and must is ambiguous. While Hacquard & Wellwood do not look for expressive-descriptive (or subjective-objective) difference in interpretation for might, their findings are still illuminating: 98 "Relative to their own distributions, might is significantly less frequent in antecedents of conditionals and matrix questions than can, but more frequent in complements of attitude verbs". (Hacquard & Wellwood, 2012: 5) "Epistemic must is significantly less frequent than root must in antecedents of conditionals and questions. . . In the complements of attitude verbs, however, their distribution does not differ significantly". (Hacquard & Wellwood, 2012: 6) While these findings do not provide a decisive answer as to the underlying reason why epistemic modals are less frequent in certain embedding environments, they support the account proposed in this dissertation by backing up the claim that epistemic modals are hard to embed, unlike root modals. The findings are also compatible with the predictions of the account proposed in this dissertation i.e. that at least for a subset of epistemic modals there will be problems embedding (of course, the size of this subset cannot be determined by these data). Interestingly, Hacquard & Wellwood's data do contain some indication for an expressivedescriptive distinction (subjective-objective in their terms): "epistemics could be acceptable in questions and antecedents of conditionals if the epistemic claim can be interpreted relative to someone aside from the speaker’s knowledge state, or that of a larger group that includes the speaker. Papafragou (2006) argues that such epistemics anchored to the collective knowledge of the speaker’s community constitute what Lyons (1977) called objective epistemics. Because such epistemics are not anchored to the speaker’s sole knowledge (a subjective epistemic use), the speaker can felicitously use them in questions or suppositions (is it/if it is possible given what the community knows that p). Do we find any evidence for such a view? A few instances of might in antecedents of conditionals do hint at such an objective stance, where might seems anchored to an implicit generic perspective…" (Hacquard & Wellwood, 2012: 19) However, Hacquard & Wellwood refrain from stating that this is indeed the case: "Absent further context, however, it is difficult to state with certainty whether the epistemics we found in questions and antecedents of conditionals have to be interpreted objectively, or otherwise anchored to someone other than the speaker". (Hacquard & Wellwood, 2012: 19) 99 As discussed previously, context indeed plays an important factor in establishing whether the use is expressive or descriptive, and this is the reason for conducting the abovementioned experiment by Cohen, Wolf and Taube. Of course, such issues are only now starting to be checked empirically in various ways; hopefully more empirical data will be gathered in the future that will shed more light on the matter. 3.2.2.2 Agreement-disagreement As mentioned in chapter 2, agreement/disagreement cases provide a good diagnostic for non-asserted content. Disagreements, for example, as opposed to negation, target a previously performed speech act, thereby may help distinguish between the content that has been asserted and the content that hasn't. The following examples are taken from Papafragou (2006), the first regarding expressive EM: (35) A: This professor must be smart. B: Is that so? ≠ Is it the case that this professor must be smart? = Is it the case that this professor is smart? b. B: That's not true!/ I don’t believe it. ≠ It's not true/ I don’t believe that this professor must be smart. = It's not true/ I don’t believe that this professor is smart. c. B: I agree. ≠ I agree that this professor must be smart. = I agree that this professor is smart. The same principle applies to possibility EM: (36) Speaker: This might be the best movie ever. 100 a. Hearer: Is that so? ≠ Is it the case that this might be the best movie ever? = Is it the case that this is the best movie ever? b. Hearer: That's not true!/I don’t believe it. ≠ That's not true!/ I don’t believe that this might be the best movie ever. = That's not true!/ I don’t believe that this is the best movie ever. c. Hearer: I agree. ≠ I agree that this might be the best movie ever. = I agree that this isthe best movie ever. These data contrast with descriptive EM, as characterized by weather claims (Papafragou, 2006): (37) Speaker: (According to the weather forecast) it might rain tomorrow. a. Hearer: Is that so? = Is it the case that it might rain tomorrow? ≠ Is it the case that it will rain tomorrow? b. Hearer: That's not true!/I don’t believe it. = That's not true!/I don’t believe that it mightrain tomorrow. ≠ That's not true!/I don’t believe that it will rain tomorrow. c. Hearer: I agree. = I agree that it might rain tomorrow. ≠ I agree that it will rain tomorrow. 101 MacFarlane (2011) also discusses agreement-disagreement cases, via the possibility of eavesdroppers (in the following scenario, it is the reader), i.e. individuals who are not directly participating in the conversation but nonetheless have interest and opinions with regards to it: (38) First case: You overhear George and Sally talking in the coffee line. Sally says, “I don’t know anything that would rule out Joe’s being in Boston right now” (or perhaps, more colloquially, “For all I know, Joe’s in Boston”). You think to yourself: I know that Joe isn’t in Boston, because I just saw him an hour ago here in Berkeley. Question: Did Sally speak falsely? (39) Second case: Scene as before. Sally says, “Joe might be in Boston right now.” You think to yourself: Joe can’t be in Boston; I just saw him an hour ago here in Berkeley. Question: Did Sally speak falsely? MacFarlane's example is used in order to motivate the assessment-sensitivity of epistemic modals, but I suggest an alternative explanation – the might used in MacFarlane's scenario is an expressive epistemic modal. As is discussed in the previous chapter, asserting a possibility expressive epistemic modal is asserting the prejacent with a low degree of strength. Thus, when a hearer evaluates Sally's assertion, she evaluates the propositional content under negotiation, and this content is false in light of the hearer's knowledge. Another conversational test, previously used in chapter 2, is the 'reason to assert': (40) Scenario A: Dave is a driver who is looking for the person he needs to pick up from the airport. He knows that this person's name is Nina and he knows that 102 Nina has blue eyes. Many people start to come out the gates, and Dave spots a woman with blue eyes. He then says to a fellow driver standing nearby: Dave: This might be Nina, my passenger. Driver2: Why do you say that? Dave: Well, she has blue eyes. Scenario B: Ted is looking for a wife. He has a list of criterions that this person needs to fulfill and one of these criterions is for her to have blue eyes. When walking down the street with his friend Marshall, he sees a woman with blue eyes. Ted: This woman might be my future wife. Marshall: Why do you say that? Ted: ?Well, she has blue eyes. Scenario A is intended to bring about the descriptive reading of the modal and scenario B the expressive. The difference in the appropriateness of 'reason to assert' replies relates to the content of each assertion. In Scenario A, the content is 'it's possible that this is Nina', which is true. There is indeed such a possibility in light of the information available to Dave. In scenario B on the other hand, the content is 'this woman will be my future wife', albeit with a low degree of strength. It seems weird to offer this content for update on the basis of very little evidence. 103 3.2.2.3 Scope The different scopes of epistemic modals and root modals with regards to negation have been discussed previously in this chapter, but there are a few more points to be made. First, interactions between modals and negation is not a simple matter, as stated in Hacquard (2011): "Several factors, beyond the epistemic/root distinction, seem to conspire to make the interaction of modals with negation a complex matter, such as the possibility/necessity distinction (Cormack & Smith, 2002), the position of negation, which varies crosslinguistically, and idiosyncrasies of various modal auxiliaries." That being said, the account proposed in this dissertation dictates that if there is a case in which an epistemic modal can be embedded under negation, this epistemic modal can't be expressive. And, following the claim that the expressive reading is the default in English (see section 3.2.2.5), we can predict that when an EM is embeddable and the context doesn’t indicate expressiveness, the EM is descriptive. Here are examples of epistemic modals that can be embedded under negation (Hacquard, 2011): (41) a. Jane doesn’t have to be at home. b. Jane need not be home. c. Jane can’t be home. In order to check if these specific lexical EM are specialized for expressive or descriptive meaning, we apply our previously used embedding tests, under the assumption that ambiguous EM need an additional expressively oriented context in order to be unembeddable: (42) Conditionals: a. If Jane has to be at home, I will drive her back. b. If Jane needs to walk back home, I will let her walk instead. 104 c. If Jane can walk such a long way, I should start walking too. (43) Factives: a. It's surprising that Jane has to be at home since her parents didn't say anything. b. It's sad that Jane needs to walk back home since her legs hurt. c. It's nice to know that Jane can walk such a long way even though her legs hurt. (44) Assent/dissent: Speaker: Jane has to/needs to/can be at home. Hearer: Is that so?/I don't believe it./I agree. = Is it the case that Jane has to/ needs to/ can be at home? = I don’t believe that Jane has to/ needs to/ can be at home. = I agree that Jane has to/ needs to/ can be at home. If we assume, following Kratzer (1991) that can is root modal, there is no wonder that it can be easily embedded and is part of the content assented to or dissented from. It is interesting to note that can acquires an epistemic interpretation embedded under negation, but the question of how this happens is not relevant to this discussion. What is important is that this epistemic meaning is descriptive, and that the other two epistemic modals pattern with can, thereby making them descriptive EM as well. A prominent account which addresses the scope of epistemic modality is von Fintel & Iatridou (2003). In this account many cases in which epistemic modals take wide scope over various quantifiers are discussed. An example scenario (von Fintel & Iatridou, 2003: 175): "We are standing in front of an undergraduate residence at the Institute. Some lights are on and some are off. We don’t know where particular students live but we know that they are all conscientious and turn their lights off when they leave. So, we clearly know that not all of the students are out (some lights are on and they wouldn’t be on if the students 105 were away). It could in fact be that all of them are home (the ones whose lights are off may already be asleep). But it is also possible that some of them are away. Since we don’t know which student goes with which light, for every particular student it is compatible with our evidence that he or she has left. With this background, consider the following sentence: (45) Every student may have left." If scope relations were free, this utterance would have two possible scope readings. Either the epistemic modal takes narrow scope, i.e. 'for every student it is (epistemically) possible that this student has left', or narrow, i.e. 'it is (epistemically) possible that every student left'. The narrow scope reading is the one compatible with the scenario, but for some reason this reading is not accessible, and the judgments are that (45) is false. Von Fintel & Iatridou propose the following Epistemic Containment Principle (ECP): (46) At LF, a quantifier cannot bind its trace across an epistemic modal. *Qi . . . [Epistemic Modal (. . . ti . . .)] The result of which is that epistemic modals can't have narrow scope when interacting with quantifiers. Another example of this principle at work is: (47) #Every student might be the tallest person in the department. (von Fintel & Iatridou, 2003 : 176) However, Tancredi (2007) notes that if an utterance is modified by phrases such as 'objectively speaking' (effectively making the descriptive reading salient) the ECP doesn’t seem to be in effect: (48) Objectively speaking, every student may have left. >> The following is also counterexample for (47): (49) (As is widely known) every person in our fund-raising events might be the richest person in the country. >>>> richest 106 In light of these counterexamples, the ECP should be modified. Instead of a principle that applies to all epistemic modals, it should be regarded as a principle that applies to expressive ones. 3.2.2.4 Diachronic evidence From a historic perspective, epistemic modals went through a meaning shift in English. Traugott (1989) discusses the development of these meanings and arrives to the conclusion that: “the modals in question show not only the well-known shift from non-epistemic to epistemic, but also a shift from relatively weak subjective epistemicity to more strongly subjective epistemicity. This finding supports the claim that meanings tend to move toward greater subjectivity…”(Traugott, 1989: 43) Traugott views all epistemic modals as subjective; therefore the term 'weak subjective epistemicity' corresponds to descriptive EM and the term 'more strongly subjective epistemicity' corresponds to expressive EM. This diachronic development is part of a continuum, when deontic meanings precede epistemic meanings and descriptive epistemic meanings precede expressive ones. The following are a few historic examples in which EM behave strictly descriptively49: (50) Embedding under a conditional (Traugott & Dasher, 2002: 127): …& raDe aefter Dam, gif hit mot gewiderian, mederan settan. and quickly after that, if it may be-fine, madder plant "and quickly after that, if the weather may be fine, [one can] plant madder". 950-1050, LawGer 12, 454. 49 This evidence is, of course, not conclusive since the claim of this dissertation is that even today EM have both expressive and descriptive readings. But it serves to show a pattern, i.e. to suggest that the default reading of EM in the past was descriptive while today the default reading is expressive. 107 (51) A modal adverb used descriptively (Traugott, 1989: 46): “You wrote so probably that hyt put me in a feare of daungerys to come” What you wrote was so probable that your words made me afraid of dangers to come.1535 Starkey Let., England xxx. (52) A speaker-oriented adverb (of the evidential type) used descriptively (Traugott, 1989: 46): “Yif thay finde evidently that I have doon extorcion” If they find from evidence that I have performed extortions 1429 Will Braybroke in Ess.AST 5:298. 3.2.2.5 Synchronic evidence According to Lyons (1977) English EM are interpreted subjectively by default, i.e. the expressive reading is the predominant one all things being equal. Returning to von Fintel & Iatridou's (2003) ECP, the following utterance obeys the ECP, and therefore, the only available reading here is one in which the epistemic modal takes wide scope, i.e. 'it is (epistemically) possible that all the students left but not everyone has'. Consequently, (53) is infelicitous: (53) #Every student may have left but not every one of them has. (von Fintel & Iatridou, 2003: 46) 108 Unlike English in which the expressive reading of EM is the default reading, Dutch is a language in which the descriptive50 reading is the default (Huitink, 2008), and indeed Dutch EM are not subject to the ECP: (54) Iedere student kan vertrokken zijn, maar niet iedere student is vertrokken. Every student may left be but not every student be left Every student may have left but not every one of them has.>> (Huitink, 2008: 3) Interestingly, in Chinese epistemic modals are marked for scope by a special particle DOU (Lin, 2012), thus both scope readings are available: (55) a. Mei-ge xuesheng keneng dou likai-le Every student may >> DOU leave. b. Mei-ge xuesheng dou keneng likai-le. Every student DOU may leave. >> Even more interestingly, it turns out that wide scope Chinese EM are subjective and narrow scope Chinese EM are objective (Lin, p.c). The following examples, using questions (Lin, p.c), support this: (56) #Mei-ge xuesheng dou keneng mai-guo shenme? every-CL student DOU may buy-ASP what 'What thing might every student have bought?' (57) Mei-ge xuesheng keneng dou mai-guo shenme? every-CL student may DOU buy-ASP what 'What might be the thing x, such that every student have bought x?' 50 Objective in Huitink (2008) terms. 109 While a translation of epistemic modality scoping in a high position relative to an interrogative is not easy, the important thing is that wide scope EM are worse than narrow scope EM, in questions. Hopefully more research will be done in the future with regards to DOU marked epistemic modals. To sum up this section, a wide range of data is presented in support of the distinction between two types of epistemic modal, expressive and descriptive. None of the camps mentioned in section 3.2.1, i.e the truthconditional or the non-truth-conditional can account the full range of data presented in this section. However, there are theories that discuss and attempt to account for the subjective-objective distinction and some of the data presented here. These theories are discussed in the next section. 3.3 Addressing previous accounts of the expressive- descriptive51 distinction 3.3.1 Lyons (1977) Lyons (1977), discussed in section 3.2.1, has introduced the subjective-objective distinction with regards to epistemic modals. While Lyons' insight is significant, there are reasons to believe that 'subjective' and 'objective' are not the optimal terms for this distinction. For one, as stated in Palmer (1986: 11) : “there is no distinction between […] what is logically true and what the speaker believes, as a matter of fact, to be true”, therefore Lyons' distinction between the speaker's objective EM claim about 'truth in the world' and the speaker's subjective EM belief does not exist. Portner (2009: 122), discussing Palmer states: "Palmer (1986) notes that if we define epistemic modality in terms of the speaker’s (or speaker’s and hearer’s) knowledge or strength of commitment, and subjectivity in terms of the evaluation of a proposition by the speaker (or speaker and 51 Since the distinction between 'expressive' and 'descriptive' EM is coined in this dissertation, most of the accounts discussed in this section use Lyons' terms 'subjective' and 'objective'. I will retain the original terms when referring to other works, and use my own terms otherwise. 110 hearer), epistemic modality will always be subjective in this sense. The closest thing to objective epistemic modality on Palmer’s view would be alethic modality". Indeed, there is an intuitive problem regarding the very notion of what it means to be an 'objective epistemic modal', since any speaker asserting an epistemic modal ultimately bases it on personal knowledge which must be subjective. Secondly, as discussed in Nuyts (2001), Lyons' subjective-objective distinction is vague and can have different interpretations (Nuyts, 2001: 393) : A. Subjectivity vs. objectivity: "… the dimension concerns the quality of the speaker's evidence for an epistemic evaluation: does (s)he have good, mathematically or formally reliable evidence (i.e., objectivity), or does (s)he have poor or vague, intuitive evidence (i.e., subjectivity)." B. Subjectivity vs. inter-subjectivity: "Alternatively, the dimension can be defined in terms of a difference in the status of the evidence and the epistemic evaluation based on that evidence from the perspective of the (knowledge of the) interaction partners in that situation. The alternatives within this dimension could then be phrased as follows: one pole involves the speaker's indication that (s)he alone knows (or has access to) the evidence and draws conclusions from it; the other pole involves his/her indication that the evidence is known to (or accessible by) a larger group of people who share the same conclusion based on it". Indeed, there is confusion in the literature regarding these two different ways of drawing the subjective-objective distinction, as will be seen in the next subsections. Nuyts further claims that based on corpora studies conducted in Dutch and German, the second type of distinction, i.e. subjective vs. intersubjective, is more appropriate. In light of these concerns, the theory proposed in this dissertation draws the distinction differently. Both types of epistemic modality are subjective, but the difference lies in the speech-act performance level – Lyons' 'subjective EM' are performed as expressions of doubt or certainty, i.e. are expressives that modify the strength of the speech act, and 111 Lyons' objective EM' are performed (with a default non-modified degree of certainty) as descriptive claims about epistemic possibilities or necessities. Another criticism of Lyons, raised in Portner (2009), is that Lyons' formalism, repeated here, is not clear: (58) Alfred may be unmarried. (59) Subjective: Poss + it-is-so + [[ Alfred is unmarried ]] (60) Objective: I-say-so + Poss + [[ Alfred is unmarried ]] Portner (2009: 123): "two points remain unclear: First, do the Poss in [(59)] and the poss in [(60)] have anything in common? And, second, what is the nature of the speech act that is in [(59)]? If the speaker is not asserting, what exactly is she doing?" The theory proposed in this dissertation answers both of these questions, as it represents both expressive and descriptive EM as the same lexical item (a probability measure) with different scopes, and accounts for the nature of the speech act as a degree of assertion, i.e. an assertion with a modified degree of strength. 