University of Southern California Dissertations and Theses

What "ought" ought to mean

(USC Thesis Other)

What "ought" ought to mean

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content What “Ought” Ought to Mean by Junhyo Lee A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (PHILOSOPHY) May 2022 Copyright 2022 Junhyo Lee ii Acknowledgments During my time at USC, the philosophy department has been an invaluable source of advice, motivation, inspiration, and friendship. Over the years, my advisors and colleagues have contributed to my work so many ways that I can’t possibly account for everything they have done. I’d like to first thank my committee. It has been a privilege to work with each of them, and I greatly appreciate all the time and effort they have given me. Above all, I wish to thank Mark Schroeder. A significant part of my dissertation directly engages with Mark’s work on the semantics of ‘ought’. The first chapter of the dissertation began as a final paper in his seminar. Also, every time I’ve stumbled, Mark has been there to help me get back on my feet. Stephen Finlay has always believed in me and taken my work seriously, no matter how jumbled it looked. Without his support, I would never have been able to take this project as far as I have. It’s always fun to talk with Fabrizio Cariani. Everytime I meet with him, I feel I learn something new. Alexis Wellwood came to our department at the perfect time for me. She has provided extremely thoughtful comments on the drafts of almost every paper I have written. She always provides me with warm feedback, advice, and encouragement no matter what I work on. I am also indebted to Deniz Rudin for stepping in to help me as my outside committee member. Beyond the members of my committee, I’d like to thank Robin Jeshion, Barry Schein, and Scott Soames for giving me a foundation in my graduate courses, so that I could make the most out of graduate school. Jeffrey Russell was my first-year mentor, and Gabriel Uzquiano was the DGS of the philosophy department most of my time at USC. I am grateful to both of them for their guidance during my first few years in the program. I am also grateful to Ralph Wedgwood for his iii guidance and support when writing this dissertation. Discussions with him have significantly improved this project. I also have benefited immensely from my fellow graduate students at USC. I’ve grown so much as a philosopher in virtue of the many conversations I’ve had with my peers. Firstly, I am incredibly thankful for having the best cohort ever: Frank Hong, Paul Garofalo, Eleonore Neufeld, Sean Donahue, and Daniel Pallies. I feel lucky to have spent graduate school with them, and hope there are many more cohort dinners in our future. I owe particular thanks to Anthony Nguyen. He is not only an important part of my other big project on generics, but has also gone over most of my work with me. Without our conversations, I would definitely have taken much longer and achieved much less with my projects. I’m indebted as well to conversations with Zachary Goodsell, Weng Kin San, Yasha Sapir, Jamie Castillo-Gamboa, Betul Erbasi, Haley Wei, Irene Bosco, Megha Devraj, Antonio Maria Cleani, Laura Nicoara, Quyen Pham, Kenneth Silver, Shyam Nair, Wooram Lee, Nathan Howard, and probably others I’ve unintentionally left out. I have no doubt that there are still so many more people whose conversations in the grad lounge, in the courtyard, in seminars, at tea parties, and in Spec Society helped shape my work over the years. In the office, Natalie Schaad, Cynthia Lugo, Nikolai Costalas, and Amanda Velasco have always been there to answer my questions, guided me through paperwork mazes, and corrected my mistakes patiently. I’m grateful for that. Finally, I would like to thank my wonderful parents. All that I have ever accomplished has been thanks to you. Thank you and I’m sorry that I’ve been so far away for so long, but I hope you will be proud of what I’ve tried to accomplish during my time out here in the U.S. iv Table of Contents Acknowledgements ...................................................................................................................... ii Abstract ........................................................................................................................................ vi 1. “Ought” and Intensionality..................................................................................................... 1 1.1. Introduction ....................................................................................................................... 1 1.2. The intensionality test ....................................................................................................... 9 1.3. A challenge to the intensionality test ............................................................................... 12 1.4. Resistance strategies ......................................................................................................... 15 1.4.1. Eternalism .............................................................................................................. 15 1.4.2. Implicit modal operators ........................................................................................ 17 1.4.3. Implicit subjunctive morpheme ............................................................................. 19 1.5. Implications ...................................................................................................................... 21 1.5.1. “Ought” and propositions ….................................................................................. 21 1.5.2. “Ought” and “reason” ............................................................................................ 24 1.6. Conclusion ........................................................................................................................ 30 2. “Ought”, Agents, and Events ................................................................................................. 32 2.1. Introduction ...................................................................................................................... 32 2.2. Conflicting evidence ......................................................................................................... 34 2.2.1. The embedding argument ...................................................................................... 34 2.2.2. The passivization argument ................................................................................... 37 2.2.3. Chrisman’s and Lee’s linguistic tests .................................................................... 38 2.3. Reconciling the conflicting evidence ............................................................................... 42 2.3.1. Capturing the deliberative-evaluative distinction .................................................. 42 2.3.2. Solving the puzzle .................................................................................................. 45 2.4. Other linguistic tests ......................................................................................................... 49 2.5. Direct and indirect deliberative “ought”s ......................................................................... 51 2.6. Deliberative “ought” implies ability “can” ...................................................................... 54 2.7. Comparison ...................................................................................................................... 57 2.7.1. The stit account ...................................................................................................... 57 2.7.2. Chrisman’s practition-based account ..................................................................... 58 2.8. Conclusion ........................................................................................................................ 59 3. “Ought” and Intervals ........................................................................................................... 61 3.1. Introduction ...................................................................................................................... 61 3.2. Background ...................................................................................................................... 62 3.3. Interval modifiers ............................................................................................................. 63 3.4. Existing accounts .............................................................................................................. 65 3.4.1. Kratzer’s account ................................................................................................... 66 3.4.2. Cariani’s account ................................................................................................... 68 v 3.4.3. Probabilistic semantics ........................................................................................... 69 3.4.4. Decision-theoretic semantics ................................................................................. 71 3.5. A difference-based semantics for “ought” ....................................................................... 73 3.6. The unity of epistemic and deontic “ought”s ................................................................... 74 3.6.1. Two puzzles about “ought” .................................................................................... 75 3.6.2. Positivity and intermediacy .................................................................................... 76 3.7. Conclusion ........................................................................................................................ 79 References ................................................................................................................................... 80 vi Abstract This dissertation defends the uniformity of “ought” by developing a new analysis of deontic “ought”. Philosophers and linguists have paid particular attention to the meaning of “ought” for various reasons. For example, moral philosophers have been interested because they are concerned with the nature of obligations and practical deliberations, and “ought” traditionally has been used to express our moral obligations and practical deliberations. Semanticists have been interested because “ought” can be used to express such a wide range of interpretations: epistemic, prudential, deliberative, and evaluative, among others. Logicians have been interested because the logic of “ought” seems to change, depending on its interpretation. In my dissertation, I confront three of the most significant challenges to the classic view on “ought”-statements and show how we can solve those problems without losing the main advantages of the classic view. The classic view is committed to the following three theses: (i) “ought” uniformly expresses a propositional operator, (ii) “ought” is evaluated relative to possible worlds, and (iii) “ought” is associated with an ordinal scale. Each chapter of my dissertation is a self-standing paper and focuses on a different aspect of the classic view. The first two chapters of the dissertation concern what is sometimes called the distinction between “ought-to-do”(= deliberative) and “ought-to-be”(= evaluative) interpretations of “ought”. Compare, for example, (1) “John ought not to kill innocent people” to (2) “There ought to be world peace.” (1) typically expresses a deliberative reading and can be paraphrased as “John is obliged to, or has most reason to, not kill innocent people. (2) typically expresses an evaluative reading and can be paraphrased as “It’d be good for there to be world peace.” Some philosophers claim that this distinction requires us to posit an ambiguity in “ought” itself. But in Chapter 1 “‘Ought’ vii and Intensionality”, I develop a new test, which I call the intensionality test, to verify this claim. This test suggests that “ought” uniformly expresses a propositional operator. In Chapter 2 “‘Ought,’ Agents, and Events”, I consider the main arguments for the ambiguity of “ought”, one of which is that pairs like (3) “Bill ought to kiss Lucy” and (4) “It ought to be that Bill kisses Lucy” do not have the same range of readings. This is alleged to be a fatal obstacle for the propositional operator view of “ought” because (3) and (4) have truth-conditionally equivalent prejacents. Many responses to this argument have been offered in print, but all of them fail to appreciate its central flaw. This flaw is that they have failed to appreciate that despite their truth-conditional equivalence, the prejacents of (3) and (4) differ in their event structure. My solution suggests that “ought” is sensitive to events, not possible worlds. It therefore defends the first, most powerful, feature of the classic view by relaxing its second, less well-motivated feature. My view also sheds light on why the moral principle that “ought” implies “can” holds. In Chapter 3 “‘Ought’ and Intervals”, I tackle the question “What kind of scale is ‘ought’ associated with?” It’s generally accepted that there are at least four levels of measurement: nominal, ordinal, interval, and ratio. Most theories of “ought” claim that “ought” is associated with an ordinal scale. For example, on the classic semantics, we can ask whether one world is better than another. But we cannot ask how much one world is better than another. In this paper, I present some new pieces of data and argue that “ought” requires at least an interval scale. The scale to which “ought” is sensitive turns out to have a number of other very important applications such as why the logic of “ought” may change depending on its interpretation. To summarize, each chapter has the following implication on semantic theorizing of “ought”: Ch1: “Ought” always operates on propositions, not actions. viii Ch2: The contextual parameters for “ought” (e.g., conversational backgrounds in the classical semantics, alternative sets in the contrastive semantics) are sensitive to events, not worlds. Ch3: “Ought” requires at least an interval scale, not an ordinal scale. These implications may not determine a unique semantics of “ought” but put significant constraints on semantic theorizing of “ought”. But it would be important to make sure that there are at least some semantic theories that satisfy all the constraints above. For concreteness, let’s see how the account developed in chapter 3 can be adjusted to satisfy all those constraints. Let me start with the following two Kratzerian definitions of “ought”: (1) ⟦“Ought: ϕ”⟧ f,g,w is defined iff f is circumstantial or epistemic, If defined, = 1 iff "w' Î BEST(f, g, w). ⟦ϕ⟧ f,g,w' = 1. (Kratzer, 1981, 1991; Portner, 2009) (2) ⟦“Ought”⟧ is defined iff f is circumstantial or epistemic, If defined, = lf. lg. lϕ. lw. "w' Î BEST(f, g, w). ⟦ϕ⟧(w') = 1. (Heim & von Fintel, 2011) (3) ⟦“Ought”⟧ is defined iff f is circumstantial or epistemic, If defined, = lf. lg. lϕ. le. "w' Î BEST(f, g, e). ⟦ϕ⟧(w') = 1. (Hacquard, 2009, 2010) On the classical semantics, “ought” is evaluated relative to conversational backgrounds such as the modal base f, the ordering source g. There are two ways to implement them in a formal system. They can be represented as parameters at the index in Lewisian terms (Lewis, 1980) or the circumstance of evaluation in Kaplanian terms (Kaplan, 1989). (1) is an example where conversational backgrounds are represented as parameters. On the other hand, they can be represented as arguments at the level of object language. (2) is an example where they are represented as arguments of “ought”. Taking the second approach, Hacquard proposes replacing ix the world variable w with the event variable e, and reaches (3). However, (1) – (3) all employ an ordinal scale, so they don’t satisfy the last condition. According to the account developed in chapter 3, “ought” is evaluated relative to alternative sets. Just as conversational backgrounds can be represented as either parameters or arguments, alternative sets also can be represented as either of them. (4) ⟦“Ought: ϕ”⟧ ALT,w,c is defined iff µc is positive or intermediate, If defined, = 1 iff "ψÎALT(w). µc(ϕ) – µc(ψ) > θc. (5) ⟦“Ought”⟧ c is defined iff µc is positive or intermediate, If defined, = lALT. lϕ. lw. "ψÎALT(w). µc(ϕ) – µc(ψ) > θc. (4) is an example where the set of alternatives ALT is represented as a parameter, and (5) is where it is represented as an argument of “ought”. By replacing the world variable w with the event variable e, we can arrive at (6). (6) ⟦“Ought”⟧ c is defined iff µc is positive or intermediate, If defined, = lALT. lϕ. le. "ψÎALT(e). µc(ϕ) – µc(ψ) > θc. According to (6), “ought” operates on propositions, ALT is a function of events, and “ought” is evaluated on an interval scale. This is because subtraction requires interval scales. Thus, (6) satisfies all the constraints discussed above. CHAPTER 1 “Ought” and Intensionality Abstract This paper is about the syntactic structures of deontic “ought”s. There are two debates on this topic. The first concerns whether epistemic and deontic “ought”s are associated with different syntactic structures. The second concerns whether two different senses of deontic “ought”s – “ought-to-do”s and “ought-to-be”s – are associated with different syntactic constructions. 1 This paper focuses on the second debate. Some philosophers and linguists have embraced what has been called the ambiguity view, according to which deontic “ought”s are ambiguous between a propositional operator and a relation between an agent and an action/proposition. 2 The main goal of this paper is to present some new negative evidence – what I call the intensionality test – against the ambiguity view. 1. Introduction Moral philosophers have been interested in the nature of obligations and practical deliberations. Since “ought” traditionally has been used to express moral obligations and practical deliberations, we might be able to get some insight about the nature of obligations and practical deliberations if we get a clearer understanding of the exact meaning of “ought.” 1 For the “ought to do”/“ought to be” distinction, see Chisholm (1964), Geach (1982), Jackson (1985), Feldman (1986), Brennan (1993), and Bhatt (1998). Different theories have used different terminology for this distinction. For example, this distinction corresponds to Humberstone’s (1971) agent-implicating and situational “ought”s, Wedgwood’s (2006) practical and political “ought”s, Price’s (2008) agential and situational “ought”s, and Schroeder’s (2011) deliberative and evaluative “ought”s, etc. 2 See Harman (1973), Geach (1982), Brennan (1993), Price (2008), Schroeder (2011), Broome (2013), and Wedgwood (2006). 1 Moral philosophers have, at least tacitly, assumed that at least some sense of “ought” is closely connected to moral obligations and practical deliberations. However, there is one potential problem with this approach. That is, “ought” can be used to express a wide range of interpretations. So, we need to first settle what sense of “ought” is the one that matters to moral philosophers. For example, Schroeder (2011) distinguishes between deliberative (= “ought-to-do”) and evaluative (= “ought-to-be”) senses of “ought” and claims that the deliberative sense of “ought” is the one that is intimately related to practical deliberations and obligations. Following Harman (1973), Chrisman (2012) introduces a more fine- grained distinction among normative senses of “ought” (i.e., evaluative, prudential, and ethical “ought”s) and claims that prudential and ethical “ought”s are particularly important to the projects of moral philosophers. The following sentences can be used to illustrate the different flavors of modality that “ought” can apparently contribute (Chrisman 2012; Harman 1973; Schroeder 2011): (1) a. Epistemic (or pseudo-epistemic): John ought to be home by now. 3 b. Bouletic: John ought to come to the party (because I like him). c. Teleological: In order to go to Santa Monica, John ought to take Train A. d. Evaluative: There ought to be world peace. e. Prudential: John ought to stop smoking. f. Ethical: John ought not to kill innocent people. I’ll treat (1a-c) as non-normative. I’ll focus on (1d-f), which express clearly normative readings of “ought.” The prudential “ought” roughly means that it is best for the agent’s 3 “Ought” is generally thought to take a true epistemic reading. But Yalcin (2016) challenges this commonly- held view. He argues that the seemingly epistemic “ought” expresses a normality reading and calls this reading a pseudo-epistemic reading. 2 interest that s/he performs some action, so (1e) can be paraphrased as “It is best for John’s interest that he stops smoking.” The ethical “ought” roughly means that the agent is obliged, or has most reason, to do some action. (1f) can be paraphrased as “John is obliged not to kill innocent people.” The evaluative “ought” roughly means that it would be good for a certain event to happen (or for a certain state to obtain). (1d) conveys something on a par with the claim that it would be good for there to be world peace. (1d) can be true even if it is practically impossible to obtain world peace. Among these readings, (1e-f) seems particularly important to moral philosophers. This is because (1e-f) are agential in the sense that they describe some action that the agent ought to do. On the other hand, (1d) is non-agential or situational in the sense that it just expresses a desirable state of affairs that ought to be the case. More specifically, Schroeder (2011) suggests five hallmarks of deliberative “ought”s (= “ought-to-do”s) and each reading can be examined in light of these hallmarks: (i) deliberative “ought”s matter directly for advice (MacFarlane & Kolodny 2010). When someone comes to you with the question of what to do, if you know what s/he deliberatively ought to do, you know what is advisable for him/her. (ii) Deliberative “ought”s close practical deliberation (Ross 2010). If you know what one deliberatively ought to do, then that settles the question of what to do. (iii) Deliberative “ought”s are closely related to the notion of accountability (Broome 2013). If an agent S deliberatively ought to do some action X, S is accountable if S fails to perform X. (iv) The deliberative “ought” implies “can.” (Moore 1922; Humberstone 1971; Wedgwood 2007) and (v) Deliberative “ought”s are closely related to the notion of obligation. 3 The ethical “ought” satisfies all of the five hallmarks. (1f) is appropriate for advice for John, settles John’s deliberative question, implies that John is accountable if he kills innocent people, implies that John has an ability to not kill innocent people, and can convey something on a par with the claim that John has an obligation to not kill innocent people. The prudential “ought” does not express an agent’s obligation, so it does not satisfy (v). (1e) has nothing to do with John’s moral obligations. But the prudential “ought” still satisfies (i)–(iv) and seems closely connected to practical deliberations. (1e) can be given as advice to John, settles John’s deliberative question, implies that John is accountable if he does not stop smoking, implies that John has an ability to stop smoking. In this paper, following Chrisman (2012), I will classify prudential and ethical “ought”s as deliberative “ought”s. Perhaps, Hallmarks (i)-(iv) that both prudential and ethical “ought”s share are more probative of deliberative “ought”s than Hallmark (v). If so, (i)-(iv) are hallmarks of deliberative “ought”s and (v) is just a hallmark of the ethical “ought.” On the most straightforward interpretation of Kratzer’s modal semantics (Kratzer 1977, 1981, 1991, 2012), which is a dominant paradigm in the semantics of modality, the deliberative–evaluative distinction is not lexically encoded. “Ought” is treated as a propositional operator. 4 Various flavors of “ought” are derived via what Kratzer calls 4 However, this doesn’t mean that the Kratzerian account is not extendable to control modal verbs. That is, even if “ought-to-do” modals turn out to be control verbs, there is a way to accommodate the control sense of “ought” in the Kratzerian framework (Fintel & Heim 2011; Kaufmann 2012). If “ought-to-do” modals are control verbs, they can be treated like propositional attitude verbs such as “want” and “believe.” Let B be the set of worlds compatible with what the subject x believes at the world w. Given (1), the lexical entry for “believe” can be defined like (2). (1) B = λxe. λws. {w¢: w¢ is compatible with what x believes in w} (2) ⟦“believe”⟧ = λp. λxe. λws. "wÎB(w)(x) [p(w) = 1] According to (2), “x believes that p” is true at w iff p is true at all worlds compatible with x’s beliefs at w. Similarly, let f be the set of worlds compatible with the subject x’s obligations/desires /purposes at w. Given (3), the lexical entry for the control “ought” can be defined like (4). (3) a. fdeontic = λxe. λws. {w¢: w¢ is compatible with x’s obligations in w} b. fbouletic = λxe. λws. {w¢: w¢ is compatible with x’s desires in w} 4 conversational backgrounds. The basic idea is that necessity operators like “ought: p” can be glossed as “In view of x, it is necessary that p.” Here the conversational background is implicit, and its value should be supplied by context. In this approach, each sentence in (1) can be glossed as follows, where the content of x in the “in view of”-phrase gives descriptive content to the intended modal flavor: (2) a. Epistemic: In view of what is known, it is necessary that John is home. b. Bouletic: In view of what is desired, it is necessary that John comes to the party. c. Teleological: In view of what is aimed at, it is necessary that John takes Train A. d. Evaluative: In view of what is good, it is necessary that there is world peace. e. Prudential: In view of what is good for John, it is necessary that he stops smoking. f. Ethical: In view of what is moral, it is necessary that John does not kill innocent people. On this picture, “ought” is considered uniformly raising. It would be useful for understanding the debate, then, to mention some linguistic background on the control-raising distinction (Jackendoff 1972; Bach 1979; Postal 1970, 1974; Chomsky 1981, 1982; Carnie 2013). Raising verbs express propositional operators and so take only one propositional argument. On the other hand, control verbs 5 express relations between c. fteleological = λxe. λws. {w¢: w¢ is compatible with x’s goals in w} (4) ⟦“oughtcontrol”⟧ = λf<e,>>. λp. λxe. λws. "wÎf(w)(x) [p(w) = 1] (4) is a simplified lexical entry for the control “ought” in the sense that it’s not doubly relativized. But it shouldn’t be difficult to implement this idea in doubly-relative modal semantics. The main difference between (2) and (4) is whether the domain accessible from the base world w is context-sensitive or not. Also, while the domain for the raising “ought” is a function of a world, the domain for the control “ought” is a function of an individual and a world. Compare (4) with (5), which is a simplified lexical entry for the raising “ought.” (5) ⟦“oughtraising”⟧ = λf>>. λp. λws. "wÎf(w) [p(w) = 1] 5 More specifically, we are interested in whether the deliberative “ought” is a subject control verb. There are two kinds of control verbs: subject control and object control verbs. Subject control verbs (e.g., “want,” “try,” “expect,” etc.) express two-place relations between an external agent and a property/proposition. Object 5 an agent and a proposition/property and so take two arguments: one for agents/experiencers and the other for propositions/properties. Here are some examples of control and raising verbs, along with sketches of their underlying syntactic structures (LFs) that are evocative of their distinct argument-taking properties: (3) a. John seems to/is likely to dance. (Raising) b. Seem/likely [John to dance]. (LF for (3a)) (4) a. John wants to/tries to dance. (Control) b. Johni wants/tries [PROi to dance] (LF for (4a)) “Seem” and “likely” are typical examples of raising verbs. “Want” and “try” are typical examples of control verbs. Raising and control constructions may look superficially similar, but they are structurally different. The subject in a raising construction is not the subject of a raising verb. They appear as the subject of the embedded verb and move up to a higher position. For example, “John” in (3a) is base-generated as the subject of “dance” and then moves up from the subject position of “dance” (= the specifier (Spec) of a verbal phrase (VP)) to the subject position of the tensed “seem” (= the specifier of a tense phrase (TP)) for syntactic reasons. 6 That is, all clauses must have grammatical subjects (in Spec TP) (= Chomsky’s (1982) Extended Projection Principle). On the other hand, the subject in a control construction is the subject of a control verb. In (4a), “John” is base-generated as the subject of “want,” not as the subject of “dance.” The subject of “dance” is saturated by an implicit pronoun called PRO, and “John” determines what this pronoun refers to. control verbs (e.g., “advise,” “order,” “tell,” etc.) express three-place relations between an external agent, an internal theme, and a property/proposition. 6 Strictly speaking, (3a) involves at least two movements. “John” begins in Spec VP where it receives the external theta-role of “dance.” And then it moves up through the embedded Spec TP to the matrix Spec TP. (Raising) [TP Johni seems [TP ti to [VP ti dance]]]. 6 This is why “John” in (4a) is interpreted as both the subject of “want” and the subject of “dance.” 7 As can be seen from (2), the orthodox view is that “ought” does not carry any argument for agents and so is uniformly raising. But Schroeder challenges this orthodox view. He argues that the deliberative “ought” has a control syntax while the evaluative “ought” has a raising syntax. 8 However, Chrisman (2012) provides some evidence against the ambiguity view of “ought” and argues that it is uniformly raising. He proposes two linguistic tests to examine the ambiguity view and shows that it fails the tests. If there is a control sense of “ought,” then this means that the control “ought” carries an external argument. 9 The arguments of a verb can be divided into two types. The first is called the external argument, and it is the argument place for subjects. The second is called the internal argument and is the argument place for direct and indirect objects. 10 The existence of the external argument makes some interesting predictions. As Chrisman (2012) noted, control verbs usually allow for 1) “er”-nominalization (e.g. “trier,” “wanter”), and 7 On the traditional account of control constructions, control constructions involve an implicit pronoun PRO. In this paper, I’ll present my arguments within the traditional framework that posits PRO. But on some alternative approach, which is usually called the movement theory of control (e.g., Hornstein 1999; Manzini & Roussou 2000), PRO doesn’t exist and control constructions are explained in terms of movement. However, even on this approach, there is a control-raising distinction. In control constructions, the subject receives two theta-roles: one from the main verb (e.g., “want” in (4)) and one from the embedded verb (e.g., “dance” in (4)). In raising constructions, the subject receives only one theta-role. That is, one from the embedded verb (e.g., “dance” in (3)). 8 This paper mainly concerns arguments against the ambiguity view. But Schroeder provides some positive arguments for the ambiguity view. These arguments will be briefly discussed later in this paper. 9 According to the traditional lexicalist approach to argument structures (Chomsky 1981; Baker 1988, 1997; Dowty 1989, 1990; Levin & Rappaport Hovav 1995), the information about the argument structure of a verb such as how many arguments and what kinds of arguments a verb has is encoded in the lexical meaning of a verb. In this paper, I formulate my arguments within this traditional framework. But it should be possible to re-formulate them in other frameworks as long as they make sense of the argument-modifier distinction. For some alternative approaches to argument structures, see Hale & Keyser (2002), Harley (1995), Borer (2005), and Ramchand (2008). 10 Alternatively, one can say that the external argument is not an argument of a verb, but an argument of another functional projection (Marantz 1984; Kratzer 1996). For how the external argument can be represented in LFs, see fn.15. 7 2) passivization (e.g. “it was tried,” “it was wanted”). 11 Thus, if there is a control sense of “ought,” it is expected to allow for these two linguistic processes. However, “ought” does not admit of either “er”-nominalization or passivization. That is, both “er"-nominalization (i.e. *“oughter”) and passivization (i.e. *“it is oughted”) lead to ungrammaticality. In response to Chrisman’s tests, proponents of the ambiguity view might appeal to an etymological explanation for this data. It is often mentioned that “ought” was derived from the past form of “own” (von Fintel & Iatridou 2008; Finlay & Snedegar 2014). The base form of “own” is subject to both “er”-nominalization and passivization (e.g., “owner,” “owned”). However, the past form of “own” is not subject to either “er”-nominalization or passivization (e.g., *“owned-er,” *“owned-ed”). The etymological explanation is also consistent with the fact that “ought” cannot be tensed (e.g., *“oughted”). Since the past forms of control verbs can still have external arguments while they are not subject to “er”- nominalization and passivization, one might claim that Chrisman’s observation that “ought” does not allow for “er”-nominalization and passivization should not be considered decisive evidence against the ambiguity view. 12 The etymological explanation might provide an easy way out for the ambiguity view, but it doesn’t help to avoid the result of the tests that I’ll propose in the next section. 11 “-er” can be thought of as picking out the external argument of a verb. Also, the passive morpheme “-en” is generally thought of as transforming the argument structure of a verb by deleting/absorbing the external agent theta role of an active verb (Baker, Johnson and Roberts 1989, Carnie 2006, Chomsky 1981, Jaeggli 1986). 12 It has often been pointed out (Chrisman 2012; Finlay and Snedegar 2014, ft.39) that the deliberative- evaluative distinction can be found not merely in “ought”-sentences but in all deontic modals including “should” and “have to.” If this is the case, the ambiguity view will need to be committed to the ambiguity in other deontic modals as well. However, it is not clear how the etymological explanation can generalize to other deontic modals such as “should” and “have to.” The base form of “should” is “shall,” but “shall” does not allow for er-nominalization (i.e., *“shaller”) and passivization (i.e., *“shalled”). If the base form “shall” does not carry any argument for agents, “should” also does not carry any argument for agents. Furthermore, since “have to” is simply not past-tensed, the etymological explanation cannot be applied to “have to.” 8 2. The Intensionality Test In this section, I propose a new linguistic test to verify the control sense of “ought.” If there is a control sense of “ought,” then it should carry an external argument. The external argument of a control verb is expected to be extensional. The external arguments (i.e., the subject positions) of control verbs (e.g., “want,” “believe,” etc.) are extensional in the sense that the existential commitments of names, pronouns, and existential quantifiers are active. 13,14 For example, let's suppose that “Newman” is the descriptive name for the first- born baby in the 22 nd century. Consider the following: (5) Obama wants to meet Newman. (6) #Newman wants to meet Obama. (5) can be true depending on Obama’s desires. But (6) is semantically anomalous because (6) is committed to the existence of Newman but Newman does not exist relative to the present time (or the utterance time). 15 Now, let’s see if the deliberative “ought” patterns like “want.” Let’s stipulate that “Newman C” refers to the baby that will be first-born in China in the 22 nd century and 13 What I mean by this is the extension of “exist.” “Exist” can be tensed (e.g., “existed,” “exists,” “will exist,” etc.), and the extension of “exist” varies depending on time. For example, (5) is present-tensed, so if (5) is true this implies that Obama exists at the present time. 14 Quine (1960) proposes another test for extensionality. That is, if a linguistic context is extensional, two co- referential terms can be substituted without change in truth-value. I’ll come back to this test in Section 3. 15 One possible explanation of this phenomenon is that the external argument is not an argument of the main verb, but an argument of another projection (Marantz 1984; Kratzer 1996). In this view, control verbs do not carry external arguments. Nonetheless, control verbs impose certain requirements on the argument structure of the clauses they appear in, so they can appear only in the structure in which the external argument is projected. In this view, the logical forms for (5) and (6) can be represented as follows: (5*) $e [Ag/Exp(e)=Obamai & Want(e, PROi meets Newman)] (6*) $e [Ag/Exp(e)=Newmani & Want(e, PROi meets Obama)] (5*) can be read as “there is some event e such that the agent/experiencer of e is Obama and e is the event of wanting to meet Newman.” (6*) can be read as “there is some event e such that the agent/experiencer of e is Newman and e is the event of wanting to meet Obama.” In (5*), “Newman” appears in the scope of the intensional operator “want.” On the other hand, in (6*), “Newman” appears outside of the scope of “want.” 9 “Newman A” refers to the baby that will be first born in the U.S in the 22 nd century. Let’s consider the following examples: (7) a. President Xi: Newman C ought to join the Communist Party of China (in order to get a good job in China). (Prudential “Ought”) b. President Xi: Newman C ought to protect and serve the Communist Party of China. (Ethical “Ought”) c. #Newman C wants/tries/expects/attempts/is reluctant/is eager to protect and serve the Communist Party of China. (8) a. President Trump: Newman A ought to stay at Trump Hotel for vacation (because I’ll give him a special discount). (Prudential “Ought”) b. President Trump: Newman A ought to make America great again. (Ethical “Ought”) c. #Newman A wants/tries/expects/attempts/is reluctant/is eager to make America great again. While (7c) is anomalous, (7a) and (7b) are fine. (7a) and (7b) seem to express deliberative readings. (7a), as a prudential “ought,” satisfies hallmarks (i)–(iv). If someone comes to President Xi with the question of what Newman C ought to do (to get a good job in China), he can give (7a) as advice and it will settle the issue of what Newman C (prudentially) ought to do. Newman C will be accountable if s/he doesn’t act according to Xi’s advice. Xi’s utterance presupposes that Newman C will have an ability to act according to his advice. If not, Xi’s utterance will count as meaningless or false. Furthermore, (7b) that expresses an ethical “ought” has all the five hallmarks of the deliberative “ought.” If someone comes to Xi with the question of what Newman C (morally) ought to do, he can 10 give (7b) as advice and it will settle the issue of what to do. Newman C will be accountable if s/he doesn’t act according to Xi’s advice. Xi’s utterance presupposes that Newman C will have an ability to act according to his advice. By uttering (7b), Xi can express what he thinks Newman C is morally required to do as Chinese. If deliberative “ought”s are control, the subject positions of (7a) and (7b) are predicted to be extensional, and so (7a) and (7b) should pattern like (6) and (7c). But they do not. 16 In a similar vein, (8a) and (8b) can be interpreted as expressing prudential and ethical “ought”s. If President Trump is asked to give advice on what Newman A ought to do on his vacation, Trump can give (8a) as advice and it will settle the issue of what Newman A (prudentially) ought to do. Newman A will be accountable if s/he doesn’t act according to Trump’s advice. Trump’s utterance presupposes that Newman A will have an ability to act according to his advice. Furthermore, (8b) seems to express an ethical “ought.” If someone comes to Trump with the question of what Newman A (morally) ought to do, he can give (8b) as advice and it will settle the issue of what to do. Newman A will be accountable if s/he doesn’t act according to Trump’s advice. Trump’s utterance presupposes that Newman A will have an ability to act according to his advice. By uttering (8b), Trump can express what he thinks Newman A is morally required to do as American. Thus, (8a) and (8b) express deliberative “ought”s. If deliberative “ought”s have the control syntax, the subject 16 One might present the following as a potential counterexample to the intensionality test. Suppose I am watching a group of schoolchildren and they all want to eat ice cream. Let "Fast" denote whichever of the children will first finish their assignment. It seems that "Fast wants to eat ice cream" sounds fine. Everyone wants to eat ice cream, so Fast (whoever s/he might be) wants to eat ice cream. However, the main reason this inference sounds fine is that it is presupposed that Fast exists now but we don’t know yet who s/he is. Suppose that “Newman” denotes the first-born baby in the 22 nd century, so s/he doesn’t exist now. Compare the above inference with the following: Everyone wants to eat ice cream, so #/?Newman wants to eat ice cream. This inference does not sound as good as the first one. This is because it is not presupposed that Newman exists now and so Newman cannot be included in the domain of “everyone.” 11 positions of (8a) and (8b) should be extensional and so (8a) and (8b) are expected to behave like (6) and (8c). But they do not. 3. A Challenge to the Intensionality Test In Section 2, I argued that deliberative “ought”s are not control because their subject positions are not extensional. I appealed to the fact that even when deliberative “ought”s are true, this doesn’t imply that their subjects exist at the present time (or the utterance time). One might challenge the validity of this test by claiming that there is an additional test for intensionality (= non-extensionality) according to which the subject positions of deliberative “ought”s are extensional. 17 One might claim that if a linguistic context is intensional, there must be two co-referential terms “S1” and “S2” such that “S1 ought to do ϕ” and “S2 ought to do ϕ” have distinct truth-values (Quine 1960). But we cannot find such co-referential terms. Therefore, the subject positions of “ought” are extensional. Let me call this the substitution test. In this section, I consider the substitution test and argue that since deliberative “ought”s pattern like intensional operators such as “necessary” and “possible” with respect to this test, deliberative “ought”s are intensional as well. The substitution test can be divided into two cases: one in which co-referential terms are rigid designators and one in which co-referential terms are non-rigid designators. When co-referential terms are rigid designators, the substitution test is not just a test for intensionality. Consider the following: (9) a. Lois Lane believes that Superman is Superman. (T) b. Lois Lane believes that Superman is Clark Kent. (F) 17 I thank an anonymous referee for bringing this objection to my attention. 12 (10) a. It is necessary that Superman is Superman. (T) b. It is necessary that Superman is Clark Kent. (T) I believe that there is no dispute that “necessary” is intensional. But there is still a contrast between (9) and (10). When two co-referential terms are rigid designators (e.g., “Superman” and “Clark Kent”), these two terms can be substituted without change in truth-value in (10), but not in (9). Philosophers have tried to capture this by saying “believe” is hyper- intensional and “necessary” is merely intensional. If an operator is hyper-intensional, it is intensional, but not vice versa. Thus, the contrast between (9) and (10) suggests that “believe” is stronger than “necessary” in that while “believe” is hyper-intensional, “necessary” is merely intensional. The considerations in Section 2 suggest that the deliberative “ought” is at least intensional. But this leaves open the question whether it is intensional or hyper-intensional. I’ll remain neutral on this issue. But if the following examples sound plausible, this might suggest that “ought” is hyper-intensional. (11) a. Superman ought to reveal his true identity. (T) b. Clark Kent ought to reveal his true identity. (?) (12) a. Superman can fly. (T) b. Clark Kent can fly. (?) However, even if (11) and (12) are not plausible, that doesn’t mean that the deliberative “ought” is extensional. This is because when co-referential terms are rigid designators, the substitution test is not just a test for intensionality but for hyper-intensionality. Let’s now consider the cases in which co-referential terms are non-rigid designators. When two co-referential terms are non-rigid designators, “necessary” passes the 13 substitution test. Given that Socrates is the greatest teacher of Plato, (13) is true but (14) is not. (13) It is necessary that Socrates is Socrates. (T) (14) It is necessary that Socrates is the greatest teacher of Plato. (F) “Ought” patterns like “necessary” in this respect. Suppose that Lois Lane met Superman at the Daily Planet yesterday and he happened to be the tallest man she met at that place. (15) Superman ought to fly and stop Lex Luthor from destroying the world. (16) The tallest man who Lois Lane met yesterday at the Daily Planet ought to fly and stop Lex Luthor from destroying the world. (16) has at least two readings: (i) one in which “the” takes scope over “ought” and (ii) one in which “ought” takes scope over “the.” Each one can be paraphrased as follows: (i) The > Ought: The tallest man who Lois Lane met yesterday at the Daily Planet at the actual world @ is such that for all deontically ideal worlds w' that is accessible from @, he flies and stops Lex Luthor from destroying the world at w'. (ii) Ought > The: For all deontically ideal worlds w' accessible from @, the tallest man who Lois Lane met yesterday at the Daily Planet at w' flies and stops Lex Luthor from destroying the world at w'. On the first reading, it is the tallest man Lois Lane met at the actual world who has an obligation to stop Lex Luthor. On the second reading, it is the tallest man Lois Lane met at the ideal worlds who has an obligation to stop Lex Luthor. We are interested in the second reading. Since (16) is false on the second reading, this shows that “ought” and “necessary” pattern alike with respect to the substitution test. If “necessary” is intensional, “ought” should be intensional, too. 14 4. Resistance Strategies In the previous section, I considered a potential challenge to the intensionality test. In this section, given that the intensionality test is a legitimate test for intensionality, I consider three possible ways to respond to the intensionality test and argue that none of them save the ambiguity view from the intensionality test. The first says that what the intensionality test shows is that a certain version of eternalism is true. The second says that “ought” comes with an independent operator functioning like “will” or “would.” The last says that “ought” itself consists of two intensional operators. 18 4.1. Eternalism Let’s start with the first response. One might claim that what (7) and (8) show is that a certain version of eternalism is true, according to which “Newman C” (or “Newman A”) does refer to a presently existing entity. That is, this view bites the bullet and accepts that Newman C (or Newman A) exists at the present time (or the utterance time). According to this view, Newman C (or Newman A) is now non-concrete and only in the 22nd century will be concrete. On this view, the contrast between “ought” and “want” might be explained as follows: (7c) (or (8c)) sounds bad because Newman C (or Newman A) is non-concrete, so s/he lacks desires. However, (7a) and (7b) (or (8a) and (8b)) sound fine because Newman C (or Newman A) as a non-concrete entity still can have obligations. 18 An anonymous referee mentioned the first strategy. And I developed the last two ideas in conversation with Schroeder (p.c.), though he is not officially committed to any of them. 15 I’d like to point out three things about this particular version of eternalism. First, eternalism seems clearly not appropriate to capture our linguistic judgments about sentences like the following: (17) Everything always exists. (F) (18) Newman C exists now. (F) (19) Dodos (or Socrates) once existed but no longer exist. (T) (20) People are concrete objects. (T) (17) and (18) are intuitively false, but eternalism should say that they are true. (19) sounds true, but eternalism should say that it is false because its second part is false. (20) sounds almost analytically true, but eternalism should say that it is false because people can exist as non-concrete objects. Secondly, there is a more general worry. It seems reasonable to think that an individual exists at some possible worlds but not at other worlds. Similarly, it seems reasonable to think that an individual exists at some times but not at other times. If one says every object exists at all possible worlds, it will collapse the contingency/necessity distinction. Similarly, eternalism seems to collapse the temporal contingency/temporal necessity distinction. Lastly, it seems that eternalism doesn’t fare well with the ethical principle that “ought” in the sense of practical deliberation/obligation implies “can” in the sense of ability. If Newman as a non-concrete object has an obligation to join the communist party, this principle will imply that Newman as a non-concrete object has an ability to join the communist party. But this seems clearly false, because non-concrete objects don’t have an ability to join the communist party. Thus, eternalism is not compatible with the principle that “ought” implies “can.” 16 4.2. Implicit Modal Operators Let’s move on to the next response, according to which “ought” comes with the implicit “will” or “would.” This is motivated by the consideration that “ought” sentences can express either a future-tense thought that makes sense only when it is evaluated relative to a particular future time (“will”), or a hypothetical thought that makes sense only under a certain hypothesis (“would”). In this view, “ought” and “will” (or “would”) are base- generated in the same syntactic position, but “will” moves up to a higher position to take scope over the “ought”-sentence. On this view, the logical forms of (7a, b) and (8a, b) can be given as follows: (21) a. Will/Would: Newman C ought to join the Communist Party of China. b. Will/Would: Newman C ought to protect and serve the Communist Party of China. c. Will/Would: Newman A ought to stay at Trump Hotel for vacation. d. Will/Would: Newman A ought to make America great again. In (21), the descriptive names “Newman C” and “Newman A” appear in the intensional context, because they are under the scope of the phonologically null “will” or “would.” This response is based on the assumption that “ought” and “will” can appear in the same syntactic structure. However, it is simply syntactically impossible for “ought” and “will” to appear in sequence, because they are in complementary distribution (Carnie 2013). Two words are in complementary distribution when they never appear in the same linguistic context. This is evidenced by the fact that the sequences of “will” (or “would”) and “ought” lead to ungrammaticality (i.e., *“will ought,” *“ought will”). In generative syntax, if two words are in complementary distribution, this is taken as suggesting that they 17 belong to the same syntactic category and so they compete for the same syntactic position. One might try to avoid this objection by re-interpreting the hidden “will” or “would” as a text-level operator such as Heim’s existential closure. According to Heim (1982), definite and indefinite descriptions are merely variables, but they get their quantificational force from the text-level existential quantifier binding all free variables in its scope. Similarly, one might posit a text level modal operator that functions like “will” or “would.” In this view, “will” or “would” does not compete for the same syntactic position with “ought,” so it avoids the objection raised above. However, even if there is such a text-level modal operator, it still fails to explain the asymmetry between “ought”-sentences and other control constructions. Let’s compare (7a, b) and (7c) again. 19 (7) a. President Xi: Newman C ought to join the Communist Party of China (in order to get a good job in China). (Prudential “Ought”) b. President Xi: Newman C ought to protect and serve the Communist Party of China. (Ethical “Ought”) c. #Newman C wants/tries/expects/attempts/is reluctant/is eager to protect and serve the Communist Party of China. Whereas (7a, b) sound fine, (7c) does not. If there is a modal closure for control constructions, the operation must be able to save (7c) from semantic anomaly. In order to explain the asymmetry between “ought”-sentences and other control constructions, one would need a modal closure only for “ought”-sentences. However, with no independent motivation, positing a modal closure operating only on “ought” sentences would be simply 19 We can make the same point with (8a, b) and (8c) 18 ad-hoc. 4.3. Implicit Subjunctive Morpheme The last response is that “ought” itself consists of two intensional operators. It seems that the best way to implement this idea is to appeal to the compositional semantics of weak necessity modals. In many languages, a weak necessity modal is composed by combining a strong necessity modal with a subjunctive (or counterfactual) morphology that appears in the consequent of a subjunctive conditional. That is, a weak necessity modal is just a strong necessity modal with a subjunctive marker (von Fintel and Iatridou 2008). On this picture, “ought” consists of two parts: the strong necessity modal MUST/HAVE TO and the subjunctive morpheme Øsub. When the strong necessity modal is used alone, it is phonologically realized as “must” or “have to.” When it is combined with the subjunctive morpheme Øsub, it is phonologically realized as “ought” or “should.” If this is the case with English, then the logical forms of (7a, b) and (8a, b) may be understood as the following: (22) a. Øsub [Newman C MUST/HAS TO join the Communist Party of China] b. Øsub [Newman C MUST/HAS TO protect and serve the Communist Party of China] c. Øsub [Newman A MUST/HAS TO stay at Trump Hotel for vacation] d. Øsub [Newman A MUST/HAS TO make America great again] On this view, “Newman C” and “Newman A” are under the scope of “Øsub,” so they appear in the intensional context. Also, the asymmetry between (7a, b) and (7c) can be explained by the availability of the subjunctive morpheme Øsub. The control “ought” is distinguished from other control verbs in the respect that it always carries the subjunctive morpheme 19 “Øsub,” and this element is what is responsible for the intensionality of the subject position of the control “ought.” This response may provide an explanation of the asymmetry between (7a, b) and (7c) without being committed to an ad-hoc semantic operation. However, it makes some incorrect predictions. First, since it is the subjunctive morphology part which is responsible for the intensionality of the external argument of the control “ought” and strong necessity modals do not carry any subjunctive morphology, it predicts that the external argument of a strong necessity modal is extensional. That is, it predicts that (23) should pattern like (7c) and (8c). But they do not. 20 (23) a. Newman C must/has to join the Communist Party of China. b. Newman C must/has to protect and serve the Communist Party of China. c. Newman A must/has to stay at Trump Hotel for vacation. d. Newman A must/has to make America great again. Second, this response conflicts with Chrisman’s tests. On this view, strong necessity modals such as “must” and “have to” are syntactically ambiguous in the sense that they can be associated with either a control or raising construction. Then, strong necessity modals are expected to pass Chrisman’s tests. That is, they should allow for “er”-nominalization and passivization. But they do not pass these tests. If we add “-er” or “-ed” to “must” or “have to,” they lead to ungrammaticality (i.e. *“must-er,” *“have-er,” *“have to-er” (“er”- nominalization), *“must-ed”, *“be had to” (passivization)). 21 20 According to the response, the deliberative reading is due to the control construction of “ought”, and it is the strong necessity part that gives rise to the control construction. So, it predicts that strong necessity modals are ambiguous between raising and control constructions and thus they have deliberative readings. 21 One might wonder if the etymological explanation for “ought” is also available for strong necessity modals. It might work for “must” but not for “have to”. “Have to” is a strong necessity modal, but it is clearly not past-tensed in itself. It can take the past-tense morpheme (i.e., “had to”). 20 5. Implications In this section, I explore three philosophical implications that the intensionality test has on the debate. The first two have to do with the propositional view on “ought.” The last has to do with the relation between “ought” and “reason.” 5.1. “Ought” and Propositions First, the intensionality of the subject position of “ought” undermines any relational view of the deliberative “ought.” There are broadly three types of explanations for the deliberative “ought.” The first is that the deliberative “ought” expresses a relation of an agent and a property (Schroeder 2011). The second is that the deliberative “ought” expresses a relation of an agent and a proposition (Wedgwood 2006 22 ). The last is that the deliberative “ought” expresses a propositional operator with no agentive argument (Chrisman 2015; Finlay & Snedegar 2014; Horty 2001; Horty & Belnap 1995). The intensionality of the subject position of “ought” is evidence against the first two views (namely, the relational views), because they are both committed to the existence of an external argument, and external arguments are extensional unless they are under the scope of an intensional operator. Wedgwood (2006) himself presents his view as a kind of a propositional operator view. He claims that while the evaluative “ought” (or the political “ought” in Wedgwood’s terms) is not indexed, the deliberative “ought” (or the practical “ought” in his terms) is implicitly 22 Wedgwood himself claims that his view is a propositional operator view. But I argue that there is a discrepancy between what he says and what he has to say. I argue that his view collapses into a relational view, so he has to say that his view is relational. 21 indexed. 23 That is, the deliberative “ought” carries an index for agents. According to Wedgwood (2006), the logical form of the deliberative “ought” is like “O(p),” where “O” is a one-place propositional operator that takes propositions as its argument and “A” is an index for agents. However, I have two worries about the status of this agent index. First, in generative grammar, indices are generally attached to only (broadly construed) nominal expressions such as proper names, pronouns, and common nouns. This has to do with the historical fact that indices were first introduced in the binding theory in order to explain the phenomenon of syntactic binding (Baker 2003, p.96). It is not clear how indexation can extend to other categories such as modal verbs and whether there is any independent motivation for such an extension. Second, Wedgwood (2006) claims that the agent index can stand in the binding relationship to a quantifier. That is, the agent index can be bound by a higher quantifier. Thus, “∃x O<x>(Px)” is a well-formed and interpretable formula in Wedgwood (2006). However, in the generative grammar tradition, this formula is not legitimate. If the agent index is a genuine index, it does not take its own syntactic node. If the agent index does not take its own syntactic node, it is not visible to any syntactic mechanisms and thus cannot be in the binding relationship to a quantifier (Collins 2007). 24,25 Thus, if Wedgwood 23 In his original suggestion, “ought” carries two types of indices: the agent and time indexes. But in this paper, I will ignore the time index. 24 I am assuming that binding is essentially a syntactic phenomenon, or at least requires some kind of syntactic mechanism. 25 It might be worth noting that Heim and Kratzer (1998) posit an index as a node as a result of quantifier raising. In their semantic theory, the movement of a quantifier phrase has two effects: (i) it leaves a trace t with an index i and (ii) it introduces an index i right below the moved phrase, which functions as a variable binder that binds the trace ti. It seems mysterious how an index alone can take a node and function as a variable binder. In response to this problem, Heim and Kratzer might claim that quantifier-raising (i) leaves a trace ti and (ii) introduces a variable binder OPi that binds the trace ti. However, the strategy of positing an empty operator has been questioned by Collins (2017) and Pietroski (2018). 