3.3.2 Kratzer (1981) Kratzer discusses Lyons' subjective-objective distinction via the following scenario (Kratzer, 1981: 57): “Imagine that Lenz, who often has bad luck, is going to leave the Old World by boat, today, on Friday thirteenth. On hearing about this, someone might utter one of the following sentences52: (61) Wahrscheinlich sinkt das Schiff probably sinks the boat. Probably, the boat will sink. 52 Kratzer's examples are numbered (42)-(45). 112 (62) Es ist wahrscheinlich, daß das Schiff sinkt. it is probable that the boat sinks. It is probable that the boat will sink. (63) Das Schiff wird (bestimmt) sinken. The boat will (certainly) sink. (64) Das Schiff dürfte sinken. The boat sink. It is probable that the boat will sink. The first two examples use the same modal wahrscheinlich in two different manners, subjective and objective. Note that the subjective manner is translated by Kratzer using the modal adverb, and the objective manner using the modal adjective. The last two examples use different modals, wird serving for the subjective example and dürfte for the objective. Thus, Krstzer's example demonstrate how some modals may be ambiguous between expressive and descriptive readings (as I claim English auxiliary modals are) while other modals are specialized for either expressive or descriptive uses (as I claim English modal adverbs and adjectives are). Kratzer explains the subjective-objective distinction by different stereotypical ordering sources. Since a stereotypical ordering source is normative, the propositions it contains are dictated by the way society views what is 'normal'. Of course, there are many ways to view propositions as 'normal' and some of them may be contradictory. Thus, the same person (the hearer) might regard a modalized utterance from two points of view – an objective one utilizing an objective stereotypical ordering source, or a subjective one utilizing a subjective stereotypical ordering source. The former may contain propositions regarding 'established facts about the boat, the technical equipment nowadays or the weather' (Kratzer, 1981: 58)53. The latter may contain superstitions. 53 Note that Kratzer addresses the subjectivity-objectivity divide and not subjectivity-intersubjectivity. 113 This account of the subjective-objective distinction is problematic as it relies on social norms, as discussed in Portner (2009: 124-125): "Kratzer’s discussion is probably not adequate as a general theory of the difference between subjective and objective epistemic modality. She describes the difference between subjective and objective ordering sources in a highly culture-specific way, since it depends on the status of certain beliefs as superstitions. Perhaps in another culture, the kinds of beliefs that we consider to be superstitions are esteemed as the most indisputable kind of knowledge. What we need is a general theory which lets us classify some reasons as objective and others as subjective on the basis of how those reasons are viewed, employed, and responded to by speakers of the language in question. For example, the fact that among German and English speakers one can dismissively say “that’s a mere superstition” shows that fear of Friday the thirteenth is subjective, and the fact that one cannot dismissively say “that’s a mere scientific fact” shows that a satellite image of a giant iceberg is objective. But in another language community, the relevant distinction between kinds of reasons might be quite different". However, in a later paper Kratzer (1991) provides an adjusted account of the differences between the expressive EM wird and the descriptive EM dürfte54, not in terms of subjective vs. objective stereotypical ordering sources, but rather in terms of a doxastic ordering source for wird, vs. a stereotypical ordering source for dürfte. Hence, making expressive EM rely on subjective or personal beliefs and descriptive EM rely on more objective or socially accepted norms. While this takes care of the criticism raised in Portner, it leaves open another problem – truth conditionality. Kratzer's account is compositional, and as such does not account for the non-truth-conditional properties of expressive EM explored in section 3.2.2. There are, however, several theories working in the classical framework that discuss and attempt to account for the expressive-descriptive distinction. These theories are addressed in the following subsections. (Huitink, 2008) 54 The terms 'expressive' and 'descriptive' are, of course, not used in Kratzer (1991). 114 3.3.3 Papafragou (2006) Similarly to Kratzer (1981), Papafragou (2006) explains subjectivity and objectivity in terms of subtypes of epistemic conversational backgrounds. Unlike Kratzer, Papafragou's discussion of Lyons' objective epistemic modality relates to the intersubjective aspect. This aspect is formalized using a group of 'knowers' which epistemic modals are relative to. According to Papafragou, an epistemic modal is true iff for every individual who is a member of this contextually-determined group there is some epistemically accessible world in which the prejacent is true. Formally: (65) [| might |]c,i = 1 iff xGc: w’fx(i): [||]c,<w’,ti>=1 Gc : contextually based group of individuals whose knowledge is relevant. fx: the set of worlds compatible with what x knows. The subjective reading of EM is a subtype of the objective reading, i.e. a case in which the speaker is the only member of Gc. The first problem of Papafragou's account, discussed in Portner (2009: 161), is that the truth conditions presented in the above formula require for every individual to be epistemically aware of the possibility that . Therefore, the objectively modalized sentence ‘it might rain tomorrow’ turns out false if even one meteorologist fails to have access to a crucial piece of data (because, for example, his internet connection is down). The second problem of Papafragou's account arises from the treatment of EM embeddings. Papafragou explains the conditional embedding problems by suggesting that ‘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 her belief-set’ (Papafragou 2006:1696). She presents the following example: (66) ?If I conclude that the Earth is flat, then I’m in trouble. However, consider the following case, in which Columbus on his way to the new world, utters the following sentence: 115 (67) If I examine my knowledge and beliefs and this examination leads me to the conclusion that the earth is not flat, I should proceed with my journey. In here, Columbus performs a mental evaluation which is done with respect to his belief set, as explicitly stated in the antecedent. Thus, according to Papafragou, this utterance should be infelicitous, but it isn't. With regards to embedding under factives, Papafragou states that ‘The complement of (factive) attitude verbs…not environments in which the speaker’s evaluation of a proposition with respect to her current beliefs could feature’ (Papafragou 2006:1697). But the following example shows otherwise: (68) It is surprising that after examining my knowledge and beliefs I find myself agreeing with my colleague, but I have to admit that this is indeed the case. In here, the complement of the factive surprising contains the speaker’s evaluation with respect to her beliefs, and again while Papafragou's account predicts infelicity, there isn't any here. 3.3.4 Portner (2007; 2009) Portner (2007; 2009) provides an account in which propositions are not only asserted but also shared. Portner adopts Stalnaker's (1978) notion of the common ground (CG) as a set of propositions, and the notion of context update as adding a proposition to this set. To this Stalnakerian picture, Portner adds the CPS – Common Propositional Space, which is a set of propositions in which the participants in the conversation are mutually interested, which are not necessarily presupposed. Portner (2007) defines the CPS in the following manner: (69) The CPS is the set of propositions of which the participants in the conversation are mutually aware. Mutually aware: A is aware of φ, B is aware of φ, A is aware that B is aware of φ, B is aware that A is aware of φ, etc. 116 Portner represents the 'conversational score' by apair consisting of the common ground and the CPS: <cg , cps>, where cg ⊆cps. The context update effect of might assertions according to Portner is: (70) Update potential of might: for any sentence φ of the form might , the update potential of φ used in context c with modal base f and ordering source g, [|φ|]c, f,g, is: <cg , cps>[|φ|] c, f,g= <cg', cps'>where (i) cg' = cg ∪{[| φ |] c, f,g}, and (ii) cps'= cps ∪ {[||] c, f,g} ∪{[| φ |] c, f,g} In prose, an EM update takes a conversational score containing a certain CG and CPS, and returns a new conversational score in which the CG is added the fully modalized proposition, and the CPS is added both the modalized proposition and the prejacent. Thus, Portner captures the 'proposal nature' of EM assertions, i.e. raising possibilities. Portner accounts for assent-dissent in the following way: since both the modalized proposition and the prejacent are added to the CPS, hearers can in principle assent to or dissent from either of them. The exact proposition assented to or dissented from is the one which is more relevant. Portner accounts for the embedding problems by proposing that objective EM have an ordering source which is a subset of the CPS, while subjective EM have an ordering source which is a subset of the speaker's beliefs. Embedding problems for subjective EM arise from two possible reasons – either the speaker’s beliefs are not available in these contexts, or that both conditionals and factives are dependent upon propositions within the CPS and thus can't be defined for subjective EM. Portner's theory provides an elegant and interesting approach to the context update effect of EM. However, as noted by Portner himself, it is still preliminary: "we certainly need a better understanding of the CPS" (Portner, 2009: 176). Also noted by Portner (2007) is that the explanation of embedding problems for subjective EM in terms of the unavailability of the speaker’s beliefs is under question because bouletic modals, which reflect subjective desires, can be embedded. 117 Another problem arises with regards to the assent-dissent explanation, i.e. that the speaker can assent to or dissent from either of the propositions added to the CPS. This is problematic because standard cases of assertion do not work in this manner. For example, a scenario in which Reut and her friend Moshe are talking on the phone and Moshe is with a mutual friend, Assaf, standing in front of a window 20 story high. Moshe says to Reut: (71) Assaf has just jumped out of the window! As defined in Portner (2007), the CPS is a set of propositions that conversational participants are aware of. The above assertion certainly makes both conversational participants aware of the proposition 'Assaf is going to die' as an implication of (71). This proposition is, of course, very much relevant. However, if Reut dissents from Moshe's assertion, the dissent does not seem to target this proposition55: (72) a. Reut: That's not true! (  It's not true that Assaf is going to die) 3.3.5 Anand & Hacquard (2008)56 Anand & Hacquard (2008) provide a truth-conditional account of EM within the classical tradition, i.e. in terms of quantifier over possible worlds. However, unlike Kratzer (1981, 1991), here modals are relative to an event of evaluation and not a world, following Hacquard (2006, 2008). While the theory provided in this paper does not aim to account directly for the subjective-objective distinction, it does raise an interesting point with regards to attitudes of acceptance that permit EM within their scope. Anand & Hacquard make a distinction between doxastic attitudes (e.g. believe, think, know) and proffering attitudes (e.g. claim, argue, demonstrate) and claim that the former impose a subjective interpretation of EM and the latter an objective interpretation. The subjectivity of doxastic attitudes is derived from their being based on the belief of the speaker and the 55 Though, the implication of the proposition that the dissent targets is that Assaf is going to die, and therefore it is also dissented from, but this is a different matter. 56 At the time of printing of this dissertation, a more recent version of Anand & Hacquard (2013) has been published. A proper account of Anand & Hacquard's recent work will have to await future work. 118 objectivity of proffering attitudes is derived from their goal to make their complements become part of the common ground. Anand & Hacquard's theory very elegantly accounts for the difference between doxastic attitudes and proffering attitudes and nicely derives the subjective stance involved in the former and the objective stance involved in the latter. I will not go into the intricacies of their account of attitudes here, but rather wish to raise a few points regarding the basis of the theory. This theory is based on Hacquard (2006, 2008), in which epistemically modalized assertions are doxastic, i.e. based on the belief of the speaker. This is problematic, as stated in Kratzer (2012: 55): "according to Hacquard, the possibilities of modals appearing in higher regions are keyed to the epistemic possibilities of speakers […] the last conclusion does not seem to be entirely correct". The reason for that is that EM do not necessarily have to relate to either belief or knowledge, as in the example discussed in chapter 2, repeated here: (73) According to the newspaper reports, the thief must have entered through the kitchen window. (74) According to the testimony we heard, the defendant must be suffering from severe emotional distress. In the theory proposed here, EM do not necessarily relate only to doxastic states – the probability spaces that the theory relates to can be either information states (doxastic), i.e. defined over the beliefs of individuals or 'evidential' probability spaces that are defined over the available evidence57. Another problem is the matter of vacuous quantification in Anand & Hacquard's formalism. Formally, a standard non-modalized utterance is always implicitly modalized and it quantifies over the speaker’s doxastic alternatives, i.e. belief worlds in the following manner: 57 A probability space that is defined over evidence contains propositions pertaining to the evidence and conditional probabilities that stand for the likelihood of events given the evidence. 119 (75) a. It is raining. b. e0 [CP ASSERT e0 [TP it is raining] ] ] c. w’CON(e0) [ it is raining(w’)] When e0is a default root event binder which binds the asserting event ASSERT, and CON(e0) is the intersection of the content of set of e0, i.e. the set of propositions composing the speaker's beliefs. In prose, the formula states that in every world which is a member of the speaker’s doxastic alternatives, it is raining. An epistemically modified utterance therefore imposes extra quantification over the speaker's doxastic alternatives: (76) a. It might be raining. b. e0 [CP ASSERT e0 [ModP might f (e0) [TP be raining] ] ] c. w’CON(e0) [w’’CON(e0) it is raining(w’’)] d. w’CON(e0) [it is raining(w’)] In prose, for every world which composes the speaker's belief worlds there is some world which is a member of the speaker's belief worlds in which it is raining. Since (c) is a vacuous quantification, the formula amounts to (d), i.e. that there is some world which composes the speaker's belief worlds in which it is raining. While this is not a big problem, there is a matter of cognitive efficiency that should be taken under consideration, i.e. why should conversational participants go through extra processing every time they assert and hear utterances containing epistemic modals? A theory that accounts for EM without such vacuous quantification is more desirable. Lastly, while Anand & Hacquard do not provide a representation for necessity EM, following the bovementioned account of might, it should look like this: (77) a. It must be raining. 120 b. e0 [CP ASSERT e0 [ModPmust f (e0) [TP be raining] ] ] c. w’CON(e0) [w’’CON(e0) it is raining(w’’)] d. w’CON(e0) [it is raining(w’)] Except from the vacuous quantification which exists here as well, note that the truth conditions of necessity EM and the truth conditions of non-modalized assertions are exactly the same, because non-modalized assertions in this theory are already epistemically modalized. This poses a problem since it is desirable to account for the difference between necessity EM and standard non-modalized assertions. 3.3.6 von Fintel & Gillies (2011) and Cohen (2013) Cohen (2013) provides an account of the subjective-objective distinction by expanding von Fintel & Gillies (2011), in which EM utterances put into play a cloud of propositions. For instance, the following is an utterance and a cloud: (78) Utterance: The dog might be on the lawn. Cloud: C is a cloud of contexts such that each context c  C, specifies a relevant group of 'knowers' G relevant to the utterance. Thus, the utterance above puts into play the following cloud (containing possibly more propositions, as required by context): a) [| The dog might be on the lawn |]c=speaker b) [| The dog might be on the lawn |]c=hearer c) [| The dog might be on the lawn |]c=speaker+hearer d) [| The dog might be on the lawn |]c=speaker+hearer+eavesdropper There can be more propositions in the cloud, but importantly the speaker has to be able to assert at least one of these propositions in order to put the cloud into play. After the 121 assertion is performed, the speaker can respond to any proposition in the cloud, which by default is the one he finds the most relevant. While von Fintel and Gillies (2011) do not discuss embedding of epistemic modals, Cohen (2013) does. According to Cohen (2013) subjective EM are semi-assertions that put into play von Fintel & Gillies' cloud, and hence they require an assertive context. Thus, subjective EM are acceptable under verbs of saying and belief that provide an assertive context , but not under conditionals or presuppositional factives. Intersubjective58 EM are formalized in Cohen (2013) as mixture models of competent reasoners, along the lines of the mixture model discussed in chapter 4, i.e. each reasoner i assigns a degree of belief to the asserted proposition. This degree is represented as a probability measure: Pi(). Each reasoner is also assigned a weight wi, indicating how 'reasonable' this reasoner is considered to be. Thus, the final probability value59 is: (79) P(  ) =  wiPi(  ) n i=1 Importantly, intersubjective EM do not put into play a cloud of propositions and are therefore not semi-assertions. Thus, they do not require an assertive context and do not have restrictions on embedding. Cohen's account provides a very good option for incorporating the expressive-descriptive (or subjective-intersubjective) distinction into the classical framework, and may well be the correct account of epistemic modals, but there are some matters that need to be settled before that can occur. The first is that this account still needs to explain the scope data, i.e. why expressive (or subjective) EM scope above quantifiers while descriptive (or intersubjective) EM scope below. The second is that this account views the difference between EM types as lexical. Therefore, it runs into the problems that originally motivated the classical Kratzerian account, i.e. solving the puzzle of multiple lexical entries for different modals. 58 59 Cohen (2013) opts for the intersubjective interpretation of Lyons' objective epistemic modality. The sum of all weights is 1, therefore this is a probability measure. 122 3.3.7 Huitink (2008) Huitink (2008) is also a good account of the subjective-objective distinction that suffers from the lexical problem, i.e. representing subjective and objective EM in terms of different lexical items. Huitink presents an ambiguity account along the lines of Lyons (1977), i.e. subjective EM are non-truth-conditional modifiers of illocutionary force while objective EM are truth-conditional markers of propositional possibility. Formally, Huitink treats speech acts as commitment spaces (Krifka, 2001) and draws the subjective-objective distinction in the following manner: (80) [|mays|] = λc. c + the speaker is weakly committed to  (81) [[mayo]] = λb. b is compatible with  In prose, a subjective mays that is applied to a proposition  modifies commitment spaces c such that the speaker is weakly committed to , i.e.  is compatible with the speaker's beliefs. An objective mayo that is applied to a proposition  modifies the set of beliefs b of the speaker or a group of believers60, such that b is compatible with . Objective EM are thus propositional hence easily embedded while subjective EM are speech-act modifiers thus not easily embedded. Furthermore, objective EM have narrow scope while subjective EM have wide scope since quantifiers are truth-conditional, hence cannot scope over subjective EM. Huitink's (2008) account is a very good theory of the subjective-objective distinction which, like the theory proposed in this dissertation, is based on Lyons' (1977) insight regarding the dual nature of EM. However, it inherits the same problems raised against Lyons (1977), discussed in section 3.3.1, i.e. that both types of EM are essentially subjective, and that the nature of the connection between mayo and mays is unclear. This account also suffers from the same problem of Cohen (2013), i.e. that it explains the 60 Hence, Huitink's interpretation of objective EM is intersubjective. 123 difference between subjective and objective EM in terms of different lexical items, which is undesirable for a modal theory. 3.3.8 Tancredi (2007) Tancredi (2007) proposes a multi-model account in which each conversational participant has a separate model of interpretation, containing a different set of beliefs. An nonmodalized utterance in this account is formalized as: (82) [[[US]]]a,w is true iff ∃M M is compatible with the beliefs of a in w and [[S]]M,a = 1 Triple brackets stand for the interpretation of an utterance, i.e. an attribution of a model of beliefs to the speaker, while double brackets stand for the interpretation of what was uttered, i.e. for calculating truth conditions. The formula states that an utterance U of sentence S according to speaker a in world w, is true iff there is some model M compatible with the beliefs of the speaker in which the truth value of the asserted sentence is 1. Tancredi accounts for the subjective-objective condition by claiming that Lyons' objective EM are actually metaphysical modals and that all epistemic modals are actually doxastic. Importantly doxastic modals, unlike all other modals – metaphysical and deontic included, are interpreted in a model compatible with the speaker’s beliefs, as (82), and thus do not employ a world parameter. Quantifiers, however, do need a world parameter, in order to fix their domain. As stated in Tancredi (2007: 11): "The quantified expression every detective, for example, will range over a very different set of individuals within the context of a Columbo show, a Sherlock Holmes novel and a real-world crime investigation. If quantifiers are not given a world parameter for this purpose, their interpretation becomes undefined." This is the reason why when doxastic modals interact with quantifiers they always take wide scope – doxastic modals are undefined otherwise. 124 There are problems with Tancredi's account. First, claiming that there are no objective EM but only metaphysical is doubtful. As shown in Cohen (2013), the following utterance has two readings: (83) Simple microorganisms may live on Mars. a. Metaphysical/circumstantial: if we place simple microorganisms on Mars, it is possible that they will survive. b. Intersubjective epistemic: it is compatible with evidence available to experts that simple microorganisms live on Mars. If an objective reading is available it needs to be accounted for. It is possible though that Tancredi's notion of Lyons' objective EM differs from the intersubjective notion mentioned above, and this difference explains Tancredi's metaphysical account. Indeed, Palmer (1986), who also rejects the notion of objective EM, considers it to be more like alethic modality, i.e. a purely logical modality pertaining to necessary or possible truth, which is very close to metaphysical modality. A more serious problem that arises from substituting objective EM with metaphysical modality is that the latter belongs to root modality, and as such is expected to have narrow scope with regards to quantifiers, as discussed in section 3.1. Thus, Tancredi's account does not provide an alternative to the classical Kratzerian account as the latter is able to account for this behavior (cf. Brennan, 1993; Hacquard, 2006). 3.4 Expressivity As discussed in chapter 2, expressives have six properties (Potts, 2007). The following subsections will follow these properties and show that expressive epistemic modal auxiliaries pattern with expressives and that descriptive epistemic modals auxiliaries do not, in a manner similar to chapter 2. 125 A. Independence Expressive EM are independent from the asserted propositional content, hence are not assented to or dissented from: (84) Sue: This might be the best movie ever. Dave(a): That's true/I agree (this is the best movie ever). Dave(b): That's not true (it isn't the best movie)! Descriptive EM are part of the asserted propositional content, hence are assented to or dissented from: (85) Meteorologist: There is a cold front approaching, it might rain tomorrow. Hearer(a): That's true/I agree (#it will be raining). Hearer(b): That's not true (#it won't be raining tomorrow)! B. Nondisplaceability. Regarding embedding under simple negation – this option is not open for modal auxiliaries, which relates to Potts' (2007) claim that there are cases in which nondisplaceability is so strong that even syntactic embedding is impossible, mentioned in chapter 2. However, it is possible to negate the whole speech act, as in (b) and (c) below. In this case, such negations may target descriptive modals but have problems targeting expressive ones. And examples (d) and (e) show the differences in embedding under conditionals: (86) a. *This isn't might be the best movie ever. b. It's just false that this must be the best movie ever (?but it might be ) c. It's just false that it must rain tomorrow (but it might) d. #If Max might be lonely, his wife will be worried. (Papafragou, 2006) 126 e. If it might rain tomorrow, people should take their umbrellas. (Papafragou, 2006) As for embedding under modality: (87) a. Maybe this might be the best movie ever. (?Then again, maybe it must be the best movie ever) b. Maybe it might rain tomorrow. (Then again, maybe it must rain tomorrow) While the descriptive EM can be embedded under another EM, embedding the expressive EM yields a non-embedded model concord reading61. C. Perspective dependence Under default conditions, expressives are bound to the perspective of the speaker. Hence, the following examples show the difference between expressive and descriptive might with regards to such perspective: (88) a. #That bastard Kresge is famous, but I personally think that he's a good guy. b. #Max might be lonely, but I personally think that he isn't. c. It might rain tomorrow, but I personally think that it won't. 61 See chapter 2 section 2.4 on modal concord. 127 D. Descriptive ineffability There is nothing to be said about descriptive ineffability here that hasn't been already said in chapter 2 section 2.3.1D. E. Immediacy As discussed in chapter 2, expressives have immediate effect, hence cannot be easily applied to events that take place beyond speech time. The following example shows how, again, expressive EM pattern with other expressives while descriptive EM do not: (89) a. That bastard Kresge was late for work yesterday. (= the speaker believes today that Kresge is a bastard) b. John might have been lonely yesterday. (= the speaker believes today that John was possibly lonely yesterday) Before presenting the next example, here is some context (based on a scenario of von Fintel & Gillies, 2008: 87): Sophie is looking for some ice cream and checks the freezer. There is none in there. Asked why she opened the freezer, she replies: c. There might have been ice cream in the freezer yesterday. (= the speaker reports that yesterday there was an epistemic possibility of ice cream in the freezer) This is a descriptive claim about the possibility that existed in the past in light of past evidence, and not an expressive claim about belief, since at the time of utterance Sophie does not believe that it is possible (or was possible) that there is ice cream in the freezer (she knows there isn't any). 128 F. Repeatability In order not to repeat myself, see chapter 2 section 2.3.1 on repeatability. 3.5 Formal representations The formal representations in this section are based on the formal account presented in chapter 2, section 2.3.2. 3.5.1 Expressive and descriptive epistemic modals Descriptive epistemic modal auxiliaries modify the propositional content, hence their representation is the same as descriptive epistemic modal adjectives: (90) a. It might rain tomorrow62. b. Ax P (P(rain) > 0) ≥ high In prose, the speaker asserts the propositional content 'it might rain tomorrow', represented by the probability degree of greater than 0. The degree of strength for this assertion is the default, i.e. equal to or greater than high, signifying a high degree of belief or confidence of the speaker in this assertion. The next is a representation of a descriptive necessity EM: (91) a. It must rain tomorrow. b. Ax P (P(rain) = 1) ≥ high Here, the speaker asserts the propositional content 'it must rain tomorrow', represented by the probability degree of 1. The degree of strength for this assertion is again the default, 62 An epistemically modified sentence without any context can be read either expressively or descriptively. This holds even for the weather utterances used in Lyons (1977) and Papafragou (2006). The choice to use weather utterance as examples of descriptive EM is just a matter of convenience. 129 i.e. equal to or greater than high, since assertions of descriptive EM are standard assertions performed with the standard degree of strength. Expressive epistemic modal auxiliaries modify the speech act, just like modal adverbs. Their representation is thus the same as epistemic modal adverbs: (92) a. Max might be lonely. b. Ax P (lonely(m)) > 0 (93) a. Max must be lonely. b. Ax P (lonely(m)) = 1 In prose, both the first and second formulas state that the speaker asserts the propositional content 'Max is lonely'. The difference between possibility and necessity expressive EM lies in the degree of belief or confidence of the speaker in the asserted proposition. For expressive possibility EM this degree is low (greater than 0) while for descriptive necessity EM this degree is very high (1), even higher than the degree of strength for standard non-modalized assertions. 3.6 Syntactic implications63 Descriptive EM are syntactically located lower than expressive EM, which explains embbedability patterns and relative scopes. The following subsections discuss these syntactic positions via the phenomenon of A-movement reconstruction. 3.6.1 A-movement reconstruction A-movement reconstruction is a phenomenon in which an A-moved quantifier can scope both in its raised and base positions, as in the following example from May (1977): 63 This section is based on Wolf & Spector Shirtz (2012). 130 (94) Some politiciani is likely ti to address John's constituency. Since likely is a raising predicate, the existential some politician moves to subject position and leaves a trace at its original location. Interestingly, the existential can be interpreted for scope in both of these positions, yielding two readings, either the wide scope (>>likely) 'there exists some politician who is likely to address John's constituency', or the narrow scope (likely >>) 'it is likely that some politician (or other) will address John's constituency'. While (94) provides prima facie motivation for A-movement reconstruction, the literature regarding this phenomenon is divided. Views or A-reconstruction range from acceptance (May, 1977; Wurmbrand & Bobaljik, 1999) to rejection (Chomsky, 1995; Lasnik, 1999) with intermediate views in the middle, for example Iatridou & Sichel (2011) who claim that A-reconstruction is possible except for some cases, i.e. Neg-DPs to be discussed shortly. While it is beyond the scope of this dissertation to provide a full fledged account of the discussion concerning A-reconstruction, this section will touch on several aspects of it, such as Lasnik's (1999) arguments against A-reconstruction, and provide a preliminary syntactic account that will incorporate this work's theory of epistemic modals at the syntax-semantics interface. The tentative explanation offered is that while theories of scope reconstruction in A-movement focus on the A-moved quantifier, it might be beneficial to explore another research direction – movement of the raising predicate itself i.e. the epistemic modal. Lasnik (1999) discusses two cases in which A-reconstruction does not take place. The first is when the raised DP is under negation (henceforth termed NegDP, following Iatridou & Sichel, 2011) and the second is under proportional modifiers: (95) No one is certain to solve the problem. >>certain / *certain >> (96) Every building is 10% likely to collapse. 131 >> 10%likely / * 10%likely >> (95) lacks the narrow scope reading for the NegDP, i.e. 'it is certain that no one will solve the problem'. (96) lacks the narrow scope reading for the universal, i.e. 'it is 10% likely that every building will collapse'. Iatridou & Sichel (2011), propose a solution for Lasnik's first problem, i.e. NegDPs, in the form of the following generalization: (97) Generalization A: The scope of a subject NegDP with respect to scopal predicates such as modals, raising, and ECM (exceptional-Case-marking) predicates reflects the relative scope of these predicates with respect to the marker expressing sentential negation. a. When these predicates scope above negation, they also scope above a subject NegDP. b. When they scope under sentential negation, they also scope under a subject NegDP. (Iatridou & Sichel, 2011: 597) Indeed, as shown in Iatridou & Sichel, the following predicates scope below sentential negation (the a examples), and also below NegDPs (the b examples): (98) a. It is not certain that he will win. >> certain b. No one is certain to win. >> certain (99) a. It is not proven/shown/demonstrated that the theory is false. >> V b. No theory was proven / shown / demonstrated to be false. >> V And predicates that can scope above sentential negation scope above NegDPs. Such are neg-raising predicates, which allow matrix negation to be interpreted within the embedded clause (Gajewski, 2005; Horn, 1978): 132 (100) a. No doctor appears to be present. V >> b. No doctor seems to be present. V >> c. No student is believed to have witnessed that crime. V >> However, as claimed in Iatridou & Sichel, the cases above are not examples of reconstruction for the following reasons: first, reconstruction yields ambiguity and there’s none here. Second, the type of modal should not affect reconstruction but it does affect the scope here. 3.6.2 The proposed account Wolf & Spector-Shirtz (2012) agree with Iatridou & Sichel – the cases above are not due to the reconstruction of the quantifier. Rather, they are due to the movement of the modal. As for whether A-reconstruction exists or not, we remain agnostic pending further research. The syntactic account we propose is based on Rizzi (1997), who proposes the following internal structure for the left periphery: (101) ForceP > (TopP) > FocP > FinP > TP Based on close correspondence between sentential negation and focus, [neg] is generated as a feature of T (Rizzi 1997) or in NegP (Pollock 1997, Haegeman 1995) and moves to FocP to take sentential scope, yielding the following structure: 133 (102) As for EM syntactic scope, we propose a QR account in which the EM moves, in order to derive various syntactic scopes. This is a feasible option64, since modals are known to move (Iatridou & Zeijlstra, 2009). For one, they are always tensed, i.e. subject to V to T movement and furthermore some modals undergo SUBJ-AUX inversion, i.e. T to C movement. Modals are, of course, quantifiers - whether over possible worlds as in the Kratzerian approach or over probability spaces as in the view proposed in this dissertation. Thus, we posit that modals move by QR (cf. Lin, 2012). According to our account, Descriptive EM, e.g. certain, correspond to Iatridou and Sichel’s (2011) non Neg-raising predicates, which scope below sentential negation. These modals move covertly to FinP (or ModalP in Lin 2012) and take the TP as their complement, yielding the following structure: 64 Another feasible option, not explored here but one we do not object to, is to derive the syntactic locations via base generation (Cinque, 1999). 134 (103) Thus the scope ¬>> certain / *certain >> ¬  is accounted for. Expressive EM, e.g. seem, correspond to Iatridou and Sichel’s (2011) Neg-raising predicates, which scope above sentential negation. Expressive EM, being speech-act modifiers move covertly to ForceP, in order to check illocutionary force (cf. Lin 2012), yielding the following structure: (104) 135 Thus the scope *¬>> seem / seem >> ¬  is accounted for. As discussed in chapter 2, many modals have a (context-dependent) tendency toward being either expressive or descriptive, e.g. modal adverbs' tendency toward expressive readings and modal adjectives' tendency toward descriptive. Other modals such as auxiliaries discussed in this chapter, and modal such as likely do not have a strong tendency either way. Thus they can freely scope above or below sentential negation: (105) No one is likely to solve the problem.>> likely / likely >> In terms of semantic representation: (106) No politician is likely to visit Israel. Descriptive EM assertion: Ax P ( y (Politician (y) & P( Visit ( y, Israel)) > 0.5 )) ≥ high Expressive EM assertion: Ax P (y (Politician (y) & Visit ( y, Israel))) > 0.5 As for Lasnik's problem of proportional modifiers repeated here: (107) Every building is 10% likely to collapse. >>10%likely/ #10%likely >> We propose that the use of explicit probabilities is strongly connected to objective chances which are descriptive. However - the context can be manipulated into a subjective reading of probabilities (i.e. personal probabilities) which makes the expressive reading possible. For example, there is a boxing match and Nick has placed two bets, each bet on one of the boxers. He then excitedly asserts: (108) One of my bets is 100% likely to succeed! The reading in which Nick is 100% certain that one of his bets will succeed is now available. 