22 needs a place for agents in his theory, it should appear as an argument of “ought.” If so, his view collapses into a relational view, and the logical form of an “ought” sentence should be like “O(S, p).” 26 Both Finlay and Snedegar (2014) and Chrisman (2015) can be taken as a kind of a propositional/sentential operator view. Finlay and Snedegar (2014) argue that “ought” is a two place propositional operator: one argument is for propositions and the other is for sets of alternative propositions. Chrisman (2015) argues that “ought” is a sentential operator, but it can take either a propositional or imperatival content. He tries to capture evaluative readings with the former and deliberative readings with the latter. Since both views are not committed to the external argument of “ought,” they are compatible with the intensionality test. They can argue that the subject position of “ought” is intensional, because “ought” takes scope over the whole content which the subject is part of. The stit view (Horty 2001; Horty & Belnap 1995), according to which there is an implicit agency operator stit (= “see to it that”) in the prejacent and this is what is responsible for the deliberative “ought,” also is compatible with the intensionality test. On the stit view, given that the prejacent is the proposition that Bill kisses Lucy, the evaluative reading of “ought” can be represented as “Ought: Bill kisses Lucy,” and the deliberative reading of “ought” can be represented as “Ought: Bill sees to it that he kisses Lucy.” Since “ought” functions as a propositional operator in either reading, the stit view doesn’t have 26 Wedgwood’s view might be revised as follows: On Kratzer’s account, conversational backgrounds are functions of possible worlds (i.e., “f(w)”). One alternative to this is to say that conversational backgrounds are functions of individuals and times (e.g., “f(x, t)”). If his view is revised this way, it can remain as a propositional operator view, because individuals are treated as an argument of conversational backgrounds of “ought,” not “ought” itself. 23 to posit any external argument for agents. As long as it doesn’t posit any external argument for agents, it is compatible with the intensionality test. The second implication is that the considerations discussed so far undermine some view on weak necessity modals. For example, one might want to combine the compositional account of weak necessity modals with the control constructions of strong necessity modals. However, as has been discussed in Section 4.3., this view wrongly predicts that the subject position of a strong necessity modal is extensional. 5.2. “Ought” and “Reason” The last implication of the intensionality test that I’ll discuss concerns the relation between “ought” and “reason.” The notions of ought and reason seem to be intimately connected. For example, many philosophers paraphrase the deliberative “ought” as “has most reason to.” Schroeder tries to capture this intuition by saying that “ought” and “reason” take the same type of entity as their argument. He says: It is plausible to suppose that whatever the deliberative “ought” relates agents to is the same sort of thing as whatever reasons relate agents to. If “ought” relates agents to actions, then reasons count in favor of actions. If “ought” relates agents to propositions, then it is natural to think that, strictly speaking, reasons will count in favor of propositions.” (Schroeder 2011, p.36). If “ought” is a propositional operator, it does not relate an agent to anything. Still, Schroeder’s remark on the relationship between “ought” and “reason” can be understood as saying that the object of the deliberative “ought” is the same sort of thing as whatever reasons count in favor of. Thus, if we accept his conjecture, it suggests that what reasons 24 count in favor of is a proposition, because the intensionality test suggests that the object of the deliberative “ought” is a proposition. The intensionality test, together with Schroeder’s conjecture, implies that reasons count in favor of propositions, not actions. Let’s call this the propositional view. But Schroeder (2007) presents some objections against the propositional view. His main contention is that the propositional view has no plausible propositional candidate for agent-neutral reason ascriptions. Schroeder introduces the distinction between agent-neutral and agent relational reasons and calls sentences like (12) in which the agent of an action is not explicit agent- neutral reason ascriptions and sentences like (13) in which the agent of an action is explicit agent-relational reason ascriptions. 27 (24) a. The fact that Katie needs help is a reason to help Katie. b. There is a reason to help Katie. (25) a. The fact that there will be dancing at the party is a reason for Ronnie to go to the party. b. There is a reason for Ronnie to go to the party. 28 Schroeder focuses on (24b) and considers two propositional candidates for the propositional view. The first candidate is the proposition that everyone helps Katie. If the reason operator takes this proposition as its argument, (24b) should express something equivalent to “there is a reason for everyone to help Katie.” The problem with this is that 27 These examples are from Schroeder (2007). 28 On Schroeder’s account, reasons count in favor of actions/properties, rather than propositions and “reason” is ambiguous between a triadic relation between a fact, an agent, and an action (for agent-relative readings) and a dyadic relation between a fact and an action (for agent-neutral readings). On his account, the logical forms for (24) and (25) can be represented as follows: (24a*) Reason (p, Φ), where p is a fact and Φ is an action/property. (24b*) ∃p Reason (p, Φ), where Φ is an action/property. (25a*) Reason (p, S, Φ), where p is a fact, S is an agent, and Φ is an action/property. (25b*) ∃p Reason (p, S, Φ), where S is an agent, Φ is an action/property. 25 there can be an agent-neutral reason to help Katie even if there is no reason for everyone to help Katie. After all, in general, if someone needs help, s/he doesn’t need a help from everyone. S/he would just need enough help. The second candidate that Schroeder considers is the proposition that someone helps Katie. If this is right, (24b) is equivalent to “there is a reason for someone to help Katie.” However, the problem with this is that it fails to validate the following entailment: (26) a. There is a reason to help Katie. b. There is a reason for you to help Katie. Schroeder argues that if there is an agent-neutral reason to help Katie, it entails that there is a reason for each person to help Katie. In particular, (26a) seems to entail that there is a reason for you to help Katie. But the second candidate does not provide any clear explanation for this inference. Schroeder’s objection to the propositional view may look devastating, and he claims that the propositional view is clearly the worst of the available views. I agree with Schroeder in that neither candidate he considers is a plausible option for the propositional view. However, this doesn’t mean that the propositional view is wrong. This is because the propositional view actually has an appropriate propositional candidate for agent-neutral reasons. I propose that agent-neutral reason ascriptions take the proposition that one helps Katie. On this view, (24b) is equivalent to “there is a reason for one to help Katie.” More specifically, the logical form of (24b) can be represented as follows: (27) There is a reason PROarb to help Katie. 29 29 I remain neutral on whether “reason” expresses a triadic relation or a dyadic relation. Still, the view developed here is distinguished from Schroeder’s view in that (i) reasons count in favor of propositions, 26 “PROarb” is an unpronounced pronoun for arbitrary objects (Bhatt and Pancheva 2017; Carnie 2006; Epstein 1984). There are broadly two views on the interpretation of “PROarb.” The first says that “PROarb” refers to an arbitrary object (Fine 1985; King 1991). This requires us to introduce a new type of objects into our ontology. But what is important for the purposes of this paper is that, in this approach, the following inference counts as valid (Fine 1985; King 1991): P1. Fa, where “a” refers to an arbitrary object. C. "x Fx If the above inference is valid, the inference from (28a) to (28b) is valid. Since the inference from (28b) to (28c) is valid by universal instantiation, the inference from (28a) to (28c) is valid. (28) a. There is a reason PROarb to help Katie. b. For every x, there is a reason for x to help Katie. c. There is a reason for you to help Katie. The second approach is that “PROarb” is interpreted as a free variable and it should be bound by a higher operator (Epstein 1984; Moltmann 2006, 2010; Bhatt and Pancheva 2017; Lebeaux 2009). Epstein (1984) posits a default universal quantifier and argues that PROarb is bound by this universal quantifier. Moltmann (2006, 2010) and Bhatt and Pancheva (2017) suggest that the arbitrary interpretation of PROarb comes from the presence of a generic operator, which at least sometimes can express a universal quantificational force or something equivalent. 30 On this view, the logical form of (26a) is rather than actions, and (ii) agent-neutral reason ascriptions are special cases of agent-relative reason ascriptions so that reason ascriptions are not syntactically ambiguous. 30 Lebeaux (2009) also posits an implicit operator that binds PROarb but is not committed to what this operator is. 27 like (28b), and thus it entails (28c). In either approach, the inference in question is validated and thus this proposal satisfies Schroeder’s desideratum. Furthermore, the proposal presented here avoids the problem that the first candidate faces. That is, (27) doesn’t entail that there is a reason for everyone to help Katie. This would be best explained in terms of scope ambiguity. While “everyone” is embedded under the reason operator in “there is a reason for everyone to help Katie,” the quantifier “every” takes scope over the reason operator in (28b). So, (28b) does not induce the collective reading that there is a reason for everyone to help Katie. A similar phenomenon can be found in the following construction: (29) a. It is fun to play basketball. b. It is fun [PROarb to play basketball]. (30) a. "x It is fun [for x to play basketball] (» For everyone it is the case that if they play basketball, it is fun for them.) b. It is fun "x [x to play basketball] (» If everyone plays basketball, it is fun.) Epstein (1984) analyzes the logical form of (29a) as (29b) and notes that the interpretation of (29b) is (30a), not (30b). In the same vein, the interpretation of (27) is (31a), not (31b). (31) a. "x there is a reason for x to help Katie. b. There is a reason "x [x to help Katie]. 31 Lastly, I’d like to mention some syntactic evidence for the proposal. Let’s consider the following sentences: 31 The difference between (31a) and (31b) may be captured by the following logical forms: (31a*) "x ∃p Reason (p, x helps Katie) (LF for individual readings) (31b*) ∃p Reason (p, "x[x helps Katie]) (LF for collective readings) 28 (32) a. There is a reason to love oneself. b. There is a reason PROarb to love oneself. (33) a. There is a reason to love *himself/*themselves. b. There is a reason for everyone to love himself/themselves. c. There is a reason for someone to love himself/themselves. Reflexive pronouns like “oneself”, “himself” must be bound by an antecedent in their local domain (i.e., Chomsky’s (1981) Binding Condition A). Otherwise, the whole sentence containing the reflexive pronoun results in ungrammaticality. For example, an imperative sentence “Love yourself!” is considered carrying an unpronounced antecedent functioning like “you” and it binds the reflexive pronoun “yourself.” If the standard binding theory is right, (32a) suggests that there is an unpronounced pronoun PROarb which functions like “one.” 32 Also, (33a) provides evidence against Schroeder’s first and second candidates. If a quantifier such as “everyone” or “someone” can appear as the antecedent, (33a) should be grammatical. But it is not. This suggests that quantifiers like “everyone” and “someone” never appear in the subject position of an agent-neutral reason ascription. I have presented an appropriate propositional candidate for the propositional view. Thus, Schroeder’s objections are not decisive evidence against the propositional view. 32 One might wonder whether PRO is mandatory or optional. On the traditional account of control constructions that posits PRO, PRO seems to be mandatory. This is because syntactic theories such as the theta theory and the Extended Projection Principle (EPP) require the subject position of a clause to be always occupied. 29 6. Conclusion Schroeder (2011) claims that the deontic “ought” is syntactically ambiguous between control and raising constructions. However, Schroeder’s account fails to pass the tests proposed by Chrisman (2012) and me. These tests all test for the presence of an external argument. “-er” can be thought of as picking out the external argument of a control verb. The passive morpheme “-en” can be thought of as transforming the argument structure of a control verb by deleting/absorbing the external argument. Also, it seems that the external argument is extensional because it appears outside the scope of an intensional operator. Of course, Schroeder provides some positive arguments for the ambiguity view. For example, “Bill ought to kiss Lucy” seems to express both deliberative and evaluative readings, whereas “It ought to be that Bill kisses Lucy” seems to express only an evaluative reading. This presents a prima facie challenge to the uniformity view. In response to this argument, Finlay and Snedegar (2014) develops a pragmatic account based on the contrastive semantics, according to which “ought”-sentences are evaluated relative to a set of alternatives. “Ought”-sentences carry a default presumption that they are relativized to agential alternative sets, which gives rise to deliberative readings. Sentences like “It ought to be that Bill kisses Lucy” resist deliberative readings, because the speaker violates Grice’s Efficiency Maxim (i.e., “Avoid unnecessary prolixity”) by using inefficient sentences and thus indicates that the speaker’s intended reading is not the default one. Chrisman (2015) also defends the uniformity view by arguing that “ought” can take either a propositional or imperatival content. If it takes a propositional content, it has an evaluative reading. If it takes an imperatival content, it has a deliberative reading. “Ought” in “Bill ought to kiss Lucy” can take either a propositional or an imperatival 30 content, so this sentence has both evaluative and deliberative readings. On the other hand, since “that Bill kisses Lucy” denotes a propositional content, “ought” in “It ought to be that Bill kisses Lucy” can take only a propositional content, so this sentence expresses only an evaluative reading. Given that there are at least some alternative explanations of the positive arguments, the negative evidence proposed here suggests that the uniformity view is more empirically adequate than the ambiguity view. At the very least, the negative evidence does put pressure on the proponents of the ambiguity view to explain why the purported control “ought” behaves differently with control verbs. 33 33 This chapter appears as Lee (2021) in Synthese. Many thanks to Elli Neufeld, Frank Hong, Anthony Nguyen, David Boylan, Barry Schein, Alexis Wellwood, Deniz Rudin, Fabrizio Cariani, Mark Schroeder, Steve Finlay, Ralph Wedgwood, and the audience at the American Philosophical Association (APA) Eastern Division meeting, 2020 for very helpful discussions. Further thanks to two anonymous referees for the journal for very thorough and insightful comments. 31 CHAPTER 2 “Ought,” Agents, and Events Abstract There are broadly two views on the “ought-to-do” (= deliberative “ought”) and “ought- to-be” (= evaluative “ought”) distinction. Some philosophers and linguists have embraced what is usually called the uniformity view, according to which “ought” has just one linguistic meaning. On this view, “ought” always expresses a propositional operator with no argument place for an agent. However, others reject this view and maintain that “ought” at least sometimes expresses a relation between an agent and an action or between an agent and a proposition, so at least some sense of “ought” carries an argument place for an agent. On this view, “ought” is ambiguous between a propositional operator and a relation. This is usually called the ambiguity view. The main aim of this essay is to develop a novel strategy to defend the uniformity view over the ambiguity view. The main arguments for the ambiguity view, at least implicitly, employ the assumption that if p and q are truth-conditionally equivalent, “Ought: p” and “Ought: q” must have exactly the same range of readings. I challenge this assumption and argue that even if p and q are truth-conditionally equivalent, “Ought: p” and “Ought: q” may have different readings. This is because p and q can have different logical forms (= event structures) and so can have access to different sets of modal flavors. 1. Introduction In the literature on deontic modality and deontic logic, it is common to distinguish between “ought- to-do”s and “ought-to-be”s (Chisholm 1964; Geach 1982; Jackson 1985; Feldman 1986; Brennan 32 1993; Bhatt 1998). 1,2 Intuitively, “ought-to-be”s describe states that ought to be the case, while “ought-to-do”s describe actions that an agent ought to do. 3 For example, “There ought to be world peace” roughly says that it’d be good/desirable for world peace to obtain. On the other hand, “Bill ought to kiss Lucy” can express both “ought-to-do” and “ought-to-be” readings. On an “ought-to- do” reading of the sentence, we view Bill’s kissing Lucy as an action that the agent Bill can perform and assess this as a possible course of action for him. On an “ought-to-be” reading of the sentence, we view Bill’s kissing Lucy as merely something that can happen and assess the goodness/desirability of such an event. One thing to note here is that these readings are not completely determined by a sentence’s superficial syntactic form. As we can see from the above example, “Bill ought to kiss Lucy” is of the form “S ought to Φ” but can express an “ought-to-be” reading. Also, some sentences do not have an “ought-to-do” reading even though they are of the form “S ought to Φ.” For example, Schroeder (2011) says that “Larry ought to win the lottery” does not express an “ought-to-do” (or deliberative in Schroeder’s terms) reading (unless Larry has an ability to manipulate the lottery). Also, Humberstone (1971) says that “Prisoners should not be beaten up by prison staff” has only an “ought-to-be” (or situational in Humberstone’s terms) reading. Some philosophers and linguists have embraced what is usually called the uniformity view, according to which “ought” has just one linguistic meaning (Chisholm 1964; Williams 1981; 1 Of course, “ought” can be used to express not only deontic readings but also non-normative epistemic (or pseudo- epistemic) readings. For example, “John ought to be home by now” does not express any normative meaning. This sense of “ought” can be approximately paraphrased as “it is probable that” (Wedgwood 2006; Finlay and Snedegar 2014) or “it is normal that” (Yalcin 2016). 2 Note that different theories have used different terminology to discuss this distinction. It corresponds to Schroeder’s (2011) deliberative and evaluative “ought”s, Humberstone’s (1971) agent-implicating and situational “ought”s, Price’s (2008) agential and situational “ought”s, Finlay and Snedegar’s (2014) agential and non-agential “ought”s, and Chrisman’s (2015) agentive and non-agentive “ought”s. 3 In this paper, I mainly focus on action verbs (e.g., “kiss,” “run”). But I briefly discuss a possible extension of my solution to attitude verbs (e.g., “believe”) and emotion verbs (e.g., “feel angry”) in footnote 18. 33 Kratzer 1981). On this view, “ought” always expresses a propositional operator with no argument place for an agent. However, others reject this view and maintain that “ought” at least sometimes expresses a relation between an agent and an action (Harman 1973; Geach 1982; Brennan 1993; Price 2008; Schroeder 2011) or between an agent and a proposition (Broome 2013; Wedgwood 2006), so some sense of “ought” carries an argument place for an agent. On this view, “ought” is ambiguous between a propositional operator and a relation. This is usually called the ambiguity view. According to the ambiguity view, a sentence of the form “S ought to Φ” can be associated with at least two logical forms. When “ought” is interpreted as a propositional operator, the logical form of “S ought to Φ” is represented as “Ought1: p (= S Φs)” and the sentence has an “ought-to- be” reading. When “ought” is interpreted as a relation, the logical form of “S ought to Φ” is represented as “Ought2 (S, Φ)” and the sentence has an “ought-to-do” reading. The main aim of this essay is to develop a novel strategy to defend the uniformity view over the ambiguity view. The main arguments for the ambiguity view, at least implicitly, employ the assumption that if p and q are truth-conditionally equivalent, “Ought: p” and “Ought: q” must have exactly the same readings. I challenge this assumption and argue that even if p and q are truth- conditionally equivalent, “Ought: p” and “Ought: q” may have different readings. This is because p and q can have different logical forms (= event structures) and so can have access to different sets of modal flavors. 2. Conflicting Evidence 2.1. The Embedding Argument Let’s first start with two main arguments for the ambiguity view, presented in the philosophical literature (Harman 1973; Geach 1982; Jackson 1985; Price 2008; Schroeder 2011). I’ll follow 34 Schroeder’s (2011) presentation of these arguments. The first argument is what I call the embedding argument. Consider the following: (1) a. Bill ought to kiss Lucy. b. It ought to be (the case) that Bill kisses Lucy. While in (1a) “ought” takes “Bill kisses Lucy” directly, in (1b) “Bill kisses Lucy” is embedded under “it is (the case) that.” Schroeder observes that while (1a) has both deliberative and evaluative readings, (1b) has only an evaluative reading. 4,5 Schroeder (2011) suggests five hallmarks of the deliberative “ought”: (i) deliberative “ought”s matter directly for advice. When someone comes to you with the question of what to do, if you know what s/he deliberatively ought to do, you know what is advisable for him/her. (ii) Deliberative “ought”s close practical deliberation. If you know what one deliberatively ought to do, then that settles the question of what to do. (iii) Deliberative “ought”s are closely related to the notion of accountability. If an agent S deliberatively ought to do some action X, S is accountable if S fails to perform X. (iv) The deliberative “ought” implies the ability “can.” If an agent S 4 From now on, I’ll mainly use “deliberative” for “ought-to-do” and “evaluative” for “ought-to-be.” 5 Finlay and Snedegar (2014) challenge this data and claim that (1b) also has a deliberative reading in a suitable context. They consider the following scenario: in a dating game show, the host is about to reveal which of the contestants is obliged to give Lucy a kiss. They say that in this context, it is natural to interpret “It ought to be that…Bill kisses Lucy” or “Lucy ought to be kissed by…Bill” as having a deliberative reading. I’d like to point out two things: First, there is a pause in each sentence. Finlay and Snedegar implicitly assume that pauses are semantically insignificant. But it seems that the pause in each sentence has the effect of placing focus on “Bill.” Also, the pause indicates that something unusual is going on in the context. Second, the context is designed to make salient the question “Who ought to kiss Lucy?” Just “Bill” is sufficient to give an answer to this question. But “Bill” is not a full sentence and it doesn’t semantically express either a deliberative or evaluative reading. The host might be able to use “It ought to be that…Bill kisses Lucy” or “Lucy ought to be kissed by…Bill” to convey something equivalent to “Bill ought to kiss Lucy” in the context (especially when the focus is placed on “Bill”). But this isn’t evidence that those sentences semantically express deliberative readings, as the fact that “Bill” conveys something equivalent to “Bill ought to kiss Lucy” in this context isn’t evidence that “Bill” semantically expresses a deliberative reading. 35 deliberatively ought to do some action X, it has to actually be in S’s power to do X. (v) Deliberative “ought”s are closely related to the notion of obligation. 6 Schroeder argues that the five hallmarks of deliberative “ought”s arise for (1a), but not for (1b). What (1b) says can be roughly paraphrased as “It’d be good/desirable for a certain event (= Bill’s kissing Lucy) to happen.” Thus, (1b) is not appropriate as advice for Bill, does not settle Bill’s deliberative question, does not imply that Bill is accountable if Bill does not kiss Lucy, does not imply that Bill is able to kiss Lucy, and is not on a par with the claim that Bill has an obligation to kiss Lucy. Proponents of the ambiguity view claim that the uniformity view has difficulty explaining the contrast between (1a) and (1b). They, at least implicitly, assume what I call Equivalence: (Equivalence) If p and q are truth-conditionally equivalent, “Ought: p” and “Ought: q” have exactly the same readings. Given Equivalence, the argument goes like this: Suppose that “ought” is uniform and it expresses a propositional operator. It seems obvious that “p” and “it is the case that p” are truth-conditionally equivalent. If so, by Equivalence, “Ought: p” (= (1a)) and “Ought: it is (the case) that p” (= (1b)) must have exactly the same readings. But (1a) and (1b) have different readings. While the deliberative reading for Bill is available for (1a), the same reading is not available for (1b). Thus, “ought” is not uniform. On the other hand, the ambiguity view has a ready explanation for this data. (1a) has both evaluative and deliberative readings, because “ought” can be interpreted as either a propositional operator or a relation. When “ought” is interpreted as a propositional operator, (1a) has an 6 Schroeder says only that the deliberative “ought” is closely connected to the notion of obligation. This is because “ought” is both weaker and stronger than the notion of obligation: weaker in that obligations are strict in a way that “ought” is not (cf. the contrast between “ought” and “must”) and stronger in that obligations can conflict but “ought” is an all-things-considered. 36 evaluative reading. When “ought” is interpreted as a relation, (1a) has a deliberative reading. On the other hand, (1b) has only an evaluative reading. This is because (1b) has an expletive subject and so the relational sense of “ought” is not available for (1b). If “ought” is interpreted as a relation between an agent and an action/proposition, the argument place for an agent must be saturated by a semantically significant subject. 2.2. The Passivization Argument Let’s move on to the next argument: the passivization argument. Consider the following: (2) a. Bill ought to kiss Lucy. b. Lucy ought to be kissed by Bill. Given that “Bill kisses Lucy” and “Lucy is kissed by Bill” are truth-conditionally equivalent, if “ought” is uniformly a propositional operator, (2a) and (2b) must have the same readings (by Equivalence). However, Schroeder argues that while (2a) has a deliberative reading for Bill, (2b) does not. That is, (2a) is appropriate as advice for Bill, settles Bill’s deliberative question, implies that Bill is accountable if he does not kiss Lucy, implies that Bill is able to kiss Lucy, and can convey something on a par with the claim that Bill has an obligation to kiss Lucy. However, (2b) does not have such a reading, because it has none of these properties. While Schroeder focuses on the deliberative reading of (2a) and argues that the same reading is not available for (2b), I’d like to point out that the ambiguity view makes more predictions than just that. If the ambiguity view is correct, (2b) should also have a deliberative reading. As “Lucy wants to be kissed by Bill” expresses Lucy’s desire to be kissed by Bill, (2b) should express Lucy’s obligation to be kissed by Bill. That is, the ambiguity view should say that (2a) and (2b) just have different deliberative readings. However, to my ear, (2b) has no deliberative reading for Lucy. 37 Deliberative “ought”s are mainly for advice. If (2b) can be given as advice for Lucy, being kissed by Bill should be eligible to be an option in decision theory. But in normal contexts, being kissed by Bill (similarly, winning the lottery) cannot be an option in decision theory. One might claim that being kissed by Bill can be an option for Lucy in some special contexts in which she can manipulate Bill (or the lottery). But even in those contexts, being kissed by Bill (or winning the lottery) is not a true option because the subject exercises only indirect control over Bill (or the lottery). The true options that Lucy can choose are whether to manipulate Bill or not (or whether to manipulate the lottery or not). So, I conclude that (2b) lacks the deliberative reading for Lucy. This suggests that the contrast between (2a) and (2b) is not due to the relational sense of “ought.” However, since the judgments are subtle, I will not try to draw any firm conclusions from them. 7 2.3. Chrisman’s and Lee’s Linguistic Tests In response to the ambiguity view, Chrisman (2012) and Lee (2021) present some competing evidence that “ought” does not have any relational sense. Since it is crucial to understand the raising-control distinction to understand their evidence, I’ll first explain this distinction. Consider the following examples: (3) Bill seems to/is likely to kiss Lucy. (Raising) (4) Bill wants to/tries to kiss Lucy. (Control) Raising and control verbs may look alike on the surface, but they are structurally different (Bach 1979; Jackendoff 1972; Postal 1974; Chomsky 1993; Carnie 2012). While a raising verb takes a single propositional argument, a control verb takes two arguments: one for an agent/experiencer 7 Schroeder (p.c.) agreed with my judgments about the deliberative reading of (2b). He said that the deliberative “ought” might carry the constraint that it takes only actions as its argument. If so, since being kissed by Bill is not an action, there should be no deliberative reading available for (2b). See Lee (2021) for an argument against the view that “ought” operates on actions. 