136 Chapter 465 Clarity 4.1 Clarity and belief Intuitively, if something is clear, then it is not vague. As the dictionary (Collins COBUILD Advanced Learner's English Dictionary) succinctly puts it: “vague ≠ clear”. But this is of course quite compatible with the fact that clear itself is a vague predicate. As pointed out in Barker and Taranto (2003): "the vagueness of clear is easy to prove, since it is possible to explicitly talk about the degree to which a proposition is clear66: (1) a. It is becoming clear that Mary is a doctor. b. It is reasonably clear that Mary is a doctor. c. It is very clear that Mary is a doctor. d. It is painfully clear that Mary is a doctor." (Barker and Taranto, 2003: 17) Barker (2009) argues further for the vagueness of clarity by demonstrating that it is susceptible to the sorites paradox. The sorites paradox applies to cases in which it is very difficult to determine the boundaries of a predicate. For example, the classically vague predicate bald. It is indisputable that an individual with absolutely no hair is bald. But what about an individual with 1 hair fibre? Or 2 hairs? It is practically impossible to determine the point in which a person changes from being bald to non-bald, i.e. the cutoff point in which a person with n number of hairs is bald while the same person with n+1 hairs is not. Moreover, we are inclined to say that no matter the number of hairs that a person has, if it is true that he is bald with n hairs, then he will still be bald with n+1 65 This chapter is based on Wolf & Cohen (2011). However, there are points of dissimilarity e.g. the distinction between expressive and descriptive clarity. Other points of departure will be mentioned along the way. 66 Barker & Taranto's example number 22. 137 hairs. Of course, if we keep adding hairs at some point the person will be non-bald, and the paradox arises from accepting both this and the conditional. Applying the same conditional to clarity yields similar results. Imagine your friend George standing right in front of you, n centimeters away. From this distance, it is clear that this is your friend George. If it is clear that this is George at n centimeters away, then surely it is clear that this is George from n+1 centimeters away. And, of course, if George is moved far enough, at some point you will cease to recognize him, but there is no cutoff point n in which it is clear that this is George while in n+1 it is not clear. But what does clarity mean? Barker and Taranto argue that It is clear that p does not, in fact, entail p, and take this fact as evidence for basing the interpretation of clarity on belief. Specifically, they propose that (2) is true iff the degree of belief of the speaker and hearer in the truth of the proposition that Abby is a doctor is greater than some vague standard.67 (2) It is clear that Abby is a doctor. Their explanation for the fact that (2) does not entail (3) is simple: believing something does not make it true. (3) Abby is a doctor. 4.2 Clarity and justification In a later paper, Barker (2009) argues that belief is not, in fact, appropriate as an account of clarity. He points out that belief is neither a necessary nor a sufficient condition for clarity. Belief is not necessary for clarity, according to Barker, since both speaker and hearer might believe that there is life on Mars, yet assent to (4). (4) It is reasonably clear that Mars is barren of life. 67 Since clarity is vague, so is the standard for clarity. Analogously, one is tall if one’s height is greater than some standard, but, as can be seen by the sorites paradox, defining this standard precisely is problematic. 138 Belief is not a sufficient condition either, because both speaker and hearer may believe in the existence of God, yet still deny (5): (5) It is clear that God exists. Instead of a theory based on belief, Barker proposes an account based on justification. Unlike belief, justification is objective: given some body of evidence, and some relevant standards, it is an objective fact whether some conclusion is justified. Barker’s formalization of his theory is inspired by Kratzer’s (1981, 1991) account of epistemic modals, discussed in chapters 2 and 3. Putting it generally, according to Kratzer: (6) must p is true iff p holds in all epistemically accessible worlds whose “degree of normality” is sufficiently high.68 However, Barker notes that clarity is not the same as epistemic modality. One difference between clarity and must that is particularly relevant here is that the former, but not the latter, requires the relevant knowledge to be publicly available. Suppose there are no publicly available facts indicating that Abby is a doctor, but the speaker has some relevant private knowledge about her profession, say because she has seen the medical tools Abby is carrying in her briefcase. In this case (7) would be perfectly acceptable, in fact true. (7) Abby must be a doctor.69 In contrast, Barker claims that (2) would be false in the situation described. We agree with Barker that (2) would not be unproblematically true in such a case, though, in our judgment, (2) would actually be odd rather than simply false, so we 68 Degrees of normality are formalized relative to a stereotypical ordering source. This example utilizes the expressive use of the modal, in which the speaker may perform an assertion based solely on her private beliefs. The descriptive use works differently. For example, imagine a scenario (originally with regards to objective must, in Cohen, 2013) in which John is the only person who sees a monster under his bed. In this scenario the descriptive necessity EM in (i) is odd: (i) #It is certain that there is a monster under my bed. 69 139 believe that the requirement for the evidence to be publicly available is a presupposition (we will return to this issue in section 4.5.1.3 below). Given the difference between clarity and epistemic modality, Barker modifies Kratzer’s definition as follows. Rather than expressing quantification over epistemically possible worlds, he proposes that clarity expresses quantification over worlds compatible with publicly available evidence. In order to account for the vagueness of clarity, 70 he proposes a vague standard of skepticism, d(clear). According to Barker’s proposed definition, (2) is true iff in all worlds consistent with publicly available evidence whose “degree of normality” is greater than d(clear), Abby is a doctor. This formalization captures Barker’s suggestion that (2) is true iff it is justified, on the basis of publicly available evidence, to conclude that Abby is a doctor. 4.3 Back to belief? Barker uses examples (4) and (5) to argue that belief is too subjective to account for clarity. We agree with his judgments concerning these examples: indeed, a very similar (if slightly weaker) example is attested:71 (8) Most people, including most theists, agree that even if God exists, it is not clear that God exists. However, there are at least two reasons why an account based on belief is attractive nonetheless: personal and adverbial clarity. 70 71 Barker claims that epistemic must is not vague. From James A. Keller (2007), Problems of Evil and the Power of God, Aldershot: Ashgate, p. 31. 140 4.3.1 Personal clarity Barker distinguishes between simple clarity, exemplified by (2), and personal clarity, as in (9). (9) It is clear to me/you/John that Abby is a doctor. Of course, something can be clear to one but not clear to another. But what does it mean for some proposition to be clear to X but not to Y? Barker claims that this can happen in one of two cases. One such case is when X and Y have access to different bodies of (publicly available) evidence. Although the evidence is publicly available, it is still possible that X was exposed to it, but Y was not. In this case, X may be justified in concluding that Abby is a doctor, while Y may not be. The second case in which, according to Barker, a proposition may be clear to X but not to Y, is when X and Y have different standards of skepticism, i.e. different values for the threshold d(clear). X may be credulous, while Y may be harder to convince, and this is why a proposition would be clear to one but not to the other. However, these cases do not exhaust the space of possibilities. Suppose both Alex and Bill see a photograph of Abby wearing a stethoscope and smoking, and they know nothing else about her: the photograph consists of all their relevant knowledge concerning Abby. The following exchange is not unnatural, and neither of them can be accused of uttering a falsehood: (10) Alex: It is clear to me that Abby is a doctor (because she is wearing a stethoscope). Bill: It is clear to me that she's not (because she has an unhealthy habit) Note that, by hypothesis, Alex and Bill share the same publicly available evidence—the photograph. Thus, their disagreement cannot be explained by a difference in the publicly available evidence that Alex and Bill have access to. 141 Can their disagreement be explained as a difference in their respective standards of skepticisms? Well, it is possible that they might differ on their respective values for d(clear), but this still would not explain their disagreement. Suppose Bill is more skeptical than Alex; then the proposition that Abby is a doctor may not be clear to Bill, but its negation will not be clear to him either. Hence, he may be justified in uttering (11), but not (10). (11) It is not clear to me that Abby is a doctor. How, then, can we explain the disagreement between Alex and Bill? Note again that neither of them can be accused of uttering a falsehood; this means that, in general, a disagreement on clarity cannot really be resolved. Consider the following attested example, in which Kyburg (1976, p. 366) discusses his disagreement with Levi: (12) We have here a simple conflict of intuitions, and in the last analysis it may be that you pays your money and you takes your choice. It seems quite clear to me that [Kyburg’s claim]. The opposite is quite clear to Levi. Barker notes this property of personal clarity, and points out its similarity, in this respect, to predicates of personal taste: disagreement on taste is a faultless disagreement (Lasersohn 2005; Kölbel 2003). In the following discourse, just as in (10), it is hard to accuse any of the interlocutors of uttering a falsehood. (13) Alex: This chili is tasty. Bill: This chili isn't tasty. Barker argues that, in both (10) and (13), the disagreement comes from the fact that Alex and Bill have different relevant standards: "Even assuming that two discourse participants agree on the facts in the world, including the degree of tallness, clarity, or tastiness of some object, they can still differ on what they consider to be the prevailing vague standard" (Barker, 2009 : 270). 142 However, degrees of tastiness or clarity, unlike degrees of tallness, are not “facts in the world”. Thus, two people may agree on Mary’s height, but disagree on whether this height is above the cutoff point for being considered “tall”; in contrast, if they agree on the degree of tastiness, fun, or clarity of something, they cannot disagree on whether it is tasty, fun, or clear. Lasersohn (2008) makes this point explicitly: "What seems crucial for disagreements over taste is not the location of the cutoff point, but the assignment of degrees. Different people may assign markedly different degrees of fun or tastiness to the same items, and may differ radically in the relative order of these items on the fun or tastiness scale; but no objective “matter of fact” would seem to select any one of these assignments or orderings as the correct one. John and Mary may disagree whether skydiving is fun, not because they both realize it is fun to degree d and differ as to whether d is sufficiently high to count as fun, but because John (who enjoys a good thrill) assigns it a high degree of fun, while Mary (who is terrified of falling) does the opposite" (Lasersohn, 2008: 308) In other words, what Alex and Bill disagree on, when they disagree on tastiness or clarity, is not merely the defining standard (often called the cutoff point) of the vague predicates tasty or clear, but on the degrees of tastiness or clarity assigned72. We conclude, then, that Barker’s justification-based account does not really explain the puzzle of personal clarity. In contrast, a belief-based account can provide a natural explanation (cf. Barker and Taranto 2003): Alex and Bill reason in different ways, and come to believe different things on the basis of the same evidence. The meaning of personal clarity, then, is based on belief. It would be strange indeed if the meaning of simple clarity were based on radically different principles; hence the meaning of simple clarity ought to be based on belief too. An account of clarity, then, both simple and personal, must make crucial use of belief. 72 Predicates of personal taste are discussed on chapter 5. 143 4.3.2 Adverbial vs. adjectival clarity Barker uses adverbial (clearly) and adjectival (it is clear that) clarity interchangeably, and does not distinguish between their meanings. But it turns out that subtle differences between the two do exist. As discussed in chapter 2, Piñón (2006) discusses adverbial and adjectival modals and notes three differences between them, under negation, conditionals and questions. If we apply Piñon’s observations to clarity, it turns out that adverbial and adjectival expressions of clarity behave differently. Moreover, clearly patterns with modal adverbs, while clear patterns with modal adjectives. Under negation: (14) a. It’s improbable/impossible that Abby is a doctor. b. *Improbably/Impossibly, Abby is a doctor. (15) a. It’s unclear whether Abby is a doctor. b.*Unclearly, Abby is a doctor. In antecedents of conditionals: (16) a. If it’s possible/probable/clear that Abby will be a doctor, then I, too, should apply to medical school. b. *If Abby will possibly/probably/clearly be a doctor, then I, too, should apply to medical school. 144 Under questions: (17) a. Is it possible/probable/clear that Abby is the best doctor in the hospital? b. ??Is Abby possibly/probably/clearly the best doctor in the hospital?73 Additional evidence for Piñon's claim comes from the fact that overt indicators of illocutionary strength, e.g. certainly or presumably, preclude the use of modal adverbs, but not modal adjectives (cf. Krifka 2007): (18) a. Certainly/presumably it is possible/probable/clear that Abby is a doctor. b. *Certainly/presumably Abby is possibly/probably/clearly a doctor. We accept Piñon’s account and apply it to clarity. But if clearly modifies the speaker’s belief in the assertion, its account must make essential use of belief. And since the meaning of clear, while not the same as clearly, is obviously related to it, its account must involve belief too. Hence, both adverbial and adjectival clarity must be based on belief, and the difference between the two is that adjectival clarity assertions are based on descriptive belief while adverbial clarity assertions are based on expressive belief. Let us put this point somewhat differently: even if Barker’s theory successfully accounted for the truth conditions of clarity, it is hard to see how it would be able to explain the differences between adjectival and adverbial clarity. If this is right and the difference depends crucially on belief, it follows that an account of clarity must involve belief. 4.3.3 Desiderata In conclusion, therefore, since there are good reasons to propose a theory that uniformly accounts for both simple and personal clarity, and since there are also good reasons to account uniformly for both clear and clearly, the desired theory must be based on belief. However, Barker’s persuasive arguments that belief is too subjective to explain clarity shouldn't be ignored. Therein lies a dilemma: the account must be based on belief, yet 73 Sentence (17b) is, of course, much better in a context in which it echoes (and questions) a previous utterance. But note that no such context is necessary to make (17a) good. 145 maintain the sense of objectivity that is provided by Barker’s justification-based theory. In other words, we need to “objectivize” belief. How can this be done? To answer this question, let us look more closely into the meaning of belief. 4.4 Belief and probability 4.4.1 The Lottery Paradox As discussed in chapter 1, there are two types of belief – expressive belief is a measure of personal judgment and descriptive belief is a measure of the available evidence. The two types of belief are represented probabilistically. However, Barker, considers the possibility of accounting for clarity in terms of probability, and rejects it: "Certainly clarity does not involve degrees of probability, as shown by the standard lottery scenario (Kyburg 1961). If your chances of winning a lottery are inversely proportional to the total number of tickets, I cannot assert You will lose (at least, I can’t assert it on the basis of knowledge), no matter how many other lottery tickets have been sold. Nor can I appropriately say It is clear that you will lose, even if you and I both believe that you will lose" (Barker, 2009 : 264). Firstly, note that Barker's claim holds for adjectival clarity but not adverbial, as it seems more appropriate to assert: (19) You will clearly lose. Therefore, the problem is with descriptive probability and not expressive 74. Indeed, any theory that treats belief as a degree of probability needs to face up to Kyburg’s lottery paradox. Suppose there are n tickets in a fair lottery, where n is a large number. Then, the statement that ticket #1 will lose is extremely probable, hence can reasonably be believed. The same holds for the statement that ticket #2 will lose, and so on, all the 74 Note that this is a deviation from Wolf & Cohen (2011), in which the claim that clearly is expressive and clear is descriptive is not made. 146 way to ticket #n. However, the conjunction of all those beliefs is equivalent to saying that no ticket will win, and this, of course, is false and cannot rationally be believed. This paradox is usually taken to indicate the difference between belief and acceptance. While belief is graded, and one can believe a proposition to a high or low degree, acceptance of a proposition is all-or-nothing, and while dependent on belief, is not determined by it. Taking this perspective, the lottery paradox is not without a solution. In fact, several well known solutions to the paradox have been proposed. This chapter is not the place to debate the relative merits of these solutions, but whichever solution is adopted, one should be able to apply it to a belief-based theory of clarity, hence overcoming Barker’s objection. For example, according to several popular approaches to the paradox (e.g., Pollock 1995; Ryan 1996; Douven 2002) acceptance is defeasible: a high probability statement is accepted unless it is defeated by some other proposition (or set of propositions) that is at least equally probable. In the case of the paradox this means that the belief that ticket #i will lose, while highly probable, will be defeated by the equally probable propositions that other tickets will lose and the (non-defeasible) knowledge that one ticket will win; consequently, this statement will not be accepted. Under the plausible assumption that assertion presupposes acceptance, and, in particular, assertion of adjectival clarity presupposes acceptance, it follows that, indeed, both You will lose and It is clear that you will lose will be unassertable. Having a high level of belief, then, is a necessary condition for asserting adjectival clarity75, but not a sufficient one. It is clear that  may be unassertable if is not accepted: if there is a belief ’ that is incompatible with  but at least as likely as it. Note that when the statement is not defeated, its high probability is sufficient for acceptability, and clarity is assertable, as attested on the Web:76 75 In the original Wolf & Cohen (2011) having a high level of belief is a necessary condition for asserting all types of clarity, not just adjectival. 147 (20) Stop Losing Money at the Casino and Instantly Discover PROVEN WINNING Formulas… If you don’t have this knowledge, it is clear that you will lose. In this context, the sentence is a habitual: it does not say that the hearer will lose on a specific occasion, but rather that the hearer will lose in general. As is well known, habituals allow exceptions: thus, the sentence allows for the possibility of an occasional win. Therefore, the statement is not defeated: it is not contradictory to say that all gamblers will habitually lose, even if on every lottery some gambler is guaranteed to win. Consequently, (20) is assertable, in fact true: since the odds are always in favor of the house, the probability that if you don’t use the “proven winning formulas” you will habitually lose is quite high.77 4.4.2 Probability and normality Recall that Barker’s theory relies crucially on the notion of a “normal” world. One of the problems with normality-based theories is that it is notoriously hard to define what counts as normal. Consequently, it is very hard to know what the predictions of a normalitybased theory are, in order to confirm or refute it.78 Still, it is possible, following the literature on normality, try to make some reasonable assumptions. Barker makes it clear that normality and high probability are distinct. Granted, what is normal is usually also probable, and what is probable is usually also normal. But the two can diverge, and something can be normal yet unlikely, or vice versa. 