38 and one for a property/proposition. So, while raising verbs express propositional operators, control verbs express relations between an agent/experiencer and a property/proposition. Thus, the underlying structures of (3) and (4) can be represented as follows: (5) seem/likely [Bill to kiss Lucy]. (6) [Billi] wants/tries [PROi to kiss Lucy]. “Seem” and “likely” are called raising verbs because “Bill” appears as the subject of “kiss” and moves up from the subject position of “kiss” to the subject position of “seem” for syntactic reasons (that is, English sentences require grammatical subjects). On the other hand, “want” and “try” are called control verbs. “Bill” appears as the subject of “want,” and it determines what the unpronounced pronoun “PRO” refers to. In (6), “Bill” and “PRO” are co-indexed, which means that they are co-referential. This is how syntactic theories capture the fact that “Bill” is interpreted as both the subject of “want” and the subject of “kiss.” Thus, one crucial difference between (5) and (6) is that while “Bill” is the semantic subject of “want,” it is not the semantic subject of “seem.” According to the theta theory, which was first proposed by Chomsky (1993) as part of his syntactic theory, a verb (or more generally, predicates including adjectives, common nouns, prepositional phrases, etc.) encodes the lexical information about how many arguments and what types of arguments it syntactically requires. This is called the argument structure of a verb. On this view, raising and control verbs have different argument structures. The argument structure of a verb is usually represented by what is called a theta-grid. The followings are sample theta-grids for “seem” and “try” (Carnie 2012): 39 “seem” “Bill seems to kiss Lucy” Theme (Proposition) Theme (Proposition) Bill to kiss Lucy “try” “Bill tries to kiss Lucy” Agent Theme (Proposition) Agent Theme (Proposition) Billi PROi to kiss Lucy In the above examples, “seem” has only one argument for a proposition. On the other hand, “try” takes two arguments: one for an agent and one for a proposition. The argument place for an agent/experiencer is called the external argument. And the argument place for a property/proposition is called the internal argument. Let’s now look at Chrisman’s and Lee’s linguist tests. Their tests test for an external argument. Chrisman observes that control verbs usually allow for (i) “er”-nominalization (e.g., “trier,” “wanter”), and (ii) passivization (e.g., “it was tried,” “it was wanted”). The function of “-er” can be thought of as picking out the external argument of a control verb. Also, linguists generally think of passivization as a process that transforms the argument structure of a control verb by saturating or deleting its external argument (Jaeggli 1986; Baker, Johnson & Roberts 1989; Chomsky 1993; Carnie 2012). For example, the following illustrates how the passive morpheme “-en” turns a two- place function “kiss” into a one-place function “be kissed” by saturating its external argument: “kiss” “Bill kisses Lucy” Agent Theme Agent Theme Bill Lucy “kiss”+“-en” (→ “kissed”) “Lucy was kissed” Agent Theme Agent Theme -en -en Lucy In the above example, the agent theta-role is assigned to “-en,” so “be kissed” can have only one argument place. Another way to understand the function of “-en” is to see it as deleting the external 40 argument, in which case the theta-grid of “kiss-en” will have only one column for a theme. Thus, if there is a relational sense of “ought,” it is expected to pass “er”-nominalization and passivization. However, “ought” fails these two tests: “ought” does not admit of either “er”-nominalization (i.e., *“oughter”) or passivization (i.e., *“it is oughted”). Furthermore, Lee proposes a semantic test for the existence of an external argument. He argument goes like this: If the deliberative “ought” is relational, its subject position should be extensional. But it’s not. Therefore, the deliberative “ought” is not relational. For concreteness, let’s consider the following example, where “Newman C” is a descriptive name for the future baby that will be first-born in China in the 22 nd century. (7) a. President Xi: Newman C ought to protect and serve the Communist Party of China. b. #Newman C wants/tries/expects/attempts/is reluctant/is eager to protect and serve the Communist Party of China. He argues that (7a) expresses a deliberative “ought.” That is, it has all of the five hallmarks of the deliberative “ought.” If someone comes to Xi with the question of what Newman C morally ought to do, he can give (7a) as advice and it will settle the issue of what to do. Newman C will be accountable if s/he doesn’t act according to Xi’s advice. Xi’s utterance presupposes that Newman C will have an ability to act according to his advice. Also, by uttering (7a), Xi can express what he thinks Newman C is morally required to do as Chinese. However, the problem is that if the deliberative “ought” is relational, (7a) is predicted to behave like (7b). That is, the subject position of (7a) is predicted to be extensional. But it’s not. 41 3. Reconciling the Conflicting Evidence So far, we have seen that there is conflicting evidence as to whether the deontic “ought” is syntactically ambiguous. In this section, I propose a novel solution to this conflicting evidence. To reconcile the conflict, I’ll use Kratzer-Hacquard modal semantics (Kratzer 1981, 1991; Hacquard 2010) and Chomsky’s theta-theory (Chomsky 1993). 8 Since there are a lot of moving parts, I’ll proceed in two steps. I first introduce Kratzer-Hacquard modal semantics and show how this theory can be refined to capture the deliberative-evaluative distinction (Section 3.1.). Then, I show how the refined theory can be combined with Chomsky’s theta theory to explain the conflicting evidence (Section 3.2.). 3.1. Capturing the Deliberative-Evaluative Distinction In Kratzer’s (1981, 1991) semantics, modal words express quantifiers over possible worlds. For example, “ought: p” can be glossed as “all possible worlds in the domain are p-worlds.” The domain of a modal is determined by two premise sets (= sets of propositions) (i.e., what Kratzer calls conversational backgrounds): the modal base f and the ordering source g. The modal base f determines which worlds are accessible from the base world w. This preliminary domain is represented as Çf(w) (i.e., the set of worlds in which all propositions in f(w) are true). The ordering source g induces an ordering on the accessible worlds (i.e., worlds in Çf(w)). Thus, more precisely, the meaning of “ought: p” can be glossed as “all worlds in Çf(w) that are best ranked by g(w) are p-worlds.” 8 For concreteness, I formulate my solution in the Kratzer-Hacquard framework. But it’d be worth noting that the solution can be applied to any theory if it has a similar device to Kratzer’s conversational backgrounds (e.g., alternative sets in contrastive semantics (Snedegar 2012; Finlay 2014; Cariani 2016)) and this device is sensitive to events. 42 “Ought”-statements are known to express a wide range of modal flavors, including epistemic, stereotypical, deontic, bouletic, teleological, lawful, prudential, etc. In Kratzer’s account, these flavors are determined by the modal base f and ordering source g. For example, she says that if a modal takes a circumstantial modal base (= a set of relevant facts) and a deontic ordering source (= a set of moral principles), it has a deontic reading. The following are Kratzer’s conversational backgrounds for deontic readings: Kratzer’s World-Relative Conversational Backgrounds Çfcirc(w): {w¢: w¢ is compatible with the circumstances of w} (circumstantial modal base) gdeon(w): {p: p is what is moral at w} (deontic ordering source) In Kratzer’s account, f and g are functions of worlds. But for some independent reasons, Hacquard proposes that f and g are functions of events. On her view, conversational backgrounds for deontic “ought”s can be given as follows: Hacquard’s Event-Relative Conversational Backgrounds Çfcirc(e): {w¢: w¢ is compatible with the circumstances of e} (circumstantial modal base) gdeon(e): {p: p is what is moral at e} (deontic ordering source) As we can see from the above, Kratzer and Hacquard do not distinguish between deliberative and evaluative “ought.” To capture the deliberative-evaluative distinction, I propose that circumstantial modal bases are divided into two types: agential and non-agential. Deontic ordering sources are correspondingly divided into two types: deliberative and evaluative. On this picture, if a modal takes an agential modal base and a deliberative ordering source, it has a deliberative reading. If it takes a non-agential modal base and an evaluative ordering source, it has an evaluative 43 reading. 9 More specifically, I propose the following as the conversational backgrounds for deliberative and evaluative “ought”s, where “Ag(e)” refers to the agent of event e: Theta-Role Sensitive Conversational Backgrounds Çfagent(e): {w¢: w¢ is compatible with Ag(e)’s circumstances at time(e)}(agential modal base) 10 gdeli(e): {p: p is one of Ag(e)’s moral obligations at time(e)} (deliberative ordering source) Çfnon-agent(e): {w¢: w¢ is compatible with the circumstances of e} (non-agential modal base) geval(e): {p: p is what is good/desirable at e} (evaluative ordering source) On my proposal, the agent of an event “Ag(e)” may show up in conversational backgrounds. The main difference between deliberative and evaluative “ought”s is that while the agential modal base and the deliberative ordering source make reference to the agent of an event, the non-agential modal base and the evaluative ordering source do not. As can be seen from the above, this proposal draws on two main ideas: (i) event-relativity and (ii) agent theta-role. If conversational backgrounds are functions of events, the agent of an event can be well-defined and show up in conversational backgrounds (cf. the agent of a world cannot be well-defined). The key to my solution is that the notion of an agent that I am employing here is syntactic: the one who is assigned the agent theta-role by the argument structure of a verb. For 9 At this point, one might wonder if we really need two types of modal bases. The agential/non-agential modal bases will turn out to be useful in explaining two different senses of “can” (i.e., ability and circumstantial) and the ethical principle that “ought” implies “can.” I’ll come back to this issue in Section 6. Also, one might wonder if an agential modal base can combine with an evaluative ordering source and a non- agential modal base with a deliberative ordering source. My answer is no. Even in Kratzer’s semantics, not every combination is possible. She says a circumstantial modal base is restricted to a deontic/bouletic/teleological ordering source and an epistemic modal base is restricted to a doxastic/stereotypical ordering source (Kratzer 1991). Similarly, an agential modal base is restricted to a deliberative ordering source and a non-agential modal base is restricted to an evaluative ordering source. 10 One important difference between agential and non-agential modal bases is that while an agential modal base can include propositions like “Ag(e) has a (first-order) intention to performs e” and “Ag(e) has a (second-order) intention for this (first-order) intention to be the direct cause of e,” etc., a non-agential modal base cannot. This feature can be used to capture the unique agentive flavor of deliberative “ought”s. Chrisman (2015, p.115) says “Intuitively, for example, there is a difference between (i) someone bringing it about that Bill kisses Lucy by spiking his cocktail in a way that foreseeably leads to this happening without Bill’s choosing to kiss Lucy (= ought to be) and (ii) Bill’s just performing the action of kissing Lucy, as we might say, directly and actively (= ought to do).” (The italics are mine.) 44 example, “kiss” carries two arguments: one for an agent and one for a theme. In “Bill kisses Lucy,” Bill is the one who is assigned the agent theta role by the argument structure of the verb “kiss.” Here, “Ag” denotes a function from an event to the agent of the event. I’ll use “Ag(e)=Bill” to represent the information that the agent of event e is Bill. Similarly, I’ll use “Th(e)=Lucy” to represent the information that the theme of event e is Lucy. 3.2. Solving the Puzzle Let’s now turn to how theta-role sensitive conversational backgrounds can be used to explain the data presented so far. Let’s first consider the main arguments for the ambiguity view: the embedding argument and the passivization argument. The main motivation for the ambiguity view is that the uniformity view appears to have difficulty explaining the contrast between (1a) and (1b), and between (2a) and (2b). (1) a. Bill ought to kiss Lucy. b. It ought to be that Bill kisses Lucy. (2) a. Bill ought to kiss Lucy. b. Lucy ought to be kissed by Bill. While (1a) has both deliberative and evaluative readings, (1b) has only an evaluative reading. Also, (2a) has a deliberative reading for Bill, but (2b) doesn’t. Let’s first consider the contrast between (1a) and (1b). In this paper, I don’t go deep into the compositional details of “ought”-statements. But I’d like to point out that my proposal can just piggyback on Hacquard’s (2009, 2010) compositional semantics. On Hacquard’s compositional semantics, deontic modals appear below tense and aspect. Aspect introduces the existential quantifier over events and relates the time of an event to the time provided by tense. Here, I will 45 be intentionally omitting tense-related matters. If so, the logical form for (1a) can be given as follows: (1a) Bill ought to kiss Lucy. (1a*) $e [e is in w & Ought<Çf(e), g(e)>: e is a kissing event & Ag(e) = Bill & Th(e) = Lucy]. (1a*) can be read as “there is some event e in w such that in all ideal worlds w* accessible from e, e is a kissing event and the agent of e is Bill and the theme of e is Lucy.” 11 In (1a*), “Ag(e)” is well-defined. The verbal event that is fed into the modal base and ordering source is a kissing event, whose agent is Bill. So, both agential and non-agential conversational backgrounds are available for (1a) and both deliberative and evaluative readings are available for it (Remember that agential conversational backgrounds make reference to the agent theta-role of the event of a main verb. So, if “Ag(e)” is not well defined, agential conversational backgrounds are not available for “ought.”). On the other hand, the logical form for (1b) can be represented as (1b*). 11 One might worry about the status of w in (1a*). Given that the event description (e.g., “kiss”) provided by a verb generally holds across worlds (e.g., this seems to hold except when a modal takes a counterfactual morpheme), if w is the actual world, (1a*) seems to imply that “Bill kisses Lucy” is true in the actual world. Answer: deontic modals generally appear below aspect and there are two kinds of aspects: perfective and imperfective. With perfective, w is the actual world, so a modal statement “O:p” implies the truth of p. With imperfective, there is an additional layer of modality, so w denotes the most normal worlds in which everything that is normal happens or the most expected worlds in which everything that is expected/supposed to happen happens. So, a statement “O:p” doesn’t imply the truth of p. This device is designed to explain the following kind of cross-linguistic data (Hacquard 2009): (1) a. Pour aller au zoo, Jane pouvait prendre le train. (⇏ Jane took the train) To go to the zoo, Jane can-past-impf take the train. b. Pour aller au zoo, Jane a pu prendre le train. (⇒ Jane took the train) To go to the zoo, Jane can-past-pfv take the train (2) a. Imperfective: ∀w' ∈ BEST(@). $e [e is in w' & Can: Jane’s taking the train(e)] b. Perfective: $e [e is in @ & Can: Jane’s taking the train(e)] (1) is a French data. The only difference between (1a) and (1b) is whether the French “can” takes the perfective or imperfective aspects. While (1b) implies that Jane took the train at the actual world, (1a) doesn’t. This difference can be captured by (2), where @ is the actual world. This effect is widely observed in languages that have an overt aspectual distinction (e.g., French, Italian, Catalan, Bulgarian, Greek, Hindi, etc.). Because English doesn’t have an overt aspectual distinction, it is hard to detect this effect. But with the help of temporal adverbs, we can get a feel for the data in English. (3) a. In her twenties, Jane was able to swim across Lake Balaton, though she never did. b. Yesterday, Jane was able to swim across Lake Balaton, ??but she didn’t. (Piñón 2003). 46 (1b) It ought to be that Bill kisses Lucy. (1b*) $e [e is in w & Ought<Çf(e), g(e)>: e is a being-the-case-that-Bill-kisses-Lucy event] (1b*) can be read as “there is some event e in w such that in all ideal worlds w* accessible from e, e is an event/state of being the case that Bill kisses Lucy.” The verbal event that is fed into the modal base and ordering source is a being-the-case-that-p event. Since the verb “be the case that p” does not assign any agent theta role, “Ag(e)” is not well-defined. So, agential conversational backgrounds are not available for (1b). Therefore, my account correctly captures the contrast between (1a) and (1b). Let’s move on to the contrast between (2a) and (2b). (2a) has a deliberative reading for Bill, because “Ag(e)” is well-defined and it refers to Bill. However, the same reading is not available for (2b), because “Ag(e)” is not well-defined for (2b). The logical form for (2b) can be given as (2b*): (2b) Lucy ought to be kissed by Bill. (2b*) $e [e is in w & Ought<Çf(e), g(e)>: e is a being-kissed event & Th(e) = Lucy & e is by Bill] (2b*) can be read as “there is some event e in w such that in all ideal worlds w* accessible from e, e is a being-kissed event and the theme of e is Lucy and e is done by Bill.” The verbal event that the conversational backgrounds take is a being-kissed event. “Be kissed” is the passive form of “kiss.” Since the passive morpheme “-en” absorbs (or is assigned) the agent theta role of the main verb (Section 2.3.), “be kissed” does not assign any agent theta role, so “Ag(e)” is not well-defined. Of course, the agent may be represented by the optional prepositional phrase “by Bill.” But importantly adjuncts are modifiers, not arguments. Since modifiers are not assigned any theta roles by verbs, 12 the agent role is not syntactically assigned to Bill in (2b*) and so agential 12 On Chomsky’s theta theory, there is a fundamental difference between arguments and adjuncts. The argument- adjunct distinction is required to explain why “Bill kissed Lucy yesterday” is grammatical but “Bill kissed Lucy Mary” 47 conversational backgrounds are not available for (2b). Therefore, my uniformity account correctly predicts the observed contrast between (2a) and (2b). So far, we have seen how we can explain the main arguments for the ambiguity view within the uniformity view. The arguments for the ambiguity view assume Equivalence. But my solution shows how Equivalence can be falsified. For example, even if “p” (= Bill kisses Lucy) and “It is the case that p” are truth-conditionally equivalent, “Ought: p” and “Ought: it is the case that p” may have different readings, because “p” and “it is the case that p” have different event structures. While the agent of a kissing event is syntactically well-defined, the agent of a being-the-case-that- p event is not well defined. Similarly, even if Bill’s kissing Lucy and Lucy’s being kissed by Bill are truth-conditionally equivalent, “Ought: Bill kisses Lucy” and “Ought: Lucy is kissed by Bill” may have different readings, because a kissing event syntactically carries an agent theta role, a being-kissed event does not. Moreover, my solution makes correct predictions with regards to Chrisman’s and Lee’s linguistic tests. Their tests all test for an external argument (i.e., an argument place for an agent/experiencer). But in my account, “ought” does not carry any external argument. So, “ought” is correctly predicted to fail their linguistic tests. More specifically, the function of “-er” is to pick out the external argument of a verb. Also, the function of the passive morpheme “-en” is to transform a two-place function into a one place function by deleting the external argument. Since “ought” does not carry an external argument, it doesn’t allow for “er”-nominalization and passivization. Furthermore, since “ought” is a propositional operator, (7a) doesn’t have to behave isn’t. The latter is ungrammatical, because “kiss” can take only two arguments, but they are already saturated by “Bill” and “Lucy.” The former is grammatical because “yesterday” is interpreted as an adjunct modifier and it doesn’t have to be assigned any theta-role. 48 like (7b). (7a) can be legitimate, because “Newman C” appears in the scope of the intensional operator “ought” and so the subject position of “ought” can be intensional. 4. Other Linguistic Tests In Section 3, I provided an alternative explanation for the main arguments for the ambiguity view. In addition to these, Schroeder argues that deliberative “ought”s behave like control verbs in standard linguistic tests to distinguish between raising and control verbs. In this section, I argue that the analysis developed here fares equally well with these linguistic tests. He considers three linguistic tests: the expletive subject test, the idiomatic subject test, and the thematic role test. These tests all test for theta roles, and I have shown how conversational backgrounds can be sensitive to theta roles. Thus, the view developed here should get the tests right. Let’s first consider the expletive subject test. While raising verbs admit of expletive subjects such as “it” or “there,” control verbs do not. 13 (8) a. It seemed/*wanted/ought to be assumed that he is capable. b. There seemed/*wanted/ought to be world peace. As can be seen in (8), raising verbs (e.g., “seem”) can take expletive subjects, but control verbs (e.g., “want”) cannot. So, the ambiguity view predicts that “ought”-sentences with expletive subjects will be acceptable but will have only evaluative readings. This prediction sounds correct, and my analysis makes the same prediction. Agential conversational backgrounds are available only when the main verb assigns the agent theta role to its subject. Since expletive subjects are not assigned any theta-roles, agential conversational backgrounds are not available for (8a) and (8b), and so they have only evaluative readings. 13 Examples (8)-(10) are from Schroeder (2011). 49 The second is the idiomatic subject test. Idiomatic meanings are preserved under raising verbs, but not under control verbs. (9) a. All hell seemed/*wanted/ought to break loose. b. The cat seemed/*wanted/ought to get his tongue. As illustrated by (9), idiomatic readings are preserved under raising verbs such as “seem,” but if a control verb such as “want” is inserted in the middle of an idiom, it can have only a non-idiomatic reading. An idiom is treated as a single expression in the sense that its meaning is not composed of the meanings of its constituents. This explains why control verbs such as “want” do not preserve the idiomatic meanings. Control verbs like “want” require the subject argument and thus require “all hell” and “the cat” to have semantic significance. The ambiguity view predicts that sentences in which “ought” is inserted in the middle of an idiom will admit only of evaluative readings. Again, I think this prediction is correct, but my analysis makes the same prediction. Agential readings exploit the agent theta role of the main verb. Since agential readings require “all hell” and “the cat” to be assigned the agent theta role by the main verbs, the idiomatic meanings are not preserved under agential readings. Lastly, let’s take the thematic role test. While raising verbs take a subject of any kind (e.g., meetings, cities, theorems, days, etc.), control verbs don’t. (10) a. The meeting seemed/*wanted/ought to start at noon. b. Jerusalem seemed/*wanted/ought to be divided between Israel and Palestine. c. Fermat’s Last Theorem seemed/*wanted/ought to be provable. d. Yesterday seemed/*wanted/ought to have been forgotten. While raising verbs do not assign any theta roles to the subject position, control verbs assign particular theta roles to the subject position. As a result, the subjects of control verbs need to be of 50 the right category to match the theta roles assigned by the verb. The ambiguity view predicts that “ought”-sentences can take non-agential subjects but they will have only evaluative readings. Again, I believe this prediction is correct, but my analysis makes the same prediction. Agential modal bases and ordering sources make reference to the agent of the event of the main verb. Since non-agential subjects such as meetings, cities, theorems, and days are not of the right type to be assigned the agent theta role, agential readings are not available for “ought”-sentences with non- agential subjects. Therefore, the uniformity view developed here correctly predicts and explains why the deliberative “ought” behaves like control verbs in the standard linguistic tests for control verbs. This is because those tests are actually tests for theta-roles and the deliberative “ought” is sensitive to theta-roles. 5. Direct and Indirect Deliberative “Ought”s In this section, I consider the direct and indirect deliberative distinction, which has been first introduced in Schroeder (2011) but challenged by many authors (e.g., Chrisman 2012; Finlay & Snedegar 2014), and discuss how to extend my framework to accommodate the distinction. Consider the following: (11) Bill ought to kiss Lucy. (Deliberative) (12) a. (To the caterers) We are expecting fifty guests tonight. There ought to be 50 chairs in the living room by 5 p.m. (Bhatt 1998) b. Wedding Advice: Invitations ought to go out by post. c. Cooking Advice: Rump roast ought to cook slowly. d. Editorial Advice: Americanized spelling ought to be eliminated. (Chrisman 2012) 51 (13) Larry ought to win the lottery. (Evaluative) 14 When (12a) is uttered to the caterers, it is naturally interpreted as imposing an obligation to them to bring about a certain state in the world. Similarly, (12b–d) can be used in the context of giving advice. In the literature, some (e.g., Hacquard 2009) draw the line between (11) and (12) and classify (12) as “ought-to-be”s. On this distinction, the difference between “ought-to do”s and “ought to be”s lies in whether the obligation in question is attributed to the subject (= “ought to do”) or the addressee (= “ought to be”). But most (e.g., Brennan 1993; Bhatt 1998; Chrisman 2012; Wedgwood 2006) draw the line between (12) and (13) and classify (12) as “ought-to-do”s. I am more sympathetic with the latter view, because (12a-d) have all of the five hallmarks of the deliberative “ought.” I will call sentences like (11) direct deliberative “ought”s and sentences like (12a–d) indirect deliberative “ought”s. The direct-indirect deliberative distinction was first introduced by Schroeder (2011), but he doesn’t provide a concrete way to draw this distinction. I propose that direct deliberative “ought”s are those in which the agent role is assigned to the subject by syntax (i.e., the argument structure of a verb) whereas indirect deliberative “ought”s are those in which the agent role is assigned to the addressee (or a constituent of an utterance context such as the speaker) merely by context. In order to capture the meaning of an indirect deliberative “ought,” I let the ordering source g be relativized to utterance contexts, as well as events. 15 If g can take an utterance context as its input, the addressee of the utterance context can be well-defined and show up in the conversational backgrounds. Let “Add” be a function from a context to the addressee of 14 “Win” is an interesting case because its subject is assigned neither an agent role nor an experiencer role. Instead, it seems that “win” assigns the beneficiary role to its subject (Levin 1993) 15 More generally, g is a function of a part of a world, including events and utterance contexts. On my view, an utterance context consists of at least a speaker, an addressee, a topic, the common ground, etc. 52 the context. So, “Add(c)” refers to the addressee of context c. If so, the ordering sources for deontic “ought”s can be revised as follows: Deontic Ordering Sources gdir-deli(e): {p: p is one of Ag(e)’s moral obligations at time(e)} (direct deliberative) gindi-deli(c): {p: p is one of Add(c)’s moral obligations at time(c)} (indirect deliberative) geval(e): {p: p is what is good/desirable at e} (evaluative) The main difference between direct and indirect deliberative ordering sources is that while “Ag” is a function of the event of a main verb so the agent role is directly assigned to the subject by syntax, “Add” is a function of an utterance context so the agent role should be indirectly assigned to the addressee by context. Since indirect deliberative “ought”s are sensitive to utterance contexts, the logical form for them should be modified accordingly. For example, the logical form for (12a) can be given as follows: (12a*) λc. [c is in @ & Ought<Çf(c), g(c)>: $e [50 chairs (e)]. In (12a*), “c” represents an utterance context which is a part of the actual world @ and “ought” appears above the existential quantifier over events (= Aspect). So, f and g are not bound by the event quantifier but bound by the utterance context. Given that the ordering source in the utterance context gindi-deli(c) is just {Add(c) makes sure that there are 50 chairs in the living room by 5p.m.