76 From http://www.casinobooksonline.com/. Of course, the sentence also has an implicature that if you do use these formulas, you will not lose. We will not comment on the truth of this implicature… 78 See Cohen (1999, pp. 20-24) for more on this and other arguments against a normality-based account of generics. 77 148 For example, consider how Barker accounts for the lottery scenario. He argues that: "in the lottery example, It is clear that you will lose is correctly predicted to be false.79 The reason is that that world in which you improbably win the lottery is just as normal as the many worlds in which you do not win. It follows that there is no way for there to be only losing worlds above the clarity threshold" (Barker, 2009: 269) The same ought to apply to (20): the worlds in which you do not lose are much less likely than worlds in which you do, but just as normal. Hence, the sentence would wrongly be predicted to be false. Suppose, for another example, that two students submitted identical exams. It is, in fact, possible that this was a coincidence. But while coincidence is, by definition, improbable, it is not abnormal. A normal world, in which both students worked in perfectly normal ways on their exams, does not become less normal if, by coincidence, their answers happen to be identical. Thus, Barker’s approach would predict that (21) is false. (21) Cleary/It is clear that you cheated on the exam. According to our judgment, however, (21) is true, and we would be quite justified in saying it to the students. One of the characteristics of normal worlds that is usually agreed on is that, in such worlds, everything happens as it should. For example, machines perform as directed. Indeed, Krifka et al (1995), who favorably discuss a normality-based approach to another phenomenon (genericity), interpret sentences involving a machine “with respect to a modal base and ordering source, where the machine performs the action for which it was designed” (p. 54). There are cases, however, when, based on our past experience, it is very likely that a machine will not function in such a `normal' manner. In such cases, Barker would predict 79 In fact, our judgment is that the sentence is inappropriate rather than simply false, providing further evidence that it constitutes a case of presupposition failure. 149 that we cannot use clearly, yet such use is attested. Consider the following passage from a post on The DVD Forum:80 (22) Bitterly disappointed—my Xbox has started crashing quite regularly. I’m not sure what to do now—this is the 2nd Xbox I'll have to send back in 10 months… It will clearly go wrong again. Although it is not a normal thing for machines to break down regularly, the speaker believes, based on past experiences, that it is likely. Of course, one might claim that, in this context, worlds in which machines break down somehow are normal. But this would be tantamount to stripping the notion of normality of any meaning: if anything can be considered normal, then a theory of normality has no predictive power. 4.4.3 Probability and logic In our formalization, we need to incorporate probability into our logical formalism. We use a variation on Halpern's (1990) logic, which combines first order logic with probability. A probability structure is a tuple <D,W,,F>. D is a domain and W is a set of possible worlds.  is a valuation function such that for each world wW, (w) assigns to the symbols of the language appropriate extensions. F={f1, f2,...} is a set of discrete probability functions over W. We introduce distinguished propositional functions Pi(), whose intended interpretation is the probability of  as judged by i. A probability judgment without an index defaults to the probability judgment of the speaker. Formally, for any proposition , set of worlds W, model M, world w and assignment function g: (23) ║Pi()║M,w,g = fi ({wW | ( M,w,g) ⊨ }) 80 http://www.thedvdforums.com/forums/showthread.php?t=451805&page=22 150 The next section presents an account of clarity that is based on the theory presented in the previous chapters. 4.5 The interpretation of clarity 4.5.1 Simple clarity 4.5.1.1 A second mixture model As recalled from chapter 1, conversational participants assign probability values to propositions after performing a cognitive process that takes into consideration various sources of evidence pertaining to said propositions, which is represented by a mixture model. This section discusses how this already established process may be applied again, this time incorporating discourse participants as sources of evidence pertaining to the clarity of propositions. Consider again (2), repeated below: (24) It is clear that Abby is a doctor. What does the speaker who utters this sentence mean? We have argued that she is reporting a probable belief that is not defeated by a belief that is more or equally probable. But Barker has shown that this cannot be the speaker's belief, or the belief shared by the speaker and hearer. Whose belief is reported, then? We argue that the speaker is referring to the beliefs of people with sound judgment, people who are knowledgeable about the matter, who can reason rationally, and whose judgment is to be trusted. Thus, when someone utters (24), she is saying that people with sound judgment would come to believe, on the basis of the available evidence, that Abby is a doctor. Note that although what the speaker is reporting is a belief, it is not her subjective belief. In fact, it is not the subjective belief of any one person. In this sense, belief is “objectivized.” 151 Of course, if the speaker says that people with sound judgment believe something, she would usually believe it herself too. It would normally be odd for someone to acknowledge that people whose judgment she trusts believe P, yet deny that she believes P herself. Such a statement would be odd, in fact quite similar to Moore’s paradox: (25) #It is clear that Abby is a doctor, but I don’t believe it. However, there are cases in which such utterances are acceptable. This is the case when people choose (not) to believe something despite their better judgment, while acknowledging the irrationality of such a belief (and in such cases the word still is often used, highlighting the contrast between the objectivized belief and the speaker's personal belief). The following sentences involve both adverbial and adjectival clarity: (26) a. It was clear that they were going to kill him, but we still didn’t believe it. b. I'm such a fool. She's clearly not into me, but I still believe I have a chance. c. I realize I'm a big fuddy-duddy. This young woman is clearly the right material for our college, but I still have my doubts. To further demonstrate this point, here is an attested example:81 (27) It was clear that emotions were mounting, but every feeling was denied and every intervention considered incorrect, negated or ignored. We account for Barker's (4) and (5), repeated below, in the same way. (28) It is reasonably clear that Mars is barren of life. (29) It is clear that God exists. The speaker who utters (28) may personally believe that there is life on Mars, while acknowledging that the experts deny this. Conversely, someone who denies (29) may believe in God, without necessarily believing that people who reason rationally would conclude, on the basis of the evidence, that God exists. In the former case, the probability 81 From: Oremland, J.D., Windholz, E. (1971). `Some Specific Transference, Counter transference and Supervisory Problems in the Analysis of a Narcissistic Personality’ Int. J. Psycho-Anal., 52:267-275. 152 value of the mixture will be higher than d(clear), while the strength of belief of the speaker in the proposition that Mars is barren of life may be lower than HIGH.82 In the latter case, the value of the mixture will be lower than d(clear), while the strength of belief of the speaker in the proposition that God exists may be higher than HIGH. We formalize the notion of “good reasoners” using a second mixture model. We define a probability mixture over the judgments of possible individual reasoners. Each individual reasoner 1jm is assigned a weight, vj, indicating how good a reasoner he or she is. And we define Pjustification to be the weighted sum of these individual probabilities: Pjustification(  ) =  vjPj(  ) m (30) j=1 Of course, the probability Pjustification then participates in the personal mixture model process, helping the hearer decide whether or not to accept . Note that (30) considers not only good reasoners, but also bad ones, although the weights associated with bad reasoners will be small, hence their effect on the resulting probability will also be small. However, in a context in which there are many bad reasoners, their combined judgments may be significant. Consider the following attested example: (31) It is clear that a strictly monotonically increasing function is one-to-one. In the context of the mathematics text from which it is taken,83(31) is true: the readers of the book are assumed to be versed in mathematical reasoning, and for them this mathematical statement is, presumably, clear. However, in most other contexts, (31) would be false, because reasoners who are able to affirm the statement are few and far between; although each one of them would be associated with a relatively high weight, the non-mathematically oriented reasoners, although associated with low weights, will win by sheer numbers. 82 As Paul Egré (pc) points out to us, this discrepancy between the speaker’s belief and the beliefs of good reasoners is probably indicated by the word reasonably, for without it (4) would become considerably worse. 83 S. G. Krantz (2009) A Guide to Real Variables, Mathematical Association of America. 153 Clear 4.5.1.2 Using Pjustification, we can now represent the assertion of (2) as follows: (32) Ax P (Pjustification(doctor(a)) ≥ d(clear)) ≥ high This formula means that it is justified to conclude that Abby is a doctor (i.e. “good” reasoners would conclude that Abby is a doctor), and the speaker believes to a high degree that it is justified to conclude that Abby is a doctor. As discussed above, in addition there is a presupposition that the belief is not defeated: there is no proposition ’ Pjustification(’)Pjustification(doctor(a)). that is incompatible with doctor(a) and Note that, as desired, (32) does not entail that Abby is a doctor: the good reasoners, however good they are, may be wrong. However, even though there is no entailment, there is a strong feeling that if (2) is true, then Abby is a doctor. Indeed, as we have seen above, the speaker wouldn’t normally express clarity if she knew that the proposition in the scope of clarity is false, or may be false. In this, clarity behaves in a way that is similar to factives. This is not to say that the predicate clear is a factive—in fact, Barker argues that it is not. He discusses the following examples84: (33) a. We know that p, though p might be false. b. We believe that p, though p might be false. c. It is clear that p, though p might be false. "Thus for a factive such as know, [(33a)] is internally inconsistent. In contrast, in [(33b)] we can allow that our belief might be mistaken without any inconsistency. Clarity 84 Example number (3) in Barker (2009). 154 behaves more like belief in this respect than like knowledge, as expected on the belief theory of clarity" (Barker 2009: 255).85 However, although we agree with Barker that (33c) is not contradictory, it is certainly not a common thing to say. And (34), where p is denied, is downright odd: (34) #It is clear that p, though p is false. We suggest that this is so because if it is justified to conclude something, i.e. if the experts conclude something, it is very likely (though not necessarily) true. In other ways, clarity actually does pattern with factive verbs. 86 For example, it is well established that factives like know can take interrogative complements, while nonfactives, like believe, cannot: (35) a. I know who the murderer is. b. I know whether the murderer is John. (36) a. *I believe who the murderer is. b. *I believe whether the murderer is John. Note that clear behaves like a factive, in that it, too, allows an interrogative complement: (37) a. It’s clear who the murderer is. b. It’s clear whether the murderer is John. Thus, while clear is not a factive, it is, in many ways, similar to factives. In fact, Égré (2008) defines a category of predicates he calls veridicals,87 which subsumes both factives and clear. 85 Since he argues against the belief theory of clarity, Barker then accounts for these facts in a different way: the standard for clarity can be low enough so as to make the sentence non-contradictory. 86 We are indebted to Hubert Truckenbrodt (pc) for this observation. 155 4.5.1.3 Publicly available evidence Sentences in the form of (34) become even worse if they rely on evidence that is not publicly available: (38) ##It was clear that I was in jail, when in fact I had broken out of jail. Although it is quite possible that the best minds thought that the speaker was still in jail, this was because the relevant evidence was kept secret, and was not publicly available. Recall from section 4.2 that Barker argues that clarity must refer to publicly available evidence, and we have proposed that this requirement is a presupposition rather than part of the truth conditions of clarity. We can now say more about the origins of this presupposition. Since we use a mixture model of a number of reasoners, they cannot be assumed to rely on private evidence, which would be accessible, at most, to only one of them. Consequently, we propose that clarity carries a presupposition to the effect that the relevant evidence is publicly available. This presupposition explains why (38) is completely out. Note that Barker’s theory makes the wrong predictions concerning (38): it follows from it that the sentence is acceptable and true. This is because for Barker, the requirement of publicly available evidence is part of the truth conditions of clarity, rather than a presupposition: (38) is predicted to say that in all normal worlds compatible with publicly available evidence, the speaker is in jail. But, if people do not realize that the speaker escaped from jail, then in all worlds (hence in all normal worlds) consistent with publicly available evidence, the speaker is in jail, and (38) ought to be unproblematically true. 87 Egré defines veridicals to be predicates that entail their complements in positive contexts; thus, according to him, (2) does entail that Abby is a doctor. 156 4.5.1.4 Clearly Following Piñon’s (2006) account of modal adverbs, we suggest that clearly modifies the strength, rather than content, of the assertion. Hence, the assertion of (39a) can be represented as (39b): (39) a. Abby is clearly a doctor. b. Ax Pjustification (doctor(a)) ≥ d(clear) In prose, the speaker asserts that Abby is a doctor based on the degree of strength that surpasses the standard of clarity, i.e. that it is justified to conclude that Abby is a doctor. It is important to stress that even though Pjustification is an intersubjective mixture model, it is used expressively here88. The nature of this expressiveness stems from clearly not asserted, i.e. not part of the descriptive propositional content. Thus, assertions of adverbial clarity, just like assertions of modal adverbs and expressive modal auxiliaries, are expressions of the degree of strength by which the assertion is performed, only here this degree relies on the sources of evidence that compose the mixture model of Pjustification. An important point regarding expressive EM is that the expressiveness is derived from the speaker's degree of belief in the asserted proposition, which is a component of the illocutionary act, specifically the sincerity condition of assertion. However, there is also a perlocutionary effect (cf. Austin, 1962) on the hearer. The effect is incorporated into the hearer's cognitive process of evaluating the asserted proposition in light of the hearer's previously existing beliefs, various sources of evidence pertaining to the proposition which the speaker is added to, and the speaker's degree of assertion. Hence, while degrees of assertion are based on belief, they become by virtue of use something more than just degrees of belief – they become the more general degrees of strength. The difference between these two notions can be made clear by the following example: A few friends are supposed to meet at a certain time. While everyone shows up as planned, John arrives late. John says: 88 This is a departure from Wolf & Cohen (2011) in which adverbial clarity is not expressive. 157 (40) I might be late for the meeting, but I brought cookies89. The EM might can't be descriptive, as the truth of the matter is that John is late, not that it is possible that he is. Thus, by virtue of Grice's (1975) quantity, the descriptive reading is infelicitous. The EM modal can be expressive, but in this case the degree of strength can't be a degree of belief, since in this scenario John knows as much as anybody else that he is late, so he can't sincerely assert that he is late with a low degree of belief. But John can utilize the perlocutionary effect of expressive might as an assertion performed with a low degree of strength, to lessen the effect of the first conjunct of his utterance. The representation of (40) is: (41) Ax P( late(x, meeting)) > 0 & Ax P( bring (x, cookies)) ≥ high This is a conjunction of speech acts rather than propositions (cf. Cohen & Krifka, 2011 on speech-act conjunction). The first speech act utilizes the expressive EM as a device for the speaker to admit something with a low degree of strength, thus lowering the chances of it being accepted for context update. Following Barker, we require that (39) not entail that Abby is a doctor. But we must proceed with care, since clearly modifies the entire speech act, in this case assertion, rather than just its propositional content. What does it mean for an assertion, rather than a proposition, to entail something? What would it mean for the assertion in (39) to entail the assertion in (42), whose representation is (43)? (42) Abby is a doctor. (43) Ax P(doctor(a)) ≥ high Note that since in our system assertions can be made at varying degrees of strength, an entailment relation between the propositional contents of two utterances is not sufficient for an entailment relation between the assertions. Consider the following example: (44) a. Abby is certainly a doctor. b. Abby is a doctor. 89 I'm thankful to Tyler Peterson (personal communication) for giving me the idea for this example. 158 c. Abby is presumably a doctor. Although the propositional contents of all three assertions are the same, intuitively (44c) does not entail (44b), which does not entail (44a). A useful notion in this context is illocutionary entailment, as defined by Searle and Vanderveken (1985): an illocutionary act A1 illocutionarily entails A2 if it is impossible to perform A1 without thereby performing A2. For example, a directive illocutionarily entails a suggestion, because it has the same content, but a higher degree of strength. But (44c) has a lower degree of strength than (44b), and, consequently, the former does not entail the latter. Illocutionary entailment does hold in the other direction, though, and (44a) illocutionary entails (44b) and (44c)90. Now our question becomes: does (39) illocutionarily entail (42)? The answer to this question is no: (39) does not necessarily have a higher degree of strength than (42). If good reasoners believe that Abby is a doctor to a degree higher than d(clear), the speaker may believe this too, but does not have to. She certainly does not have to believe this to a degree greater than HIGH. Therefore, it is possible for the speaker to perform the assertion in (39), without thereby performing the assertion in (42). 4.5.2 Returning to personal clarity Let us now move to personal clarity: (45) It is clear to X that Abby is a doctor. It is tempting to say that (45) expresses nothing more than X’s subjective belief. We could say that X believes to a degree of at least d(clear) that Abby is a doctor, and the 90 There is a pragmatic implication, that will not be discussed here, which causes (44a) to seem not to illocutionary entail (44b), i.e. to be weaker. This is the same implication that causes must assertions to seem weaker than non-modalized assertion. However, as argued in von Fintel & Gillies (2010), must (and, as seen in this dissertation, certainly as well) only seems to be weaker, but is actually stronger than the nonmodified counterparts. 159 speaker believes to at least a high degree that X believes to a degree of at least d(clear) that Abby is a doctor: (46) Ay P (Px(doctor(a)) ≥ d(clear)) ≥ high Such an approach, however, would be incorrect. It turns out that not only simple clarity, but even personal clarity, must be “objectivized”. Just like with simple clarity, the evidence for a statement of personal clarity must also be publicly available. Suppose I have secret knowledge that Mother Teresa, whom everybody admires, is actually corrupt. It would still be strange to say (47a); in contrast, (47b) would be perfectly fine. (47) a. #It is (secretly) clear to me that Mother Theresa is a crook. b. It is (secretly) known to me that Mother Theresa is a crook. Moreover, just like simple clarity, we do not normally use personal clarity if we know that the proposition in the scope of clarity is false. Thus, even if Mary is convinced that the Earth is flat, it would be strange (other than by way of irony) to utter: (48) #It is clear to Mary that the Earth is flat.91 Indeed, just like simple clarity, personal clarity patterns with factives in that it allows interrogative complements: (49) a. It is clear to the police who the murderer is. b. It is clear to the police whether John is the murderer. Consequently, it makes sense to represent personal clarity, just like simple clarity, as a mixture model. The to-PP has the role of a speech act modifier.92 It modifies the sincerity conditions of the utterance, indicating that the belief that competent reasoners conclude In contrast, (i) is perfectly fine; but of course if  seems clear, it means that  is not clear. (i) It seems clear to Mary that the Earth is flat. 92 We are thankful to Manfred Krifka for suggesting this idea to us. 91 160 that Abby is a doctor is ascribed to X, rather than to the speaker. Thus, (45) receives the following representation: (50) Ay Px (Pjustification(doctor(a)) ≥ d(clear)) ≥ high Note that (50) is very similar to (32), the representation of simple clarity: it says that it is justified to conclude that Abby is a doctor, and that this is believed to a high degree; the difference is that in (32) this belief is ascribed to the speaker, whereas in (50) it is ascribed to X. This representation of personal clarity explains its objective feel, and why (47a) and (48) are odd. In expressions of personal clarity, X is also the one who assigns the weights to the reasoners indicating their competence since the mixture model probability is embedded under the speech act strength probability. Thus, when Alex and Bill disagree in (10), repeated below, their disagreement is over the assignment of these weights: Alex thinks that good reasoners should concentrate on Abby’s appearance, while Bill thinks that they ought to reach their conclusions based on her behavior. (51) Alex: It is clear to me that Abby is a doctor (because she is wearing a stethoscope). Bill: It is clear to me that she's not (because she has an unhealthy habit).93 Our theory can explain why personal clarity is possible with clear but not with clearly: as we have seen (cf. example (18)), the adverb clearly is already, in itself, a speech act modifier. We have also seen that one such modifier cannot be inside the scope of another: if the to-PP is also a speech act modifier, it cannot co-occur with clearly. Our view of personal clarity is reminiscent of Anand’s (2008) intuition concerning predicates of personal taste; he points out that faultless disagreement is possible even when an overt experiencer is indicated: 93 One implication of this view is that (2) and (ii) have the same meaning. (ii) It is clear to me that Abby is a doctor. We believe that this is, in fact, the case, and the difference is that (ii) is used to indicate contrast, much like the use of an overt pronoun in a pro-drop language (Enç 1986). 161 (52) a. X: This is tasty for me. Y: But you're wrong. It's disgusting. b. X: This is boring for me. Y: But you're wrong. It's not boring. Anand contrasts these examples with constructions that unambiguously express a perceptual experience, where faultless disagreement is impossible: (53) a. X: I like this. Y: #But you're wrong. It's disgusting. b. X: This bores me. Y: #But you're wrong. It's not boring. On the basis of such examples, Anand proposes that predicates of personal taste, both with and without overt experiencers, express the perception of “normal” perceivers. See chapter 5 for an alternative account of personal taste predicates. 4.6 The origins of the mixture model In this chapter we propose an account of clarity, both adjectival and adverbial, both simple and personal. Adjectival clarity modifies the propositional content of the utterance, by stating that the proposition is believed by “good” reasoners. Adverbial clarity does not modify the propositional content of the utterance, but rather its sincerity conditions: it indicates that the speaker believes that “good” reasoners would reach the conclusion expressed by the propositional content of the assertion. Both simple and personal clarity are expressions of 162 “objectivized” belief. The former ascribes this belief to the speaker, whereas the latter ascribes it to the person overtly indicated. Thus, the theory we present in this chapter is an analysis of one particular vague property, clarity, in terms of collective judgments of good reasoners. But why does this property have such a meaning? Why is its meaning dependent on the judgment of competent agents? We have seen in the previous section that there are grounds to believe that another class of vague predicate, namely predicates of personal taste, can be accounted for in similar ways. But this only pushes the question one step further: why do the meanings of these predicates involve a mixture model? We do not know the answer to this question, but there is one speculation that appears particularly promising. As discussed above, both clarity and predicates of personal taste are vague. Perhaps, then, vagueness itself requires the use of mixture models: perhaps any vague property ought to be handled in this way. There is at least one established theory of vagueness that makes a suggestion along these lines. Wright (1987) says: "For an object to be (definitely) red is for it to be the case that the opinion of each of a sufficient number of competent and attentive subjects…would be that it was red." (p. 244). This idea is formalized in terms of a mixture model of the judgments of competent and attentive subjects: if the value of the mixture passes a certain threshold, the object is red, tasty, or clear, otherwise it is not. If Wright’s theory is on the right track, there is nothing so special about clarity, and its meaning comes simply from its vagueness: for an object to be red is for competent observers to judge that it is red, and for a proposition to be clear is for competent reasoners to judge that it is true. The next chapter applies this idea to predicates of personal taste. 163 Chapter 594 Predicates of personal taste 5.1 Faultless disagreement 5.1.1 Basic intuitions The intuitive interpretation of predicates of personal taste (henceforth PPT) e.g. tasty, fun, interesting, is that they convey the speaker's point of view. Indeed, it would seem odd if predicates of personal taste are anything other than personal. Therefore, at a first glance the truth conditions of utterances such as ‎the following should be just as personal – (1) is true iff Suzy finds the cake tasty and (2) is true iff James finds the rollercoaster ride fun: (1) Suzy: This cake is tasty. (2) James: This rollercoaster ride is fun. Intuition, however, runs into trouble when trying to account for cases of disagreement (cf. Kölbel, 2003; Lasersohn, 2005): (3) Suzy: This cake is tasty. James: No, it's not! The intuitive account can't explain such disagreements, since there isn't any semantic contradiction between Suzy finding the cake tasty and James not finding it so. In other words, the intuitive account leaves James nothing to disagree about. Yet, as this type 94 This chapter is based on Wolf (2011). 164 of disagreement is prevalent in everyday talk, there obviously is something to disagree about. But what is it and how can it be accurately represented? Alongside the disagreement problem, there is another one – while Suzy and James are clearly disagreeing, this is not an ordinary type of disagreement about 'facts in the world'. For example, if Suzy and James disagree about whether the cake contains sugar, one of them speaks the truth and the other utters a falsehood. The person who is wrong is at fault and should mend his beliefs accordingly. However, in the type of dispute depicted above, both Suzy and James are fully entitled to their claim and none can (or should) be blamed for uttering a falsehood. This is the problem of faultlessness, which an adequate theory of PPT should account for as well. The need to account for cases of faultless disagreement motivates an approach that distinguishes between the content of an utterance and the assignment of truth values, which is Lasersohn's (2005) relativist theory of PPT, discussed in the next subsection. 5.1.2 Lasersohn 2005 Lasersohn's (2005) seminal paper presents a theory that extends Kaplan's (1989) framework. Kaplan distinguishes between the content and the character of utterances, when the latter is a function from contexts to contents and the former is a proposition, i.e. a function from world/time indices <w,t> to truth values. Lasersohn adds a third index, the judge, representing the individual whose taste determines the truth value for utterances containing PPT: (4) [|the cake is tasty|]w,t,j = 1 iff the cake is tasty at world w, time t, according to judge j. The introduction of a judge index solves the faultlessness problem – each conversational participant's claim is true, albeit with regards to a different judge, therefore no one is at fault. The disagreement is accounted for by contradictory 165 contents – there is no triple index of evaluation (world, time, judge) in which both contents are true, i.e. the content of the utterance 'the cake is not tasty' indeed contradicts the content of the utterance 'the cake is tasty' . Lasersohn's theory is at odds with a competing approach to PPT, contextualism. This approach, based on the same principles as the Kratzerian theory of modality discussed in previous chapters, holds that an asserted proposition may contain context dependent elements. These elements include modals, predicates of personal taste, and many others. Both approaches to PPT are similar in that both assume some context dependency is involved, and that this context dependency stems from perspectival information, i.e. someone's perspective determines the truth value for PPT. Therefore, both approaches employ some type of judge. The main difference between the approaches is that for contextualism the judge is an implicit argument which is part of the content, while for relativism the judge is an index of evaluation which comes into play after the content has been established and before this content is assigned a truth value. That is, Lasersohn's theory distinguishes between content (i.e. the bare proposition the cake is tasty which is not relativized to a judge) and the assignment of truth values, while contextualist theories do not (i.e. the proposition expressed already contains a judge). This distinction allows Lasersohn's theory to account for both faultlessness and disagreement, while contextualist theories can't account for disagreement. This is because in disputes such as (3), a contextualist representation of Suzy's utterance is the content [the cake is tasty for Suzy] and for James' utterance the representation is the content [the cake is tasty for James]. These contents do not contradict each other. The contextualism-relativism debate will not be further discussed in this chapter. For one, as shown in Stojanovic (2007), both approaches are semantically equivalent, i.e. putting faultless disagreement cases aside, it doesn’t matter whether an utterance containing implicit arguments such as PPT is evaluated for truth by a contextualist or a relativist semantic interpretation – if these interpretations are applied to the same indexical parameters at the same point of view, the truth conditions will be the same. Secondly, the basic premise in both approaches is that PPT are essentially subjective 166 (i.e. both account for PPT using some sort of judge, be it an index or an implicit argument). Therefore, the problems that will be discussed in the next section, which stem from this subjectivity, are equally challenging to both approaches95. 5.1.3 Problems of Lasersohn 2005 5.1.3.1 The pragmatic problem The default usage of PPT, according to Lasersohn's account, involves an autocentric stance, i.e. the speaker is the judge whose taste determines the truth value. This raises a pragmatic problem. Being a default case means that unless contextually specified otherwise, each conversational participant naturally expects the judge in a PPT utterance to be the speaker (Stojanovic, 2007). In our case, James would naturally take Suzy to be expressing her own taste, and vice versa. But if both conversational participants recognize that they are expressing an individual taste, why would they wish to argue? Note that when the judge is made explicit there is no actual dispute, as apparent by the infelicity of the following: (5) Suzy: This cake is tasty for me96. #John: No it's not, this cake is disgusting for me! The autocentric stance is not a viable notion. But perhaps a different stance is at play? Maybe the disagreement stems from both conversational participants taking each other's utterance to mean something other than a subjective point of view? Lasersohn suggests a different stance which can be used, an exocentric stance, in which the judge is someone other than the speaker. This type of stance is apparent in (but not exclusive to) utterances such as: 95 Except for the semantic problem, which is specific to Lasersohn's account. While the phrase "for me" is predominantly used in this type of examples, it is interesting to note that "to me" is more frequently used (331,000 vs. 55,000) as attested by 'google' search and might sound more natural to some readers. I will nonetheless follow common practice and use the former. 96 167 (6) Suzy: This cat food is tasty, because my cat can't get enough of it The judge in (6) is not Suzy, but the cat. We may even, quite reasonably, assume that Suzy has never tasted the cat food herself. What, then, about disputes involving exocentric stance? (7) Suzy: This cat food is tasty. James: No, it's not! An exocentric stance is not the default case, thus an argument may arise. However, in this case the argument boils down to a misunderstanding. Assuming Suzy uses an exocentric stance as in the previous example, James can either use an autocentric stance or an exocentric one in which either the aforementioned cat or some other individual is the judge. If James employs an autocentric stance, then the 'dispute' is that Suzy claims that the cat finds the cat food tasty and James claims that James doesn't find it tasty, which is not really a dispute but rather a misunderstanding. Once the different stance is resolved (for instance, once Suzy realizes that James has actually eaten the cat food and found it disgusting) Suzy and James will understand that there's no real argument going on, since it's not feasible to compare cat taste standards with human. As recalled, explaining both disagreement and faultlessness is a main desiderata of PPT theories, thus losing disagreement is a high price to pay. Moreover, if there is no disagreement, there is no reason to posit a judge index. If James employs an exocentric stance in which the judge is some individual other than Suzy's cat (for instance, if James uses his own cat as a judge) we will be left with the original pragmatic problem, since making this judge explicit again eliminates the dispute: (8) Suzy: This cat food is tasty for my cat 168 #James: No it's not, this cat food is disgusting for mine! Finally, if James employs an exocentric stance in which the judge is the same as Suzy's (i.e. Suzy's cat), Lasersohn's theory predicts that there is a real dispute but also that this dispute is not faultless. In this case, James' claim contradicts Suzy's claim since there is no world, time, judge index in which both claims are true. However, both of them can't be right at the same time – the cat either finds the cat food tasty, or it doesn't. It is only a matter of finding out whose claim is correct, and in that case one of the disputing parties will be wrong i.e. at fault. And again, explaining both disagreement and faultlessness is a main desiderata of PPT theories, thus losing faultlessness is a high price to pay. To conclude, the pragmatic problem is that it doesn't matter which judge the context assigns – the theory either predicts no disagreement or no faultlessness. 5.1.3.2 The semantic problem Disagreement in Lasersohn's theory is accounted for in the usual semantic sense, i.e. at the level of content - both conversational participants assert contradictory contents. And, since Lasersohn adds a judge to Kaplan's original indices, contradiction occurs when there is no world, time and judge index in which both contents are true. This raises a semantic problem, since it is perfectly possible to assert contradictory contents in a Kaplanian framework in such a way that doesn't constitute any disagreement. The following example (based on Recanati, 2007, slightly modified) serves to show this point: (9) Suzy (on Sunday morning): It is raining James (on Monday evening): It is not raining 169 Suzy asserts a content which is true at a certain world/time index <w, t>97. James asserts a content which is true at a different time, i.e. the world/time index is <w, t'>. The contents are contradictory since there is no index in which both contents are true. Yet, since Suzy's assertion is evaluated for truth at a different time index than James', (similarly to our original example (3), in which Suzy's assertion is evaluated for truth by a different judge) any dispute they might have concerning the state of rain (for instance, if Suzy's utterance was left as a voice message on James' cell phone and James mistakes this message to co-occur with his own temporal location) will be due to a misunderstanding and thus not constitute a genuine disagreement. Since both (9) and (3) are cases in which two utterances semantically contradict one another, and since (9) does not constitute a real disagreement, there is no support to Lasersohn's claim that disagreement is to be represented by contradictory contents in a Kaplanian framework. The semantic problem, therefore, is that the semantic mechanism employed by Lasersohn in order to explain cases of faultless disagreement cannot explain disagreement. 5.1.3.3 The logical problem As recalled, both main theories of PPT, relativism and contextualism, employ an implicit judge (whether as an index of evaluation or as a propositional unarticulated constituent). It is puzzling, therefore, that utterances involving PPT partake in logical inferences such as modus ponens98, whose validity is not affected by any judge. (10) Suzy: If the cake is tasty, I will buy two slices. James (the baker): The cake is tasty. Suzy: Then I will buy two slices. 97 Since these utterances do not contain PPT, there is no need for a judge index Based on Geach's (1965) criticism of emotivist meta-ethical theories with regards to claims that moral predicates such as right and wrong express the speaker's positive or negative emotional attitudes towards the prejacent. Theories that treat PPT utterances as containing an indexical (or implicit) judge are open to the same line of criticism. 