}, (12a*) can be read as “In all ideal worlds in which the addressee makes sure that there are 50 chairs in the living room by 5p.m), there is an event/state of there being 50 chairs in the living room by 5p.m.” This seems to correctly capture the intuitive meaning of (12a). To summarize, “ought” can be bound by either an event or an utterance context. If “ought” appears below Aspect (= the event quantifier), it is bound by the event provided by the main verb, 53 so Ag(e) can show up in the conversational backgrounds and this gives rise to a direct deliberative “ought.” On the other hand, if “ought” appears above Aspect (= the event quantifier), it is bound by the utterance context, so Add(c) can show up in the conversational backgrounds and this gives rise to an indirect deliberative “ought.” Thus far, we have seen how to capture the direct-indirect deliberative distinction within the uniformity view. This gives the uniformity view an important advantage over the ambiguity view, because this solution is not available to the ambiguity view and it is not clear how the ambiguity view can capture the direct-indirect deliberative distinction and accommodate data like (12). 6. Deliberative “Ought” Implies Ability “Can” In this section, I argue that the view developed here gives a natural account of the ethical principle that “‘ought’ implies ‘can’”. This is the principle that “S ought to Φ” in the sense of practical deliberation implies “S can Φ” in the sense of ability. In this section, I show how my analysis can be extended to capture the ability sense of “can” and then why the deliberative “ought” implies the ability “can.” Let’s first consider the following two senses of “can”: (14) Bill can lift this table. (ability) (15) Hydrangeas can grow here. (circumstantial) (14) is naturally interpreted as expressing Bill’s ability to lift the table. I’ll call this the ability reading, and “can” on the ability reading can be roughly paraphrased as be able to. (15) is naturally interpreted as expressing circumstantial modality. It concerns what is possible, given a certain circumstance. I’ll call this the circumstantial reading, and “can” on the circumstantial reading can 54 be glossed as it is possible that. For example, (15) can naturally be given as an answer to the question in the following scenario: Suppose I acquire a piece of land in a far-away country and discover that soil and climate are very much like at home, where hydrangeas prosper everywhere. Since hydrangeas are my favorite plants, I wonder whether they would grow in this place and inquire about it. The answer is (15). In such a situation, the proposition expressed by (15) is true. It is true regardless of whether it is or isn’t likely that there are already hydrangeas in the country we are considering. All that matters is climate, soil, the special properties of hydrangeas, and the like. (Kratzer 1991) On Kratzer’s account, (15) expresses a pure circumstantial reading, which is characterized by a circumstantial modal base and an empty ordering source. The circumstantial modal base for (15) may include all relevant facts that hold at a certain circumstance such as the special properties of hydrangeas, the climate and soil of the land, etc. So, (15) roughly means that, given hydrangeas’ special properties, hydrangeas’ growing is compatible with the climate and soil of the land. One interesting point is that the arguments for the ambiguity of “ought” can be replicated with the ability “can.” That is, the ability-circumstantial “can”s pattern like the deliberative-evaluative “ought”s. Consider the following examples: (16) a. Bill can kiss Lucy. (ability/circumstantial) b. It can be that Bill kisses Lucy. (circumstantial) (17) a. Bill can kiss Lucy. (Bill’s ability O) b. Lucy can be kissed by Bill. (Bill’s ability ´) (18) a. The expletive subject test: There seemed/*wanted/can be world peace. b. The idiomatic subject test: The cat seemed/*wanted/can get his tongue. 55 c. The thematic role test: The meeting seemed/*wanted/can start at noon. An ability reading is available for (16a), but not for (16b). Also, (17a) is naturally interpreted as expressing Bill’s ability but the same reading is not available for (17b). “Can”-sentences in (18) are acceptable but they have only circumstantial readings. Moreover, “can” does not pass Chrisman’s “er”-nominalization (i.e., *“can-(n)er”) and passivization tests (i.e., *“can-(n)ed”). The analysis developed in this paper gives a natural explanation for (16)-(18) without positing lexical ambiguity. On my view, circumstantial modal bases are divided into two kinds: agential and non-agential. I propose that while the circumstantial “can” is characterized by a non-agential modal base and an empty ordering source, the ability “can” is characterized by an agential modal base and an empty ordering source. Agential and non-agential modal bases are repeated below: Agential and Non-Agential Modal Bases Çfcirc-agent(e): {w¢: w¢ is compatible with Ag(e)’s circumstances at time(e)} 16 Çfcirc-non-agent(e): {w¢: w¢ is compatible with the circumstances of e} A non-agential modal base may include all relevant facts that hold at a certain circumstance. Similarly, an agential modal base may include all relevant facts about the agent and his/her surroundings such as the agent’s physical abilities and all relevant properties of the circumstance that the agent is located in. For example, the modal base for (14) may include propositions about Bill’s physical strength (e.g., “Bill is strong/weak”), Bill’s physical conditions (e.g., “he is not injured”), the weight of the table (e.g., “the table weighs 50lbs”), etc. 16 The notion of ability seems intimately related to the notion of control. Philosophers often distinguish between two kinds of control: control over actions and control over beliefs (cf. Hieronymi’s (2006) manipulative and evaluative control, Owen’s (2000) voluntary and reflective control). If this is the case, we might need a more fine-grained distinction than this. I am sympathetic with this distinction, and there seem to be multiple ways to accommodate this distinction in my framework. For example, one way to accommodate this distinction is to say that other theta-roles such as the experiencer of an event Ex(e) may show up in the conversational backgrounds and that while the agent role gives rise to a manipulative/voluntary control reading, the experiencer role gives rise to an evaluative/reflective reading. But I will not further pursue this issue here. 56 Since the arguments for the ambiguity view can be replicated with the ability “can,” if the ambiguity view is correct, it will require us to be committed to the ambiguity of “can” as well. But my analysis doesn’t have to be committed to the ambiguity of “can.” This gives my analysis an important advantage over the ambiguity view. Furthermore, although my analysis is in line with Kratzer’s framework, it’s an improvement on her account. First, since Kratzer doesn’t provide any account of the ability “can,” my analysis fills in the needed gap left in Kratzer’s account. Second, even though it is widely accepted that the deliberative “ought” implies the ability “can,” it has been unclear, at least on Kratzer’s account, how and why they stand in this relationship. My account gives a direct answer to this question. The deliberative “ought” implies the ability “can,” because they share the same agential modal base. The difference between them arises from different ordering sources. While the deliberative “ought” takes a deliberative ordering source, the ability “can” takes an empty ordering source. 17 7. Comparison Due to the limitations of space, I’ll briefly compare my analysis with two pre-existing competitors. They are different versions of the uniformity view, but they are not empirically adequate. 7.1. The stit account The stit account (Horty 2001; Horty & Belnap 1995) posits an implicit agency operator stit (= “see to it that”) in the prejacent and says that (1a) has ambiguous readings because its prejacent is ambiguous. The evaluative reading of (1a) arises from “Ought: S Φs” and the deliberative reading 17 My analysis also implies that the evaluative “ought” implies the circumstantial “can.” This is because they share the same non-agential modal base. The difference between them arises because while the evaluative “ought” takes an evaluative ordering source, the circumstantial “can” takes an empty ordering source. 57 of (1a) arises from “Ought: stit: S Φs.” However, Schroeder (2011) correctly points out that the stit account falls prey to the overgeneration problem. Since it attributes the deliberative-evaluative distinction to the ambiguity in the prejacent, it predicts that (1b) should have both deliberative and evaluative readings. But (1b) does not have a deliberative reading. The solution developed here is not subject to this problem. On my view, (1a) has both deliberative and evaluative readings because both agential and non-agential conversational backgrounds are accessible from a kissing event. On the other hand, (1b) has only an evaluative reading because agential conversational backgrounds are not accessible from the being-the-case- that-p event. 7.2. Chrisman’s practition-based account Chrisman (2015) defends the uniformity view by arguing that “ought” can take either a declarative content (= propositions) or an imperatival content (= what he calls practitions). If “ought” takes a declarative content, it has an evaluative reading. If it takes an imperatival content, it has a deliberative reading. Chrisman’s account is similar to the stit account in that it attributes the deliberative-evaluative distinction to the ambiguity in the prejacent. However, it’s an improvement on the stit account, because it has a potential to explain the asymmetry in readings between (1a) and (1b). (1a) has both deliberative and evaluative readings because the prejacent “Bill to kiss Lucy” can express either a proposition (or whatever is expressed by “Bill kisses Lucy”) or a practition (or whatever is expressed by “Bill, kiss Lucy!”). (1b) has only an evaluative reading because “that p” can express only a proposition. However, Chrisman’s account fails to explain indirect deliberative “ought”s (Section 5). Let’s consider the following examples again (Bhatt 1998; Chrisman 2012): 58 (19) a. There ought to be 50 chairs in the living room by 5 p.m. b. #There, be 50 chairs in the living room by 5 p.m! (20) a. Invitations ought to go out by post. (Wedding Advice) b. #Invitations, go out by post! (21) a. Rump roast ought to cook slowly. (Cooking Advice) b. #Rump roast, cook slowly! (22) a. Americanized spelling ought to be eliminated. (Editorial Advice) b. #Americanized spelling, be eliminated! Chrisman (2012) says that the above examples express deliberative “ought”s. If so, “ought” should take an imperatival content expressed by its prejacent. However, according to Chrisman’s (2015) analysis, the imperatival content of each prejacent will be the same as what is expressed by (19b)– (22b). But the prejacent of each sentence does not express a proper imperatival content. For example, “there” in (19b) is an expletive subject, so it does not refer to any addressee. Similarly, (20b)–(22b) presuppose/imply that the subjects (i.e., invitations, rump roast, Americanized spelling) are addressees. But they are not proper objects to be an addressee. On my view, (19)– (22) express indirect deliberative readings. They can be interpreted as imposing an obligation or giving advice to the addressee of the utterance context Add(c). But it is not clear how the direct- indirect deliberative distinction can be captured in his analysis. 8. Conclusion In this paper, I presented conflicting evidence as to whether deontic “ought”s are syntactically ambiguous, and developed a novel strategy to reconcile the conflicting evidence within the uniformity view. The main arguments for the ambiguity view implicitly assume Equivalence (i.e., 59 If p and q are truth-conditionally equivalent, “Ought: p” and “Ought: q” must have the same readings), but this principle is not always true because the contextual parameters (e.g., Kratzer’s conversational backgrounds) for “ought” are sensitive to the theta roles of an event and p and q may have different event structures. After presenting the solution, I argued that my analysis successfully explains why the deliberative “ought” behaves like a control verb in the standard tests to distinguish between raising and control verbs, and showed how my analysis can be extended to capture the direct-indirect deliberative distinction and how it can capture the moral principle that the deliberative “ought” implies the ability “can.” 18 18 In this paper, I have focused on action verbs (e.g., “kiss”). But my framework can be extended to attitude verbs (e.g., “believe,” “want,” etc.) and emotion verbs (e.g., “feel outrage,” “fear,” etc.) as well. Consider the following: (Doxastic “Ought”) You ought to believe that the Earth revolves around the Sun. (Feldman 2001) (Emotional “Ought”) You ought to feel outrage about genocide. (Chrisman 2008) One might argue that doxastic and emotional “ought”s express only some of the five hallmarks for the deliberative “ought,” so they express a normative reading that is neither a deliberative nor an evaluative reading. For example, one might argue that they cannot be given as advice to someone, because we do not have the same sort of control over our attitudes or emotions as we do over our ordinary actions (cf. Hieronymi’s (2006) manipulative and evaluative control, Owen’s (2000) voluntary and reflective control). Nevertheless, one can be accountable if they fail to believe the Earth revolves around the Sun or feel outrage about genocide. Attitude verbs and emotion verbs are generally taken as carrying the experiencer theta-role (Jackendoff 1972; Grimshaw 1990; Carnie 2012). If so, this third reading can be captured if agents and experiencers are subject to different kinds of norms. For example, suppose that there are at least three kinds of norms: norms governing agents’ actions (= action norms), norms governing experiencers’ mental states, (= mental state norms), norms evaluating states of affairs in the world (= worldly state norms). Then, ordering sources for deontic “ought”s can be modified accordingly: Deontic Ordering Sources gdeli(e): {p: p is one of Ag(e)’s action norms at time(e)} gdoxa/emo(e): {p: p is one of Exp(e)’s mental state norms at time(e)} geval(e): {p: p is what is good/desirable according to worldly state norms at e} 60 CHAPTER 3 “Ought” and Intervals Abstract Over the last few decades, the semantics of “ought” has proliferated along with an expanding diversity of conceptions of both what and how we communicate with deontic modals. Examples include the classical semantics, contrastive semantics, and conflict accounts. However, most of these views do share one striking feature – the idea that “ought” claims are evaluated on an ordinal scale. I argue in this paper that this assumption is mistaken. By considering some new data that haven’t been considered before, I show why deontic “ought” claims require at least an interval scale, and why this creates an adequacy problem for a wide range of existing views. I will also show, by considering two other proposals, that simply incorporating interval scales into our semantics does not suffice. I will argue that we need the difference between values, not just values on an interval scale. In closing, I will show how my account of “ought”-claims can be extended to capture the unity of epistemic and deontic “ought”-claims. 1. Introduction Accounts of the semantics of “ought” have proliferated over the last few decades, and with them, an expanding diversity of conceptions of both what we communicate with deontic modals, and how. The classical semantics, 1 contrastive semantics, 2 and conflict account, 3 to name just three, all tell us very different things about the logic and argument structure of “ought”, and about what 1 Lewis (1973), Kratzer (1981, 1991, 2012), von Fintel (2012). 2 Sloman (1970), Jackson (1985), Finlay (2009, 2014), Snedegar (2012), Cariani (2013, 2016), Lassiter (2011, 2017), Blumberg & Hawthorne (2021). 3 Fraassen (1973), Kratzer (1977), von Fintel (2012), Horty (2012), Swanson (2016). 61 it takes to express an “ought” claim in context. But nearly all of these views have something striking in common – the idea that “ought” claims deploy an ordinal scale. This pervasive assumption, I will argue in this paper, is a mistake. Drawing on some new data that has not been considered before, I will show that deontic “ought” claims require at least an interval scale, and why this creates a problem of adequacy for a range of existing views. But I will also show, by considering two other proposals, that it is not enough to introduce interval scales into our semantics. I will argue that we need the difference between values, not just values on interval scales. After developing a new analysis of “ought”-claims, I will close by showing how my account can be extended to capture the unity of epistemic and deontic “ought”s. 2. Background Before presenting the new data, let me explain some key notions such as “interval scale”, “ratio scale”, etc. If you are familiar with these notions, you can just skip to the next section. Basically, there are four levels of scales: nominal, ordinal, interval, and ratio scales (Stevens, 1946; Krantz et al, 1971; Sassoon 2010). This four-level distinction is determined by the formal properties of scales. Each scale is characterized by equality (= classifying), rank-ordering, equality of differences, and equality of ratios. One thing to note here is that the characteristics of each type of scale are cumulative. That is, each has the properties of lower levels and adds some new properties. Nominal scales involve categorization, but no orderings. An example of a nominal scale is the color of hair. “Black”, “brown”, “red”, etc. describe the color of hair, but there is no natural rank order between them, and so “greater than” or “less than” is not well-defined over those colors. We can only talk about whether two things have the same color or not. 62 Ordinal scales involve orderings. For example, the results of Olympic games such as gold, silver, and bronze medals constitute an ordinal scale. On an ordinal scale, we can decide which ranking value is better than another, but it’s not meaningful to ask how close or far apart two ranking values are. Interval scales have orderings and differences. On an interval scale, if one value is higher than another, we can measure the difference between the two values. For example, Centigrade and Fahrenheit temperature scales are interval scales. On Centigrade and Fahrenheit scales, we can decide not only which degree is greater than another but also how much one degree is greater than another. But interval scales have no true zero points. For example, the zero degree on the Centigrade scale is just the water freezing point. That doesn’t represent the absolute lack of heat. The same point is represented as 32 degrees on the Fahrenheit scale, which is an equally legitimate measurement of temperature. Other examples of interval scales include IQ test scores, dates on a calendar, and expected utilities, etc. Ratio scales have orderings, differences, and true zero points. On ratio scales, we can measure the ratio of two values. For example, degrees on the Kelvin temperature scale constitute a ratio scale. The zero point on the Kelvin scale represents the absolute lack of heat. Thus, while “20°C is twice as hot as 10°C” is not true because 0°C doesn’t mean the absolute zero point, “20°K is twice as hot as 10°K” is true because 20°K is two times farther away from the absolute zero point than 10°K. Other examples of ratio scales include age, weight, and height, etc. 3. Interval Modifiers In this section, I will present a new piece of data. It concerns an “ought” that takes part in a comparative construction (i.e., comparative “ought”). It will show that comparative “ought”s can 63 be modified by interval modifiers. But before considering comparative “ought”-claims, let’s first look at what interval modifiers are. (1) a. Bill is taller than Mary is. b. Bill is 2 inches/much/slightly taller than Mary is (interval modifiers). c. Bill is twice/three times as tall as Mary is (ratio modifiers). (1a)–(1c) are all constructions of degree comparison. But there is an important difference between them. (1a) is just a regular comparative construction. To capture the meaning of (1a), we need at least an ordinal scale. For example, when “height” denotes a function from an individual to their height, the meaning of (1a) can be captured like: height (Bill) > height (Mary). (1b) is a comparative construction with an interval modifier. To capture the meaning of (1b), we need at least an interval scale. Measure phrases like “2 inches” are naturally interpreted as predicates of differences (von Stechow, 1984; Schwarzschild, 2005; Sassoon, 2010). So, (1b) is naturally interpreted as: Bill’s height is strictly greater than Mary’s height and the difference between them is at least 2 inches (i.e., height(Bill) – height(Mary) ≥ 2 inches). “Much” and “slightly” appear in the same syntactic position as measure phrases like “2 inches”. Thus, if “2 inches” is a predicate of differences, this provides good reason to think that “much” and “slightly” are predicates of differences as well. (1c) is a comparative construction with a ratio modifier. To capture the meaning of (1c), we need at least a ratio scale. (1c) can be paraphrased as: Bill’s height is strictly greater than Mary’s height and the ratio between them is at least 2 (i.e., !"#$!%('#(() !"#$!%(*+,-) ≥ 2). Since “tall” allows for both interval and ratio modifiers, this suggests that the height scale associated with “tall” must be a ratio. Let’s now consider whether deontic “ought” can take part in a comparative construction and, if so, whether they can take interval and ratio modification. 64 (2) a. Bill ought to save his daughter more than he ought to save a stranger. b. Bill ought to save his daughter much/slightly more than he ought to save a stranger (interval modifiers). c. #Bill ought to save his daughter twice/three times as much as he ought to save a stranger (ratio modifiers). As we can see from (2a), deontic “ought” can take part in a construction of degree comparison. Furthermore, deontic “ought” can take interval modifiers but not ratio modifiers. Recent work on “ought” (e.g., Lassiter 2011, 2017; Portner & Rubinstein 2016) has noticed that “ought” can take part in a construction of degree comparison (e.g., (2a)). But they haven’t highlighted the fact that “ought” allows for interval modifiers (e.g., (2b)). “Much” and “slightly” in (2b) are naturally interpreted as predicates of the difference between two degrees. Let p be the proposition that Bill saves his daughter and q the proposition that Bill saves a stranger. Then, (2b) is naturally paraphrased as: p’s degree on the relevant scale is strictly greater than q’s degree and the difference between the two degrees is much/slight. This suggests that “ought” is associated with at least an interval scale. This result raises a significant challenge to a wide range of major theories of “ought”, because they employ an ordinal scale and we cannot measure the difference between two values on an ordinal scale. I will discuss this issue in the next section. 4. Existing Accounts In this section, I will consider four of the most prominent accounts of deontic “ought”. More specifically, in Sections 3.1 and 3.2, I consider two accounts that employ an ordinal scale. In Sections 3.3 and 3.4, I examine two accounts that employ an interval scale. I argue that none of those accounts provide a satisfactory explanation of empirical data. 65 4.1. Kratzer’s account I’ll first consider the classical semantics, according to which “ought” expresses a universal quantifier over possible worlds. For concreteness, let’s take a Kratzerian definition of “ought” (Lewis, 1973; Kratzer 1981, 1991, 2012; von Fintel, 2012). The classical semantics for “ought” (3) ⟦“Ought: 𝜙”⟧ f,g,w = 1 iff "w' Î BEST(f, g, w). ⟦𝜙⟧ f,g,w' = 1. On this view, “Ought: 𝜙” is true relative to the modal base f and the ordering source g iff all best worlds determined by f and g are 𝜙-worlds. As can be seen from this, the classical semantics is based on orderings over worlds. Since f and g play a key role in determining orderings on worlds, let me briefly explain what they are and how they determine orderings on worlds. Basically, f and g are sets of propositions. 4 The main function of f is to provide a domain Çf (= a set of worlds in which all propositions in f are true) and the main function of g is to impose an ordering on the worlds in the domain provided by f. Based on f and g, the relative ranking of worlds is defined as: for two worlds u and v, u is at least as good as v (i.e., u ≥g v) iff all propositions in g that are true in v are also true in u. In other words, u is at least as good as v iff u satisfies all the propositions that v satisfies. Based on the world ordering ≥g, the best worlds are defined as the set of worlds that no other world outranks. The world ordering ≥g and the selection function BEST can formally be defined as follows: (4) For any u, v in Çf, u ³g v iff {fÎg: vÎf} Í {fÎg: uÎf}. (5) BEST(f, g, w) = {uÎÇf: "vÎÇf [v ³g u ® u ³g v]} 4 For simplicity, I am ignoring relativization to worlds. 66 One significant problem with the classical semantics is that it is not clear whether and how it can accommodate the data in (2). When it comes to (2a), “Ought: 𝜙 more than Ought: 𝜓” can be understood as expressing a relation between two degrees. That is, (2a) compares (i) the degree that the proposition that Bill saves his daughter on the relevant scale and (ii) the degree that the proposition that Bill saves a stranger has on the same scale. To explain (2a), proponents of the classical semantics can claim that the truth condition of “Ought: 𝜙 as much as Ought: 𝜓” is just that the best-𝜙 world is at least as good as the best 𝜓-world (Villalta, 2008). 5 Based on this, they can claim that “Ought: 𝜙 more than Ought: 𝜓” is true just in case the best-𝜙 world is as good as the best 𝜓-world, but the best-𝜓 world is not as good as the best 𝜙-world. However, the classical semantics fails to explain (2b) (i.e., a comparative “ought” with an interval modifier). Given that p is the proposition that Bill saves his daughter and q the proposition that Bill saves a stranger, the truth-condition of (2b) can be given as: (i) p’s degree on the relevant scale is strictly greater than q’s degree and (ii) the difference between the two degrees is mush/slight. However, the classical semantics, as it stands, has no obvious way to capture the second part of this truth-condition. This is because, on the classical semantics, the world-ordering ≥g is ordinal. 6 That is, for any two worlds w1 and w2, even if w1 and w2 are connected and w1 is better than w2, the ordering relation ≥g doesn’t determine how much w1 is better than w2. This suggests that in order to capture the data like (2b), “ought” must take at least an interval scale. 7 5 This is inspired by Kratzer’s (1981, 1991) notion of comparative possibility, according to which ϕ is at least as good a possibility as ψ (relative to f and g) iff for all u ÎÇf, if u Î ψ, then there is a world v ÎÇf such that v ³g u and vÎ ϕ. An equivalent condition is that there is no ψ-world in the modal base that outranks all ϕ-worlds. Assuming the limit assumption, it can be simplified as: the best ϕ-world is at least as highly ranked (by f and g) as the best ψ-world. Based on this, “better possibility” can be defined as follows: ϕ is a better possibility than ψ iff ϕ is as good a possibility as ψ but ψ is not as good a possibility as ϕ. Given the limit assumption, this is equivalent to the condition that the best ϕ-world outranks the best ψ-world. 6 The world-ordering in the classical semantics can be not connected. If so, it will be even weaker than an ordinal scale. 7 According to conflict accounts (Fraassen 1973; Kratzer 1977; von Fintel 2012; Horty 2012; Swanson 2016), “ought: ϕ” is true relative to g iff ϕ follows from some maximally consistent subset of the ordering source g. I am not going 67 4.2. Cariani’s account The next account that I’ll consider is Cariani’s (2013, 2016) account. The main difference between Kratzer’s and Cariani’s accounts is that while the former is based on ordinal orderings on worlds, the latter is based on ordinal orderings on alternatives (= propositions). On Cariani’s account, “ought” involves relativization to alternatives. This is modeled by adding a new parameter ALT, which is a set of alternatives. On his view, the definition of “ought” can be given as follows: Cariani’s (2016) semantics for “ought” (6) ⟦“Ought: 𝜙”⟧ f,ALT,g,w = 1 iff "w' Î ÈBEST(f, ALT, g, w). ⟦𝜙⟧ f,ALT,g,w' = 1 On this view, “Ought: 𝜙” is true relative to f, ALT, and g iff all worlds that belong to the best alternatives are 𝜙-worlds. In other words, “Ought: 𝜙” is true when all best alternatives entail 𝜙. ALT is a set of mutually exclusive propositions that are considered live options in light of the modal base f. That is, it partitions the possible space provided by the modal base (i.e., Çf(w)). The main function of g is to impose an ordering on the propositions in ALT. The relative ranking of alternatives is determined as follows: for two alternatives a and b, a is at least as good as b, (= a ³g b) iff a satisfies every member of g that b does. 8 Based on ³g, the selection function BEST picks out the most ideal propositions in light of g. The ordering ³g, the selection function BEST, and the union of best propositions ÈBEST can be formally defined as follows: (7) For any a, b in ALT, a ³g b iff {πÎg: bÎπ} Í {πÎg: aÎπ} (8) BEST(f, ALT, g, w) = {aÎALT: "bÎALT [b ³g a ® a ³g b]} to discuss the conflict accounts, because they employ the same world-ordering as the classical semantics, and so they face the same problems as the classical semantics does. 8 On Kratzer’s account, propositions order possible worlds, so g is a set of propositions. On Cariani’s account, sets of propositions order alternatives (= propositions), so g is a set of sets of propositions. 68 (9) ÈBEST(f, ALT, g, w) = {w: $pÎBEST(f, ALT, g, w) [wÎp]} Let’s now consider if Cariani’s semantics can capture the data in (2). To explain (2a), Cariani might propose that “Ought: 𝜙 as much as Ought: 𝜓” is true iff 𝜙 is at least as good as 𝜓. Based on this, he can suggest that “Ought: 𝜙 more than Ought: 𝜓” is true iff 𝜙 is as good as 𝜓, but 𝜓 is not as good as 𝜙. Thus, like Krater’s account, Cariani’s account may fare well with (2a). However, Cariani’s account doesn’t fare well with (2b). The problem is that, like the ordering relation between worlds in the classical semantics, the ordering relation between alternatives on Cariani’s account is ordinal. Given that p is the proposition that Bill saves his daughter and q the proposition that Bill saves a stranger, (2b) requires measuring p’s degree, measuring q’s degree, and measuring the difference between the two degrees. However, since we cannot measure the difference between two alternatives on Cariani’s ordinal ordering on alternatives, Cariani’s semantics does not explain the data like (2b). 