98 170 When tasty is embedded in the antecedent of the conditional, it isn't relativized to any judge. However, James' reply contains an unembedded occurrence of tasty in which, by default assumption, the judge is James. How can Suzy draw any logical conclusion from these two premises? One option open for a relativist is that utterances containing embedded PPT convey a bare proposition, and the judge index ,indeed an index of evaluation, is specified only when evaluated (in Macfarlane's (2014) terms, assessment sensitive). Thus, the judge in Suzy's utterance is left open until assessed – and James' assertion provides this assessment. A similar example of assessment sensitivity is occurrence of PPT in interrogatives: (11) Suzy: Is the cake tasty? James: Yes, it is. Again, the judge of tasty can be said to be left unspecified, until assessed by James. In here it is even more obvious that Suzy intends the judge to be James, as the question is directed at him. But now we are left with the same problem we had when we started – why should Suzy take James' taste in matters of cakes to be anything other than a personal opinion? Why should it affect her modus ponens conclusion or provide her with any knowledge with regards of the tastiness of the cake for her? 5.1.4 Recanati (2007) Recanati (2007) accepts the basic ingredients of Lasersohn's theory but argues that the judge of PPT cannot be the speaker. Instead, utterances such as (1) should mean "the cake is tasty for us", when us is "the community to which the speaker and his audience belong"99 (Recanati 2007: 91). Recanati's theory is able to account for the pragmatic, semantic and logical problems discussed above: the pragmatic problem is accounted for since both discourse 99 Determined by contextual domain restriction on a universal quantifier. 171 participants use the same judge and therefore it is understandable that they are arguing. The semantic problem is accounted for since in this case semantic contradiction and disagreement are related to each other - both Suzy and James argue about the same thing, i.e. whether the cake is tasty for both of them. The logical problem is accounted for since there is no puzzle that an inference is valid when all premises use the same judge index. However, this theory suffers from problems of its own. The first of which is that it fails to account for faultlessness, since the judge is the same for both dispute participants. And, as discussed previously, accounting for faultlessness is a main desiderata for PPT theories, thus losing it is a high price to pay. The second problem arises from Recanati's distinction between utterances like (3) and utterances that make the judge explicit, such as "the cake is tasty for me". The latter is, by Recanati, a weaker claim since it's entailed by the community reading. This is the reason why when Suzy's assertion is challenged by James she can retreat to the weaker claim, thereby avoiding making a mistake100: (12) Suzy: This cake is tasty James: No, it's not! Suzy: I meant, this cake is tasty for me. Note that negation scopes over the implicit “for us”, interpreted as a universal quantifier in this case. Otherwise, James' utterance can have the reading in which for both James and Suzy - the cake is not tasty, which is undesired since it disregards Suzy's original assertion. That is, if Suzy's original assertion means 'the cake is tasty for us' James can object on grounds of his own taste (i.e. the cake is not tasty for James, therefore it is not true that the cake is tasty for both James and Suzy) but James can't include Suzy's taste in his objection. Thus a narrow scope negation is out. The wide scope reading, in which it is not the case that the cake is tasty for both Suzy and 100 According to Recanati Suzy's first assertion is surely a mistake since James is part of us, therefore the reply in (12) automatically turns the original assertion false. 172 James, makes this dialogue felicitous and consistent. Suzy's second assertion (i.e. the weaker claim that the cake is tasty for her alone) is consistent as well. However, by these lights, (13) should be felicitous since it is a conjunction of 'it is not the case that the cake is tasty for us' with the weaker claim 'but it is tasty for me'. But in fact (13) is as infelicitous as (14) (predicted to be infelicitous by Recanati's theory), which is a conjunction of the strong claim 'the cake is not tasty for me' (entailing 'not tasty for us') with the claim 'the cake is tasty for us': (13) #This cake is not tasty, but it is tasty for me. (14) #This cake is not tasty for me, but it is tasty. 5.2 Objectivized predicates The previous sections have shown that a judge index (whether in a relativist or contextualist framework) cannot solve the PPT puzzles and cannot adequately account for faultless disagreement. This section presents a theory that solves these puzzles via treating PPT as objectivized predicates. 5.2.1 Cohen (2010) – how to "objectivize" a subjective judgment PPT are evaluative predicates. Therefore, an existing theory that accounts for evaluatives may come in handy. The case of Clear, discussed in the previous chapter provides such a theory. Clarity shares the following properties with PPT and other evaluatives (cf. Cohen, 2010): A. Gradability: (15) a. It is very/reasonably/rather clear that Abby is a doctor. b. The cake is very/reasonably/rather tasty. 173 c. The play is very/reasonably/rather interesting. B. The use of comparatives and superlatives: (16) a. Clearer, clearest. b. Tastier, tastiest. c. More interesting, most interesting. C. Modification by an overt experiencer: (17) a. It is clear to me/to you/to John that Abby is a doctor. b. The cake is tasty for me/for you/for John c. The play is interesting for me/for you/for John D. Most importantly, faultless disagreement: (18) Suzy: It is clear that Abby is a doctor. James: No it's not! In light of the similarities and the fact that both clear and PPT are evaluatives, Cohen (2010) offers a treatment of PPT along the lines of clarity– tasty is an objectivized predicate, whose truth conditions depend on the opinions of various individuals, specifically those that the speaker considers to be good evaluators of taste. To treat PPT as objectivized predicates is to agree with Recanati's (2007) observation that community plays an important role in the determination of PPT's truth conditions while solving the problems that were posed previously. 174 5.2.2 A formal account of objectivized predicates The formalization is along the lines presented in chapter 1, i.e. a probabilistic speechact based account. As discussed in chapter 4, the notion of 'good reasoners' is represented by using a probabilistic mixture model, which is defined over the judgments of possible individual reasoners. Each individual reasoner 1< i < n is assigned a weight, wi, indicating how good a reasoner he or she is. Pmixture is the weighted sum of these individual probabilities: (19) P( )   wiPi ( ) n i 1 The sum of all weights is one, thus Pmixture is a probability function. Cohen (2010) notices the parallel between clarity and PPT and provides a formal account of PPT in terms of the mixture model – the idea is that each individual i contributes to the mixture model a personal degree of belief with regards to tasty. This degree is a probability measure representing the subjective probability value that individual i assigns to the proposition. These individuals are then assigned weights wi, indicating their perceived reliability in the eyes of the speaker. If an individual (including the speaker herself) is considered to be non-reliable in matters of taste, her weight will be low and if an individual is considered an expert (such as a known connoisseur) in matters of taste her weight will be high. The final probability value is then computed as the weighted sum of the probabilities that were assigned: (20) Pmixture(tasty (cake)) =  wi  Pi (tasty (cake)) n i=1 To preserve the context-dependency facet of tasty, a vague threshold is employed. The threshold of tasty is associated with a delineation function (Lewis, 1970) d(tasty), which returns a standard of taste that varies according to context. In some contexts (say, a meeting of fine-cuisine critics) this standard will have to be very high, while in 175 other contexts (a meeting of friends at a local hotdog stand) this standard can be lower. Then, "the cake is tasty" is true iff the value of the mixture model is greater than d(tasty): w n (21) i i=1 Pi(tasty (cake))  d (tasty ) 5.3 Accounting for puzzles and problems 5.3.1 Faultless disagreement The original discussion between Suzy and James is repeated here: (22) Suzy: This cake is tasty James: No, it's not! Suzy and James are in disagreement with each other. This disagreement is semantically represented as contradictory content: w n (23) i i=1 w n (24) i Pi(tasty (cake))  d (tasty ) Pi(tasty (cake))  d (tasty ) i=1 In prose, Suzy asserts the propositional content 'good evaluators of taste will conclude that the cake is tasty' and James asserts the content 'good evaluators of taste will not conclude that the cake is tasty'. Faultlessness is accounted for as well, since the composition of the mixture model can be assigned different weights by different speakers, hence both dispute participants can't be blamed to reach different conclusions and none of them is at fault. 176 It is important to note that the above formulas fit in the context update theory presented in previous chapters. Specifically, the asserted contents here, just like the asserted contents discussed previously, are just one part of the assertion operator. Thus, the full representation of Suzy's assertion is: (25) ASuzy P (  wi  Pi(tasty (cake))  d (tasty ) ) ≥ high n i=1 In order to keep things simple, the full representation is henceforth omitted. 5.3.2 Logical inference Repeating (10): (26) Suzy: If the cake is tasty, I will buy two slices. James: The cake is tasty. Suzy states that if good evaluators of taste will conclude that the cake is tasty, she will buy two slices. James asserts that good evaluators of taste will conclude so. Suzy and James' definition of what counts as 'good evaluators' may differ, and therefore the manner by which they reach the conclusion that the cake is tasty is different, but – once a conclusion is reached, it is the same for both, i.e. the updated content is the same. For simplicity, let's assume that Suzy and James' conversation is one in which any assertion is accepted and updated into the common ground. Suzy asserts the conditional, thus the conditional is updated. James then asserts the simple PPT utterance, and it is updated as well. The formulas for Suzy's and James' assertions are respectably: (27) a.  wi  Pi(tasty (cake))  d (tasty )  buy (Suzy, 2slices) n i=1 177 w n b. i=1 i Pi(tasty (cake))  d (tasty ) Once James' assertion is updated, the common ground contains only information states in which it is true that good evaluators will conclude that the cake is tasty. Since this is the case, modus ponens is in effect and Suzy should buy two slices. Suzy may, of course, choose not to accept James' utterance, but this is a different conversational story explained by the conversational theory presented in this dissertation. 5.3.3 Recanati's problem Recall that Recanati's problem was that both of the following are infelicitous: (28) # This cake is not tasty, but it is tasty for me. (29) # This cake is not tasty for me, but it is tasty In the theory proposed here, when an overt experiencer is mentioned, no mixture model is involved and the probability value is fixed to that of the experiencer101. Therefore (28) is represented as (30) and (29) as (31): (30) Pmixture(tasty(cake)) < d(tasty) & Pspeaker(tasty(cake)) ≥ d(tasty) (31) Pspeaker(tasty(cake)) < d(tasty) & Pmixture(tasty(cake)) ≥ d(tasty) As can be seen in the above formulas, the claim in (28) is that while good evaluators of taste will not conclude that the cake is tasty, the speaker finds the cake to be tasty. The claim in (29) is that while the speaker finds the cake not to be tasty, a mixture model composed of good evaluators of taste will conclude otherwise. 101 But see Cohen (2010) for an alternative proposal. 178 The reason why both cases are infelicitous may stem from Grice's (1975) Maxim of Quality: "Contribute only what you know to be true. Do not say false things. Do not say things for which you lack evidence". Recall that once the threshold of the mixture model is surpassed (or not) the result is a standard non-probabilistic truth value, but this is not the case with the non-mixturemodel speaker-based assertion. I.e. the assertion with the mixture model is updated as 'the cake is tasty' (mixture model implicit) and the assertion corresponding to the speaker is updated as 'the cake is tasty for the speaker' (probability value explicit). Therefore, in the first case the speaker asserts one thing i.e. proposes to make 'the cake is not tasty' common ground, and then asserts that she personally believes otherwise, thus breaking the maxim of quality by asserting the first conjunct. In the second case the speaker asserts that she does not personally believe that the cake is tasty but then proposes to make 'the cake is tasty' common ground, again violating the maxim by asserting something she believes to be false. 179 Conclusion The central claim of this dissertation is that degrees of assertion, i.e. degrees of strength by which assertions are performed, are essential to the semantic and pragmatic research of many conversational phenomena. This is supported by data concerning modal adverbs, modal adjectives, and modal auxiliaries, the predicate clear and predicates of personal taste. I argue that these constructions grammaticalize a distinction between expressive and descriptive meaning, when the latter is propositional and the former operates at the speech act level and encodes degrees of assertion as part of the expressive dimension (cf. Potts, 2007) of the utterance. This distinction is represented by different types of probability. For example, the case of epistemic modals (EM) – expressive EM are represented by a probability distribution that corresponds to belief as subjective judgment, and descriptive EM are represented by a probability distribution that corresponds to belief as a measure of evidence. The formal account is also unified, as different types of EM are represented probabilistically by the same lexical root – all possibility EM (e.g. may, might, possibly, maybe) have the probability degree of 'greater than 0', all intermediate EM (e.g. probable, probably, likely) have the probability degree of 'greater than 0.5' and all necessity EM (e.g. must, certain, certainly) have the probability degree of 1. The conversational theory presented in this dissertation accounts for the different context update effects of expressive and descriptive meanings via a novel discourse element, the Negotiation Zone (NZ) and a novel assertion operator that distinguishes between the propositional content and the degree of strength of the assertion. A performance of an assertion by a speaker places the assertion, represented by the assertion operator, in the NZ pending acceptance or rejection. While the assertion is under negotiation, the hearer activates a mixture model that takes into consideration 180 various sources of evidence pertaining to the asserted proposition. One of these sources of evidence is the speaker, and the degree of strength by which the speaker asserts the proposition, together with the speaker's weight (a contextually determined degree that stands for the value of the speaker as a source of evidence pertaining to the asserted proposition) affects the result of the mixture model and thus the chances of the assertion to be accepted (or rejected, or neither). Assertions are performed with a default degree of strength of equal to or greater than high. Thus standard non-modalized assertions place an assertion operator that contains the asserted propositional content with a high degree of strength. Descriptively modified assertions place an assertion operator that contains the descriptively modified propositional content with a high degree of strength, and expressively modified assertions place an assertion operator that contains the asserted propositional content with the expressively modified degree of strength. Expressive and descriptive meanings have different conversational effects. In terms of context update potential, all descriptively modified assertions have the same strength as standard non-modalized assertion and thus have an equal chance of being accepted for context update, while expressively modified assertions have varied degrees of strength. Thus, for example, possibility expressive EM have a lower chance of being accepted for context update by the hearer than standard and descriptively modified assertions while necessity expressive EM have a higher chance. In terms of context update effect, all expressively modified assertions contribute the same propositional content as standard non-modalized assertion and thus have an equal effect on the context if they are accepted for context update, while descriptively modified assertions have varied propositional contents. Thus, for example, possibility descriptive EM have a weaker effect on the context than standard and expressively modified assertions since the set of probability spaces in which 'it is possible that p' is a superset of the set of probability spaces in which 'it is true that p'. An important point regarding expressive EM is that the expressiveness is derived from the speaker's degree of belief in the asserted proposition, which is a component 181 of the illocutionary act, specifically the sincerity condition of assertion. However, as discussed in chapter 4, section 4.5.1.4, there is also a perlocutionary effect (cf. Austin, 1962) on the hearer. The effect is incorporated into the hearer's cognitive process of evaluating the asserted proposition in light of the hearer's previously existing beliefs, various sources of evidence pertaining to the proposition which the speaker is added to, and the speaker's degree of assertion. Hence, while degrees of assertion are based on belief, they become by virtue of use something more than just degrees of belief – they become the more general degrees of strength. Once we see that it is possible for degrees of assertion to be degrees of strength rather than just degrees of belief, we can apply them to other modals and not just epistemic. As mentioned in chapter 3, root modality has been claimed to have subjective and objective readings (Tobin, 1996; Palmer, 1986; Kolodny & MacFarlane, 2007), thus, this is an option worth exploring in future research. This research direction may even shed light on the control-raising discussion for modals. As recalled from chapter 3, some theories (e.g. Hofmann, 1966; Brennan, 1993) claim that the syntactic composition of root modals is explained via a control structure while epistemic modals have a raising structure, while other theories (e.g. Bhatt 1998; Wurmbrand, 1999) show that many deontic modals have a raising structure and therefore claim that the root-raising distinction isn't valid for characterizing epistemic vs. deontic modals. It may turn out that the root-raising distinction is valid for characterizing expressive vs. descriptive modals. Another possible future research direction is the connection between epistemic modality and evidentiality. As mentioned in chapter 3, (cf. Portner, 2009, for a broad discussion of the subject), there is a debate in the literature regarding the connection between epistemic modals and evidentials. Some theories claim that the nature of the connection is that evidentials are the more basic type and EM are derived from them (cf. Westmoreland 1995; Drubig, 2001), while other theories claim that it is the other way around (Matthewson et al. 2006; McCready & Ogata, 2007). The theory proposed in this dissertation treats both types of EM as encoding sources of evidence, 182 as part of the mixture model process. Thus, it follows from this theory that evidentials can be truth conditional (corresponding to Matthewson et al. 2006 and McCready & Ogata, 2007) if they are used descriptively, and can also be non-truth-conditional (corresponding to Faller, 2002; Westmoreland 1995 and Drubig, 2001) if they are used expressively. Adding the expressive-descriptive distinction may benefit the discussion on the connection between epistemic modality and evidentiality. The theory proposed in this dissertation can also be further applied to deal with vagueness. As explored in chapters 4 and 5, the theory was applied to the vague predicate clear and to the vague predicates of personal taste. A proposal has been raised, by which the mixture model may be beneficial in dealing with other vague expressions, which is a matter of future research. Another area of research that the theory proposed in this dissertation can be applied to deal with is conditionals. As discussed in chapter 2, Cohen (2010) proposes a theory of conditional based on the distinction between propositional content and degrees of strength, in which conditionals are illocutionary operators that modify the strength of the sincerity condition of the speech act. That is to say, when a speaker asserts the conditional  >  , she asserts  with a sincerity condition that the belief in the probability P( | ) is high. The syntactic direction, explored in chapter 3, can also be further expanded. The phenomenon of A-reconstruction has been explored, and the account provided was both semantic and syntactic. Further research may shed more light on Areconstruction specifically, but also more generally on the syntax-semantics interface with relations to modality. In conclusion, I hope that in this dissertation I have been able to demonstrate that the conceptually simple idea, namely that one may assert a sentence with varying degrees of strength, has wide ranging applications to a variety of linguistic issues. 