4.3. Probabilistic semantics Thus far, I have considered two accounts that employ ordinal scales and showed that they fail to capture the meaning of (2b) (i.e., a comparative “ought” with an interval modifier). In the next two subsections, I’ll consider two accounts that employ at least interval scales. Those accounts may fare well with (2b), but that is not enough to fully capture our intuitions. The first analysis that I consider is Finlay’s (2009, 2014, 2016) probabilistic account of deontic “ought”. Probabilistic semantics for “ought” (10) ⟦“Ought: ϕ”⟧ ALT = 1 iff "ψÎALT. Pr(ϕ/E) > Pr(ψ/E). 69 According to (10), “Ought: ϕ” is true relative to ALT iff the probability of ϕ given the relevant end E is greater than the probability of each alternative given E. Since Finlay’s account deploys the probability scale and the probability scale is a ratio, his account fares well with (2a) and (2b). Proponents of the probabilistic semantics can say that the truth-condition of (2a) is given as the probability of Bill’s saving his daughter given the relevant end E is greater than the probability of Bill’s saving a stranger given E (i.e., Pr(Bill saves his daughter/E) > Pr(Bill saves a stranger/E)). More importantly, they can say that the truth-condition of (2b) is given as the difference between the probability of Bill’s saving his daughter given E and the probability of Bill’s saving a stranger given E is much/slight (i.e., Pr(Bill saves his daughter/E) – Pr(Bill saves a stranger/E) is much/slight). However, this account faces the problem of overgeneration. It may successfully capture why (2a) and (2b) sound fine, but it doesn’t capture why (2c) sounds anomalous. It seems reasonable to think that expressions like “probable” or “likely” are associated with the probability scale (Lassiter, 2011, 2017). Given that we throw a fair coin, all of the following sentences sound true. (11) a. The die’s coming up an even number is more probable than the die’s coming up six. b. The die’s coming up an even number is much more probable than the die’s coming up six (interval modifier). c. The die’s coming up an even number is three times as probable as the die’s coming up six (ratio modifier). The probability scale is a ratio, so this explains why (11c) is legitimate. If deontic “ought” takes the probability scale, it seems mysterious why (2c) sounds anomalous while (11c) sounds fine. 70 4.4. Decision-theoretic semantics The next analysis that I consider is decision theoretic accounts of deontic “ought” (Goble, 1996; Lassiter, 2011, 2017), according to which deontic “ought” takes the expected value (EV) scale. Decision-theoretic semantics for “ought” (12) ⟦“Ought: ϕ”⟧ ALT = 1 iff "ψÎALT. EV(ϕ) > EV(ψ). According to (12), “Ought: ϕ” is true relative to ALT iff the expected value of ϕ is greater than the expected value of each alternative in ALT. Since on standard decision theory expected utilities constitute an interval scale (von Neumann & Morgenstern, 2007; Peterson, 2017), this account successfully explains the data in (2). (2a) can be paraphrased as the expected value of Bill’s saving his daughter is greater than the expected value of Bill’s saving a stranger. (2b) can be paraphrased as the difference between the expected value of Bill’s saving his daughter and the expected value of Bill’s saving a stranger is much/slight (i.e., EV(Bill saves his daughter) – EV(Bill saves a stranger) is much/slight). Decision theoretic accounts successfully capture the meanings of these two, because the expected value scale is an interval and interval scales have orderings and differences. Furthermore, they successfully capture (2c) as well. (2c) can be paraphrased as the ratio between the expected value of Bill’s saving his daughter and the expected value of Bill’s saving a stranger is two (i.e., ./('#(( 1+2"1 !#1 3+4$!%",) ./('#(( 1+2"1 + 1%,+5$",) ≥ 2). But since the expected value scale is not a ratio, it does not have ratios between values. Just as “20°C is twice as hot as 10°C” is not true, (2c) cannot be true. Decision theoretic accounts fare well with (2). This is because they are the only accounts that employ non-ratio, interval scales. However, it is not enough to introduce non-ratio, interval scales into our semantics to fully capture our judgments on “ought”-claims. Let’s consider the following two cases: 71 Lottery 1: Mary is asked to choose between two winning lottery tickets: A and B. She knows that both are winning tickets, but she does not know what the prize of each one is. If she chooses A, she will win 1 million dollars. If she chooses B, she will win 10 dollars. (13) Mary ought to choose A. (13) sounds true. This is intuitively so, because the winning prize of A is much higher than the winning prize of B. Now, let’s consider a variant of the case. Lottery 2: Mary is asked to choose between two winning lottery tickets: A and B. She knows that both are winning tickets, but she does not know what the prize of each one is. If she chooses A, she will win 1 million dollars. If she chooses B, she will win 1 million dollars and 1 cent. (14) Mary ought to choose B. My informants had mixed opinions on (14). Some reported that (14) is true in Lottery 2. Others reported that it is not. The data that I wish to explain is the fact that people have mixed opinions on (14). However, (12) doesn’t explain why some people are inclined to judge (14) as not true in Lottery 2. It predicts that (14) is simply true in Lottery 2. One thing to note here is that epistemic “ought” patterns similarly with deontic “ought” in this respect. Consider the following: Raffle: Suppose that Alice is holding 401 tickets in a raffle with 1000 tickets in total. Bob is holding 400 tickets. 199 other people are holding a single ticket. There is only one winner in the raffle. (15) Alice ought to win. People seem to have mixed opinions on (15), as well as (14). That is, some people think that (15) is true, but others think that it is not. Lottery 2 and Raffle suggest that it is not enough just to 72 introduce interval scales into the semantics for “ought” to fully capture our judgments on “ought”- claims. In the next section, I will discuss how to accommodate both comparative “ought”s with interval modifiers (= (2)) and Lottery 2 (= (14)). 5. A Difference-Based Semantics for “Ought” In this section, I develop a theory of “ought”, which explains both comparative “ought”s with interval modifiers and Lottery 2. Then, in Section 6, I discuss how to extend the account developed here to preserve the unity of epistemic and deontic “ought”s. The central idea is that subtraction is encoded in the meaning of “ought”. Since subtraction requires measuring the difference between two values and only interval scales have differences, the following entry requires that “ought” should take interval scales. Difference-Based Semantics for “Ought” (16) ⟦“Ought: ϕ”⟧ ALT,c = 1 iff "ψÎALT. µc(ϕ) – µc(ψ) > θc. According to (16), “ought” is not associated with a particular scale. Instead, its scale is contextually provided. On this view, “Ought: ϕ” is true relative to ALT iff the difference between the degree of ϕ on the contextually given scale and the degree of each option in ALT is greater than the contextual threshold θc. Let’s first consider how this account captures Lottery 2. According to (16), the meaning of “ought” is defined in terms of differences. That is, “ought” measures the difference between two values, and this difference needs to meet a certain contextual threshold. This provides a natural explanation for why people have mixed opinions on Lottery 2. If the threshold for differences (= θc) is zero, people judge (14) to be true. If the threshold is some substantial degree, people judge 73 (14) to be not true. This is because the one-cent difference is too small to meet the contextual threshold. Next, let’s consider how this account allows us to explain comparative “ought”s with interval modifiers. On this view, (2a) can be glossed as the difference between the degree of Bill’s saving his daughter and the degree of Bill’s saving a stranger on the contextually given scale is non-zero (i.e., µc(Bill leaves) – µc(Mary leaves) > 0). (2b) can be glossed as the difference between the degree of Bill’s saving his daughter and the degree of Bill’s saving a stranger is much/slight (i.e., µc(Bill leaves) – µc(Mary leaves) is much/slight). Since (16) requires the scale of “ought” to be an interval, this account successfully captures those meanings. (2c) can be glossed as the ratio between the degree of Bill’s saving his daughter and the degree of Bill’s saving a stranger is two (i.e., 6('#(( ("+2"1) 6(*+,- ("+2"1) ≥ 2). One thing to note here is that (16) does not prohibit “ought” from taking a ratio scale. This means that according to (16), if “ought” takes a ratio scale, it will allow for ratio modifiers, and if it takes a non-ratio, interval scale, it will allow for only interval modifiers. On this view, (2c) sounds bad because deontic “ought”s generally take non-ratio, interval scales such as expected value, importance, goodness, desirability, obligatoriness scales. I will come back to this issue in Section 6. 6. The Unity of Epistemic and Deontic “Ought”s In this section, I show how the account developed here can be extended to accommodate the unity of epistemic and deontic “ought”s. In Section 6.1., I present two puzzles about epistemic and deontic “ought”s, which appear to undermine the unity of “ought”. Then, in Section 6.2., I add a presupposition to my account, and show how this presupposition can be used to explain those puzzles while preserving the unity of “ought”. 74 6.1. Two Puzzles about “Ought” The two puzzles that I discuss concern the asymmetrical behaviors of “ought”. The first concerns its behaviors with respect to ratio modifiers. Compare (2) (= deontic “ought”) with (17) (= epistemic “ought”). While deontic “ought” doesn’t allow for ratio modifiers, epistemic “ought” does. (2) a. Bill ought to save his daughter more than he ought to save a stranger. b. Bill ought to save his daughter much/slightly more than he ought to save a stranger (interval modifiers). c. #Bill ought to save his daughter twice/three times as much as he ought to save a stranger (ratio modifiers). (17) a. The die ought to come up an even number more than the die ought to come up six. b. The die ought to come up an even number much more than the die ought to come up six (interval modifier). c. The die ought to come up an even number three times as much as the die ought to come up six (ratio modifier). The second puzzle concerns the logic of “ought”. More specifically, it concerns its behavior with respect to a logical principle called Union: 9 (18) a. Ought: 𝜙 as much as Ought: 𝜓. b. Ought: 𝜙 as much as Ought: 𝜒. c. ⊨ Ought: 𝜙 as much as Ought: (𝜓 ∨ 𝜒). 9 The analogous principle for “probably” was first mentioned in Yalcin (2010). After that, Lassiter (2011, 2017) discussed Union for “probable” and “good”. Here, I extend the principle to “ought”. 75 Union holds for deontic “ought”s but not for epistemic “ought”s. Let’s first consider epistemic Union. Given that we flip a fair coin, the following argument doesn’t sound valid. The premises are clearly true, but the conclusion is false. Epistemic Union (19) a. The coin ought to come up heads as much as the coin ought to come up tails. b. The coin ought to come up heads as much as the coin ought to come up heads. c. The coin ought to come up heads as much as the coin ought to come up heads or tails. On the other hand, deontic Union is valid. Suppose that Bill has recently experienced many misfortunes. He has lost his job. His child has recently been diagnosed with a serious disease. But he doesn’t have enough savings for his child’s treatment. On the other hand, Mary is living a peaceful life with her family. Bill and Mary both bought a single lottery ticket. If any of them should win, Bill should win the lottery. Deontic Union (20) a. Bill ought to win as much as Mary ought to win. b. Bill ought to win as much as Bill ought to win. c. Bill ought to win as much as Bill or Mary ought to win. Unlike epistemic Union, the above inference sounds fine. The premises are clearly true, and the conclusion is true as well. 6.2. Positivity and Intermediacy The two puzzles presented in the previous section raise a challenge against the unity of “ought”. To solve those puzzles, I propose adding a constraint on the scales of “ought”. More specifically, if the scales of “ought” are restricted to either positive or intermediate scales, we can give a natural 76 explanation of the two puzzles. So, we have good reason to believe that “ought” takes either positive or intermediate scales. Let’s begin with what positivity and intermediacy mean. A scale S is positive when for any two propositions, their disjunction (or join) has a greater degree of the property than either. For example, the probability scale is positive. For any two propositions p and q, the probability of p or q is greater than the probability of p and the probability of q, unless either of them has a zero probability. A scale S is intermediate when for any two propositions, their disjunction has an intermediate degree of the property between the degree of the best and the degree of the worst. For example, the goodness scale is intermediate. For any two propositions p and q, the goodness of p or q is between the goodness of p and the goodness of q. If the goodness of p is greater than the goodness of q, the goodness of p or q is greater than the goodness of q but smaller than the goodness of p. These two properties can be formally defined as: (21) a. A scale S is positive iff for any 𝜙, 𝜓, 𝜇 S(𝜙 ∨ 𝜓) ≥ 𝜇 S(𝜙). b. A scale S is intermediate iff for any 𝜙, 𝜓, 𝜇 S(𝜙) ≥ 𝜇 S(𝜙 ∨ 𝜓) ≥ 𝜇 S(𝜓) or 𝜇 S(𝜓) ≥ 𝜇 S(𝜙 ∨ 𝜓) ≥ 𝜇 S(𝜙). Based on these two notions, I propose that “ought” carries a constraint that it takes either a positive or intermediate scale. On my view, “ought” is not associated with a particular scale (cf., “tall”) but just restricts the range of scales. If it takes a positive scale, it has an epistemic reading. If it takes an intermediate scale, it has a deontic reading. This constraint can be captured as a presupposition or defined-ness condition on “ought”-statements, and the definition of “ought” can be revised as the following: (22) ⟦“Ought: ϕ”⟧ ALT,c is defined iff µc is positive or intermediate if defined, = 1 iff "ψÎALT. µc(ϕ) – µc(ψ) > θc. 77 Let’s first consider how the above constraint can be used to explain the asymmetric behavior of “ought” with respect to ratio modifiers. It seems reasonable to think that “ought” in (17) takes the probability scale. 10 Since the probability scale is a ratio, it shouldn’t be surprising that “ought” in (17) is compatible with ratio modifiers. On the other hand, deontic “ought” takes intermediate scales (e.g., “good”, “obligatory”, “desirable”, “important”, etc.) and intermediate scales are generally a non-ratio. Consider the following: (23) a. #Saving your daughter is twice/three times as good as saving a stranger. b. #Saving your daughter is twice/three times as obligatory as saving a stranger. c. #Saving your daughter is twice/three times as desirable as saving a stranger. d. #Saving your daughter is twice/three times as important as saving a stranger. As (23) shows, the scales of goodness, obligatoriness, desirability, and importance are not compatible with ratio modifiers. So, when “ought” takes one of those scales, it is expected to be incompatible with ratio modifiers. Let’s move on to Union. Let me begin with epistemic Union. Since epistemic “ought” ranges over positive scales, even if (18a) and (18b) are true and so the degree of 𝜙 is at least as great as the degree of 𝜓 and the degree of 𝜒, there is no guarantee that the degree of 𝜙 is at least as great as the degree of 𝜓 ∨ 𝜒. This is because by positivity, the degree of 𝜓 ∨ 𝜒 is generally greater than the degree of 𝜓 and the degree of 𝜒. Let’s now consider deontic Union. Since deontic “ought” is intermediate, (A3) is true. Suppose also that for reductio, (18a) and (18b) are true but (18c) is false. Then, we can show that the following four assumptions are inconsistent. 10 Yalcin (2016) claims that the seemingly epistemic “ought” expresses a normality reading, which he calls a pseudo- epistemic reading. (22) lets me remain neutral on this issue. This is because the normality scale is positive, and so (22) is compatible with either the probability or normality scale. 78 (A1) 𝜇 D(𝜙) ≥ 𝜇 D(𝜓). (by (18a)) (A2) 𝜇 D(𝜙) ≥ 𝜇 D(𝜒). (by (18b)) (A3) (i) 𝜇 D(𝜓) ≥ 𝜇 D(𝜓 ∨ 𝜒) ≥ 𝜇 D(𝜒) or (ii) 𝜇 D(𝜒) ≥ 𝜇 D(𝜓 ∨ 𝜒) ≥ 𝜇 D(𝜓). (by Intermediacy) (A4) 𝜇 D(𝜓 ∨ 𝜒) >𝜇 D(𝜙). (By the negation of (18c)) Given Intermediacy (= (A3)), either (i) or (ii) is the case. Suppose that (i) is the case. Then, 𝜇 D(𝜓) ≥ 𝜇 D(𝜓 ∨ 𝜒). By (A1) and (A4), 𝜇 D(𝜓 ∨ 𝜒) > 𝜇 D(𝜓). But this is a contradiction. Thus, (i) cannot be the case. Suppose that (ii) is the case. Then, 𝜇 D(𝜒) ≥ 𝜇 D(𝜓 ∨ 𝜒). By (A2) and (A4), 𝜇 D(𝜓 ∨ 𝜒) > 𝜇 D(𝜒). But this is a contradiction. Thus, (ii) cannot be the case. Therefore, if (A1), (A2), (A3) are true, (A4) must be false and the conclusion of Union must be true. That is, given Intermediacy, deontic Union is valid. 7. Conclusion A wide range of theories including the classical semantics, Cariani’s contrastive semantics, and conflict accounts share something striking in common. It is the idea that “ought” claims deploy an ordinal scale. However, I argued in this paper that this pervasive assumption is mistaken. Presenting some new data, I showed that deontic “ought” claims require at least an interval scale, and why this creates a problem of adequacy for a wide range of existing views. Furthermore, I argued, by considering the probabilistic and decision theoretic views, that it is not enough just to introduce interval scales into our semantics. I developed and defended a difference-based semantics on “ought”-claims, and showed how this account can be extended to capture the unity of epistemic and deontic “ought”s. 79 References Bach, E. (1979). Control in Montague grammar. Linguistic Inquiry, 10, 515-31. Baker, M. (1988). Incorporation: A theory of grammatical function changing. Chicago: University of Chicago Press. Baker, M. (1997). Thematic roles and syntactic structure. In L. Haegeman (ed.), The elements of grammar (pp. 73-137). Dordrecht: Kluwer. Baker, M. (2003). Lexical categories: Verbs, nouns, and adjectives. Cambridge University Press. Baker, M., Johnson, K., & Roberts, I. (1989). Passive arguments raised. Linguistic Inquiry, 20, 219–51. Bhatt, R. (1998). Obligation and possession. In H. Harley (ed.), Proceedings of the UPenn/MIT Workshop on argument structure and aspect (pp. 21–40). MITWPL. Bhatt, R., & Pancheva, R. (2017). Implicit arguments. In M. Everaert, & H. C. van Riemsdijk (eds.), The wiley blackwell companion to syntax (2 nd edition, pp. 1-35). John Wiley & Sons, Inc. Blumberg, K., & Hawthorne, J. (2021). Inheritance: Professor Procrastinate and the logic of obligation. Philosophy and Phenomenological Research (online first). Borer, H. (2005). Structuring sense. New York: Oxford University Press. Brennan, V. (1993). Root and epistemic modal auxiliary verbs. Ph.D. Dissertation, UMass, Amherst. Broome, J. (2013). Rationality through reasoning. Wiley-Blackwell. Cariani, F. (2013). ‘Ought’ and resolution semantics. Nous, 47(3), 534–558. Cariani, F. (2016). Deontic modals and probabilities: one theory to rule them all? In N. Charlow & M. Chrisman (eds.), Deontic modality (pp.11–46). New York: Oxford University Press. Cariani, F., & Santorio, P. (2018). Will done better: Selection semantics, future credence, and indeterminacy. Mind, 127, 129–165. Carnie, A. (2013). Syntax: A generative introduction (3 rd edition). Wiley-Blackwell. Chisholm, R. (1964). The ethics of requirement. American Philosophical Quarterly, 1, 147–153. Chomsky, N. (1981). Lectures on government and binding. Foris Publications. 80 Chomsky, N. (1982). Some concepts and consequences of the theory of government and binding. Cambridge, Mass.: MIT Press. Chrisman, M. (2008). Ought to believe. The Journal of Philosophy, 105(7), 346–370. Chrisman, M. (2012). ‘Ought’ and control. Australasian Journal of Philosophy, 90(3), 433–451. Chrisman, M. (2015). The meaning of “ought”. Oxford University Press. Collins, J. (2007). Syntax, more or less. Mind, 116, 805–50. Collins, J. (2017). The semantics and ontology of the average American. Journal of Sematnics, 34(3), 373–405. Condoravdi, C. (2002). Temporal interpretation of modals: Modals for the present and for the past. In D. Beaver, L. Casillas, B. Clark, & S. Kaufmann (eds.), The construction of meaning. Palo Alto, CA: CSLI Publications. Dowty, D. (1989). On the semantic content of the notion ‘thematic role’. In B. Partee, G. Chierchia, & R. Turner (eds.), Properties, types and meaning (pp. 69–129). Dordrecht: Reidel. Dowty, D. (1990). Thematic proto-roles and argument selection. Language, 67, 547–619. Epstein, S. D. (1984). Quantifier-pro and the LF representation of PROarb. Linguistic Inquiry, 15(3), 499–505. Feldman, F. (1986). Doing the best we can: An essay in informal deontic logic. Dordrecht, Boston; D. Reidel Publishing Company. Feldman, R. (2001). Voluntary belief and epistemic evaluation. In M. Steup (ed.), Knowledge, Truth, and Duty (pp. 77–92). Oxford: Oxford University Press. Fine, K. (1985). Reasoning with arbitrary objects. New York: Blackwell. Finlay, S. (2009). Oughts and ends. Philosophical Studies, 143(3), 315-340. Finlay, S. (2014). Confusion of tongues: A theory of normative language. New York: Oxford University Press. Finlay, S. (2016). Ought: out of order. In N. Charlow & M. Chrisman (eds.), Deontic modality (pp. 169-199). New York: Oxford University Press. Finlay, S., & Snedegar, J. (2014). One ought too many. Philosophy and Phenomenological Research, 89(1), 102–124. 81 von Fintel, K. (2012). The best we can (expect to) get? challenges to the classic semantics for deontic modals. Paper presented in a session on deontic modals at the Central APA (2/17/2012). von Fintel, K., & Heim, I. (2011). Intensional semantics. Unpublished Lecture Notes. von Fintel, K., & Iatridou, S. (2008). How to say ought in foreign: The composition of weak necessity modals. In J. Gueron, & J. Lecarme (eds.), Time and modality (pp.115–141). Springer. van Fraassen, B. (1973). Values and the heart’s command. Journal of Philosophy, 70(1), 5–19. Geach, P. (1982). Whatever happened to deontic logic? Philosophia, 11, 1-12. Goble, L. (1996). Utilitarian deontic logic. Philosophical Studies, 82(3), 317–357. Grimshaw, J. (1990). Argument structure. Cambridge, MA: MIT Press. Hacquard, V. (2009). On the interaction of aspect and modal auxiliaries. Linguistics and Philosophy, 32, 279-315. Hacquard, V. (2010). On the event relativity of modal auxiliaries. Natural Language Semantics, 18, 79–114. Hale, K., & Keyser, S. J. (2002). Prolegomenon to a theory of argument structure. Cambridge, Mass.: MIT Press. Harley, H. (1995). Subjects, events, and licensing. Ph.D. Dissertation. MIT. Harman, G. (1973). Review of ‘the significance of sense: meaning, modality, and Morality’. The Philosophical Review, 82(2), 235–9. Heim, I. (1982). The semantics of definite and indefinite noun phrases. Ph.D. Dissertation. UMass, Amherst. Heim, I. (2010). Degree operators and scope. In B. Jackson & T. Matthews (eds.), Proceedings of SALT 10. (pp. 40–64). Ithaca, NY: Cornell University. Heim, I., & Kratzer, A. (1998). Semantics in Generative Grammar. Malden, MA: Blackwell. Hieronymi, P. (2006). Controlling attitudes. Pacific Philosophical Quarterly, 87(1), 45–74. Hornstein, N. (1999). Movement and control. Linguistic Inquiry, 30, 69–96. Horty, J. (2001). Agency and deontic logic. Oxford: Oxford University Press. 82 Horty, J. (2012). Reasons as defaults. Oxford University Press. Horty, J., & Belnap, N. (1995). The deliberative stit: A study of action, omission, ability, and obligation. Journal of Philosophical Logic, 24, 583–644. Humberstone, L. (1971). Two sorts of ‘ought’s. Analysis, 32, 8–11. Jackendoff, R. (1972). Semantic interpretation in generative grammar. Cambridge, MA: MIT Press. Jackson, F. (1985). On the semantics and logic of obligation. Mind, 94, 177–195. Jaeggli, O. (1986). Passive. Linguistic Inquiry, 17, 587–622. Kaplan, D. (1989). Demonstratives. In J, Almog, J, Perry, & H, Wettstein (eds.), Themes from Kaplan (pp. 481 – 563). New York: Oxford University Press. Kaufmann, M. (2012). Interpreting imperatives. New York: Springer. Kennedy, C. (1999). Projecting the adjectives. New York: Garland. Kennedy, C. (2007). Vagueness and grammar: The semantics of relative and absolute gradable adjectives. Linguistics and Philosophy, 30(1), 1–45. Kennedy, C & McNally, L. (2005). Scale structure, degree modification, and the semantics of gradable predicates. Language, 81(2), 345–381. King, J. (1991). Instantial terms, anaphora, and arbitrary objects. Philosophical Studies, 61, 239– 265. Krantz, D. H., Luce, R. D., Suppes, P., & Tversky, A. (1971). Foundations of measurement: Additive and polynomial representations. San Diego, London: Academic Press, University of Fribourg. Kratzer, A. (1977). What ‘must’ and ‘can’ must and can mean. Linguistics and Philosophy, 1, 337-55. Kratzer, A. (1981). The notional category of modality. In H. J. Eikmeyer, & H. Rieser (eds.), Words, worlds, and contexts. Berlin and New York: de Gruyter. Kratzer, A. (1991). Modality. In A. von Stechow, & D. Wunderlich (eds.), Semantics: An international handbook of contemporary research (pp. 639–650). Berlin: de Gruyter. Kratzer, A. (1996). Severing the External Argument from its Verb. In J. Rooryck, & L. Zaring (eds.), Phrase structure and the lexicon (pp.109–137). Springer. 83 Kratzer, A. (2012). Modals and conditionals. Oxford University Press. Lebeaux, D. (2009). Where does binding theory apply? Cambridge, Mass.: MIT Press. Lassiter, D. (2011). Measurement and modality: The scalar basis of modal semantics. Ph.D. dissertation. New York University. Lassiter, D. (2017). Graded modality. Oxford University Press. Lee, J. (2021). “Ought” and intensionality. Synthese, 199, 4621–4643. Levin, B. (1993). English verb classes and alternations. Chicago, London: The University of Chicago Press. Levin, B., & Rappaport Hovav, M. (1995). Unaccusativity: At the syntax-lexical semantics interface. Cambridge, Mass.: MIT Press. Lewis, D. (1973). Counterfactuals. Oxford: Blackwell. Lewis, D. (1980). Index, context, and content. In S, Kanger & S, Öhman (eds.), Philosophy and grammar (pp. 79–100). Dordrecht: Reidel. MacFarlane, J., & Kolodny, N. (2010). Ifs and oughts. Journal of Philosophy, 107, 115–43. Manzini, R., & Roussou, A. (2000). A minimalist theory of A-movement and control. Lingua, 110, 409–447. Marantz, A. (1984). On the nature of grammatical relations. MIT Press. Moltmann, F. (2006). Generic one, arbitrary PRO, and the first person. Natural Language Semantics, 14, 257–281. Moltmann, F. (2010). Generalizing detached self-reference and the semantics of generic one. Mind and Language, 25(4), 440–473. Moore, G. E. (1922). Philosophical studies. London: Kegan Paul. von Neumann, J. & Morgenstern, O. (2007). Theory of games and economic behavior (60 th anniversary commemorative edition). Princeton: Princeton University Press. Peterson, M. (2017). An introduction to decision theory (2 nd edition). Cambridge University Press. Pietroski, P. (2018). Conjoining meanings. Oxford: Oxford University Press. Piñón, C. (2003). Being able to. In G. Garding & M. Tsujimura (eds.), WCCFL 22 Proceedings (384–397). Somerville, MA: Cascadilla Press. 84 Portner, P. (2009). Modality. Oxford: Oxford University Press. Portner, P & Rubinstein. A. (2016). Extreme and non-extreme deontic modals. In M. Chrisman & N. Charlow (eds.), Deontic modality. (pp. 256–282). Oxford University Press. Postal, P. (1970). On coreferential complement subject deletion. Linguistic Inquiry, 1, 439-500. Postal, P. (1974). On raising. Cambridge, MA: MIT Press. Price, A. (2008). Contextuality in practical reason. Oxford: Oxford University Press. Quine, W.V.O. (1960). Word and object. Cambridge, Mass.: The MIT Press. Ramchand, G. C. (2008). Verb meaning and the lexicon: A first phase syntax. New York: Cambridge University Press. Ross, J. 2010. Personal and impersonal obligation. Journal of Philosophical Logic, 39, 307–23. Sassoon, G. (2010). Measurement theory in linguistics. Synthese, 174, 151–180. Schroeder, M. (2007). Reasons and agent-neutrality. Philosophical Studies, 135, 279–306. Schroeder, M. (2011). Ought, agents, and actions. Philosophical Review, 120, 1–41. Schwarzschild, R. (2005). Measure phrases as modifiers of adjectives. Recherches Linguistiques de Vincennes, 34, 207-228. Sloman, A. (1970). ‘Ought’ and ‘better.’ Mind, 79, 385–394. Snedegar, J. (2012). Contrastive semantics for deontic modals. In M. Blaauw (ed.), Contrastivism in philosophy: New perspectives. London: Routledge. von Stechow, A. (1984). Comparing semantic theories of comparison. Journal of Semantics, 3, 1–77. Stevens, S. S. (1946). On the theory of scales of measurement. Science, 103, 677–680. Swanson, E. (2016). The application of constraint semantics to the language of subjective uncertainty. Journal of Philosophical Logic, 45(2), 121–146. Villalta, E. (2008). Mood and gradability: An investigation of the subjunctive mood in Spanish. Linguistics and Philosophy, 31(4), 467–522. Wedgwood, R. (2006). The meaning of ‘ought’. In Russ Shafer-Landau (Ed.), Oxford studies in metaethics (vol. 1, pp.127–160). Oxford: Oxford University Press. 85 Wedgwood, R. (2007). The nature of normativity. Oxford: Oxford University Press. Wellwood, A. (2019). The meaning of “more”. Oxford University Press. Williams, B. (1981). Ought and moral obligation. In Moral Luck (pp. 114–123). Oxford: Oxford University Press. Wurmbrand, S. (1999). Modal verbs must be raising verbs. In S. Bird, A. Carnie, J. Haugen, & P. Norquest (eds.), The proceedings of the 18 th west coast conference on formal linguistics (pp. 599–612). Somerville, MA: Cascadilla Press. Yalcin, S. (2010). Probability operators. Philosophy Compass, 5(11), 916–937. Yalcin, S. (2016). Modalities of normality. In M. Chrisman, & N. Charlow (eds.), Deontic modality (pp. 230–255). Oxford University Press. Zhang, L., & Ling, J. (2021). The semantics of comparatives: A difference-based approach. Journal of Semantics, 38, 249–303. 86