183 References Anand, P., & Brasoveanu, A. (2010). Modal concord as modal modification. In Proceedings of Sinn und Bedeutung Vol. 14 (pp. 19–36). Anand, P., & Hacquard, V. (2008). Epistemics with attitude. In Proceedings of SALT, Vol.18 (pp. 37–54). Anand, P., & Hacquard, V. (2013). Epistemics and Attitudes. Semantics and Pragmatics, 6(8), 1–59. Retrieved from http://dx.doi.org/10.3765/sp.6.8 Anscombe, G. E. M. (1957). Intention. Harvard University Press. Austin, J. L. (1962). How to do things with words. Oxford University Press. Bellert, I. (1977). On semantic and distributional properties of sentential adverbs. Linguistic Inquiry, 8(2), 337–351. Bhatt, R. (1998). Obligation and possession. In H. Harley (Ed.), Papers from the UPenn/MIT Roundtable on Argument Structure and Aspect, MITWPL 32 (pp. 21–40). Cambridge, MA. Braithwaite, R. B. (1932). The nature of believing. In Proceedings of the Aristotelian Society vol.32. London: Harrison and Sons. Brennan, V. (1993). Root and Epistemic modal auxiliary verbs. Ph.D. Dissertation. University of Massachusetts, Amherst. Carlson, L. (1982). Dialogue Games: An Approach to Discourse Analysis. MIT. Chierchia, G. (2004). Scalar implicatures, polarity phenomena and the syntax/pragmatics interface. In A. Belletti (Ed.), Structures and Beyond (pp. 39– 103). Oxford: Oxford University Press. Chomsky, N. (1995). Categories and transformations. In The Minimalist Program (pp. 219–394). Cambridge, Mass.: MIT Press. Cinque, G. (1999). Adverbs and functional heads: A cross-linguistic perspective. Oxford: Oxford University Press. Cohen, A. (1996). Think generic! The meaning and use of generic sentences. Carnegie Mellon University. 184 Cohen, A. (2009). Probability in Semantics. Language and Linguistics Compass, 3(1), 265–281. Cohen, A. (2010). Conditionals as illocutionary operators. In 6th International Symposium of Cognition, Logic and Communication Formal Semantics and Pragmatics: Discourse, Context, and Models. Riga. Cohen, A. (2012). Generics as Modals. Recherches Linguistiques de Vincennes, 1, 63–82. Cohen, A. (2013). Embedded Epistemic Modals: An Experimental and Theoretical Investigation. In The International Congress of Linguists (ICL). Geneva. Cohen, A. (2014). A Tasty Mixture. Ms. Ben-Gurion University of the Negev. Cohen, A., & Krifka, M. (2011). Superlative Quantifiers as Meta Speech Acts. In B. Partee, M. Glanzberg, & J. Skiltters (Eds.), The baltic yearbook of cognition, logic and communication (pp. 1–56). Cohen, L. J. (1989). An introduction to the philosophy of induction and probability. Oxford: Clarendon Press. Cormack, A., & Smith, N. (2002). Modals and negation in English. In S. Barbiers, F. Beukema, & W. van der Wurff (Eds.), Modality and its Interaction with the Verbal System (pp. 133–163). Amsterdam: John Benjamins. Davidson, D. (1967). The logical form of action sentences. Essays on Actions and Events, 5, 105–148. Davis, C. (2009). Decisions, dynamics and the Japanese particle yo. Journal of Semantics, 26(4), 329–366. Davis, C., Potts, C., & Speas, M. (2007). The pragmatic values of evidential sentences. Proceedings of SALT, 17(1976), 71–88. De Finetti, B. (1977). Probabilities of Probabilities: A Real Problem or a Misunderstanding? In A. Aykac & C. Brumat (Eds.), New Directions in the Application of Bayesian Methods (pp. 1–10). Amsterdam. Dekker, P. (1993). Transsentential Meditations, ups and downs in dynamic semantics. University of Amsterdam. Döring, S. (2012). The focus sensitivity of sentence adverbs. In Proceedings of ConSOLE XIX (pp. 201–214). 185 Drubig, H. (2001). On the syntactic form of epistemic modality. Ms., University of Tbingen. Ernst, T. (2009). Speaker-oriented adverbs. Natural Language & Linguistic Theory, 27(3), 497–544. Erteschik-Shir, N. (1997). The Dynamics of Focus Structure. Cambridge: Cambridge University Press. Faller, M. (2002). Semantics and Pragmatics of Evidentials in Cuzco Quechua. PhD thesis. Stanford University. Farkas, D. F., & Bruce, K. B. (2010). On reacting to assertions and polar questions. Journal of Semantics, 27(1), 81–118. von Fintel, K. (1997). Bare Plurals, bare Conditionals, and Only. Journal of Semantics, 14(1), 1–56. von Fintel, K., & Gillies, A. S. (2010). Must . . . stay . . . strong! Natural Language Semantics, 18(4), 351–383. von Fintel, K., & Gillies, A. S. (2011). Might made right. In A. Egan & B. Weatherson (Eds.), Epistemic Modality (pp. 108–130). Oxford: Oxford University Press. von Fintel, K., & Iatridou, S. (2003). Epistemic Containment. Linguistic Inquiry, 34, 173–198. von Fintel, K., & Iatridou, S. (2004). Seminar Notes. Ms. Frege, G. (1956). The thought: A logical inquiry. Mind, 65(259), 289–311. Gajewski, J. (2005). Neg-Raising: Presupposition and polarity. Doctoral dissertation. MIT. Geach, P. (1965). Assertion. Philosophical Review, 74, 449–465. Geurts, B. (2007). Really fucking brilliant. Theoretical Linguistics, 33, 209–214. Geurts, B., & Huitink, J. (2006). Modal concord. In P. Dekker & H. Zeijlstra (Eds.), Concord Phenomena and the Syntax Semantics Interface, ESSLLI. Malaga. Giannakidou, A. (1999). Affective dependencies. Linguistics and Philosophy, 22, 367–421. 186 Gillies, D. (2000). Philosophical Theories of Probability. London: Routledge. Ginzburg, J. (1996a). Dynamics and the semantics of dialogue. In J. Seligman & D. Westerstahl (Eds.), Logic, language and computation. Stanford: CSLI. Ginzburg, J. (1996b). Interrogatives: Questions, facts and dialogue. In The handbook of contemporary semantic theory. Oxford: Blackwell. Grice, H. P. (1975). Logic and conversation. In P. Cole & J. L. Morgan (Eds.), Syntax and Semantics, volume 3: Speech acts (pp. 225–242). New York: Seminar Press. Groenendijk, J., & Stokhof, M. (1975). Modality and conversational information. Theoretical Linguistics, 2, 61–112. Groenendijk, J., & Stokhof, M. (1990). Partitioning Logical Space, 02(August). Groenendijk, J., Stokhof, M., & Veltman, F. (1995). Coreference and contextually restricted quantification. In Proceedings of SALT vol.5 (pp. 112–129). Gunlogson, C. (2013). True to form: Rising and falling declaratives as questions in English. Routledge. Gutzmann, D. (2012). Use-conditional meaning. Studies in multidimensional semantics. Doctoral dissertation. University of Frankfurt. Hacking, I. (2001). An Introduction to Probability and Inductive Logic. History & Philosophy of Logic (Vol. 24, p. xvii, 302 p.). Cambridge University Press. Hacquard, V. (2006). Aspects of Modality. Massachusetts Institute of Technology. PhD dissertation. Hacquard, V. (2010). On the event relativity of modal auxiliaries. Natural Language Semantics, 18(1), 79–114. Hacquard, V. (2011). Modality. In C. Maienborn, K. von Heusinger, & P. Portner (Eds.), Semantics: an international handbook of natural language meaning (pp. 1484–1515). Berlin: Mouton de Gruyter. Hacquard, V., & Wellwood, A. (2012). Embedding epistemic modals in English. Semantics and Pragmatics, 5(4), 1–29. Hajek, A. (2011). Interpretations of Probability. Stanford Encyclopedia of Philosophy, (Dec 19, 2011). Retrieved from http://plato.stanford.edu/entries/probabilityinterpret/ 187 Halliday, M. (1970). Functional diversity in language as seen from a consideration of modality and mood in English. Foundations of Language, 6, 322–361. Halpern, J. Y., & Fagin, R. (1992). Two views of belief: belief as generalized probability and belief as evidence. Artificial Intelligence, 54(3), 275–317. Heim, I. (1983). File change semantics and the familiarity theory of definiteness. In R. Bäuerle, C. Schwarze, & A. von Stechow (Eds.), Meaning, Use and Interpretation of Language (pp. 164–189). Berlin: de Gruyter. Hintikka, J. (1961). Modality and quantification. Theoria, 27, 119–128. Hofmann, R. (1966). Past tense replacement and the modal system. (Harvard University Computation Lab, Report 17). Cambridge, MA. Horn, L. (1978). Remarks on Neg-raising. In P. Cole (Ed.), Syntax and Semantics 9: Pragmatics (pp. 129–220). New York: Academic Press. Horn, L. (1996). Exclusive company: only and the dynamics of vertical inference. Journal of Semantics, 13, 1–40. Huitink, J. (2008). Scoping over epistemics in English and in Dutch. In Current issues in unity and diversity of languages. (pp. 2077–2089). Linguistic Society Korea. Huitink, J. (2012). Modal Concord: A Case Study of Dutch. Journal of Semantics, 29(3), 403–437. doi:10.1093/jos/ffr012 Iatridou, S., & Sichel, I. (2011). Negative DPs, A-Movement, and Scope Diminishment. Linguistic Inquiry, 42, 595–629. doi:10.1162/LING_a_00062 Iatridou, S., & Zeijlstra, H. (2009). On the scopal interaction of negation and deontic modals. In M. Aloni & E. Al. (Eds.), Proceedings of Amsterdam Colloquium 2009 (pp. 315–324). Jackendoff, R. (1972). Semantic Interpretation in Generative Grammar. Studies in linguistics series 2 (Vol. R.). Cambridge, MA: The MIT Press. Jacobson, S. (1978). On the use, meaning, and syntax of English preverbal adverbs. Acta Universitatis Stockholmiensis. Stockholm Studies in English, (44), 5–160. Kadmon, N., & Landman, F. (1993). Any. Linguistics and Philosophy, 16, 353–422. Kamp, H. (1981). A theory of truth and semantic representation. In J. Groenendijk, T. Janssen, & M. Stokhof (Eds.), Formal Methods in the Study of Language (pp. 277–322). Amsterdam: Mathematisch Centrum. 188 Kaplan, D. (1989). Demonstratives. In J. Almog, J. Perry, & H. Wettstein (Eds.), Themes from Kaplan (pp. 481–563). Oxford University Press. Kaplan, D. (1999). The meaning of ouch and oops: Explorations in the theory of meaning as use. Manuscript. Kennedy, C. (2007). Vagueness and Grammar: The Semantics of Relative and Absolute Gradable Adjectives. Linguistics and Philosophy, 30(1), 1–45. Kölbel, M. (2003). Faultless Disagreement. Proceedings of the Aristotelian Society, 104, 53–73. Kolodny, N., & MacFarlane, J. (2007). Ought: Between Objective and Subjective. Unpublished Manuscript, 1–40. Kratzer, A. (1977). What “Must” and “Can” Must and Can Mean. Linguistics and Philosophy, 1, 337–355. Kratzer, A. (1981). The notional category of modality. In H.-J. Eikmeyer & H. Rieser (Eds.), Words, Worlds, and Contexts: New Approaches in Word Semantics (pp. 38–74). Berlin: Walter de Gruyter. Kratzer, A. (1991). Modality. In A. von Stechow & D. Wunderlich (Eds.), Semantik/Semantics: An International Handbook of Contemporary Research (pp. 639–50). Berlin: de Gruyter. Kratzer, A. (2012). Modals and conditionals: New and revised perspectives. Oxford: Oxford University Press. Krifka, M. (2001). Quantification into question acts. Natural Language Semantics, 9, 1–40. Kripke, S. A. (1963). Semantical analysis of modal logic I: Normal modal propositional calculi. Zeitschrift Für Mathematische Logik Und Grundlagen Der Mathematik, 9, 67–96. Kyburg, H. J. (1988). Higher order probabilities and intervals. International Journal of Approximate Reasoning, 2(3), 195–209. Lackey, J. (2007). Norms of assertion. Noûs, 41(4), 594–626. Landman, F. (1984). Data semantics for attitude reports. Logique et Analyse, 106, 165–192. 189 Lasersohn, P. (2005). Context dependence, disagreement and predicates of personal taste. Linguistics and Philosophy, 28, 643–686. Lasnik, H. (1999). Chains of arguments. In S. D. Epstein & N. Hornstein (Eds.), Working minimalism (pp. 189–215). Cambridge, MA: MIT Press. Lassiter, D. (2011a). Gradable epistemic modals, probability, and scale structure. In N. Li & D. Lutz (Eds.), Semantics and Linguistic Theory (SALT) (Vol. ZZ, pp. 197–215). Lassiter, D. (2011b). Measurement and Modality: The Scalar Basis of Modal Semantics. New York University Dissertation. Lewis, D. (1970). General semantics. Synthese, 12, 18–67. Lewis, D. (1976). Probabilities of Conditionals and Conditional Probabilities. The Philosophical Review, LXXXV, 297–315. Lewis, D. (1979). Scorekeeping in a language game. Journal of Philosophical Logic, 8(1), 339–359. Lin, S.-Y. (2012). On asymmetric epistemic interference. Retrieved from http://ling.auf.net/lingbuzz/001508 Linebarger, C. M. (1987). Negative polarity and grammatical representation. Linguistics and Philosophy, 10(3), 325–387. Lyons, J. (1977). Semantics. Cambridge: Cambridge University Press. Macfarlane, J. (2010). What is assertion? In J. Brown & H. Cappelen (Eds.), Assertion - new philosophical essays (pp. 79–96). Oxford and New York: Oxford University Press. MacFarlane, J. (2011). Epistemic Modals Are Assessment-Sensitive. In A. Egan & B. Weatherson (Eds.), Epistemic Modality (pp. 144–178). Oxford University Press. Macfarlane, J. (2014). Assessment Sensitivity: Relative Truth and Its Applications. Oxford: Oxford University Press. Matthewson, L., Rullmann, H., & Davis, H. (2008). Evidentials as epistemic modals: Evidence from St’át’imcets. In J. Van Craenebroeck & J. Rooryck (Eds.), The Linguistic Variation Yearbook, 7 (pp. 201–254). Amsterdam: John Benjamins. May, R. (1977). The grammar of quantification. Doctoral dissertation. MIT, Cambridge, Mass. 190 McCready, E., & Ogata, N. (2007). Evidentiality, modality and probability. Linguistics and Philosophy, 30(2), 147–206. McLachlan, G., & Peel, D. (2000). Finite Mixture Models. New York: Wiley. Mellor, D. H. (2005). Probability: A Philosophical Introduction. New York: Routledge. Moore, G. E. (1942). A reply to my critics. In P. A. Schlipp (Ed.), The Philosophy of G. E. Moore. Evanston: Northwestern University. Nilsen, Ø. (2004). Domains for adverbs. Lingua, 114(6), 809–847. Nuyts, J. (2001a). Epistemic Modality, Language, and Conceptualization: A cognitive-pragmatic perspective. Human Cognitive Processing. John Benjamins. Nuyts, J. (2001b). Subjectivity as an evidential dimension in epistemic modal expressions. Journal of Pragmatics, 33, 383–400. Palmer, F. R. (1986). Mood and Modality. Summer Institute of Linguistics Inc (Vol. 5, p. 258). Cambridge: Cambridge University Press. Papafragou, A. (2006). Epistemic modality and truth conditions. Lingua, 116(10), 1688–1702. Pearl, J. (1987). Do we need higher-order probabilities and, if so, what do they mean? In AAAI Workshop on Uncertainty in AL Seattle, Washington (pp. 170–179). Perkins, M. R. (1983). Modal expressions in English. London: Pinter. Perlmutter, D. (1971). Deep and Surface Structure Constraints in Syntax. New York: Holt, Rinehart and Winston. Piñón, C. (2006). Modal adverbs again. In Honoring Anita Mittwoch on her 80th birthday, Syntax, Lexicon, and Event Structure, The Hebrew University of Jerusalem (pp. 1–6). Piñón, C. (2009). The thing about modal adverbs. Snippets, (20). Portner, P. (2009). Modality. Oxford: Oxford University Press. Potts, C. (2005). The Logic of Conventional Implictures (Oxford Stu.). Oxford: Oxford University Press. Potts, C. (2007). The expressive dimension. Theoretical Linguistics, 33, 165–198. 191 Recanati, F. (2007). Perspectival Thought : a Plea for (Moderate) Relativism. Oxford: Clarendon Press. Reinhart, T. (1981). Pragmatics and Linguistics: An Analysis of Sentence Topics. Philosophica, 27, 53–94. Rizzi, L. (1997). The fine structure of the left periphery. In L. Haegeman (Ed.), Elements of Grammar (pp. 281–337). Dordrecht: Kluwer. Roberts, C. (1996). Information structure in discourse: Towards an integrated formal theory of pragmatics. In H. Yoon & A. Kathol (Eds.), Working Papers in Linguistics-Ohio State University Department of Linguistics (pp. 91–136). Ross, J. (1969). Auxiliaries as main verbs. In W. Todd (Ed.), Studies in Philosophical Linguistics (Series 1) (pp. 77–102). Evanston: Great Expectations Press. Searle, J. R. (1969). Speech Acts: An Essay in the Philosophy of Language. East (Vol. 0, p. 203). Searle, J. R. (1976). A classification of illocutionary acts. Language in Society, 5, 1– 23. Searle, J. R., & Vanderveken, D. (1985). Foundations of Illocutionary Logic. Cambridge: Cambridge University Press. Skyrms, B. (1980). Causal Necessity. New Haven: Yale University Press. Stalnaker, R. (1968). A theory of Conditionals. In N. Rescher (Ed.), Studies in Logical theory (pp. 98–112). Oxford: Blackwell. Stalnaker, R. (1978). Assertion. Cole, Peter, Ed.: Pragmatics. Syntax and Semantics . New York: Academic Press, ., 9, 315–332. Stenius, E. (1967). Mood and language-game. Synthese, 17, 254–274. Stojanovic, I. (2007). Talking about taste: Disagreement, implicit arguments, and relative truth. Linguistics and Philosophy, 30, 691–706. Tancredi, C. (2007). A Multi-Model Modal Theory of I-Semantics. Part I: Modals. MS. University of Tokyo. Retrieved from http://semarch.linguistics.fas.nyu.edu/Archive/WFkYjU1Y/Multi-Model Modal Theory.pdf Tobin, Y. (1996). Must vs. Have (got) to: a Sign-Oriented Approach. In E. Borochovsky & Y. Tobin (Eds.), Proceedings of the Eleventh and Twelfth 192 Annual Meetings of the Societatis Linguisticae Europaeae Sodaicium Israelense. XIX-XXVI. Traugott, E. C. (1989). On the Rise of Epistemic Meanings in English: An Example of Subjectification in Semantic Change. Language, 65, 31–55. Traugott, E. C., & Dasher, R. B. (2002). Regularity in Semantic Change. Cambridge Studies in Linguistics (p. 362). Van Rooij, R. (2003). Negative polarity items in questions: Strength as relevance. Journal of Semantics, 20(3), 239–273. Van Rooij, R. (2006). Attitudes and changing contexts. Springer Netherlands. Vanderveken, D. (1990). Meaning and speech acts. Volume 1: Principles of language use. Cambridge: Cambridge University Press. Vanderveken, D. (1991). Meaning and speech acts. Volume 2: Formal semantics of success and satisfacton. Cambridge: Cambridge University Press. Veltman, F. (1996). Defaults in Update Semantics. Journal of Philosophical Logic, 25, 221–261. Wang, P. (1993). Belief revision in probability theory. In Proceedings of the Ninth international conference on Uncertainty in artificial intelligence. Morgan Kaufmann Publishers. Westmoreland, R. (1998). Information and Intonation in Natural Language Modality. Ph.D. Dissertation. Indiana University. Wittgenstein, L. (1953). Philosophical Investigations. New York: Macmillan. Wolf, L. (2011). An Interpersonal Approach to Predicates of Personal Taste. In Y. Falk (Ed.), Proceedings of the Israel Association of Theoretical Linguistics 26. Wolf, L., & Cohen, A. (2009). Modal Adverbs as Negotiation Chips. International Journal for Language Data Processing, Sprache Und Datenverabeitung., 33(12), 169–177. Wolf, L., & Cohen, A. (2011). Clarity as objectivized belief. In P. Egre & N. Klinedinst (Eds.), Vagueness and Language Use (Palgrave s., pp. 165–190). Basingstoke: Palgrave Macmillan. 193 Wolf, L., & Spector Shirtz, I. (2012). A-Reconstruction Puzzles from an ExpressiveDescriptive Perspective. In 9th Workshop on Syntax and Semantics. University of Siena. Wurmbrand, S. (1999). Modal verbs must be raising verbs. In S. Bird, A. Carnie, J. Haugen, & P. Norquest (Eds.), Proceedings of West Coast Conference in Formal Linguistics 18 (pp. 599–612). Somerville, MA: Cascadilla Press. Wurmbrand, S., & Bobaljik, J. (1999). Modals, Raising and A-Reconstruction. In Colloquium Leiden University. Leiden. Yalcin, S. (2010). Probability Operators. Philosophy Compass, 5, 916–937. Yalcin, S. (2012). Bayesian Expressivism. In Proceedings of the Aristotelian Society (Hardback) (Vol. 112, pp. 123–160). Blackwell Publishing Ltd. Zeijlstra, H. (2007). Modal concord. In Proceedings of SALT Vol. 17 (pp. 317–332). 194