Abstract (if available)

Abstract This dissertation defends the uniformity of “ought” by developing a new analysis of deontic “ought”. Philosophers and linguists have paid particular attention to the meaning of “ought” for various reasons. For example, moral philosophers have been interested because they are concerned with the nature of obligations and practical deliberations, and “ought” traditionally has been used to express our moral obligations and practical deliberations. Semanticists have been interested because “ought” can be used to express such a wide range of interpretations: epistemic, prudential, deliberative, and evaluative, among others. Logicians have been interested because the logic of “ought” seems to change, depending on its interpretation.
In my dissertation, I confront three of the most significant challenges to the classic view on “ought”-statements and show how we can solve those problems without losing the main advantages of the classic view. The classic view is committed to the following three theses: (i) “ought” uniformly expresses a propositional operator, (ii) “ought” is evaluated relative to possible worlds, and (iii) “ought” is associated with an ordinal scale. Each chapter of my dissertation is a self-standing paper and focuses on a different aspect of the classic view.
The first two chapters of the dissertation concern what is sometimes called the distinction between “ought-to-do”(= deliberative) and “ought-to-be”(= evaluative) interpretations of “ought”. Compare, for example, (1) “John ought not to kill innocent people” to (2) “There ought to be world peace.” (1) typically expresses a deliberative reading and can be paraphrased as “John is obliged to, or has most reason to, not kill innocent people. (2) typically expresses an evaluative reading and can be paraphrased as “It’d be good for there to be world peace.” Some philosophers claim that this distinction requires us to posit an ambiguity in “ought” itself. But in Chapter 1 “‘Ought’ and Intensionality”, I develop a new test, which I call the intensionality test, to verify this claim. This test suggests that “ought” uniformly expresses a propositional operator.
In Chapter 2 “‘Ought,’ Agents, and Events”, I consider the main arguments for the ambiguity of “ought”, one of which is that pairs like (3) “Bill ought to kiss Lucy” and (4) “It ought to be that Bill kisses Lucy” do not have the same range of readings. This is alleged to be a fatal obstacle for the propositional operator view of “ought” because (3) and (4) have truth-conditionally equivalent prejacents. Many responses to this argument have been offered in print, but all of them fail to appreciate its central flaw. This flaw is that they have failed to appreciate that despite their truth-conditional equivalence, the prejacents of (3) and (4) differ in their event structure. My solution suggests that “ought” is sensitive to events, not possible worlds. It therefore defends the first, most powerful, feature of the classic view by relaxing its second, less well-motivated feature. My view also sheds light on why the moral principle that “ought” implies “can” holds.
In Chapter 3 “‘Ought’ and Intervals”, I tackle the question “What kind of scale is ‘ought’ associated with?” It’s generally accepted that there are at least four levels of measurement: nominal, ordinal, interval, and ratio. Most theories of “ought” claim that “ought” is associated with an ordinal scale. For example, on the classic semantics, we can ask whether one world is better than another. But we cannot ask how much one world is better than another. In this paper, I present some new pieces of data and argue that “ought” requires at least an interval scale. The scale to which “ought” is sensitive turns out to have a number of other very important applications such as why the logic of “ought” may change depending on its interpretation.
To summarize, each chapter has the following implication on semantic theorizing of “ought”:
Ch1: “Ought” always operates on propositions, not actions.
Ch2: The contextual parameters for “ought” (e.g., conversational backgrounds in the classical semantics, alternative sets in the contrastive semantics) are sensitive to events, not worlds.
Ch3: “Ought” requires at least an interval scale, not an ordinal scale.
These implications may not determine a unique semantics of “ought” but put significant constraints on semantic theorizing of “ought”.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Constraining assertion: an account of context-sensitivity

PDF

Reasons, obligations, and the structure of good reasoning

PDF

Telling each other what to do: on imperative language

PDF

Conceptually permissive attitudes

PDF

The grammar of individuation, number and measurement

PDF

Reasoning with uncertainty and epistemic modals

PDF

Rationality and the primacy of the occurrent

PDF

Processing the dynamicity of events in language

PDF

Units of agency in ethics

PDF

Contrastive reasons

PDF

Meaningfulness, rules, and use-conditional semantics

PDF

Responding to harm

PDF

A perceptual model of evaluative knowledge

PDF

The grammar of correction

PDF

Getting our act(s) together: the normativity of diachronic choice

PDF

Public justification beyond legitimacy

PDF

Beliefs that wrong

PDF

A deontological explanation of accessibilism

PDF

Vagueness, ontology, and the social world

PDF

Positivist realism

Asset Metadata

Creator Lee, Junhyo (author)

Core Title What "ought" ought to mean

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Philosophy

Degree Conferral Date 2022-05

Publication Date 04/19/2022

Defense Date 05/13/2022

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag alternatives,context-sensitivity,control syntax,deliberative "ought",deontic modality,evaluative "ought",events,intensionality,interval,OAI-PMH Harvest,ordinal,raising syntax,ratio,scale,theta-roles

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Finlay, Stephen (committee chair), Schroeder, Mark (committee chair), Cariani, Fabrizio (committee member), Rudin, Deniz (committee member), Wellwood, Alexis (committee member)

Creator Email junhyole@usc.edu,junhyolee85@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC111014555

Unique identifier UC111014555

Document Type Dissertation

Format application/pdf (imt)

Rights Lee, Junhyo

Type texts

Source 20220419-usctheses-batch-929 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu