Subjective Meaning: Alternatives to Relativism 9783110402001, 9783110374728

A dish may be delicious, a painting beautiful, a piece of information justified. Whether the attributed properties "

194 60 1MB

English Pages 259 [260] Year 2016

Report DMCA / Copyright


Polecaj historie

Subjective Meaning: Alternatives to Relativism
 9783110402001, 9783110374728

Table of contents :
Subjective meaning: An introduction
If expressivism is fun, go for it!
Doing without judge dependence
Predicates of personal taste and the evidential step
Contextualism and disagreement about taste
Two kinds of subjectivity
Evaluative propositions and subjective judgments
Predicates of experience
Propositions and implicit arguments carry a default general point of view
Subjective meaning and modality

Citation preview

Subjective Meaning

Linguistische Arbeiten

Edited by Klaus von Heusinger, Gereon Müller, Ingo Plag, Beatrice Primus, Elisabeth Stark and Richard Wiese

Volume 559

Subjective Meaning Alternatives to Relativism Edited by Cécile Meier and Janneke van Wijnbergen-Huitink

ISBN 978-3-11-037472-8 e-ISBN (PDF) 978-3-11-040200-1 e-ISBN (EPUB) 978-3-11-040211-7 ISSN 0344-6727 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutschen Nationalbibliografie; detailed bibliographic data are available on the Internet at © 2016 Walter de Gruyter GmbH, Berlin/Boston Satz: le-tex publishing services GmbH, Leipzig Printing: CPI books GmbH, Leck ♾ Printed on acid-free paper Printed in Germany

Contents Preface | VII Janneke van Wijnbergen-Huitink Subjective meaning: An introduction | 1 Daniel Gutzmann If expressivism is fun, go for it! | 21 Sanna Hirvonen Doing without judge dependence | 47 Lavi Wolf Predicates of personal taste and the evidential step | 69 Dan Zeman Contextualism and disagreement about taste | 91 Christopher Kennedy Two kinds of subjectivity | 105 Carla Umbach Evaluative propositions and subjective judgments | 127 Christine Gunlogson und Gregory Carlson Predicates of experience | 169 Tom Roeper Propositions and implicit arguments carry a default general point of view | 201 Michael Hegarty Subjective meaning and modality | 227 Index | 249

Preface This volume grew out of a workshop Subjective meaning: Alternatives to relativism that was held at the Humboldt-University in Berlin as part of the 32nd meeting of the Deutsche Gesellschaft für Sprachwissenschaft (DGfS) on February 23-26, 2010 with the invited speakers, Christopher Kennedy (University of Chicago) and Kai von Fintel (Massachusetts Institute of Technology). A satellite workshop was organized at Frankfurt University on February 22, 2010, with just the same two invited speakers. The contributions to the workshop and this volume all circle around one question: What does it mean to pronounce a subjective judgment? The goal of the workshop was to bring together scholars that agreed to disagree with one of the predominant explanations to the so-called “problem of faultless disagreement”, namely relativism; the idea that the truth of a sentence depends on the opinion of the person who makes a judgment of its truth value rather than on objective matters of fact. All articles in this volume explore new, non-relativistic explanations for faultless disagreement and thereby seek to settle the debate about the indispensableness of relativism, which received a lot of attention recently, both within linguistics and the philosophy of language. One variety of the problem of faultless disagreement occurs with “predicates of personal taste” like yummy or yucky. A judgment that contains a predicate of personal taste may be contradicted without any fault. Spinach may be yucky from the point of view of one person and yummy from the point of view of another. Another variety of the problem of faultless disagreement occurs with “epistemic modals” like might, for example. A dialogue about the whereabouts of some keys may start like this: “Did you see my keys?”—“They might be on your desk.” The person looking for her keys may contradict the last utterance: “No, they aren’t, I already checked the desk”. But she does not mean that the speaker suggesting a search on the desk made a false statement. One of the dialogue partners just didn’t have enough information about the keys and considered something possible that the other partner already knew to be impossible (See Kai von Fintel’s presentation on accessed March 16,2016). Proponents of relativism solve the problem of faultless disagreement for both varieties of subjective meaning in a similar manner: They relate the notion of truth to the perspective of a judge (see MacFarlane, 2007 for epistemic modals, and Lasersohn, 2005 for predicates of personal taste). Indeed, the two varieties have in common that subjective judgments are somehow not really false (although contradicted).

VIII | Preface

One of the results of the workshop, however, was that the two varieties of subjective judgments are of a fundamentally different kind: We may argue endlessly who is right in case of a judgment about personal taste. Typically, such arguments may never be settled. But in the case of a subjective judgment with an epistemic modal, we never argue who is right. This suggests that subjectivity is not a homogeneous phenomenon, but may come in different guises. Accordingly, different tests for subjectivity need not play out the same. One such diagnostic is the acceptability of embedding under a subjective attitude verb like find. It turns out that not all expressions over which we can faultlessly disagree pass the find-test. Epistemic modals are a case in point (#I find it might be raining). The contributions in this volume emphasize the similarities between predicates of personal taste and epistemic modals on the one hand and more objective adjectives and modals on the other hand. They take these as a starting point for the analysis. It is argued that putative subjective expressions display a great variety of uses in faultless disagreement scenarios. In many cases, the speaker seems to intend to make a claim about the preferences of a group, or even about the preferences of people in general, rather than just speaking for herself. Some people conclude from this that faultless disagreement is an illusion. At the very least, this makes subjective expressions much more mundane than previously thought, casting doubt on the contention that subjectivity isn’t amenable to a treatment in terms of standard formal semantics. Finally, several papers in this volume highlight the pragmatic peculiarities of judgments about personal taste and epistemic modal judgments in dialogue situations. While one can disagree with a speaker who calls something yucky, her statement becomes immune to denial when she explicitly relativizes her statement to her own perspective, for instance by saying yucky for me. Again, a relativistic semantics appears unnecessary, if we can explain the restrictions on felicitous (dis)agreement in general pragmatic terms. We would like to thank all contributors to the workshops and to this book for their enthusiasm to undertake new and fruitful investigations in this fascinating area of research, leading to such an excellent set of papers. Special thanks go to Jeanette Nüssli-Gut (ETH Zürich), who introduced the Berlin workshop with a talk on “Language in food sensory science: Methodological considerations” and let us all experience one of the five basic tastes (together with sweet, sour, bitter and salty): umami (from Japanese umai, which may be interpreted as a subjective valuation ‘delicious’, or as describing the actual sensory experience ‘meaty taste’). Sensory science, as applied to food industry, on the one hand rests on objective judgments of trained people and on the other hand collects subjective, hedonistic judgments of naive consumers. Nüssli-Gut’s talk illustrated how this last type of judgment is influenced by various factors, including age and cultural background,

Preface |


and how retasting may contribute to settling the debates that may ensue due to individual differences between consumers. We heartily thank several external anonymous peer reviewers for their judgment and advice. The quality of the papers is also their merit. Furthermore, we would like to thank the Freunde und Förderer der Universität, Frankfurt for their generous financial support of the workshops in Frankfurt and Berlin. We thank Klaus von Heusinger, our series editor, for comments on the content of the volume, his continued support and encouragement, Susanne Trissler and Ayline Heller for their help in preparing the final manuscript, Daniel Gietz (De Gruyter’s editor) for the always friendly feedback and patience, and Stephan Korell at le-tex, for quickly responding to our technical questions. Cécile Meier and Janneke van Wijnbergen-Huitink Frankfurt am Main and Nijmegen, March 2016

References Lasersohn, Peter. 2005. Context dependence, disagreement, and predicates of personal taste. Linguistics and Philosophy 28(6). 643–686. MacFarlane, John. 2007. Relativism and disageement. Philosophical Studies 132(1). 17–31.

Janneke van Wijnbergen-Huitink

Subjective meaning: An introduction Abstract: This introductory chapter traces some of the considerations on the basis of which relativistic approaches to subjective meaning became en vogue. In doing so, the chapter provides an overview of the relevant linguistic and philosophical issues when developing a treatment of subjectivity. In addition, the subsequent chapters of this volume are positioned with respect to these issues.

1 The case for relativism Subjective meaning is expressed by sentences whose truth intuitively depends on personal preferences, rather than on objective matters of fact. Prime examples involve so-called predicates of personal taste such as tasty and fun: (1)

a. b.

This cake is tasty. Roller coasters are fun.

The truth of sentences like (1a) and (1b), it seems, can vary from one speaker to the next. An adequate description of the meaning of these sentences should thus take into account the person on whose opinion the statement is based. In what follows, we will call this person “the judge”. As a first stab, we might postulate that the judge is an argument of predicates of personal taste and that their judge argument slot is usually filled with a covert PP featuring a deictic pronoun referring to the speaker: (2)

a. b.

[[tasty]]c,w = λx.λy.x tastes good to y in w [[tasty (for me)]]c,w = λx.x tastes good to the speaker of c in w

Then, the truth-value of sentences like (1a) depends on who the speaker is. If Ann utters the sentence, it might be true, whereas the very same sentence could be false when said by Bob. In this sense, then, the truth of (1a) varies with the judge. Because the identity of the judge thus varies with the utterance context, this type of analysis is commonly termed “contextualism”.

Janneke van Wijnbergen-Huitink, Department of Philosophy, Theology and Religious Studies, Radboud University Nijmegen, Postbus 9103, 6500 HD Nijmegen, The Netherlands, e-mail: [email protected]

2 | Janneke van Wijnbergen-Huitink

Lasersohn (2005) famously argues that contextualism is unsatisfactory. The problem is that it fails to explain that we can have “disagreements” about taste: (3)

a. b.

Ann: This cake is tasty! Bob: No, it isn’t. It’s disgusting.

Here Bob denies what Ann said. But according to contextualim, he has little reason to do so. Given the contextualist analysis in (2b), statements containing predicates of personal taste express different propositions relative to different speakers. It follows that what Bob said does not contradict the proposition expressed by Ann. Therefore, the two are not disagreeing any more than they would be if Ann said: “I am a doctor” and Bob said: “I am not a doctor”. Why then, should Bob deny what Ann said in (3)? As an alternative, Lasersohn proposes that the judge does not come in when determining the content of sentences like (1a), but only later, when determining their truth-value. This means that truth is relativized, such that sentences are never plain true or false, but they are always true or false relative to a specific judge. Accordingly, this proposal is referred to as “relativism” (other proponents of relativism include Bylinina, 2015; Egan, Hawthorne & Weatherson, 2005; Kölbel, 2002, 2004; MacFarlane, 2011; Stephenson, 2007b). More specifically, Lasersohn (2005, 2009) adds a “judge index of evaluation” to the familiar world index:¹ (4)

[[tasty]]c,w,j = λx.x tastes good to j in w

Rather than treating the judge as an implicit argument of the predicate, relativism treats it as a parameter in the circumstance of evaluation. This has the effect that the proposition expressed by (1a) can be true-for-Ann, but false-for-Bob. At the same time, the proposition cannot be both true and false judged from the same perspective. There is thus a contradiction at the level of (Kaplanian) content (though contents are now sets of world judge pairs rather than sets of worlds), and this conflict accounts for the sense of disagreement in disputes like (3). Although in this system all sentences are evaluated relative to a judge index, in many cases the identity of the judge makes no difference. That is, objective sentences like (5), which express matters of fact, have the same truth-value for all judges.

1 Relativism isn’t new. In fact, in the early 1970s, it was quite common for semanticists to employ individuals as evaluation indices. For example, Lewis (1970) included an index for the speaker. Of course, since Kaplan (1989), we prefer to think of the speaker as a contextual parameter, rather than as an index of evaluation. See e.g., Glanzberg (2007) for a discussion of this development.

Subjective meaning: An introduction


This cake contains sugar.


[[contains sugar]]c,w,j = λx. x contains sugar in w.



The denotation of the predicate contains sugar, given in (6), does not refer to the judge index, so only the facts in w will be relevant for its evaluation, and if (6) is true-for-Ann, it must also be true-for-Bob. The identity of the judge may vary, depending on the perspective that is taken by the speaker. Normally when we assert or assess something, we adopt an “autocentric” perspective, taking ourselves to be the relevant judge. But we may also take an “exocentric” stance, which happens in free indirect discourse (where the judge is the person whose thoughts are expressed), in questions (where the judge is the addressee) and when we speak for someone else, typically someone who is unable to report his own perspective, as in: (7)

a. b.

Ann: How’s that new cat food you bought? Bob: I think it’s tasty, because the cat has eaten a lot of it.


a. b.

Ann: How did Billy enjoy the rides? Bob: Well, the merry-go-round was fun, but the water slide was a little too scary.

In (7), Bob presumably did not try the cat food himself. And although in (8) Bob may have accompanied Billy on the rides, it is clear that Bob’s experiences are not under discussion, and what he said is true just in case Billy enjoyed the merry-goround (Bob may have found it rather dull).² To sum up, although contextualism and relativism yield the same truthpredictions (cf. Stojanovic, 2007), they are not equivalent. They make diverging predictions for disagreements like (3). The relativist analysis offers a straightforward explanation for this phenomenon, in terms of Ann and Bob having uttered sentences with contradictory contents.³ According to Lasersohn, no other theory provides as good an account of disagreements about taste. Others, however, think 2 Contextualism would deal with such examples in terms of the predicate combining with different judge PPs. For example, Bob’s statement in (7) is to be analyzed as I think it’s tasty (to the cat). 3 When predicates of personal taste occur in embedded positions, the two theories also come apart. Consider: (i)

Everyone had a tasty dessert.

It seems this can be true even though the people quantified over may have very different standards of taste. If indeed the judge can vary with the interpretation of the quantifier, this prima facie speaks in favor of contextualism (unless one is willing to analyze quantifiers as shifting the judge

4 | Janneke van Wijnbergen-Huitink

that more sophisticated forms of contextualism come close enough (e.g. Barker, 2013; von Fintel & Gillies, 2008, 2011; Glanzberg, 2007; Moltmann, 2010; Pearson, 2013; Stojanovic, 2007; Sundell, 2011). Note that contextualism isn’t married to the judge being the speaker, but the judge argument slot of predicates of personal taste may be filled with material referring to a group of judges (This cake is tasty (for us / for most people)). However, as we will see in the next section, while such a move may predict disagreement, it cannot predict that disagreements about taste are “faultless”, in the sense that both interlocutors have said something true. Some scholars doubt the explanatory power of the relativist move (e.g. Moltmann, 2010; Stojanovic, 2007). Presumably, Ann and Bob are competent speakers who fully understand the meaning of predicates of personal taste like tasty and disgusting. If so, they are aware of the fact that whenever these predicates are correctly applied to some object, there is some perspective relative to which the application was made. Furthermore, they realize that assertions are normally made from an autocentric perspective. But then, why should Bob deny what Ann said? After all, as a competent speaker he understands that she is merely asserting what tastes good to her. However, relativism need not be understood as an explanation of our disagreement intuitions.⁴ Stephenson (2007b) even explicitly denies that this is her objective: I should clarify that when I say that disagreement is possible in a certain dialogue, I mean very narrowly that expressions like no (it isn’t) and nuh-uh are allowed. I don’t mean that we have an intuition that the speakers are disagreeing about something, which may be a broader phenomenon. I also don’t mean that the disagreement is necessarily a rational or sensible one to engage in. (Stephenson, 2007b, 493).

index). Upon closer scrutiny, however, judge-shifting is not uniformly allowed in quantificational structures, so it is unclear how much weight we should assign to examples like (i). (ii)

#Everyone danced with an attractive girl, but I thought all girls at the party were ugly.

4 In fact, as a theory about our disagreement intuitions, relativism seems to miss the mark. As pointed out by Huvenes (2012) and Sundell (2011), for example, we may have the intuition that two people disagree, even in the absence of a contradiction at the propositional level. For instance, in the next dialogue Bob explicitly states that he takes himself to be in disagreement with Ann. But at the same time, it seems obvious that he cannot be denying the content of what she said, because this would imply that he thinks that she is confused about her own taste: (i)

Ann: I like this chili. Bob: I disagree, it is way too hot for me.

Huveness and Sundell propose that disagreement merely requires conflicting attitudes about the same thing.

Subjective meaning: An introduction



So, according to Stephenson, for Bob to be able to felicitously respond to Ann with no, it has to be the case that he assents to the negation of the proposition that she assents to. And this is what relativism elegantly delivers, but what the simplest form of contextualism fails to derive. Lasersohn and Stephenson agree that contradiction drives linguistic denial, and spell this out at the level of truth conditional content.⁵ However, not everyone would agree that truth conditions alone exhaust the range of natural language meanings. For example, in this volume, Daniel Gutzmann argues in If Expressivism is fun, go for it! that taste disputes are rather driven by a contradiction at the level of use-conditional content, which is the level of meaning needed to capture the contribution of expressives like Yuck!. The remainder of this chapter is structured as follows. Section 2 compares relativism to a more sophisticated contextualist analysis, where sentences like (1a) express generic readings of a special, first person oriented kind. We will see that the two theories differ in their account of the nature of disagreements like (3). After that, Section 3 investigates a wider range of subjective expressions, including epistemic modals and gradable adjectives, and points to contrasts in their acceptability under the attitude find. We will discuss the possibility that these contrasts reflect two kinds of subjectivity, one for which relativism is the right analysis, and one for which contextualism fits, and we will raise some problems for this view. Finally, Section 4 argues that at least some differences between taste predicates and epistemic modals are best analyzed as pragmatic rather than semantic differences.

2 Generic uses and faultless disagreement Relativists put a great deal of emphasis on cases where taste statements are anchored to individual judges, but especially in taste disputes, it seems that we aim to convince others, and in that case it cannot be just our own preferences that matter. Instead, taste disputes arguably involve generic uses of predicates of personal taste that are about the tastes of normal judges (Keshet, 2005; Moltmann, 2010; Pearson, 2013). This is backed up by investigations reported by Wiggins & Potter (2003) (brought to my attention by Crespo & Fernández 2011), who based on a cor-

5 Stephenson (2007b) departs from Lasersohn, however, in treating tasty as a two-placed predicate, whose judge argument may be filled by a covert element PRO referring to the judge index of evaluation, but also by pro, denoting a contextually salient individual, which is what happens in examples like (7) and (8).

6 | Janneke van Wijnbergen-Huitink

pus of family mealtime conversations conclude that unmodified taste statements (This wine is very nice) and those that are explicitly relativized (I like this wine) have different discourse functions. The latter mere expressions of subjective preference are typically used to account for future activity (I might, for instance, go to the store later to buy another bottle of wine because I like it), whereas the former are typically used to persuade others to try the relevant food or drink, indicating that the meaning of these statements doesn’t exclusively involve the speaker’s preferences. One way to spell this out semantically is by construing tasty as a two-place predicate (as in (2a)), which may combine with a variable that is bound by a covert generic quantifier: (9)

[[Gen x [This cake is tasty (to x)]]]c,w = 1, iff this cake tastes good to all normal people

If this contextualist analysis is correct for Ann’s taste statement in (3), it is no longer mysterious why Bob can deny what she said. As long as Bob is included in the domain of quantification, the mere fact that he dislikes the cake contradicts what Ann said. However, Lasersohn (2005) rejects this possibility, because it seems to disconnect the proposition expressed from the experiences of the speaker. This wrongly predicts that there is nothing anomalous about statements like: (10)

a. #This is not fun at all, although I’m having fun doing it. b. #This cake is tasty, but I don’t like it.

But of course we do not talk this way. In contrast, Lasersohn accounts for such examples by assuming that we typically assert from an autocentric perspective. Addressing this worry, Moltmann (2010) and Pearson (2013) argue that the involved in statements with predicates of personal taste is of a special, first person oriented kind. When I call something tasty, I generalize from my own experience, to that of other people, based on their presumed similarity to me. I do the same when I say things like: (11)

This stuff is a good remedy; it worked for me. (= Hegarty, 2016, example (11), this volume)

Subjective meaning: An introduction

| 7

Moltmann (2010) claims that in English, this type of generalization is expressed by generic one. She points out that the use of one typically implies that the sentence holds for the speaker, i.e. (12a) implies (12b):⁶ (12)

a. b.

One can see the picture from the entrance. I can see the picture from the entrance.

Similarly, it is only a small leap from This cake is tasty to This cake is tasty to me, explaining the anomaly of (10a) and (10b).⁷ Moltmann and Pearson relate first person orientedness to de se interpretation, in such a way that the interpretation of an unmodified taste statement like This cake is tasty involves self-ascription of a property with generic content, reflected in quantification over people that the judge identifies with: (13)

λx. this cake tastes good to all normal people that x identifies with

Self-ascription of this property involves putting oneself in the shoes of the people in the domain of quantification, on the assumption that we are alike in taste. People always identify with themselves, and in ordinary settings also with their interlocutors (see Pearson, 2013 for further details about the “identify with” relation, and its status in her theory versus Moltmann’s 2010 proposal). One consequence of this de se analysis is that although taste statements aim for objectivity, they can only be based on the speaker’s own, private experiences, even when she puts herself in the shoes of others. Taste statements are thus subjective in the sense that their assertion is warranted by first personal grounds. Note however that this theory requires that the contents of taste statements be properties. A principled way to ensure this would be to adopt the relativist framework, where the addition of a judge index renders all sentence contents properties (see e.g. Stephenson’s 2007b treatment of attitudes de se). Wanting to avoid relativism, Moltmann and Pearson resort to complicating the syntax by positing λ-abstractors. This raises the question just how non-relativistic their generic ac-

6 Note, however, that first-person oriented one is not possible in combination with predicates of personal taste: (i)

#That was fun for one.

It is not clear why this is not possible since, according to Moltmann (2010), we only need our own experiences in order to use one. 7 There are counterexamples, like: One can see me from the entrance or when predicates of personal taste occur in the scope of evidentials, e.g. Apparently, this cake is tasty. See Moltmann (2010) and Pearson (2013) for further discussion.

8 | Janneke van Wijnbergen-Huitink

count really is, but an in-depth comparison between these two strategies is beyond the scope of this chapter.⁸ Relativists will probably not dispute that predicates of personal taste allow generic interpretations, but they do deny that such an interpretation (always) figures in dialogues like (3). Rather, relativists think of the disagreement between Ann and Bob as “faultless”, in the sense that the interlocutors have both said something true. Contextualists, on the other hand, view disputes like (3) as “genuine” disagreements, where either Ann or Bob must have said something false. Either Ann has the property of being an x such that all normal people she identifies with find this cake tasty, or she does not. Contextualists take faultless disagreement to be a mistaken intuition, that results from not distinguishing between the generic use and a personal use where the judge argument is filled by a silent deictic pronoun, such that only the speaker’s tastes matter (Keshet, 2005; Moltmann, 2010; Stojanovic, 2007). It seems that this latter, personal use is unambiguously expressed by sentences like: (14)

a. b.

This cake is tasty to me. I find this cake tasty.

When used generically, there can be disagreement. But when we think that this disagreement is faultless, we have moved to the personal uses, which are mere reports of the speaker’s experiences and cannot fail to be true (assuming the reports are sincerely made). The personal and the generic use are only simultaneously available to a third party, so faultless disagreement can only be real from a bird’s eye perspective, not from the perspective of the disagreeing parties. To sum up, the main point of divergence between relativism and contextualism is the status of faultless disagreement. Relativists claim that it really is possible that Ann and Bob are both using tasty to express just their own personal opinion, realize that they do this, and yet take themselves to be disagreeing. Contextualists claim that this is an illusion. Several contributions to this volume tie into this debate. In particular, in Subjective meaning and modality Michael Hegarty proposes to extend the generic account to epistemic modals. In Propositions and implicit arguments carry a default “general point of view” Tom Roeper offers experimental evidence that young children will readily overcome their own perspective to assess statements from a general point of view, suggesting the generic use has default status. In Doing without judge dependence, Sanna Hirvonen addresses the metaphysical worries

8 It is perhaps telling that Egan (2014), who proposes a de se account for judgments about comic properties, takes the de se view to be essentially relativist.

Subjective meaning: An introduction



connected to such objective construals, namely that they wrongly seem to suggest that there are such things as taste properties within the objects themselves that make such claims true. In Predicates of personal taste and the evidential step Lavi Wolf agrees that Ann and Bob aim for more objectivity than the relativist proposal allows, but he doesn’t think one should spell this out in terms of a (special kind) of , but in terms of the weighted sum of the degree of belief conversationalists attach to statements like This cake is tasty. Finally, in Contextualism and disagreement about taste, Dan Zeman emphasizes that contextualist solutions must deny the core relativist intuition of faultless disagreement, and argues that this is a bad thing. Shouldn’t semantics account for speaker’s intuitions instead of explaining them away as resting on a misjudgment of uniformity? Clearly, this should be a last resort strategy. In the next section, we will discuss the embedding possibilities for a wider range of subjective expressions, suggesting subjectivity comes in different varieties. The choice between relativism and contextualism then might depend on the kind of subjectivity expressed. It could be that each is right for a specific kind of subjective meaning.

3 Different kinds of subjectivity? 3.1 Acceptability under find How can we identify subjectivity? In the previous section, we have already encountered one test for subjectivity, to wit, the possibility for faultless disagreements. But there are other diagnostics. For example, Sæbø (2009) points out that the attitude predicate find seems to select complements that depend on personal opinion rather than on objective facts: (15)

a. I find this cake tasty. b. #I find this cake contains sugar. c. #I find Wilma pregnant.

This raises the question how acceptability under find relates to the other test for subjectivity, the possibility for faultless disagreement. Are all predicates that give rise to faultless disagreement acceptable under find? The behavior of epistemic modals like might makes it clear that the answer to this question is negative. Epistemic modals do engage in faultless disagreements (Egan et al., 2005; MacFarlane, 2011; Stephenson, 2007b):

10 | Janneke van Wijnbergen-Huitink


Ann: Bob:

Bill might be in his office. No, he can’t be. He never works on Friday.

We have the intuition that Ann is talking about what possibilities are compatible with the evidence available to her. At the same time, Bob can felicitously disagree with what she said.⁹ But epistemic modals cannot be embedded under find: (17)

#I find that it might be raining this afternoon.

The same holds for epistemic must (cf. Sæbø, 2009): (18)

#Fred finds that it must have rained yesterday.

The corpus study by Lande (2009) failed to find any epistemic modals licensed in the complement of subjective attitude verbs, nor in the German, nor in the Norwegian part of the corpus. This strongly suggests that in as far as epistemic modals are subjective, they are not subjective in the sense that they can be embedded under subjective attitudes.¹⁰ Stephenson (2007c) proposes that find requires that the subject have direct experience of the complement proposition. This would rule out (17) and (18) because epistemic modals are suggestive of indirect, inferential evidence (von Fintel & Gillies 2010 make this point for epistemic must). However, it seems wrong that find places such a requirement:

9 There is a possible confound: Bob might be expressing that Ann was wrong about the prejacent proposition that Bill is in his office. But relativists claim that the target is the modalized proposition. 10 Reis (2013) offers several counterexamples found on the internet, and concludes that German existential epistemic modals can be embedded under finden. For example: (i)

Du findest, dass man nicht wissen kann, ob er gefoltert worden ist oder nicht. Lit: ‘You find that it is impossible to know whether he was tortured or not.’

However, I find it difficult to make sure that the modals in her examples do indeed express epistemic modality in these contexts, rather than opportunity/ability. Relatedly, at least in Dutch, onwaarschijnlijk ‘unlikely’ is fine under a subjective attitude: (ii)

Ik vind het onwaarschijnlijk dat niemand het ongeluk gezien heeft. ‘I find it unlikely that nobody saw the accident.’

But again it is not clear that the meaning of this expression involves an inference based on the knowledge available to the speaker.

Subjective meaning: An introduction


| 11

a. #Homer finds himself gay. (Sæbø, 2009, example (7)) b. #I find that I am hungry. c. Homer finds that the death penalty should be abolished.

Homer presumably has direct experience of his sexual orientation. In spite of this, (19a) is out. In contrast, deontic modality is not experiential, but (19c) is fine. Sæbø (2009, p. 24) concludes that epistemic modals “evidently do not depend on the same sort of assessment as do predicates of personal taste”. Extending these observations, Kennedy (2013) points out that dimensional like tall pattern with epistemic modals in this respect, while evaluative adjectives like attractive seem to form one group with taste predicates (the dimensionalevaluative distinction goes back to Bierwisch, 1987, 1989). Two people can faultlessly disagree about dimensional adjectives (Barker, 2013; Richard, 2004): (20)

a. b.

Ann: Fred is tall. Bob: No, he’s not.

The intended reading is one where Ann and Bob have the same comparison class in mind (so their disagreement does not depend on Ann comparing Fred to the average man, and Bob to the average basketball player), but they just disagree about how tall Fred needs to be to count as tall. They may do so, while they agree on Fred’s actual height. In that case, there is no fact of the matter that will decide which of Ann and Bob is right. Moreover, neither Ann nor Bob seems to have privileged authority to decide where the cut-off point for tall lies. At the same time, dimensional gradable adjectives, in their positive form, are slightly degraded under find, while the comparative form is completely unacceptable, again indicating that faultless disagreement and find do not track the same kind of subjectivity:¹¹ (21)

a. ??I find Billy tall. b. *I find Billy taller than Fred.

Things are different for evaluative gradable adjectives. Evaluative adjectives like pretty and boring differ from tall and its kin in that they intuitively do not mea-

11 In contrast, in Dutch, one can without any degradedness say things like: (i)

Ik vind Jan lang. ‘I find Jan tall.’

suggesting that there are subtle differences between the subjective attitude verbs of different languages.

12 | Janneke van Wijnbergen-Huitink

sure properties that are inherent in the objects themselves, but that are ascribed to them by some individual. Evaluative predicates pattern with taste predicates like tasty in that they can be embedded under find, both in their positive and comparative form. (22)

a. b.

I find Johnny Depp more attractive than Brad Pitt. I find this lecture boring.

The reader can verify for herself that such predicates engage in faultless disagreements as well, or see Bylinina (2015) for some relevant illustrations. To sum up, the above contrasts suggest that subjectivity is not a homogeneous phenomenon. Rather, epistemic modals and dimensional adjectives are not subjective in the way that evaluative adjectives and taste predicates are, given that only the latter display the kind of subjectivity that find is sensitive to. How does this relate to the debate between contextualists and relativists? Could it be that both are right, each for a specific kind of subjectivity? How can we decide which analysis fits which kind of subjectivity?

3.2 Semantic implications In an intriguing paper, Sæbø (2009) argues that find selects for a linguistically represented judge argument, thus presupposing a contextualist analysis of expressions that are acceptable under this attitude (the argument comes from embedded coordinations and is not discussed here). The contextualist analysis for tasty is repeated here for convenience: (23)

[[tasty]]c,w = λx.λy.x tastes good to y in w

If This cake is tasty occurs unembedded, the predicate’s judge argument is (per default) filled by a covert judge PP for me. However, Sæbø proposes that when it gets embedded under find, the attitude verb fills this argument slot with its subject. In sentences like Mary finds this cake tasty, the attitude verb thus takes a complement that, as a stand alone sentence, cannot be assigned a truth value, because the judge argument is missing. The corresponding denotation for find is given in (24a), and (24b) derives the meaning for Mary finds this cake tasty: (24)

a. b.

[[find]]c,w = λP⟨e,t⟩ .λx.P(x) [[Mary finds this cake tasty]]c,w = ([[find]]c,w ([[tasty]]c,w ([[this cake]]c,w )))([[Mary]]c,w ) = (λP⟨e,t⟩ .λx.P(x)(λx.λy.x tastes good to y in w (this cake))) (Mary) = this cake tastes good to Mary in w

Subjective meaning: An introduction



On this analysis, objective predicates, as in (15b) and (15c), are unacceptable under find due to a type-mismatch between the attitude and its complement, which fails to denote a set of judges that make it true. Note though that this cannot be the full story about find, for one would like to exclude that the attitude combines with any intransitive predicate, or else Mary finds walks could mean that Mary walks. Judge argument slots thus seem to have a special status, even in contextualist accounts, suggesting that the relativist alternative should not be dismissed too quickly (see Bouchard, 2012; Reis, 2013, for further assessment of Sæbø’s proposal). If Sæbø is right, and acceptability under find is best explained by having an argument place for a judge, then the contrasts between epistemic modals and dimensional adjectives on the one hand, and taste predicates and evaluative adjectives on the other, corresponds to a valency distinction. The latter take a judge argument, and are thus subjective in a contextualist way, but the former are subjective in a way that is not linguistically represented, suggesting that relativism provides the right analysis for this type of subjectivity. However, this story faces at least three problems. First, as pointed out by Bylinina (2015), the contrast between dimensional and evaluative adjectives under find ought to be much more pronounced, if unacceptability relates to a type-theoretic conflict. Second, predicates of personal taste do but evaluative adjectives in English do not take judge PPs, which is unexpected if the subjectivity of both kinds of expressions is manifested in the presence of an extra argument: (25)

a. That was fun / tasty / for John. b. ??Mary is brave/silly/smart for/to John.

This contrast cannot be attributed to idiosyncrasy, since Bylinina shows it is crosslinguistically stable.¹² For instance, in German, a dative argument corresponding to the judge (the so-called dativus iudicantis) is not licensed with evaluative adjectives: (26)


Diese Frage ist mir peinlich. Lit: ‘This question is for me embarrassing.’ b. *Dieser Student ist mir schlau. Lit: ‘This student for me is smart.’

12 Although it seems some idiosyncrasies must nevertheless be allowed. For example, Dutch lekker ‘tasty’ does not take an overt judge argument, and even in English, ‘Mary is beautiful to John’ doesn’t sound too bad, cf. Pedersen (2012).

14 | Janneke van Wijnbergen-Huitink

Bylinina (2015) argues that taking a judge PP does not reveal anything about subjectivity, but relates to the semantic dependence on an experience instead, which implies the presence of an experiencer argument. She favors a uniform, relativistic conception of subjectivity. Finally, it seems that Sæbø’s contextualist account of find places epistemic modals and predicates of personal taste in precisely the wrong camp. Recall from Section 2 that contextualism and relativism make different predictions regarding disagreement. Only relativism can make room for faultless disagreement. Hence, if epistemic modals were subjective in a relativist way, one would expect disputes about them to be clearly faultless. However, as said in the Preface, we do not feel that disputes about epistemic modals are about who is right. Rather, this is the hallmark of disputes with taste predicates. We will address this contrast in the next section, and we will argue that it may receive a pragmatic explanation. In the present volume, in Two kinds of subjectivity Christopher Kennedy further defends his claim that dimensional and evaluative adjectives are subjective in different ways and explains the difference in terms of quantitative and qualitative interpretations, as described for color terms in Kennedy & McNally (2005). In Evaluative propositions and subjective judgments Carla Umbach denies that acceptability under German finden ‘find’ relates to subjectivity, and proposes instead that finden selects for a complement that is open to interpretation by the subject, suggesting a reinterpretation of the contrasts observed by Kennedy.

4 The pragmatics of subjectivity It is not just challenging to describe the truth conditions of subjective expressions, but their use conditions are puzzling as well. In particular, predicates of personal taste that occur embedded under first person find are immune to linguistic denial (cf. Crespo & Fernández, 2011). In such disputes, we get the impression that Bob knows Ann’s preferences better than she does herself: (27)

Ann: I find this cake tasty. Bob: #No! It is disgusting.¹³

Similar observations apply to predicates of personal taste with overt first person judge PPs (cf. Moltmann, 2010):

13 If you have trouble getting the relevant judgment, make sure that you read Bob’s denial as targeting the entire attitude claim rather than just the embedded complement that this cake is tasty.

Subjective meaning: An introduction


| 15

Ann: This cake tastes good to me. Bob: #No! This cake tastes disgusting to me.

Again, Bob has no business denying what Ann said. Lasersohn (2005) does not discuss disputes like this. But assuming that subjective attitudes and judge PPs fix the identity of the judge, he would predict that Ann forces Bob to take an exocentric perspective when assessing her claim. So when Bob evaluates her sentence, he has to put himself in her shoes and then check whether the cake is tasty. It seems plausible that it is exactly this what makes Bob’s denial odd. Bob can never reach a better judgment in this way than Ann did herself when she said: “I find this cake tasty / This cake is tasty to me”, since Ann has inherent authority with respect to her own tastes. In their chapter Predicates of experience, Christine Gunlogson and Gregory Carlson spell this out in a more general theory of linguistic denial and evidential considerations. Their main claim is that for linguistic denial to be licensed, the denying party has to have evidence that is at least as good as that of the person whose claim is being challenged. Direct, perceptual evidence outweighs all other types, and since Ann’s statements in (27) and (28) are obviously made on such unassailable, direct evidence, Bob cannot challenge her claim. A different approach is advocated by Carla Umbach, who claims in Evaluative propositions and subjective judgments that statements like This cake is tasty and I find this cake tasty have different discourse functions. The former but not the latter are intended to update the common ground. Although Gunlogson and Carlson limit their discussion to predicates of experience, it seems that careful attention to evidential considerations might help to solve some puzzles concerning epistemic modals as well. It is often pointed out in the literature that there is something special about faultless disagreements with epistemic modals, to wit, once Bob reveals his knowledge, it would be natural for Ann to change her mind and retract her previous claim, and plain odd if she would stand by it (MacFarlane, 2011): (29)

Ann: Bob: Ann: Ann’:

Bill might be in his office. No, he can’t be. He never works on Friday. O, then I guess I was wrong. #Still, I insist that he might be in his office.

Put differently, we do not tend to continue arguing about who is right with epistemic modals. As pointed out by Stephenson (2007b), this sets epistemic modals apart from predicates of personal taste, for which retraction would be plain odd, and about which our disputes may never be settled:

16 | Janneke van Wijnbergen-Huitink


Ann: This cake is tasty. Bob: No, it isn’t. It’s disgusting. Ann: #O, then I guess I was wrong.

It could be argued that retraction indicates that Ann’s initial might-statement was simply false (even though her claim doesn’t obviously rely on a mistaken belief) (cf. Bouchard, 2012; Wright, 2007; Willer, 2013). This implies that the disagreement between Ann and Bob was never faultless to begin with, eliminating the need for a judge-dependent semantics.¹⁴ On the other hand, given an evidential theory of linguistic negation à la Gunlogson and Carlson, retraction need not be related to the truth-value of Ann’s might-claim. Recall that they propose that judge selection depends on evidential considerations. Generally, the judge with better evidence is the stronger candidate. Although Gunlogson and Carlson are mostly concerned with direct versus indirect evidence, it seems innocuous to say that a judge with complete background information is in a better position to be the judge than someone with only partial information. Then, retraction is the result of Ann recognizing that Bob has better evidence than she does. That is, one might understand Ann’s O, then I guess I was wrong to mean that she was wrong to take herself to be an apt judge. No such thing happens in disputes about taste like (30), because there Ann and Bob are typically on equal footing. Relatedly, Dietz (2008) notes that there is an asymmetry between the interlocutors in faultless disagreement cases with epistemic modals. Generally, the denying party knows more than the speaker who made the might-claim. Once the order of the speakers is reversed, the disagreement becomes absurd: (31)

Bob: Bill cannot be in his office. Ann: #No! He might be in his office. Ann’: What makes you say that?

14 Knobe & Yalcin (2014) claim that retraction data are not robust. In particular, they report that people do not like the following retraction: (i)

Ann and Bob are looking for a set of keys. Ann says to Bob, “The keys might be in the drawer.” Bob looks in the drawer. The keys are not there. Bob becomes agitated. He says, “They’re not. Do you still stand by your claim?” Ann replies, “I take what I said back—I was wrong.”

But here Ann states that she takes back what she said, as if her statement wasn’t a useful contribution, where in fact it prompted Bob to look in the drawer and rule out a possible location of the keys.

Subjective meaning: An introduction



Dietz thinks that cases like this pose a threat to relativist analyses, but even though Bob does not share the evidence for his claim, it is not as if Ann gets nothing from his utterance.¹⁵ In fact, she learns that it is not compatible with Bob’s knowledge that Bill is in his office, hence that there is something to be known, not yet known by her, that rules this out. Following Gunlogson and Carlson, this means that she is not in a position to deny what Bob said. This is a case where an exocentric perspective is actually better than her autocentric perspective. To sum up, disagreements with epistemic modals display different characteristics than disagreements with predicates of personal taste. This could be taken as evidence that the two kinds of expressions are semantically subjective in different ways. However, the contrasts may also be given a different explanation, in terms of the pragmatic mechanism fixing the identity of the judge.

5 Conclusion Relativism provides an elegant account of subjective meaning. Yet it seems safe to say that the notion of relative truth is somewhat difficult to swallow, and in recent years we have witnessed a quest for less radical alternatives. This introductory chapter identified several strands in prima facie promising contextualist analyses where relativism seems to loom (i.e. the first person oriented view, Sæbø’s contextualist analysis of selection by find). So the jury is still out on whether we can completely avoid relativism. At the very least, however, the efforts by linguists and philosophers to find new solutions to the problem of faultless disagreement have gained us a better understanding of subjectivity. Acknowledgment: Special thanks go to Cécile Meier and Klaus von Heusinger for their thoughtful comments on a earlier version of this chapter. I am also grateful to all authors of the subsequent chapters in this book. Your inspiring work has truly shaped my thoughts about relativism and its alternatives.

15 This point is also forcefully made by Krzyźanowska et al. (2014) regarding Gibbardian standoffs with indicative conditionals, which are standardly treated as epistemic modals (Kratzer, 1991, 2012) and for which Stephenson (2007a) proposes a relativist semantics, to which von Fintel & Gillies (2008) in turn object that it has problems dealing with assessor asymmetry.

18 | Janneke van Wijnbergen-Huitink

References Barker, Chris. 2013. Negotiating taste. Inquiry: An Interdisciplinary Journal of Philosophy 56(2–3). 240–257. Bierwisch, Manfred. 1987. Semantik der Graduierung. In Manfred Bierwisch & Ewald Lang (eds.), Grammatische und konzeptuelle Aspekte von Dimensionsadjektiven, 91–286. Berlin: Akademie Verlag. Bierwisch, Manfred. 1989. The semantics of gradation. In Manfred Bierwisch & Ewald Lang (eds.), Dimensional adjectives: Grammatical structure and conceptual interpretation, 71– 261. Berlin: Springer. Bouchard, David-Etienne. 2012. Long-distance degree quantification and the grammar of subjectivity. Montreal, QC: McGill University dissertation. Bylinina, Lisa. 2015. Judge-dependence in degree constructions. diait7bjdjwpjbs/jos8draft.pdf?dl=0 (accessed March 2, 2016). Crespo, Inés & Raquel Fernández. 2011. Expressing taste in dialogue. In Ron Artstein, Mark Core, David DeVault, Kallirroi Georgila, Elsi Kaiser & Amanda Stent (eds.), Semdial 2011 (Los Angelogue) Proceedings of the 15th Workshop on the Semantics and Pragmatics of Dialogue, 84–93. Los Angeles. Dietz, Richard. 2008. Epistemic modals and correct disagreement. In Manuel García-Carpintero & Max Kölbel (eds.), Relative truth, 239–262. Oxford: Oxford University Press. Egan, Andy. 2014. There is something funny about comedy: A case study in faultless disagreement. Erkenntnis 79(1 Supplement). 73–100. Egan, Andy, John Hawthorne & Brian Weatherson. 2005. Epistemic modals in context. In Gerhard Preyer & Georg Peter (eds.), Contextualism in philosophy: Knowledge, meaning and truth, 131–168. Oxford: Oxford University Press. von Fintel, Kai & Anthony Gillies. 2008. CIA leaks. The Philosophical Review 117(1). 77–98. von Fintel, Kai & Anthony Gillies. 2010. ‘Must’ ...stay ...strong! Natural Language Semantics 18(4). 351–383. von Fintel, Kai & Anthony Gillies. 2011. ‘Might’ made right. In Andy Egan & Brian Weatherson (eds.), Epistemic modality, 108–130. Oxford: Oxford University Press. Glanzberg, Michael. 2007. Context, content, and relativism. Philosophical Studies 136. 1–29. Hegarty, Michael. 2016. Subjective meaning and modality. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 227–248. Berlin: De Gruyter. [This volume]. Huvenes, Torfinn. 2012. Varieties of disagreements and predicates of taste. Australasian Journal of Philosophy 90. 167–181. Kaplan, David. 1989. Demonstratives. In Joseph Almog, John Perry & Howard K. Wettstein (eds.), Themes from Kaplan, 481–565. Oxford: Oxford University Press. Kennedy, Christopher. 2013. Two sources of subjectivity: Qualitative assessment and dimensional uncertainty. Inquiry: An Interdisciplinary Journal of Philosophy 56(2–3). 258–277. Kennedy, Christopher & Louise McNally. 2005. Scale structure, degree modification and the semantics of gradable predicates. Language 81. 345–381. Keshet, Ezra. 2005. A matter of taste. Unpublished manuscript, MIT. Knobe, Joshua & Seth Yalcin. 2014. Epistemic modals and context: Experimental data. Semantics and Pragmatics 7(10). 1–21. Kölbel, Max. 2002. Truth without objectivity. London: Routledge.

Subjective meaning: An introduction

| 19

Kölbel, Max. 2004. Faultless disagreement. Proceedings of the Aristotelian Society 104. 53–73. Kratzer, Angelika. 1991. Modality/Conditionals. In Arnim von Stechow & Dieter Wunderlich (eds.), Semantik: Ein internationales Handbuch der zeitgenössischen Forschung/Semantics: An international Handbook of Contemporary Research, 639–656. Berlin: De Gruyter. Kratzer, Angelika. 2012. Modals and conditionals. Oxford: Oxford University Press. Krzyźanowska, Karolina, Sylvia Wenmackers & Igor Douven. 2014. Rethinking Gibbard’s riverboat argument. Studia Logica 102(4). 771–792. Lande, Inna. 2009. Subjektive Einstellungsberichte im Deutschen: Einstellungsverben und beurteilungsabhängige Prädikate. Oslo: University of Oslo MA thesis. Lasersohn, Peter. 2005. Context dependence, disagreement, and predicates of personal taste. Linguistics and Philosophy 28(6). 643–686. Lasersohn, Peter. 2009. Relative truth, speaker commitment, and control of implicit arguments. Synthese 166. 359–374. Lewis, David. 1970. General semantics. Synthese 22. 18–67. MacFarlane, John. 2011. Epistemic modals are assessment-sensitive. In Andy Egan & Brian Weatherson (eds.), Epistemic modality, 144–178. Oxford: Oxford University Press. Moltmann, Friederike. 2010. Relative truth and the first person. Philosophical Studies 150. 187–220. Pearson, Hazel. 2013. A judge-free semantics for PPTs. Journal of Semantics 30. 103–154. Pedersen, Walter. 2012. Faultless disagreement and semantic relativity. http://people. (accessed June 9, 2014). Reis, Marga. 2013. Dt. ‘finden’ und “subjektive Bedeutung”. Linguistische Berichte 236. 389– 426. Richard, Mark. 2004. Contextualism and relativism. Philosophical Studies 119. 215–242. Sæbø, Kjell Johan. 2009. Judgment ascriptions. Linguistics and Philosophy 32(4). 327–352. Stephenson, Tamina. 2007a. Indicative conditionals have relative truth conditions. Proceedings from the Annual meeting of the Chicago Linguistic Society 43. 231–242. Stephenson, Tamina. 2007b. Judge dependence, epistemic modals, and predicates of personal taste. Linguistics and Philosophy 30(4). 487–525. Stephenson, Tamina. 2007c. Towards a theory of subjective meaning. Cambridge, MA: MIT dissertation. Stojanovic, Isidora. 2007. Talking about taste: Disagreement, implicit arguments, and relative truth. Linguistics and Philosophy 30(6). 691–706. Sundell, Timothy. 2011. Disagreements about taste. Philosophical Studies 155(2). 267–288. Wiggins, Sally & Jonathan Potter. 2003. Attitudes and evaluative practices: Category vs. item and subjective vs. objective constructions in everyday food assessments. British Journal of Social Psychology 42. 513–531. Willer, Malte. 2013. Dynamics of epistemic modality. The Philosophical Review 122(1). 45–92. Wright, Crispin. 2007. New age relativism and epistemic possibility: The question of evidence. Philosophical Studies 17(1). 262–283.

Daniel Gutzmann

If expressivism is fun, go for it! Towards an expressive account of predicates of personal taste Abstract: Predicates of personal taste like tasty give rise to the puzzle of “faultless disagreement”: with taste predicates, two speakers can disagree while both saying something true. In an influential paper, Lasersohn (2005) dismisses various approaches to the puzzle of faultless disagreement, before coming up with a contextualist solution. This paper examines “expressivism”, one of the dismissed options, and shows that the arguments Lasersohn provides against it are not conclusive. It then develops a hybrid expressive approach that combines a simple indexical semantics in the truth-conditional dimension with a deontic attitude in the use-conditional dimension. The first dimension accounts for the faultlessness, as both speakers express what is tasty for them, while the disagreement happens at the use-conditional layer which concers what should count as tasty in the context of use.

1 Predicates of personal taste, faultless disagreement, and relativism In his influential paper, Lasersohn (2005) addresses some of the puzzles that are raised by predicates of personal taste for traditional truth-conditional semantics in the Frege-Tarskian tradition. Intuitively, the main problem that predicates of personal taste (PPTs) pose for a truth-conditional analysis is that they are not subject to an objective truth evaluation upon which truth-conditional frameworks are based. Consider, for instance, an utterance of a simple declarative sentence, like (1), that does not contain any subjective expression. (1)

Tom likes tofu.

For this sentence, it is straightforward to state the truth conditions. An utterance of (1) is true if Tom likes tofu. This condition is supposed to capture the meaning of (1) by defining the proposition: the set of worlds in which (1) is true. (2)

{w : Tom likes tofu in w}

Daniel Gutzmann, Institute of Linguistics, University of Frankfurt, Norbert-Wollheim-Platz 1, 60629 Frankfurt, Germany, email: [email protected]

22 | Daniel Gutzmann

In case of PPTs, it is however more complicated, as it is not intuively clear what the truth conditions shall be. Consider the following examples that illustrate the case of PPTs. (3)

a. b. c.

Roller coasters are fun. Cats are cute. Tofu is tasty.

(Lasersohn, 2005, 643)

The problem, according to Lasersohn’s reasoning, is that a straightforward truthconditional analysis is not available. Take, for instance, the following proposal for the truth conditions of (3c). (4)

#“Tofu is tasty” is true iff tofu is tasty.

While (4) seems intuitive, it completely ignores the important intuition that PPTs are “subjective”.¹ It seems to be a matter of my personal taste of food whether I would judge, for instance, (3c) to be true or not. There arguably is no objective fact that can make me think that (3c) is false if I experience tofu as tasty. From the subjective meaning of PPTs it follows that it is perfectly reasonable to assume that (3c) is true when uttered by tofu lover Tom, while the same utterance would be false if uttered by tofu hater Hanna. Now, consider what happens when Tom and Hanna sit down to discuss tofu. (5)

a. b.

Tom: Tofu is tasty. Hanna: No, tofu isn’t tasty at all.

Given the subjectivity of their respective utterances, we are inclined to say that both Tom and Hanna are saying something true. Since whether tofu is tasty or not is a matter of their respective tastes and since their tastes seem to differ, they both are justified in asserting different judgments regarding the tastiness of tofu. Crucially, there is another intuition regarding exchanges like (5), one that, as we will see, is actually in conflict with the intuition of subjectivity. When Tom and Hanna both assert their respective assessment in a dialogue like (5), there is a strong intuition that they disagree with each other and there actually is an argument going on between the two of them. Taken together, these two intuitions give rise to what became to be known as the puzzle of “faultless disagreement” (Kölbel, 2004). In a situation like (1), nei-

1 As a reviewer pointed out, this does only hold if we assume that the tasty used in the metalanguage means tasty tout court.

If expressivism is fun, go for it! |


ther Tom nor Hanna make a fault by asserting that tofu is or is not tasty.² Still, they disagree with each other. This poses a problem for simple versions of truthconditional semantics. If Tom and Hanna both assert something true, both utterances, Tofu is tasty and Tofu is not tasty, have to be true. However, capturing the intuition of Tom and Hanna contradicting each other requires that there is a contradiction at the level of semantic content. The dilemma is that the later requirement would imply that only one of the assertions can be true, which, in turn, contradicts the first intuition. Relating PPTs to other linguistic expressions, the puzzle posed by faultless disagreement can be traced back to the observation that for other classes of expressions, faultlessness [±F] and disagreement [±D] do not go hand in hand. First, consider the following example that illustrates that PPTs are not as subjective as indexical utterances. (6)

Indexicals a. b.

[+F, −D]

Tom: I eat tons of tofu. Hanna: I do not eat tons of tofu.

If Tom eats tons of tofu and Hanna does not, both are right. That is, the dialogue in (6) is faultless. However, there is no disagreement taking place in (6), because, at the level of semantic content, they express two different propositions. That is, if one just stays at the level of Kaplanian character (Kaplan, 1989), the two utterances may seem to be, in some sense, contradictory: if they were uttered in the same context, and hence by the same speaker, they would actually express contradictory propositions. However, if uttered by two different speakers, as in (6), they express two independent propositions. This is illustrated by the infelicity of using a negative response particle in Hanna’s utterance that signals disagreement or contradiction. (7)

a. b.

Tom: I eat tons of tofu. Hanna: #No, I do not eat tons of tofu.

That is, while indexicals can be said to be subjective because they may receive a different interpretation for each speaker and context, they do not give rise to faultless disagreement as PPTs do. For this reason, a hidden indexical analysis along the lines of (8) cannot work, as it would not account for the disagreement, even if it correctly predicts faultlessness. (8)

⟦tasty⟧c,w = λx.x is tasty for the speaker of c in w

2 That is, as long as their assertion actually matches their respective taste, of course.

24 | Daniel Gutzmann

If (8) really were all there is to the meaning of tasty, (5) would actually mean the same as the following variant. (9)

a. b.

Tom: Tofu is tasty for me. Hanna: Tofu is not tasty for me.

As is the case for (6), Tom and Hanna both may be right in (9), but they do not seem to disagree with each other: Tom is saying something about his taste and Hanna is saying something about her taste and of course, their tastes may differ. In some sense, PPTs hence seem to be more objective than indexicals. However, they still are not as objective as ordinary (objective) predicates. For instance, consider the following dialogue involving two simple, objective declaratives. (10)

Objective predicates a. b.

[−F, +D]

Tom: Tofu was first invented in Japan. Hanna: No, tofu was first invented in China.

In this case, it is clear that either Tom or Hanna is right and the other one is mistaken in their beliefs about the origin of tofu. That is, (10) is not a faultless exchange. Therefore, it is also clear that there is an actual dispute going on in (10) as Tom and Hanna are arguing about where tofu was first invented. That is, while (10) involves disagreement, it is certainly not a faultless situation. In this sense, PPTs are more subjective than ordinary predicates. Indexicals and objective predicates therefore suggest that disagreement requires one party to be mistaken and that faultlessness and disagreement hence are complementarily distributed. The main challenge posed by PPTs is therefore how to get [+F, +D]. Besides the strategy of denying that faultless disagreement actually is the correct diagnosis of examples like (5) and explaining our intuitions, there are several routes one can take to achieve the goal to account for faultless disagreement as induced by PPTs. In his influential 2005 paper, Lasersohn discusses some solutions. He dismisses them all and settles on an “relativist” account. His approach enriches the traditional two-dimensional Kaplanian approach by introducing an additional “judge” parameter to which the truth of an utterance is relativized at the level of content. According to this approach, (5a) and (5b) would express the same content, since the content of tasty does not change from context to context. That is, the content of, say, (5a) is that tofu is tasty. However, its extension (in the form of its truth value) is relativized to the judge and hence depends on whether

If expressivism is fun, go for it! |


tofu is tasty for Tom.³ That is, even if (5a) and (5b) express the same content, they may nevertheless differ in their truth value (see van Wijnbergen-Huitink (2016) [this volume] for an overview and discussion of Lasersohn’s approach). One of the alternative options that Lasersohn rejects is “expressivism”, according to which utterances involving PPTs are akin to expressive speech acts (Searle, 1969). The aim of this article is to show that despite Lasersohn’s dismissal, such an approach is not only viable, but even fares better than his approach in some respects. In order to do so, I will first sketch his arguments against the expressive approach to PPTs, before I will rebut them. Once his arguments are off the table, I will sketch what such an expressive approach might look like. To spell out the sketched proposal formally, I will make use of the developments and progresses that have been achieved in our understanding of expressive language since Lasersohn (2005) formulated his case against expressivism.

2 Lasersohn’s arguments against expressivism As illustrated in the last section, the problem of faultless disagreement arises from the assumption that PPTs are true or false. Expressivism solves the problem by denying just that basic assumption of a truth-conditional analysis of PPTs. This corresponds to the fourth option for a solution to faultless disagreement that Lasersohn discusses (and rejects): (11)

Option 4: Deny that truth and falsity are involved. (Lasersohn, 2005, 656)

Since, according to Lasersohn, the property of being true or false does only apply to the class of assertive speech acts, one way to spell out (11) is to assume that PPTutterances are not proper assertions that present a proposition as true. Instead, utterances as in (3) are considered as “non-assertive acts of affective expression” (Lasersohn, 2005, 656). That is, this perspective renders the PPT-utterances in (12) akin to purely expressive utterances of interjections as in (13). (12)

a. b.

This is fun! This is tasty!


a. b.

Whee! Mm-mm.

3 That is, the relativist account treats predicates of personal taste like content-based indexicals, so to speak, rather than ordinary character-based indexicals.

26 | Daniel Gutzmann

In the parlance of Searle’s (1969) classification of speech acts, PPT-utterances are hence expressive speech acts, instead of being assertive ones. This is motivated by the observation that expressives as those in (13) share some crucial properties with PPTs. Expressives display a subjective attitude or sensation that has an immediate effect that cannot easily be challanged by the hearer.⁴ If PPTs can receive a plausible expressive analysis, the problem of faultless disagreement would more or less vanish, as there is actually no contradiction involved in seemingly conflicting exchanges like (5). However, Lasersohn (2005, 656f.) only sketches the possibility of an expressive, non-truth-conditional approach to PPTs in order to dismiss it in favor of his perspective. He does this by presenting three arguments against expressivism for PPT-utterances. As we will see, each of his arguments is based on how expressive utterances are supposed to work in logical reasoning.

2.1 Argument 1: Deniability The first argument Lasersohn (2005, 657) presents against expressivism concerns the fact that two speakers can disagree regarding PPT-judgments as in (5), which I repeat here. (5)

a. b.

Tom: Tofu is tasty. Hanna: No, tofu isn’t tasty at all.

In contrast to PPT-utterances, genuine affective speech acts like Mm-mm cannot directly be denied in the same straightforward way, neither with That’s not true nor a plain No as illustrated in (14). (14)

a. b.

Tom: Mm-mm! Hanna: ??That’s not true! This isn’t tasty at all! Tom: Mm-mm! Hanna: ?No! This isn’t tasty at all!

According to Lasersohn, this shows that expressives do not have the same status as PPT-utterances because the former cannot be denied, whereas the later can, as shown by (5).

4 See, among others, Potts (2007b); Gutzmann (2013) for overviews of characteristic features of expressive items.

If expressivism is fun, go for it!



2.2 Argument 2: Embeddability While Lasersohn’s first argument concerns the interaction and different behavior of PPT- and expressive utterances with respect to denial in dialogue, his second argument against expressivism concerns their interaction with truth-conditional connectives. Prima facie, sentences involving PPTs can appear embedded under truth-conditional operators and can participate in common patterns of logical reasoning like modus ponens (Lasersohn, 2005, 657). (15)

Modus ponens with PPT-utterances a. b. c.

If there is a loop, the roller coaster is fun. There is a loop. Therefore, the roller coaster is fun.

If sentences like The roller coaster is fun actually do not have a truth value, arccording to Lasersohn, this behavior should be unexpected, as logical deduction as in (15) should only be possible with truth-conditional content.

2.3 Argument 3: Contradiction Lasersohn’s (2005) last argument against an analysis of PPT-utterances as nontruth-conditional, expressive speech acts goes into the heart of the problem of faultless disagreement. He argues that expressivism actually cannot account for the D in faultless disagreement. Since the notion of contradiction is based on the notion of truth, it is hard to see in what sense two expressive speech acts could contradict each other. Again, as we have seen above for denial, the notion of contradiction only seems to make sense for assertive speech acts, and if PPTs are expressive utterances, we should expect them inapt for making contradictions, contrary to the intuition of faultless disagreement. I think that all these three objections are not lethal to an expressivist approach. The main reason why I am not taking Lasersohn’s arguments to be convincing is that it is not clear that what he presents as evidence really establishes what he intends to show and whether the arguments warrant the conclusions he draws. In the following, I will therefore examine his arguments more closely, extending the scope of investigation beyond the affective use of interjections to more complex expressive speech acts as well as to other non-assertive speech acts.

28 | Daniel Gutzmann

3 In defense of expressivism Having presented Lasersohn’s (2005) objections against an expressive perspective on PPTs, I present some counterarguments in this section that are intended to illustrate that his arguments are not conclusive enough to rule out expressivism as a viable option in order to tackle PPTs and the puzzle of faultless disagreement. I will start by going through Lasersohn’s arguments one by one, mirroring the order of the last section, before turning to the bigger question of what this may tell us about the logic of non-truth-conditional meaning and reasoning.

3.1 Deniability The argument from deniability is based on the application of a That’s not true-test, whose premise is that you can deny an utterance by such direct denial phrases if and only if what you are targeting is truth-conditional. Hence, the test is supposed to be able to establish that an utterance is truth-conditional (or not). However, the bi-conditional that underlies the test and Lasersohn’s application of his argument does seem to be too strong. There are many instances in which such a phrase can be used even if we are dealing with an arguably non-assertoric speech act. While the use of such phrases may be strange as an reaction to mere interjections as mm-mm or whee, more complex expressive speech acts like congratulations, condolences and others seem to license such phrases if the hearer wants to challenge the speaker’s sincerity. Besides expressive acts, a similar use of denials is also possible with commissive speech acts like promises. (16)

A: B:

I congratulate you! That’s not true! I know that you are actually really envious of me.


A: B:

I promise you to come to your party. That’s not true! I know that you are not planing to come, because you’ll not even be in town then.

Another problem of the That’s not true-test is that it can also be used to target just a subpart of an assertion. (18)

A: B:

{I think that, Bill says that} Tom loves tofu. That’s not true! Tom doesn’t love tofu.

While the embedded content still is relevant for the truth conditions of the entire utterance, it is not asserted, which also casts doubt on the assumption that the test

If expressivism is fun, go for it! |


actually tests for assertions. Overall, it is therefore far from clear that the possibility of using the That’s not true-test is sufficient to establish the truth-conditional and assertoric nature of an utterance. The same is true the other way around as the impossibility to use the That’s not true-test does not necessarily establish nontruth-conditionality. Consider, for instance, non-restrictive relative clauses. (19)

a. Tom, who is the world’s biggest tofu lover, is a great poker player. b. #No, that’s not true! (He doesn’t like tofu.)

While it is not possible to use the That’s not true-test to deny the content of the non-restrictive relative clause in (19), the majority of the literature agrees that it conveys truth-conditional content. The fact that it cannot be targeted by a direct denial is commonly ascribed to the fact that the content of the non-restrictive relative clause is not at-issue (Potts, 2005; Gutzmann, 2015; AnderBois et al., 2015; Koev, 2013) or only a secondary assertion (Horn, 2002, 2013). That is, That’s not true rather seems to test for the way the content in question behaves in discourse pragmatic terms, instead of testing for truth-conditionality (Simons et al., 2010; Tonhauser et al., 2013).

3.2 Embeddability The argument from embeddability is actually more an argument concerning the status of expressive content as part of logical reasoning. It is based on the implicit assumption that being part of what looks like a logical deduction is evidence for the content in question being truth-conditional and assertoric content. However, without any further argumentation and evidence, this conclusion seems to be too hasty, as there are many non-assertoric speech acts that can take part in patterns of reasoning that very much look like the logical deduction Lasersohn presents as evidence for his argument against expressivism. Take, for instance, directive speech acts as expressed by imperatives. As the following example shows, directive speech acts can occur in the consequence of a conditional and hence be concluded upon by what resembles the use of modus ponens in (15). (20)

Modus ponens with non-assertoric speech acts: directives a. b. c.

If the roller coaster has a loop, go for it. The roller coaster has a loop. Therefore, go for it.

30 | Daniel Gutzmann

Similar observations can be made about commissive speech acts like promises, which can also be inferred by a modus-ponens like argument. (21)

Modus ponens with non-assertoric speech acts: commissives a. b. c.

If the roller coaster has a loop, I promise to ride it with you. The roller coaster has a loop. Therefore, I promise to ride it with you.

Likewise and more to the point of this paper, not only directive but also expressive speech acts can be part of such reasoning processes as well. (22)

Modus ponens with non-assertoric speech acts: expressive speech acts a. b. c.

If the roller coaster has a loop, I congratulate you for being brave. The roller coaster has a loop. Therefore, I congratulate you for being brave.

The weak spot of the argument from embeddability lies in the fact that it is not really clear what the reasoning processes Lasersohn (2005) refers to actually amount to. If he actually has logical deductions in mind in the sense of classical logic, i.e., inferences that are truth preserving, then expressive speech acts, which arguably do not receive a truth value, cannot be part of such inference patterns by definition. However, this is trivial and, without further argumentation, it is not clear what this shall tell us about the nature of PPTs, as it is not obvious whether the patterns in (15) that Lasersohn uses to support his argument are patterns of truth-preserving inferences. In fact, as we will see below, they actually do not seem to be. In any case, the examples in (20)–(22) show that some kind of inference is licensed even if non-truth-conditional utterances are involved. This, in turn, shows that there are other deductive systems available in natural language reasoning that are not directly tied to truth preservation in the straightforward sense of classical logic. Therefore, the fact that PPT-utterances can be part of such inference patterns is not conclusive evidence for the fact that they are truthconditional, as it is not clear whether the inference in (15) is of the strictly logical variety or whether it actually is more akin to (22). The problem, of course, is that there is no well-established deductive system for non-assertoric content like there is for assertive utterances. However, there are some attempts to provide such a system. For instance, in his famous manuscript on the meaning of ouch and oops, Kaplan (1999) sketches what such a system may look like. He argues, for instance, that expressive adjectives like damn can be part

If expressivism is fun, go for it! |


of deductive systems that go beyond logical validity based on truth (Kaplan, 1999, 22). (23)

Reasoning with expressives a. b. c.

Kaplan was promoted. I have a derogatory attitude toward Kaplan. Therefore, that damn Kaplan was promoted.

Without presenting a fully worked out theory for drawing such inference, Kaplan nevertheless draws a broad picture of how one can incorporate the nontruth-conditional meaning of expressives like damn, which—following Récanati (2004)—may be dubbed “use-conditional meaning”, into a deductive system (Gutzmann, 2015). The basic idea is that validity in such a system is not based solely on the notion of truth in the traditional sense, but rather on what Kaplan calls “truth-plus” or “truth with an attitude” (Kaplan, 1999, 10). This extended notion not only encompasses the ordinary truth-conditional content, but also incorporates the information conventionally expressed by expressives and other use-conditional expressions, like the negative speaker attitude conveyed by the use of damn. Under this view, logical validity then is not based on preservation of truth (in the classical narrow sense), but on what Kaplan calls “information delimination”. (24)

Semantic validity as information delimination

(Kaplan, 1999, 10)

There must be no semantic information in the conclusion that is not already contained in the premises. This notion of semantic validity can easily capture the intuition that of the two following inferences, only the former seems valid, while the later seems invalid.⁵ (25)

That damn Kaplan was promoted. Kaplan was promoted.


Kaplan was promoted.  That damn Kaplan was promoted.

Kaplan does not intend his considerations as a full-fledged account of reasoning with expressive content, nor do I, but this brief detour shall at least make clear that it seems possible to extend our standard deductive tool beyond just truthpreservation such that inferences like those in (23) or (25) can receive a formal

5 These are Kaplan’s (1999, 8) Argument 1 and 2.

32 | Daniel Gutzmann

analysis. I refer the reader to Gutzmann (2015) for details on how this can be achieved. For now, this brief detour should suffice to cast doubt on Lasersohn’s (2005) argument from embeddability. There is an additional argument to be made at this point. As Wolf (2016) [this volume] points out, the example Lasersohn (2005) choses does not generalize to other inference patterns. Note that in (15), the PPT is in the consequent of the conditional. If, however, a PPT occurs in the antecedent, the conclusion does not necessarily follow if the two premises are uttered by different speakers. Consider the following example (Wolf, 2016, example (12), p. 75). (27)

a. b. c.

Suzy: If the roller coaster is fun, I will buy a ticket. James: The roller coaster is fun.  Suzy: Therefore, I will buy a ticket.

The problem, of course, is that if Suzy’s and James’ taste regarding roller coasters differ, say Suzy loves fast roller coasters with loops and stuff, while James enjoys the slower ones that go through a nice garden, then the fact that James takes the roller coaster under discussion to be fun will not lead Suzy to the conclusion that she shall buy a ticket. This is similar to what happens if we switch speakers in a deduction involving expressive adjectives. In contrast to the inference in (23), which is valid under information delimination if we ascribe the premises and conclusion to the same speaker, a variant parallel to (27) is not. (28)

a. b. c.

Tom: Kaplan was promoted. Hanna: I have a derogatory attitude toward Kaplan.  Tom: Therefore, that damn Kaplan was promoted.

Note that the fact that the speaker changes should not matter if we are dealing with plain (non-indexical) truth-conditional utterances. (29)

a. b. c.

Tom: If Tofu is made from soy, it is healthy. Hanna: Tofu is made from soy. Tom: Therefore, tofu is healthy.

Of course, the reason why (27) as well as (26) are invalid is their subjective nature. However, in contrast to how they behave in the disagreement case, both PPTs and expressives pattern exactly like indexicals in the inferences. (30)

a. b. c.

Tom: If I am sad, I eat a lot of tofu. Hanna: I am sad.  Tom: Therefore, I eat a lot of tofu.

If expressivism is fun, go for it! |


This, at least, suggests that PPT-utterances are more akin to indexicals than they are to plain truth-conditional items. As I will argue below, PPT are actually indexical in their nature, at least one half of them, and this can explain why switching speakers in an argument can render it invalid like it does for indexicals.

3.3 Contradiction Lasersohn’s last argument against an expressive approach to PPTs, the argument from contradiction, is based on the assumption that only utterances that are true or false and make proper assertions can contradict each other (Lasersohn, 2005, 658). However, there are also non-assertoric speech acts that can be taken as contradicting each other. As before, consider directive speech acts first. (31)

Contradictions with non-assertoric speech acts: directives a. b.

Tom to Suzy: Take a ride on this roller coaster! Hanna to Suzy: Do not take a ride on this roller coaster!

Even if we are clearly not dealing with assertions, Tom and Hanna intuitively contradict each other in (31) insofar as the recommendations they give to Suzy are in conflict with each other. Suzy can only fulfill one of them. If we not artificially constrain the notion of contradiction to assertion by stipulation, this is as perfect a contradiction as we can get. And if used as a technical assertion-only notion of contradiction that excludes examples like (31), it could not be used as an argument for the assertive status of PPTs anymore. Besides directives, we can get a similar feel of contradiction in purely expressive speech acts. Consider the following example, which uses interjections instead of PPTs, but otherwise comes very close to the dialogue in (5) in terms of who commits and (dis)agrees to what. (32)

Contradictions with non-assertoric speech acts: expressive speech acts a. b.

Tom [eating tofu]: Mm-mm! Hanna [eating from the same tofu]: Ugh!

If we refrain from invoking the notion of truth, (32) seems to involve a similar kind of faultless disagreement as (5) does. Both Tom and Hanna display their taste for the tofu eaten and are justified in doing so (if they are sincere), while, simultaneously, they seem to disagree on whether the tofu is tasty or not. This means that either one gives up the assumption that contradiction and disagreement must be based on truth values and assertions, or one must assume that expressive utter-

34 | Daniel Gutzmann

ances as in (32) receive a truth-conditional interpretation after all. In either case, the argument from contradiction would become vacuous. Interestingly, Lasersohn (2005, 683) himself admits that even in the relativist solution he presents, the nature of the disagreement in (5) is left “somewhat mysterious”. He suggests that two sentences contradict each other if they “cannot both be accommodated under a single perspective”. Of course, this seems to be exactly what is going on in cases like (32). A single speaker cannot utter both interjections regarding the same target without conveying conflicting information. (33)

Tom [eating tofu]: #Mm-mm! Ugh!

Similar observations can be made regarding other non-assertoric speech acts. In each case, we can say that they contradict each other, if they cannot be felicitously performed by the same speaker. (34)

a. b.

Directives: Tom: #Take the roller coaster. Don’t take the roller coaster. Commissives: Tom: #I promise that you’ll get a bike for Christmas. I promise that you won’t get a bike for Christmas.

And of course, contradiction with plain assertions are also covered by this approach to contradiction. (35)

Tom: #Tofu is made from soy. Tofu is not made from soy.

This intuitive idea of a contradiction can be made more precise and subsumed under the notion of information delimination once spelled-out more formally. For now, we can conclude that also the argument from contradiction does not rule out an expressive approach to PPTs.

4 An expressive approach to PPTs Having rebutted Lasersohn’s (2005) three objections against expressivism, an expressive approach to the meaning of PPTs becomes a viable option again. In this section, I provide a first outlook of what such an expressive account may look like. I draw on recent work on the treatment of expressive content by Potts (2005, 2007a,b) and others (McCready, 2010; Gutzmann, 2015) that provides formal frameworks that draw on the basic insights of Kaplan’s (1999) suggestions of how to provide a proper semantic approach to expressive meaning. As we will see, the general strategy to account for PPTs and faultless disagreement in an expressive setting is to distribute the workload to two different meaning dimensions.

If expressivism is fun, go for it! |


While the faultlessness is established by the subjective truth-conditional content of an PPT-utterance, the disagreement happens on a use-conditional layer that, for PPTs, is normative and hence up to dispute. Before we can spell out the details of such a two-dimensional approach, we first have to lay out the foundations that we need in order to deal with use-conditional content.

4.1 Use-conditional meaning and hybrid semantics The basic idea of Kaplan’s approach is the observation that truth conditions alone cannot capture the whole range of conventional meaning in natural language. For certain expressions of natural language, a correct Semantic Theory would state rules of use rather than something like a concept expressed. (Kaplan, 1999, 6, emphasis mine)

However, in contrast to more radical theories of “meaning as use”, the twist of Kaplan’s agenda is that the (conventional) non-truth-conditional aspects of meaning are perfectly amenable to a treatment by the methods of formal semantics, once we recognize that the meaning of expressives and other use-conditional items can be captured by asking under which circumstances such expressions are felicitously used. Consider, for instance, the example of the interjection oops which Kaplan uses to illustrate his idea (Kaplan, 1999, 17). Instead of asking what makes an utterance of Oops! true, we ask in which contexts it can felicitously be uttered. For illustration, consider the use conditions given by Kaplan (1999, 17). (36)

“Oops!” is felicitously used, if the speaker just observed a minor mishap.

It is not important here whether (36) perfectly captures the use conditions of oops. What is crucial, is that, at its core, the way the use conditions in (36) are specified is quite akin to the traditional ways of specifying truth conditions. Displaying (36) side by side to a truth-conditional schema highlights the parallels. (37)




“Snow is white” 2 is true, 3 iff snow is white. “Oops!” 2 is felicitously used, 3 iff the speaker observed a minor mishap.

Both schemata connect an expression (in line 1) with a condition (in line 3) that is supposed to capture the meaning of the expression. The crucial difference is

36 | Daniel Gutzmann

the way in which the connection is established (by line 2): the “mode of expression”, as Kaplan (1999, 41) calls it. In case of (37), the condition is connected to the expression by the notion of truth, while it is felicity in the condition in (38). The crucial point about the parallelism between (37) and (38) is that even in the use-conditional case, the meaning is captured by a condition and a condition is something sentential that can hold or not for a particular use. This enables us to give even these non-truth-conditional contents a proper treatment, “since once we have the expressive content in sentential form, we can bring out our old toolbox, and go to work” (Kaplan, 1999, 9). That is, just like truth conditions as in (37) define a proposition, the use conditions in (38) do the same as well. One important difference though is that use, in contrast to truth, is always tied to being actually used. While the ordinary truth conditions defined by (37) can be given as set of worlds as in (39), we can therefore think of the use-conditional proposition that (38) establishes as a set of contexts as in (40).⁶ (39)

⟦(37)1 ⟧t = {w : snow is white in w}


⟦(38)1 ⟧u = {c : c S observed a minor mishap in c W }

Within this basic framework, the semantic content of an ordinary assertive utterance hence is given by a set of possible worlds and contexts for indexicals, while expressive speech acts denote a set of contexts. From this, we can then specify their truth and use conditions respectively. An assertion is true if the actual world is in the set of possible worlds denoted by the semantic content of the sentence. Likewise, an expressive speech act is justified or felicitous if the actual utterance context is in the set of contexts denoted by the semantic content of the sentence. We can easily state this prose formally. (41)

a. b.

“Snow is white” is true in w@ if w@ ∈ {w : snow is white in w}. “Oops!” is felicitous in c@ if c@ ∈ {c : c S observed a minor mishap in c W }.

So far, we have pretended that truth- and use-conditional utterances are completely separated and express only one kind of content. While this may be true for rather reduced utterances like Oops!, many utterances become hybrid speech

6 The superscripts distinguish between the truth-conditional and the use-conditional content of an expression. For further motivation for using contexts for modeling use conditions, see Gutzmann (2015) or Predelli (2013). Note that I treat contexts as a Kaplan-style tuple of (at least) a speaker c S , a judge c J and a world c W . Therefore, we cannot simply say that a particular fact holds for the context c, but have to say that it holds in c W (the world of the context).

If expressivism is fun, go for it! |


acts if they involve both truth- and use-conditional expressions. Consider, for instance, expressive adjectives inside otherwise assertive utterances. (42)

That damn Kaplan got promoted.

Due to the presence of the expressive damn, an utterance like (42) involves an assertoric truth-conditional part as well as an expressive use-conditional part. That is, it denotes both a set of worlds as its truth-conditional content and a set of contexts as its use-conditional content. (43)

a. b.

⟦(42)⟧t = {w : Kaplan got promoted in w} ⟦(42)⟧u = {c : c S dislikes Kaplan in c W }

I call expressions like (42) that encode both kinds of meaning “hybrid” expressions.⁷ Cases like (42) become hybrid expressions because of the inclusion of a use-conditional expression in an otherwise truth-conditional expression. But there are also lexically hybrid expressions that come with both meaning dimensions right from the lexicon. Consider the following German examples taken from Gutzmann (2011, 131). (44)

a. b. c.

Köter ‘cur’ ⇝ dog + expressing a negative attitude Bulle ‘cop’ ⇝ policeman + expressing a negative attitude Tussi ‘bimbo’ ⇝ girl + expressing a negative attitude

Since hybrid expressions mix both kinds of content, their meaning must be given by both truth and use conditions if we do not want to leave one meaning dimension unaccounted for. (45)

a. b.

(42) is true in w@ if w@ ∈ {w : Kaplan got promoted in w} (42) is felicitous in c@ if c ∈ {c : c S dislikes Kaplan in c W }

I call the general idea to employ both truth and use conditions alongside each other in order to capture the entire meaning of natural language expressions “hybrid semantics”. Without delving into the depths of a formal framework for hybrid semantics that calculates both meaning dimensions in a compositional fashion, the general format of hybrid semantics that abstracts away from the internal calculations can be given as follows (see Gutzmann 2015, 154).

7 For discussion and overviews over the varieties of hybrid content, see McCready (2010) and Gutzmann (2011, 2013).

38 | Daniel Gutzmann


Hybrid semantics ⟦S⟧ = ⟨⟦S⟧t , ⟦S⟧u ⟩ a. b.

S is true in w@ if w@ ∈ ⟦S⟧t . S is felicitous in c@ if c@ ∈ ⟦S⟧u .

It should be noted that hybrid semantics is a general idea that is relatively independent of the actual formal implementation. I present my own suggestion in Gutzmann (2015), which is based on previous work inspired by the work of Potts (2005, 2007b), but in principle, the hybrid semantics template in (46) can receive a formal underpinning in more or less any other formal framework.

4.2 A hybrid approach to PPTs Having laid out the basics of the use-conditional interpretation of expressive items and hybrid semantics, we can now turn back to PPTs and how the problem of faultless disagreement can be tackled within this account. The core idea of an account of PPTs along these lines is to distribute faultless disagreement to the two meaning dimensions available in a hybrid semantic setting. That is, on one level, we account for the intuition of faultlessness, and on the other we account for the intuition of disagreement. Understood in this way, PPTs are lexically hybrid expressions, like those in (44), as they are conventionally specified to contribute to both the truth- and use-conditional meaning dimension. Let us start with the question of what we would like to have on the truthconditional layer. First, as expressed by the faultlessness property, every speaker that sincerely utters a PPT-utterance intuitively expresses some truth about matters of her personal taste. Moreover, as examples (27)–(30) have shown, PPTs behave similarly to expressives and indexicals when used in inferences drawn from utterances of different speakers. All this speaks for giving a simple, indexical semantics for PPTs that relativizes the meaning of a PPT on the character level (in contrast to Lasersohn’s content-level relativism), similar to the first proposal in (8), where ⟦. . .⟧t signals truth-conditional content. (47)

⟦tasty⟧t,w,c = λx.x is tasty for the judge of c in w.

Note that I adopted the judge parameter as the parameter to which the PPT is relativized instead of the speaker. This complies with Potts’s (2007b) suggestion to adopt the judge parameter for expressives and keeps the intuition that sometimes, PPTs may have relativization points other than the speaker. In most cases, how-

If expressivism is fun, go for it!



ever, the judge will be the speaker, so that (47) will give rise to the following truth conditions for (5). (48)

a. b.

Tom: Tofu is tasty. ⇒ TC: Tofu is tasty for Tom. Hanna: No, tofu isn’t tasty at all. ⇒ TC: Tofu is not tasty for Hanna.

Obviously, the two propositions denoted by the truth-conditional content of (48a) and (48b) are perfectly compatible with each other and hence can actually both be true simultaneously. We can also already see how this approach accounts for the invalidity of the argument in (27). Under the presented analysis, the truth-conditional content of the premises and conclusion is actually more like in the following argument, which clearly is invalid. (49)

a. b. c.

If the roller coaster is fun for Suzy, Suzy will buy a ticket. The roller coaster is fun for James.  Therefore, Suzy will buy a ticket

Only under the addition of the premise that James’ taste of roller coasters is sufficiently similar to Suzy’s own roller-coaster preferences, one can accept (49). However, if this is indeed the case, the plain-PPT variant in (27) becomes a better argument as well. (50)

a. b. c. d.

Suzy: If the roller coaster is fun, I will buy a ticket. James: The roller coaster is fun. James’ taste regarding roller coasters is pretty much like Suzy’s. Suzy: Therefore, I will buy a ticket

The simple indexical analysis of the truth-conditional analysis, can also help to explain why the following utterances seem to be contradictory (see also Wolf, 2016, this volume). (51)

a. #Tofu is not tasty, but it is tasty for me. b. #Tofu is not tasty for me, but it is tasty.

However, as correctly identified by Lasersohn (2005), such an indexical analysis leaves the intuition about disagreement completely unaccounted for. This is where the multidimensionality of hybrid semantics kicks in. We can keep the indexical analysis on the truth-conditional layer and account for the intuition of contradiction in the use-conditional dimension of PPT-utterances. The intuition here is that PPT-utterances often carry a normative component about what shall be considered as tasty. Such an intuition does not only hold for PPT-utterances,

40 | Daniel Gutzmann

but for ethical judgments in general. Actually, the emotive approach to ethical terms proposed by Stevenson (1937) provides a good starting point for fleshing out a hybrid analysis of PPTs. He argues, that even though ethical judgments involve a descriptive component, their “major use is not to indicate facts, but to create an influence. Instead of merely describing people’s interests, they change or intensify them” (Stevenson, 1937, 18). For instance, when Tom says Eating meat is bad to Hanna, he does not just want to let her know that he (or a contextually inferable group of people) disapproves of eating meat. Rather, Tom is attempting to get her to disapprove of it as well (cf. Stevenson, 1937, 19). Taking this as inspiration, I propose to render the expressive component of a PPT-statement as an affective expression of a deontic attitude, where ⟦. . .⟧u signals use-conditional content. (52)

⟦tasty⟧u,w,c = λx.x shall count as tasty in c W .

(to be revised)

That is, while a truth-conditional disagreement about, say, whether tofu is made from soy or not, is a disagreement about objective facts in the world, the disagreement induced by PPTs like tasty is about what shall count as tasty in the utterance context. (53)

a. b.

Tom: Tofu is tasty. ⇒ UC: Tofu shall count as tasty. Hanna: Tofu isn’t tasty. ⇒ UC: Tofu shall not count as tasty.

Crucially, there seems to be no matter of fact that can settle whether tofu shall count as tasty in the context of (53). Again, this way of thinking about the disagreement that takes place in dialogues like (53), is anticipated by Stevenson’s (1937, 27) approach to ethical terms, who argues that two speakers arguing about whether something is good, disagree insofar as each one tries to redirect the other’s judgment. He compares this also to exchanges as the following, in which the speakers can be viewed to disagree in the same sense. (54)

a. b.

Tom: I want to eat tofu tonight. Hanna: I want to eat broccoli.

Even though they do not disagree in the technical sense—(54a) and (54b) express two independent propositions—Tom and Hanna disagree in their dinner wishes and try to align the other’s interest with their own by their utterance. Taking the two dimensions of PPTs together, we therefore arrive at the following truth and use conditions for our example PPT-utterance. (55)

Tofu is tasty is true in context c and world w if the judge of c likes tofu in w and it is felicitous in c if tofu shall count as tasty in c W .

If expressivism is fun, go for it! |


That is, the hybrid approach to PPTs combines a subjective statement of one’s own taste with a normative use condition about what should be considered as tasty in the current context.⁸ The latter aspect, of course, is a perfectly fine issue to disagree upon.

4.3 Deontic force, negation, and reasoning with PPTs A last aspect I want to address here at least briefly is the question of where the deontic, use-conditional meaning dimension of PPTs comes from and how it behaves in terms of projection behavior. Stevenson (1937, 23) argues that expressive, or, as he calls it, “emotive” meaning, “is a tendency of a word, arising through the history of its usage, to produce (result from) ‘affective’ responses in people.” Crucially, this tendency can even be observed with variants of PPT-utterances in which the speaker takes care to just relativize them to her own perspective, for instance, by making the judge explicit. (56)

a. b.

Tom: I know many people hate tofu, but it is tasty for me. Hanna: No, tofu is not tasty.

Even if Tom tries to make clear that he is just making a statement about his own statement, Hanna can still felicitously disagree with that. This contrasts with plain objective utterances. Consider the following dialogue that is parallel to (56). (57)

a. b.

Tom: Many people have a silver car, but I have a green one. Hanna: #No, I have a silver car.

Similar observations can be made if the PPT-sentence is embedded under an attitude predicate. Even in such cases, rejecting the embedded PPT-judgment still is viable. (58)

a. b.

Tom: I find tofu tasty / I think tofu is tasty. Hanna: No, tofu isn’t tasty.

It therefore seems to be the case that the deontic force of the use-conditional dimension of PPTs is always active in such constructions. This is correctly predicted by the projection behavior that is commonly ascribed to use-conditional content

8 “The use of PPTs is normative, their meaning is not” (Ariel Cohen, in his talk at the DGfS workshop that lead to this volume). Compare this again also to Stevenson’s (1937, 25) ideas, who compares This is good with I do like it; do so as well, which comes pretty close to the hybrid analysis suggested here.

42 | Daniel Gutzmann

(Potts, 2007b; Gutzmann, 2015). However, there is a complication. Consider again the first premise in the inference in (15) that Lasersohn (2005, 657) uses for his argument from embeddabilty. (59)

If there is a loop, the roller coaster is fun.

Interestingly, it is not only the truth-conditional component of the PPT that becomes unasserted in the consequent of a conditional—that is, (59) does not make any statement about whether the roller coaster is fun for the speaker or judge— also the deontic use-conditional interpretation vanishes. This seems to be a counterargument against treating it as an expressive component, as expressives usually escape conditionalization, as in the following example. (60)

If he publishes another paper, that damn Kaplan will be promoted.

In contrast to the PPT in (59), the expressive damn in (60) still has its expressive force so that the entire utterance still displays a negative speaker attitude towards Kaplan. A related problem is that, in contrast to expressives, PPT can easily be negated, which cancels the positive deontic reading. First, consider the following example in which an expressive adjective occurs in a negated sentence. (61)

That damn Kaplan did not get promoted. a. b.

TC: Kaplan did not get promoted. UC: The speaker has a negative attitude towards Kaplan.

The negation in this example only targets the truth-conditional content, but crucially leaves the use-conditional dimension untouched. That is, (61) expresses that Kaplan did not get promoted and that the speaker (or judge) as a negative attitude towards Kaplan. Again, this is different from how PPTs behave under under negation. To see this, consider Hanna’s response from (5) that denied Tom’s positive judgment about the tastiness of tofu. (62)

Hanna: Tofu isn’t tasty at all. a. b.

TC: Tofu is not tasty for Hanna. UC: Tofu shall not count as tasty.

Crucially, the negation in (62) does not only target the truth-conditional content that tofu is not tasty for Hanna, but it is also active in the use-conditional dimension. Obviously, an utterance of (62) does not express that tofu shall count as tasty, but that tofu shall not count as tasty. Otherwise, there would not be disagreement in (5) and Hanna’s utterance should be almost self-contradictory.

If expressivism is fun, go for it! |


Prima facie, these observations seem to speak against an expressive analysis of PPTs. However, we can solve this obstacle, if we think a bit about the source of the deontic component. As already detailed in the quote from Stevenson (1937) given above, using PPTs seems to produce affective responses. The reason may be that if one makes a PPT-utterance, one is first and foremost making a claim about one’s own taste, as captured by the indexical truth-conditional component. However, due to its subjective nature, such a statement is relatively uninformative for the hearer, at least if she is not engaged in finding out about the speaker’s taste. That is, there may be pragmatic forces that drive PPTs and other ethic predicates like good to develop the deontic use-conditional dimension of meaning. Even if this may be rather speculative and awaits further diachronic research, it is clear that even synchronically, the deontic force of a PPT depends on the assertion of its truth-conditional component. When you do not assert that, say, tofu is tasty for you, you neither make a deontic claim that it shall count as tasty. If we therefore link the use-conditional component of PPTs to the truth-conditional content, we can explain why their deontic force vanishes under negation and conditionalization. (63)

⟦tasty⟧u,c,w = λx.[x is tasty for the judge of c W ↔ x shall count as tasty in c].

The use conditions for PPTs are therefore that one shall only assert the subjective truth-conditional content of a PPT-sentence, iff one believes that the taste judgment shall objectively hold in the utterance context. The bi-conditional in (63) gives us the desired results. In positive sentences that asssert the subjective truthcondtional content of a PPT, we can derive a positive deontic attitude from the use-conditional content. For instance, consider Tom’s utterance and the inference licensed by its truth- and use-conditional content. (64)

Tom: Tofu is tasty. a. b. c.

Tofu is tasty for Tom. Tofu is tasty for Tom ↔ Tofu shall count as tasty. TC+UC: Tofu shall count as tasty

Likewise, in negative taste judgments like Hanna’s denial in (5), we can derive a negated deontic attitude if we take the two dimensions together. (65)

Hanna: Tofu is not tasty at all. a. b. c.

¬(Tofu is tasty for Hanna). Tofu is tasty for Hanna ↔ Tofu shall count as tasty. ¬(Tofu shall count as tasty).

44 | Daniel Gutzmann

Depending on the utterance context and what is known about Hanna’s food preferences, the result then may be strengthened pragmatically to the stronger negative deontic attitude that tofu shall count as not tasty. In contrast, if a PPT-sentence is used in the antecedent of a conditional, the two dimensions of content do neither warrant an inference to a positive attitude nor an inference to a negative one, which is exactly what we observed above. (66)

Tom: If tofu is tasty, I’ll buy some. a. b. c.

If tofu is tasty for Tom, Tom will buy some. Tofu is tasty for Tom ↔ Tofu shall count as tasty.  Tofu shall count as tasty

Furthermore, stating the semantics of the use-conditional dimension in this way, still can explain why PPTs embedded under attitudes, as in (58), still may enfold their deontic force. Due to the subjective character of PPTs and that speakers in most cases have privileged epistemic access to their taste, asserting that one finds tofu tasty comes very close to asserting that it is tasty for oneself. In a similar vein, a PPT can still receive a deontic reading even if it is embedded under a third person attitude report, presupposing that the subject of the attitude report is considered to be the contextual judge. Imagine, for instance, two laymen discussing whether tofu is tasty without having tried it yet. If Mary is a person whose taste is considered to be good, the following utterance can come with the use condition that tofu shall count as tasty in the utterance context, given the additional premise in (67c) that Mary is right in her belief about her taste. (67)

Mary thinks that tofu is tasty. a. b. c. d.

Mary thinks that tofu is tasty for her. Tofu is tasty for Mary ↔ Tofu shall count as tasty Mary thinks that tofu is tasty for her → Tofu is tasty for Mary. Tofu shall count as tasty

This also complies with the intuition about the function of referring to external judges in order to settle disputes about taste.

5 Conclusion In his paper on PPTs, Lasersohn (2005) presents three arguments against an expressive approach to PPTs and the problem of faultless disagreement. All three arguments, which I presented briefly in Section 2, are not conclusive. If one

If expressivism is fun, go for it!



actually considers expressive speech acts beyond mere interjections, as I have done in Section 3, all his tests point out many similarities between PPTs and non-assertoric speech acts. This led me to explore the possibility of an expressive approach in more detail. The core idea of my proposal is to use a hybrid approach to PPTs that deals with the two sides of faultless disagreement on different meaning dimensions. Truth-conditionally, PPT-utterances assert a simple, subjective taste-judgment, which accounts for the intuition of faultlessness. On the use-conditional layer, however, such utterances give rise to a deontic attitude towards what shall count as tasty in the utterance context. This is what speakers actually disagree about when they argue about taste as in (5). With such a multi-propositional approach, we therefore “can have our subjective cake [. . . ] and eat it objectively, too” (von Fintel, at the workshop that lead to this volume). The argument Tom and Hanna have is therefore an argument about what should count as tasty in the context that involves both speakers. They argue because establishing a common context is an important point of communication. After all, the analysis I sketched in this paper could be regarded as both “expressivism” and what Lasersohn (2005, §4.4) calls a “metacontextual-conflict account”. Finally, taking a closer look at the inference patterns associated with PPT-utterances, I refined the use-conditional dimension by making it dependent on the assertion of the truth-conditional dimension. This not only captures the intuition about the pragmatic origin of the deontic character of PPTs, but also gets their inference and projection behavior right. Acknowledgment: I would like to thank the audience of the DGfS workshop on Subjective Meaning at Berlin for helpful discussion. Thanks also to Erik Stei, who was the first one to get me interested in this topic, and Carla Umbach and Lavi Wolf, with whom I had insightful discussions of these matters. An anonymous reviewer provided very helpful comments that helped to improve the first draft of the paper. The same is also true for Janneke van Wijnbergen-Huitink and Cécile Meier. Working with them was fun! Needless to say, that all remaining, untasteful or unfunny mistakes are my own.

References AnderBois, Scott, Adrian Brasoveanu & Robert Henderson. 2015. At-issue proposals and appositive impositions in discourse. Journal of Semantics 32(1). 93–138. Gutzmann, Daniel. 2011. Expressive modifiers & mixed expressives. In Olivier Bonami & Patricia Cabredo-Hofherr (eds.), Empirical issues in syntax and semantics 8, 123–141. Paris: Colloqué de Syntax et Sémantique à Paris.

46 | Daniel Gutzmann

Gutzmann, Daniel. 2013. Expressives and beyond. In Daniel Gutzmann & Hans-Martin Gärtner (eds.), Beyond Expressives, 1–58. Leiden: Brill. Gutzmann, Daniel. 2015. Use-conditional meaning (Oxford Studies in Semantics and Pragmatics 6). Oxford University Press. Horn, Laurence R. 2002. Assertoric inertia and NPI licensing. CLS 38(2). 55–82. Horn, Laurence R. 2013. I love me some datives: Expressive meaning, free datives, and fimplicature. In Daniel Gutzmann & Hans-Martin Gärtner (eds.), Beyond expressives, 151– 199. Leiden: Brill. Kaplan, David. 1989. Demonstratives. In Joseph Almog, John Perry & Howard Wettstein (eds.), Themes from Kaplan, 481–563. Oxford: Oxford University Press. Kaplan, David. 1999. The meaning of ouch and oops. Unpublished manuscript. Koev, Todor. 2013. Apposition and the structure of discourse. PhD thesis. New Brunswick, NJ: Rutgers University. Kölbel, Max. 2004. Faultless disagreement. Proceedings of the Aristotelian Society 104(1). 53–73. Lasersohn, Peter. 2005. Context dependence, disagreement, and predicates of personal taste. Linguistics and Philosophy 28(6). 643–686. McCready, Eric. 2010. Varieties of conventional implicature. Semantics and Pragmatics 3(8). 1–57. Potts, Christopher. 2005. The logic of conventional implicature (Oxford Studies in Theoretical Linguistics 7). Oxford: Oxford University Press. Potts, Christopher. 2007a. The centrality of expressive indices. Theoretical Linguistics 33(2). 255–268. Potts, Christopher. 2007b. The expressive dimension. Theoretical Linguistics 33(2). 165–197. Predelli, Stefano. 2013. Meaning without truth. Oxford: Oxford University Press. Récanati, François. 2004. Pragmatics and semantics. In Laurence R. Horn & Gregory Ward (eds.), The handbook of pragmatics, 442–462. Oxford: Blackwell. Searle, John R. 1969. Speech acts. Cambridge: Cambridge University Press. Simons, Mandy, Judith Tonhauser, David I. Beaver & Craige Roberts. 2010. What projects and why. Proceedings of SALT 20. 309–327. Stevenson, Charles Leslie. 1937. The emotive meaning of ethical terms. Mind 46(181). 14–31. Tonhauser, Judith, David Beaver, Craige Roberts & Mandy Simons. 2013. Toward a taxonomy of projective content. Language 89(1). 66–109. van Wijnbergen-Huitink, Janneke. 2016. Subjective meaning: An introduction. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 1–20. Berlin: De Gruyter. [This volume] Wolf, Lavi. 2016. Predicates of personal taste and the evidential step. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 69–90. Berlin: De Gruyter. [This volume]

Sanna Hirvonen

Doing without judge dependence Abstract: Semantic relativists hold that disagreements of taste are “faultless disagreements”, i.e. the speakers are expressing contradictory contents while neither of them is at fault. This paper argues that we should distinguish between subjectivist and objectivist uses of predicates of taste and only the former fits the pattern of faultless disagreements. Objectivist uses are made on the basis of an objectivist “folk” theory of taste which holds that there is an objective truth of the matter and hence, only one of the disagreeing parties can be right. Moreover, they constitute the majority of uses and hence the semantic theory should primarily be concerned with them. The problem is that objectivism as a metaphysical theory of taste is quite plausibly false. I argue that if one accepts the so-called “principle of semantic competence” (Stojanovic, 2007) as the recent theorists systematically do, the semantics of predicates of taste should be objectivist (i.e. non-judge-dependent). Objectivist semantics coupled with the falsity of metaphysical objectivism about taste leads to an error theory of taste discourses.

1 Introduction The problems behind giving a semantic analysis of predicates of taste are due to the following conflict. Most theorists take taste to be subjective, and hence they agree with the dictum of the ancient Romans: De gustibus non est disputandum. However, looking at the actual uses of taste predicates shows that disagreement is perfectly commonplace. The challenge is thus considered to be that we must give an analysis of taste predicates which makes them subjective—or “judgedependent” in the recent terminology—but which can also explain how disagreement is possible. However, the problem with many recent semantic analyses of taste predicates is the assumption that there is just one kind of use of taste predicates, and those uses are both subjective and allow for the possibility of disagreements. This paper has two aims. First, I argue that there are two distinct uses of predicates of taste, subjectivist and objectivist uses, and only the former can be explained by accounts that posit judge dependence. Subjectivist uses only express Sanna Hirvonen, Department of Philosophy, University College London, Gower Street, London WC1E 6BT, Great Britain, e-mail: [email protected]

48 | Sanna Hirvonen

the speaker’s own taste, and there can be no genuine disagreements when the predicates are used in the subjective way. In contrast, objectivist uses allow for genuine disagreements, and they are made on the basis of a belief that there exists a unique standard of taste that holds for all persons. Moreover, objectivist uses constitute the majority of uses and hence they should be the data that forms the basis for the semantics of predicates of taste. Second, I discuss what happens if we accept that the objectivist uses are the default uses but there are no objective taste properties which could make the judgments true. I argue that if one accepts the so-called “principle of semantic competence” which holds that speakers know the truth-conditions of their utterances, then one should accept an error theory of taste discourse. In other words, objectivist judgments of taste aim to attribute objective properties but since those properties don’t exist, the judgments are systematically false. The plan of the paper is as follows. The first section argues that there are the two uses of predicates of taste which can be distinguished on the basis of the different behavioural dispositions they give rise to. The subjectivist uses give the impression that the speaker is really talking about their own taste, and they underlie the intuition that disagreements of taste are faultless. Relativists like Kölbel (2003) and Lasersohn (2005) have exclusively focused on the subjectivist uses. But there are uses which do not fit the subjectivist model, and which I call “objectivist” because I argue that they are made on the basis of an underlying, objectivist folk theory of taste (Section 3). Objectivist uses lead to very different behaviours than subjectivist uses. For example, if two speakers are making objectivist judgments and they disagree, they don’t take the disagreement to be faultless. On the contrary, they believe there is a judge-independent truth about the matter and hence one of them has to be wrong. They may provide arguments and reasons for their judgment, aim to convince the other, be prepared to give up their view if they are shown to be wrong, and retract their past judgments as false if they now hold a different opinion. Moreover, I will argue that the objectivist uses constitute the majority of uses of predicates of taste, and hence the semantic account must build on those uses, not on the more marginal subjectivist ones. Section 3.2 argues that objectivist uses cannot be explained by subjectivist theories which posit judge dependence. Section 3.3 discusses the claim that objectivism is false, and moreover, it is so evidently false that speakers couldn’t possibly be making objectivist judgments. I argue that the latter claim is without grounds. The last part (Section 4) discusses the consequences of supposing both that objectivism as a metaphysical thesis is false, and that objectivism is the folk theory which underlies the majority of uses of predicates of taste. Section 4.1 discusses the reasons to choose between subjectivist and objectivist semantics under that

Doing without judge dependence | 49

supposition, and concludes that if one accepts the principle of semantic competence one ought to accept an error theory of taste. Section 4.2 discusses the consequences of the error theory of taste to communication, and argues that even if objectivist metaphysics about taste is false, there is no reason to be alarmed by the error involved in objectivist taste statements. First, false or truth-valueless utterances can be perfectly meaningful as illustrated by our normal talk about non-existent entities such as zombies or Pegasus. Secondly, objectivist judgments of taste are not far from being true. Hence even the empirical error involved is negligible in most contexts. After all, a very large number of our ordinary statements are only approximations of truth.

2 The two uses of taste predicates 2.1 Subjectivist uses There are many moves one can make in discussions and disputes about taste. At one extreme one merely aims at stating something about one’s own taste—call these “the subjectivist uses”. One makes an explicitly subjectivist judgment of taste by using various linguistic tools for subjectivising one’s judgment, for example: (1)

To me curry wurst is delicious.


I find curry wurst delicious.

However, the standard view is that a bare judgment of taste like “Curry wurst is delicious” may also be implicitly subjective, i.e. its content, truth-conditions or its pragmatically conveyed content may be about the speaker.¹ Arguably, subjectivist uses are also revealed by the speaker’s behaviour in situations where a disagreement might arise or one’s past judgments are brought up. Subjectivist uses won’t give rise to serious, ongoing disagreements because what a subjectivist use of a judgment of taste conveys is truth-conditionally or pragmatically equivalent to an explicitly subjective judgment. Let me illustrate. Suppose Alice finds oysters disgusting. She is now at a restaurant with Bob where we can imagine the following exchange to take place:

1 On this point most of the recent relativist and contextualist theories agree, and their disagreement concerns rather the correct semantic framework.

50 | Sanna Hirvonen


a. b. c.

Alice (pointing at oysters): Look, oysters! They’re disgusting. Bob: No they’re not, they’re delicious. Alice: I find them pretty gross.

Suppose that in (3a) Alice only intended to express her own disgust towards oysters. Since she didn’t make an explicitly subjective judgment Bob can negate the sentence she used. However, since Alice never meant to talk about anyone else’s taste, she doesn’t insist on the disgustingness of oysters and instead she reformulates her judgment in an explicitly subjective form in (3c). By doing that she signals that she isn’t disagreeing with Bob but only making a statement about her own taste. The exchange (3a)–(3b) is an instance of the kind of “faultless disagreement” that the relativist and contextualist theories typically focus on. Alice and Bob are making judgments that seem to contradict, but we feel that both are entitled to their opinion and neither is at fault. In some situations one’s past, conflicting judgment of taste may be brought up, leading one to retract a past judgment if one’s taste has changed. MacFarlane (2009; 2014) has used the fact that people may retract their past judgments of taste as an argument for his brand of relativism. However, an account of predicates of taste must also explain why sometimes people don’t retract their judgments. Here is an example. Some years have passed since Alice and Bob’s previous dialogue. Alice has come to love oysters, and she is again in a restaurant with Bob who doesn’t know about her change of heart. (4)

a. b. c.

Alice: Oysters are so delicious! Bob: What? Last time we met you said they’re disgusting. Alice: I know, my taste has changed. They were disgusting to me then.

Here Alice again emphasises that her previous judgment was about her taste back then, rather than taking back what she said as false. In what kind of situations are predicates of taste used subjectively? First, a person who thinks that “there’s no accounting for taste” will only make subjectivist judgments of taste. She is well aware of only ever aiming to speak about her own taste and consequently she will never engage in serious disagreements, and she doesn’t retract her past judgments (unless she realises that her taste was temporarily aberrant then, e.g. as a consequence of being sick). But also, an objectivist about taste may make a subjectivist judgment when she only wants to talk about her own taste. Let us next look at what kind of people objectivists about taste are.

Doing without judge dependence | 51

2.2 Objectivist uses There is another kind of use of judgments of taste which lead to very different behaviour than the subjectivist uses. I don’t think there is anything controversial in accepting the existence of the conversational behaviours that I’ll describe; however, what is controversial is to treat them as “objectivist uses” of predicates of taste. An objectivist use as I define it is a judgment which aims to state facts about taste that are independent of the taste of the speaker or of any other particular individual or group. For example, if Matti says “Krombacher is better than Becks” and intends his use to be objectivist, he isn’t only saying that Krombacher is better than Becks to him, or to him and some relevant group. Instead he takes it to be a fact about the beers that Krombacher is better than Becks, for any person (as long as they don’t have a bizarre taste).² The linguistic behaviour that objectivist uses lead to contrast strongly with the subjectivist uses. First and foremost, serious and ongoing disagreements are possible almost uniquely with objectivist uses.³ By ongoing disagreement I mean one that goes on after the speakers have discovered that they hold opposing opinions, and by serious that the speakers take themselves to truly disagree, and do not hold that no one is at fault. In other words, speakers do not take objectivist disagreements to be faultless. Let us call them “genuine disagreements”. What are the characteristics of a genuine disagreement? Let us look at Matti’s objectivist judgment “Krombacher is better than Becks” again. In case someone disagrees with him he will defend his statement. He can do that by offering arguments, by not taking the opponent’s point of view seriously (e.g., by taking him or her to be insufficiently sophisticated regarding beer), by specifying what for him constitutes a good beer and so on. If the person Matti is discussing with makes objectivist judgments as well, the disagreement may go on for a long time without them coming to an agreement while both continue to think they are right. Alternatively one of them may become convinced and change her mind; or they may come to think that they don’t know how to resolve the dispute and end it at that; or they may even ask a third person’s opinion, or consult an expert. In short, the disagreement follows the pattern typical to disagreements about ordinary facts.

2 Note: good has many other senses than being gustatorily good but our examples only intend to use the gustatory sense of good and better. 3 There are cases of ongoing disagreements which are not a consequence of objectivist uses, but of a desire to e.g. calibrate the use of words in the context (see e.g. Sundell, 2011). However, I think such disagreements are marginal.)

52 | Sanna Hirvonen

Another behavioural disposition that goes with objectivist judgments is retracting one’s judgments once one has changed one’s mind. For example, suppose Matti later comes to think that actually Becks is better than Krombacher. If someone now reminds him that he used to think the opposite, he will take back his judgment by saying e.g. “I know, I was wrong”. It’s worth noting that such cases are disagreements with one’s past self. Since in retracting one admits that one was wrong, that’s more evidence that there are disagreements of taste where someone is taken to be wrong (in these cases, the speaker’s own past self). What is worth noting is that objectivist uses are particularly challenging for a theory of semantics of predicates of taste which takes the existence of faultless disagreements as the basic datum to be explained. Indeed, the main contextualist and relativist theorists (e.g. Glanzberg, 2007; Lasersohn, 2005; Kölbel, 2003; MacFarlane, 2014) don’t acknowledge the existence of objectivist uses as I have described them at all. The reason is obvious; whereas subjectivist uses invite to relativise truth or to posit a hidden parameter or free enrichment, objectivist uses are much harder to account for as we’ll see later. Before we return to the two uses and their significance to the semantics of predicates of taste, let us discuss what underlies the two uses of taste.

3 Objectivism about taste 3.1 The folk view about taste In this section I will argue that objectivist uses are pretty much the default use of predicates of taste, and that they are a consequence of a simple and appealing theory of taste—our “folk” theory so to speak, because it is the theory we usually grow up with and only give up after reflection. For our purposes we don’t need to attempt at a precise characterisation of it but generally the theory includes something like the following statements: Objectivism about value 1. Many objects have value properties. 2. Humans track the value properties of objects. 3. A person can be a more or less accurate tracker of those properties. 4. A value judgment predicates value properties. 5. One can be wrong in one’s value judgments. 6. Value judgments are true iff they correctly track the value properties.

Doing without judge dependence | 53

The statements are about value properties and value judgments generally and so apply to e.g. aesthetic or moral judgments as well. The idea is that objectivist uses of judgments of taste are manifestations of “implicit beliefs” in the objectivist theory as applied to taste-related evaluative properties like deliciousness, fun, or tastiness. To clarify, I don’t mean that people have explicitly formed a theory in their mind which they try to express by their judgments of taste. Rather, people might not be able to state the folk theory, but their linguistic behaviors make most sense when we assume that they believe the theory. That is the idea behind implicit beliefs. The statements above speak abstractly of tracking value properties, and the method by which one tracks them varies between the subject matters. For example, properties like deliciousness and tastiness are tracked by the senses of smell, taste and touch which together create the perception of flavours. For aesthetic judgments, traditionally philosophers have spoken of “the faculty of taste” which is exercised when making an aesthetic judgment about say, beauty. Generally, the notions of a good or bad taste are linked with the idea of being able to correctly track the taste-related value properties. A person with a bad taste attributes value where none is or simply fails to track the value of an object. Since our grounds of judgments of taste are our sensory abilities, we should rather expect people to be objectivists about taste. For example, we all agree that people with normal vision track the visual properties of the world. Not the visual properties to someone, but just the visual properties. In other words, there is a fact about the matter of how something looks or sounds, so it’s rather natural to think that there are also facts about the evaluative properties which are tracked by our other senses. For illustration about objectivist thinking in practice, let us return to Matti and his objectivist judgment “Krombacher is better than Becks”. Why he says that is because he believes he can track the properties that make beers good, in other words, that he has a sufficiently good taste regarding beers to make an objectivist judgment. Why he thinks he has a good taste might be because he is experienced in beer tasting, because he is confident that he is naturally apt at it, or because his evaluations about beers are generally met with agreement; his reasons are not important, but typically they play a role in determining his behaviour if his statement is challenged. Naturally Matti also thinks that what he said is true, and it is true because Krombacher is better than Becks; and again, not just to him, but to everyone (whether or not they realise it). Unless Matti has an exceptionally high opinion of himself as a beer taster he is also ready to concede that he might be wrong. He could end up changing his mind if say, his favourite beer expert announced that Becks was better than Krombacher, or all of his usually reliable friends told him

54 | Sanna Hirvonen

so. In that case he would most likely retreat to a subjectivist statement by saying e.g. “Ok, maybe Becks is actually better but I still like Krombacher more”. So, the first and strongest argument for objectivism as the folk view about taste is that it’s the best explanation for genuine disagreements of taste. But I think other facts about the culture surrounding taste support the view as well. The following are examples of cases where the people’s behaviours are best explained by attributing them folk objectivist beliefs. There are plenty of disagreements of taste where people basically get angry if you try ending their disagreement with saying “There’s no disputing about taste” (try saying that to fans of food, drink or music). I think they get angry because they think you’re wrong: there is disputing about taste. One can induce long discussions in Italians by asking them questions like “Which is the best kind of pasta?”. It’s hard to believe that the topic would be so fascinating if they took it to be merely a matter of everyone expressing their personal preferences. Or watch the French talk hours about food and try making sense of it thinking that they’re all just talking about their own taste. Or, imagine that you’re having a meal at a relative’s or a friend’s place and they ask how the meal was. If people were subjectivists about taste, it should be ok to say “I didn’t like it”. But by saying that you would actually be implicating that the food wasn’t good, objectively speaking, thereby upsetting the cook. The very idea of good and bad taste makes sense only on the background of objectivism: If taste was just a matter of personal likes and dislikes, what would it mean that someone has a bad taste? Admittedly, we can talk of bad taste ironically or as an exaggeration. But nevertheless, there is even a certain amount of consensus regarding what counts as bad taste. For example, Stern & Stern (1990) offer a fascinating look into visual and cultural phenomena that are broadly considered to be of bad taste—ironically perhaps, but if taste was merely a personal matter, talk of bad taste would not work, even ironically. The importance of matters of taste is shown in the pleasure we take in discussing and disagreeing over the values of things: which foods or drinks are good and why, which clothes are stylish or cool, which activities are most fun. I’ve argued that genuine disagreements make best sense of the basis of objectivism, but so do serious discussions about taste. The point of explaining to someone why one shouldn’t add ketchup to a bolognese sauce is typically not only that one personally doesn’t like it, but that there is something inherently (gustatorily) bad about the combination. Hence, by convincing another of what one considers a truth about taste (e.g. “There should be no ketchup in bolognese sauce”) one aims to make them better off. That’s the point of truths: they’re informative, and knowing truths usually makes life easier. In contrast, truths about one’s personal taste are irrelevant to pretty much anyone else. Supposing that judgments of taste are

Doing without judge dependence |


always subjective fails to make sense of the importance we grant to discussions about taste in our lives. Admittedly it is difficult to tell genuinely objectivist judgments from mere exaggeration or loose talk. But still, objectivism about taste is not only an unreflective attitude people lapse into when drinking beer but a default approach to taste which is manifest in various forms in our societies. Any travel guide is filled with authoritative statements about the gustatory or other taste-related values of places, activities and restaurants, and the same holds of other guidebooks and reviews in general. People follow the advice because they trust that the critics are sophisticated, knowledgeable and able to discriminate the good from the bad. If the advice leads one for example to eat food that seriously displeases one, a typical response is to think that the critic doesn’t know what she is talking about; she doesn’t have a good taste, and shouldn’t be working on a profession that requires it. Let me give an example of the kind of cultural phenomenon that I take to manifest objectivism about taste. Here’s a quote from an article by the French correspondent of the BBC news: Take one picture I saw recently (...) It was supposed to be a bitter-sweet comedy (...) What we got was two hours of inconsequential, plotless twaddle (...) Or take the comedy Le Vilain (The Villain) which has just come out. It is billed as an offbeat caper (...) my 14year-old daughter and I both agreed, on emerging, that there was one minor drawback: it wasn’t funny. (...) My point is that (...) what we had been led to believe was that these films were actually pretty remarkable. (...) It was the serious critics in Le Monde, Le Figaro and elsewhere, who used adjectives like hilarious, tender, burlesque, complex, original.” (

The quote is from an article that aims to show how the French culture is doing poorly because they refuse to see its poorness, and instead even the serious critics praise mediocre works of art. The author is so confident in his own and his 14year-old daughter’s abilities to discern what is genuinely funny that he concludes that all the critics are either self-deluded or intentionally misleading because they want to maintain that French movies are still great. (The author even admits later in the article that others in the audience did find it funny: “I may not have found Le Vilain funny, but a lot of people in the audience were in stitches.”) Again, ironic or not, the journalist’s very argument on the poor state of French culture relies on his own ability to track the funniness of films (an ability he considers superior to the abilities of all those others in the audience who were in stitches). If subjectivism about taste was indeed the default view as most philosophers hold, the author would have simply concluded after the film that his sense of humour is different from that of the French. But instead, he goes for an elaborate

56 | Sanna Hirvonen

conspiracy theory of lying critics because he believes that they really do evaluate the film as he does—as not funny. I’ve argued that objectivism is the folk view. But I also want to add that there shouldn’t be anything surprising that it is the folk view. After all, our tastes are pretty much alike. Compare matters of taste to some clearly objective matters of fact like whether a certain economical or historical theory is correct. In the latter case there are major epistemological obstacles in finding out the truth, but clearly there is a truth. Likewise we may suppose that it’s difficult to find the truth about taste but often we do agree, and when we don’t, we need to keep searching the truth. We know that with matters of taste education and experience change us and make our tastes more refined. So it is a rather natural assumption that we would converge in our tastes if only we had the same experiences. We all know that adolescents have a different taste to adults and even if it’s partly due to random fashions and their level of development, it’s also due to the fact that they’ve had little experience of the relevant objects of taste. Suppose one accepts that folk objectivism makes best sense of most uses of judgments of taste. What needs to be addressed is the question about the relation of objectivism to people’s grounds for their judgments of taste. After all, as the relativists emphasise, we are faultless because we judge on the basis of our own personal experiences of which we cannot be mistaken. I don’t think there’s a conflict there. Think about vision for example. We all see things pretty much the same way, but sometimes we misperceive. Our judgments about the looks of things are grounded in our subjective visual experiences, but we think that everyone perceives alike and therefore subjective experience is a guide to objective properties. The same line of thinking applies to taste properties. In other words, the folk epistemology of taste holds that overall one’s personal experiences are reliable but not infallible guides to the objective taste properties. This ends my argument that objectivist thinking underlies most uses of judgments of taste. But, one may ask, if objectivism is the default approach to taste, then how come subjectivism is so dominant in philosophy and linguistics that the current semantic accounts focus almost exclusively on the transparently subjective uses like faultless disagreements? Maybe because subjectivism is a consequence of reflection and theorising. I’ve argued above that objectivism is a very natural position to have since judgments of taste are based on perception just like judgments about looks or sounds. It’s only when one begins to really think about taste, or perhaps witness too many unsolvable disagreements, or read about facts of flavour perception that one becomes a subjectivist. I have described two uses of predicates of taste which display rather different behaviour. But I should add that we need not think that there are only purely

Doing without judge dependence | 57

objectivist and subjectivist uses. Some uses may be somewhere in between, and the similarity of tastes plays a crucial role here. As long as one remains within one’s taste peers (persons with a very similar taste) it is easy to make objectivist statements since the peers will tend to agree. But the more varied the group of discussants is regarding their tastes, the more it encourages the subjectivist stance simply because no agreement seems to be reachable. Furthermore, some predicates may standardly be used more subjectively than others just because there is less agreement about the phenomenon (fun might be the best example).

3.2 Subjectivist semantics cannot explain objectivist uses Suppose objectivist uses are the default uses, as I have argued. As mentioned, the recent trend has been to emphasise the subjectivist uses and to give semantics which can explain them (as illustrated by relativists and contextualists alike). Let us call the theories which posit judge dependence (either of content or of truthconditions) “subjectivist semantics”, and the theories that don’t posit it “objectivist semantics”. Subjectivist semantics does not have an explanation for objectivist uses. Most importantly, subjectivist accounts cannot explain genuine disagreements of taste without positing something like normative or practical disagreement. Let me briefly explain why. Traditionally it is assumed that a competent language user has implicit knowledge of the truth-conditions of the sentences of the language. That knowledge manifests itself in speakers’ truth-value intuitions, for example when someone makes a judgment of taste. The assumption is often repeated in the literature on predicates of taste as well. For example, Stojanovic (2007) formulates it as follows: Semantic Competence (SC): Speakers of English are semantically competent with predicates of taste: they master their meaning and truth conditions. (Stojanovic, 2007, 696).

In a typical disagreement about taste a hearer takes the speakers’ utterance that contains a taste predicate to be false. That shows that the taste predicate is not relativised to the speaker since were it the case, the hearer would very rarely if ever be in a position to take the speakers’ utterance to be false. To illustrate, if Peter is in the pub with Matti when he says that Krombacher is better than Becks, it is a perfectly normal reaction from Peter’s part to say (or to think) “That’s false”. If the proposition expressed by Matti was “Krombacher is better than Becks for Matti”, it would not be a normal reaction for Peter. That is enough according to the

58 | Sanna Hirvonen

traditional methodology to strongly support the view that the truth-conditions do not contain the speaker. It should be noted that a moderate relativist cannot both claim that speakers know the truth-conditions and that they genuinely disagree either. Even if the role of the judge in the semantics is formalised as a parameter of the Kaplanian circumstance of evaluation or a Lewisian index rather than an aspect of the content expressed, it is part of the truth-conditions. And hence, as long as we assume speakers to have implicit knowledge of truth-conditions, no subjectivist theory is able to explain the truth-value intuitions speakers have in disagreements.⁴ I have argued that what explains Peter’s truth-value intuition in the above example is his belief that Matti made a false empirical claim about the beers. His utterance would have been true if Krombacher was a better-tasting beer than Becks; there is no relativisation to the speaker or a group.⁵ And Peter thinks Matti got it wrong. But now, does such a theory get into trouble explaining the subjectivist uses? No. Whichever account one has for the objectivist uses, the subjectivist uses can always be explained by free enrichment (if one accepts the existence of free enrichment). In other words, one may hold that the subjective uses are intended as such, and consequently the hearer enriches a bare judgment of taste by mentally adding e.g. for the speaker, resulting in an enriched, explicitly subjective judgment of taste like “For Matti, Krombacher is better than Becks”.⁶ But there is no way that one could use the pragmatic story to the other direction, i.e. to take the subjectivist uses as the default ones and explain the objectivist uses pragmatically. Since the subjectivist uses are the “sophisticated” ones, we may assume that the speaker really does have an intention to subjectivise her judgment. In contrast, if objectivist uses are the default uses, it doesn’t really make

4 The situation is a bit more complex in a radical relativist picture, but they face a closely related problem. Suppose that speaker-hearers know that assessment-sensitivity is true of predicates of taste, as MacFarlane assumes (see e.g. MacFarlane, 2007). Then they also know that speakers make judgments that are true as assessed by themselves, the speakers. Thus, even if a judgment is false as assessed by a certain hearer, she has no reason to insist on what’s true to themselves rather than to the speaker. It’s thus mysterious what goes on in genuine disagreements of taste. MacFarlane (2007) suggests that the disagreement is not about truth, but an attempt to coordinate beliefs. For criticism, see Hirvonen (2014). 5 Predicates of taste are gradable adjectives so they do refer implicitly to something like a comparison class, and in the case of adjectives like good, one needs to select a dimension of goodness (gustatory, health-related, moral etc.) But neither of those amount to anything like relativisation to judges. 6 This is pretty much the position of Kent Bach (2009; 2011) about epistemic modals and predicates of taste.

Doing without judge dependence | 59

sense to suppose that they come with a specific intention to objectivise a subjectivist judgment.

3.3 The challenge to objectivist metaphysics of taste Suppose we conclude that a simple theory which explains a wealth of data is indeed objectivist semantics as the basic account, on top of which one may add either ambiguity or a pragmatic story to account for the subjectivist uses. For example, Moltmann (2010) and Pearson (2013) have argued for semantics which can explain the objectivist uses. According to them, e.g. tasty means roughly, tasty to people in general.⁷ Where’s the problem? In the metaphysics. There is a consensus among many authors that there simply are no such taste properties that could make true the objectivist judgments. Moreover, that’s not an esoteric fact but something that is expected to be known by everyone. The recent theories for predicates of taste are thus driven by the presuppositions that objectivist metaphysics is false and that everyone knows that it’s false. Moreover, since there clearly are subjectivist uses of predicates of taste, the attention has been on them since they pose no metaphysical problems. For example, Lasersohn (2005) states: The concern is that with sentences like these [i.e. judgments of personal taste], there really is no fact of the matter, so it is not clear what to say about whether sentences like these are true or false, or what their truth conditions are.” (Lasersohn, 2005, 644, emphasis mine)

But is it obvious that there is no fact of the matter? Suppose that Pearson’s analysis is correct. Then a judgment of personal taste, e.g. “Anchovies are delicious”, is true iff anchovies are delicious to people in general. There is a fact about that. It’s important to keep in mind that objectivism doesn’t require that the value properties are independent of people altogether. The metaphysical worry is accompanied by an epistemological one: if objectivism turned out to be true, i.e. that there are facts about deliciousness, funniness or tastiness, then those facts are extremely hard or impossible to know, as illustrated by the persistence of disagreements. MacFarlane states the epistemological worry as follows: Although I have known some objectivists about “tasty”, most people seem to recoil from the view. They do not think that there is a “fact of the matter” about whether a thing is tasty in the way that there is a fact of the matter about whether it is red or deciduous or acidic.

7 See van Wijnbergen-Huitink (2016, 6-7) [this volume] for a summary of their views.

60 | Sanna Hirvonen

What underlies this intuition, I suggest, is a realization that if “tasty”, like “poisonous”, expresses an objective property of things, then our ordinary methods for deciding which things to call “tasty” are radically defective. What methods are these? To a pretty good first approximation, we call a food “tasty” when we find its taste pleasing, and “not tasty” when we do not. (MacFarlane, 2014, 3)

Let me try to spell out MacFarlane’s argument a bit. If we suppose that objectivism was true, then we would have to conclude that our own tastes are a very fallible method to find out the objective taste properties of things. Why? Presumably because we disagree about taste so much, and if objectivism is true, in every disagreement of taste at most one can be correct. Perhaps the idea is that if we believed in objectivism, we would be much more careful in making our judgments because we know the large possibility of error. But we do make judgments of taste all the time on the basis of our own tastes, and hence we don’t believe in objectivism. First, if that’s the argument, I think MacFarlane intellectualises people too much. Being often wrong would only deter the careful types from making assertions. And, there are normally no consequences of making a judgment of taste that turns out to be considered false so there’s no reason to be cautious. Secondly, there is no conflict between the idea that we make judgments on the basis of our own taste and that objectivism is true; all we need is the assumption that normally our senses correctly track the value properties of things. Since in a genuine disagreement of taste it’s often very hard to conclude who if any is right, people won’t often be shown wrong. Also, the objectivist will emphasise that we need to be careful in distinguishing between a justified and a true assertion. Since almost the only method we have to make judgments of taste is our own taste, we are justified in making them (unless we know our taste is idiosyncratic). In that sense even objectivist can admit that a judgment of taste is “faultless”. But it doesn’t follow that old any judgment of taste is true; that depends on whether the agent indeed tracks the value properties. Thus, the objectivist will hold that one may be justified in taking something to be fun or delicious given one’s limited previous experiences, but whether something is fun or delicious is not just a matter of one’s current likes and dislikes. Experience can teach us, and just as we come to change many of our beliefs regarding the “objective matters of fact” (e.g. that the funny furniture in the living room is called a divan, or that whales aren’t fish), we come to change our beliefs about taste. Typically we look back at our earlier selves and feel relief that we have learnt so much about all the delicious foods there are, the different ways to have fun or the different ways people can be good-looking. In contrast, it is a challenge

Doing without judge dependence |


for the subjectivist to explain how learning is possible. If one is always right when making a taste judgment, only change can take place, not development. Perhaps the most crucial difference between objectivism and subjectivism is hence epistemological. The subjectivist insists that we only know how things taste to us, and if we took our judgments of taste as objectivist judgments, we would have to acknowledge that we’re very often wrong. The objectivist in contrast advocates caution and emphasises the possibility of mistakes in taste judgments. Moreover, the objectivist is necessarily committed to its being somewhat difficult for us to find out facts about taste which is why there is all the disagreement. A comparison with objectivism about morality is useful: few people give up moral realism and endorse moral relativism even though it’s notoriously difficult to come to an agreement about what the right moral account is. To conclude, the opponents of objectivism and objectivist semantics would have to do two things: First, to show that objectivism really is absolutely implausible, and second, that it’s implausibility is so evident that we cannot assume that people actually use predicates of taste to make objectivist judgments. But whether objectivism is true or not, what really matters for our purposes is that its not easy to see that it is false. If it isn’t, then I think there are strong grounds to argue that objectivism is the folk view that underlies many or most judgments of taste.

4 An error theory of objectivist taste discourse 4.1 Objectivist vs. subjectivist semantics If the subjectivist uses are in fact the marginal ones, used systematically only by a minority of speakers, the semantic account should be primarily concerned with accounting for the objectivist uses. What we will consider next is the relation between the metaphysics and the semantics. Suppose that objectivist metaphysics is false and hence—as the subjectivist intuition states—there really is no fact of the matter that could make objectivist judgments of taste true. Nevertheless people make their judgments believing the objectivist theory to be true. Which semantic theory would be correct in that situation; one that does justice to people’s beliefs, or one that does justice to how the reality is? If the former, then people’s positive judgments of taste would turn out to be false since they attribute objective value properties that don’t exist. The possibility isn’t much considered in the literature, and for example Lasersohn (2005) dismisses it as follows:

62 | Sanna Hirvonen

I assume that our primary goal is to give an analysis of sentences containing predicates of personal taste which assigns them a coherent semantics. [footnote 4: Therefore I do not consider “error” theories, which assign such sentences incoherent or automatically false readings.] (Lasersohn, 2005, 650)

However, I will argue that if the majority of the speakers use predicates of taste in the objectivist way, then there’s a very straightforward argument to show that the semantics must be objectivist as well. As was mentioned, the recent theorists of the semantics of taste predicates are uniformly committed to the principle of semantic competence which holds that speakers have implicit knowledge of the truth-conditions of their judgments.⁸ If, as we are supposing, a majority of the speakers make objectivist judgments but we give predicates of taste subjectivist semantics for metaphysical reasons, then the majority of the speakers would not master their truth-conditions. Rather than competently evaluate judgments of taste relative to themselves or some other relevant individual, they would believe that their truth depends on the objective taste properties which actually don’t exist. That’s a possible view and we will consider it briefly below, but let me emphasise that it’s not consistent with endorsing the principle of semantic competence. If one does endorse the principle of semantic competence, then the semantics has to match people’s implicit beliefs about the truth-conditions of their judgments. And if those beliefs are objectivist, the semantics will be objectivist. But since the world fails to make the objectivist judgments non-vacuously true, we end up with an error theory about taste discourse.⁹ But should that by itself repudiate objectivist semantics? There is a strong prima facie reason why not: by moving from objectivist semantics of taste to the error theory of taste we have effectively moved from doing semantics to doing metaphysics. But one way to see the job of semantics is that it is concerned with giving truth-conditions, not truth values. Itis highly controversial to argue that our current semantics reflects how the world actually is rather than how we take it to be. Let us look at the possible objections to the error theory in more detail. Some might resist an error theory because they believe in an inherent connection between semantics and metaphysics, and that language guides us in

8 See Zeman (2016) [this volume] for a discussion of giving up the principle of semantic competence and the consequent semantic blindness. 9 There are many details about which sentences end up being false, depending on one’s view about predicates with empty extensions. The standard view holds that if nothing is P then sentences of the form (∀x)(Px → Qx) (every P is Q) are vacuously true, as well as their negations (∀x)(Px → ¬Qx) (no P is Q).

Doing without judge dependence | 63

our metaphysics. It’s not a perfectly reliable guide as shown by names for nonexistents and empty predicates. But when it comes to a non-theoretical predicate such as good, there has to be some property which makes those predications true. However, we should be aware of the dangers of tying semantics and metaphysics too closely together. We clearly don’t want to rule out empirical errors that speakers make since otherwise we would be committed to statements about e.g. astrology and different religions being true (and hence also to true contradictions, e.g. “There is just one God and there are many Gods”). But we happily accept an error theory about astrology talk. So is there some principled difference between astrology statements and objectivist statements about taste that would make us accept an error theory of the former but not of the latter? The claim that the error theorists makes is that many people are committed to a flawed empirical theory, objectivism about taste, and hence in both the astrology and objectivist taste cases the mistake is empirical, not conceptual or linguistic. The speakers are assuming things to be a certain way but they are not, and that discrepancy between their beliefs and the truth shows itself in e.g. the trusted horoscope giving false predictions, or it turning out that despite of massive training in wine tasting people’s value judgments of them still differ. The person inclining towards taste subjectivism might again insist on a difference between the epistemological status of those statements. False beliefs about religions or astrology are due to the person having accepted a false theory, but taste properties are directly perceived or experienced and hence we cannot be mistaken about them. But that’s clearly not true about religious experience. The person who thinks he feels the presence of a demon has an equally direct experience as the person who tastes jellied eel and thinks it’s objectively disgusting. Both are wrong in what they think the experience is about (whereas talk about mistakes regarding the experience itself perhaps doesn’t make sense). The error theorist can thus agree that there is nothing wrong with people’s perceptions of various objects of taste. But where they do go wrong is in conceptualising those experiences incorrectly (e.g. as an experience of something objectively disgusting) which results in being liable to draw false inferences about it. For example, the believer in demons may infer that since there’s a demon, Satan exists, and the objectivist about disgusting may infer that jellied eel is disgusting to any person with a good taste. The mistake of the taste objectivist is probably a less serious mistake of the two cases above. But other than that there seems to be no principled reason to say that the person who thinks he perceives a demon is making an empirical mistake whereas the person who thinks he perceives objective disgustingness is not. As long as we admit that a whole area of discourse may be built on mistaken empiri-

64 | Sanna Hirvonen

cal claims we have no reason to rule out that being the case regarding objectivist taste discourses. Another argument against error theories can be traced to interpretationist views on the theory of meaning, and more specifically to the emphasis on the principle of charity (see e.g. Davidson, 2001). The principle holds that when interpreting an agent, one ought to maximise the amount of true beliefs she has (that is, maximise beliefs that are true by the interpreter’s lights). Clearly the principle isn’t supposed to apply to whatever, and for example there is no way that astrological statements could be made to come out true while also retaining the normal semantic values of the expressions in them in other contexts. But in the case of taste we can do better since as we’ve seen, the taste objectivist isn’t committing very serious errors. So given our enlightened subjectivist theoretical point of view, we’ll interpret the objectivist taste utterances as subjectivist utterances. Hence we can end up with a contextualist or moderate relativist theory according to which the truth-conditions do refer to the speaker after all. But, the strategy has its limits. We’ll still be left with false beliefs and statements: those that a hearer such as Peter has regarding what Matti said, and worse, those that speakers like Matti have regarding their own utterances. Peter thought that Matti was making an objectivist statement, and so did Matti. So if we insist on being charitable, we get the problem known from epistemic contextualism of our semantics being committed to attributing speakers with “semantic blindness” (Schiffer, 1996; Hawthorne, 2004). In other words, charity recommends us to relativise taste predicates to speakers since that makes speakers’ utterances true. But the speakers don’t know that that is what they said because they believe they made a speaker-independent, objectivist judgment. Since radical interpretation and charity applies to both beliefs and language, speakers who use taste predicates objectively are also mistaken about their beliefs about their own beliefs. For example, Matti actually only believes that Krombacher is better than Becks to him, not that it’s better for everyone (which is what he believes he believes). That manifests itself in mistaken inferences, e.g. when Matti thinks that if Krombacher is better than Becks, then any person with a good taste prefers Krombacher. (Here his thought content in fact refers to himself, and it doesn’t follow that if Krombacher is better than Becks to Matti, any person with a good taste prefers Krombacher.) And his false beliefs about other people’s utterances manifest in his behaviour, e.g. when Matti disagrees with a person who states that Becks is better than Krombacher. To summarise, the friend of the principle of charity should be aware of the radical consequences the principle has. First, it is possibly committed to speakers having false second-order beliefs (beliefs about their own beliefs). Second, it commits speakers to making false inferences and to misinterpret others’ utterances.

Doing without judge dependence | 65

Third, it denies that speakers have implicit knowledge of truth-conditions which is why they can make mistaken truth-value evaluations. To conclude this section, we’ve seen that appeals to the connection between language and metaphysics is no reason to resist an error theory of taste, and reliance on the principle of charity has radical consequences both in terms of the kind of error it attributes (not empirical but linguistic), and because it rejects the methodology according to which the truth-value intuitions of the speakers should be respected. Therefore, even if objectivism about taste is false, if we want to correctly account for what people believe (including beliefs about what they say), we should go for the semantics that corresponds to their views on taste. In contrast, subjectivist semantics severs the link between what people say and what they believe they say and is hence committed to denying that speakers have knowledge of truth-conditions of judgments of taste. Let me next argue that from the point of view of successful communication there is not much difference whether objectivism about taste turns out to be true or false, and hence an error theory is not problematic in that respect either.

4.2 False but nearly true utterances Let us first consider the well-known and commonplace cases of utterances with questionable truth-values. There are no zombies, yet I can successfully convey something quite specific by remarking to a friend who has had a rough night that he looks like a zombie. Did I say something true? There are no zombies so literally speaking it’s not possible to look like a zombie. Same question can be posed about our talk of Pegasus, Donald Duck, demons, telepathy and so on. So the first point to consider is whether making a false or truth-valueless claim has consequences for communication. My friend understood what I meant, and all of us would understand judgments made about the above-mentioned non-existents. When people didn’t yet know that witches don’t exist, they still had perfectly interpretable discourses about witches. Most of their utterances about witches were of course false (e.g. “The witch who lives next door has left a dead marmot on my doorstep”, “I must have been cursed by a witch” etc.), but they managed to communicate exactly what they wanted: beliefs about witches, those dangerous beings with supernatural powers. What is crucial to see is that for the purposes of communication what matters is primarily that the content is understood, not that it is true (even if true judgments are way better from the point of view of sharing information). If you and I are believers in witches and you tell me you have been cursed, I become alarmed

66 | Sanna Hirvonen

and maybe recommend you to contact another witch who will perform a countercurse. No communication breakdown takes place because we share our strange set of false beliefs. And that is the situation regarding objectivism about taste (assuming that it’s false): taste objectivists are embarked on their common search for the unattainable goal, the objective truth about taste. Communication only fails if the speakers disagree about what exists, which is why a discussion about taste between an objectivist and subjectivist is bound to turn philosophical. However, in the case of taste what makes the situation happier than that of the witch-believers is that the utterances of taste objectivists are not far from the truth.¹⁰ First, people are after all similar in their tastes so often an objectivist taste statement will be true of a large number of people, even if it fails as an objectivist statement that is about everyone (or about people generally). Second, people like to hang out with others with very similar tastes. In such a group the objectivist statement will at the best case hold of everyone so it’s almost as good as true. Likewise, if two taste peers disagree, they may very well come to an agreement because their arguments really make sense given their shared background tastes (such arguments would of course leave cold a person with a very different taste). Objectivism is thus close enough to being true without it causing any real disruptions in our ordinary lives. And arguably, a very large number of our ordinary statements are not quite true anyway. For example, is France hexagonal and Italy boot-shaped? Was it really 3pm or four past three when you answered the question about what time it was? Do you really know that you’ll stay home tonight (after all, your home might have burnt down)? Exaggeration and sloppiness are totally commonplace but both contribute to falsehoods: our utterances are very often false but nearly true. Does it matter whether that is due to a slight empirical error, exaggeration or sloppiness? I can’t see why it would as long as the falsehood is close enough to truth.

10 Gilbert Harman’s views about morality are close to the view suggested here about taste (Harman & Thomson, 1996; Harman, 2000). He thinks our moral statements are false since they presuppose the truth of moral objectivism, and hence we should revise our moral talk and accept that at best we have moral statements that are true relative to communities. Revision might be easier to achieve in the taste case due to the various existing ways of relativising one’s taste statements (e.g. by saying “to me this is fun”).

Doing without judge dependence |


5 Conclusions What I’ve argued in this paper is that most speakers are committed to objectivism about taste and that is the best explanation of “genuine” disagreements about taste and other ways in which we talk about taste. Subjectivist uses exist as well, but when the speaker is making a subjectivist statement, she will not engage in disagreements or serious conversations about the topic. Hence, no use of taste predicates is at once both subjective and such that it can be disagreed with (except in the rare cases where the speaker is mistaken about her own taste). I then addressed the subjectivist challenge to objectivism about taste, arguing that their case is rather weak against objectivism, and especially against the idea that objectivism is the folk view (even if it is actually false). Finally, I argued that even if objectivism about taste turns out to be false there are strong reasons in favour of objectivist semantics for taste predicates. If one is convinced that semantics should be done based on what people take the truth-conditions to be, then an error theory seems inevitable. Otherwise semantics ends up severing the link between linguistic meaning and beliefs, and has to posit semantic blindness. Finally, I have argued that an error theory of taste has hardly any consequences for successful communication, and no consequences for communication between people who believe in objectivism. And given that the error in objectivism about taste is rather harmless, the objectivist judgments about taste are close enough to being true so that no harm is done by such talk. Acknowledgment: I want to thank two anonymous reviewers, Julien Dutant, Michael G.F. Martin and the audience at the Subjective Meaning Workshop for helpful comments and discussions.

References Bach, Kent. 2009. Relatively speaking. Bach, Kent. 2011. Perspectives on possibilities: Contextualism, relativism, or what? In Andy Egan & Brian Weatherson (eds.), Epistemic modality, 19–59. Oxford: Oxford University Press. Davidson, Donald. 2001. Subjective, intersubjective, objective. Oxford: Clarendon Press. Glanzberg, Michael. 2007. Context, content and relativism. Philosophical Studies 136. 1–29. Harman, Gilbert. 2000. Explaining value and other essays in moral philosophy. Oxford: Oxford University Press.

68 | Sanna Hirvonen

Harman, Gilbert & Judith Jarvis Thomson. 1996. Moral relativism and moral objectivity. Oxford: Blackwell. Hawthorne, John. 2004. Knowledge and lotteries. Oxford: Oxford University Press. Hirvonen, Sanna. 2014. Predicates of taste and perspective dependence. London: University College London dissertation. Kölbel, Max. 2003. Faultless disagreement. Proceedings of the Aristotelian Society 104. 53–73. Lasersohn, Peter. 2005. Context dependence, disagreement, and predicates of personal taste. Linguistics and Philosophy 28(6). 643–686. MacFarlane, John. 2007. Relativism and disagreement. Philosophical Studies 132. 17–31. MacFarlane, John. 2009. Nonindexical contextualism. Synthese 166. 231–350. MacFarlane, John. 2014. Assessment sensitivity: Relative truth and its applications. Oxford: Oxford University Press. Moltmann, Friederike. 2010. Relative truth and the first person. Philosophical Studies 150. 187–220. Pearson, Hazel. 2013. A judge-free semantics for predicates of personal taste. Journal of Semantics 30. 103–154. Schiffer, Stephen. 1996. Contextualist solutions to scepticism. Proceedings of the Aristotelian Society 96. 317–333. Stern, Jane & Michael Stern. 1990. The encyclopedia of bad taste. New York: HarperCollins Publishers Ltd. Stojanovic, Isidora. 2007. Talking about taste: Disagreement, implicit arguments and relative truth. Linguistics and Philosophy 30(6). 691–706. Sundell, Timothy. 2011. Disagreements about taste. Philosophical Studies 155. 267–288. van Wijnbergen-Huitink, Janneke. 2016. Subjective meaning: An introduction. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 1–19. Berlin: De Gruyter. [This volume]. Zeman, Dan. 2016. Contextualism and disagreement about taste. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 91–104. Berlin: De Gruyter. [This volume].

Lavi Wolf

Predicates of personal taste and the evidential step Abstract: Predicates of personal taste give rise to faultless disagreement, meaning that they encode a duality that allows them to invoke objective disagreement and to subjectively refrain from assigning blame. The claim of this paper is that this duality interacts with the evidential step, a conversational move from an assertion that functions as a source of evidence pertaining to a proposition, to the context update of this proposition. Assertions with predicates of personal taste fail to make the evidential step due to their irreducible perspective dependency, and are consequently reinterpreted and updated as an objectivized mixture model. The paper presents a conversational theory that is able to account for faultless disagreement as well as for new and old problems and puzzles previously unaccounted for.

1 Introduction The intuitive notion of predicates of personal taste (henceforth PPTs) such as tasty is that they convey the speaker’s point of view, and therefore should be differentiated from objective predicates e.g. made in Israel. But if that were the case, i.e. if a person who says that something is tasty is really saying that she¹ finds it tasty, then what sense can we make of the following exchange? (1)

Suzy: This cake is tasty. James: No, it’s not!

Such cases of faultless disagreement (cf. Kölbel, 2004; Lasersohn, 2005) are so called because of two problems they raise. The first is that while Suzy and James are clearly disagreeing, this cannot be accounted for by the intuitive notion because disagreement requires semantic contradiction and there isn’t any contradiction between Suzy finding the cake tasty and James not finding it so. In other

1 Masculine and feminine pronouns will alternate randomly. Lavi Wolf, The Language, Logic and Cognition Center (LLCC), The Hebrew University of Jerusalem, Mount Scopus, 91905 Jerusalem, Israel, e-mail: [email protected]

70 | Lavi Wolf

words, the intuitive notion leaves James nothing to disagree about. The second problem is that while Suzy and James are indeed having a dispute, this is not an ordinary type of disagreement about “facts in the world”. For example, if Suzy and James disagree about whether or not the cake is made in Israel, one of them speaks the truth and the other utters a falsehood. The person who is wrong is at fault and should mend his beliefs accordingly. However, in the type of dispute depicted above, both Suzy and James are fully entitled to their claim and none can (or should) be blamed for uttering a falsehood. This is the problem of faultlessness, which an adequate theory of PPTs should account for as well. Cases of faultless disagreement are a motivation for Lasersohn’s (2005) relativist account, as well as the contextualism-relativism debate that arose with regards to PPTs following Lasersohn’s paper. They are very nicely discussed in the introduction chapter of this volume (van Wijnbergen-Huitink, 2016). I will, therefore, assume familiarity with these approaches and proceed ahead. For the purposes of this paper, contextualism and relativism are both aspects of the same view,² the subjective evaluation view, according to which PPT are perspectivedependent on a judge and the default cases of faultless disagreement stem from this role being assigned to the speaker, i.e. default cases of PPT are evaluated from a subjective perspective. Whether this judge is represented by some covert indexical which is part of the propositional content or as an index of evaluation will not concern this paper. I therefore focus on Lasersohn (2005) as a prime example of the subjective evaluation approach, in Section 2. The rest of Section 2 discusses Récanati (2007) as an example of an objective evaluation approach³ and Cohen’s (2010) objectivized evaluation. Section 3 presents the conversational theory of Wolf (2015), the epistemic step and the evidential step which is the source of the faultless disagreement phenomenon for PPT. Section 4 answers the questions raised in previous sections and Section 5 concludes the paper.

2 Indeed, as claimed in Stojanovic (2007), both approaches are semantically equivalent, i.e. it doesn’t matter whether an utterance containing a PPT is evaluated for truth by a contextualist or a relativist semantic interpretation—if they are applied to the same indexical parameters at the same point of view, the truth conditions will be the same. 3 The term objective is used here in the intersubjective sense, i.e. an evaluation which is shared by all individuals under consideration.

Predicates of personal taste and the evidential step |


2 Background 2.1 Subjective evaluation The important advantage of Lasersohn (2005) is that it solves the faultless disagreement puzzle, i.e. is able to explain why disputes concerning PPTs are possible while at the same time no dispute party is at fault. Lasersohn’s seminal paper presents a theory that extends Kaplan’s (1989) framework. Kaplan distinguishes between the content and the character of utterances, when the latter is a function from contexts to contents and the former is a proposition, i.e. a function from world/time indices ⟨w, t⟩ to truth values. Lasersohn adds a third index, the judge, representing the individual whose taste determines the truth value for utterances containing PPTs: (2)

⟦the cake is tasty⟧w,t,j = 1 iff the cake is tasty at world w, time t, according to judge j.

The introduction of a judge index solves the faultlessness problem—each conversational participant’s claim is true, albeit with regards to a different judge, therefore no one is at fault. The disagreement is accounted for by contradictory contents—there is no triple index of evaluation (world, time, judge) in which both contents are true, i.e. the content of the utterance “the cake is not tasty” indeed contradicts the content of the utterance “the cake is tasty”. This is an advantage over contextualist accounts which cannot explain disagreement (cf. van Wijnbergen-Huitink, 2016, this volume). However, this theory has several problems, which are discussed in the following sections.

2.1.1 The pragmatic problem The default use of PPT, according to Lasersohn’s account, involves an “autocentric stance”, i.e. the speaker is the judge whose taste determines the truth value. This raises a pragmatic problem. Being a default case means that unless contextually specified otherwise, every conversational participant naturally expects the judge in a PPT utterance to be the speaker (Stojanovic, 2007). Hence, conversational participants should pragmatically assign the speaker the judge role by default. In our case, James should naturally take Suzy to be expressing her own taste, and vice versa. But if both conversational participants indeed recognize that they each are expressing an individual taste, why would they wish to argue? To be more specific, even though disagreement can be explained semantically in this system, it

72 | Lavi Wolf

can’t be motivated pragmatically. Note that when the judge is made explicit there is no actual dispute, as apparent by the infelicity of the following (cf. also Crespo & Fernández, 2011; Gunlogson & Carlson, 2016; Umbach, 2016): (3)

Suzy: This cake is tasty for me. James: #No it’s not, this cake is disgusting for me!

The autocentric stance is not a viable notion pragmatically, then. But perhaps a different stance is at play? Maybe the disagreement stems from both conversational participants taking each other’s utterance to mean something other than a subjective point of view? Lasersohn suggests a different stance which can be used, an exocentric stance, in which the judge is someone other than the speaker. This type of stance is apparent in (but not exclusive to) utterances such as: (4)


This cat food is tasty, because my cat can’t get enough of it.

The judge in (4) is not Suzy, but the cat. We may even, quite reasonably, assume that Suzy has never tasted the cat food herself. What, then, about disputes involving exocentric stance? (5)

Suzy: This cat food is tasty. James: No, it’s not!

An exocentric stance is not the default case, thus an argument may arise. However, in this case the argument boils down to a misunderstanding. Assuming Suzy uses an exocentric stance as in the previous example, James can either use an autocentric stance or an exocentric one in which either the aforementioned cat or some other individual is the judge. If James employs an autocentric stance, then the “dispute” is that Suzy claims that the cat finds the cat food tasty and James claims that James doesn’t find it tasty, which is not really a dispute but rather a misunderstanding. Once the different stance is resolved (for instance, once Suzy realizes that James has actually eaten the cat food and found it disgusting) Suzy and James will understand that there’s no real argument going on, since it’s not feasible to compare cat taste standards with human. As recalled, explaining both disagreement and faultlessness is a main desideratum of PPT theories, thus losing disagreement is a high price to pay. If James employs an exocentric stance in which the judge is some individual other than Suzy’s cat (for instance, if James uses his own cat as a judge) we will be left with a similar pragmatic problem, since making this judge explicit again eliminates the dispute:

Predicates of personal taste and the evidential step | 73


Suzy: This cat food is tasty for my cat. James: #No it’s not, this cat food is disgusting for mine!

Finally, if James employs an exocentric stance in which the judge is the same as Suzy’s (i.e. Suzy’s cat), according to Lasersohn’s theory there is a real dispute but it is not faultless. In this case, James’ claim contradicts Suzy’s claim since there is no world, time, judge index in which both claims are true. However, both of them can’t be right at the same time—the cat either finds the cat food tasty, or it doesn’t. It is only a matter of finding out whose claim is correct, and in that case one of the disputing parties will be wrong, i.e. at fault. And again, explaining both disagreement and faultlessness is a main desideratum of PPT theories, thus losing faultlessness is a high price to pay. To conclude, it doesn’t matter which judge the context assigns—the theory either predicts no disagreement or no faultlessness.

2.1.2 The semantic problem Disagreement in Lasersohn’s theory is accounted for in the usual semantic sense, i.e. at the level of content—both conversational participants assert contradictory contents. And, since Lasersohn adds a judge to Kaplan’s original indices, contradiction occurs when there is no world, time and judge index in which both contents are true. This raises a semantic problem, since it is perfectly possible to assert contradictory contents in a Kaplanian framework in such a way that doesn’t constitute any disagreement. The following example (based on Récanati, 2007, slightly modified) serves to show this point: (7)

Suzy (on Sunday morning): James (on Monday evening):

It is raining. It is not raining.

Suzy asserts a content which is true at a certain world-time index ⟨w, t⟩.⁴ James asserts a content which is true at a different time, i.e. the world-time index is ⟨w, t󸀠 ⟩. The contents are contradictory since there is no index in which both contents are true. Yet, the characters are different since Suzy’s assertion is evaluated for truth at a different time index than James’, (similarly to our original example (1), in which Suzy’s assertion is evaluated for truth by a different judge). Thus any dispute they might have concerning the state of rain (for instance, if Suzy’s utterance was left as a voice message on James’ cell phone and James mistakes this message

4 Since these utterances do not contain PPTs, there is no need for a judge index.

74 | Lavi Wolf

to co-occur with his own temporal location) will be due to a misunderstanding and not constitute a genuine disagreement. Since both (7) and (1) are cases in which two utterances semantically contradict one another, and since (7) does not constitute a real disagreement, there is no support to Lasersohn’s claim that disagreement is to be represented by contradictory contents in a Kaplanian framework.

2.1.3 The Frege-Geach problem Lasersohn (2005, 656) argues that PPTs may occur embedded under truth-conditional operators e.g. conditionals, and thus participate in logical deduction such as the following modus ponens: (8)

If there is a loop, the roller coaster is fun. There is a loop. Therefore, the roller coaster is fun.

This argument is based on the Frege-Geach problem (cf. Geach, 1965), raised against emotivist metaethical theories with regards to the claim that moral predicates, e.g. right and wrong, express the speaker’s positive or negative emotional attitude towards the prejacent. Lasersohn uses the same argument against expressivism and claims that the modus ponens serves to show that PPTs are truth conditional, hence not expressive, and accountable by his theory. Lasersohn’s argument is disputed by Gutzmann (2016) [this volume], who shows that there are cases in which non-assertoric speech acts participate in logical deduction: (9)


Modus ponens with non-assertoric speech acts: imperatives If the roller coaster has a loop, go for it. The roller coaster has a loop. Therefore, go for it. Modus ponens with non-assertoric speech acts: expressive speech acts If the roller coaster had a loop, congratulations for being brave! The roller coaster had a loop. Therefore, congratulations for being brave!

But there is another flaw in Lasersohn’s argument. Note that in both examples the PPT and the non-assertoric speech act is used in the apodosis of the conditional. Matters are different for both PPTs and non-assertoric speech acts when used in the protasis. Non-assertoric speech acts are completely out:

Predicates of personal taste and the evidential step | 75


#If go for it / congratulations then you earned my respect.

PPTs are not completely out, which may indicate that they are truth conditional after all. But the following lacks the logical certainty of Lasersohn’s example: (12)

Suzy: If the roller coaster is fun, I will buy a ticket. James: The roller coaster is fun. Suzy: Therefore, I will buy a ticket.

In this instance, unlike the former, it seems that the premises do not necessitate the conclusion, which means it lacks the deductive strength of the former. In order to see the difference in a clearer manner, imagine a conversational scenario in which Suzy is talking to Jeff, a person whose opinion Suzy values, and refers the first assertion to him—“if the roller coaster is fun, I will buy a ticket”. Enters James, whose opinion Suzy doesn’t value in any way, and asserts that the roller coaster is fun. Will Suzy conclude from this that she should buy a ticket? Not necessarily. Now, imagine the same scenario with the same conversational participants, but now Suzy’s first assertion is “if there is a loop, the roller coaster is fun”, and James’ assertion is that there is a loop. Does Suzy’s opinion of James affect Suzy’s conclusion? If PPTs are indeed truth conditional and explained in terms of Lasersohn’s theory, then it shouldn’t matter whether they appear in the protasis or apodosis of conditionals. In light of the problems presented in the previous sections perhaps what we need is a theory which is more objective. Such a theory, by Récanati (2007), is discussed in the following section.

2.2 Objective evaluation Récanati (2007) accepts the basic ingredients of Lasersohn’s theory but argues that the default judge of PPTs cannot be the speaker. Instead, utterances such as (1), repeated below, should mean as This cake is tasty for us, where us is “the community to which the speaker and his audience belong” (Récanati, 2007, 91): (13)

Suzy: This cake is tasty. James: No, it’s not!

Recanati’s theory is able to account for the pragmatic, semantic and logical problems that Lasersohn’s theory suffers from, as discussed above. The pragmatic problem is accounted for since both discourse participants use the same judge and therefore it is understandable that they are arguing. The semantic problem is accounted for since in this case semantic contradiction and disagreement are re-

76 | Lavi Wolf

lated to each other—both Suzy and James argue about the same thing, i.e. whether the cake is tasty for both of them. The logical problem is accounted for since both the embedded and non-embedded PPT in the deduction pertain to the same judge index. However, this theory has problems of its own. The first is that it fails to account for faultlessness since the judge is the same for both dispute participants. And, as discussed previously, accounting for faultlessness is a main desideratum for PPTs theories, thus losing it is a high price to pay. The second problem arises from Recanati’s distinction between utterances like (1) and utterances that make the judge explicit, such as “the cake is tasty for me”. The latter is, by Recanati, a weaker claim since it’s entailed by the community reading. Because of this, when Suzy’s assertion is challenged by James, she can retreat to the weaker claim, thereby avoiding making a mistake.⁵ (14)

Suzy: This cake is tasty. James: No, it’s not! Suzy: I meant, this cake is tasty for me.

Note that negation has to scope over the implicit for us, which is interpreted as a universal quantifier. Otherwise, James utterance will have the potential reading in which it is true for both James and Suzy that the cake is not tasty. This is undesired since it overrules Suzy’s original assertion. That is, if Suzy’s original assertion means the same as This cake is tasty for us, James can object on grounds of his own taste (i.e. the cake is not tasty for James, therefore it is not true that the cake is tasty for both James and Suzy), but James can’t include Suzy’s taste in his objection. Thus a narrow scope negation is out. The wide scope reading, in which it is not the case that the cake is tasty for both Suzy and James, makes this dialogue felicitous and consistent. Suzy’s second assertion (i.e. the weaker claim that the cake is tasty for her alone) is consistent as well. However, in light of this, (15) should be felicitous since it is a conjunction of “it is not the case that the cake is tasty for us” with the weaker claim “but it is tasty for me”. However, (15) is as infelicitous as (16) (which is predicted to be infelicitous by Récanati’s theory) which is a conjunction of the strong claim “the cake is not tasty for me” (entailing that the cake is not tasty for us) with the claim “the cake is tasty for us”:

5 According to Récanati Suzy’s first assertion is surely a mistake since James is part of us, therefore the reply in (14) automatically makes the original assertion false.

Predicates of personal taste and the evidential step |


#This cake is not tasty, but it is tasty for me.


#This cake is not tasty for me, but it is tasty.


So far we can see that there are problems with both the subjective and the objective evaluation approaches. The following section presents a theory which combines both, Cohen’s (2010) “objectivized approach”.

2.3 Objectivized evaluation In an attempt to come up with a theory of PPTs that does not suffer from the subjective or objective problems,⁶ Cohen (2010) presents an objectivized theory of PPTs based on Wolf & Cohen’s (2011) theory of the predicate clear. Cohen discusses the similarity between clarity and PPTs—both are gradable, can be modified by comparatives and overt experiencers, used as superlatives, and most importantly exhibit faultless disagreement: (17)

Suzy: It is clear that Abby is a doctor. James: No it’s not!

In light of the similarities and the fact that clear and PPTs are both evaluatives, Cohen (2010) offers a treatment of PPTs along the lines of clarity—PPTs are objectivized predicates, whose truth conditions depend on the opinions of various individuals, specifically those that the speaker considers to be good evaluators of taste. The notion of “good evaluators” is formalized by a probabilistic mixture model, which is defined over possible individuals: (18)

P mixture(φ) = ∑ni=1 w i P i (φ)

The idea is that each individual i contributes to the mixture model a personal degree of belief with regards to tasty. This degree is a probability measure representing the subjective probability value that individual i assigns to the proposition. These individuals are then assigned weights w i , indicating their perceived reliability in the eyes of the speaker. If an individual (including the speaker herself) is considered to be non-reliable in matters of taste, her weight will be low and if an individual is considered an expert (such as a known connoisseur) in matters

6 Generic theories of PPTs also do not suffer from the subjective and objective problems. For a discussion and rejection of the generic solution, cf. Cohen (2010).

78 | Lavi Wolf

of taste, her weight will be high. The final probability value is then computed as the weighted sum of the probabilities that were assigned. In the case of tasty: (19)

P mixture((tasty)(cake)) = ∑ni=1 w i P i ((tasty)(cake))

Since PPTs are vague, Cohen uses a delineation function (Lewis, 1970) d(tasty), which returns a standard of taste that varies according to context. In some contexts (say, a meeting of fine-cuisine critics) this standard will have to be very high, while in other contexts (a meeting of friends at a local hotdog stand) this standard can be lower. Thus, This cake is tasty is true iff the value of the mixture model is greater than d(tasty): (20)

∑ni=1 w i P i ((tasty)(cake)) > d(tasty)

2.3.1 Objectivization and diagonalization While the mixture model presented in the previous section is truth conditional, it does not really provide the conditions of truth for PPTs, i.e. it does not tell us whether it is true that, in our example, the cake is tasty. Rather, it provides information with regards to the tastes of individuals which are contextually under consideration. In this sense the mixture model operates less like a horizontal proposition and more like a diagonal one, which is “the proposition that is true at i for any i if and only if what is expressed in the utterance at i is true at i” (Stalnaker, 1978, 81). The mixture model takes into consideration different points of view i, i.e. individuals with different standards of taste and checks for each one whether the interpretation of tasty in the eyes of that individual is true for that individual. Diagonal propositions are different from horizontal ones in many ways. One of them is that the diagonalization process is not activated out of the blue. In order for diagonalization to take effect there has to be some violation of conversational rules that forces utterances to be reinterpreted. Such a situation in which conversational rules are broken and conversational participants are not able to update a standard horizontal proposition arises when conversational participants perform PPT assertions. This conversational situation, I submit, is the cause for the activation of the mixture model and is the source of faultless disagreement in PPT. In order to see the conditions that give rise to the activation of the mixture model we need to look into the inner workings of conversation and the steps that conversational participants take in order to accept and update new information. These steps are discussed in the following sections.

Predicates of personal taste and the evidential step | 79

3 Conversational elements 3.1 The epistemic step Sauerland (2004, 2005) discusses a gap between what the Gricean (1975) maxims are able to account for and the conversational implicatures that are actually conversationally derived. The maxim of quality, for instance, is not enough to derive scalar implicatures and there is a need for an extra step on the hearer’s part in order to get there. Regard the following (Sauerland, 2005): (21)

Maxim of Quantity: Make the most informative statement that you know to be true.


The Philharmonic played many of Beethoven’s symphonies. a. b.

Primary Implicature: The speaker is not sure that the Philharmonic played all of Beethoven’s symphonies. Secondary Implicature: The speaker is sure that the Philharmonic did not play all of Beethoven’s symphonies.

The main utterance in (22) generates both a primary implicature (22a) and a secondary one (22b). The maxim of quantity can account for the primary implicature but in order to derive the secondary implicature there is a need for a further step— the epistemic step. Sauerland accounts for this step via a contextual assumption about the epistemic state of the speaker, i.e. that she either knows/believes that the alternative is true or that she knows/believes that it is false⁷: (23)

K speaker (the Philharmonic played all of Beethoven’s symphonies) ∨ K speaker (the Philharmonic didn’t play all of Beethoven’s symphonies)

The first disjunct is out due to the primary implicature, therefore the hearer infers the secondary implicature. This account of the epistemic step is useful and insightful, but it doesn’t quite take us all the way to where we need to get, i.e. to the implicature itself. Taking it step by step, the first one is to pragmatically infer about what the speaker doesn’t know for certain. The second one (the epistemic step) is to get rid of uncertainties and infer about what the speaker does know for certain. But in order to get from there to the implicature itself, i.e. that the Philharmonic did not play all of Beethoven’s symphonies, we also need to get rid of the speaker’s epistemic state altogether. 7 The K-operator expresses epistemic certainty (Hintikka, 1961).

80 | Lavi Wolf

Another example of this step in action, this time depicting the well-known move from inclusive to exclusive disjunction: (24)

The Philharmonic played Beethoven’s or Mozart’s symphonies.

The truth conditional meaning is the inclusive disjunction, i.e. that the Philharmonic played either Beethoven’s or Mozart’s symphonies, or both. Since the speaker doesn’t utter the conjunction and since the conjunction is semantically stronger, i.e. entails the disjunction, the maxim of quantity leads the hearer to infer that the speaker isn’t certain that the conjunction is true. Applying the epistemic step here leads to the conclusion that the speaker is certain that the conjunction is false. But, once more, the speaker’s epistemic state has nothing to do with the eventual implicature that the hearer updates, which is the exclusive disjunction. In order to derive the exclusive disjunction for (24), and the implicature that the Philharmonic played many but not all of Beethoven’s symphonies for (22), we need one more step which I term the evidential step.⁸

3.2 The evidential step Wolf (2015) discusses the context update process of assertion, which involves a speaker who performs the speech act and a hearer who has to decide whether to accept, reject, further discuss, or agree to disagree about it. In order to make such a decision, the hearer needs to takes into account various sources of evidence at her disposal. These sources may include direct knowledge e.g. perception, deductive processes e.g. inference, or reported information e.g. hearsay.⁹ Once these sources are being considered, the hearer decides whether to take the evidential step and make the transition from the evidence that was presented with regards to the truth of the proposition, to the truth of the proposition. The formal apparatus utilizes the following assertion operator: (25)

A x ⟨S, C⟩

The first argument, S, stands for the degree of strength by which the assertion is performed and the second argument C is the assertion’s propositional content.

8 The examples used here are about scalar implicature but, as will be seen ahead, the evidential step applies to every type of assertion. 9 If this classification of sources of evidence brings grammatical evidentials to the mind of the reader, this is not a coincidence. I believe that evidentials may indeed be accounted for in terms of this view of context update, but will say no more about this here.

Predicates of personal taste and the evidential step | 81

The order of arguments stands for relative scope, i.e. the degree of strength scopes over the propositional content. Thus, a shorthand representation of this assertion operator in probabilistic terms, i.e. the assertion of propositional content φ with a degree of strength that is the value v of a probability function P, is: (26)

A x P(φ) = v

An example for a standard process of assertion: (27)

Suzy: The cake is on the shelf. A Suzy P(on-shelf(cake)) ≥ high

Suzy asserts the propositional content “the cake is on the shelf”, with a default degree of strength for assertion which is equal to or greater than high, which stands for some probabilistic value close to 1 (based on a standard norm of assertion that the speaker highly believes in what she asserts). Once Suzy puts forward the assertion, the hearer has to consider it. This process of consideration is formalized as a mixture model, similar to the one described in the previous sections, composed of sources of evidence pertaining to the proposition. (28)

P(φ) = ∑ni=1 w i P i (φ)

If P(φ) exceeds the hearer’s threshold of acceptance¹⁰, the proposition is accepted and updated into the common ground; if P(¬φ) exceeds this threshold, the proposition is rejected. Otherwise it is left in the Negotiation Zone, a conversational repository of items under negotiation (cf. Wolf, 2015) which will not be discussed here. The first important point regarding this process is that the speaker, Suzy, is but one of many sources of evidence pertaining to the asserted proposition, and that each source is a point of view with a distinct probability space and probability function. The second important point is that if the hearer chooses to accept the assertion, he makes the move from an evidence-dependent or evidence-relative proposition to a truth conditional one, which is not relative to the beliefs of any individual. The next section will show how this distinction plays a role in faultless disagreement.

10 The assumption is that, by default, the threshold of acceptance is the same degree as the default for standard assertion, i.e. high. However, this threshold can change (cf. Davis et al., 2007; Davis, 2009).

82 | Lavi Wolf

4 Solutions to the problems 4.1 Faultless disagreement PPTs require some perceiver to do the evaluation. A cake can’t be tasty if no one tastes it and a roller coaster can’t be fun if no one experiences the ride. As Lasersohn (2005, 643) puts it, they “aren’t about matters of fact, but are really just matters of opinion”. However, under the assumption that every assertion conveys the degree of belief of the speaker alongside the propositional content, PPTs in standard assertions always have a default evaluator who is the speaker. Therefore, I propose the following conversational addition to Cohen (2010): at the time of assertion PPTs behave just like objective predicates and it is only later, at the stage of context update, that they fail to make the evidential step and need to be reinterpreted as the mixture model. In order to see this process at work step by step: (29)

Suzy: This cake is tasty. A Suzy P(tasty(cake)) ≥ high¹¹

Suzy asserts the content “the cake is tasty” with the degree of strength, which is Suzy’s degree of belief, of equal to or greater than high. This degree, recall, is derived from a mixture model of various sources of evidence that Suzy is aware of, which pertain to the taste of the cake. Note that the degree of strength of the assertion is expressed and not asserted as part of the propositional content. At the next stage the hearer, James, evaluates Suzy’s assertion in light of his own internal mixture model with the evidence available to him. Suzy’s assertion participates in this mixture model as an additional source of evidence pertaining to the proposition “the cake is tasty”. Note that at this stage every source of evidence acts as a judge of sorts, an evaluator. If the probability value of James’ mixture model surpasses his threshold of acceptance, James will accept Suzy’s assertion: (30)

Suzy: This cake is tasty. A Suzy P(tasty(cake)) ≥ high James: I agree / That’s right / Yes, it is! A James P(tasty(cake)) ≥ high

The next stage is the evidential step in which the asserted propositional content is updated into the common ground. However, the content is just the bare evaluator-

11 This degree is d(tasty) in Cohen (2010).

Predicates of personal taste and the evidential step |


free proposition that the cake is tasty. This content has no truth-value on its own and needs to be coupled with some evaluator. However, as was argued in the previous sections, this evaluator cannot be fully subjective or objective. Thus, in order to save the context from becoming defective and preserve the coordination of the presuppositions of conversational participants, the updated propositional content is reinterpreted as the mixture model: (31)

P mixture(tasty(cake)) ≥ high

Importantly—this mixture model is truth-conditional (not expressed). It stands for the proposition that is true in every world in the context set, that good evaluators of taste will find the cake tasty. Moving on to cases of faultless disagreement: (32)

Suzy: This cake is tasty. A Suzy P(tasty(cake)) ≥ high James: I disagree / That’s wrong / No, it isn’t! A James P(¬tasty(cake)) ≥ high

Both conversational participants assert contradictory contents with the same degree of belief, which accounts for the disagreement. Of course, both contents can’t be updated into the common ground since such an update will be: (33)

P mixture(tasty(cake)) ≥ high & P mixture(¬tasty(cake)) ≥ high

which is a contradiction. Yet, Suzy and James perform the assertion from a subjective point of view, i.e the degree of belief defined over the probability space of the speaker. Both Suzy and James are entitled to draw their own conclusions based on the evidence available to them and perform an assertion in light of these conclusions. If the evidence available to each of them differs or the weights they assign to the sources of evidence differ, it is none of their fault. This accounts for the faultlessness.

4.2 The Frege-Geach problem Repeating (12): (34)

Suzy: If the roller coaster is fun, I will buy a ticket. James: The roller coaster is fun. Suzy: Therefore, I will buy a ticket.

84 | Lavi Wolf

A representation of Suzy’s first assertion in (34) is: (35)

Suzy: If the roller coaster is fun, I will buy a ticket. A Suzy P(P(buy(Suzy, ticket)|fun(roller-coaster)) = 1) ≥ high

And a representation of James’ is: (36)

James: The roller coaster is fun. A James P(fun(roller-coaster)) ≥ high

Now, in order for a modus ponens argument to be valid the following features have to be maintained (cf. Schroeder, 2009): A. If the premises are true, then the conclusion is true as well. B. It is inconsistent to accept each of the premises and deny their conclusion. C. Accepting the premises commits someone, in some sense, to accepting their conclusion. The first feature is maintained, since if it is true that the probability of Suzy buying a ticket given that the roller coaster is fun is 1, and it is true that the roller coaster is fun, then it is true that Suzy will buy a ticket. Things get a bit more complicated in the next two features, since under the conversational picture that we’re describing, there is a difference between asserting a proposition and accepting it. We may assume that if Suzy’s assertion is accepted than the updated content will be the conditional probability, since this operator makes the PPT relative to a probability space. But we can’t assume an update of the bare proposition “the roller coaster is fun” for the reasons described above. We therefore need to update James’ assertion as a mixture model. But now, both assertions can be accepted at the same time without inconsistency, because the embedded PPT is not evaluated by the same probability space as the nonembedded PPT. For the same reason, accepting the premises doesn’t necessarily commit someone to accepting their conclusion. Compare with the case of PPTs in the apodosis, repeating (8): (37)

Suzy: If the roller coaster has a loop, the roller coaster is fun. A Suzy P(P(fun(roller-coaster)|loop(roller-coaster)) = 1) ≥ high James: The roller coaster has a loop. A James P(loop(roller-coaster)) ≥ high Therefore, the roller coaster is fun.

An update of the proposition “the roller coaster has a loop” does not require reinterpretation as the mixture model. Thus, it is inconsistent to accept each of the

Predicates of personal taste and the evidential step |


premises and deny their conclusion, and accepting the premises commits someone to accepting their conclusion.

4.3 Récanati (2007) The problem in Récanati (2007), recall, is that the theory predicts (38) to be felicitous: (38) #This cake is not tasty, but it is tasty for me. (39) #This cake is not tasty for me, but it is tasty. In the theory proposed here, when an overt experiencer is mentioned, there is no need for mixture model to be involved since the probability value is fixed to that of the experiencer.¹² Therefore, at the level of the propositional content, (38) is represented as (40) and (39) as (41): (40)

P mixture(¬tasty(cake)) ≥ high & P speaker(tasty(cake)) ≥ high


P speaker(¬tasty(cake)) ≥ high & P mixture(tasty(cake)) ≥ high

As can be seen above, the content in (38) is that while good evaluators of taste will conclude that the cake is not tasty, the speaker finds the cake to be tasty. The content in (39) is that while the speaker finds the cake to be not tasty, a mixture model composed of good evaluators of taste will conclude that it is. The reason why both cases are infelicitous stems from the speaker being a member of the mixture model. Hence, both of these assertions imply that the speaker is assigned a low weight within this mixture model, since in both cases the speaker’s evaluation does not affect the result. Combining this with the pragmatic assumption that individuals usually hold their own taste judgments in a higher regard than they hold others’ results in an infelicitous utterance. Note that if the speaker explicitly cancels this assumption, the utterances become better: (42)

Don’t count on my taste, I have a very unusual one, you see. This cake is not tasty, but it is tasty for me.


Don’t count on my taste, I have a very unusual one, you see. This cake is not tasty for me, but it is tasty.

12 But see Cohen (2010) for an alternative proposal for overt experiencers.

86 | Lavi Wolf

4.4 Revisiting intuition This paper opens with the intuitively clear contrast between PPT and objective predicates, yet claims that at the time of utterance, i.e. the assertion performance, PPTs behave just like objective predicates. This requires revisiting the initial intuition in a bit of detail. In order to see that these two predicates are not as different as we perceive them to be, we need a scenario in which some conversational assumptions that are usually taken for granted do not play a role. Take the following two sentences¹³, the first containing the objective predicate doctor and the second the PPT funny: (44)

a. b.

Abby is a doctor. Abby is funny.

Suppose that Suzy and James see a photograph of a woman named Abby whom they have never met (and never will), wearing a stethoscope and oversized comical sunglasses. Both of them know nothing about her: the photograph and the name consist of all their relevant knowledge concerning Abby. At this point Suzy may assert either of the sentences above, and James may object to either. Both of these reactions would constitute faultless disagreement—Suzy and James (are entitled to) draw different conclusions from the same evidence. Suzy is entitled to think that Abby is a doctor just as much as she is entitled to think that Abby is funny, and James is entitled to disagree. The sharp contrast between these two predicates seems to blur in this scenario. The distinctions between them are derived from conversational assumptions that affect the evidential step. Once these assumptions are lifted, both predicates fail this step. The first conversational assumption that is flouted in this scenario is that Suzy and James are both cooperative (Grice, 1975), i.e. obeying maxims such as quantity which dictates providing the full information with regards to the asserted content. This would require Suzy and James to hedge their assertion with additions such as I think that . . . or going by the picture here . . . . If this maxim is indeed flouted, this would be a bad move, both conversationally and theoretically. However, quantity also dictates that speakers should not provide too much information, and another maxim, manner, dictates that speakers should be brief. So the scenario is not flouting the maxims after all, just obviating. But even if the maxims were flouted, then the same problem would present itself to PPT assertions as well, i.e. that in order

13 The following example is based on a scenario in Wolf & Cohen (2011) addressing an example from Barker (2009).

Predicates of personal taste and the evidential step | 87

to preserve cooperation speakers would be forced to hedge PPT assertions e.g. According to my taste / I think that the cake is tasty. Needless to say, theories of PPT and faultless disagreement (including mine) rely on the premise that this is not the case. Another conversational assumption that is obviated is that speakers are well informed (Sauerland, 2004; Spector, 2003)—in simple terms, that they know what they are talking about, which is not the case here. And another is that the truth of objective predicates can be validated, as opposed to the truth of PPT. The last two assumptions are, of course, not always the case in everyday life, therefore not always the case in everyday linguistic exchanges. But the important point here is that obviating these assumptions leads to an intuitive blurring between objective predicates and PPT, which suggests that the main difference between these predicates is pragmatic and empirical, not lexical, i.e. at the stage of assertion the two predicates are the same. Interestingly, there is a discussion in the philosophical tradition regarding mental content which leads to a similar conclusion. Putnam (1975) discusses natural kind terms such as water and gold, and shows via several influential thought experiments how people use these terms in everyday life without really knowing what they refer to, for example the underlying microstructure of water and gold. This leads to the conclusion that the mental content of these predicates is contingent and empirical rather than objective. Moreover, when people use such terms without knowing the underlying microstructure, they must somehow defer to experts who do know it as part of a “linguistic division of labor”. Burge (1979) also discusses semantic deference in examples such as the following: an individual holds the belief that he has arthritis in the thigh. This belief’s truth value is dependent upon the content assigned to the term arthritis, which is itself dependent upon the way this term is used in the linguistic community of the speaker. If, for example, this linguistic community uses the term to refer to rheumatoid ailments in general, the content of the belief would have been such that the belief would have been true in the world of the individual. These two theories not only support the claim that there is no lexical difference between PPT and objective predicates, they also lend support to the mixture model conversational mechanism, in which assertions are evaluated in light of various individuals which are assigned weights according to their degree of perceived reliability, i.e. experts.

88 | Lavi Wolf

5 Conclusion PPTs pose a problem for semantics. The claim of this paper is that at the heart of this problem lies the transition from subjectivity, i.e. the speaker’s point of view and epistemic state from which an assertion is generated, to objectivity, i.e. propositional information that can be shared by conversational participants and employed for further inferences as conversation progresses. The conversational theory of Wolf (2015) is able to represent this transition via the evidential step which is inherent in all assertions, in which hearers interpret partial information received from various sources of evidence and turn it into truth-conditional propositions. PPTs, unlike objective predicates, cannot go through the evidential step since the source of evidence in the case of PPTs is inseparable. This is a pragmatic and not a lexical constraint, as is demonstrated in Section 4.4. They are therefore reinterpreted in a diagonalization-like manner, as claims about context rather than about the world, via a probabilistic mixture model. This theory thus provides an account of the dual nature of PPTs— predicates whose origin is a subjective perspective which are used objectively to convey truth-conditional information. Acknowledgment: I would like to thank Janneke van Wijnbergen-Huitink and Cécile Meier, the organizers of the “Subjective meaning: alternatives to relativism” workshop, and the participants for the inspiring and creative atmosphere. I am also very thankful to Ariel Cohen, Robert van Rooij and Galit Sassoon for helpful discussions on this topic.

References Barker, Chris. 2009. Clarity and the grammar of skepticism. Mind and Language 24(3). 253– 273. Burge, Tyler. 1979. Individualism and the mental. Midwest Studies in Philosophy 4(1). 73–121. Cohen, Ariel. 2010. A tasty mixture. Unpublished manuscript. Ben-Gurion University of the Negev. Crespo, Inés & Raquel Fernández. 2011. Expressing taste in dialogue. In Ron Artstein, Mark Core, David DeVault, Kallirroi Georgila, Elsi Kaiser & Amanda Stent (eds.), Semdial 2011 (Los Angelogue) Proceedings of the 15th Workshop on the Semantics and Pragmatics of Dialogue, 84–93. Los Angeles. Davis, Christopher. 2009. Decisions, dynamics and the Japanese particle yo. Journal of Semantics 26(4). 329–366. Davis, Christopher, Christopher Potts & Margaret Speas. 2007. The pragmatic values of evidential sentences. Proceedings of SALT 17. 71–88.

Predicates of personal taste and the evidential step | 89

Geach, Peter. 1965. Assertion. The Philosophical Review 74. 449–465. Grice, Herbert Paul. 1975. Logic and conversation. In Peter Cole & Jerry L. Morgan (eds.), Syntax and semantics, volume 3: Speech acts, 225–242. New York: Academic Press. Gunlogson, Christine & Gregory Carlson. 2016. Predicates of experience. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 169–200. Berlin: De Gruyter. [This volume]. Gutzmann, Daniel. 2016. If expressivism is fun, go for it! In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 21–46. Berlin: De Gruyter. [This volume]. Hintikka, Jaakko. 1961. Modality and quantification. Theoria 27. 119–128. Kaplan, David. 1989. Demonstratives. In Joseph Almog, John Perry & Howard K. Wettstein (eds.), Themes from Kaplan, 481–565. Oxford: Oxford University Press. Kölbel, Max. 2004. Faultless disagreement. Proceedings of the Aristotelian Society 104. 53–73. Lasersohn, Peter. 2005. Context dependence, disagreement, and predicates of personal taste. Linguistics and Philosophy 28(6). 643–686. Lewis, David. 1970. General semantics. Synthese 22. 18–67. Putnam, Hilary. 1975. The meaning of ‘meaning’. In Keith Gunderson (ed.), Minnesota studies in the philosophy of science 7: Language, mind and knowledge, 131–193. Minneapolis: University of Minnesota Press. Récanati, François. 2007. Perspectival thought: A defense of (moderate) relativism. Oxford: Oxford University Press. Sauerland, Uli. 2004. Scalar implicatures in complex sentences. Linguistics and Philosophy 27(3). 367–391. Sauerland, Uli. 2005. The epistemic step. Talk presented at the conference “Experimental Pragmatics” held at Cambridge University, April 14–16, 2005. Schroeder, Mark. 2009. Hybrid expressivism: Virtues and vices. Ethics 119(2). 257–309. Spector, Benjamin. 2003. Scalar implicatures: Exhaustivity and Gricean reasoning. In Balder ten Cate (ed.), Proceedings of the the eight esslli student session, 277–288. https://users. (accessed March 16, 2016). Stalnaker, Robert. 1978. Assertion. In Peter Cole (ed.), Syntax and semantics, volume 9: Pragmatics, 315–332. New York: Academic Press. Stojanovic, Isidora. 2007. Talking about taste: Disagreement, implicit arguments, and relative truth. Linguistics and Philosophy 30(6). 691–706. Umbach, Carla. 2016. Evaluative propositions and subjective judgments. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 127–168. Berlin: De Gruyter. [This volume]. van Wijnbergen-Huitink, Janneke. 2016. Subjective meaning: An introduction. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 1–20. Berlin: De Gruyter. [This volume]. Wolf, Lavi. 2015. Degrees of assertion. Beer Sheva: Ben-Gurion University of the Negev dissertation. Wolf, Lavi & Ariel Cohen. 2011. Clarity as objectivized belief. In Paul Egré & Nathan Klinedinst (eds.), Vagueness and language use, 165–190. Basingstoke: Palgrave Macmillan.

Dan Zeman

Contextualism and disagreement about taste Abstract: In this paper I investigate a certain contextualist answer to the problem raised for the view posed by the phenomenon of faultless disagreement: namely, that it cannot account for disagreement in ordinary exchanges involving predicates of personal taste. I argue that the answer investigated either misses the target, ignoring the relevant cases on which the relativist challenge is based or that it has to appeal to semantic blindness, a move that has certain costs. In addition, I argue that the same holds for a specific variant of contextualist view – the presuppositional view proposed by López de Sa.

1 Introduction One of the most vivid debates within contemporary philosophy of language and linguistics is that between contextualism and relativism about a series of expressions such as predicates of personal taste, epistemic modals, know and its kin, moral terms etc. One of the elements at the centre of this debate has been the phenomenon of “faultless disagreement”, which has been thought to pose a serious problem for contextualism and lend support to relativism. In this paper, I investigate a certain contextualist answer to the problem raised by that phenomenon (namely, that contextualism cannot account for disagreement in ordinary exchanges involving predicates of personal taste) and argue that it faces a dilemma. I also address a particular variant of contextualism (López de Sa’s presuppositional view) and show that the same happens in this case too. I start with clarifying the sense in which the labels “contextualism” and “relativism” will be used in this paper. In Section 3, I introduce the phenomenon of faultless disagreement and the problem it raises for contextualism. I also specify the notion of disagreement that relativists rely on in posing the problem. Then, in Section 4, I offer an overview of the debate I’m concerned with and of the answers contextualists have given to the problem mentioned. The next sections show that the particular answer I focus on has overlooked the cases the relativist bases her challenge on (Section 5) and that appeal to semantic blindness (a move that some contextualDan Zeman, Department of Linguistics and Basque Studies, University of the Basque Country, Unibertsitateko Ibilbidea, 5, 01006 Vitoria-Gasteiz, Spain, e-mail: [email protected]

92 | Dan Zeman

ists make) comes with a cost (Section 6). In the last section I briefly present López de Sa’s particular brand of contextualism and argue that it has the same problems as the other versions.

2 Contextualism vs. relativism: defining the notions The labels “contextualism” and “relativism” have each been used to refer to more than one position; it is thus important to clarify right from the outset what exactly is meant by them here. To get a better handle on the way I will use these terms, I adopt the familiar Kaplanian picture of semantics, according to which the assignment of a truth value to an utterance is done in two stages. At stage one, the utterance is assigned a content, by the resolution of indexicals and other contextsensitive expressions in the context of utterance; at stage two, that content is evaluated with respect to the circumstance of evaluation of the context of utterance— unless the sentence uttered contains operators that shift the circumstance to the one specified by the operators in question. Thus, in Kaplan’s (1989) view, context plays a double role. On one hand, it provides elements that get into the content of utterances: call this, following MacFarlane (2009), the “content-determinative” role of context. On the other hand, context provides elements in the circumstances of evaluation with respect to which utterances are to be evaluated: call this, again following MacFarlane (2009) the “circumstance-determinative” role of context. The elements provided by the two kinds of contexts are sometimes called determinants of truth value (see Récanati, 2007)—e.g. a time or the speaker. Thus, I will call “contextualism” about a certain expression any view according to which context has a content-determinative role with respect to a determinant of truth value corresponding to the expression in question. I will call “relativism” about a certain expression any view according to which context has a circumstancedeterminative role with respect to a determinant of truth value corresponding to the expression in question. Now, the expressions I’m concerned with in this paper are predicates of personal taste. Although there is some disagreement about which is the determinant of truth value corresponding to such expressions, I will take here that determinant to be a standard of taste.¹ I will thus call “contextual-

1 Other options are: an experiencer, a judge (Lasersohn, 2005; Stephenson, 2007b) or a perspective (Kölbel, 2004b). In this relation, see also the introduction in van Wijnbergen (2016) [this volume].

Contextualism and disagreement about taste

| 93

ism about taste” any view according to which context has a content-determinative role with respect to standards of taste and “relativism about taste” any view according to which context has a circumstance-determinative role with respect to standards of taste. This way of carving the logical space will lump together very different positions under a single label; however, as far as I can see, this will not have any bearing on the issues raised in this paper.²

3 Faultless disagreement and the disagreement problem The phenomenon that has been invoked by many proponents of relativism as the main motivation for the view is the phenomenon of faultless disagreement. Cases of faultless disagreement are cases in which, although there is a perceived disagreement between two parties (or, as philosophers like to put it, an intuition to that effect), none of them is at fault in any significant way. To use the example of predicates of personal taste, imagine Alice and Bill in the middle of a ride on the Formula Rossa, the fastest roller coaster in the world.³ In the following exchange (1)

Alice: This is fun! Bill: No, it’s not fun. It’s terrible!

there is a tendency to interpret Alice and Bill both as having a disagreement (signaled by the expression No at the beginning of B’s line) and as being faultless.⁴ Starting from such an intuition of faultless disagreement, the argument against contextualism (or at least against very simple versions of it) has been that, since it treats utterances such as that of Alice and Bill in (1) as making the standard of taste according to which the ride is fun or not part of the content of those utterances (so that the content of Alice’s utterance is that the ride on the Formula Rossa is fun according to Alice’s standard of taste, while the content of Bill’s utterance is that the ride on the Formula Rossa is not fun according to Bill’s

2 Thus, under “contextualism”, as I defined the term, one will find views as different as Stanley’s (2007) and Récanati’s (2010). On the other hand, under “relativism” one will find both more moderate views such as Kölbel’s (2004b), Kompa’s (2005), Récanati’s (2007) or Broogaard’s (2008) and more radical views such as MacFarlane’s (2014) or Lasersohn’s (2005; 2008). 3 At least, according to Wikipedia: (May 28, 2014). 4 See Wright (1992) and Kölbel (2004a) for more about this phenomenon, as well as many of the papers in García-Carpintero & Kölbel (2008).

94 | Dan Zeman

standard of taste), it fails to capture the perceived disagreement between Alice and Bill, despite capturing the fact that they are perceived as faultless. Critics of contextualism such as Kölbel (2004b) and Lasersohn (2005) have thus concluded that contextualism cannot account for faultless disagreement, and thereby that a crucial feature of our use of predicates of personal taste is left unaccounted for. Since the contextualists’ failure consists in not being able to account for the disagreement part of the phenomenon, I will refer to the problem with the phrase “the disagreement problem”, leaving aside the faultlessness aspect of the phenomenon. A legitimate question that arises at this point is what exactly the notion of disagreement employed by the relativist in raising her challenge is. From the many notions of disagreement that can be found in the literature, I will rely here on one notion that can be extracted from the characterization given by Kölbel to the phenomenon of faultless disagreement. According to Kölbel: A faultless disagreement is a situation where there is a thinker A, a thinker B, and a proposition (content of judgment) p, such that: (a) A believes (judges) that p and B believes (judges) that not-p. (b) Neither A nor B has made a mistake (is at fault). (Kölbel, 2004a, 53–54) The notion of disagreement used corresponds to clause (a) above. Let us flag it for further reference: Simple Disagreement: A and B disagree if A believes (judges) that p and B believes (judges) that non-p. Although simple disagreement is not without problems in sorting out intuitive cases of disagreement from cases in which disagreement is lacking (see, for example, MacFarlane, 2007 and Marques, 2014a), I will use it for purposes of distinguishing between the main contextualist answers to the disagreement problem. As we will see in the next section, each such answer could be characterized by a certain attitude towards simple disagreement.

4 Contextualist answers to the disagreement problem Before getting to the answers alluded to above, let me put the debate surrounding the disagreement problem in a larger context. As I understand it, the debate between contextualism and relativism I’m concerned with has gone through three stages. The first stage consists in the apparition of contextualist views about

Contextualism and disagreement about taste

| 95

certain expressions, most prominently know and its kin (Cohen, 1986; DeRose, 1992)—but also about moral terms (for example Dreier, 1990; Harman, 1996)—as a reaction to invariantist treatment of such expressions.⁵ And although predicates of personal taste haven’t been in the spotlight that much, it is fair to say that at this stage, one form or other of contextualism has been considered the orthodoxy when it comes to their semantics too. Stage two consists in the appearance of relativist doctrines challenging contextualism as the right semantics for the target expressions. While vindicating the invariantist view when it comes to the semantic content of utterances containing the target expressions, relativism sides with contextualism in highlighting the context-sensitivity such expressions exhibit, differing from it in the way such context-sensitivity is accounted for—namely, by assigning different roles to context (see Section 2) and thus by employing different mechanisms of context resolution. This is where faultless disagreement and data like the exchange in (1) come into the picture: they are supposed to form the basis of an argument in favor or relativism and against contextualism for predicates of personal taste and other expressions the semantics of which has been taken so far to be contextualist. The third stage of the dispute I’m considering consists in the contextualist reaction to the relativist challenge. When it comes to the specific problem posed by exchanges like (1), there are two possible reactions to it. One is to deny that the intuition of disagreement is present in such exchanges. Now, it is common sense that the extent to which intuitions should be relied upon in philosophical theorizing is a delicate issue, and certainly much more needs to be said about the conditions under which they could be trusted. In this particular case, though, the intuition of disagreement seems to be pretty robust and thus, I take it, cannot be simply ignored. And, indeed, it hasn’t: the more widespread reaction among contextualists to the problem posed by exchanges like (1) has been to agree that the intuition of disagreement is present but to show that, despite the relativist’s contention, contextualism has the resources to account for it and thus provide a solution to the disagreement problem. Here, three contextualist answers can be distinguished on the basis of the proponents’ attitude towards simple disagreement. Thus, one answer was to accept simple disagreement and to show that, even if understood in this way, contextualism can yield disagreement. This way of answering the disagreement problem appeals to certain uses of predicates of personal taste that the relativist has largely ignored in raising her challenge. The

5 Of course, expressivism was also a prominent view. For a locus classicus of expressivism, see Blackburn, 1984. Recently, hybrid versions of expressivism have been proposed for predicates of personal taste too - see, for example, Gutzmann (2016) [this volume].

96 | Dan Zeman

idea is that disagreement can be accounted for in the case of those uses of predicates of personal taste and thus that the problem of disagreement is solved. The second answer consists in denying that the condition for disagreement specified by simple disagreement (namely, that disagreement involves contradictory contents) is necessary for disagreement. This answer has been accompanied by the unearthing of a host of phenomena that all have the right of being counted as disagreement (for a summary, see Cohnitz & Marques, 2014). In particular, several authors have claimed that the disagreement perceived in exchanges like (1) is of a non-doxastic nature and thus that we need to appeal to different notions of disagreement than the one characterized by simple disagreement to capture it. Thus, it has been claimed that the disagreement in exchanges like (1) is one in attitude, as opposed to belief (Huvenes, 2012, 2014; Marques, 2014a; López de Sa, 2015) or of a practical nature (Stojanovic, 2012; Marques, 2014b; Marques & García-Carpintero, 2014). Given that such an understanding of disagreement is compatible with a contextualist semantics for predicates of personal taste, the disagreement problem is solved. The third answer has been to claim that the disagreement in exchanges like (1) is not semantic (thus not involving the semantic content of the utterances of the interlocutors), but pragmatic. Sundell (2011) (but see also Umbach, 2016 [this volume]) lists a number of ways in which exchanges like (1) could be interpreted that all locate disagreement at a different level than that of semantic content. Thus, there is disagreement about presuppositions (a view also put forward, in different forms, by López de Sa, 2007, 2008, 2015 and Parsons, 2013), disagreement about implicatures (for example Finlay, 2005), manner disagreement (about the words used or their pronunciation), character disagreement (about the meaning of the words) and context disagreement (about what context the interlocutors are in). As such, this approach need not deny that simple disagreement holds, but if so it does with respect to a different level than the semantic one. Given that such an understanding of disagreement is compatible with a contextualist semantics for predicates of personal taste, the disagreement problem is, again, solved. Although the last two answers have been proven to be popular with contextualists lately, I will not be concerned with them in this paper. I will instead focus on the first answer to the disagreement problem and inquire into whether it is satisfactory. My aim will be to show that contextualists that take this route have either disregarded the crucial cases, their answer thus missing the point of the relativist challenge or their response involves appeal to some form or other of semantic blindness on the part of the speakers, thus incurring additional theoretical costs.

Contextualism and disagreement about taste

| 97

5 Relevant and irrelevant cases As I briefly mentioned above, the contextualist strategy pertaining to this answer to the disagreement problem has been to draw attention to uses of predicates of personal taste in which contextualism can account for disagreement in exchanges such as (1), even if disagreement is conceived along the lines of simple disagreement. Thus, various authors have claimed that disagreement can easily be accounted for if one focuses on cases in which predicates of personal taste are put to a exocentric use⁶, when one makes a claim about someone else’s preferences (Glanzberg, 2007; Cappelen & Hawthorne, 2009), a group use, when one makes a claim about a group’s preferences (Cappelen & Hawthorne, 2009; Huvenes, 2012) or a generic use, when one makes a claim about the preferences of the entire humankind or a suitably restricted subset thereof (Stojanovic, 2007; Cappelen & Hawthorne, 2009; Moltmann, 2010). Since these authors present such uses in the form of a list, it may be useful for us to make such a list as well. Building on the one provided by Cappelen & Hawthorne (2009, 110-111) for an exchange similar to (1), the following cases in which contextualism is able to account for disagreement, even if conceived along the lines of simple disagreement, are possible (the labels in square brackets are mine): (i) The speaker is using fun autocentrically, the hearer realizes this, but exocentrically points out that the relevant event is not fun for the original speaker. [Exocentric use] (ii) The speaker is claiming that the referent of this is fun for a group that includes the interlocutor. While it is fun for the speaker, it is not fun for certain other members of the group. Here the interlocutor is quite within his rights to correct the speaker. Once corrected, the speaker will in that case not stick to his guns unless he feels the alleged counterevidence is faulty. [Group use] (iii) The original speaker is using fun generically—that is, intending to make a claim about what people in general find fun, and the hearer uses fun in the same way. [Generic use] However, none of the cases on this list are the ones that the relativist bases (or should base) her challenge on. What I take the crucial cases for the relativist to be are, perhaps among others, the following (for the same exchange as before):

6 For the difference between exocentric and egocentric/autocentric uses and a detailed discussion of them, see Lasersohn (2005) and Stephenson (2007a).

98 | Dan Zeman

(iv) The original speaker is using fun autocentrically and the hearer also uses fun autocentrically (even if he realizes that the original speaker uses fun autocentrically). [Individual-Egocentric use]⁷ (v) The speaker is claiming that the referent of this is fun for a group that does not include her interlocutor. The interlocutor, in her turn, is claiming that the referent of this is fun for a group that does not include the original speaker. [Group-Egocentric use]⁸ Since cases (iv)–(v) are as natural as cases (i)–(iii), it is not enough to point out that there are some cases in which disagreement is accounted for without managing to account for it in all cases. And, while it is true that in initial formulations of the disagreement problem the issue of the variety of uses predicates of personal taste can be put to did not arise, and perhaps as a consequence exocentric, group and generic uses of such predicates have been ignored (the notable exceptions here being Lasersohn, 2005; Stephenson, 2007b), the important thing to note is that the cases on which the relativist should base her challenge on (that is, cases (iv) and (v)), are not on the contextualist’s list. It is pretty obvious that the contextualist cannot account for disagreement in (iv)–(v) as easy as she can in cases (i)–(iii). Assuming that relativism can account both for cases (iv)–(v) and for cases (i)–(iii) (there is nothing in relativism precluding a principled treatment of exocentric, group and generic uses of predicates of personal taste), it follows that, with respect to the disagreement problem as formulated here, relativism is in a better position than contextualism.⁹ Of course, the contextualist could deny that the intuition of disagreement is present in specific cases like (v) and (vi), but this strikes me as a rather ad-hoc move. For what this comes down to is to claim that the intuition is missing precisely in the cases that are crucial for the relativist challenge. Not only does this strategy seem fishy, but by following it contextualists jeopardize their own right to appeal to intuitions—an appeal that has played an important role in their opposition to invariantism.

7 This, I take it, is the use most relativists have put predicates of personal taste to in raising the challenge for contextualism, even though this hasn’t always been explicit. For explicitly appealing to such a use, see Lasersohn (2005). 8 Appeal to this use is not widespread, but see López de Sa (2008). 9 This is not to say that relativism is not without problems when it comes to accounting for disagreement. A series of authors have questioned relativism’s, in both its forms, capacity to do so (among others Iacona, 2008; Rosenkranz, 2008; García-Carpintero, 2008; Moltmann, 2010; Francén, 2010; Marques, 2014b). My aim here is not do defend relativism from such allegations; it is, rather, to show that, at least when it comes to cases like (iv) and (v), relativism has no problems, while the particular contextualist answer to the disagreement problem investigated does.

Contextualism and disagreement about taste |


6 Semantic blindness and its costs The upshot of the section above is that the contextualist answer to the disagreement problem considered above is, at best, incomplete. This creates the need from the contextualist part to say something more about cases like (iv) and (v). Now, in some recent approaches to this issue, the contextualist strategy has been to treat the disagreement in cases like (iv) and (v) by employing a different notion of disagreement than the one captured by simple disagreement (the second answer mentioned in Section 4). However, such a strategy is not the only one, and some contextualists have been content to appeal to semantic blindness in connection to the cases we are interested in. Although this is not a novel point¹⁰, in what follows I want to stress that the cost for the contextualist that such a strategy comes with is not negligible and that the acknowledgment of this point might have consequences for the overall comparison between contextualism and rival views. First, however, let us ask what semantic blindness is. In a nutshell, semantic blindness involves the attribution, by the theorist, in a systematic and large-scale way, of ignorance of how some expressions in a language work to otherwise competent speakers of that language. Appeal to semantic blindness in semantics is not new. It has been, for example, the preferred solution of some traditional contextualists (featuring prominently in DeRose’s 2006 defense of epistemic contextualism). In connection to predicates of personal taste, such a strategy has been appealed to by, for example, Cappelen & Hawthorne (2009). Thus, in connection to a semantics of hot, which they take as a model for a semantics of predicates of personal taste, they write: [T]he contextualist will explain the relevant datum by appeal to a dose of semantic blindness (. . . ) [O]wing to misjudgments about semantic uniformity, some disagreement judgments are accepted when they ought not be. (Cappelen & Hawthorne, 2009, 118)

Having to posit semantic blindness to explain away certain phenomena, however, was found by many philosophers (perhaps most prominently by Schiffer, 1996) to be an unattractive feature of a semantic view. In a more recent expression of this attitude, Hawthorne writes: “Obviously, we do not wish to posit semantic blindness when we don’t have to. Other things being equal, a semantic theory that didn’t claim semantic blindness on the part of competent speakers would be a better theory.” (Hawthorne, 2004, 109) Now, of course, the crucial issue is whether

10 See Baker (2012) for making the point that adopting semantic blindness is not an innocent concession on the contextualist part, and Alexander & Westerståhl (2010) for making the same point in connection to Cappelen & Hawthorne’s (2009) specific proposal.

100 | Dan Zeman

other things are equal. A full comparison between contextualism and relativism is beyond the scope of this paper, but let me note that one major advantage contextualists have claimed their view has over relativism is that it doesn’t complicate the semantic machinery unnecessarily, and thus that it doesn’t incur additional theoretical costs. But if it turns out that there is a need to appeal to semantic blindness on the contextualist part, this advantage is not on as strong a footing as contextualists would like it to be. Acknowledging this fact is a precondition of a realistic weighting of the strengths of the view. This is so even if, as some authors claim (for example Montminy, 2009; Kindermann, 2013), relativism itself has to appeal to some form of blindness: a fair and accurate comparison of the two views cannot be done without being clear on the theoretical costs of each of them. As I said, I won’t provide such a comparison here, but the point that the costs of appealing to semantic blindness need to be made explicit remains.¹¹

7 Presuppositional contextualism and presuppositional blindness Although it belongs to what I called in Section 4 pragmatic strategies, in this final section I investigate an interesting form of contextualism that claims to yield disagreement in cases like (iv) and (v), even if disagreement is conceived along the lines of simple disagreement. Thus, in a series of papers, López de Sa (2007, 2008, 2015) has contended that disagreement is possible because predicates of personal taste trigger a “presupposition of commonality” (a view also put forward, but not endorsed, in Kölbel, 2007): when interlocutors enter in exchanges like (1) above, they presuppose that they are alike in their tastes—that is, they presuppose that a unique standard of taste is shared in the conversation and that the disagreement is about whether (in the specific case of (1)) the ride on the Formula Rossa is fun or not according to that standard. Such a move is compatible with a contextualist semantics for predicates of personal taste: while the semantic content expressed by the utterances of the two interlocutors will contain the standard according to which the ride on the Formula Rossa is fun, so that the content of Alice’s utterance is that the ride on the Formula Rossa is fun according to Alice’s standard of taste, while the content of Bill’s utterance is that the ride on the Formula Rossa is not fun according to Bill’s standard of taste, the presupposition of commonality en-

11 See, however, Hirvonen (2016) [this volume] for discussing semantic blindness and for sketching an error-theory for predicates of personal taste.

Contextualism and disagreement about taste

| 101

sures that, in effect, the interlocutors disagree. The disagreement problem is thus solved, without ignoring the relevant cases and (allegedly) without appealing to semantic blindness. One can find several criticisms of López de Sa’s view in the literature. Thus, MacFarlane (2007) has complained that the presuppositional view, with its understanding of presupposition in pragmatic terms, is fitted for conversations, but that disagreement extends beyond such cases. Baker (2012) has applied a series of standard tests for presupposition to discourses involving predicates of personal taste and has concluded that they fail. Marques & García-Carpintero (2014) argue that presuppositional contextualism has troubles with cases in which the interlocutors are “enlightened” about their different tastes. Here, however, I will not take any sides in this dispute. Rather, what I want to point out is that, in connection to cases (iv) and (v), López de Sa, as the other contextualists proposing the answer investigated in this paper, is (tacitly) appealing to a form of blindness. On standard versions of contextualism, semantic blindness is related to a lack of knowledge on the speakers’ part with respect to the meaning of words in their own language. On presuppositional contextualism, the form of blindness involved is related to a lack of knowledge on the speakers’ part with respect to the presuppositions triggered by words in their own language. Since, on this view, predicates of personal taste function like presupposition triggers, it follows that the perceived disagreement in cases in which the interlocutors don’t presuppose that they share a standard of taste stems from their sheer ignorance about the presuppositions triggered by the predicates of personal taste they use. Not only does this appeal to presuppositional blindness burden the contextualist in the same way as semantic blindness burdens the proponent of standard versions, but it also strikes me as implausible for the reason that such ignorance doesn’t seem to be present in other cases of presupposition-triggering expressions. One way to bring this point to light is to compare the following exchanges between Alice and Bob—this time Bob playing the role of a presuppositional contextualist: (2)


a. b. c.

Alice: The cat is on the mat. Bob: So, you presuppose that there is exactly one cat and exactly one mat around. Alice (puzzled): Yes, I do. That’s why I said “the cat” and “the mat”.

a. b. c.

Alice: John started smoking again. Bob: So, you presuppose that John has smoked before. Alice (puzzled): Yes, I do. That’s why I said “again”.

102 | Dan Zeman


a. b. c.

Alice: This is fun.¹² Bob: So, you presuppose we share a standard of taste. Alice (intrigued): No, I don’t. I actually know you cannot stand the Formula Rossa. Why would you ask that?!

The comparison between (2) and (3), on one hand, and (4), on the other shows that presuppositional blindness appears only in the case of predicates of personal taste. This is peculiar, given that people don’t seem in general to be oblivious to the presuppositions triggered by other words in their own language. What this suggests, I take it, is that there is no basis for claiming that such presuppositional blindness exists. If one takes this to be the only explanation for why people take themselves to disagree in cases (iv) and (v)¹³, what this shows is that the presupposition of commonality might not be there in the first place. It seems, then, that both standard versions of contextualism and the presuppositional variant face the same dilemma: they either ignore the relevant cases on which the disagreement problem is based or they appeal to semantic blindness, thus incurring additional theoretical costs. A full comparison between contextualism and relativism is beyond the scope of this paper, but, with respect to the disagreement problem as I understand it here, the particular contextualist answer I have investigated doesn’t seem to be the best way to go. Acknowledgment: I’m grateful to the editors of this volume, Janneke van Wijnbergen-Huitink and Cécile Meier, for constant support and valuable feedback. I would also like to thank an anonymous reviewer for comments and suggestions. Research for this paper has been financially supported by a MINECO Juan de la Cierva grant (JCI-2012-12974) and by the Semantic Content and Conversational Dynamics project (FFI2012-37658) at the University of Barcelona.

12 I’m assuming here that this refers to the Formula Rossa and that Alice’s use of fun is egocentric. 13 Recently, López de Sa (2015) has employed a new strategy to deal with cases (iv) and (v): appeal to disagreement in attitude. While I want to be neutral here about whether the strategy as a whole works, I have two misgivings about its adoption by López de Sa. First, by adopting it, he provides a non-unitary treatment of disagreement in cases like (iv) and (v), employing one strategy (the presupposition of commonality) when the interlocutors share a standard of taste, and another strategy (appeal to disagreement in attitude) when they don’t. Second, I disagree with the motivation López de Sa gives for the appeal to disagreement in attitude. According to the new view, there is a difference between disagreement existing and disagreement being expressed, the claim being that in cases like (iv) and (v) it is not possible to express disagreement (despite the interlocutors having different attitudes towards the object of evaluation, and thus disagreeing). But it seems to me obvious that the disagreement is easily expressed in such cases (in fact, I take it that this is precisely what Alice and Bob do in the exchange focused on in this paper).

Contextualism and disagreement about taste |


References Alexander, Almér & Dag Westerståhl. 2010. Review of relativism and monadic truth. Linguistics and Philosophy 33(1). 37–50. Baker, Carl. 2012. Indexical contextualism and the challenges from disagreement. Philosophical Studies 157(1). 107–123. Broogaard, Berit. 2008. Moral contextualism and moral relativism. The Philosophical Quarterly 58(232). 385–409. Cappelen, Herman & John Hawthorne. 2009. Relativism and monadic truth. Oxford: Oxford University Press. Cohen, Stewart. 1986. Knowledge and context. Journal of Philosophy 83(10). 574–583. Cohnitz, Daniel & Teresa Marques. 2014. Disagreements. Erkenntnis 79(1). 1–10. DeRose, Keith. 1992. Contextualism and knowledge attributions. Philosophy and Phenomenological Research 52(4). 913–929. Dreier, James. 1990. Internalism and speaker relativism. Ethics 101(1). 6–26. Finlay, Stephen. 2005. Value and implicature. Philosopher’s Imprint 5(4). 1–20. Francén, Ragnar. 2010. No deep disagreement for new relativists. Philosophical Studies 151(1). 19–37. García-Carpintero, Manuel. 2008. Relativism, vagueness and what is said. In Manuel GarcíaCarpintero & Max Kölbel (eds.), Relative truth, 129–154. Oxford: Oxford University Press. García-Carpintero, Manuel & Max Kölbel (eds.). 2008. Relative truth. Oxford: Oxford University Press. Glanzberg, Michael. 2007. Context, content, and relativism. Philosophical Studies 136. 1–29. Gutzmann, Daniel. 2016. If expressivism is fun, go for it! In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 21–46. Berlin: De Gruyter. [This volume]. Harman, Gilbert. 1996. Moral relativism. In Gilbert Harman & Judith Jarvis Thompson (eds.), Moral relativism and moral objectivity, 1–64. Oxford: Blackwell. Hawthorne, John. 2004. Knowledge and lotteries. Oxford: Clarendon Press. Hirvonen, Sanna. 2016. Doing without judge dependence. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 47–68. Berlin: De Gruyter. [This volume]. Huvenes, Torfinn. 2012. Varieties of disagreements and predicates of taste. Australasian Journal of Philosophy 90. 167–181. Huvenes, Torfinn. 2014. Disagreement without error. Erkenntnis 79(1). 143–154. Iacona, Andrea. 2008. Faultless or disagreement. In Manuel García-Carpintero & Max Kölbel (eds.), Relative truth, 287–295. Oxford: Oxford University Press. Kaplan, David. 1989. Demonstratives. In Joseph Almog, John Perry & Howard K. Wettstein (eds.), Themes from Kaplan, 481–565. Oxford: Oxford University Press. Kindermann, Dirk. 2013. Relativism, sceptical paradox and semantic blindness. Philosophical Studies 162(3). 585–603. Kölbel, Max. 2004a. Faultless disagreement. Proceedings of the Aristotelian Society 104. 53–73. Kölbel, Max. 2004b. Indexical relativism vs. genuine relativism. International Journal of Philosophical Studies 12(3). 297–313. Kölbel, Max. 2007. How to spell out genuine relativism and how to defend indexical relativism. International Journal of Philosophical Studies 15(2). 281–288.

104 | Dan Zeman

Kompa, Nikola. 2005. The semantics of knowledge attributions. Acta Analytica 20(1). 16–28. Lasersohn, Peter. 2005. Context dependence, disagreement, and predicates of personal taste. Linguistics and Philosophy 28(6). 643–686. Lasersohn, Peter. 2008. Quantification and perspective in relativist semantics. Philosophical Perspectives 22(1). 305–337. López de Sa, Dan. 2007. The many relativisms and the question of disagreement. International Journal of Philosophical Studies 15(2). 269–279. López de Sa, Dan. 2008. Presuppositions of commonality: An indexical relativist account of disagreement. In Manuel García-Carpintero & Max Kölbel (eds.), Relative truth, 297–310. Oxford: Oxford University Press. López de Sa, Dan. 2015. Expressing disagreement: A presuppositional indexical contextualist relativist account. Erkenntnis 80(1). 153–165. MacFarlane, John. 2007. Relativism and disageement. Philosophical Studies 132(1). 17–31. MacFarlane, John. 2009. Nonindexical contextualism. Synthese 166(2). 231–250. MacFarlane, John. 2014. Assessment sensitivity: Relative truth and its applications. Oxford: Oxford University Press. Marques, Teresa. 2014a. Doxastic disagreement. Erkenntnis 79(1). 121–142. Marques, Teresa. 2014b. Relative correctness. Philosophical Studies 167(2). 361–373. Marques, Teresa & Manuel García-Carpintero. 2014. Disagreement about taste: Communality presuppositions and coordination. Australasian Journal of Philosophy 92(4). 701–723. Moltmann, Friederike. 2010. Relative truth and the first person. Philosophical Studies 150. 187–220. Montminy, Martin. 2009. Contextualism, relativism and ordinary speakers’ judgments. Philosophical Studies 143(3). 341–356. Parsons, Josh. 2013. Presupposition, disagreement, and predicates of taste. Proceedings of the Aristotelian Society 113(2). 163–173. Récanati, François. 2007. Perspectival thought: A defense of (moderate) relativism. Oxford: Oxford University Press. Récanati, François. 2010. Truth-conditional pragmatics. Oxford: Oxford University Press. Rosenkranz, Sven. 2008. Frege, relativism and faultless disagreement. In Manuel GarcíaCarpintero & Max Kölbel (eds.), Relative truth, 225–237. Oxford: Oxford University Press. Schiffer, Stephen. 1996. Contextualist solutions to scepticism. Proceedings of the Aristotelian Society 96. 317–333. Stanley, Jason. 2007. Language in context. Oxford: Oxford University Press. Stephenson, Tamina. 2007a. Judge dependence, epistemic modals, and predicates of personal taste. Linguistics and Philosophy 30(4). 487–525. Stephenson, Tamina. 2007b. Towards a theory of subjective meaning. Cambridge, MA: MIT dissertation. Stojanovic, Isidora. 2007. Talking about taste: Disagreement, implicit arguments, and relative truth. Linguistics and Philosophy 30(6). 691–706. Stojanovic, Isidora. 2012. Emotional disagreement. Dialogue 51(1). 99–117. Sundell, Timothy. 2011. Disagreements about taste. Philosophical Studies 155(2). 267–288. Umbach, Carla. 2016. Evaluative propositions and subjective judgments. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 127–168. Berlin: De Gruyter. [This volume]. Wright, Crispin. 1992. Truth and objectivity. Cambridge: Cambridge University Press.

Christopher Kennedy

Two kinds of subjectivity Abstract: What makes a predicate subjective? To a certain extent at least, this question is independent of the question of how (or whether) subjectivity should be captured by semantic theory, in the sense that any theory that recognizes a subjective/nonsubjective distinction in the first place should be able to say what kinds of features constitute the difference between the subjective and the nonsubjective terms of the language. However, it is often the case that a better understanding of fine-grained details of the lexical semantic properties of particular classes of terms can inform our theoretical understanding and analysis of the categories and constructions that those terms enter into. The aim of this paper is to begin to develop such an understanding. My strategy will be to focus on one empirical domain, scalar predicates, and to look for patterns of distribution and interpretation of expressions that track the subjective/nonsubjective distinction. The conclusion will be that there are (at least) two kinds of subjectivity which are distinguished in that one affects distribution and one does not: subjectivity that is associated with vagueness generally, and subjectivity that is associated with predicates that involve qualitative assessment, independent of vagueness. A caveat: focusing on scalar predicates allows me to carry out a fine-grained examination of lexical semantic details, but it remains to be seen whether the conclusions that I reach about subjectivity in this domain can also help us understand subjectivity in modals, conditionals, and other kinds of constructions.

1 Subjectivity and selection In order to investigate subjectivity, we need to know how to identify it. The most common diagnostic for subjectivity involves truth assessment: if two opposing sentences can be used to contradict each other without implying that one of the speakers has said something false, they are thought to involve some form of subjectivity. This kind of test is exemplified by the “faultless disagreement” pattern in (1) (Kölbel, 2002). (1)

a. b.

Anna: Trippa alla romana is tasty. Beatrice: Trippa alla romana is not tasty.

Christopher Kennedy, Department of Linguistics, University of Chicago, 1010 E. 59th St., Chicago, IL 60637, USA, e-mail: [email protected]

106 | Christopher Kennedy

Beatrice’s utterance in (1b) is understood as contradicting Anna’s utterance in (1a), and so represents a kind of disagreement, yet we have the clear sense that both Anna and Beatrice are in some sense right, and so the disagreement is “faultless”. This situation contrasts with the one in (2), in which we have a case of disagreement in which one of the speakers (Anna) is without question wrong, because trippa alla romana is made from the stomach lining of a cow or other animal. (2)

a. b.

Anna: Trippa alla romana is a vegetarian dish. Beatrice: Trippa alla romana is not a vegetarian dish.

The scenario in (1) also contrasts with the one in (3), where we imagine a context in which Anna and Beatrice are looking at menus that, unknown to each other, differ on the dish that will follow the one they are currently eating: for Anna, it is trippa alla romana; for Beatrice, it is saltimbocca. (3)

a. b.

Anna: Trippa alla romana is next on the menu. Beatrice: Trippa alla romana is not next on the menu.

As in (1) and unlike (2), neither Anna nor Beatrice make a false assertion; but in (3), this is because they are simply saying different things: there is no real disagreement in the first place. Faultless disagreement and other phenomena related to truth assessment sensitivity have played a central role in theorizing about subjectivity, and in particular in deciding between contextualist and relativist accounts of the phenomenon. According to both accounts, subjectivity—and corresponding faultlessness in disagreements like (1)—arises when a predicate is relativized to a “judge”, which is typically the speaker in simple assertions like the ones under discussion here. In a relativist analysis, the character and content of a subjective predicate are fixed, but its extension is judge-dependent (see for example Lasersohn, 2005, 2009; Stephenson, 2007; Kölbel, 2002, 2009). On this view, Anna and Beatrice truly disagree about whether the same property can be predicated of trippa alla romana; at the same time, their two utterances can both be true in virtue of the fact that the extension of tasty relativized to Anna may include trippa alla romana, while the extension of tasty relativized to Beatrice may not. In a contextualist account, in contrast, the character of a subjective predicate is constant, but its content and extension are judge-dependent (see for example Glanzberg, 2007; Stojanovic, 2007). Faultlessness in (1) is due to the fact that the property claimed to hold of trippa alla romana in Anna’s utterance in (1a) is different from the property denied to hold of trippa alla romana in Beatrice’s utterance in (1b): roughly tasty for Anna vs. tasty for Beatrice. The problem for this view is

Two kinds of subjectivity

| 107

that it does not provide a clear explanation of why Anna and Beatrice appear to be in disagreement. In particular, without saying more, it fails to explain the difference between the discourse in (1) and the one in (3), in which a standard contextualist analysis of next derives exactly the right result: next in Anna’s mouth means next on Anna’s menu and next in Beatrice’s mouth means next on Beatrice’s menu, so there is no disagreement (and no fault). While patterns of truth assessment like the faultless disagreement paradigm provide us with a diagnostic for subjectivity and may also help distinguish competing theoretical analyses, they do not help us decide whether subjectivity correlates with some feature of the linguistic representation (cf. Stojanovic, 2007). To answer this question, we need to ask whether there are patterns of interaction between subjective predicates and other expressions. The one I want to focus on here involves subjective attitude verbs such as English find in the construction find x pred, which is discussed in detail by Sæbø (2009). As shown by the pattern of acceptability in (4), this construction requires the predicate that heads the small clause complement of find to be subjective: replacing the subjective predicate tasty with a non-subjective predicate like a vegetarian dish or tasty for Beatrice (in which for Beatrice makes the judge explicit and renders the predicate nonsubjective) results in unacceptability. (4)

a. Anna finds trippa alla romana tasty. b. ??Anna finds trippa alla romana to be a vegetarian dish. c. ??Anna finds trippa alla romana tasty for Beatrice.

Speaking descriptively, these examples show that the subject of find must be understood as the judge of the embedded predicate—(4a) means roughly the same thing as Trippa alla romana is tasty to Anna—and when this condition is not met, the resulting sentence is unacceptable. Find is not the only verb whose subject can be understood as the judge of an embedded subjective term; the subjects of doxastic attitude verbs such as think and believe, discussed by Stephenson (2007), can also be understood in this way. (6a), for example, is more or less synonymous with (4a). (5)

a. b. c.

Anna believes trippa alla romana to be tasty. Anna believes trippa alla romana to be a vegetarian dish. Anna believes trippa alla romana to be tasty for Beatrice.

However, unlike find, believe can also embed non-subjective predicates, as shown by the acceptability of (5b–c). Similarly, the contrast between (6a) and (6b) shows that only the subject of find is obligatorily understood to be the judge of the embedded predicate. (6a) is acceptable because a contextual individual (the cat) can

108 | Christopher Kennedy

be understood to be the judge of what counts as tasty, rather than the individual denoted by the subject of the attitude verb (Anna); (6b) is odd because Anna must be understood as the judge, which implies that she has been eating the cat food. (6)

a. Anna thinks/believes the cat food is tasty (because the cat ate it all up). b. ??Anna finds the cat food tasty (because the cat ate it all up).

Based on the contrast in (6), Stephenson (2007) proposes that find means the same thing as think or believe, but has an extra requirement that the doxastic anchor have direct experience of the embedded proposition. However, as Sæbø (2009) points out, this analysis doesn’t explain the unacceptability of (7b), given the assumption that Homer has direct experience of his sexual orientation. (7)

a. Homer thinks/believes he is gay. b. ??Homer finds himself (to be) gay.

Sæbø argues that (7b) shows that find actually selects for a subjective predicate, and (in accord with the intuition stated above) fixes the judge of the embedded predicate to the semantic value of its subject. On this view, find does not itself introduce any truth-conditional content; it is instead a “radical judge-shifter”. Sæbø provides both a relativist and a contextualist implementation of the analysis. In the relativist variant, find causes the extension of the embedded predicate to be determined relative to its subject. On this view, the problem with (7b) is that the contribution of find is completely vacuous: since gay is non-subjective, its extension is judge-independent, and fixing its judge parameter to Homer makes no difference in meaning. In the contextualist version of Sæbø’s analysis, subjective predicates are type-wise distinct from non-subjective predicates, in having an extra judge argument: tasty is type ⟨e, ⟨e, t⟩⟩, where the most external argument is the judge; gay is type ⟨e, t⟩. The function of find, on this view, is to bind the value of the judge argument of the embedded predicate to the matrix subject, and (7b) is unacceptable since the non-subjective predicate gay lacks a judge argument position: this is a case of type-mismatch. In his paper, Sæbø discusses several patterns of data involving find and related verbs in other languages which support the hypothesis that there is a typetheoretic difference between the predicates that are acceptable under find and those that are not, and uses this result to draw conclusions about the right theory of subjectivity, arguing for a contextualist analysis of the sort proposed in Stojanovic (2007). In this paper, I want to accept Sæbø’s conclusions about semantic type, but step back from the contextualism/relativism debate, and instead just use the subjective attitude verb construction as a way of probing the linguistic representation of subjectivity. In particular, I want to ask whether all predicates that

Two kinds of subjectivity



are subjective in virtue of their behavior in faultless disagreement contexts are also acceptable in the x find y pred construction. If so, we may conclude that subjectivity uniformly has type-theoretic consequences. If not—as I believe the facts suggest—then we must recognize more than one form of subjectivity: one that involves a type-theoretic aspect of meaning, and one that does not.

2 Subjective standards vs. subjective evaluations Richard (2004) observes that predicates of personal taste are not the only kinds of scalar predicates that display faultless disagreement effects. Instead, most (and maybe all) vague scalar predicates can give rise to this pattern: (8)

a. b.

Anna: “Carla is rich/heavy/old/tall.” Beatrice: “No she’s not!”

Richard points out that there are two ways of understanding the dialogues in (8). On one interpretation, Anna and Beatrice appear to disagree because they are each implicitly comparing Carla to different groups of people or “comparison classes”, e.g. for rich, the bottom 99% of the American population (for Anna) vs. the top 1% (for Beatrice). But this is not a real disagreement: this situation is analogous to Anna and Beatrice’s discussion about what is next on the menu in (3). The other way of understanding these dialogues is one in which Anna and Beatrice are both implicitly comparing Carla to the same comparison classes but nevertheless have different assessments about whether she counts as rich, thin, heavy, and so forth. This constitutes a true case of faultless disagreement, so we must conclude that vague predicates in general are subjective, not just vague predicates of personal taste. In particular, even vague predicates which are used to describe (in principle) externally observable and measurable features of an object — such as its wealth (rich), its weight (heavy), its age (old), its height (tall), and so forth—can be subjective. However, when we look at acceptability under find, we see that there is a subtle contrast between “dimensional” vague predicates such as the ones in (8a) and predicates of personal taste and other “evaluative” vague predicates. (These categories come from Bierwisch, 1989.) Consider the latter group first. (9a) illustrates the acceptability of taste predicates under find, which we have already discussed; (9b) shows that other kinds of evaluative predicates are also acceptable here. (9)

a. b.

Anna finds her bowl of pasta tasty/delicious/disgusting. Anna finds Carla stimulating/annoying/boring/tedious.

110 | Christopher Kennedy

In contrast, similar examples with dimensional vague predicates are odd:¹ (10)

a. ??Anna finds her bowl of pasta big/large/small/cold. b. ??Anna finds Carla rich/heavy/old/tall.

This contrast is a subtle one, but a systematic examination of a range of data supports the conclusion that there is a real distinction here. First, these examples can all be improved by adding an adverb that is derived from an unquestionably subjective adjective: (11)

a. b.

Anna finds her bowl of pasta {surprisingly, remarkably, unusually} big/large/small/cold. Anna finds Carla {annoyingly, disgustingly, irritatingly} rich/heavy/ old/tall.

This makes sense if the adverb makes the whole predicate subjective (in the relevant way), which in turn implies that the predicate without the adverb is not subjective (in the relevant way). Second, if an adjective has both a dimensional and an evaluative sense, find disambiguates to the latter. Consider adjectives like heavy, light and dense, used to describe a piece of cake. In the sentences in (12a), these adjectives can be understood in two distinct ways: as describing objective properties of the cake (its weight or density), or as describing a subjective assessment of the quality of the cake, made in virtue of the experience of tasting it. (12)

a. b.

This piece of cake is heavy/light/dense. I find this piece of cake heavy/light/dense.

The examples in (12b), however, have only the latter reading: these sentences could be felicitously uttered after tasting the cake, but they would be very odd ways to report measurements of its physical properties. This is made particularly clear by the contrast in (13): the use of the measure phrase in (13b) blocks the

1 Two comments. First, note that there is a different sense of find meaning ‘discover’ which is acceptable when the embedded predicate is stage level: When she returned to the table, Anna found her bowl of pasta cold/??big. Because this sense of find is an achievement verb, while the one I am interested in is stative, I keep my example sentences in the simple present form in order to filter it out. Second, as we will see below, some dimensional predicates can also have evaluative meanings, and to the extent that the examples in (10) can be understood in this way, they are fine, but unambigously evaluative. We will examine clear cases of this distinction below.

Two kinds of subjectivity



subjective understanding that is available in (13a), and the resulting sentence is anomalous. (13)

a. I find this frosting thick. b. ??I find this frosting 2cm thick.

Finally, and most significantly, the examples in (10) can all be made perfectly acceptable by replacing the verb find with the verb consider, as in (14). (14)

a. b.

Anna considers her bowl of pasta big/large/small/cold. Anna considers Carla rich/thin/heavy/old/young/short.

The following pair illustrates this context in a particularly clear way. In a situation in which a passenger is checking his luggage for a flight, the attendant could report the fact that his bag exceeds the standard weight allowance by uttering (15a), but not by uttering (15b). (15)


I’m sorry, sir, but the airline considers this bag heavy. You will have to pay an extra baggage fee. b. ??I’m sorry, sir, but the airline finds this bag heavy. You will have to pay an extra baggage fee.

Similarly, in the discourse in (16), Anna can use consider to report her subjective assessment of whether Carla counts as tall in the context (one in which the vague standard is based on the needs of the team), but the corresponding sentence with find sounds strange. (16)

a. Anna: We need a tall woman to play center on the basketball team. b. Beatrice: What about Carla? Is she tall? c. Anna: I don’t consider her tall. d. ??Anna: I don’t find her tall.

Syntactically, consider and find have almost identical distributions, and in constructions in which they are fully interchangeable, they appear to have quite similar semantic affects: (17a-b) sound synonymous, and in particular both are understood in a way that relativizes the embedded predicate to the surface subject. (17)

a. b.

Anna finds the pasta tasty / beautifully presented. Anna considers the pasta tasty / beautifully presented.

However, there is a crucial difference between find and consider: like think and believe, consider does not require its complement predicate to be subjective (though it allows for subjectivity, like think and believe):

112 | Christopher Kennedy


a. b.

Homer considers/??finds himself gay. Homer considers/??finds trippa alla romana to be a vegetarian dish.

These facts indicate that despite the similarity in meaning between the constructions in which find and consider occur, only the former is a “radical judge shifter” in Sæbø’s sense.² We may then conclude, based on the systematic difference in acceptability in dimensional and evaluative vague predicates under find and consider, that although the kind of subjectivity manifested by these predicates can give rise to faultless disagreement, it is not linguistically encoded in the same way.³ This conclusion can be both strengthened and refined by examining a second set of facts, which highlight another difference between subjectivity in dimensional predicates and subjectivity in evaluative predicates. We have already seen that both classes of predicates give rise to faultless disagreement effects in their vague, positive forms. When we turn to forms that are not vague, such as comparatives, we see that evaluative predicates still give rise to faultless disagreement:

2 That is, only find has a denotation such that the sole function of the expression is to fix the judge. Consider, like other attitude verbs, can fix the judge of a subjective predicate embedded in its complement, but it does not have a meaning that requires its complement to introduce subjectivity. 3 There is, evidently, some degree of cross-linguistic variation here, or perhaps more likely, different subjective attitude verbs in different languages have slightly different properties. Sæbø (2009) shows that in Norwegian, the subjective attitude verb synes (which he glosses as ‘seem’) is incompatible with nonsubjective predicates, but is nevertheless acceptable with vague dimensional predicates, as shown in (i). (Here I follow Sæbø in just writing subjective attitude verb in the free translation, to indicate the difficulty of providing true translations of these constructions.) (i)

a. ??Mange forskere synes at dinosaurene ble utryddet av et voldsomt Many researchers seem that dinosaurs were extinguished by a violent kometnedslag for 65 millioner år siden. comet.impact for 65 million years since ‘Many scientists subjective attitude verb that the dinosaurs were extinguished by a major comet impact 65 million years ago.’ b. De synes det er langt til lege. they seem it is far to doctor ‘They subjective attitude verb it is far to the doctor.’

On the other hand, it may be the case that the dimensional predicates that Sæbø considers in his examples are like the ones discussed above in having both an objective and a subjective sense, with only the latter available in (ib). (The experience of distance can certainly be subjective; see the English examples with long below.) This should be a point for future investigation.

Two kinds of subjectivity

| 113


a. b.

Anna: The tripe is tastier than the haggis. Beatrice: No, the haggis is tastier than the tripe.


a. b.

Anna: Skiing is the most fun! Beatrice: No, skating is the most fun!


a. b.

Anna: Carla is more stimulating/annoying/boring/tedious than David. Beatrice: No, David is more stimulating/annoying/boring/tedious than Carla.

Comparative forms of dimensional predicates, however, do not give rise to faultless disagreement: in the following dialogues, it is clear that one of Anna or Beatrice is right and the other is wrong. (22)

a. b.

Anna: The tripe is colder than the haggis. Beatrice: No, the haggis is colder than the tripe.


a. b.

Anna: Skiing is the most expensive! Beatrice: No, skating is the most expensive!


a. b.

Anna: Carla is richer/taller/heavier/older than David. Beatrice: No, David is richer/taller/heavier/older than Carla.

The same pattern appears in interactions with subjective attitude verbs. Comparative forms of evaluative predicates can be embedded under find, but comparative forms of dimensional predicates cannot be (cf. Sæbø, 2009): (25)

a. b. c.

Anna finds the tripe tastier than the haggis. Beatrice finds skating the most fun. Anna finds Carla is more stimulating/annoying/boring/tedious than David.


a. ??Anna finds the tripe colder than the haggis. b. ??Beatrice finds skating the most expensive. c. ??Anna finds Carla richer/taller/heavier/older than David.

Turning to adjectives that have both evaluative and dimensional senses, we see that this polysemy is retained in the comparative form: the disagreement between Anna and Beatrice in (27) can be understood either as a disagreement about their quantitative measurements of the cakes’ weight or density (the dimensional senses of the adjectives), or as an argument about their qualitative assessments of the cakes’ imprint on their taste/digestion (the evaluative senses of the adjectives).

114 | Christopher Kennedy


a. b.

Anna: This cake is heavier/lighter/denser than that one. Beatrice: No, that cake is heavier/lighter/denser than this one.

However, this disagreement is faultless only on the latter understanding, for example if Anna and Beatrice are food critics. If instead they are food scientists reporting on a set of culinary experiments, then their disagreement is not faultless: one of them is right and the other is wrong, because her measurements were off, or she misread her instruments, or whatever. Similarly, if we report Anna’s utterance in (27a) with (28a), it is ambiguous whether we are reporting on her taste experience (as a food critic) or on her measurements (as a food scientist). (28b), on the other hand, can only be understood as a description of Anna’s subjective taste experience. (28)

a. b.

Anna thinks that this cake is heavier/lighter/denser than that one. Anna finds this cake heavier/lighter/denser than that one.

This kind of polysemy between an evaluative/qualitative/subjective sense and a dimensional/quantitative/objective sense appears in other classes of predicates as well (i.e., not only predicates that can be used to describe aspects of taste experiences), with similar results. Consider, for example, the dialogue in (29), uttered in a context in which it is an objective fact that the flight from Chicago to Tokyo takes 13 hours and 5 minutes, while the flight from Chicago to Hong Kong takes 15 hours and 40 minutes. (29)

a. b.

Anna: The flight from Chicago to Hong Kong is longer than the one from Chicago to Tokyo. Beatrice: No, the flight from Chicago to Tokyo is longer than the one from Chicago to Hong Kong.

There is one reading of (29) in which Anna is right and Beatrice is wrong. There is, however, a second reading in which their disagreement is faultless, but the disagreement has to do with their subjective experiences of the flight time, rather than with the objective durations of the flights. On the latter, subjective reading, Beatrice could justify her claim in virtue of the fact that she has to fly coach from Chicago to Tokyo, but gets to fly first class from Chicago to Hong Kong, and we could report her view using either (30a) or (30b). On the former, objective reading, in which Beatrice is wrong about the objective difference in flight time, only (30a) is an appropriate description of her mental state.

Two kinds of subjectivity


a. b.

| 115

Beatrice thinks that the flight from Chicago to Hong Kong is longer than the flight from Chicago to Tokyo. Beatrice finds the flight from Chicago to Hong Kong longer than the flight from Chicago to Tokyo.

The facts that we have examined so far support two generalizations. First, vague, positive form scalar predicates give rise to faultless disagreement effects across the board, but only evaluative positive form scalar predicates are acceptable under find. Second, only evaluative scalar predicates in the comparative form (which is not vague; see Kennedy, 2011) show faultless disagreement effects and acceptability under find; comparative forms of dimensional scalar predicates do not. The conclusion to be drawn from these generalizations is that the kind of subjectivity associated with vagueness—subjectivity about whether or not an object meets a standard for satisfaction of the predicate—must be distinguished from the kind of subjectivity associated with evaluativity—subjectivity about an assessment of an object’s qualities—because only the latter licenses embedding under subjective attitude verbs. In particular, the facts suggest that the component of meaning that is responsible for making a predicate one that expresses a qualitative assessment has a representational, type-theoretic status that the component of meaning that is responsible for making a predicate one that expresses a relation to a vague standard does not. Let us now see how this observation fits in to current thinking about vagueness and evaluativity.

3 Vagueness Richard’s (2004) observation that vague predicates in general give rise to faultless disagreement effects is, from one perspective, unsurprising, as a number of researchers have argued for an essentially relativistic semantics for vague predicates. For example, Bogusławski (1975) proposes that a vague scalar predicate is true of an object just in case the value of the object on the relevant scalar continuum is “conspicuous”, “noteworthy” or “sufficient to attract attention”. Similarly, Fara (2000) claims that an object satisfies a vague scalar predicate just in case the degree to which it manifests the relevant scalar property is “significant”, and uses this semantic analysis to explain the phenomenological properties of vague predicates. Both of these characterizations imply that the interpretation of a vague

116 | Christopher Kennedy

form is relativized to some agent: the entity relative to whom conspicuousness, noteworthiness, or significance is assessed.⁴ Moreover, given common assumptions about the compositional semantics of gradable predicates, the fact that the positive form of adjectives like rich, tall and old gives rise to faultless disagreement while the comparative form does not, is also unsurprising. This is because a core semantic difference between the positive and comparative forms—one that must be captured by any empirically adequate semantic analysis of gradable predicates—is that the latter lacks whatever semantic (or pragmatic) features give rise to the vagueness of the former, and simply expresses an asymmetric ordering relation. In a Fara-style analysis, for example, tall expresses the interest-relative, vague property of having a degree of height that significantly exceeds some threshold (for a comparison class); taller than David, on the other hand, expresses the non-relativistic, non-vague property of having a degree of height that exceeds David’s degree of height. In many analyses of gradable adjectives and their various forms, this difference between positive and comparative is accounted for by hypothesizing that the adjectival root does not express a property on its own, but instead just introduces a mapping between individuals and scalar values, or degrees.⁵ This basic idea is implemented in different ways in the literature; since these differences are not relevant to the questions I am asking here, I will assume without argument the implementation in Kennedy (1999), in which gradable adjectives denote functions from objects to degrees: they combine with an individual and return a measure of how much that individual manifests a particular type of scalar value, such as height, age or wealth, as shown in (31). (31)

a. b. c.

[[tall]] = height⟨e,d⟩ [[old]] = age⟨e,d⟩ [[rich]] = wealth⟨e,d⟩

4 This position has been called into question by Stanley (2003) on the grounds that we can have beliefs about the truth or falsity of a sentence like Mt. Everest is tall without having beliefs about any agent relative to whom Mt. Everest’s height is supposed to be conspicuous, noteworthy or significant. 5 Every semantic analysis of positive and comparative gradable predicates must include some way of deriving the difference in vagueness between the two forms. The approach that I present here is one way of doing so, but there are others which maintain the view that adjectives (both positive and comparative) denote properties (see for example Wheeler, 1972; Kamp, 1975; Klein, 1980; van Rooij, 2011). I adopt a degree-based, decompositional semantics of gradable adjectives here mainly because it allows for a transparent characterization of the difference between positive and comparative adjectival predicates, but I suspect that my central claims could be restated in terms of a different set of initial assumptions.

Two kinds of subjectivity

| 117

Gradable adjectives are converted into properties through composition with degree morphology, which in turn gives rise to the observed differences in meaning between different forms of the adjective. Among the inventory of degree morphology in English is the comparative morpheme comp (realized as the degree word more or the suffix -er) and, by hypothesis, a phonologically null positive form morpheme pos. It is the difference in meaning between these two morphemes that derives the differences in meaning between the positive and comparative forms, and which can in turn be used to explain their difference in faultless disagreement contexts. The comparative morpheme maps the adjective onto a property that is true of an object iff its degree on the relevant scale exceeds the degree introduced by the than-constituent, as given in (32). (32)

[[comp]] = λg⟨e,d⟩ λd than λx.g(x) ≻ d than

The positive morpheme, on the other hand, maps the adjective onto a property that is true of an object iff its degree on the relevant scale is above a contextually appropriate threshold or “standard”, e.g. one that represents the minimum value required to have a significant or noteworthy degree of the relevant property in the context of utterance. Its denotation is given in (33), where stnd(g) represents “the standard appropriate for the kind of measurement encoded by g,” which, for the sake of this discussion, I am assuming to be computed in a Fara-style way. (33)

[[pos]] = λg⟨e,d⟩ λx.g(x) ⪰ stnd(g)

The denotations of the comparative and positive forms of the adjectival root tall, according to this analysis, are as shown in (34). (34)

a. b.

[[comp]]([[tall]]) = λd than λx.height(x) ≻ d than [[pos]]([[tall]]) = λx.height(x) ⪰ stnd(g)

The comparative denotes a property that is true of an object just in case its height exceeds the degree introduced by the than-constituent. This is a precise property, and is moreover fully objective, since whether it holds of an object or not is fully determined by facts about that object’s height. The positive, on the other hand, denotes the property of having a height that exceeds a standard of significance or noteworthiness, which is both vague (Fara, 2000), and certainly subjective, since whether it holds of an object depends not only on that object’s height, but also on some subjective assessment of significance (Richard, 2004). With these semantic assumptions in hand, we predict the following pattern: positive form adjectives should be subjective; comparative form adjectives should not be. If we restrict our empirical focus to dimensional adjectives and our diag-

118 | Christopher Kennedy

nostics to faultless disagreement, this is indeed what we see, but once we broaden our empirical domain to include evaluative adjectives, and add subjective attitude verbs to our set of diagnostics, the picture becomes more complex. Table 1 summarizes the pattern we observed in the previous section. Table 1. Adjectives and Subjectivity

dimensional evaluative





What this table makes clear is that vagueness (i.e., positive form semantics) is a sufficient condition for faultless disagreement (=FD) effects with scalar predicates, but not a necessary one. Vagueness is neither a necessary nor a sufficient condition for embedding under find (=FIND), however; in fact, vagueness does not correlate with acceptability under find at all. Instead, whether a scalar predicate is acceptable under find correlates with the evaluative/dimensional distinction: based on the data we have seen, evaluativity is both a necessary and a sufficient condition for embedding under find. The behavior of comparative forms of evaluative adjectives shows moreover that evaluativity is a sufficient condition for faultless disagreement. One conclusion to be drawn from these facts, if Sæbø (2009) is correct that a predicate is acceptable as the complement of find just in case it semantically selects for a judge (i.e., if its semantic type is such that it requires saturation by an individual-denoting expression which is interpreted as the source of subjective assessment), is that the kind of subjectivity introduced by vagueness is not one that is reflected in semantic type. Putting things in terms of the compositional assumptions introduced above: the judge argument is not introduced by pos, but rather by the adjective, since the crucial factor determining acceptability under find is not positive vs. comparative (vague vs. not vague), a distinction that comes from the functional elements, but rather evaluative vs. dimensional, a distinction that comes from the lexical items.⁶ There are a number of plausible ways of explaining and analyzing the subjectivity of vagueness: relativistically, in virtue of a particular semantics of the posi-

6 Sæbø actually draws the opposite conclusion, based on facts like those discussed in footnote 3; what we need to do now is reassess the examples he discusses with the evaluative/qualitative vs. dimensional/quantitative distinction in mind.

Two kinds of subjectivity

| 119

tive form (Richard, 2004); at a more discourse/pragmatic level, in virtue of the dynamics and uncertainty of vague standards (Barker, 2002); or in some other way which does not have the same type-theoretic consequences as the kind of subjectivity that we see with (both positive and comparative) evaluative predicates. Identifying the best model of this kind of subjectivity is not something I will discuss here (see Kennedy, 2013); instead, in this paper I want to focus on the question of how to model the kind of subjectivity manifested by evaluative predicates. We have seen that evaluative predicates must be distinguished from dimensional ones, in both their positive and comparative forms, in a way that licenses embedding under find. Following Sæbø, I take this distinction to be a type-theoretic one; the semantic question that now needs to be answered is what this distinction corresponds to in terms of meaning.⁷ In the next section, I will propose an answer to this question that builds on previous work by Kennedy & McNally (2010) on the semantics of color adjectives.

4 Evaluativity as qualitative assessment One of the conclusions of the previous section, based on the interaction of adjective class and acceptability of embedding under the subjective attitude verb find, is that the dimensional/evaluative distinction is a type-theoretic one. The most straightforward way of capturing this distinction is to say that it is lexical, and indeed, this is the route taken in Bierwisch (1989), the most comprehensive discussion of the dimensional/evaluative distinction in the literature, but unfortunately not one that will help to explain the patterns we have seen in this paper. Specifically, Bierwisch proposes that the difference between dimensional and evaluative adjectives is that the former lexically encode degree functions, much in the way described in the previous section, while the latter are underlyingly “regular” (type ⟨e, t⟩) properties. Their gradable meanings are derived using a special mapping function that turns them into degree predicates, and which accounts for various entailment patterns that Bierwisch wants to explain, but which does not introduce anything like a judge argument, and so does not give us a way of explaining the different pattern of acceptability of dimensional and evaluative adjectives under find.

7 Actually, it is probably not crucial that the distinction is type-theoretic: any formal distinction that provides the basis for distributional constraints would be sufficient. To keep things (probably overly) simple, though, I will assume with Sæbø that the distinction is a type-theoretic one.

120 | Christopher Kennedy

Instead, what we would like to be able to say is something like the following. Dimensional adjectives are regular type ⟨e, d⟩ measure functions of the sort listed in (31), which combine with degree morphology to derive various kinds of properties as outlined above. Evaluative adjectives, on the other hand, denote judge-dependent, “subjective measure functions” and have the semantic type ⟨e, ⟨e, d⟩⟩. However, merely saying that the two classes of adjectives differ in semantic type in this way is not enough; we also need to say what the difference in actual meaning is. In other words, we need to make some sense of what it means to say that an evaluative adjective denotes a “subjective measure function”. As a starting point, let us take a look at adjectives like salty, sugary, watery, and so forth, which are similar to heavy and dense in that they have both quantitative senses and qualitative ones.⁸ (35), for example, can be understood either as in (35a) or as in (35b). (I focus on comparatives here to eliminate any role of vagueness, but the same kinds of patterns hold for the non-comparative forms as well.) (35)

This dish is saltier than that one. a. b.

This dish contains more salt than that one. This dish has a more (subjectively) salty quality than that one

These readings are truth-conditionally distinct. If the dishes in question contain different amounts of salt, but for some reason the salt taste is undetectable, then (35) is true on the (a) reading and false on the (b) reading. If the dishes actually contain no salt, but one tastes saltier than the other for some reason, then (35) can be true on the (b) reading and false on the (a) reading. Furthermore, disagreement dialogues and find distinguish the two readings in the expected way. Anna and Beatrice’s disagreement in (36) is faultless only on the qualitative understanding of saltier, and only the qualitative understanding is present under find in (37). (36)

a. b.

Anna: This dish is saltier than that one. Beatrice: No, that dish is saltier than this one.


I find this dish saltier than that one (even though I know that it contains less salt).

8 In the discussion of heavy and dense in Section 2, I referred to their dimensional senses as “quantitative” and their evaluative senses as “qualitative”. I think that the quantitative/qualitative distinction is the more general one, of which the dimensional/evaluative distinction is a subtype, for reasons that will become clear as we move forward.

Two kinds of subjectivity

| 121

Adjectives of tastes (which are different from adjectives of personal taste), such as salty, sugary, and so forth, are derived from nouns, and in this respect are similar to another class of adjectives which show a quantitative/qualitative ambiguity: color words (Kennedy & McNally, 2010). For example, (38) can be understood either quantitatively, as in (38a), or qualitatively, as in (38b), and the two readings can be teased apart through the use of different modifiers, as shown in (39). (38)


This leaf is greener than that one. a. b.

More of this leaf is green than that one. This leaf is qualitatively closer to “pure green” than that one.

a. b.

The leaf is completely green. The leaf is perfectly green.

quantitative qualitative

Adapting Bierwisch’s (1989) analysis of evaluative adjectives, Kennedy & McNally account for this ambiguity by hypothesizing that the basic meaning of a color word is the nominal one, which they take to be the name of a color (a kind), and that there are two ways of deriving a meaning as a gradable adjective.⁹ One way involves mapping the nominal meaning to an adjective meaning that measures the quantity of (the relevant) color manifested by the object; the other way is to map the nominal meaning to an adjective meaning that measures the quality of (the relevant) color manifested by the object. Kennedy & McNally give a decompositional semantics which makes use of two primitive functions quant and qual which, when combined with a color noun meaning, derive the respective qualitative and quantitative adjective meanings, as in (40). (40)

a. b.


[[[N green]]] = green, the name of a kind [[[Aquant green]]] = λx.quant(green)(x): a function from an individual x to a degree which represents the quantity of green manifested by x [[[Aqual green]]] = λx.qual(green)(x): a function from an individual x to a degree which represents the quality of green manifested by x

I would like to suggest that predicates of tastes—and predicates that involve qualitative assessment more generally—should be analyzed in basically the same way, with one crucial modification: expressions whose meanings are defined in terms of qual have an extra argument that corresponds to the source of qualitative as-

9 Kennedy & McNally also introduce a third, non-gradable adjective meaning for color words, which is not relevant for the current discussion. See Hansen (2011) for detailed assessment of Kennedy & McNally’s proposals.

122 | Christopher Kennedy

sessment, which I will represent as an index on the qual function. The two senses of salty can then be characterized formally as in (41), where salt is the denotation of the noun salt (a kind). (41)

a. b.

[[[Aquant salty]]] = λx.quant(salt)(x) [[[Aqual salty]]] = λxλy.qual y (salt)(x)

(41a) is a measure function that maps its argument x onto a degree that represents the quantity of salt manifested by x: how much salt it contains. (41b), on the other hand, is a “subjective measure function” which maps its argument x onto a degree that represents the quality of salt manifested by x according to y—it gives a measure of y’s subjective assessment of x in relation to salt. The intuition underlying this proposal is that qualitative measurements are distinct from quantitative ones in being relativized to a source of qualitative assessment, and that this difference is reflected in the argument structure of the predicates used to make the relevant kinds of measurements. In particular, this distinction is reflected type-theoretically in a way which, following Sæbø (2009), can explain the difference in acceptability of qualitative vs. quantitative meanings under find: qualitative meanings of adjectives like salty have a judge argument which can be saturated by the surface subject of find; quantitative meanings of the same adjectives do not.¹⁰ Adjectives of tastes like salty, sugary, watery have both qualitative and quantitative interpretations, as do adjectives based on dimensional concepts like weight

10 An apparent problem for this proposal—or at least for the specific hypothesis that the qualitative/quantitative ambiguity that I have identified with the evaluative/dimensional distinction is the same as the qualitative/quantitative ambiguity that Kennedy & McNally (2010) posit for color words—is that color words are not embeddable under find, even when understood qualitatively: (i)

a. ??I find this object green/blue/red/yellow. b. I consider this object green/blue/red/yellow.

I would like to suggest that the problem with (ia) is not that the qualitative meaning of color words should be analyzed in some other way, but rather that color words are special in that their judge arguments are lexically saturated by an argument that has the semantic consequence of making the source of qualitative assessment the “normal observer;” in Lasersohn’s (2005) terms, they are obligatorily “exocentric”. This can be viewed as a consequence of the fact that although color can be manifested either quantitatively or qualitatively, the qualitative experience of color is something that is taken to be shared across individuals, rather than subject to individual variation in the way that experiences of taste, etc. are. Color adjectives do give rise to faultless disagreement patterns, as discussed in Hansen (2011), but this is arguably due to the fact that they are vague, albeit possibly in ways that involve more multidimensionality than the kinds of vague predicates discussed in Section 2.

Two kinds of subjectivity



(heavy), elapsed time (long) and density (dense), as we have seen. There are, I suspect, many other adjectives that pattern in the same way; adjectives based on substance terms, such as plastic, metallic, wooden, icy, and so forth, provide one class of examples. At the same time, there are some adjectives which appear to have only quantitative readings (e.g. tall and rich), and many which appear to have only qualitative meanings (e.g. predicates of personal taste). This could indicate that these adjectives lexicalize one meaning but not the other, but it may also simply indicate that there are concepts which are naturally mapped to both quantitative or qualitative measurements, while others are naturally mapped only to one kind of measurement or the other. For example, it is natural to talk about a subjective experience of the salt manifested by a particular bowl of soup, but a bit odd (though maybe not impossible) to talk about the subjective experience of the height manifested by a particular individual. Likewise, it is natural to talk about an objective measurement of the quantity of salt in a bowl of soup, absent any subjective experience of it, but it is not so natural to talk about the objective quantity of taste in the soup, without any subjective experience of it. In any case, I take the existence of a large class of adjectives that display a quantitative/qualitative ambiguity, and the fact that the two readings give rise to different patterns of distribution under find (as well as to other distributional differences of the sort observed by Kennedy & McNally, 2010), as evidence for the core hypothesis that the analysis presented here aims to capture: that adjective meanings can be broken down into a basic concept term that provides some dimension of measurement, plus either a quantitative or (relativized) qualitative way of measuring an object relative to that dimension. Whether the particular implementation of this hypothesis presented above is the best way to capture the qualitative/quantitative distinction remains to be seen. I myself doubt that it is, and suspect that a more explanatory account will emerge out of careful thought about the basic elements of predicate meaning, from a more sophisticated examination of the multidimensionality of qualitative assessment (as in Umbach, 2016, for example) [this volume], and also by a close examination of languages in which predicate expressions show more grammatical complexity than they do in English (see, for example, Koontz-Garboden & Francez, 2010). That said, the current proposal is at least based on a pattern of linguistic behavior that has some generality, and so hopefully represents a first step towards a more satisfying and explanatory analysis.

124 | Christopher Kennedy

5 Conclusion In this paper, I have compared different classes of scalar predicates relative to two diagnostics of subjectivity: faultless disagreement effects and acceptability under the subjective attitude verb find. We have seen that these diagnostics provide the basis for distinguishing two kinds of subjectivity: the kind of subjectivity manifested by vague predicates, which triggers faultless disagreement but does not ensure embeddability under find, and the kind of subjectivity manifested by evaluative predicates, which ensures embeddability under find and also triggers faultless disagreement, independent of vagueness. Following Sæbø (2009), I have assumed that acceptability under find is a function of semantic type, and I proposed a lexical semantic analysis of evaluativity as judge-dependent qualitative assessment that builds on Kennedy & McNally’s (2010) analysis of quantitative vs. qualitative meanings of color adjectives. This means that the analysis is committed to a contextualist semantics for the class of predicates which involve qualitative assessment, which, according to the diagnostics discussed in this paper, include predicates of personal taste. This is a conclusion already reached by Sæbø (2009) (who advocates in particular the contextualist analysis proposed in Stojanovic, 2007). An open question, then, is how to reconcile a contextualist semantics for predicates of personal taste and other predicates of qualitative assessment, which is justified on a purely grammatical basis by the interaction with subjective find, with the fact that they can be used in dialogues that appear to express genuine disagreement. (Recall from Section 1 that the problem that the faultless disagreement pattern presents for a contextualist semantics is not faultlessness, but disagreement.) Unfortunately, this is not a question that I can satisfactorily answer at this time, but let me finish this paper with a few thoughts about it. The argument against contextualism based on disagreement crucially relies on the assumption that the presence of a genuine disagreement (faultless or otherwise) in a linguistic exchange entails that the speakers engaged in the exchange are using sentences that have the same (relevant) semantic content, something that is not the case if the properties expressed by qualitative expressions vary with the judge. However, this “shared content” assumption has been challenged in recent work by Sundell (2010) and Plunkett & Sundell (2013), who show that genuine disagreement can arise even in exchanges that involve sentences with demonstrably distinct semantic content (e.g. disagreements that involve implicatures). If genuine disagreement does not in fact entail shared content, then one of the central arguments against a contextualist account of predicates of personal taste disappears. (That said, the contextualist analysis still needs to be supplemented

Two kinds of subjectivity



with an account of e.g. the difference between “bare” taste predicates like tasty, which give rise to disagreement, and predicates with explicit judges, like tasty to me, which do not.) A final word on faultless disagreement with vague dimensional adjectives: if this is due to some aspect of the meaning of the positive form, (e.g. its interest relative semantics, if Fara’s (2000) analysis of vagueness is correct), then given that this aspect of meaning is absent from the comparative form of evaluative adjectives, but such adjectives still give rise to faultless disagreement, we must conclude that faultless disagreement can have more than one source. This would further entail that we cannot conclude from the observation that a particular expression E shows faultless disagreement effects that it has some semantic/pragmatic feature F (e.g. a relativistic meaning), because the same effects could be due to some other feature F 󸀠 (e.g. type-theoretic subjectivity). On the other hand, it may be the case that faultless disagreement in these two cases is due not to the semantics of the positive form of a dimensional adjective, nor to the type-theoretic subjectivity of a comparative evaluative adjective, but rather to some more general, non-representational property of meaning that both expressions share. In order to answer this question, we need to think more about the ways that the two classes of expressions are similar, and about the features they share with other kinds of expressions that give rise to faultless disagreement. Acknowledgment: I would like to express my thanks to the organizers of the 2010 DGfS workshop “Subjective meaning: Alternatives to relativism”, Janneke van Wijnbergen-Huitink and Cécile Meier, for stimulating me to think about this material, and to the participants in the workshop for their feedback on its first presentation. Since then, I have received very valuable comments and criticism from Lisa Bylinina, Max Kölbel, Kjell Johan Sæbø, and Carla Umbach, for which I am also very grateful.

References Barker, Chris. 2002. The dynamics of vagueness. Linguistics and Philosophy 25(1). 1–36. Bierwisch, Manfred. 1989. The semantics of gradation. In Manfred Bierwisch & Ewald Lang (eds.), Dimensional adjectives, 71–261. Berlin: Springer. Bogusławski, Andrzej. 1975. Measures are measures: In defence of the diversity of comparatives and positives. Linguistiche Berichte 36. 1–9. Fara, Delia Graff. 2000. Shifting sands: An interest-relative theory of vagueness. Philosophical Topics 20. 45–81. Glanzberg, Michael. 2007. Context, content, and relativism. Philosophical Studies 136. 1–29.

126 | Christopher Kennedy

Hansen, Nat. 2011. Color adjectives and radical contextualism. Linguistics and Philosophy 34(3). 201–221. Kamp, Hans. 1975. Two theories of adjectives. In Edward Keenan (ed.), Formal semantics of natural language, 123–155. Cambridge: Cambridge University Press. Reprinted in: Klaus von Heusinger & Alice ter Meulen (eds.), The dynamics of meaning and interpretation. Selected papers of Hans Kamp, 225–261. Leiden: Brill, 2013. Kennedy, Christopher. 1999. Projecting the adjective: The syntax and semantics of gradability and comparison. New York: Garland. Kennedy, Christopher. 2011. Vagueness and comparison. In Paul Egré & Nathan Klinedinst (eds.), Vagueness and language use, 73–97. Basingstoke: Palgrave Macmillan. Kennedy, Christopher. 2013. Two sources of subjectivity: Qualitative assessment and dimensional uncertainty. Inquiry: An Interdisciplinary Journal of Philosophy 56. 258–277. Kennedy, Christopher & Louise McNally. 2010. Color, context and compositionality. Synthese 174. 79–98. Klein, Ewan. 1980. A semantics for positive and comparative adjectives. Linguistics and Philosophy 4. 1–45. Kölbel, Max. 2002. Faultless disagreement. Proceedings of the Aristotelian Society 104. 55–73. Kölbel, Max. 2009. The evidence for relativism. Synthese 166. 375–395. Koontz-Garboden, Andrew & Itamar Francez. 2010. Possessed properties in Ulwa. Natural Language Semantics 2. 197–240. Lasersohn, Peter. 2005. Context dependence, disagreement, and predicates of personal taste. Linguistics and Philosophy 28(6). 643–686. Lasersohn, Peter. 2009. Relative truth, speaker commitment, and control of implicit arguments. Synthese 166. 359–374. Plunkett, David & Timothy Sundell. 2013. Disagreement and the semantics of normative and ethical terms. Philosopher’s Imprint 13(23). 1–37. Ms., UCLA and University of Kentucky. Richard, Mark. 2004. Contextualism and relativism. Philosophical Studies 119. 215–242. van Rooij, Robert. 2011. Implicit versus explicit comparatives. In Paul Egré & Nathan Klinedinst (eds.), Vagueness and language use, 51–72. Basingstoke: Palgrave MacMillan. Sæbø, Kjell Johan. 2009. Judgment ascriptions. Linguistics and Philosophy 32(4). 327–352. Stanley, Jason. 2003. Context, interest relativity, and the Sorites. Analysis 63(4). 269–280. Stephenson, Tamina. 2007. Towards a theory of subjective meaning. Cambridge, MA: MIT dissertation. Stojanovic, Isadora. 2007. Talking about taste: Disagreement, implicit arguments, and relative truth. Linguistics and Philosophy 30(6). 691–706. Sundell, Timothy. 2010. Disagreements about taste. Philosophical Studies 155(2). 267–288. Umbach, Carla. 2016. Evaluative propositions and subjective judgments. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 127–168. Berlin: De Gruyter. [This volume]. Wheeler, Samuel. 1972. Attributives and their modifiers. Noûs 6(4). 310–334.

Carla Umbach

Evaluative propositions and subjective judgments Abstract: This paper addresses evaluative predicates like tasty and beautiful. According to the prevailing accounts of evaluative predicates their meaning is relative to a judge/experiencer giving rise to so-called faultless disagreement. Starting from the puzzle of the competent speaker, it is argued in this paper that disagreement about matters of taste is genuine disagreement. An analysis is provided separating semantic and pragmatic aspects of evaluativity. Semantically, evaluativity is traced back to two different sources, (i) the possibility of metalinguistic usage negotiating the standard for satisfaction of the predicate and (ii) the use of multiple criteria in order to compensate for the lack of a fixed denotation. Metalinguistic usage is detailed along the lines of Barker (2002). The idea of evaluative predicates making use of multiple context-dependent criteria while relating to a constant meaning component of, e.g. commending, has been suggested by Hare (1952). Pragmatically, the analysis starts from the observation that the attitude verb finden ‘find’ in German can be used to block denial. While unembedded evaluative propositions license genuine denial, when occurring as a complement of first person finden they are presented as mere subjective judgments allowing only indirect forms of expressing disagreement. Pragmatics will be spelled out making use of the notion of individual discourse commitments in Farkas & Bruce (2010).

1 Introduction It has been claimed by various authors that predicates like tasty and fun give rise to so-called “faultless disagreement” such that participants in a dialogue assert contradictory propositions without one of them being wrong. In the dialogue in (1) Ann asserts that licorice is tasty and Ben denies her assertion claiming that it tastes terrible. Although the assertions of Ann and Ben are clearly contradictory, their disagreement in (1) appears less severe than their disagreement in (2), because in (2) either Ann or Ben must be wrong while in (1) it is intuitively possible that both are in some sense right. This intuition is commonly described as faultless disagreement. Carla Umbach, Zentrum für Allgemeine Sprachwissenschaft (ZAS), Schützenstr. 18, D-10117 Berlin, Germany, e-mail: [email protected]

128 | Carla Umbach


a. b.


a. b.

Ann: Lakritze ist lecker. ‘Licorice is tasty.’ Ben: Nein! Lakritze schmeckt eklig. ‘No, it isn’t! It tastes terrible.’ Ann: Osnabrück liegt in Dänemark. ‘Osnabrück is in Denmark.’ Ben: Nein! Es liegt in Deutschland. ‘No, it isn’t! It’s in Germany.’

The intuition of faultless disagreement in cases like (1) is usually accounted for in semantics by assuming that predicates of personal taste introduce a judge parameter determining the truth value of the propositions they occur in. The judge mostly corresponds to the speaker, e.g. Ann and Ben in (1), and is implemented either as an implicit argument provided by the context, or as an evaluation parameter in addition to world and time (Stephenson, 2007a; Lasersohn, 2005, 2009). None of these accounts, however, is suited to explain the puzzle of the competent speaker: A competent speaker of a language will know if a word involves an implicit argument or judge parameter. So why should he bother to express a denial in the first place? For example, if Ben would interpret Ann’s assertion in (1a) as saying that licorice is tasty to her, or by her judgment, why should he deny her assertion by asserting that licorice is not tasty to himself? (cf. Stojanovic, 2007; Moltmann, 2010) This puzzle casts doubt on any account involving an implicit argument or a judge parameter, and in fact it casts doubt on the very idea of faultless disagreement—why not accept the denial expressed by Ben in (1b) as genuine disagreement? In this paper, I will follow Stojanovic (2007) in considering the idea of faultless disagreement to be a misconception due to a bird’s eye perspective which is not available to either Ann or Ben. From the local perspective of the discourse participants Ann claims that licorice is tasty tout court and Ben denies this claim expressing genuine disagreement. Still, there is a clear difference between (1) and (2) since the latter describes a matter of fact while the former relates to a matter of taste. The difference between matters of fact and matters of taste is linguistically reflected in German by constraints on the attitude verb finden which combines with evaluative propositions, but not with factual propositions.¹ This is why (3) is not acceptable but (4a) is. By embedding an evaluative proposition under first

1 German finden is close in meaning to English find, but the latter is much more restricted in distribution. For this reason, German finden will also be glossed as think or consider in this paper, although this might not be the best translation.

Evaluative propositions

| 129

person finden the speaker can (i) make his claim immune against denial and, (ii), convey dissent with a previous claim without directly denying it. If, for example, Ann expresses her position embedded under finden, Ben cannot deny it, cf. (4), and if Ben counters Ann’s assertion in (5) by embedding his own position under finden, he will convey that he doesn’t agree without directly denying it. (3)

a. *Ich finde, Osnabrück liegt in Deutschland. ‘I think Osnabrück to be in Germany.’


a. b.


a. b.

Ann: Ich finde Lakritze lecker. ‘I find licorice tasty.’ Ben: #Nein! Lakritze ist eklig. ‘No! Licorice is terrible.’ Ann: Lakritze ist lecker. ‘Licorice is tasty.’ Ben: Ich finde Lakritze eklig. ‘I find licorice terrible.’

The literature on faultless disagreement focuses on more or less two predicates of personal taste, namely tasty and fun. Lasersohn explicitly rejects the idea of taking other predicates into account in order to avoid issues of aesthetics (cf. Lasersohn 2005, 645). In the present paper, basic insights will be taken from aesthetics, and the scope of predicates under consideration will be extended including predicates like schön ‘beautiful’, gut ‘good’ and also dimensional predicates like groß ‘big’. Dimensional adjectives like groß are known to diverge in semantic behavior from evaluative adjectives like schön with respect to Normbezug² (Bierwisch, 1987) and with respect to combination with measure phrases (cf. Sassoon, 2011). Surprisingly, when considering propositions embedded under finden, (unmodified) dimensional adjectives go together with evaluative adjectives—groß as well as schön are licensed in (6a). The picture changes though when considering comparative forms: schöner ‘more beautiful’ is licensed in finden complements whereas größer ‘bigger’ is not, cf. (6b).³

2 While the sentence A ist größer als B ‘A is bigger than B’ is neutral as to whether A or B are big, the sentence A ist schöner als B ‘A is more beautiful than B’ entails (in most contexts) that both A and B are beautiful; see also Rett (2008) who calls this effect “evaluativity”. 3 There is an evaluative reading of größer ‘larger’ in example (6b) which is licensed in finden complements: Ich finde die alte Wohnung größer als die neue — das muss an den kleineren Fenstern liegen ‘I find the old apartment larger than the new one which is possibly due to the small windows’. Many thanks to Marga Reis for providing this example.

130 | Carla Umbach


a. b.

Ich finde die neue Wohnung schön / groß/ *100 qm groß. ‘I think the new apartment is beautiful / large / lit.: 100 sqm large.’ Ich finde die alte Wohnung schöner/*größer als die neue. ‘I think the old apartment is more beautiful / larger than the new one.’

In this paper, an analysis of evaluative propositions—propositions with lecker/tasty, schön/beautiful, groß/big and also schöner / more beautiful as their main predicates— will be proposed in which semantic aspects of evaluativity are separated from pragmatic aspects. As for semantics, it will be argued that there are two reasons for a proposition to be evaluative. Either it is used in a metalinguistic way affecting the denotational borderline of the predicate or it is used in a descriptive way exploiting commonly agreed on criteria for satisfaction of the predicate (depending on comparison class, speaker community, context etc.). The distinction between metalinguistic and descriptive usage is taken from Barker (2002). The idea of criteria for satisfaction of evaluative predicates was proposed by Hare (1952). A semantics of evaluative predicates based on criteria is sketched making use of generalized measure functions, cf. Umbach & Gust (2014). Update will be implemented based on Krifka’s enriched notion of a common ground consisting of a pair of worlds and interpretations (cf. Krifka, 2012). As for pragmatics, evaluative propositions may—but need not—be relativized to express the speaker’s subjective position, e.g. by embedding under first person finden. (First person is mandatory, as with performative verbs.) Pragmatics will be spelled out in the discourse framework devised in Farkas & Bruce (2010) which includes, in addition to the common ground, sets of individual discourse commitments for each discourse participant. The paper is structured as follows: In Section 2, the current approaches to evaluative predicates will be summarized and the notion of taste judgment proposed by Kant (1878 [1790]) will be outlined. A classification inspired by Kant will be suggested distinguishing semantic and pragmatic aspects of evaluative propositions. Section 3 is about the semantics of evaluative propositions, distinguishing between descriptive and interpretational, i.e., metalinguistic, usage of propositions. In Section 4, the difference in pragmatics between general judgments (licensing denial) and subjective judgments (blocking denial) will be spelled out in the discourse framework in Farkas & Bruce (2010).

Evaluative propositions

| 131

2 The idea of faultless disagreement 2.1 Current positions in the literature The interpretation of evaluative predicates has been a topic of debate for the last decade in linguistics as well as philosophy. In this section, prominent positions in the literature will be reviewed focusing on two questions: (i) Does the account include an experiencer or judge, and if so, who can take that role? (ii) Is faultless disagreement acknowledged to exist? Contextualist and relativist positions will be grouped together and contrasted with metalinguistic positions. For details on the contextualism/relativism debate see the introduction to this volume by Janneke van Wijnbergen-Huitink. Lasersohn (2005) starts from the intuition that what we find in dialogues like (1) is faultless disagreement⁴, that is, Ben denies the proposition asserted by Ann and, at the same time, neither Ann nor Ben are wrong. Lasersohn rejects the option that predicates of personal taste involve an implicit argument such that tasty means tasty for some salient individual because the content (in a Kaplanian framework) of Ann’s and Ben’s assertions would be different (Licorice is tasty to Ann / Licorice is tasty to Ben) thereby not accounting for the fact that Ben’s reply is a denial. In view of this problem, Lasersohn suggests assuming that the content of Ann’s and Ben’s utterances in (1) is in fact the same (modulo negation) and considering the truth of propositions involving taste predicates as being relative to a judge index, in addition to the world index. His approach is implemented in a Kaplanian framework, treating the content of a sentence as a set of world-individuals pairs (instead of a set of worlds). Thus the truth or falsity of propositions involving taste predicates depends not only on the world of evaluation but also on a “standard of taste” or judge. Technically, taste predicates are one-place predicates and differ from regular predicates only in their judge-dependent valuation. There are three possibilities for the judge index corresponding to three perspectives: (i) The autocentric perspective is the perspective of the individual assessing the sentence, i.e., the speaker or hearer—this is the default case; (ii) the exocentric perspective results from the speaker taking the perspective of some other individual, e.g. in free indirect discourse and in questions; (iii) the acentric perspective is like a bird’s eye view. If a sentence is assessed from an acentric perspective, there is no judge and thus no determinate truth value. There is an interesting remark about the acentric perspective in Lasersohn’s (2009, 6) paper: “It is perhaps worth not-

4 The notion of faultless disagreement was coined by Kölbel (2002).

132 | Carla Umbach

ing that it is only when we adopt an acentric stance that ‘faultless disagreement’ really seems faultless.” This implies that, from the perspective of the discourse participants, disagreement is not faultless, which is the position in the present paper. Stephenson (2007a) presents another judge-dependent account. She starts from the observation that taste predicates are close to epistemic modals (e.g. It might be raining): While predicates of personal taste raise the question of whose taste is relevant, epistemic modals raise the question of whose knowledge is expressed. Stephenson makes use of Lasersohn’s system allowing, however, for an implicit argument interpretation in addition to the judge-as-index interpretation.⁵ In addition to the semantic interpretation of taste predicates, Stephenson gives a pragmatic account of taste judgments in dialogue. She posits a common ground which represents the mutual beliefs of the discourse participants and is updated in the course of the communication (Stalnaker, 1978, 2002). The common ground consists of a set of pairs of worlds and judges (instead of a set of worlds). Judgments of taste are handled by two principles, “actual judge” and “norm of assertion”. The actual judge represents the group of participants in the conversation and is the judge for all world-judge pairs in the common ground. Thus there must be consensus concerning matters of taste throughout the common ground. The norm of assertion, on the other hand, allows the speaker to assert a sentence even if he does not believe that the sentence is true as judged by the group of conversational participants. He is only required to believe that the sentence is true as judged by himself. If the assertion is accepted by the group and added to the common ground, the judge parameter of the proposition will be shifted to the “actual judge”. Judge parameters are implemented like world parameters: For every conversation, there is an actual world and an actual judge. Judges that differ from the actual one occur only temporarily and disappear as soon as the common ground is established, thereby preserving consensus.

5 Since Stephenson postulates that the autocentric perspective is obligatory, the implicit argument interpretation is required to account for examples where the agent differs from the individual whose taste is under consideration: (i)

a. b.

Mary: How’s that new brand of cat food you bought? Sam: I think it’s tasty, because the cat has eaten a lot of it.

At first sight the option of an implicit argument interpretation appears to address a home-made problem, which does not occur in Lasersohn’s system—he simply argues for an exocentric perspective of the agent of the attitude in cases like (i). Still, including both an implicit argument and a judge parameter makes it possible to implement judge-dependence in a centered-world account, the judge being the center, cf. Stephenson (2010).

Evaluative propositions

| 133

Moltmann (2010) also accepts the existence of faultless disagreement. However, she suggests accounting for this intuition not by relativizing the truth value of the proposition to a judge/experiencer but by “grasping the propositional content in a first-personal way, namely by applying the predicate to everyone in the domain as if to oneself” (Moltmann, 2010, 1). This idea is based on her account of first-person-based genericity where sentences involving generic one are understood such that the speaker identifies himself with each individual in the quantified domain. Simplifying considerably, Moltmann’s approach can be said to interpret the sentence Licorice is tasty such that it means One considers/finds licorice tasty. This sentence in itself has absolute truth conditions, is either true or false. But since different speakers may attribute different properties to the individuals he/she identifies with, truth values may differ between speakers, which is, following Moltmann, the reason for the intuition of faultless disagreement. Another first person genericity account is presented in Pearson (2013) linking it to the semantics of attitudes de se. In Pearson’s account first person indexicals (implicit experiencer arguments) and the judge parameter, which are distinct ways of representing subjectivity in relativist semantics, are conflated. This leads to a different view on identification: While in Moltmann’s account the speaker simulates the individual he/she identifies with, in Pearson’s account the speaker empathizes with that individual. Similar to Moltman and Pearson, Anand (2008) claims that propositions involving taste predicates are in some sense generic. According to Anand, they require universal quantification over normal individuals. Cohen (2010) argues that this leads to a number of problems and replaces the notion of normal individuals by the notion of competent perceivers, where perceivers are assigned different weights in determining the probability that the proposition is true. So when John and Mary disagree on a judgment of taste, they disagree on what counts as a competent perceiver and what the appropriate weights are, which is genuine disagreement. The intuition of faultlessness is explained by the fact that John and Mary are, of course, entitled to have different assessments of perceivers. John’s reason for contradicting Mary in a dialogue like (1) is to make her correct her assessment of competent perceivers. He does not want to “to convince her that she likes roller coasters [or licorice] but that she ought to like them” (Cohen, 2010, 2), since this is what the majority of competent perceivers do. Note that this is a normative position (see also Wolf, 2016, this volume). Analyses that reject the idea that judgments of taste involve an implicit experiencer or perceiver or judge mainly refer to some dispute about the interpretation of the predicates. This possibility is, e.g. entertained by Egan (2010), although he finally dismisses this option in favor of a de se account of propositional attitudes, and by Sundell (2011). Sundell distinguishes between “character disagreement” (about the character of the expression), “content disagreement” (about the truth

134 | Carla Umbach

of the content) and “context disagreement” (about the relevant context). It is suggested that disagreement about matters of taste is not about the content of the asserted propositions, and not about the character of the predicates, but about the context, that is, about the contextual standards of tasty, fun etc. Another analysis rejecting judge-dependence is Barker (2013). Barker considers disagreement about matters of taste to be disagreement about “the discourse”. This includes contextual standards of the predicates. But it also includes norms saying which aspects are relevant in determining whether a predicate applies. The intuition of faultlessness is explained by the fact that none of the participants has privileged authority over the discourse. Barker implements this analysis making use of the distinction between descriptive usage and metalinguistic usage of a proposition proposed in Barker (2002). To distinguish between these two usages, a notion of common ground is required that consists of worlds and discourses, and update may affect either one. The analysis in the current paper will be based on the distinction between the two usages without, however, adopting Barker’s (2013) claim that disagreement about matters of taste is restricted to “the discourse”, i.e., is always metalinguistic. Metalinguistic usage will be called “interpretational” in this paper (cf. Section 3.1). Let us finally consider Stojanovic (2007; 2012). She rejects the idea of faultless disagreement arguing that discourse participants either genuinely disagree, or they are both right but their disagreement boils down to a misunderstanding. This is demonstrated in (7). When Ben denies her assertion, Ann has two ways to reply. In (7c) she insists on her original claim, while in (7d) she retreats to a weaker position.⁶ Ann’s reply in (7c) makes it obvious that she considers their disagreement to be genuine, in that one of them has to be wrong, while her reply in (d) indicates a misunderstanding: What Ann meant to say is just that licorice is tasty to her. (7)

a. b. c. d.

Ann1: Licorice is tasty. Ben: No, it isn’t! It tastes terrible. Ann2: It is tasty. And it’s not just that I find it tasty; it’s tasty tout court. Ann2󸀠 : OK. To my taste, licorice is tasty; that’s all I’m saying.

Stojanovic furthermore argues that, even if there were faultless disagreement, relativism would fail to account for it, since semantically competent speakers of English would be aware that taste predicates are judge-dependent, and thus not bother to deny the assertion. (This is the puzzle of the competent speaker, cf. Section 1). In her (2012) paper, Stojanovic focuses on expressions of emotion (sad, as

6 The wording of Ann’s replies is directly adopted from Stojanovic’s (2007) examples (5) and (3).

Evaluative propositions

| 135

in It is sad that. . . ), which are similar to predicates of taste in giving rise to the intuition of faultlessness. While still rejecting the idea that disagreement may be faultless, she elaborates her position, combining contextualist and metalinguistic features such that the lexical meaning of the expressions is underspecified with respect to experiencer, respects, comparison class and threshold. In addition, the concepts associated with emotion predicates (as well as taste predicates) are open-ended. So even if discourse participants specify the required parameters, there may be residual disagreement which is metalinguistic in nature.

2.2 Kant’s notion of judgments of taste In his Kritik der Urteilskraft (1878 [1790]) Kant characterizes judgments of taste— about beauty (das Schöne)—and judgments of agreeableness—about pleasure (das Angenehme) in a way surprisingly relevant for linguistic interpretation.⁷, ⁸ In Kant’s system, first, judgments of taste and agreeableness are distinguished from judgments about factual matters. The latter are about properties of objects, whereas the former are about properties the subject ascribes to the object. Next, judgments of taste (in Kant’s terminology) are distinguished from judgments of agreeableness. (In linguistics, there is a tendency to narrow down the notion of taste to gustatory taste which, in Kant’s terminology, is a matter of agreeableness, see below.) Judgments of agreeableness don’t claim to be generally valid—others need not share our judgment: In Ansehung des Angenehmen bescheidet sich ein jeder: daß sein Urteil, welches er auf ein Privatgefühl gründet, und wodurch er von einem Gegenstande sagt, daß er ihm gefalle, sich auch bloß auf seine Person einschränke. Daher ist er es gern zufrieden, daß, wenn er sagt: der Kanariensekt ist angenehm, ihm ein anderer den Ausdruck verbessere und ihn erinnere, er solle sagen: er ist mir angenehm; [. . . ] Darüber in der Absicht zu streiten und das Urteil anderer, welches von dem unsrigen verschieden ist, gleich als ob es diesem logisch entgegengesetzt wäre, für unrichtig zu schelten, wäre Torheit; und in Ansehung des Angenehmen gilt also der Grundsatz: Ein jeder hat seinen besonderen Geschmack. (Kant, 1878 [1790], 54)⁹

7 Very many thanks to Peter Bosch for bringing Kant’s theory on aesthetics judgments to my attention. 8 In presenting Kant’s ideas, extensive use will be made of the clear and comprehensible article on aesthetic judgment by Zangwill (2014) in the Stanford Encyclopedia of Philosophy. 9 “As regards the Agreeable, everyone concedes that his judgment, which he bases on a subjective feeling, and in which he declares that an object pleases him, is restricted merely to himself personally. Thus he does not take it amiss if, when he says that Canary wine is agreeable, another corrects the expression and reminds him that he ought to say: It is agreeable to me. [. . . ] To quar-

136 | Carla Umbach

In contrast, judgments of taste demand, following Kant, general validity and thus come with a normative claim—we insist on others agreeing with our taste: Mit dem Schönen ist es ganz anders bewandt. Es wäre [. . . ] lächerlich, wenn jemand, der sich auf seinen Geschmack etwas einbildete, sich damit zu rechtfertigen gedächte: dieser Gegenstand (das Gebäude, was wir sehen, das Kleid, was jener trägt, das Konzert, was wir hören, das Gedicht, welches zur Beurteilung aufgestellt ist) ist für mich schön. [. . . ] Reiz und Annehmlichkeit mag für ihn vieles haben, darum bekümmert sich niemand; wenn er aber etwas für schön ausgibt, so mutet er andern eben dasselbe Wohlgefallen zu: er urteilt nicht bloß für sich, sondern für jedermann, und spricht alsdann von der Schönheit, als wäre sie eine Eigenschaft der Dinge. Er sagt daher, die Sache ist schön, und rechnet nicht etwa darum auf Anderer Einstimmung in sein Urteil des Wohlgefallens, weil er sie mehrmalen mit dem seinigen einstimmig befunden hat, sondern fordert es von ihnen. Er tadelt sie, wenn sie anders urteilen, und spricht ihnen den Geschmack ab [. . . ] und sofern kann man nicht sagen: Ein jeder hat seinen besonderen Geschmack. (Kant, 1878 [1790], 55)¹⁰

Kant’s characterization of the two types of judgments is strikingly close to linguistic form. His judgments of the agreeableness come with an implicit experiencer argument (“he does not take it amiss if, [. . . ] [another] reminds him that he ought to say: It is agreeable to me.”) and they don’t license denial. In contrast, his judgments of taste do not allow for an experiencer argument (“be ridiculous if any one [. . . ] were to think of justifying himself by saying: This object [. . . ] is beautiful for me.”) and they do license denial. Kant’s characterization seems to suggest that this is a lexical distinction: There are predicates like schön ‘beautiful’ that give rise to general judgments and there are predicates like angenehm ‘pleasant’ express-

rel over such points with the idea of condemning another’s judgment as incorrect when it differs from our own, as if the opposition between the two judgments were logical, would be folly. With the Agreeable, therefore, the axiom holds good: Everyone has his own taste” (Translation by J. C. Meredith, Moment_2). 10 “The Beautiful stands on quite a different footing. It would, on the contrary, be ridiculous if any one who prides himself on his taste were to think of justifying himself by saying: This object (the building we see, the dress that person has on, the concert we hear, the poem submitted to our criticism) is beautiful for me. [. . . ] Many things may for him possess charm and agreeableness—no one cares about that; but when he puts a thing on a pedestal and calls it beautiful, he demands the same delight from others. He judges not merely for himself, but for all men, and then speaks of beauty as if it were a property of things. Thus he says that the thing is beautiful; and it is not as if he counts on others agreeing with him in his judgment of liking owing to his having found them in such agreement on a number of occasions, but he demands this agreement of them. He blames them if they judge differently, and denies them taste, which he still requires of them as something they ought to have; and to this extent it is not open to men to say: Everyone has his own taste.” (Translation by J.C. Meredith, Part_1/Section_1/Book_1/Moment_2)

Evaluative propositions

| 137

ing mere personal feelings. In the recent literature it has, in fact, been pointed out that in the case of predicates reporting internal experience subjects cannot be mistaken, for example, I am freezing or I am in pain. It may be suspected that the paradigm example tasty belongs to that category justifying a judge-dependent analysis à la Lasersohn (2005) or Stephenson (2007a; 2007b). But since people fight about matters of gustatory taste as they fight about matters of beauty, it may also be of the type of judgments claiming general validity (note that German lecker ‘tasty’ does not license an experiencer argument). We will leave this question for further research (see, e.g. the experimental study by Meier, 2012) and focus on predicates like schön ‘beautiful’ in this paper. There is no room for something akin to faultless disagreement in Kant’s system. The speaker expresses genuine disagreement. But since this judgment is not about of matters of fact, it is essentially normative (“when he puts a thing on a pedestal and calls it beautiful, he demands the same delight from others.”). According to Kant, normativity is a core characteristic of taste judgments explaining the possibility of genuine denial (for the notion of normativity and its role in Kant’s system see Zangwill, 2014). Thus, an analysis of taste judgments (in the sense of Kant) has to refer to normativity with respect to language use.¹¹

2.3 Semantic vs. pragmatic aspects of evaluativity In this paper it will be assumed that there is no faultless disagreement and no misunderstanding—if you claim that licorice is tasty and I say no, this is genuine disagreement. The idea of faultless disagreement presupposes a bird’s eye perspective, but for participants in a discourse there is no bird’s eye perspective. Still, there is the intuition that judgments concerning matters of taste are in some sense weaker than judgments concerning matters of fact. The reason for this intuition is—as argued in this paper—that judgments about matters of taste can be presented as being merely subjective, for example, in German by embedding the proposition in question under the attitude verb finden (‘find’/‘consider’/‘think’) (cf. Section 4). Judgments presented in this way by the speaker (first person) are immune against denial, as shown by the infelicity of denial in (4) in the intro-

11 In the approach in Stephenson (2007a), normativity is not explicitly mentioned. However, in her account the judge-dependent evaluation loses relativity as soon as the proposition enters the common ground since for a judge-dependent statement to enter the common ground the judge must be substituted by the actual judge of the conversation. Thus a competent speaker will intend that his statement be adopted by the overall group of discourse participants, which is a normative intention.

138 | Carla Umbach

duction. However, in contrast to what is claimed in relativist accounts, judgments about matters of taste can also claim general validity, and in that case license genuine denial. Therefore, from a pragmatic point of view, general judgments (licensing denial) will be distinguished from subjective judgments (blocking denial). From a semantic point of view, being licensed as a complement of the attitude verb finden appears, at first sight, suited to separate propositions about matters of taste from those about matters of fact. This was in fact suggested in earlier versions of this paper. It proved wrong for two reasons: First, the characteristic German finden is sensitive to is not evaluativity but instead interpretational (metalinguistic) usage of the proposition (cf. Section 3.1). Secondly, it became evident that evaluative propositions have not only an interpretational usage but also a descriptive one. These two insights led to an analysis that emphasizes the opposition between descriptive and interpretational usage of propositions instead of that between propositions about matters of fact and propositions about matters of taste. The hypothesis in this paper is shown in the classification schema in Table 1. The horizontal dimension is semantic relating to the usage of the propositions. Propositions with (the positive form of) evaluative predicates like beautiful can be used either way. Propositions with (the positive form of) dimensional predicates, like tall, can also be used either way but only the interpretational usage appears evaluative (and can be embedded under finden) while the descriptive usage expresses a factual matter presupposing a given standard for satisfaction of the predicate. Considering comparative forms, propositions with evaluative predicates can again be used either way. Comparatives of dimensional predicates, in contrast, do not license an interpretational usage. This picture entails that a proposition can be evaluative for two different reasons: It can be evaluative in the sense of having an interpretational usage manipulating standards, but it can also be evaluative when having a descriptive usage, provided the predicate is lexically evaluative (gradable but non-dimensional). Surprisingly, these two reasons for being evaluative partition the data analogous to the two kinds of subjectivity suggested by Kennedy (2016) [this volume], even if his explanation is completely different (Section 4.1). Distinguishing descriptive usage and interpretational usage leads to different modes of updating the common ground: While the former leads to updating worlds, the latter leads to updating interpretations (or “discourses” in the sense of Barker). Implementing this distinction requires a notion of common ground including interpretations in addition to worlds. The semantic dimension will be detailed in Section 3. The vertical dimension in Table 1 is pragmatic relating to the nature of the judgment made by the speaker. When making a general judgment, the speaker

Evaluative propositions



demands that the proposition in question is included in the common ground of the conversation. In contrast, when making a subjective judgment, the speaker does not demand that the proposition is included in the common ground. It may instead stay a mere individual discourse commitment. This will explain the blocking of denial effect. Details of the pragmatic dimension will be discussed in Section 4. Finally, the classification in Table 1 raises the question of whether descriptively used propositions can be presented as subjective judgments, that is, is the left lower cell inhabited? We will leave this question for future research and focus on the instances shown in the table. Table 1. Descriptive vs. interpretational usage — general vs. subjective judgments descriptive usage

interpretational usage evaluative propositions

general judgments

A ist schön ‘A is beautiful’

A ist schön ‘A is beautiful’

A ist schöner als B ‘A is more beautiful than B’

A ist schöner als B ‘A is more beautiful than B’

A ist groß. ‘A is tall’

A ist groß ‘A is tall’

A ist größer als B ‘A is taller than B’

-Ich finde A schön ‘I think A is beautiful’

subjective judgments


Ich finde A schöner als B ‘I think A is more beautiful than B’ Ich finde A groß ‘I think A is tall’

140 | Carla Umbach

3 The semantic dimension In this section, the semantic dimension of evaluativity will be investigated. We will start with the difference between descriptive and interpretational usages of propositions (3.1). Four dialogues will be considered, involving descriptive and interpretational uses of propositions with dimensional and evaluative predicates. It will be proven that dimensional as well as evaluative predicates license descriptive and interpretational usages (3.2). The puzzle of how to explain descriptive usages of evaluative propositions without postulating a judge or experiencer is solved along the lines of Hare (1952) (Section 3.3). In Section 3.4 a brief sketch is presented of how Hare’s notion of criteria can be implemented in a formal framework. In Section 3.5 evaluative update is defined based on Krifka (2012). Throughout this section we will assume a degree based analysis of gradability such that the positive form relates to a standard defining the borderline of the predicate (e.g. Kennedy, 1999).

3.1 Descriptive vs. interpretational usage of propositions In Barker (2002) it is shown that propositions based on gradable predicates have two types of uses, a descriptive and a metalinguistic one (called interpretational in this paper). In its descriptive use the sentence Feynman is tall provides information about Feynman’s height by asserting that he exceeds a contextually given standard of tallness. In its metalinguistic use the sentence provides information about what counts as tall in that context by asserting that Feynman does. As Barker points out, Asserting and accepting a token of the sentence Feynman is tall can resolve some portion of the mutual uncertainty associated with the applicability of the predicate tall. More specifically, it eliminates from further consideration the possibility that the vague standard of absolute tallness might be greater than the maximal degree of Feynman’s height. (Barker, 2002, 1)

The analysis is implemented in a dynamic framework such that the common ground consists of pairs of worlds and discourses and update may affect both. Barker’s distinction of a descriptive and a metalinguistic/interpretational usage aims at gradable predicates. But it also applies to non-gradable ones. Imagine that in a context of delivery of goods the speaker points at a bulky package. In this situation the proposition in (8) is used descriptively. In contrast, in the context of Max Black’s chair museum (Black, 1937) the speaker pointing to a borderline case

Evaluative propositions



in between chairs and non-chairs the proposition is used in an interpretational way. (8)

This is a chair.

It is generally assumed that the distinction between the two usages correlates with the distinction between topic and comment (see, e.g. Krifka, 2012). In its descriptive usage the sentence Feynman is tall is about Feynman, and thus Feynman is the topic of the proposition. In its interpretational usage the sentence is about the standard of comparison of tallness in the context and thus the standard of tallness is the topic of the proposition. This is demonstrated in (9)/(10) where the comment part is marked as being accented (which is a useful indicator for comments in this type of sentences). (9) (10)

A: What about Feynman? B: Feynman is TALL. A: What about the standard of height in this community? B: FEYNMAN is tall.

There are, however, clear cases of interpretational usages of sentences that are not about an expression but about an individual. Consider the sentences in (11). In the context of delivery of goods the bulky package the speaker points at is most naturally the topic of the sentence, cf. (11a). In the context of the chair museum it can be either way. If the denotation of chair is under debate, then the object pointed at is the comment, cf. (11b). If the names of the objects in the museum are under debate then the object pointed at is the topic, cf. (11c). In (11a), the proposition is used descriptively informing the hearer about the contents of the bulky package. In (11b), the proposition is used interpretationally providing an example of the denotation of chair. In (11c), the proposition is also used interpretationally, the speaker asserting that the name of the object is chair. (11)

a. b. c.

A (pointing to a bulky package): What about this? B: This is a CHAIR. A: What about the meaning of chair? Which objects count as chairs? B (pointing to a borderline chair): THIS is a chair. A (pointing to a borderline chair): What about this object? What is it called? B: This is a CHAIR.

Interpretational usage of propositions is not acceptable in normal conversation because speakers know the meaning of the words in their language. In fact, (11b,c)

142 | Carla Umbach

require a situation with borderline cases, like the chair museum, or a learning context. In the case of gradable expressions, however, the interpretational usage is unmarked since the denotation is not fixed in advance in the language and instead varies depending on context and comparison class. The observation concerning the predicate chair in (11) can straightforwardly be carried over to gradable expressions. Suppose a secretary has to order shirts for the school’s soccer team and there are only two sizes available. She doesn’t know Feynman personally and thus asks the trainer What about Feynman? In this context the answer Feynman is tall has a descriptive usage. Suppose, on the other hand, that the teacher has to split students into two groups and asks the trainer for advice. If her question is Who counts as tall on the school soccer team?, the answer Feynman is tall has an interpretational usage and the intonation is as in (10). Likewise, if her question is What about Feynman?, the answer Feynman is tall has an interpretational usage the reason being that the teacher knows Feynman personally and thus does not ask what his actual height is but instead asks for advice whether he should be considered as tall. In this situation she negotiates the standard of tallness while still talking about Feynman. The soccer team example is evidence again that the distinction between descriptive and interpretational usage need not correlate with the distinction between topic and comment. Even if used interpretationally, the proposition may be about Feynman the speaker requesting that the standard of tallness is such that Feynman is included. This is a normative claim about the denotation of a predicate of the language. For nongradable predicates like chair it would be acceptable only in borderline cases. For gradable adjectives like tall normative claims are a common means of negotiating the standard of comparison. Anticipating the discussion of the German attitude verb finden in Section 4.1, finden does not embed propositions about matters of fact, cf. (12a). Complements of finden may be evaluative propositions, cf. (12b), but also definitional sentences and normative sentences, even if clearly not evaluative as in (12c,d), cf. Sæbø (2009) and Reis (2013). (12)

a. *Ich finde, Osnabrück liegt in Dänemark. ‘I consider Osnabrück to be in Denmark.’ b. Ich finde, das Bild ist schön. ‘I find the picture beautiful.’ c. Ich finde, das ist ein Stuhl. ‘I consider this a chair.’ d. Ich finde, indirekte Steuern sollen abgeschafft werden. I think indirect taxes should be abolished.’

Evaluative propositions

| 143

In order to account for the range of finden complements, Reis (2013) (referring to Ducrot, 1980) suggests that finden induces a reading of the complement proposition such that it represents one of a number of alternative interpretations of a given (presupposed) state of affairs (for details see Section 4.1). Reis is sceptical as to whether her interpretational account of finden complements relates to Barker’s (2002) interpretational usage of propositions. It can be observed, however, that when embedded under finden complements must be used interpretationally. This is demonstrated by the fact that contexts inducing the descriptive usage block embedding under finden, while contexts inducing the interpretational usage license embedding under finden, cf. (13). (13)





(Speaker pointing to a bulky package in a delivery of goods) #Ich finde, das ist ein Stuhl. ‘I consider this as a chair.’ (Speaker pointing to a borderline case in Black’s chair museum) Ich finde, das ist ein Stuhl.¹² ‘I consider this as a chair.’ (Speaker asked which size Feynman’s shirt has to be) #Ich finde Feynman ist groß. ‘I think Feynman is tall.’ (Speaker asked whether Feynman should be counted as tall) Ich finde Feynman ist groß.¹³ ‘I think Feynman is tall.’

To conclude, following Barker (2002) we will distinguish between a descriptive and an interpretational usage of propositions without, however, assuming that this distinction correlates with the distinction between the topic and the comment of a proposition. While descriptive uses concern the way the world is, interpretational uses concern the way expressions are used. But even then one can talk about an expression claiming that it applies to some individual, and one can talk about an individual claiming that it should count as an instance of a certain expression. So the notion of interpretational usage in this paper is slightly broader than what is commonly considered as metalinguistic, including propositions where the topic is not (the meaning of) an expression.

12 Many thanks to Markus Kracht for providing me with this example. 13 Note by the way that the variant where the topic is the standard (instead of Feynman) is not good when embedded under finden: A: What about the standard of height in this community? B: ? Ich finde FEYNMAN groß. ‘I think Feynman is tall.’

144 | Carla Umbach

3.2 Four dialogues Let us compare the four dialogues in (14)–(17). They take place in two different situations. In the first situation, Ann informs Ben on the phone about a sculpture she saw in an exhibition, and Ben has not seen it himself. To avoid complications it will be assumed that the definite NP the sculpture is anaphoric, that is, Ann and Ben have been talking about the sculpture before. In the second situation, both Ann and Ben are in the exhibition standing vis-à-vis the sculpture. In each of these situations Ann and Ben have a dispute involving a dimensional adjective and a dispute involving an evaluative adjective. (14)

Dialogue 1: phone/tall a. b. c.


Dialogue 2: vis-à-vis/tall a. b.


Ann: Die Skulptur ist groß. ‘The sculpture is tall.’ Ben: Nein, ist sie nicht. (Ich finde sie klein.) ‘No, it is not. (I think it is small.)’

Dialogue 3: phone/beautiful a. b. c.


Ann: Die Skulptur ist groß. ‘The sculpture is tall.’ Ben: ??Nein, ist sie nicht. ‘No, it is not.’ Ben󸀠 : Nein, sie kann nicht groß sein. Sie hätte sonst nicht durch die Tür gepasst. ‘No, it can’t be tall. Otherwise it wouldn’t have passed through the door.’

Ann: Die Skulptur ist schön. ‘The sculpture is beautiful.’ Ben: ??Nein, ist sie nicht. ‘No, it is not.’ Ben󸀠 : Nein, sie kann nicht schön sein, sonst hätte meine Tante sie nicht verkauft. ‘No, it can’t be beautiful. Otherwise my aunt wouldn’t have sold it.’

Dialogue 4: vis-à-vis/beautiful a. b.

Ann: Die Skulptur ist schön. ‘The sculpture is beautiful.’ Ben: Nein, ist sie nicht. (Ich finde sie hässlich.) ‘No, it is not. (I find it ugly.)’

Evaluative propositions

| 145

Consider the phone/tall dialogue in (14). To strengthen the case, assume that Ben needs to know the height of the sculpture because he has to transport it back to the studio. In using the dimensional adjective groß (‘tall’), Ann informs Ben about a property of the sculpture: Its height exceeds the standard height of sculptures presented in an art gallery. This is a descriptive use of the proposition Die Skulptur ist groß ‘The sculpture is tall’. Since Ben cannot see the sculpture, he can deny her assertion only by citing indirect evidence, for example, by arguing that the sculpture must be smaller than the height of the door of the exhibition hall. In contrast, in the vis-à-vis/tall dialogue in (15), Ann cannot reasonably try to inform Ben about the height of the sculpture when using the dimensional adjective groß (‘tall’). Instead, she asserts that the sculpture is included in what should be considered as tall for a sculpture in this context. Accordingly, when denying her assertion, Ben does not question her measuring skills but her assessment of what counts as tall for a sculpture in that context. This is what we refer to as interpretational usage. Now consider the dialogues with the evaluative adjective beautiful. In the visà-vis situation in (17) there seems to be no difference between the dimensional adjective and the evaluative one. Since their sensory experience will be the same (more or less), Ann’s assertion can only address the question of what the adequate standard of comparison for the beauty of sculptures in this context is.¹⁴ Direct denial is fine but, as in the dimensional case, Ben doesn’t question Ann’s perception of the sculpture but her assessment of what counts as beautiful for a sculpture in that particular context. So (17) is a case of interpretational usage, analogous to (15). Finally, what is Ann doing in (16) when using the evaluative adjective on the phone? Is she conveying the information that the sculpture has some property that makes it exceed a contextually fixed standard of beauty? If so, her assertion should be paraphrased as saying “There is an n such that the beauty of the sculpture is n beauty-units, which is above the generally accepted standard of comparison for the beauty of sculptures in this context”. But there are no beauty units inherent 14 The assumption that the sensory experience of Ann and Ben is more or less the same is controversial. One of the reviewers pointed out that, at least, it does not generalize to judgments of gustatory taste—the sensory experience of licorice may change from one subject to another. Similarly, men seem to be less sensitive to bitterness than women. Another reviewer referred to statements like I am freezing reporting the internal experience of the speaker and suggested handling beautiful in the same way, that is, reporting the internal experience of the speaker in view of the sculpture. However, I am freezing differs from The sculpture is beautiful in not licensing denial, not even if faultless. Moreover, the focus in this paper is on predicates like schön which cannot be a matter of internal experience of the speaker. Otherwise, any aesthetic debate would be pointless.

146 | Carla Umbach

to the sculpture. Unlike a certain degree of height, a certain degree of beauty is not a property of an object and instead a property ascribed by the speaker as if it where inherent—this is what we learned from Kant. The absence of a measurable beauty property seems to suggest that the proposition in (16) cannot have a descriptive usage because it cannot convey information about the world, that is, the sculpture. Still, we feel uneasy with the conclusion that Ann in (16) discusses the standard of beauty. Doesn’t she at least inform Ben about certain characteristics of the sculpture? Suppose Ann and Ben firmly agree in disliking sculptures with a rusty surface. Doesn’t Ben then learn from Ann that the sculpture under debate does not have a rusty surface?¹⁵ From the point of view that beauty is not a property of the object, a descriptive usage of evaluative propositions is at first sight puzzling. The puzzle will be solved with the help of Hare’s (1952) notion of criteria in the next section. It will be explained why propositions with schön/beautiful may license a descriptive usage in addition to the interpretational one and, moreover, why the comparative schöner / more beautiful also licenses both usages, while the comparative größer/taller does not.

3.3 Descriptive usage of evaluative propositions In his book The Language of Morals Hare investigates prescriptive language including imperatives and value judgments.¹⁶ His prime example is the predicate good. There is no property shared by good things—a good motor car and a good picture and a good meal have nothing in common apart from being good. So there is no denotational meaning of good. But there is what Hare calls the commending function of good: calling a motor car or a picture or a meal good means commending it. The commending function is called the evaluative meaning component of good. In addition to the evaluative meaning component there is, following Hare, a descriptive meaning component. We will call it quasi-denotational to avoid confusion with the notion of descriptive usage of a proposition. Although there is no property denoted by good there are criteria, relative to comparison class, speaker community, time etc., establishing a standard for something to be called good. The criteria relate to factual properties thereby creating a—highly context-

15 Very many thanks to my colleagues from ZAS, and in particular Stephanie Solt, for convincing me that evaluative propositions can convey descriptive information. 16 Very many thanks also to Vera Hoffmann-Kolss, who brought this book to my attention.

Evaluative propositions



dependent—quasi-denotational meaning. This is why value judgments may provide factual information. Hare’s example is good motor car. Suppose someone has been told that a particular car M is a good motor car. Suppose, moreover, this person knows nothing about M, but he knows what the accepted standard of goodness in motor cars is: He will complain that I have misled him, if he subsequently discovers that M will not go over 30 m.p.h., or uses as much oil as petrol, or is covered with rust, or has large holes in the roof. His reason for complaining will be the same as it would have been if I had said that the car was red and he subsequently discovered that it was black. I should have led him to expect the motor-car to be of a certain description when in fact it was of a quite different description. (Hare, 1952, 113)

One consequence of the lack of a regular denotation is that the predicate good, in contrast to red, cannot be taught by ostension—there is no common property to infer from examples as diverse as good motor cars, good pictures and good meals. Restricting the domain to motor cars someone may be taught by ostension to distinguish good motor cars from bad ones. Still, he will not have learned that calling something good means commending it. Another consequence of the lack of a regular denotation is the fact that the predicate good, in contrast to red, cannot be redefined in the sense of changing the denotation. It may happen that motor-cars will in the near future change considerably in design [. . . ]. It may be that then we shall cease giving the name ‘a good motor-car’ to a car that now would rightly and with the concurrence of all be allowed that name. [. . . ] we may begin to say ‘No cars of the nineteen-fifties were really good; there weren’t any good ones till 1960’. (Hare, 1952, 113)

Such a shift in meaning is not a redefinition but a change of standard and makes essential use of the fact that the evaluative meaning of good, that is, the commending function, stays constant. [. . . ] we are doing what would be called, if ‘good’ were a purely descriptive word, redefining it. But we cannot call it that, for the evaluative meaning remains constant; we are rather altering the standard. (Hare, 1952, 119)

Consider the sentence in (18). Suppose the addressee is familiar with Honda Civics. But he knows nothing about criteria for good motor cars. Then (18) provides information on which criteria count as good-making criteria, that is, it provides information about the standard of good in the context of motor cars (plus speaker community, time, etc.)—being a good motor car is being like a Honda Civic. This is what we call an interpretational usage of a proposition. Now suppose that the addressee is not familiar with Honda Civics. But he agrees with the speaker on

148 | Carla Umbach

what criteria for good motor cars are. Then (18) provides information on Honda Civics, namely that they satisfy the agreed-on criteria for good motor cars. This is the descriptive usage of a proposition. So even for evaluative propositions there is a descriptive usage in addition to the interpretational usage. (18)

A Honda Civic is a good motor car.

Transferring Hare’s analysis of good to our prime example, beautiful let us assume that the evaluative meaning component is, as before, given by commending, maybe commending as to appearance. There is no denotation but there are, for example, criteria for what counts as a beautiful apartment and criteria for what counts as a beautiful sculpture (depending on speaker community, time etc.). Criteria may directly be related to factual properties. Let us assume that for an apartment to count as beautiful it has to be provided with a parquet floor, stucco ceilings and an unobstructed view over the roofs of the city. Such criteria directly establish a quasi-denotational meaning. Criteria may also be related to evaluative properties, so that the quasi-denotational meaning is only indirectly established, for example, stylish furnishing. Grounding criteria in factual properties may even fail in some cases because criteria relate to subjective perceptual experience, like sensation of cold. Let us nevertheless assume for ease of exposition that criteria are sufficiently grounded to establish a quasi-denotational meaning. Coming back to the phone/beautiful dialogue in (16), Hare’s notion of criteria establishing a quasi-denotational meaning explains the puzzle of why evaluative propositions may convey factual information. Suppose in (16) that Ann and Ben agree on the criteria for a sculpture to count as beautiful, say, a sparkling surface and well-balanced proportions. Then Ann in (16) conveys the information that these criteria apply to the sculpture and if not, Ben is entitled to complain that Ann mislead him. Let us briefly compare the evaluative predicate beautiful to the dimensional predicate tall. Hare does not mention dimensional predicates, but the connection is obvious. As in the case of good and beautiful, there is no denotation of tall independent of comparison class and context. And as in the case of good and beautiful a change of standard, e.g. what counts as tall for a 12 year old girl, cannot be considered as a redefinition since it makes use of a fixed meaning component, say, outstanding in height (see the notion of “standing out” in Kennedy, 2007). The difference between tall on the one hand and beautiful on the other is located in the nature of the criteria. For tall, the criterion is “exceed a certain degree of height”. It is fixed by the lexical meaning of the word in so far as the dimension has to be the dimension of height which comes with a ratio scale. Otherwise it depends on comparison class and context. For beautiful there are multiple criteria

Evaluative propositions

| 149

and none of it is fixed by the lexical meaning of the word. Criteria can be viewed as features with certain values, e.g. ⟨surface: sparkling⟩. Features can be viewed as dimensions coming with scales of various types—ratio, ordinal, or even nominal. From this point of view criteria are (sets of) feature-value pairs or (sets of) points in multiple dimensions. So while in the case of tall there is only dimension and it is linked to a ratio scale, there are multiple dimensions in the case of beautiful linked to scales of various types (cf. Section 3.4). From this point of view, evaluative predicates like beautiful and dimensional predicates like tall are not that different. Both allow for an interpretational usage of the proposition they occur in, thereby facilitating standards to be negotiated. Both allow for a descriptive usage relying on context-dependent criteria, thereby allowing them to convey factual information. In the case of beautiful the interpretational as well as the descriptive usage count as evaluative although for different reasons—the former is evaluative in the sense of manipulating standards and the latter is evaluative in the sense of lacking a regular denotation and thus relating to multiple criteria. In the case of tall, the interpretational usage counts as evaluative in the sense of manipulating standards, as in the case of beautiful. The descriptive usage of tall, however, is not evaluative since there is only one dimension and it is fixed by the lexical meaning—it has to be height. Up to now, only positive forms of the adjectives have been considered. As for comparatives, the comparative of beautiful appears evaluative and the comparative of tall does not, cf. (19a,b). This effect can be explained by the difference between evaluative and dimensional predicates suggested above: The proposition in (19a) has a descriptive as well as an interpretational usage. The descriptive usage refers to an order established by the criteria of what counts as beautiful in the given context (for a sketch of how the order can be established see Section 3.4). The interpretational usage facilitates negotiation of criteria. As in the case of the positive form, the descriptive usage of (19a) is intuitively evaluative because it has to take recourse to varying criteria, and the interpretational usage of (19a) counts as evaluative because criteria (and the order itself) are negotiated. The proposition in (19b), however, has a descriptive usage only, and the descriptive usage is not intuitively evaluative because there is a single ratio scale dimension fixed by the lexical meaning. An interpretational usage of the proposition in (19b) is impossible because neither the dimension nor the order on a ratio scale are negotiable. Anticipating the discussion in Section 4.1 again, the fact that the positive but not the comparative form of dimensional adjectives licenses an interpretational usage is reflected by the fact that the positive but not the comparative can occur in a complement of the German attitude verb finden, cf. (20).

150 | Carla Umbach


a. b.


Diese Skulptur ist schöner als die, die wir gestern gesehen haben. ‘This sculpture is more beautiful than the one we saw yesterday.’ Diese Skulptur ist größer als die, die wir gestern gesehen haben. ‘This sculpture is taller than the one we saw yesterday.’


Ich finde diese Skulptur groß. ‘I think this sculpture is tall.’ b. *Ich finde diese Skulptur größer als die, die wir gestern gesehen haben. ‘I think this sculpture is taller than the one we saw yesterday.’

3.4 Generalized measure functions and multi-dimensional spaces In this section a brief sketch will be provided of how Hare’s idea of criteria can be implemented in formal semantics. The starting point is the notion of multidimensional attribute spaces and generalized measure functions proposed in Umbach & Gust (2014) in order to account for the meaning of similarity expressions (such, like this, similar etc.). A well-known instance of multi-dimensional spaces are conceptual spaces as suggested by Gärdenfors (2000). Unlike Gärdenfors’ conceptual spaces, which employ a quantitative similarity measure (geometrical distance), we employ a qualitative notion such that similarity is understood as indistinguishability with respect to classification functions defined on dimensions. Another fundamental difference lies in the fact that Gärdenfors’ conceptual spaces form a stand-alone system, independent of truth-conditional semantics while in Umbach & Gust (2014) attribute spaces are integrated into truth-conditional semantics. Integration makes use of a generalized version of the notion of measure functions known in degree semantics (cf. Kennedy, 1999). In the case of dimensional adjectives like large, there is a single dimension fixed by the adjective’s lexical meaning that relates to a ratio scale. In the case of evaluative adjectives there are, following Hare’s analysis, multiple criteria to be taken into account. Criteria can be seen as points on dimensions with scales of various types (ratio, ordinal, nominal). This suggests a straightforward generalization of the notion of measure functions: While common measure functions map individuals to degrees in a single ratio scale dimension, generalized measure functions map individuals pointwise into multi-dimensional spaces, where dimensions are of various types. This is shown in (21) for large and beautiful (attributed to apartments). The common measure function μ size associated with the adjective large in (21a) takes individuals to points in the size dimension, say, real numbers. The generalized measure

Evaluative propositions

| 151

function μ beauty associated with the adjective beautiful in (21b) takes individuals to points in a multi-dimensional space spanned by a floor dimension, a ceiling dimension, a size dimension etc. where the former two have nominal scales and the latter has a ratio scale as in (21a). (21)

a. b.

μsize : U → R μbeauty : U → ⟨floor, ceiling, size, . . .⟩, where μ beauty (x) = ⟨μ floor (x), μ ceiling (x), μ size (x), . . . ⟩ and μ floor (x) ∈ {parquet, marble, carpeting, . . .}, μ ceiling (x) ∈ {stucco, suspended, plasterboard, . . .}, μ size (x) ∈ R, etc.

Multi-dimensional attribute spaces are given by a set F of dimensions. Dimensions are associated with a set C(F) of classification functions defined on their points. Classification functions approximate natural language predicates on a conceptual level yielding corresponding truth values (modulo fuzzy membership). The role of classification functions is twofold. First, while generalized measure functions take individuals to points in attribute spaces, classification functions link points to regular predicates. Secondly, classification functions determine the level of granularity: Similarity is defined such that two individuals are similar with respect to a set of dimensions F and an associated set of classification functions C(F) iff all classification functions p∗ ∈ C(F) yield the same result when applied to corresponding points in the attribute space, cf. (22). This corresponds (for fixed F and C(F)) to the indistinguishability notion of similarity in rough set theory (Pawlak, 1982). In order to implement a gradable notion of similarity the sets of classification functions associated with an attribute space F are assumed to be (partially) ordered with respect to the granularity they allow for. The definition of “more similar” in (23) is such that x is more similar than y to z iff there is a more fine-grained set of classification functions according to which x is similar to z but y is not. (22)

sim(x, y, F, C(F)) iff ∀p∗ ∈ C(F) : p∗ (μ F (x)) = p∗ (μ F (y))


more-sim(x, y, z, F, C(F)) iff ∃C 󸀠 sim(x, z, F, C󸀠 (F)) & ¬sim(y, z, F, C󸀠 (F)) & C󸀠 (F) < C(F)

Evaluative adjectives, like beautiful, can be interpreted in multi-dimensional spaces analogous to dimensional adjectives. The interpretation of the positive form is relative to a standard of beauty in a set of relevant dimensions associated with a set of classification functions. An individual x is beautiful iff it is sufficiently similar to the standard (that is, similar according to the granularity of the set of classification functions), cf. (24). The interpretation of the comparative form

152 | Carla Umbach

in (25) makes use of the gradable notion of similarity: x is more beautiful than y iff x is more similar to the standard than y, that is iff there is a set of classification functions C󸀠 (F) such that x is beautiful according to C󸀠 (F) but y is not, where C 󸀠 (F) is more fine-grained than C(F).¹⁷ (Note that this is analogous to the interpretation of comparative forms in Klein, 1980). (24)

beautiful(x)F,C,std iff sim(x, std, F, C(F))¹⁸


more-beautiful(x,y)F,C,std iff more-sim(x, y, std, F, C(F))

This sketch of evaluative adjectives in multi-dimensional spaces, although very brief, shows how Hare’s idea of criteria can be implemented such that it corresponds to the idea of dimensions and standards known in degree semantics while supporting the affinity of dimensional and evaluative adjectives pointed out in the previous section. It has to be kept in mind, however, that the implementation in multi-dimensional spaces does not capture the evaluative meaning component, that is, the commending function in the case of good and beautiful, which is, following Hare, the primary meaning of these expressions.

3.5 Evaluative update Barker’s (2002) distinction between a descriptive and a metalinguistic, interpretational use of propositions is taken up in Krifka (2012) in order to provide an analysis for definitional generic sentences, which differ from descriptive generic sentences in restricting the language instead of making generalizations about the world (Madrigals are polyphonic is a definitional generic sentence whereas Madrigals are popular is a descriptive one). Krifka introduces an interpretation index in addition to the world index and defines the common ground as a pair of sets of interpretations and worlds ⟨I, W⟩. The meaning of an expression α is relative to interpretation and world, [[α]]i,w . While the world index targets factual matters, the interpretation index targets the metalinguistic contributions of expressions. The difference between descriptive propositions (Madrigals are popular) and definitional propositions (Madrigals are

17 Note that the standard cannot be dropped in the interpretation of the comparative because dimensions need not relate to ordinal/ratio scales. The standard provides the “direction” of the comparative. Since classification sets can be arbitrarily fine-grained the standard will never be “reached”, that is, beauty is still “open-scale”. 18 To be precise, it should be sim(x, μ −1 (std), F, C(F)) because the first two arguments of the sim relation are individuals and std is a point in the attribute space. The problem is technically fixed by postulating a unique Urbild.

Evaluative propositions



polyphonic) is accounted for by providing separate update rules, cf. (26a,b). The first one implements update of descriptive propositions, reducing the set of possible worlds without affecting the set of interpretations. The second one implements update of definitional propositions, reducing the set of interpretations without affecting the set of worlds. Note that the rules are asymmetric: a world w is excluded if the proposition to be updated, [[φ]], is false in w under all possible interpretations, cf. (26a), whereas an interpretation i is excluded if there is a world w such that [[φ]] is false in w under i, cf. (26b). Krifka explains the asymmetry as follows: The existential meaning rule [=26a] reflects the idea that it is not determined yet which interpretation is the one that the participants will ultimately settle on, and so all the options have to be kept open. We interpret a sentence as if it were in the scope of a possibility operator: Under the interpretations that are to be considered, [[φ]] might be true. If a proposition [[φ]] is accepted definitionally at a common ground ⟨I, W⟩, as in (15) [=26b], then the set of possible worlds stays the same, but only such interpretations i remain admissible for which the proposition [[φ]] is true in all possible worlds of the common ground. (Krifka, 2012, 377)


a. b.

⟨I, W⟩ + DES([[φ]]) = ⟨I, {w ∈ W|∃i ∈ I[[φ]]i,w }⟩ (= 14 and 15 in Krifka 2012)¹⁹ ⟨I, W⟩ + INT([[φ]]) = ⟨{i ∈ I|∀w ∈ W[[φ]]i,w }, W⟩

We will adopt Krifka’s system in order to account for the distinction between the descriptive and the interpretational usage of propositions (cf. Section 3.1). The rule in (26a) will be called “descriptive update” (DES) and the one in (26b) “interpretational update” (INT). While the world component of the common ground restricts possible worlds, the interpretation component restricts possible interpretations. This includes the interpretation of non-gradable predicates like chair and madrigal and also the interpretation of gradable adjectives, dimensionals like tall as well as evaluatives like beautiful. Considering evaluative predicates, possible interpretations provide criteria, i.e., dimensions plus standard cut-off values, determining whether an individual counts as an instance of the predicate. For dimensional predicates possible interpretations provide standard values for the single lexically fixed dimension (cf. Section 3.3). The rules in (26) are such that update is either descriptive or interpretational. This is plausible in the case of descriptive sentences as opposed to definitional ones, as considered in Krifka (2012). Definitional sentences involve only interpretational update while descriptive sentences require only descriptive update.

19 In Krifka (2012) the name in the second rule is DEF instead of INT: ⟨I, W⟩ + DEF([[φ]]) = ⟨{i ∈ I|∀w ∈ W[[φ]]i,w }, W⟩. It is changed here to highlight the connection to the descriptive and interpretational usage of propositions.

154 | Carla Umbach

However, in the case of gradable predicates, we want to account for the distinction between descriptive usage and interpretational usage of the propositions they occur in. In Section 3.1/3.2 this distinction was presented as a categorical one— propositions are used either descriptively or interpretationally. This clear-cut division was helpful in the analysis but it is presumably too strict. According to Barker (2002) propositions with gradable predicates generally provide descriptive as well as interpretational information. Moreover, the two-steps analysis of disagreement about evaluative propositions in Stojanovic (2012)—first, disagreement about contextual parameters has to be settled but then there may still be residual disagreement about the meaning of the predicates—seems to go against a clear-cut division of descriptive and interpretational usage. This suggests that in the case of gradable predicates update should be considered as including both descriptive and interpretational parts. In (27) a rule is presented combining descriptive and interpretational update such that descriptive update precedes interpretational update. The reason for that can easily be seen. Consider a common ground ⟨{w1 , w2 }, {i1 , i2 }⟩. Suppose [[φ]]i1 ,w1 , [[φ]]i2 ,w1 and [[φ]]i2 ,w2 are false while [[φ]]i1 ,w2 is true. Applying INT subsequent to DES will yield a common ground ⟨{i1 }, {w2 }⟩. In contrast, applying DES subsequent to INT will yield an empty common ground ⟨{}, {}⟩, which appears inadequate. We thus take it that aligning descriptive information among discourse participants is prior to aligning interpretations, as implemented in (27). (27)

Evaluative update ⟨I, W⟩ + DES ∘ INT([[φ]])⟩ = (⟨I, W⟩ + DES([[φ]])) + INT([[φ]])

4 The pragmatics of subjective judgments The horizontal dimension in the classification proposed in Section 2.3 relates to the semantic difference between descriptive and interpretational usage of propositions, as discussed in the previous section. The vertical dimension relates to the difference between general and subjective judgments, which is a pragmatic difference addressing the role of judgments in conversation. In this section we will first have a brief look at the meaning of finden. Next, the discourse framework of Farkas & Bruce (2010) will be presented and the pragmatics of subjective judgments will be spelled out in this framework.

Evaluative propositions

| 155

4.1 The meaning of finden The German attitude verb finden as well as the English attitude verb find block complement propositions about factual matters. This is why (3), repeated below as (28), is not acceptable. Complements of German finden have long been assumed to be restricted to evaluative propositions, as in the case of English find. In Sæbø (2009) an analysis is presented summarizing Norwegian synes, Swedish tycka, French trouver, English find and German finden under the notion of subjective attitude verbs. Sæbø focusses on the observation that these verbs require a “subjective predicate” (subsuming taste and dimensional adjectives and also deontic modals). Following his analysis, the sole function of subjective attitude verbs consists in shifting the judge such that it is identical to the agent of the attitude verb. The semantics of subjective attitude verbs is spelt out in a relativist framework, with a judge index, and also in a contextualist framework, with an implicit experiencer argument. This yields two different explanations for the incompatibility of subjective attitude verbs with judge-invariant (i.e., non-evaluative) propositions: In the relativist framework their combination would be redundant, mapping judge-invariant propositions onto themselves, and in the contextualist framework their combination would lead to a type clash. (28)

*Ich finde, Osnabrück liegt in Dänemark. ‘I consider Osnabrück to be in Denmark.’

Sæbø’s analysis predicts that embedding a proposition under first person subjective attitude verbs has no effect at all thereby failing to account for the data in (4), repeated below in (29). If embedding under first person finden has no effect, how can it then be explained that it makes the proposition immune against denial? (29)

a. b.

Ann: Ich finde Lakritze lecker. ‘I think licorice is tasty.’ Ben: #Nein! Lakritze ist eklig. ‘No! Licorice tastes terrible.’

In a manuscript of Nouwen (2007) an “impersonal” approach to taste predicates is advocated dispensing with judges as well as implicit experiencer arguments. Attitude reports with find (as a subjective attitude verb) are analyzed analogously to belief reports, by means of a particular set of worlds constituting the agent’s findworlds. An agent’s find-worlds are a subset of his belief worlds selected “based on subjective experience of the world.” (Nouwen, 2007, 6), which is taken to explain why the cat food example (cf. footnote 5) presupposes information from personal experience—if Sam finds cat food tasty, he must have tried it. The idea of find-

156 | Carla Umbach

worlds as a particular subset of belief worlds does not, however, explain the fact that finden/find cannot embed factual propositions (you can have personal access to facts of the world). More importantly, like Sæbø’s account, Nouwen’s account fails to explain the immunizing effect of first person finden. In Reis (2013) an in-depth analysis of German finden is given. Extending what is generally known about finden complements (see the data in (12) in Section 3.1), Reis presents a number of examples from corpora showing that complement propositions need not be judge-dependent.²⁰ What is more, in the case of judgedependent propositions the judge position can be occupied by an agent different from the agent of the attitude verb, cf. (30c,d). Only when focusing on small clause complements, Sæbø’s analysis of finden as a judge shifter is consistent with the data. As for the meaning of finden, Reis does not resort to some form of judgedependence but instead proposes an analysis in which the complement proposition represents one of a number of alternative interpretations: finden [. . . ] projects (i) the presupposition that there is an eventuality known to the finden subject, which is under debate and open to interpretation, (ii) a reading of the complement proposition as an interpretation of this eventuality selected by the finden subject from a set of interpretational alternatives; it is with this selection process that the finden specific subjectivity is to be identified. (Reis, 2013, 389)

Relating her analysis to the notion of interpretation in Krifka (2012)/Barker (2002), Reis is sceptical as to whether it covers her idea of interpretational alternatives, arguing that in some cases interpretational alternatives intuitively appear descriptive, such as the examples in (30) (taken from Reis, 2013). She leaves this question for further research, not commenting on the nature of descriptive interpretational alternatives. (30)


Ich finde, Berta schnarcht. (Reis’ example 37c) ‘I think Berta snores.’

20 German finden may embed complement propositions, dass-complements (ia) as well as V2 complements (ib), and also small clauses (ic). (i)

a. b. c.

Ich finde, dass das Bild schön ist. ‘I think that the picture is beautiful.’ Ich finde, das Bild ist schön. ‘I think the picture is beautiful.’ Ich finde das Bild schön. Lit: ‘I think the picture beautiful.’

Evaluative propositions






Ich finde, der Bezug ergibt sich aus der Überschrift .. . . [. . . ] (Reis’ example (38c)) ‘I think the reference is provided by the title.’ Fandest du auch, dass (für) Max das Verhalten seiner Frau peinlich war? (Reis’ example (39b)) ‘Did you also think that for Max the behavior of his wife was embarrassing?’ Ich finde für ihn war es super, denn er hat viel gelernt [. . . ], aber ich bin mit dem Kindergarten sehr unzufrieden. (Reis’ example (40c)) ‘I think it was super for him because he learned a lot, but I am not satisfied with this kindergarten.’

The analysis in Reis (2013) is closely related to a proposal made for French trouver ‘find’ in Ducrot (1980). Ducrot speaks of value judgments, however the majority of his examples are not about evaluative predicates, but instead about describing a given state of affairs in one of a number of possible ways, as in (31). Ducrot posits a distinction between primary and secondary predication which is reminiscent of the descriptive/interpretational distinction used in this paper. He points out that French trouver is restricted to interpretational uses (his primary predication) and compares trouver to a number of other attitude verbs.²¹ (31)

Je trouve que (c’est) la Grèce (qui) a (en réalité) conquis Rome. ‘I think that it was Greece that in fact conquered Rome.’

Kennedy (2016) claims that there are two kinds of subjectivity induced by gradable predicates. One is attributed to the fact that the positive form of gradables requires a context-dependent standard. The other one is attributed to the lexical meaning of evaluative predicates and is found for positive as well as comparative forms. The first kind of subjectivity is diagnosed with the help of faultless disagreement and occurs with dimensional as well as evaluative predicates. The second kind of subjectivity is diagnosed with the help of compatibility with the attitude verb find in English, and is restricted to evaluative predicates. Thus, in contrast to German finden, English find only licenses complements with evaluative predicates like tasty and beautiful but not complements with dimensional predicates like tall. The difference between descriptive and interpretational usages is not considered in Kennedy’s paper. Still, it would be interesting to see how English find be-

21 I’m very grateful to Renate Hoffmann who generously provided me with a translation of Ducrot’s paper.

158 | Carla Umbach

haves in that respect. It was argued for German finden that complement propositions have to be used interpretationally. Coming back to the phone/beautiful dialogue in (16), Ann cannot use a proposition embedded under finden when she wants to convey descriptive information relating to agreed-on criteria (e.g. that the sculpture does not have a rusty surface, cf. Section 3.2). Can she use a proposition embedded under English find in this case? In explaining the difference between the two kinds of subjectivity Kennedy refers to the difference between quantitative and qualitative interpretations described for color terms in Kennedy & McNally (2005). Dimensional predicates are said to have a quantitative interpretation, whereas evaluative predicates have a qualitative one. It seems plausible to align dimensional predicates with quantitative interpretations—after all, dimensional predicates express a measurable degree of some property. However, in the case of color terms both the quantitative and the qualitative interpretation express a measurable degree of a property, just that the former concerns the area covered by the color and the latter concerns the color value. By aligning evaluative predicates with qualitative interpretations, it is predicted that qualitative interpretations of color words are judgedependent (Kennedy makes use of a judge-dependent account of evaluativity). However, color values are clearly not judge-dependent but instead measured in hue, saturation, and brightness. It is also predicted that qualitative interpretations of color words can occur in complements of find which goes against the data. Still, the line Kennedy draws between the two kinds of subjectivity seems to match with the line drawn in this paper between propositions that are evaluative because they are used interpretationally and propositions that are evaluative because the predicate has no regular denotation (i.e. descriptive usages of propositions includingl beautiful etc.). Thus although the explanation for the two kinds of subjectivity given by Kennedy is not convincing, the distinction itself is supported by the findings in this paper. Coming back to German finden, it will be assumed in this paper at the risk of oversimplification that German finden is sensitive to the interpretational usage of the embedded proposition. As shown in Section 3.1 and 3.3 complements of finden include propositions with:²²

22 One way to spell out the semantics of finden would be this: Assume a particular type ⟨s, ⟨i, t⟩⟩ where i stands for interpretation (recall that propositions are evaluated with respect to a world and an interpretation). Assume, moreover, that the doxastic alternatives of an attitude holder mirror the separation of interpretations and worlds postulated for the common ground. The meaning of finden can then be spelt out analogous to the interpretational update in Section 3.5 such that an agent x stands in the relation finden to φ iff φ is true in every world under each interpretation maintained by the agent:

Evaluative propositions

| 159

(i) the positive form of gradable predicates—groß ‘tall’ as well as schön ‘beautiful’; (ii) the comparative form of evaluative predicates but not that of dimensional ones – schöner ‘more beautiful’ but not größer ‘taller’ (iii) non-gradable predicates like chair; where (i) and (ii) but not (iii) license small clause complements. The characteristics of finden complements in (i)-(iii) above relate to the semantics of finden. In addition, finden has a particular pragmatic effect, namely the blocking of denial effect shown in (32). This is a speech act like effect found only for first person present tense (as with performative verbs). If Ann uses first person finden, Ben cannot directly deny her assertion, cf. (32a). He can only make it obvious that he doesn’t agree by using finden himself, cf. (32b). Direct confirmation is marked, while confirmation using finden again is perfect, cf. (32c,d).²³ To avoid misunderstandings: The blocking effect of finden does not occur with second/third person and past/future tense. Ann’s utterance in (33) is a plain assertion licensing plain denial. (32)

Ann: Ich finde die Skulptur schön. ‘I think that the sculpture is beautiful.’ Ben: a. b. c.


# Nein, sie ist nicht schön. ‘No, it is not.’ Ich finde sie nicht schön. ‘I don’t think it is beautiful.’ ?Ja, sie ist schön. ‘Yes, it is.’

[[finden]] = λx e .λφ ⟨s,⟨i,t⟩⟩ .∀i ∈ π 1 (DOXx ), ∀w ∈ π 2 (DOXx ).[[φ]]i,w where π 1 ⟨I, W⟩ = I, π 2 ⟨I, W⟩ = W

This semantics accounts for the fact that finden complements must be interpretational usages. It does not, however, account for the subjective experience constraint discussed in the literature. 23 Agreement can be expressed without relativization, as shown below. But note that the confirmation particle is not licensed and verum focus or a verum particle like tatsächlich is required. (i)

a. b.

Ann: Ich finde die Skulptur schön. ‘I think the sculpture is beautiful.’ Ben: Sie ist tatsächlich schön. ‘It is in fact beautiful.’

160 | Carla Umbach

d. (33)

Ich finde sie auch schön. ‘I also think it is beautiful.’

Ann: Sue findet die Skulptur schön. ‘Sue thinks that the sculpture is beautiful.’ Ben: Nein, das tut sie nicht. Wir haben gerade darüber gesprochen. ‘No, she doesn’t. We talked about it recently.’

In the remainder of this paper the focus will be on the speech act capacity of finden, that is, on the blocking of denial effect of first person finden. It is considered as evidence that finden sentences do not express general judgments, i.e., regular assertions meant to enter the common ground of the conversation. They instead express subjective judgments indicating mere individual discourse commitments.²⁴ The discourse framework of Farkas & Bruce (2010) makes it possible to spell out the difference between general and subjective judgments thereby accounting for the blocking of denial effect of first person finden. This is shown in the next section.

4.2 The framework of Farkas & Bruce Farkas & Bruce (2010) investigate the use of polarity particles in reacting to assertions and polar questions and suggest a discourse structure extending Stalnaker’s common ground by individual—but public!—discourse commitments. While the common ground includes the discourse commitments shared by the participants, individual discourse commitments are not, or not yet, mutually shared. Assertions are understood as proposals addressing the question under discussion and are added to the common ground only after confirmation by the other discourse participants. Farkas & Bruce define a discourse structure K such that it contains: – a (possibly empty) set of propositions DC X for each participant X—propositions that X is publicly committed to, which are not shared by the participants of the conversation; – a set of propositions cg—common ground—shared by the participants of the conversation;

24 One might wonder how the semantics of finden complements brings about the pragmatic effect of blocking denial. Although a fully specified explanation cannot be given in this paper, it appears plausible that a proposition representing an interpretation of a given state of affairs— recall that complement propositions of finden are restricted to evaluative uses—may remain an individual discourse commitment. In contrast, propositions used descriptively convey information about the world and thus have to enter the common ground.

Evaluative propositions

– –



the table representing the issues to be resolved, implemented as a stack of pairs of syntactic objects and their denotations; a projected set ps comprising future common grounds projected by the elements of the table.

For ease of exposition, the table will be simplified in the remainder of this section omitting the syntactic objects, and we will ignore the projected set. Update rules operating on such discourse structures are specified for polar questions, assertion, confirmation and denial. For example, an assertion is given as an operator turning an input structure K i into an output structure K o such that the asserted proposition p is pushed onto the table and is added to the individual discourse commitments of the speaker, cf. (34a) (see Farkas & Bruce’s example (9)). Confirming an assertion p is done by an operator turning an input structure K i into an output structure K o such that p is added to the individual discourse commitments of the confirming participant, cf. (34b) (see Farkas & Bruce’s example (16)). If p is confirmed by each discourse participant, it will be an element of every individual discourse commitment set and trigger a “common ground increasing operation” moving p from the individuals commitment sets into the common ground, and removing it from the table, cf. (34c) (see Farkas & Bruce’s example (17)). Finally, (total) denial of an assertion p is done by an operator pushing ¬p on top of the table and adding ¬p to the individual discourse commitments of the denying participant, cf. (34d) (see Farkas & Bruce’s example (22)). Denial ends in a “crisis” because neither p nor ¬p can become common ground and be removed from the table. A way out of the crisis is for participants to “agree to disagree”, removing the controversial propositions from the table while leaving them in the individual discourse commitments, cf. (34e) (see Farkas & Bruce’s example (23)). (34)

a. b. c.

Assertion (p, a, K i ) = K o such that DC a,o = DC a,i ∪ {p}, and T o = push(p, T i ) Assertion_Confirmation (b, K i ) = K o such that DC b,o = DC b,i ∪ {p}, where p = top(T i ) and p ∈ DC a,i Common_Ground_Increasing: if DC X,o = DC X,i ∪ {p} for all participants X, then (i) cg o = cg i ∪ {p}, (ii) DC X,o = DC X,i − {p} for all participants X, and (iii) T o = pop(p, T i )²⁵

25 This is a simplification again. The original rules says: “pop off the table all items that have as an element of their denotation an item q that is entailed by cg o ” (Farkas & Bruce, 2010, 19).

162 | Carla Umbach

d. e.

Total_Denial (b, K i ) = K o such that DC b,o = DC b,i ∪ {¬p}, where p = top(T i ) and p ∈ DC a,i Agree_To_Disagree (K i ) = K o such that T o = pop(p, T i ) where p ∈ DC X,I and ¬p ∈ DC Y,I for at least one X and Y

The distinctive feature of the Farkas & Bruce model consists in the representation of the individual discourse commitments of the participants. This idea accounts for the fact that an assertion involves a public commitment of the speaker even if it is not accepted by the other discourse participants, which was already pointed out in Gunlogson (2001). Individual discourse commitments have a twofold task in the Farkas & Bruce system. First they serve as a temporary parking position for propositions waiting for confirmation, and secondly they serve as a representation of controversial information. The latter role is exploited by the agree-to-disagree’ rule in (34e) We will slightly extend the task of individual discourse commitments using them to represent propositions conveyed in subjective judgments, e.g. embedded under first person finden.

4.3 General and subjective judgments in the Farkas & Bruce framework In order to account for the semantic distinction between descriptive and evaluative propositions such that the former reduces possible worlds and the latter reduces possible interpretations, the common ground as well as the individual discourse commitments will be assumed to consist of pairs of interpretations and worlds, ⟨I, W⟩. Update will be done according to the rules in Section 3.5. The distinction between general and subjective judgments is a pragmatic distinction reflecting different intentions of the speaker. General judgments—with descriptive as well as evaluative propositions—are regular assertions waiting for confirmation or denial. If confirmed, the propositions are included in the common ground. If denied, the conversation is in a crisis and the issue under debate (on the table) remains unresolved, unless the participants “agree to disagree”. Subjective judgments present their propositions as mere opinions, not intended to enter the common ground. This is what makes them immune against denial, cf. (32a). The other discourse participant can express agreement and disagreement by using subjective judgments again, cf. (32b) and (32d). In the case of agreement the common-ground-increasing rule in (35c) will shift the proposition from the individual commitments sets to the common ground, as it does when general judgments are confirmed. Once the proposition is in the common ground it cannot be distinguished from propositions that entered the common ground via

Evaluative propositions

| 163

a general judgment. In the case of disagreement the controversial propositions remain individual commitments. Since subjective judgments are not intended to enter the common ground, we will assume that they don’t affect the table. This accounts for the impossibility of denial. At the same time it accounts for the intuition that there is no unresolved issue regardless of whether the participants agree or disagree with a subjective judgment. In order to extend the Farkas & Bruce system such that it covers subjective judgments a rule must be added which is analogous to the assertion rule in (34a) without, however, placing the proposition at the top of the table. This rule will be called “opinion” and is given in (35). Denial and confirmation are ruled out as reactions because there is no suitable proposition on the table. Agreement and disagreement are expressed by uttering another opinion. (35)

Opinion (p, a, K i ) = K o such that DC a,o = DC a,i ∪ {p}, and T o = T i where p is the complement of first person present tense finden²⁶

Let us finally consider the agree-to-disagree rule again. The sequence it was meant to cover in the original system is shown in (36). The crisis which would result from total denial is avoided by taking the issue from the table while leaving the individual commitments unchanged, p ∈ DC A , ¬p ∈ DC B . In (37), a controversy is shown addressing a matter of taste. In this case the crisis which would result from denial is most naturally avoided by rephrasing the controversial proposition as an opinion. Let us call this type of discourse move “retreat to opinion”. As before the controversial issue is removed from the table while leaving the individual commitments unchanged.²⁷ (36)

agree to disagree Ann: Osnabrück liegt in Dänemark. ‘Osnabrück is in Denmark.’ Ben: Nein! Es liegt in Deutschland. ‘No, it is in Germany.’ Ann: Reden wir über was anderes.

26 Subjective attitude verbs in other languages will presumably block denial in the same way as German finden, even if they differ in distribution. 27 One reviewer argued that (37) can be continued by Ann saying O.K., let us discuss our criteria of beauty again. indicating that the controversial issue is not removed from the table. It has to be noted, however, that such a continuation, although possible, is not necessary. Ann might as well start a completely unrelated topic directly after the exchange in (37). This is evidence that when continuing (37) by a request to discuss the criteria of beauty a novel issue is placed on the table which can be picked up or denied by Ben.

164 | Carla Umbach

‘Let’s not pursue this further.’ Ben: O.k. (37)

retreat to opinion Ann: Die Skulptur ist schön. ‘The sculpture is beautiful.’ Ben: Nein! Sie ist nicht schön. ‘No, it is not beautiful.’ Ann: Ich finde sie jedenfalls schön. ‘Anyway, I think it is beautiful.’ Ben: Ich nicht. ‘I don’t.’

Summarizing the handling of general and subjective judgments in the extension of the Farkas & Bruce framework suggested above, general judgments (descriptive uses as well as interpretational uses of propositions) constitute issues to be confirmed or denied. If confirmed, the proposition expressed in the judgment enters the common ground, which in the case of interpretational uses has the effect of narrowing down the set of interpretations instead of narrowing down the set of worlds. If a general judgment is denied, the controversy can be resolved either by agreeing to disagree or by retreat to opinion (provided the proposition is licensed as a complement of finden). Subjective judgments express propositions not intended to enter the common ground and are represented as individual discourse commitments without being placed on the table. This is implemented by the opinion rule in (35) which is the only deviation from the Farkas & Bruce system. Not being on the table, subjective judgments cannot be confirmed or denied. Agreement and disagreement can be expressed by using subjective judgments again. If discourse participants agree with respect to an individual commitment, it will be moved to the common ground. This is the same in the case of general judgments and of subjective judgments. So participants in a conversation can achieve compliance in two different ways: Either they make a general judgment thereby demanding acceptance, or they make a subjective judgment and wait for agreement. Subjective judgments thus provide a by-pass strategy to hopefully achieve compliance without risking denial.

Evaluative propositions

| 165

5 Conclusion Summing up, we started out from the hypothesis that faultless disagreement is a misconception: Denial of evaluative propositions is genuine disagreement. However, evaluative propositions can be presented as mere subjective judgments, making the proposition immune against denial. A semantic and a pragmatic dimension of the problem of evaluativity and faultless disagreement were considered. Semantically, descriptive uses have been distinguished from interpretational (metalinguistic) uses of propositions (cf. Barker, 2002). It was shown that if a proposition occurs as a complement of the German attitude verb finden ‘find’, it has an interpretational use. Four dialogues were examined involving descriptive and interpretational uses of propositions with dimensional and evaluative predicates. Dimensional predicates license, as well-known, descriptive as well as interpretational uses. Evaluative predicates license interpretational uses, which was expected. Surprisingly, however, evaluative predicate also license descriptive uses, thereby raising the question of how an evaluative proposition can be descriptive, given that there is no judge. The puzzle was solved with the help of Hare’s (1952) account of evaluative predicates. Hare combines a primary commending meaning component with criteria dependent on comparison class, context and speaker community which yield a quasi-denotation even if evaluative predicates do not have a regular denotation. A proposal was made on how to capture the idea of criteria in formal semantics using generalized measure functions and multi-dimensional spaces. Update was defined taking the common ground to include worlds as well as interpretations (cf. Krifka, 2012). Pragmatically, general judgments have been distinguished from subjective judgments, which are marked by, e.g. the first person present tense of finden. When using a general judgment, regardless of whether it’s about a descriptive proposition or about an evaluative one, the speaker demands that his statement is included in the common ground of the conversation. When using a subjective judgment the speaker does not demand that it enters the common ground. It presents a mere opinion that may stay an individual discourse commitments in the sense of Farkas & Bruce (2010). This implies that subjective judgments are immune against direct denial and allow only for disagreement in the form of other subjective judgments. The approach presented here does away with the notion of faultless disagreement. Denial in disputes about matters of taste is accepted as genuine denial thereby accounting for the puzzle of the competent speaker. On the other hand, judgments concerning matters of taste can be presented as subjective judgments

166 | Carla Umbach

thereby allowing discourse participants to disagree without one of them being wrong. This possibility might explain the intuition of faultless disagreement. The positions in the literature on evaluative predicates and the intuition of faultless disagreement divide in relativist and contextualist accounts on the one side and metalinguistic accounts on the other. Hare’s theory is not a relativist/contextualist account but at the same time not a purely metalinguistic account. His idea of a primary meaning component, e.g. commending in the case of good, provides a perspective on gradability highlighting the fact that contextual variability is possible only if there is a constant component. Although Hare doesn’t mention dimensional predicates, this perspective strongly suggests that dimensional predicates are a special case of evaluatives in relating to only one dimension which is, moreover, lexically fixed. There is a large number of open issues. As for semantics, the sketch of how to implement Hare’s criteria has to be detailed. This includes the question of how criteria are grounded, which will hopefully lead to overdue distinctions in the field of evaluative predicates—painful and tasty are clearly different in many respects from good and beautiful (and one might come back to Kant’s distinction of the Beautiful and the Agreeable). Another open issue concerns evaluative predicates in attributive positions, which allow for a restrictive and a non-restrictive interpretation. It has been observed that the majority of non-restrictive attributive adjectives are evaluative (Umbach, 2006), raising the question of why evaluative adjectives are more susceptible to non-restrictive interpretation than others. As for pragmatics, disputes about matters of taste have been reduced in this paper to overly simple cases. Gathering data on more realistic disputes is a top priority issue. Finally, there is the puzzle of the nature of the primary meaning component of evaluative predicates, that is, the commending function in the case of good and beautiful. It appears to be beyond the scope of denotational semantics and instead gives the impression of a speech-act-like element. But then, can there be adjectives with a speech act capacity? Acknowledgment: I presented this paper in various stages at the annual meeting of the Deutsche Gesellschaft für Sprachwissenschaft in Berlin in 2010 and at linguistic colloquia in Aarhus, Berlin, Bielefeld, Frankfurt, St. Peterburg and Tezpur. The final version differs considerably from previous ones—I’m grateful to the audiences for their valuable comments and criticism. I would also like to thank the students in my seminars on evaluative predicates in Osnabrück and in Stuttgart for lively discussions and empirical support. I’m particularly grateful to Louise McNally, Stephanie Solt, Isidora Stojanovic, Marga Reis, the editors of this volume Cécile Meier and Janneke van Wijnbergen-Huitink, the editor of the series “Linguistische Arbeiten” Klaus von Heusinger, and two anonymous reviewers for

Evaluative propositions

| 167

constructive and sophisticated criticism and support. This paper is based upon work supported by the Deutsche Forschungsgemeinschaft (UM 100/1-1).

References Anand, Pranav. 2008. Predicates of taste: Context-shift, assessment-shift or something else? Paper presented at the Second International Conference on Quotation and Meaning (ICQM 2) held in Berlin, Zentrum für Allgemeine Sprachwissenschaft (ZAS). Barker, Chris. 2002. The dynamics of vagueness. Linguistics and Philosophy 25(1). 1–36. Barker, Chris. 2013. Negotiating taste. Inquiry: An Interdisciplinary Journal of Philosophy 56(2–3). 240–257. Bierwisch, Manfred. 1987. Semantik der Graduierung. In Manfred Bierwisch & Ewald Lang (eds.), Grammatische und konzeptuelle Aspekte von Dimensionsadjektiven, 91–286. Berlin: Akademie Verlag. Black, Max. 1937. Vagueness: An exercise in logical analysis. Philosophy of Science 4. 427–455. Cohen, Ariel. 2010. A tasty mixture. Unpublished manuscript. Ben-Gurion University of the Negev. Ducrot, Oswald. 1980. Je trouve que. In Danièle Bourcier, Oswald Ducrot & Sylvie Bruxelles (eds.), Les mots du discours: essais, 57–92. Paris: Les Éd. de Minuit. Egan, Andy. 2010. Disputing about taste. In Ted Warfield & Richard Feldman (eds.), Disagreement, 247–286. Oxford: Oxford University Press. Farkas, Donka & Kim Bruce. 2010. On reactions to assertions and polar questions. Journal of Semantics 27(1). 1–37. Gärdenfors, Peter. 2000. Conceptual spaces. Cambridge, MA: MIT Press. Gunlogson, Christine. 2001. True to form: Rising and falling declaratives as questions in English. London: Routledge. Hare, Richard Mervyn. 1952. The language of morals. Oxford: Oxford University Press. Kant, Immanuel. 1878 [1790]. Kritik der Urteilskraft Mit bibliographischen Angaben versehen von Karl Kehrbach. Leipzig: Philipp Reclam. Kennedy, Christopher. 1999. Projecting the adjective: The syntax and semantics of gradability and comparison. New York: Garland. Kennedy, Christopher. 2007. Vagueness and grammar: The semantics of relative and absolute gradable adjectives. Linguistics and Philosophy 30(1). 1–45. Kennedy, Christopher. 2016. Two kinds of subjectivity. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 105–126. Berlin: De Gruyter. [This volume]. Kennedy, Christopher & Louise McNally. 2005. Scale structure, degree modification and the semantics of gradable predicates. Language 81. 345–381. Klein, Ewan. 1980. A semantics for positive and comparative adjectives. Linguistics and Philosophy 4(1). 1–45. Kölbel, Max. 2002. Truth without objectivity. London: Routledge. Krifka, Manfred. 2012. Definitional generics. In Alda Mari, Claire Beyssade & Fabio Del Prete (eds.), Genericity, 372–389. Oxford: Oxford University Press.

168 | Carla Umbach

Lasersohn, Peter. 2005. Context dependence, disagreement, and predicates of personal taste. Linguistics and Philosophy 28(6). 643–686. Lasersohn, Peter. 2009. Relative truth, speaker commitment, and control of implicit arguments. Synthese 166. 359–374. Meier, Kim. 2012. Auflösung von Konflikten aufgrund von evaluativen Urteilen in Dialogen. Stuttgart: University of Stuttgart BA thesis. Moltmann, Friederike. 2010. Relative truth and the first person. Philosophical Studies 150. 187–220. Nouwen, Rick. 2007. Predicates of (im)personal taste. Unpublished manuscript, University of Utrecht. Pawlak, Zdzislaw. 1982. Rough sets. International Journal of Parallel Programming 11(5). 341– 356. Pearson, Hazel. 2013. A judge-free semantics for PPTs. Journal of Semantics 30. 103–154. Reis, Marga. 2013. Dt. ‘finden’ und “subjektive Bedeutung”. Linguistische Berichte 236. 389– 426. Sæbø, Kjell Johan. 2009. Judgment ascriptions. Linguistics and Philosophy 32(4). 327–352. Sassoon, Galit. 2011. Be positive! Norm-related implications and beyond. In Ingo Reich, Eva Horch & Dennis Pauly (eds.), Proceedings of Sinn und Bedeutung 15, 531–546. Saarbrücken: Universaar—Saarland University Press. Stalnaker, Robert. 1978. Assertion. In Peter Cole (ed.), Syntax and semantics, volume 9: Pragmatics, 315–332. New York: Academic Press. Stalnaker, Robert. 2002. Common ground. Linguistics and Philosophy 25(5-6). 701–721. Stephenson, Tamina. 2007a. Judge dependence, epistemic modals, and predicates of personal taste. Linguistics and Philosophy 30(4). 487–525. Stephenson, Tamina. 2007b. Towards a theory of subjective meaning. Cambridge, MA: MIT dissertation. Stephenson, Tamina. 2010. Control in centered worlds. Journal of Semantics 27. 409–436. Stojanovic, Isidora. 2007. Talking about taste: Disagreement, implicit arguments, and relative truth. Linguistics and Philosophy 30(6). 691–706. Stojanovic, Isidora. 2012. Emotional disagreement. Dialogue 51(1). 99–117. Sundell, Timothy. 2011. Disagreements about taste. Philosophical Studies 155(2). 267–288. Umbach, Carla. 2006. Non-restrictive modification and backgrounding. Proceedings of the Ninth Symposium on Logic and Language, Hungarian Academy of Sciences 152 –159. Umbach, Carla & Helmar Gust. 2014. Similarity demonstratives. Lingua 149(A). 74–93. Wolf, Lavi. 2016. Predicates of personal taste and the evidential step. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 69–90. Berlin: De Gruyter. [This volume]. Zangwill, Nick. 2014. Aesthetic judgment. In Edward N. Zalta (ed.), The stanford encyclopedia of philosophy (fall 2014 edition), aesthetic-judgment.

Christine Gunlogson und Gregory Carlson

Predicates of experience Abstract: In this paper we take an evidential approach to the phenomenon of “faultless disagreement” (Kölbel, 2004) in connection with predicates of taste (Lasersohn, 2005). Our claim is that for a broad class of intuitively subjective predicates, not just the taste variety, the experiencer has unassailable evidence as to the truth or falsity of the predication, simply by virtue of being the experiencer. This is not a guarantee that the experiencer’s claims are true, but it makes them, in the usual case, immune from criticism as to their justification. We elaborate on the consequences of this view, and arrive at an evidential conception of the use of predicates of taste and experience that is guided by the dynamics of discourse pragmatics.

1 Introduction In a series of papers beginning with Lasersohn (2005), Peter Lasersohn has drawn attention to the semantics and pragmatics of what he terms “predicates of personal taste”. Paradigmatic examples include fun and tasty, as illustrated below (Lasersohn, 2005): (1)

a. b.

The roller coaster is fun (for me / for you / for Joe). This wine is tasty.

The central problem presented by Lasersohn is reconciling the apparent subjectivity of judgments of taste with the intuition that disputes about taste involve genuinely contradictory propositions. If we look at the discourse in (2), we have little difficulty in concluding that the participants are disagreeing about some matter of fact: (2)

(A and B are looking at a roller coaster) A: That roller coaster was built in 1924. B: No, it wasn’t! (=That roller coaster was not built in 1924.)

Christine Gunlogson, Gregory Carlson, Department of Linguistics, University of Rochester, Lattimore Hall 507, Rochester, NY 14627, USA, e-mail: [email protected] and [email protected]

170 | Gunlogson and Carlson

It is plausible to model the propositions asserted in such a disagreement as being contradictory. That is, the dialogue has a structure that is reducible to: A: p B: ¬p The same would appear to be the structure of the dialogue in (3). However, in this case, Mary and John each seem fully entitled to their position, in contrast to the example above: (3)

(John and Mary just got off the roller coaster) Mary: That was fun! John: No, it wasn’t!

Despite being individually entitled to their positions, Mary and John clearly disagree about the roller coaster ride, an observation that is hard to explain if there is no fact of the matter to disagree about. Lasersohn’s solution in his (2005) paper is to relativize truth values to a judge, introduced into the semantics for all sentences, though inert except in evaluation of predicates of taste. Both John’s and Mary’s statements can be true when relativized to each speaker’s judgment; they remain contradictory in the sense that both cannot be true for a single judge. Where an experiencer argument is overtly expressed in the sentence, as with the for-phrase in (4), the experiencer is obligatorily assigned the role of judge and the sentence has ordinary objective semantics: (4)

(About Bill’s ride on the roller coaster) Mary: That was fun for Bill. John: No, it wasn’t.

In (4), unlike (3), Mary and John disagree in the normal way: one has said something true and the other something false, i.e., one of them is simply wrong about Bill’s experience on the roller coaster. Thus, it is not simply the subject matter or the adjective that is used, but the character of the whole proposition expressed that gives rise to the phenomenon of faultless disagreement. In this paper we will use Lasersohn’s observations as a foundation for our own exploration and expand the picture to include the contrast between (5) and both (3) and (4): (5)

(John and Mary just got off the roller coaster) Mary: That was fun for me. John: #No, it wasn’t.

Predicates of experience | 171

Consideration of a broader range of data leads us to a set of generalizations concerning the availability of direct evidence to each of the interlocutors. We conclude that the crucial differences and similarities between the cases of “faultless disagreement” (Kölbel, 2004) exemplified in (3), ordinary disagreement in (4), and what we will term “faulty disagreement” in (5) (as well as related phenomena that come to light in the course of investigation) are best explained not by direct appeal to truth conditions themselves, but by a generalization concerning the kind of evidence that can be brought to bear on determining truth or falsity of judgments—a matter referrable to independent pragmatic principles. Predications of taste crucially involve an experiencer, who has inherent authority with respect to value judgments about his or her own state. It is not the case that only experiencers can make such judgments, but they do have an evidentiary advantage. Direct experience outweighs all other kinds of evidence and in particular, outweighs testimony from others about similar experiences, as seen in evidential systems cross-linguistically (Willett, 1988). Taking this perspective leads to a number of simplifications and predictions, which we take as support for the position. Below we extend generalizations concerning evidence beyond the boundaries of predicates of “taste”, whatever those boundaries may be, and this too we interpret as a welcome development. We begin by providing some background for an evidential point of view using ordinary predicates. Once this has been established, we then turn to a set of observations regarding fun predications both with and without for-phrase experiencers, showing the systematicity underlying the pattern in (3)–(5) and giving an explanation in evidential terms. We then discuss the applicability of our generalizations to other predicates, developing the evidential point of view further and pointing out a broader range of phenomena that can be explained from this point of view. In the conclusion we summarize our claims and identify some of the central issues that remain unresolved. Throughout the paper we set aside consideration of generic sentences based on experiencer predicates, such as Roller coasters are fun and Ice cream is tasty. These do not induce a reading where the speaker or any other particular individual is an experiencer. All our examples are to be understood episodically, in the context of a particular (imagined) event, unless otherwise noted. Our hope and expectation is that an appropriate analysis of this interpretation will lead to an understanding of the generic.¹

1 According to Roeper (2016) [this volume] generic sentences may be assertions from a default general point of view and Hegarty (2016) [this volume] relates statements with predicates of personal taste to modal sentences.

172 | Gunlogson and Carlson

2 An evidential point of view One purpose of conversation, we assume, is to exchange information. One person has information that another does not have, and the sharing of this information is at least one basic purpose of communication. If Bob knows how to get to his own house and not Barb’s, and Barb knows how to get to her house but not Bob’s, the two via conversation can arrive at a state where both know how to get to one another’s houses. This assumes of course that the propositions used to exchange this type of information are accurate, or, in other words, true. If Bob tells Barb he lives at 123 Maple Street when in fact he lives at 321 Oak Street, no exchange of truths has taken place (or at least, no direct exchange—Barb might use the statement once its falsity has been recognized to infer that Bob is a pathological liar or addle-brained). In the absence of truth, there is no sharing of information qua the normally-assumed conversational purposes. While this would suggest that if we are faithful to those purposes, we would not ever utter a proposition we did not know for certain to be so, in fact this is clearly not how things work. Grice (1975) notes this specifically in his supermaxim of Quality, which does not simply read “make your contribution true”, but rather (in part) “try to make your contribution. . . true”. What drives our willingness to share what we take to be information is our belief that we are, in fact, sharing information. We normally assume that these beliefs have some rational foundation. So, for instance, Bob meets someone at a reception who tells him her name is “Barb Jones”; maybe he even hears several others address her by that name. It seems entirely reasonable for Bob to conclude that the person’s name is, in fact, Barb Jones. So that when Bob talks with Jack, and Jack wants to know the names of various people, Bob tells him that the person by the punch bowl is Barb Jones. This is all mundane. Of course, there can be various circumstances where in fact that is not her name. Maybe she is Julie Jones, Barb’s less famous twin sister, and has given up on correcting people; maybe she’s a spy and does not want anyone to know her name. It is easy to imagine any number of reasons for Bob’s assertion about that person’s name to in fact be false. At the same time, we all agree that Bob is acting normally, that he has adequate basis to assert what he did because he has good evidence that it is so. At no point do we need to think that Bob ever assumed that he knew for certain, just that he had some kind of adequate basis. However, not all bases are equal; some are better than others. For example, between Bob and Barb Jones above, we think we’d all agree that Barb has much better basis for knowing her name than Bob, a stranger, does.

Predicates of experience | 173

We can begin to flesh out this difference in evidential weighting by examining the plausibility of certain types of conversations. Consider the following example. A has just walked in from outdoors moments ago and meets B who is sitting in a windowless sound-proofed basement room in front of a computer. B is looking at the weather radar and notices big green globs on the screen over their location. A has just come in from outside and is standing there completely dry, and even can still see outside via a window down the hall: (6)

B: It’s raining outside. A: No, it’s not.

We take it that there is a disagreement here about an objective fact, as in (2). Both A and B have evidence to support their assertions. However, we clearly judge that A is in the better position to render the verdict in the situation. It’s not that B has no evidence, just worse evidence, and so unless there are some surprises here we have not been informed of (tricks, lies, mirrors, etc.), we’d certainly put more stock in A’s assertion than B’s. In light of this, consider the following discourse: (7)

(A walks in dripping wet from outside. B is in his office looking at the weather radar) A: It’s raining outside. B (checking his computer screen): #No, it’s not.

B’s denial strikes many as rather presumptuous at best. It is not that there is no way to make this into a coherent discourse (perhaps B knows that A gets confused when he walks through lawn sprinklers), but under normal circumstances with equally rational equally sincere interlocutors, B is the one who is behaving somewhat strangely. Now, notice what has changed here between (6) and (7) is not the proposition expressed, but rather a bit of pragmatic information: under what circumstances each interlocutor is offering this information. When there is a disparity between the types of information each has access to in order to support their assertions, we can get the discourse effects we find in (7). Notice that if we change the circumstances so that the interlocutors are on a par with one another, we do not get the sense that such a response is inappropriate: (8)

(A and B have both spent all day in their windowless, sound-proofed basement offices; they are looking at current radar weather on different sites) A: It’s raining out. B (checking his screen again): No, it’s not.

174 | Gunlogson and Carlson

Now, in the case of (8), the disagreement can be resolved by appeal to better and easily obtainable evidence: go outside and look. But, at this point neither A nor B has that kind of evidence, and unless one can show the other that their computer or radar site is faulty, that he’s mistaking ground clutter for rain, etc., they are pretty much in a dead heat as far as evidence is concerned. It is the disparity in the evidence each has that makes for the oddness of (7) and the lack of disparity that accounts for the acceptability of (8). It is not the propositions expressed—they are identical. We have thus far suggested that certain types of evidence are “better” than other types, but have not provided a general characterization of the notion. We are not going to even try to do this in any contentful terms. Rather, we turn to the outlines of overt marking of evidentiality across languages. “Evidentiality” refers to morphological marking that accompanies the expression of assertions in a variety of the world’s languages to provide listeners with information about the source of the information conveyed. For example, if you heard from someone else that John hit a home run in a baseball game, you might report it using one particle, but if you saw him hit the home run, you’d use a different particle. Many languages have no such systematic morphology, including the more familiar ones, but still a great many do. To a decent approximation, the different types of evidentials found are limited to four basic types in world languages (e.g. the survey of Willett, 1988 and the summary in Speas, 2008; see also Murray, 2010). These particles can be used to distinguish claims based on inference, on hearsay, on direct perception (seeing it, hearing it, etc.), and on “direct experience”. There might be subclasses of these in a given category (e.g. seeing vs. hearing), but we assume these four basic types. These form a kind of hierarchy, in that if you use a marker indicating that you gained the information via inference (standing outside a baseball stadium, you heard John’s name announced, and shortly thereafter you see the ball sail over the fence, so “John hit a home run”), it indicates you did not have perceptual evidence (you did not see him hit the home run); however, if you use the marker indicating you saw or directly perceived it, the possibility of being able to infer it from other evidence at hand is not similarly precluded. The perceptual evidence is “better”. We are going to assume that perception and direct experience do not “compete” with one another in this way, both being the most unassailable kinds of evidence available. This does not mean that they are veridical or infallible (clearly, perception can be led astray), only that one cannot get better evidence than this.

Predicates of experience | 175

3 Direct and indirect evidence With all this in mind, we return to predicates of taste, which have as their “best” evidence direct experience. The advantage of direct experience over other types can be clearly seen in (5), repeated below, where John’s attempt to correct Mary’s assessment of her own experience on the roller coaster is strange, not just contradictory. (5)

(John and Mary just got off the roller coaster) Mary: That was fun for me. John: #No, it wasn’t.

Despite the presence of the for-phrase, (5) is not a case of ordinary disagreement. Nor is it “faultless”, and in evidential terms, it is easy to see why not. Under ordinary circumstances, whatever evidence John might have for his position regarding Mary’s state pales in comparison to her direct experience. To the extent that John’s response is acceptable in (5), he can only be understood as claiming some unusual expertise: he considers himself better at gauging Mary’s state of mind than she is. With respect to a third party’s experience, though, John and Mary are on an equal footing, which is why John’s contradictory response to Mary in (4) lacks the strange flavor it had in (5): (4)

(About Bill’s ride on the roller coaster) Mary: That was fun for Bill. John: No, it wasn’t.

Like Lasersohn, we take (4) to be a case of ordinary disagreement. Our explanation, however, does not depend on the presence or absence of an overt syntactically-expressed experiencer but on parity between the evidence available to each side. In (4), neither party has first-hand experience to trump the other’s evidence. Presumably John and Mary base their judgments on indirect evidence such as observations of Bill’s demeanor and familiarity with his inclinations. They may both have reasonable grounds, but only one of them can be correct; the other is simply mistaken. The difference between the unacceptable (5) and the faultless case in (3), then, comes down to this: in (3), John and Mary can be interpreted as each basing their individual assessments on their own individual experiences. The presence of the first person relativizing clause in (5) blocks that interpretation for John’s statement; he must be construed as challenging Mary’s testimony, and the evidence available to him does not suffice for a challenge to Mary’s judgment. (There re-

176 | Gunlogson and Carlson

main questions beyond the scope of this paper, such as why the result should be infelicity rather than coercion to a more appropriate interpretation, and how the underlying evidential principles should be understood. We touch on these briefly at the end of the paper.) In basic outline, however, the evidential story is quite straightforward and has, as we’re about to see, substantial explanatory power. Incorporating basic observations about direct evidence allows for an alternate explanation of (5) that marshals independently needed principles and requires no novel semantic machinery. To begin, we note that the evidence-based perspective brings to light an important assumption lurking in the description of (3): that the parties to the faultless disagreement have equally strong direct evidence. Compare the circumstances of the exchange in (9) to those in (3): (9)

(Mary just got off the roller coaster; John is standing by the exit waiting for her) Mary: That was fun! John: #No, it wasn’t!

(9) has much the same feel as (5), though it is identical to (3) in terms of words uttered and notably in the absence of a for-phrase. In (9), like (5), John must be taken as attempting to correct Mary’s judgment about her own experience, and this is a distinctly odd thing for him to do in the circumstances. While John may have plenty of roller coaster rides in his past to justify his own judgment of roller coaster rides, he cannot reasonably expect Mary to disregard her own experience to accept indirect evidence about his. To the extent John’s response can be made sensible, it suggests he is talking about his own direct experience of some related but different eventuality. For example, one could, with not too much of a stretch, understand him as complaining about what he was doing during the same interval that Mary was riding, namely standing around waiting. This merely reinforces the point: John’s response is comprehensible to the extent that it can be understood as something he does have direct experience of; or, in a somewhat longer reach, as a claim that he is somehow more qualified to judge Mary’s reactions than she is. In general, though, John’s response is not easily coerced or accommodated into a reading where his intention is to convince Mary that she is mistaken about her ride. It seems to come up against a relatively hard constraint on interpretation imposed by the primacy of direct experiential evidence. Mary’s rejection of John’s claim in favor of her direct evidence is very nearly a foregone conclusion.

Predicates of experience | 177

Summarizing the observations so far in terms of evidential configurations, we have the following: Table 1. Evidential match-ups Disagreement type


Mary’s evidence

John’s evidence


Faultless Ordinary Faulty

(3) (4) (5) (9)

Direct Indirect Direct

Direct Indirect Indirect


Now, if it is correct that the evidential match-ups posited above are significant, and that faulty configurations systematically lead to anomaly, then we should find reflexes not just for disagreement but for a variety of conversational moves. This is in fact the case, as we will now demonstrate.

3.1 Stating and questioning Just as it was strange for John to try to correct Mary about whether she had fun, so it is strange for him to attempt to inform her that she had fun, as in (10a). The corresponding polar question in (10b), by contrast, is fine: (10)

(John to Mary, both just off the roller coaster) a. b.

#That was fun for you. Was that fun for you?

The basics of the above pattern follow readily from straightforward assumptions about informing, questioning, and direct experience. Assuming Mary’s and John’s individual experiences grant each of them the highest possible authority as to whether they enjoyed the roller coaster or not, then it is reasonable for John to address a question to Mary about her experience, as in (10b), and not so reasonable for him to attempt to inform Mary, as in the most accessible reading of (10a).²

2 Note that tag questions such as That was fun for you, wasn’t it? are also fine here, and their patterning in general shares characteristics of statements and questions, as is natural if we assume that the tag question construction has both a declarative or “assertive” component and an interrogative one, as its form suggests. It does not seem to be the case that John is prohibited from drawing conclusions about Mary’s experience (the declarative component) just because she is better qualified to do so. Rather, it seems that John’s statement must be tempered by the ac-

178 | Gunlogson and Carlson

For concreteness in spelling out this suggestion we will use the terminology of Gunlogson (2008), who employs the notion of source in characterizing contextual states. Under Gunlogson’s proposal, use of a declarative sentence normally commits its author to the propositional content, where commitment is a narrower notion than assertion in the sense that it simply forecasts the unavailability of certain developments in the discourse without building in any propositional attitudes or intentions. From the fact of a speaker’s commitment, however, together with general principles such as (a version of) Grice’s maxim of Quality, attitudes and intentions can readily be inferred. Mediating such inferences is the (pragmatic) necessity of having a plausible source for commitments undertaken by each interlocutor, encoded in the simple interpretive principle below (from Gunlogson’s (36)): (11)

Source Principle: Commitments have sources.

The definition of source is given in (12) (from Gunlogson’s (27)): (12)

An agent α is a source for a proposition φ in a discourse d iff: a. b.

α is committed to φ; and according to the discourse context, α’s commitment to φ in d does not depend on another agent’s testimony that φ in d

Nothing said so far requires a speaker to be a source for everything he or she says; conversation would not work as well as it does for information exchange if people could only bring up matters they were already prepared to testify to. Information questions, in fact, can be understood as attempts to get one’s interlocutor to contribute material for which the questioner is ill-equipped to be a source. In the framework under discussion, it is possible as well for a commitment to be undertaken by a speaker with the prospect of getting the addressee to enroll as a source. In general, this may result in both speaker and addressee being recognized as independent sources; or it may be that the speaker contributes a declarative as candidate for a dependent commitment, with the implicit understanding that it will be withdrawn if not ratified by the better-situated addressee. (Refer to Gunlogson, 2008 for a fuller discussion of the possible configurations.)

knowledgment of Mary’s status as expert and request for confirmation, which the interrogative tag plausibly provides. It is as if the tag makes clear that John is only licensed in his conclusion if Mary ratifies it. Despite their interest we will set aside these complexities for current purposes and leave tag questions out of the mix.

Predicates of experience | 179

It is this latter case that we suggest is most relevant to the situation in (10). That is, given the acknowledged superiority of direct experience on the evidential scale, Mary must be the ultimate source for any evaluation of her experience. We need one more piece from Gunlogson (2008) to sketch a hypothesis about the contrast between (10a) and (10b) (from Gunlogson’s (39)): (13)

Rule of Initial Commitment: A speaker making a discourse commitment to φ in a context neutral with respect to φ is expected to be a source for φ.

The applicability of (13) to (10a) is fairly clear. The expectation that John is a source for the proposition that Mary had fun on the roller coaster follows from (13). Whether or not she actually did have fun, and whether or not John has some reason to think so, this proposition cannot be positioned in the discourse as contributed on John’s authority. The infelicity of (10a) is the result.³ Returning to (10b), we assume with Gunlogson (2008) that use of a polar interrogative has at a minimum the effect of signaling that its author is not a (current or prospective) source for the content. If the propositional content corresponding to the interrogative is to become common knowledge, therefore, it can only do so with someone other than the speaker as source. Mary, the addressee, is the obvious candidate and in fact the only person eligible in this case. Naturally enough, the pattern reverses when John is making reference to his own experience using first person singular pronouns. (14)

(John to Mary, both just off the roller coaster) a. That was fun for me. b. #Was that fun for me?

Again, this follows under the noncontroversial assumption that questions qua requests for information are expected to be addressed to a party supposed to be more knowledgeable than the questioner. Since John is asking about a matter on which he is the ultimate authority, (14b) is odd, while (14a) is perfectly normal. Or, in the source terminology, since John is the only interlocutor with authority to be the source, attempts to alienate that authority by using an interrogative lead to infelicity.

3 Again, why infelicity rather than coercion is the result remains open, though see Gunlogson (2008) for some relevant observations about similar results not involving taste predicates, suggesting at least that the phenomenon is general.

180 | Gunlogson and Carlson

Interestingly, a first person plural experiencer is anomalous in both cases: (15)

a. #That was fun for us. b. #Was that fun for us?

The infelicity in (15) follows under the assumption that neither party has a direct joint experience. Each individual is privileged only with respect to their own experience. Since use of us under the described circumstances necessarily includes both parties, it violates the same constraints as (10a) and (14b), respectively. That is, the attempt in (15a) to testify on behalf of another direct experiencer about the experience is the same sort of anomaly as (10a) while (15b) parallels the oddness of (14b). In the case of a third-person experiencer, where the best evidence possible is necessarily indirect, there is no inherent anomaly in using either a statement or a question: (16)

(John to Mary about Bill’s ride on the roller coaster) a. b.

That was fun for Bill. Was that fun for Bill?

This is as expected. Whether a statement or a question is more suitable will depend on the common ground between John and Mary concerning the knowledgeability of each about Bill. Note that in general, we do not expect both a question and a statement with the same descriptive content to be felicitous for the same speaker in the same context, as the preconditions for stating vs. questioning are quite distinct. Usually if one is in a position to state that φ, then the circumstances do not favor asking whether φ, and vice versa. With the statement John lays claim to knowledge about Bill, while with the question he seems to disown it. So only one of the two possibilities in (16) is likely to be felicitous in a given context. The point is, it could be either one. This is quite different from (10) and (15), where the #-marked forms are strange in ordinary circumstances and only escape strangeness in unusual ones. Summarizing so far, both statements and questions involving fun are sensitive to the presence or absence of direct evidence.⁴ Statements about the speaker’s

4 Tenny (2006) observes that predicates of direct experience in Japanese restrict their experiencer subjects to first person in the declarative and second person in the interrogative. She argues that in Japanese it is actually a grammatical restriction, and not just a pragmatic one. Thus, *You are cold is ungrammatical whereas Are you cold? is fine; I am cold is grammatical, but *Am I cold? is ungrammatical. Our analysis of English is that this similar pattern is due to pragmatics and is not, in English, a grammatical restriction.

Predicates of experience |


own direct experience, as in (17a), are routine, as are statements based on indirect evidence about someone else’s experience, as in (17c). For questions, it is the addressee’s possession of direct evidence that makes for an appropriate question, as in (18b), or the mutual lack of direct evidence as in the third party case in (18c). The use of an inclusive first person plural pronoun is not suitable for either questioning or stating if the interlocutors both have direct experience. (17)

The roller coaster was fun. . . a. for me / #for us b. #for you / #for us c. for Bill


Was the roller coaster fun. . . a. #for me / #for us? b. for you / #for us? c. for Bill?

The fact that the evidential account straightforwardly generalizes to questioning gives it a significant advantage over a semantically-oriented judge approach, under which questions require a special “exocentric” default. Since the default judge is normally the speaker, questions must be stipulated to be something out of the ordinary where the addressee becomes judge by default. The stipulation, moreover, does not lead to correct results, since whether the addressee is the best judge depends on the nature of the evidence and the situation and not just the choice of sentence type (and/or predicate). For now, one important indicator that the evidential approach is on the right track is the change we see in the patterning when the predication is predictive rather than reported, i.e., when the predication is about a future event, as in (19)– (20): (19)

The roller coaster will be fun. . . a. b. c.


for me / for us for you / for us for Bill

Will the roller coaster be fun. . . a. b. c.

for me / for us? for you / for us? for Bill?

182 | Gunlogson and Carlson

No one (leaving aside time travelers) can have direct experience now of roller coaster rides in the future. This simple observation accounts for what would otherwise be a rather mystifying contrast.⁵ A question such as (20a) requires the addressee to have some specialized knowledge of the questioner’s dispositions, to be sure, as well as some familiarity with the roller coaster. But that is just a particular case of the more general tendency of questions to be addressed to those with qualifications to provide answers. The contrast in acceptability between (19)-(20) versus the infelicitous versions in (17)-(18) is sharp. It may be thought that the patterns documented above can be reduced to a simpler generalization that pronominal arguments to questions are more likely to be second persons. While it is true that second person pronouns are quite common in questions, the more general explanation is still the availability of evidence. For example, while the first person is odd in (21a), it is perfectly fine in (21b), where the doctor’s expertise bears more directly on the diagnosis than subjective experience: (21)

a. b.

Do you / #Do I have a headache? (To a doctor) Do I have pneumonia?

For predicates of appearance, as in (22), the person presenting the appearance is not authoritative at all: (22)

Do #you / Do I look OK?

Furthermore, (21) and (22) retain their characters if the speaker describes their state of wonderment rather than posing a question: (23)

a. b.

I wonder whether #I/you have a headache. I wonder whether I have pneumonia.


I wonder whether #you/I look OK.

The point is simply that acceptability rides on who is in position to have the appropriate sort of evidence, given the nature of the predicate and its arguments, sentence type, and circumstances of utterance—it is not a rule about pronoun choice or sentence type.

5 In a similar vein, Lasersohn (2005, 672) notes the need for exocentric assignment of a judge for a future predication.

Predicates of experience | 183

3.2 Faultless and faulty agreement Suppose that Mary, for one reason or another, upon disembarking from the roller coaster, announces that the ride was fun for her in circumstances where she didn’t in fact enjoy the experience in the least. Her statement is simply false. This supposition changes the status of John’s dissent in (5), repeated below, very little. (5)

(John and Mary just got off the roller coaster) Mary: That was fun for me. John: #No, it wasn’t.

It remains odd, in the ordinary course of events, for John to correct Mary’s pronouncement about her own experience.⁶ It is difficult for a non-experiencer to challenge the testimony of an experiencer even if the latter is being transparently inaccurate. Transparently inaccurate, that is, with respect to all external indications—and that of course is the nub of the issue. There is just no obvious way for the non-experiencer to mount an effective challenge to the experiencer, even if she all but admits she is lying. This observation is consistent with the idea that the faultiness or faultlessness of John and Mary’s positions is better characterized by an appeal to evidence than to truth conditions. If Mary wants to insist that she had fun, it’s hard for John to make the case against it even if they mutually recognize that she is being insincere. The point emerges even more clearly when we consider that the effect is not limited to challenges and disagreements—faultiness can be found with affirmation and agreement as well. First consider (25), mirroring (3), where John and Mary are understood to be in agreement about their roller coaster experiences: (25)

(John and Mary just got off the roller coaster) Mary: That was fun! John: Yes, it was!

Crucially, John’s agreement in (25) makes reference to his own experience, not Mary’s. We can confirm this by noting the contrast between (25) and (26), where

6 As noted earlier, there are ways to alleviate the oddness. John’s correction improves in acceptability to the extent that it is plausible he has firm ground for judging Mary’s actual experience from his own external viewpoint. This would require not only familiarity with Mary’s typical behavior and inclinations, as with the third party judgments in (4) and (16), but a claim to better comprehension of her experience (or more honest reporting) than she herself has put forward.

184 | Gunlogson and Carlson

the assumption that both parties have direct experience is dropped. The parallel with the contrast between (3) and (9) is exact: (26)

(Mary is just off the roller coaster; John is waiting for her) Mary: That was fun! John: #Yes, it was!

If Mary’s statement is explicitly about her own experience, as in (27), the same kind of anomaly arises with John’s agreement that we saw for (5): (27)

(John and Mary just got off the roller coaster) Mary: That was fun for me! John: #Yes, it was!

If John wants to join in Mary’s enthusiasm about the ride, he must say something more along the lines of (28): (28)

(John and Mary just got off the roller coaster) Mary: That was fun for me! John: For me too!

Completing the picture, John’s agreement with Mary about someone else’s experience, as in (29), is very like the ordinary disagreement case of (4): (29)

(About Bill’s ride on the roller coaster) Mary: That was fun for Bill. John: Yes, it was.

Accounting for faultless and faulty agreement as illustrated above requires a basic assumption about the function of the affirmative particle yes and its variants. Without making a detailed argument, we will turn again to Gunlogson (2008), who observes that yes does more than indicate agreement or acceptance of some content. In uses like those above, yes conveys that the responder is affirming the stated content based on his own judgment, independent of the testimony just offered by his interlocutor. In other words, yes registers the agreeing party as a source for the content agreed to (though not necessarily the sole source).⁷

7 There is an anaphoric component involved in the interpretation of yes (as well as many uses of no) which deserves more attention than we are able to give it here. Spelling out how a polar particle relates to preceding content of course requires explicit assumptions about what that content is, which is at the center of the debate about taste predicates. There may well be insights into how taste predicates work to be gained from a more careful examination of their interaction with polar

Predicates of experience | 185

Support for this view comes from the observation that yes is odd when combined with a claim not to know or with a confirmatory question, as in (30a), but consistent with the claim to knowledge in (30b): (30)

Mary: The roller coaster is closed today. John: a. #Yes, I didn’t know that. / #Yes, it is? b. Yes, I know. / Yes, that’s right.

Receipt of new information, as in (30a), is more felicitously marked with oh (Heritage, 1984). (31) differs from (29), for instance, in conveying that Mary’s report about Bill’s experience is news to John: (31)

Mary: That was fun for Bill. John: Oh. (I didn’t know that.)

These observations lead us to expect that an oh from John will be strange in certain circumstances, in particular the situation of (25), where he is a direct experiencer. This expectation is borne out: (32)

(John and Mary just got off the roller coaster) Mary: That was fun! John: #Oh. (I didn’t know that.)

An explanation of the faultless and faulty agreement cases above follows readily on this view. John is appropriately qualified to affirm Mary’s claim in the same circumstances in which he can appropriately challenge her claim—namely, when he has evidence on a par with hers. When Mary has direct experience and John does not, anomaly results whether he attempts agreement or disagreement. And when John is a direct experiencer, he does not normally have the option of accepting Mary’s assessment as enlightenment about his own experience (except perhaps under special circumstances).

response particles. The point we wish to emphasize here, however, is that there is a more general phenomenon involving effects of yes (and no) to which taste predications as well as others are subject. The notion of source we rely on here is crucially pragmatic, operating we assume at the level of Grice’s Quality maxim rather than being implemented in the semantics. It is thus quite different in nature and broader in application than a Lasersohn-style judge parameter, since concerns of truth-telling and adequacy of speakers’ evidence figure in discourse generally, not just with certain predicates.

186 | Gunlogson and Carlson

3.3 Must and fun As a final set of data we believe our analysis sheds light on, we consider the use of epistemic modals to indicate evidentiality in English (see Stephenson, 2007; Pearson, 2013, for interactions with taste predicates). Use of a modal such as must with ordinary predicates (The dog must be in the yard) suggests that perceptual support for the claim is lacking. With predicates of taste, the effect of must is to cancel the presumption of experience. So, suppose one licks an ice cream cone and utters ?That must taste good!. Since the ice cream cone has clearly been licked, the most direct type of evidence is present, and to implicitly deny it by using must is decidedly odd. This follows from our approach, given that must has the effect of removing the presumption of direct evidence. Following von Fintel & Gillies (2010), we take it that the modal’s presence indicates that the evidence is indirect, most likely supported by inference. Returning to the roller coaster, we confirm that evidential marking is strange when the speaker of the claim is the experiencer, whether or not the experiencer is expressed: (33)

(Mary after her roller coaster ride) a. That was fun (for me)! b. #That must have been fun (for me)!

As (33) shows, speakers do not have the option of setting aside direct evidence in order to base their testimony on less-privileged evidence, such as inference. This is in keeping with the character of overt marking of evidentiality, as we remarked earlier. Furthermore, the patterns we have considered so far suggest the existence of a strong interpretive principle that if a speaker’s judgment could, in principle, be based on direct experience, and the speaker does nothing to indicate otherwise, then the speaker will be understood as having direct experiential evidence. It also follows that interlocutors who lack the direct experience will be able to use must in expressing inferences about others’ experiences based on observations of behavior; and this too seems correct. One interesting consequence of this view is that agreement and disagreement do not seem to be on a par. We observe this but do not offer an explanation at this time. The generalization seems to be that disagreement requires the dissenter’s evidence be at least as high-ranking as the original claim, and probably better. Agreement, on the other hand, is more tolerant. If Mary’s initial claim is not marked evidentially, and John’s response is so marked, as in (34) below, the dissent is clearly marked as inferior to Mary’s claim

Predicates of experience |


and odd; agreement is fine under the circumstances, even if based on weaker evidence: (34)

Mary: That was fun for me. John: #No, it must not have been. / Yes, it must have been.

In (34), John’s use of must acknowledges that he is not basing his position on direct experience. His attempt at dissent is therefore odd; the use of must conflicts with the implicit claim that his evidence must be somehow better to dispute it, in order to voice the dissent in the first place. The agreeing response is fine, however. In the case of (35), Mary first makes a statement with the modal, signaling a lower level of evidence, while John’s dissent is not so marked, allowing for the possibility of his evidence being somehow better. Even if it is in fact not any better, there are no contradicting signals in the dissent, as in the case above, and John’s dissent is fine, as well as his agreement: (35)

Mary: The roller coaster must have been fun for Bill. John: No, it wasn’t. / Yes, it was.

Finally, if both interlocutors use the modal, as in (36), the implication of an affirmative response without presumption of any better evidence is fine (perhaps both just watched Bill on his ride), but disagreement under the circumstances is once again strange. (36)

Mary: The roller coaster must have been fun for Bill. John: #No, it must not have been. / Yes, it must have been!

With regard to the disagreement in (36), it does not seem impossible in principle that two agents could arrive at different conclusions based on the same evidence, or that one party could have more and/or better evidence than what is available in the discourse situation. Still, the attempt to disagree does not on its own readily convey that John has such superior evidence. This sort of disagreement seems to necessitate further explanation and discussion of the evidence and inferences. Suppose John continued as follows: (37)

John: No it must not have been—normally Bill does like roller coasters but when I saw him earlier he wasn’t feeling well, and he still looks very pale. I’m surprised he decided to go.

Although (37) is perhaps not perfect (the combination of negation and the modal remains somewhat awkward), the additional explanation does seem to make the disagreement more palatable than in (36).

188 | Gunlogson and Carlson

In the absence of evidential marking by either, as in (38), we are back to the case of ordinary (dis)agreement, where both dissent and agreement are acceptable (as we would predict) and neither party marks their evidence as inferior. (38)

Mary: The roller coaster was fun for Bill. John: No, it wasn’t. / Yes, it was.

In sum, there is a fairly complex pattern of data on a number of fronts that appears amenable to an evidential analysis of the type we are pursuing here. On a more speculative note, we hold out some hope for the evidential approach to address a deeper question about epistemic modality. We have assumed, following von Fintel & Gillies (2010), that the connection between epistemic must and reliance on indirect evidence is lexically stipulated. But as the authors point out, this is less than satisfying given that similar patterns are widely attested cross-linguistically. From an evidential point of view, however, there may be a way to understand this linkage as more definitional than stipulative. Let us suppose that evidence from direct experience is only available in this world, the one we’re in. It certainly seems reasonable to claim that one cannot directly experience other worlds, whatever one takes the status of those other worlds to be. Putting it another way, direct experiences are, by definition, experiences of the way things are (and not primarily the way they could be, or should be, or will be. . . ). Access to representations of things other than the way they are—i.e., to modal representations—is then inevitably based on indirect evidence. From this perspective, judgments that are evaluated against modal bases may be strong in the sense of invoking necessity, like must. But they can never be at the top of the scale in evidential terms, and thus will not exhibit the immunity to challenge that direct experience confers. We leave this suggestion for further research.

4 Generalizing the experience The patterns we have discussed so far have been exemplified almost exclusively by the predicate fun. This is in part because the link between fun and experience is particularly strong, even stronger we believe in fundamental ways than with tasty, generally assumed to be in the same class. We speculate that fun belongs to a subclass of predicates that take as an argument a particular sort of eventuality (as also suggested by Pearson, 2013), an experience, and we comment briefly on some relevant differences between fun and tasty. Our present concern, though, is to outline why we believe that the phenomena we observe so far, including fault-

Predicates of experience | 189

less disagreement, can best be understood in light of a pragmatic account. This contrasts in important ways with the view expressed in Lasersohn (2005). However, there remains a need to posit propositional structures for predicates of taste that are qualitatively different from other propositional structures. Though Lasersohn includes a judge parameter for all predicates, recall that this parameter is inert in all instances apart from predicates of taste, thus distinguishing this class from the remainder. Even on a pragmatic account such as the one set forth here, there is a need for the semantics to play some kind of role in determining just what types of evidence bear most directly on the truth or falsity of certain propositions arising from various types of predication and not others. Lasersohn’s proposal that distinguishes active from inert judge parameters could be seen in this light as a means of distinguishing those propositions that require an experiencer. We aim to show that across a broad range of predicates involving an experiencer, expressed or not, the generalization holds that what matters is whether or not the interlocutors have access to the strongest form of evidence, namely direct experience. Tasty is included in this generalization but it extends to predicates not easily classified as involving taste, such as be cold, which plausibly all allow for assessment of an experience by an experiencer.⁸ Our goal is to establish that the appropriate distinctions are not structural ones singling out certain classes of predicates and hinging on the presence nor absence of an overt experiencer. Rather, this is a fully pragmatic phenomenon.

4.1 Beyond taste Recall the central cases of faultless, ordinary, and faulty disagreement discussed earlier: (3)

(John and Mary just got off the roller coaster) Mary: That was fun! John: No, it wasn’t!

8 We do not claim there are no important syntactic, semantic, and/or pragmatic distinctions between the classes of predicates mentioned. It seems to us, in fact, that there are important distinctions in behavior even between the prototypical taste predicates tasty and fun. Relevance of evidence is a general concern in language use. What is special, we suggest, is the privileged position of direct experience and how the potential for direct experience interacts with the semantics. Tasty (and red, e.g.) might seem intuitively to be as experiential as fun, but in linguistic terms they seem to behave rather differently. It would certainly be helpful to say more about those differences but we are not able to do so in this paper in any detail since we have not thus far been able to arrive at a taxonomy that is remotely satisfactory to us.

190 | Gunlogson and Carlson


(About Bill’s ride on the roller coaster) Mary: That was fun for Bill. John: No, it wasn’t.


(John and Mary just got off the roller coaster) Mary: That was fun for me. John: #No, it wasn’t.

To recap briefly, (3) is the case of faultless disagreement, where Mary and John seem to be saying something contradictory at some level, but where the position of each appears to be unimpeachable, i.e., faultless. (4) exemplifies ordinary disagreement, where Mary and John are discussing the experience of a third party, Bill, explicitly specified in the for-phrase. The last case, (5), is faulty disagreement: Here an experiencer is overtly given by the for-phrase, as in (4), but the specified experiencer is first person. The anomaly of John’s disagreement in (5) results, we claim, from his attempt to challenge Mary’s judgment based on her own direct experience (even though John has his own direct experience to fall back on). It is similar in nature to (9), where John lacks such direct experience: (9)

(Mary just got off the roller coaster; John is standing by the exit waiting for her) Mary: That was fun! John: #No it wasn’t!

Consider now the predicates cold and exhausting in the uses exemplified below. (39)

(While cross-country skiing) Mary: It’s cold out today. / This is exhausting! John: No, it’s not. / Yes, it is.

In (39), no experiencer is expressed, similar to (3). Are these cases of faultless disagreement/agreement as well? Note that cold and exhausting do not seem (intuitively) to be predicates of taste. For cold especially one might argue that the disagreement is one of fact (albeit subject to some evaluation), more like (2), repeated below, than like (3): (2)

(A and B are looking at a roller coaster) A: That roller coaster was built in 1924. B: No, it wasn’t! (=That roller coaster was not built in 1924.)

The important comparison is where a first-person experiencer is overtly given, as in (40):

Predicates of experience | 191


Mary (shivering): I’m cold. / I’m exhausted. John: #No, you’re not. / #Yes, you are.

Like the faulty disagreement case in (5), John’s attempt to disagree (or agree) with Mary’s judgment based on her own direct experience of the conditions is strange. Disagreement or agreement about a third-party experiencer, as in (41), poses no such difficulty, as expected on analogy with (4). (41)

(Watching the cat burrow under a blanket) Mary: Poor Sylvester! He’s cold. / He’s exhausted. John: No, he’s not. (He’s just playing games.) / Yes, he is.

The lesson we draw from the above is that what matters to the faultlessness or faultiness of judgments of experience is whose experience, what kind, and how the evidence available to each interlocutor stacks up. The common factor in all these cases, we submit, is a predicate that can be experiential. Taste predicates may well qualify, but the point is that the potential for direct experience is what gives rise to the patterns in question.⁹ In (40) and (41), the experiencer is expressed as the syntactic subject, a different structure from the fun predications we have been considering so far. Hard resembles fun in having an optional for-phrase for the experiencer. Though hard is not so evidently a predicate of “taste”, it patterns just like fun, as seen in (42)–(44). Assume that Mary and John are discussing an exam both have just completed: (42)

Mary: That was hard. John: No, it wasn’t. / Yes, it was.


Mary: That was hard for me. John: #No, it wasn’t. / #Yes, it was.


Mary: That was hard for Bill. John: No, it wasn’t. / Yes, it was.

The same patterns of faultless, faulty, and ordinary disagreement are reproduced above for hard, though the status of the predicate as involving “taste” is doubtful.

9 While we focus on direct experience in this paper as a source of faultless disagreement, we leave open the possibility that there may be similar phenomena that do not involve an experiencer. For instance, if A says “It may rain tomorrow” and B says “It’s not true that it may rain tomorrow”, this may constitute an instance of “faultless disagreement” even in the absence of direct experience (and, in this case, we would suggest it is inferential evidence that constitutes the “best possible” evidence). If such is the case, it would be entirely compatible with the approach taken here.

192 | Gunlogson and Carlson

If this story is on the right track, we ought to observe generalization of other patterns noted in this paper for the larger class of predicates allowing for experience. Agreement patterns paralleling the disagreement observations have already been shown in (40)–(44). We briefly exemplify a couple more patterns below: (45)

Patterns involving questions, statements, first and second person: a. b.


Mary: I’m cold. / # Am I cold? / Is Sylvester cold? John: #You’re cold. / Are you cold? / Is Sylvester cold? Mary: The test was hard (for me). / #Was the test hard (for me)? / Was the test hard for Bill? John: #The test was hard (for you). / Was the test hard (for you)? / Was the test hard for Bill?

Neutralization of question/statement contrasts with predictions about future experience: a. b.

Mary: The test will be hard (for me). / Will the test be hard (for me)? John: The test will be hard (for you). / Will the test be hard (for you)?

The reader is invited to return to the generalizations of Section 3 and test their robustness at more length.

4.2 Argument alternations The membership of the “taste” class is not altogether certain, and there is as far as we know no standard set of diagnostics. The fun predications we have been considering feature a syntactic subject that is loosely the producer of the evaluated experience, with the experiencer optionally expressed in a for-phrase. Other predicates that appear to work similarly are shown in (47): (47)

The roller coaster was enjoyable / scary / a blast / boring. . . for me / for you / for us / for Bill.

Intuitively, these predicates seem to meet the criterion of subjectivity and allow for faultless disagreement (and agreement), just like fun: (48)

(John and Mary again, exiting the roller coaster) Mary: That was quite enjoyable / scary / a blast / boring. John: No, it wasn’t! / Yes, it was!

Predicates of experience |


Many such predicates also come in other flavors in which the experiencer is the subject (and hence required), as in (49a-b), or is otherwise obligatorily expressed, as with the direct objects in (49c): (49)

a. b. c.

Mary had fun / a blast on the roller coaster. Mary enjoyed/detested the roller coaster. The roller coaster ride bored/thrilled/scared Mary.

Predications like those in (49) have overtly expressed experiencers, and so would straightforwardly (we presume) receive a judge assignment in the semantics under a Lasersohn-style approach. Such an approach works as expected for thirdparty experiencers: (50)

(Talking about Bill’s ride) Mary: Bill enjoyed the roller coaster. John: No, he didn’t! / Yes, he did!

(50) is just a case of ordinary disagreement or agreement, like (4). But when we turn to (51), where John tries to disagree with Mary’s assessment of her own experience, the by-now familiar patterns of faulty disagreement and agreement reappear: (51)

(John and Mary, exiting the roller coaster) Mary: I enjoyed that! John: #No, you didn’t! / #Yes, you did!


(Mary is getting off the roller coaster; John is waiting for her) Mary: I enjoyed that! John: #No, you didn’t! / #Yes, you did!

From what we have observed about fun, it is clear that the faulty patterns in these cases cannot be tied to the presence or absence of an overt experiencer argument. They are more amenable to the broad pragmatic approach we take here. In this section, we have observed that there is nothing privileged about the syntactic introduction of experiencers in for-phrases. Alternatives include tophrases (tasty to me), grammatical subjects (I enjoyed the opera), or objects (The movie bored her), in addition to no expression at all (That was fun!). The message we take away from this is that it does not matter what expression is employed or who the experiencer might be, but rather, that an experiencer is an individual who is in the best possible position to assess the truth or falsity of the proposition.

194 | Gunlogson and Carlson

4.3 Situational dependence Even if the list of predicates triggering judge-relative interpretation is expanded, operating on a purely semantic basis runs into a more fundamental problem alluded to earlier, namely that whether disagreement is “faultless” or not, often depends not just on the predicate but the situation under which the utterance takes place. Consider (53), which does not feature predicates of taste, but does involve differential access to the best available evidence: (53)

(Mary and John are on different continents, having a telephone conversation) Mary: It was cold here today. John: #No, it wasn’t. / #Yes, it was.

Note that John may have evidence, such as a recent weather report, that would be perfectly adequate to support his claim in other circumstances. But since Mary is there, and can go outside and see for herself, she clearly has access to better evidence than John, so John is understood as being presumptuous. However, if we change the circumstances so that Mary is known by both to have no access to first-hand information, not even weather reports, for whatever reason, then John’s response suddenly becomes quite acceptable. A further advantage of the approach we take here can be seen in situations where a participant in an event does not, for one reason or another, even have access to their own experience: (54)

(Mary was drunk and passed out at the start of the ride, remembering nothing; John was sober) John: The roller coaster was fun. Mary: #No, it wasn’t. / #Yes, it was.

Mary’s claim in (54) cannot, under the circumstances, be based on her experience of the roller coaster, and thus acquires no authority. Direct experience counts as such only if the experiencer’s testimony is based on privileged access to the experience.

4.4 Remarks on fun and tasty Observe that in the usual case, what is fun (or not) about a roller coaster is not the physical structure per se but the riding event that takes place on it. Thus the rough equivalence shown in (55):

Predicates of experience |



The roller coaster was fun. a. b.

≈ The ride (on the roller coaster) was fun. ≈ Riding the roller coaster was fun.

Other predicates exhibiting similar equivalences are illustrated in (56)–(58): (56)

a. b.

That book is boring. ≈Reading that book is boring.


a. b.

The wait was nerve-wracking. ≈Waiting was nerve-wracking.


a. b.

That chair is comfortable. ≈Sitting in the chair is comfortable.

This equivalence does not hold for tasty predications: (59)

The chili was tasty. a. b.

≉ ??The consumption of the chili was tasty. ≉ ??Eating the chili was tasty.

And there are other perception predicates that follow the model of tasty rather than fun: (60)

a. b.

The garbage stinks. ≉ ??Smelling the garbage stinks.


a. b.

The view of the ocean was beautiful. ≉ ??Viewing the ocean was beautiful.

Roughly speaking, tasty seems to be conceptualized as a property of the substance consumed, not the experiential event of consuming it. By contrast, the apparent attribution of fun to the non-sentient roller coaster structure in The roller coaster was fun is illusory, as the equivalences in (55) reveal. We suggest that fun is predicated of a characteristic experience (an event) pragmatically associated with the roller coaster, namely the riding of it. Likewise, in (56)–(58), the predicates are understood in connection with activities of reading, waiting, and sitting. The pragmatic nature of the association is important, as the nature of the activity can vary with context. The nature of the associated experience may vary under more specialized circumstances. For example, The roller coaster was fun could be uttered by an amusement park designer who is talking about which rides she most enjoyed designing; in such a case the equivalences in (55) would not hold but variants invoking the activity of designing would.

196 | Gunlogson and Carlson

Clearly more needs to be said to flesh out these differences between the fun and tasty classes, and we do not attempt that task in this paper. But even a cursory examination is enough to cast some doubt on the legitimacy of a unified class of “taste” predicates receiving special semantic treatment. We believe that what fun and tasty predications have in common, and share with a much larger class of predicates, is that they are, or at least have the capacity to be, judgments of experience.

5 Conclusion The proposal presented in this paper is that pragmatic considerations involving the quality of evidence underlie a broad range of phenomena that have in other work been attributed to truth-conditional differences. We have concentrated on pragmatic aspects of fun predications and have tried to show the central role of in agreements as well as disagreements about taste. In addition to the familiar categories of faultless disagreement, as in (3), and ordinary disagreement, shown in (4), we have argued for the importance of “faulty” disagreement and agreement, where a first person experiencer is explicitly present (for me), as in (5), repeated below: (3)

(John and Mary just got off the roller coaster) Mary: That was fun! John: No, it wasn’t!


(About Bill’s ride on the roller coaster) Mary: That was fun for Bill. John: No, it wasn’t.


(John and Mary just got off the roller coaster) Mary: That was fun for me. John: #No, it wasn’t.

The anomaly in (5) does not fall out directly from taking a subjective view of truth conditions; Mary’s claim in (5) is classified as “objectively” true or false in a purely semantic account because of the presence of the for-phrase. Without consideration of the pragmatic contribution of evidential factors, there is no ready explanation for the contrast between (5) and the case of ordinary disagreement in (4). We take the position that evidential considerations may also suffice for characterizing the faultless disagreement phenomenon (at least for experiential predicates). The strong but rather vague intuitions appealed to in assessing

Predicates of experience | 197

disagreement are, we submit, as easily construed as concerning the unassailability of one’s position as they are concerning the truth and falsity of the assertions made.¹⁰ However, we do not press the latter point in the present paper, since it is clear that there is something about the underlying semantics of the sentences and predicates involved that determines just what kinds of evidence would be relevant in the first place and we have no specific proposals to make about that issue in this paper. Our central claim is rather that the evidential approach in large part solves the problems that, for instance, the introduction of a judge parameter is designed to address, reducing (but not eliminating) the motivation for adding that dimension to the underlying semantic machinery. In addition, we suggested in Section 4 that the patterns in question extend well beyond predicates of taste, so that a pragmatic explanation in terms of evidentiality has very broad application, while a semantic mechanism confined to a certain class of lexical items appears increasingly limited in terms of data coverage. Evidentiality provides key insights into a wide range of conversational phenomena that has little directly to do with predicates of personal taste. However, in concentrating on pragmatic aspects, we have yet to directly address one of the central semantic issues concerning predicates like fun: what is the semantic value of a sentence like (3), in which no overt experiencer is specified, and how does it compare to the semantic values of sentences that do explicitly express an experiencer, whether that experiencer is one of the interlocutors, as in (5), or some third party, as in (4)? Considering just (3), we might be tempted to pursue an objective treatment of fun predications. That is, suppose that in the given circumstances, either Mary is correct or John is—there is a fact of the matter that makes reference to some property of the roller coaster ride that either it has or does not have. The pragmatic account we have outlined can be used to understand why determining the fact of the matter is fraught, a different matter. The testimony of the rollercoaster rider will normally be immune to challenge, but that immunity, we have argued, stems from the primacy of direct experience rather than requiring the assumption that the testimony is true in every case where it is justified. Sentences like those in (4) and (5), however, reveal the limitations of a purely objective approach. The availability of an overt experiencer does seem to require that the experience of fun be relativizable to a particular individual, the nature of whose experience (fun or not) determines the validity of the claim. Even if an objective semantics were to be possible for (3), in other words, (4) and (5) show that

10 We thus take a sympathetic stance towards contextualist explanations of faultless disagreement (see this volume’s introduction (van Wijnbergen-Huitink, 2016) for an overview of the contextualism/relativism debate and Umbach (2016) [this volume], who reaches a similar conclusion.

198 | Gunlogson and Carlson

we need some operative notion of “fun for X” to figure not just into the pragmatics but also into the semantic composition. This observation brings us back to a familiar dilemma: is (3) to be analyzed on analogy with (4) and (5) as having an experiencer argument that happens to be covert? Or is there some other way to characterize the explicit relativity of (4) and (5) that does not require treating (3) similarly? Lasersohn specifically designs his approach to meet the latter criterion, arguing (most convincingly, we find) against the covert argument approach. We will not revisit those arguments here. We are not at present inclined to provide a comprehensive argument against Lasersohn’s judge-semantics proposal, which we do not see as inherently incompatible with our pragmatic story. So if we retain a Lasersohn-style judge parameter, at the very least one would expect the pragmatically-driven selection of the appropriate judge for a context to coincide with evidential considerations; it follows from our view that given a choice, a participant with direct experience will be a stronger candidate for judge than one without. In addition, as we have noted throughout, the evidential account is quite general in nature and covers a range of phenomena that the judge account does not directly extend to. A goal for future work then might be to explore the possibility of a semantic account that has recourse neither to covert arguments nor to judges (or anything in the semantics playing a similar role) but that has the potential to mesh productively (i.e., non-redundantly) with the pragmatic requirements we have argued for so far.

References von Fintel, Kai & Anthony Gillies. 2010. ‘Must’ ...stay ...strong! Natural Language Semantics 18(4). 351–383. Grice, Herbert Paul. 1975. Logic and conversation. In Peter Cole & Jerry L. Morgan (eds.), Syntax and semantics, volume 3: Speech acts, 225–242. New York: Academic Press. Gunlogson, Christine. 2008. A question of commitment. Belgian Journal of Linguistics 22(1). 101–136. Hegarty, Michael. 2016. Subjective meaning and modality. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 227–228. Berlin: De Gruyter. [This volume]. Heritage, John. 1984. A change-of-state token and aspects of its sequential placement. In J. Maxwell Atkinson & John Heritage (eds.), Structures of social action: Studies in conversation analysis, 299–345. Cambridge: Cambridge University Press. Kölbel, Max. 2004. Faultless disagreement. Proceedings of the Aristotelian Society 104. 53–73. Lasersohn, Peter. 2005. Context dependence, disagreement, and predicates of personal taste. Linguistics and Philosophy 28(6). 643–86.

Predicates of experience | 199

Murray, Sarah E. 2010. Evidentiality and the structure of speech acts. New Brunswick, NJ: Rutgers University dissertation. Pearson, Hazel. 2013. A judge-free semantics for predicates of personal taste. Journal of Semantics 30(1). 103–154. Roeper, Tom. 2016. Acquisition arguments for a general point of view. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 201–226. Berlin: De Gruyter. [This volume]. Speas, Peggy. 2008. On the syntax and semantics of evidentials. Language and Linguistics Compass 2(5). 940–965. Stephenson, Tamina. 2007. Judge dependence, epistemic modals, and predicates of personal taste. Linguistics and Philosophy 30(4). 487–525. Tenny, Carol L. 2006. Evidentiality, experiencers, and the syntax of sentience in Japanese. Journal of East Asian Linguistics 15(3). 245–288. Umbach, Carla. 2016. Evaluative propositions and subjective judgments. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 127–168. Berlin: De Gruyter. [This volume]. van Wijnbergen-Huitink, Janneke. 2016. Subjective meaning: An introduction. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 1–20. Berlin: De Gruyter. [This volume]. Willett, Thomas. 1988. A cross-linguistic survey of the grammaticization of evidentiality. Studies in Language 12(1). 51–97.

Tom Roeper

Propositions and implicit arguments carry a default general point of view Acquisition evidence against relativism and subjectivity Abstract: This article claims that there is a “default general point of view” (GPOV) linked to every propositional statement, whether or not “subjectivity” appears to be relevant. Thus religious statements like Allah is great are intended to be a true statement that holds for all people whether or not you are a believer, and it is not subject to “a judge” or evidential validation. Relativism may be philosophically correct, but it is not built into the grammar. The presumption of truth allows propositions to be usefully transportable: we can transfer facts to others (the stove is on fire) under this default truth presupposition. Under this thesis, we claim that children treat all such statements, including statements with so-called “predicates of personal taste”, as propositions. Thus God is great, Johnny is big, This tastes yucky are all assertions that do not have an implicit argument as in God is great for me. This is evident in the intent behind children’s sentences like That’s easy. Naturalistic data on sure and experimental data on scenarios where children, who see a giraffe eat the top of trees, answer “no” to the question Is it easy to eat the top of trees? supports this idea. Instances of “faultless disagreement” are arguments about this intended truth and characterize child conversations (It’s yucky, No, it isn’t yucky). Adjectives that subcategorize control and factivity must be learned and therefore plausibly require implicit arguments that carry GPOV. Children initially use easy with an implied complement: easy to do, but it carries an implicit argument for everyone. They eventually project both the complement and the implicit argument when they learn the distinction between, for instance, John was glad to swim and *John was upset to swim.

1 Introduction: Intellectual history When the radio was invented, there were skeptics who said it would be of limited use because “no one has anything to say to everyone”, that is, we have no general assertions to make to a non-specified audience. This is obviously untrue, given the success of radio. Even animals communicate to an unspecified audience: a growl, Tom Roeper, Department of Linguistics, University of Massachusetts, Amherst, MA 01003, USA, email:[email protected]

202 | Tom Roeper

or a howl, or a hiss is intended for both visible and unseen predators (to every living thing) in the environment. If so starkly untrue, why did it seem plausible that radios are useless? The broad answer may be because our cultural psychological models are dominated by notions of individual exchange and self-interest. In Western culture and particularly economics, self-interest is presumed for every action, just as ego-centrism is mistakenly assumed for children. By extension, discourse models begin with the existence of a fundamental speaker-hearer relation where each person has a perspective and goals with respect to the other. This model does not easily represent, for instance, what a teacher assumes in speaking (in a distinctive classroom manner) to a whole classroom of children. Assertions, we argue to the contrary, presume a generalized speaker (“general point of view”) and a general, undifferentiated audience, much like a radio speaker. We claim here that children demonstrably assume this non-egocentric general point of view immediately, in their earliest declarative utterances, never possessing the pure ego-centrism so readily assumed. The personal perspective emphasis is further enlarged by another intellectual dogma, born of relativistic physics: every action or claim is affected by the actor and no independent reality exists outside of individual perspectives. Therefore, in addition, one can seek only an ideal description, truth is an illusion and “subjectivity” is inevitable.¹

1.1 Theoretical background Western philosophy, accordingly, has produced both “externalist” and “internalist” views of the denotational power of grammar. The externalist view claims that statements have a determinable truth-value, while the internalist view suggests that a full description must be done entirely in mental terms, which seems to invite the view that propositions are essentially “subjective”. Formal semantics representations that begin from an individual perspective lead quite naturally to the observation that all statements are in some measure altered or affected by speaker limitations and hence must be subjective. Nonetheless, the inevitable failure of individual speakers to achieve “objectivity” does not mean that the grammar itself does not, in a sense, “presume” objectivity. It is undeniable that we have asser-

1 This paper is a revision of a paper originally co-authored with Chloe Gu entitled: “Acquisition arguments for a general point of view as an alternative to relativism and subjectivity”. I am indebted to her for the searches and all of the experimental work, and development of many ideas. This version adds perspectives on implicit benefactive arguments and control.

A default general point of view

| 203

tions and factives in grammar. That is, grammar must refer to notions of “truth” whether or not that truth has an external grounding.² And this bias of grammar captures our “intentions”—we intend our assertions to reflect a general truth beyond ourselves. Therefore we argue that the much simpler default assumption (1) captures both intuitional and acquisition facts, which are the default assumption before subjective accommodations are represented in the semantics and pragmatics. There is for every assertion: (1)

A general (human) point of view (GPOV)

Why do we add “human” and what is its significance (see Landau, 2014 for discussion)? We argue that expletive sentences like the following, which show no morphology for “point of view” (POV), get a human POV despite its animal reference: (2)

Is it good to wear a dog collar?

(2) is automatically evaluated from a general human perspective, not a dog’s perspective, so the natural answer is “no”. If such a restriction is present, it suggests a biological bias that is not captureable by any application of a “pure resource logic” defineable independent of human nature. We can consider it a default presupposition. It is noteworthy that this assumption can be contextually accommodated to animals or again to something more abstract:³ (3)

a. b.

It must be delicious for the cat to eat. It must be tiring to run in circles like a hamster does.

2 See Hinzen (2007) for background and relevant discussion. 3 See Roeper & Roeper (2010) on further default assumptions about “cooperation” to explain the contrast: (i)

a. b. c. d. e.

We agreed to disagree *We disagreed to agree. We disagreed about agreeing. John and Bill avoided meeting. *John and Bill avoid to meet.

(ib) is excluded because control requires both an agent and a broad notion of cooperation, as occasionally noted elsewhere in the literature. (ic,d) can be explained under the view that an arbitrary PRO (PROarb) is present in the lower clause, with perhaps some sensitivity to the absence of future tense possibility as well. Interestingly, sentences like (ie) are frequent L2 errors, indicating the subtlety of language variation here.

204 | Tom Roeper

In (3b) the implicit controller is really neither animal nor human but a “human as if animal” entity. Properties of a GPOV are pursued in many systems (as in the current work on “predicates of personal taste” (PPT) in this volume). An “exocentric” (Lasersohn, 2005; Stephenson, 2007) property can be formally captured as an extension of personal perspectives, or by indexicals or small pro representations linked to a speaker that becomes generic (Moltmann, 2010).⁴ These perspectives lead to the necessity of a “judge parameter” to decrease the subjectivity. Similarly, suggestions at the DGfS workshop on a “normative” basis for statements move in the direction we advocate, but maintain a relativist ingredient we think may be unnecessary.⁵ In general, we argue that the best representation should not be a derivative notion within the notation for assertions, for instance from an implicit for someone to a generic (Stephenson, 2007).⁶ In addition, we argue a GPOV can be separately projected onto “implicit benefactive arguments” which are needed for control structures and PPTs, and for which syntactic representation explicitly or implicitly is required. Thus two dimensions are involved: (4)

a. b.

Assertion: GPOV is implied PPTs: implicit GPOV, benefactive [=for everyone]

Two important acquisition questions arise: 1. Where does the child start? 2. How does the child discern where PPTs exist? The problem is difficult because there is lexical and language variation, as we will illustrate. Therefore we argue that initially all sentences entail the implicit GPOV of an assertion. The child needs further evidence to make finer syntactic distinctions.

4 These are furthermore all pertinent to the evidential dimension of modality (see Kratzer, 1991). 5 It should be noted that this discussion is partly, but not fully informed of all the extensive relevant philosophical and formal semantic literature on presuppositions and PPTs. The views are primarily derived from our work in acquisition. 6 GPOV also corresponds to Speas’ (2004) notion of the “seat of knowledge”. Note that it is different from Hollebrandse (2000) where a syntactic POV Operator on a root CP position controls the co-variation of pronouns, deictics, and tense. It captures the simultaneous shifts from: Can you send me this here now to Yesterday he asked could I send him that there then with embedded semi-quotation. (See also Anand & Nevins’ 2004 arguments that things “shift together”.) These explicit syntactic POV forms are distinct from the GPOV and the implicit argument benefactives to be discussed.

A default general point of view



1.2 Assertions and PPTs The nature of PPTs is obscure in many situations. Consider the following contrast: (5)

a. b.

The meat is salty. Ocean water is salty.

(5a) is possibly stage-level and possibly subject to differences of opinion (faultless disagreement), but (5b) involves an inherent property (reflected in the compound saltwater). It would seem redundant, but possible to add: (6)

#Ocean water is salty to everyone.

Likewise expressions like: (7)

Waxed floors are shiny.

would seem awkward if it were enlarged to: (8)

#Waxed floors are shiny to everyone.

unless its universality had somehow come into question. We regard shiny as a property of the waxed floor, not a reflection on the observer’s opinion, although it entails a perceiver somewhere in its semantics. This issue relates to the ancient question originally posed by George Berkeley about the existence of unperceived objects or perceptible events: “If a tree falls in a forest and no one is around to hear it, does it make a sound?” Under our view, the GPOV is present and therefore the answer is “yes”. The very fact that it is a thorny philosophical question suggests that it should be seen as part of the entailments of sentences rather than their representation. Our approach to GPOV also suggests that philosophical interest in “faultless disagreement” is missing an important perspective by seeming to allow only the view that individuals may have separate views of the truth which applies to each of them and not others. Classic examples include: John thinks Bill is more beautiful than Fred, and Mike thinks the opposite. But again there is a GPOV lurking here. John can accept that for Mike Fred is more beautiful, without accepting the idea that Fred is more beautiful in general, because, in general, he thinks Bill is more beautiful, not only for himself, but for others. Hence the disagreement about the GPOV continues to be present, if unresolveable. The role of “fault” seems to me to disappear if there is no method of determining the truth. Thus it seems to me that the notion of faultless disagreement does not avoid the GPOV factor, nor represent another significant way to characterize such situations.

206 | Tom Roeper

It is perhaps valuable to enlarge this claim. We think the term faultless disagreement is no longer helpful because it assumes a strict speaker/hearer contrast which many interchanges do not reflect. Consider this discussion: (9)

A: B: A: B: A:

I think he is probably not athletic. Well, he plays pool well. Is that really athletic? I guess we need a criterion for athletic ability. Yes—perhaps athletic entails some full body co-ordination.

This is a discussion where a variety of relevant, but perhaps incompatible assertions are made, aimed at achieving a common understanding, by contributing information to the Common Ground for common evaluation. The concept of faultless disagreement fails to acknowledge that many assertions can be made which may be mildly incompatible but which are intended as basically true by all sides, not a reflection of individual perspectives, hence not a form of disagreement. It seems very possible that a high proportion of discussions are of this nature and we misconstrue the semantics behind them if we do not acknowledge it.

1.3 GPOV, relativism and contextualism Our position is that, aside from the philosophical conundrum, our “language” assumes a GPOV if an adjective or other expression with a perception property is used. Moreover, once again, we consider this assumption to be built into a UG interface with semantics and therefore requires no learning. Propositions may be the distinctive characteristic of human language and their reflection of a GPOV is central to their utility. There are many utterances that reflect the GPOV quite easily: (10)

a. b.

On the whole, roller coasters are fun, the owners believe. In general, hamburgers must be tasty, or sales would not be so high.

To situate our view in contrast to relativism, contextualism, and subjectivity— very nicely outlined by van Wijnbergen-Huitink (2016) [this volume] in the introduction to this volume—our view is yet different. It reverts to the seemingly untenable ancient Platonic view that there is an abstract truth which assertions seek to enunciate, “pure forms” for things and concepts. Language design allows our assertions to entail that presumptuous presupposition, hence they are easily the objects of vehement conversations. Van Wijnbergen-Huitink (2016) shows quite nicely that virtually all the efforts to employ hidden adjuncts and arguments

A default general point of view

| 207

lead to unacceptable sentences. Therefore the Platonic view that “real” truth is expressed, unmodified by a judge or a specific experiencer, seems supported by those arguments. When our meaning engages the semantics of possible worlds, then the pragmatics of situations becomes relevant. Immediately, all the other situational interpretations are linked to representations with implicit arguments. Then we are led to many finer discriminations among situations, relevant inferences, and possible control structures, to which we turn shortly. The consequence is that disagreements—unless implicit modifications are demonstrably present—are exactly disagreements about putative truths and nothing more complex. If I say “It tastes awful” and you say “It doesn’t”, the conversation suggests that we share an abstract notion of “awfulness”, even if a moment’s reflection would suggest we do not. Like No-fault car insurance, where fault often exists but it is too obscure to be worth contention, disagreements about assertions require the civilized view that the truth is hard to determine, which is distinct from there being no truth. Hence conversations often have a vehement quality, and are never really resolved. It is important to note that, where truth is not subject to disagreement, assertions have the virtue that they allow “transportability” without co-experience. If we say: (11)

The stove is on fire.

we can run round the house and let everyone know because no computation about the speaker’s truthfulness or individual perspective is necessary to warrant action (except, of course, for the boy who cried “wolf” who exactly misused the GPOV/truth aspect of propositions).

2 The role of implicit arguments in PPTs and their acquisition Evaluative predicates, like PPTs, are subject to lexical and language variation, as we explore below, and therefore they create an acquisition problem. Where does the child start? We argue that the natural starting place is to assume that all predicates are truth-bearing propositions about properties. Thus Roller coasters are fun is both a general assertion and possibly an assertion for which situational accommodations occur. Roeper (1987) argued that there are implicit agent argument controllers for many infinitives and Pearson (2013) argues that there is as well an implicit bene-

208 | Tom Roeper

factor argument represented in the semantics and implicit syntax. Primary evidence for implicit benefactors comes from the possibility of control predicates which require explicit or implicit controllers: (12)

Roller coasters are fun [to ride]. (= for someone/everyone to ride)

Note, by contrast, the oddness of the assertion of a property with control: (13)

?*Waxed floors are shiny [to see].

This is precisely because the complement in (13) is not a possible subcategorization of the adjective, since it is a property. But would a child know that? If shiny could be either a PPT or a property, then it is not inherent in the adjective.

2.1 PPTs and subjectivity Lasersohn (2005) has argued that PPTs are a central paradigm for communication and are oriented toward verbs linked to personal experience for such cases. Other predicates that seem to come from domains of personal commitment, like religion, nevertheless have a different illocutionary force, but the same copular or transitive syntax. Thus signs proclaim: (14)

God is good / Allah is great.


Jesus died for your sins.

Religious remarks, dominated by evaluative terms, are obviously intended to be “general truths”, not for the speaker alone, and worth a war to defend. The religious fanatic holds that such statements are true for you, even if you do not agree, essentially enunciating a Platonic commitment. While it is obvious to those in Western cultures that such assertions are—at a scientific level—quite cultural and relative, it is equally obvious that any presumption of relativity is an anathema to the speaker, and notably unmarked in the language: God is great ≠ God is great for me. We believe that children who have vociferous arguments over topics like (16) are presuming a notion of truth that is closer to “God is great” than a situationally modified “Roller coasters are fun”: (16)

a. b.

Tomatoes are yucky. That’s easy.

A default general point of view

| 209

And even “Roller coasters are fun”, unmodified by “for me”, has the force of a claim about a universal GPOV, contrary to the literature. It is noteworthy that children acquire forms like (16b) before ones with an explicit complement although there is always an implied complement easy [to do]. This means that before the syntax of infinitival control is available, the meaning is still accessible in some form that it would be valuable to characterize appropriately. The strongest hypothesis, perhaps the “strongest meaning” in the sense of Winter (2001), is that all PPTs are ambiguous between propositions with a GPOV (strongest meaning) and predicates that carry an implicit modifying argument. Whether the modification is present depends upon the pragmatic act of accommodation that every sentence requires. They all have one assertive reading and another reading where an implicit argument is projected that allows a finer differentiation into: for me (speaker), for someone, or for everyone. Pearson (2013) argues for a situation-centered possible world semantics which can allow this differentiation with implicit arguments of this kind, sensitive to context. We consider situation-centered possible worlds as a species of accommodation that is related to situational variables (locative) for sentences like: (17)

Does everyone have a spoon? ⇒ accommodation: at this meal

If we look at other adjectives we find as well that we need to accommodate to a situation-centered perspective that Pearson advocates, but again acknowledging a GPOV ambiguity. Inaudible provides a case study: (18)

The discussion is inaudible.

This is on one reading—the default reading—a claim about an insufficient level of sound that holds for everyone at any time or place. However if five people in the backrow of an auditorium yelled, “it is inaudible”, then one might conclude that there should be an accommodation of location to the backrow, to which someone might respond: “Sit on the floor in front of the front row and then you can hear.” So in these environments, we need to project the GPOV into the implicit argument domain (inaudible for everyone in the backrow), which in turn is modified by the pragmatics of accommodation. Note that an added infinitive would be ungrammatical (*this is inaudible to hear) because the complement is built into the meaning of audible, ruling out an explicit agent as well: *inaudible by someone while as usual allowing the more

210 | Tom Roeper

broadly available inaudible to someone.⁷ -ible unlike -able rules out any complement: *This is inaudible to inform the audience. Similarly, if a senior citizen hears a five-year-old say “Roller coasters are fun” (meaning fun for everybody), he will not leap on board, but construe it as “fun [for a five-year-old]” or just “fun [for him]”, despite the child’s intentions. We can see Rollercoasters are fun as ambiguous between a GPOV assertion and one open to accommodation in situations.

2.2 Further dimensions of control Implicit agent arguments are needed for many control predicates (Roeper, 1987, originally noted by R. Manzini, p.c.): (19)

a. b. c.

Nominalizations: the bombing of cities to destroy morale Compounds: meat-eating to gain weight -able: goods are exportable to improve profits

The agent can in fact be added: meat-eating by young girls to gain weight. Notably, implicit agents cannot be controlled as observed by Chomsky (1986): (20)

The wine was bought to be drunk.

(20) does not entail that the buyer and drinker agents must be identical, illustrating very sharply a contrast with more obscure cases where some relation between controller and PRO is clearly present (21): (21)

The leader was assassinated to preserve American power.

The individual assassin is acting for the CIA, which is seeking to preserve American power. (21) shows that a control link must be established. Such elements allow control to include part-whole and “partial control” environments, whose exact semantics are an interesting challenge, to which the whole topic of implicit arguments is relevant. Landau (2000) argues that these involve an PROarb, but if we enlarge the set of arguments to include benefactives (once called “super-equi”), which fits Pearson’s (2013) account, they can be seen as particular implicit arguments without the sharp distinction between implicit arguments that (20) illustrates:

7 There are more subtleties here. Inaudible to someone seems superior to #the floor is shiny to someone which we regard as possible as a coercion of a property into a PPT.

A default general point of view

| 211


John yelled to open the door. ⇒ yell (to someone) (for someone) to open the door


But the uncertainty at the beginning of the 2013 season faded and the lights went on to reach yet another national tournament.⁸ ⇒ lights went on (for relevant people) (PRO-for relevant people) to reach ...⁹

It is clear that we must differentiate predicates that subcategorize for control with distinct implicit arguments and those that do not, showing that this kind of control is part of the representation of passive and particular adjectives (24): (24)

a. John was eager to win / smart to win. b. *John was brilliant to win. c. *Bill was awkward to win. d. Bill was dressed to win. e. *John was peculiar to win.

It is far from obvious why smart allows control, but brilliant does not, unless we have a case where the controller is explicitly an argument of brilliant: (25)

It was brilliant of John to win.

In fact we can vary across: for me (the speaker), for someone, for anyone, for everyone depending on the situation. In (26) PRO is controlled by an implicit existential quantifier. (26)

John yelled to open the door. ⇒ for someone

But in (27) the implicit quantifier and controller takes the form of a negative polarity item. (27)

It is not easy to climb the tree. ⇒ for anyone

As we shall see in greater depth, the PRO in (27) should assume a GPOV perspective as default. This is a more semantically precise version—though obviously not

8 Thanks to I. Landau for alerting us to such examples. 9 Landau (2000) takes a narrower view of possible implicit arguments than we do here, but if the semantics literature and contributions to this volume are on the right track, then the postulation of implicit arguments for these cases becomes increasingly plausible. Landau (2014), though I have not read it carefully, moves in this direction as well.

212 | Tom Roeper

integrated into formal semantic theories—of what PROarb has been used for in syntax. Even this concept does not cover the subtlety of all felicity problems, for instance, the difference between explicit and implicit complements. We can say “Waxed floors are shiny” but it is odd to say “Waxed floors are shiny to see” (see also (13) above).

2.3 The acquisition dimension Propositions may not be representative of a child’s first utterances. Potts & Roeper (2006) claim that “expressive” small clauses are first. They are then expanded to propositional tensed sentences: (28)

he big ⇒ he is big

Adults retain comparable expressives as incredulity indicators: You idiot! which, likewise, notably lacks a copula. Once tense emerges, Jill deVilliers (p.c.) reports that a child almost immediately allows subject-verb inversion to indicate questions that entail true/false alternatives when the hearer is asked to verify or deny the proposition: (29)

Is he big?

Does a child have the implicit argument reading immediately after propositions emerge? It is not clear. In a pilot experiment where there should be bound PROarb readings, K. Carlson found something different (see Roeper, 2007): (30)

Scenario: A boy eats a bowl of cherries. Question: To eat cherries is to become empty, is that right? Answer six-year-old: No, you become full. Answer four-year-old: Yes, you eat the cherries and the bowl becomes empty.

This experimental hint—which we pursue in the experiment—suggests that the acquisition path begins with: (31)

GPOV propositions, then PROarb for control, then control is restricted by syntactically real implicit arguments.

That is, given many subtle ambiguities, it is natural that the first and starting assumption for a child is that all such predicates are simply GPOV propositions, which are exactly the same:

A default general point of view


a. b. c.



Johnny is big. Allah is great. Vegetables are yucky.

This is followed by implicit complements (That’s easy = That’s easy to do), followed by PROarb, followed by real implicit arguments.

2.4 Acquisition background Implicit arguments and implicatures have been studied in acquisition and in both instances there is evidence that children show limited ability to reliably grasp them before the age of five, particularly more sophisticated implicatures (see de Villiers & Roeper, 2011, for an overview). Thus implicit agents are not initially recognized as what separates (33a) from (33b) (see Verrips, 1996): (33)

a. b.

The apple dropped on the ground. The apple was dropped on the ground.

If for me is a comparable implicit argument, then we would predict that children would avoid sentences like That is yucky until they were able to project both invisible implicit agents and invisible benefactives, but as we shall see, truth assertions over taste predicates (PPTs) occur much earlier than implicit arguments. Neither the implicit benefactive for me nor for anyone is expressed in putative PPTs, because they are not necessarily there, unless it is demanded for control. Let us look more closely at what “transportable propositions” involve: (34)

Assertions are transferable externalized propositions about general truth.

Thus the value of these largely personal observations is that they hold under transfer as externalized assertions, as in the dinner arrangements in (35). (35)

a. b. c. d. e.

I will be 20 minutes late for dinner. Bill is hungry. Fred hates buffets. Mary’s children have to leave early. We have to consider all this.

All express individual perspectives, or POVs, in a sense, but they are expressed as available truths, hence true from a GPOV. If these views are told to Mike who tells them to John, then John can weigh them collectively and choose the right restaurant to meet at. It is clear that a “presumption of truth” is necessary to their

214 | Tom Roeper

value as “transportable propositions” where we can adjust all indexical references (I = John, we = an unclear subset) without disturbing the underlying proposition such that they are useable by a third person in making a decision. Many discussions overlook this dimension and thus may mischaracterize socalled speaker-adverbs: (36)

Unfortunately, John won the race.

While it is not unfortunate for John, it might be for the speaker, yet the sentence actually says that it is unfortunate in some absolute sense beyond both speaker and John. One might in fact challenge it by saying “It is just unfortunate for you” and get a reply “No, it is unfortunate because he always wins and others are discouraged from competing”, indicating the GPOV force of the sentence. A comparison to animals again may be instructive. Animals also often communicate to an invisible and unknown audience when they give warning cries or growls or other indications of attitude which are intended (by the biology, not by some conscious intentional system) to communicate something to other organisms that they may not see. Interestingly, in the animal system, a howl does not externalize a proposition that allows it to be transferable among members of the species.¹⁰ Acquisition evidence pertinent to this point begins with the experimentation of Carol Chomsky 1969, who presupposed without comment that children understood a GPOV for expletives when she asked 5-8 year olds “Is it easy or hard to see the doll?”, where the GPOV is entailed for adults although the child might assume it is a hearer-POV alone.

3 Children’s knowledge of POV in longitudinal data We will focus on how children grasp contrasts like the following: (37)

a. b.

John is sure to win the race. [derived from: it is sure for John to win the race] John is sure that he will win the race.

10 That is our impression, not being educated on modern ethology. Perhaps a sequence of howls among wolves has this transportable function.

A default general point of view

| 215

In (37a) it is neither the child (or speaker) nor the subject John to which the “sureness” refers, but the GPOV. An indication that it is a GPOV and not an adjunct in the grammar lies in the fact that the nominal “sureness” can retain either reading: “the sureness of the conclusion” (= GPOV), or “his sureness of the conclusion” (= personal POV). Adverbs are limited to GPOV: He will surely win. While sure is a raising verb and the underlying form of (37a) It is sure that John will win the race makes a claim about probability in the world—not just the speaker’s view, these adjectives are often seen as speaker-oriented because the word sure itself can also be an expressive (see below) that clearly carries speaker-certainty (as in Sure, I’ll help or subject-certainty as in the sure-that construction). This ambiguity could influence the child’s perception of the GPOV—in particular raising sure when there is a non-expletive subject like John. Such ambiguities make determining the acquisition path a challenge for the child. To get a glimpse of children’s usage of POV constructions, we ran a search on children’s naturalistic production of expletive constructions for raising adjectives, focusing on sure, using the CHILDES database (MacWhinney, 2000). Adam in the Brown corpus, Ross in the MacWhinney corpus and Shem in the Clark corpus were studied. In the search, we went through all utterances of Adam, Ross and Shem to look for constructions that could be categorized as a potential expletive or contained sure.¹¹ The results were then hand-checked for expletive or sure-constructions. The longitudinal data showed that expletive constructions emerged fairly early in these children (ranging from 2;5 to 4;1 years). Some were clear GPOV uses, as the interaction between the child and the investigator/caregiver revealed that the child was expressing opinions about a general fact, e.g. (38). These children also knew how to use for-PPs to express POVs for a specific person. Shem produced six expletives with for me, e.g. (39), and Ross produced four such utterances with for you or for me, e.g. (40).¹² (38)

*URS: No # he couldn’t take it apart. *CHI: Why? *CHI: It’s hard to take apart? *URS: Yes # it’s quite hard.

(Adam 3;4.01)

11 A potential expletive construction would contain a BE verb followed by an adjective, and the search command looked for this BE+adjective combination in the morphological tier of the transcriptions. 12 Copied from CHILDES: *URS = Ursula and *Chi = child; # = pause or non-grammatical sound or unintelligible piece.

216 | Tom Roeper


*SHE: Is [/] i(t)’s hard for me?


*CHI: milk is is good for you # but gum is bad for you.

(Shem 2;7.10) (Ross 3;7.03)

Not many children used the more ambiguous sure-construction during their recording sessions before age 5;5. Early sure uses usually link to the grammatical subject’s certainty (41), as the word sure appeared at the beginning of an utterance and did not form a grammatically correct sentence, but a fragment with an implied “are you . . . sure you . . . ”: (41)

*CHI:. . . sure you don’t want some?

(Adam 2;7.14)

Another early use of sure was in exclamatives (42), where the word sure appeared in second-position and expressed the speaker’s attitude towards the utterances: (42)

*CHI: oh # you sure believe. (Adam 3;2.21) *CHI: it sure is big. (Ross 3;8.18)

Others do not primarily express subject-certainty but rather emphatic real-worldcertainty, which we link to GPOV: (43)

*CHI: yeah (.) sure that could go like that.

(Ross 3;0.17)

We found a few sure-that and sure-to-constructions in Adam’s and Ross’s transcriptions. Adam produced two such utterances at 4;4 and 5;2, and Ross produced four such utterances between 3;8 and 5;5. The constructions were all grammatical, but it appeared that the grammatical subjects of the utterances were often the speaker (I) or the hearer (you). In other words, the utterances expressed a point of view from a local person. (44)

*CHI: I’m sure they’re there. *CHI: be sure to bring some water too. *CHI: are you sure it doesn’t write?

(Ross 3;8.18) (Adam 4;4.01) (Adam 5;2.12)

The corpus data suggested that children started to express the GPOV from early on (about 3.3 years), and they were able to state a POV from a specific person, even though those POVs were generally from a local person. There were few cases which involved a clear subject POV, but third person subject POV uses were very rare. This may be due to the nature of longitudinal data, which involved interaction between the child and the caregiver. The setting may prefer a local person’s POV in conversation. Nevertheless a case like Are you sure it does not write entails a question about another person’s emotional commitment to a proposition. This entails recognition of other minds (relevant to the Theory of Mind literature)

A default general point of view |


and the attribution of a proposition to that mind which may be false. The attribution of a “possible belief” seems much more sophisticated than the attribution of a false belief. Experimental methods are more versatile in exploring children’s knowledge of POV, and we developed two lines of experiments to study children’s understanding of general, speaker, and subject POVs.

4 Experiment design and methods Of the two types of experiments, the first evaluated whether children understood GPOV in expletive constructions, and the second focused on teasing apart subject/speaker POV/GPOV uses in sure-constructions. The test items were all presented in a story format. One example of the expletive stories can be found in (45). The story first stated that Johnny had problems jumping over the fence, because he was too little. It was also told to the child that other people who were taller did not have difficulty doing it. The contrast between one person (Johnny) and others would show that jumping over the fence was generally easy, though it might be difficult for some people under specific conditions. (45)

Look, there is a tall fence in the backyard. Johnny wants to jump over the fence, but it is so tall that he fails every time. But look, his sisters are tall and strong, they have no problems jumping over the fence. So: Is it easy to jump over the fence?

Stories of sure-constructions were different from stories of expletive constructions. One example of such stories can be found in (46). The story illustrated that one person had some false belief due to some accidents. Although the character in the story did not know the truth, the child as a listener knew exactly what had happened. In such a case, we had a contrast of different POVs, and would be able to test whether the child fully understood the sure question at the end of the story. (46)

This is Jacob. Jacob loves to eat peanut butter and jelly sandwiches. His favorite kind of jelly is strawberry. His mom always makes him a peanut butter and strawberry jelly sandwich for lunch, like this one here (bottom left). But last night Jacob’s mom ran out of strawberry jelly and so she had to make his sandwich with grape jelly instead. The next morning Jacob’s mom gives him his lunch bag and Jacob goes off to school, but Jacob’s mom forgot to tell him about the jelly!

218 | Tom Roeper

Four types of sure-questions can be asked after such a story. A sure-to-question as in (47a) addresses speaker’s knowledge of the general truth, but it also entails a child’s ability to carry out syntactic raising: It is sure for John to ⇒ John is sure to. This construction forces the speaker to use a GPOV to answer the question. (47b) is a sure-that construction that elicits the (grammatical) subject’s POV. The answer would be “yes” for (47b), because Jacob thought that his mother put strawberry jelly in his sandwich. (47)

a. b.

Is Jacob sure to have strawberry jelly in his peanut-butter sandwich? Answer: No. Is Jacob sure that strawberry jelly is in his peanut-butter sandwich? Answer: Yes.

The third and fourth types of sure-questions were used in part of the experiment to study children’s understanding of GPOV in sure-constructions (without syntactic raising). Both (48a) and (48b) relate to the general truth of the utterance, and the target answers are both “no”. (48)



Is it sure for Jacob to have strawberry jelly in his peanut-butter sandwich? Answer: No. Is it sure that Jacob will have strawberry jelly in his peanut-butter sandwich? Answer: No.

There were two rounds of experiments. The first round consisted of both kinds of stories. The participants in the first round were ten children aged from 3;8 to 5;2.¹³ For these ten children, each experiment session contained five expletive stories, three sure-stories and two control items. The participants were randomly assigned to two groups, and the only difference between the groups was the kind of surequestions that were asked after each sure-story. Each child would receive three sure-stories with two sure-that questions and one sure-to question, or with one sure-that question and two sure-to questions. Since there were 10 children who participated in the experiment, we would be able to gather an equal amount of sure-to answers and sure-that answers.

13 They were recruited from local daycare centers, and they were tested in their classrooms. At the time of testing, four children were younger than 4;0, five children were between 4;0 and 5;0, and one child was older than 5;0.

A default general point of view

| 219

For the second round, a total of 32 children aged from 3;7 to 7;9 participated in this round, involving older children as well.¹⁴ There were twelve such stories, and the stories were presented to children in the same order. Each story was accompanied with one of the four kinds of sure-questions. Altogether, each child would receive three sure-to questions like (47a), three sure-that questions like (47b), three GPOV sure-questions like (48a), and another three GPOV sure questions like (48b).

4.1 Results Children in our study did very well on expletive stories. Their answers, on average, were target-like for 80.4% of the time for all expletive stories. In four of the five stories, the target answers accounted for more than 80% of children’s replies. There was one story in which two children failed to offer relevant answers to the question, but only two out of 10 children gave non-target answers for that story. Overall, the error rate of children’s answers was only 19.4%. The participants not only produced target answers, but also did it consistently for most stories. Two children gave adult-like answers for all five questions. One of them was 4;4, and the other was 4;6. The other eight children missed only one question, and their age ranged from 3;8 to 5;2 (average age was 4;1).¹⁵ Some of the target answers were impressive, as a girl aged 4;6 added (49) after a target-like “yes”-answer for some stories. (49)

(but) not for Johnny

To make such a comment, the girl needed to fully understand the contrast between a GPOV and one specific person’s POV, and her own speaker-POV. Note that for the few non-target answers, children typically replied “No, I can do it” applying their own view, speaker-POV, to a GPOV construction, showing a subset of the kind of responses, which is often cited as evidence for egocentricity. One of the stories was designed to test children’s preference for a human bias in expletive constructions, and the result showed that children did apply the same bias as adults did. The story is illustrated in (50):

14 Fifteen of these children were between 3;2 and 4;11, eleven children were between 5;1 and 6;9, and six children were older than 7;0. Like in the first round, these children were recruited from local daycare centers / primary schools, and they were tested in their classrooms. 15 Not offering relevant answers counts as missing a question in this case.

220 | Tom Roeper


Look. There is a tall tree. Johnny’s sisters want to reach the top of the tree, but they are so little. But a giraffe is tall. So: Is it easy to reach the top of the tree?

The story is slightly anti-pragmatic, pushing the child toward a giraffe’s POV, but we got strong human GPOV responses. 80% of the children replied “no” (targetlike) and 20% of the children stated “yes” (non-target). The two “yes”-children were 3;8 and 3;9, while children who gave target answers to this question ranged from 3;8 to 5;2 (with an average age of 4;3). Children’s responses to sure-stories, on the other hand, were more diverse. In the first round of experiment, there were only three sure-stories, but we found that answers to sure-to questions (48a), which adopted a GPOV, were target-like 63.6% of the time for the ten children between 3;8 to 5;2. Answers to sure-that questions (48b), on the other hand, were target-like 21.4% of the time. These results were preliminary as only a small number of sure-questions were asked, but it revealed some differences in children’s responses to sure-to and sure-that constructions. In the second round of experiments, more sure-stories were presented, and we found that answers to sure-to questions (47a), on average, were target-like 69.8% of the time for all 32 children. Children between 3;7 and 4;4 gave 63.6% target answers to all sure-to questions, and children between 4;4 and 6;0 gave 61.9% target answers. However, children older than 6;0 reached 91.7% accuracy rate in giving target answers to sure-to questions. Children older than 6;0 performed a lot better than younger children in understanding the general POV in sure-to constructions. Answers to sure-that constructions (47b), which probed for the subject POV, were target-like 35% of the time for children under 6;0. 33.3% answers from children between 3;7 and 4;4 were target-like, and 36.7% answers from children between 4;4 and 6;0 were target-like. Three children answered all three sure-that questions correctly (aged 4;2, 4;4 and 5;5, respectively). Two children made one mistake (aged 4;0 and 4;2), and six children made two mistakes for all three surethat questions (aged 3;7, 3;11, 4;0, 4;3, 4;11, and 5;7, respectively). It looks like children may have made some progress from 3;7 to 6;0, but overall, they did not provide as many accurate answers as they did for sure-to constructions. An unexpected finding in the answers to sure-that constructions was that children older than 6;0 gave target-like answers only 13.3% of the time. It was not clear why older children did not perform well on this task, as they outperformed younger children in other questions we asked. It is noteworthy that this age is beyond the point where children make errors on “false belief”. Nonetheless they gave a “reality” answer which is consistent with GPOV. Since they were “worse” than younger children, we speculate that they may have imposed a syntactically more complex form of factivity sure of the fact

A default general point of view

| 221

that, as in the factive: aware that there was a sandwich in his bag which would then call for a “false presupposition” accommodation, which is distinctly harder (see Rau, 2013). For sure-stories that target for a GPOV, children on average gave target-like answers 74% of the time for (48a)-type questions, and 86% of the time for (48b)type questions. Answers from children between 3;7 and 4;4 were target-like 63.8% of the time for (48a). GPOV in sure-constructions were consistent with the findings from expletive stories. Children developed a good understanding of the GPOV at an earlier stage, and the use of GPOV was not only associated with the expletive construction.

4.2 Discussion of results Children’s responses in the two experiments showed that they did not blindly apply subjective meaning to every case, and they were able to interpret expletives with a GPOV from early on. They knew that expletives, without for-PPs, assumed truth in a common ground, and they further noticed that the common ground usually only consisted of humans. One strength of our approach is that it allows us to distinguish our GPOV from the customary reference to speaker-POV or speakerfactivity (see Guerzoni, 2003; de Cuba, 2006; Roeper, 2011). After receiving a yes or no we typically asked why. The why-questions we asked distinguished the two, as in (51): (51)

A house with a small window that is just large enough for a small boy, John, to crawl through, but no one else can. Is it hard to crawl through the window?

For instance in (51), the “yes”-children almost never answered “Yes, because I find it hard”. Instead, they would say “Yes, but not for Johnny”, for the why-question.¹⁶ On the whole, when we looked at the answers to the sure-stories, we found that a number of answers stated some facts they learned from the story, reverting to a GPOV or more narrow speaker-POV: (52)

a. b.

Is Johnny sure that he will win the race? Target answer: Yes.

16 Notice this is only true for the “yes”-children, which are children who answered the question like adults.

222 | Tom Roeper


Children’s answers: a. b.

He’s too small, he’s too small to win the race. (4;3) Because I think he’s too small and I think he doesn’t have much energy. (4;6)

Still the high error rate in sure-that-responses suggested that children might adopt a GPOV to interpret sure-that-questions. Given the contrast with infinitives, it was likely that the syntax of these constructions stood as obstacles, and children assigned a non-target underlying structure to sure-that-constructions. The real question in (54a) might be transformed to (54b) where children possibly mistook the meaning of sure in the raising construction as an indication again that it was a factive adjective like aware which, challengingly, would introduce a false presupposition (Is John aware that he will win ⇒ false presupposition): (54)

a. b.

Is John sure that he will win? Is it sure that he (John) will win?

Studies on false belief show that children often fail to link the complement clause to the subordinating matrix clause verb (e.g. de Villiers, 1995, 2000) at an early stage which serves as another kind of default (see de Villiers & Roeper, 2011). On the other hand, sure-to-constructions do not allow for such a structural reanalysis, but rather a raising-operation. Nevertheless, the experiment showed that children did perform better for sure-to-questions. (55)

Johnny is sure to win.

One reason why children produced some non-target answers, particularly children under 6.0, for sure-to-questions could be that they had overgeneralized the use of sure based on what they knew of glad where no raising is required and a factive implicature on the infinitive is entailed: (56)

a. b.

Johnny was sure to win. Johnny was glad to win.

Studies on raising predicates report that children take time in learning different kinds of raising predicates (e.g. Chomsky, 1969; Solan, 1979; Becker, 2005).¹⁷

17 Solan (1979) studied as well an interesting case of object-control for which there is no expletive that could be replaced. In (i) it is noteworthy that another kind of non-raising object control structure is needed because (ia) does not correspond to (ib): (i)

a. b.

The tiger is pretty to look at. *It is pretty to look at the tiger.

A default general point of view |


In sum, there is ample evidence from both naturalistic and experimental evidence that GPOV is available to children as a plausible default assumption.

5 Conclusion We have argued that assertions entail a presupposition of a GPOV and that all PPTs have this as a logical alternative. Therefore the “subjective” or relativist view captures only one side of the story. For children, the recognition may not be immediate that both a) the syntactic necessity of an implicit or explicit benefactive argument is called for in control structures, and b) pragmatics of accommodation must apply to implicit arguments. Therefore they initially treat all statements with predicates of taste as assertions of truth which entail a GPOV. For control structures, like those in our experiments, we have argued that children may begin with a PROarb construction that is not a specific subcategorization from particular adjectival predicates. Experience must teach them that certain adjectives, like smart require an implicit argument and take controlled factive and non-factive infinitivals. Those implicit benefactives are either GPOV, or when subject to contextual accommodation, more finely restricted in terms of time, location, and human differences, which seems to be very compatible with a centered-world account of possible world semantics. In general, we have argued that a GPOV notion built into the concept of an assertion is a default assumption for adults and children and remains as one option for all PPTs as well. The refined semantics of PPTs is a second order phenomenon that engages situational semantics and pragmatics. It is best seen as being always in contrast to the assertion of truth which entails a GPOV perspective. Acknowledgment: Thanks to the audience at the DGfS workshop on subjectivity, and Angelika Kratzer, for discussion and her broad support for the perspective adChildren may go through a stage where they prefer an object-control analysis before raising is recognized, since it is common to hear (iia), although it is ungrammatical to say (iib): (ii)

a. b.

Ice cream is delicious to eat. *It is delicious to eat ice cream.

Thus the acquisition path that covers all the forms eager/easy/pretty should be carefully established. Even in the pretty/delicious cases the adult interpretation is that delicious is not confined to the speaker. Clauss & Hartman (2015) provide evidence using nonsense adjectives, that children are very attentive to the syntactic frames.

224 | Tom Roeper

vocated here which opposes the relativist tradition. I am also especially indebted to Hazel Pearson, whose ideas have been partly incorporated here, for discussion leading to the revision of the original paper. Thanks to Chloe Gu (primary experimenter and searcher), to Barbara Partee’s seminar on implicit arguments and PROarb, and to the editors, Janneke van Wijnbergen-Huitink and Cécile Meier, and to the reviewers for many helpful suggestions. Thanks as well to Peggy Speas, Jill deVilliers, Bart Hollebrandse, and all the members of the Evidentiality Project for providing the background for this work, and, as always, the LARC acquisition community at UMass. Errors and misinterpretations are probably—depending on your point of view—my responsibility. The work was supported by the NSF Evidentiality grant to M. Speas, T. Roeper, J. deVilliers, J. Garfield.

References Anand, Pranav & Andrew Nevins. 2004. Shifty operators in changing contexts. Proceedings of SALT 14. 20–37. Becker, Misha. 2005. Learning verbs without arguments: The case of raising verbs. Journal of Psycholinguistic Research 34(2). 165–191. Chomsky, Carol. 1969. The acquisition of syntax in children from 5 to 10. Cambridge, MA: MIT Press. Chomsky, Noam. 1986. Knowledge of language: Its nature, origin, and use. New York: Praeger. Clauss, Michael & Jeremy Hartman. 2015. Syntactic cues in adjective learning. Paper presented at the Boston University Conference on Language Development (BUCLD 40) held in Boston, Boston University. de Cuba, Carlos. 2006. The adjunction prohibition and extraction from non-factive CPs. In Donald Baumer, David Montero & Michael Scanlon (eds.), Proceedings of the 25th West Coast Conference on Formal Linguistics, 123–131. Somerville, MA: Cascadilla Press. Guerzoni, Elena. 2003. Why even ask? Cambridge, MA: MIT dissertation. Hinzen, Wolfram. 2007. An essay on names and truth. Oxford: Oxford University Press. Hollebrandse, Bart. 2000. The acquisition of sequence of tense. Amherst, MA: University of Massachusetts dissertation. Kratzer, Angelika. 1991. Modality/Conditionals. In Arnim von Stechow & Dieter Wunderlich (eds.), Semantik: Ein internationales Handbuch zeitgenössischer Forschung/semantics: An international handbook of contemporary research, 639–656. Berlin: De Gruyter. Landau, Idan. 2000. Elements of control: Structure and meaning in infinitival constructions. Dordrecht: Kluwer Academic Publishers. Landau, Idan. 2014. A two-tiered theory of control. MIT Press. Unpublished manuscript. Lasersohn, Peter. 2005. Context dependence, disagreement, and predicates of personal taste. Linguistics and Philosophy 28(6). 634–686. MacWhinney, Brian. 2000. The CHILDES project: Tools for analyzing talk. Mahwah, NJ: Lawrence Erlbaum Associates.

A default general point of view

| 225

Moltmann, Friederike. 2010. Relative truth and the first person. Philosophical Studies 150. 187–220. Pearson, Hazel. 2013. A judge-free semantics for PPTs. Journal of Semantics 30. 103–154. Potts, Christopher & Tom Roeper. 2006. The narrowing acquisition path: From expressive to declarative. In Ljiljana Progovac, Kate Paesani, Eugenia Casielles & Ellen Barton (eds.), The syntax of nonsententials: Multidisciplinary perspectives, 183–201. Amsterdam: Benjamins. Rau, Jennifer. 2013. Acquisition of presupposition. Unpublished manuscript, University of Massachussetts. Roeper, Tom. 1987. Implicit arguments and the head complement relation. Linguistic Inquiry 18(2). 267–310. Roeper, Tom. 2007. The prism of grammar: How child language illuminates humanism. Cambridge, MA: MIT Press. Roeper, Tom. 2011. How the emergence of propositions separates strict interfaces from general inference. In Ingo Reich, Eva Horch & Dennis Pauly (eds.), Proceedings of Sinn und Bedeutung 15, 55–78. Saarbrücken: Universaar–Saarland University Press. Roeper, Tom & Tim Roeper. 2010. Game theory and syntactic control. In Jan-Wouter Zwart & Mark de Vries (eds.), Structure preserved. Studies in syntax for Jan Koster, 299–312. Amsterdam: Benjamins. Solan, Lawrence. 1979. The acquisition of tough movement. In Fred R. Eckman & Ashley J. Hastings (eds.), Studies in first and second language acquisition, 83–97. Rowley, MA: Newbury House. Speas, Margaret. 2004. Person (and mood and tense) and indexicality. Paper Presented at the Harvard Workshop on Indexicals, Speech Acts, and Logophors, Harvard University, Cambridge, MA. Stephenson, Tamina. 2007. Judge dependence, epistemic modals, and predicates of personal taste. Linguistics and Philosophy 30(6). 487–525. van Wijnbergen-Huitink, Janneke. 2016. Subjective meaning: An introduction. In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 1–20. Berlin: De Gruyter. [This volume]. Verrips, Maaike. 1996. Potatoes must peal: Acquisition of the passive in Dutch. The Hague: Holland Academic Graphics. de Villiers, Jill. 1995. Steps in the mastery of sentence complements. Paper presented at the Biennial Meeting of the Society for Research in Child Development, Indianapolis, IN. de Villiers, Jill. 2000. Language and theory of mind: What are the developmental relationships? In Simon Baron-Cohen, Helen Tager-Flusberg & Donald Cohen (eds.), Understanding other minds: Perspectives from autism and developmental cognitive neuroscience, 83–123. Cambridge: Cambridge University Press. de Villiers, Jill & Tom Roeper (eds.). 2011. Handbook of generative approaches to language acquisition. New York: Springer. Winter, Yoad. 2001. Plural predication and the strongest meaning hypothesis. Journal of Semantics 18. 335–365.

Michael Hegarty

Subjective meaning and modality Abstract: The case offered by Lasersohn (2005) for evaluating statements of personal taste relative to a judge, specified as a dedicated index for model-theoretic interpretation, is based on the argument that no form of mere contextual dependence on the speaker (or a group to which the speaker belongs) and reflected in the content of the utterance is adequate to explain the facts. The present work argues that Lasersohn overlooks a contextual account of the interpretation of these judgments that is adequate to explain the facts, involving generic quantification over a community to which the speaker belongs. On this view, there is no need for a further judge parameter; the speaker is given in the usual context parameter, and the community for evaluation is contextually determined through its connection with the speaker. Reviewing the literature on epistemic modal interpretation, we can likewise resist an approach to the evaluation of epistemic modal judgments as relative to the knowledge of an assessor, by appeal to knowledge distributed through a group to which the speaker belongs. On this view, there is no need for a further assessor parameter; the speaker is given within the usual context parameter, and the modal base is determined in relation to the speaker. Thus, we resist proposals which would link these two domains as exhibiting relativity of interpretation to a model-theoretic index, in favor of a proposal that both domains exhibit a standard form of contextualism.

1 Introduction Lasersohn (2005) considered and discarded a variety of approaches to the interpretation of statements of personal taste (PT), such as those in (1); these will be called PT judgments hereafter. (1)

a. b.

This is fun. This chili is tasty.

Finding no satisfactory account of the interpretation of judgments made with these predicates using model-theoretic interpretation relative to a speaker, time and possible world of evaluation, he proposed adding a judge index to the conMichael Hegarty, Law School Admission Council, 662 Penn Street, Newtown, PA 18940, USA, e-mail: [email protected]

228 | Michael Hegarty

text, so that statements using these predicates are evaluated relative to an individual, often (but not always) determined by context to be the speaker. The present article argues that there is a satisfactory model-theoretic interpretation of PT judgments that Lasersohn overlooked, relative to the standard indices of context (including speaker) and possible world, with an implicit argument over a group or community to which the speaker belongs (reducing to the speaker alone as a limiting case). But the group must be understood to have special, contextually determined properties. Implementing the semantics otherwise as Lasersohn does, the group is the denotation of a for-argument of the PT predicate. Some recent research in the semantics of modal interpretation similarly proposes an assessor for the interpretation of epistemic modal statements, so that an epistemic modal is interpreted as relative to the knowledge of the assessor. Stephenson (2005, 2007) generalizes these results and Lasersohn’s and argues that a wide range of statements are interpreted relative to a set of indices including an assessor or judge. Although the present article cannot address the full range of phenomena discussed by Stephenson, it invokes work on epistemic modality by DeRose (1991) and von Fintel & Gillies (2008) to argue that the case of epistemic modality is parallel to that of PT judgments, but along the lines proposed here—that they both involve a speaker-centered, contextually determined group, but are not relative to an assessor or judge index on model-theoretic denotations.

2 Lasersohn’s argument for a judge index Lasersohn’s argument proceeds by elimination. He considers various formulations of model-theoretic interpretation of PT judgments relative to a context consisting of a speaker index, a time index, and a world index, in which the only sort of indexicality allowed is the sort exhibited by pronouns, by which the content of the evaluated statement varies depending on the contextually determined value of a variable to potentially yield different truth values (for example, as I am a doctor varies in content from speaker to speaker, and can thus vary in truth value). These will not be comprehensively reviewed here, but a salient subset of them will be, in order to provide a point of departure for the approach to be developed in Section 3 below. First, Lasersohn considers the idea of endowing PT predicates with an implicit indexical argument, whereby fun, for example, means fun for x. I will call this an experiencer argument, since that is its thematic role. This implicit experiencer matches the overt argument in expressions such as fun for Max, but if it is left implicit, the indexical is typically resolved as the speaker. He rejects this

Subjective meaning and modality |


idea since it fails to make sense of a dialogue such as that in (2) (from Lasersohn’s examples (12) and (14)), an example of faultless disagreement. (2)

John: This chili is tasty. Mary: No, the chili is not tasty.

Lasersohn notes that if John’s utterance in (2) means tasty (the-chili, John), and Mary’s utterance means ¬tasty (the-chili, Mary), then Mary’s response is not interpreted as a contradiction to John’s, as it would be in any typical instance of the dialogue in (2). Next, Lasersohn (2005, 650) considers the idea of interpreting PT judgments as “making indexical reference to some relevant individual or group, not necessarily the speaker” to serve as experiencer. In typical examples, the group would consist of the speaker and addressee(s), or a larger group to which the speaker and addressee(s) belong. In (2), if John’s utterance is interpreted as saying that the chili is tasty for members of the group consisting of himself and Mary, then Mary’s utterance, based on Mary’s lack of appreciation for the chili, could be interpreted as asserting that the chili is not tasty for members of the group (since Mary doesn’t find it so), and thus as properly contradicting John’s claim. But Lasersohn points out that this analysis does not work when the order of the two utterances is reversed, as in (3) below (Lasersohn’s (17)), uttered as Mary and John ride a roller coaster. (3)

Mary: This is not fun. John: Oh, yes it is!

If Mary’s utterance is interpreted as asserting that the roller coaster is not fun for both of them, on the grounds that it isn’t fun for her, then John’s follow-up utterance in (3) could only be interpreted as asserting that the roller coaster is, in fact, fun for both of them. This interpretation is bizarre, but it is the only one available under this analysis. Another option considered by Lasersohn, following Bhatt & Izvorski (1995), is that when an explicit experiencer argument is not present, then PT predicates are interpreted as involving generic quantification over an implicit experiencer. Roughly, (1a) and (1b) mean that this is fun, or chili is tasty, for people in general, admitting a few exceptions, in line with other generic statements. Lasersohn argues, in effect (and I think, correctly), that this view involves a disconnect between the experience of the speaker and the reported experience, and that this disconnect is empirically disconfirmed. If (1a) and (1b) report on what people generally find fun, or tasty, then a speaker who doesn’t share the judgment should be able to use either of (1a) and (1b) in a quite neutral way, even though they don’t

230 | Michael Hegarty

share the judgment. This disconnect surfaces in examples in which a speaker says, “This is not fun, although I am having fun doing it”, (Lasersohn’s (22b)), or “This chili is tasty, even though I personally hate it”. Lasersohn seems right about this disconnect: it is odd for a speaker to issue a PT judgment that the speaker disavows unless an experiencer other than the speaker is specified. But the view that PT judgments are interpreted as generic quantification over experiencers would seem to allow this use. The flip side of this is that it should be contradictory for a speaker to issue a PT judgment, and at the same time deny that it is generically shared, as in Lasersohn’s example (23), “This is fun, but most people would hate it.” Lasersohn’s judgment, which seems correct, is that this is not a contradictory assertion, and is in fact quite imaginable in a context where the speaker is aware of the unpopularity of their own form of fun. The possibility of contradiction in (2) seems to preclude the possibility that PT predicates vary in meaning with the experiencer. If that were so, there should be no guarantee of common meaning between John’s use of tasty in (2) and Mary’s use, which would support Mary contradicting John. At the extreme, the colloquy in (2) would be akin to John asserting that he and Mary are at the bank (by which he means a financial institution), and Mary responding that no, they are not at the bank (by which she means a riverbank). When the referents of terms within S shift this way between the assertion of S, and the response “No, S is not true,” contradiction would typically not be perceived as going on. But contradiction is the typical reading of Mary’s response in (2). Lasersohn considers the possibility that PT judgments don’t have truth values, but rejects this through consideration of syntactic environments for PT judgments which seem to implicate truth values, as when they are arguments of truth predicates (e.g., John: This chili is tasty. Mary: That’s true. / That’s not true. / That entails that it has taste.), or occur in conditional sentences or as terms in classical argument forms such as modus ponens which are semantically explicated in terms of truth. Affective expressions that are plausibly not truth conditional, such as “Ummm” or “Wheee!” don’t have these properties.¹ Lasersohn allows that a PT judgment, like any other assertion, is evaluated relative to a speaker index, a time index, and a world index. Thus, I like the chili might be true when uttered by Max, and false when uttered by Kim, simply because the contents of the two utterances are different. Max asserts the content like (Max, the-chili), and Kim asserts like (Kim, the-chili). The speaker index resolves the pronouns differently, and thus leads to assertions that differ in content. The world index can lead to a difference in truth values even with the same

1 But see the critique of this argument by Gutzmann (2016) [this volume].

Subjective meaning and modality |


content. To use one of Lasersohn’s examples, The chili contains pork might be false in the actual world, but true in various other worlds. Assuming that the subject the chili and the predicate contains pork are rigid designators across worlds, the content is the same in all worlds, even though this content is true in some worlds, and false in others (assuming that the same pot of chili can be prepared with pork in one world, and without it in another). Lasersohn proposes that model-theoretic semantics be formulated so that all sentences are evaluated relative to an experiencer judge index, as well as relative to speaker, time and world indices. For non-PT predicates such as contains pork or be a doctor, the effect of the judge is neutralized by the condition that the interpretation function, which assigns an intension to each predicate, is constant on judges, assigning the same intension to such a predicate across judges. But for a PT predicate such as tasty or fun, the intension assigned will depend on the judge index. As a result, two PT judgments, such as John’s utterance of This is fun and Mary’s utterance of This is fun, can vary in truth value even if (everything else being equal) they have the same content. The judge index can be identified with the speaker, yielding strictly personal PT judgments. Or it can serve as an objective standard for PT judgments (in a specific PT domain, say, chili tasting), about which people can disagree and contradict one another, as in (2) and (3). The interpretation of an overt for-clause such as for Kim (meaning that Kim is a PT experiencer) is defined so that fun(x, for Kim) is true of x in a context, just in case there is a context in which x is fun is true relative to the judge Kim. To provide a thumbnail sketch of Lasersohn’s semantics (with a couple of simplifications and adaptations that make no difference to the force of his proposal), Lasersohn defines a logical language fragment (which he calls a “toy language”), with the elements of predicate and sentence logic, which is interpreted in a “structure” or model. Just to be clear, Lasersohn does not interpret expressions of a natural language directly in the model; implementation of a system along the lines he sketches would proceed most faithfully to his sketch by translating natural language expressions into a formal language L, and interpreting the formal language in a model. The model M = ⟨C, D, T, W, I⟩ consists of a set of utterance contexts C, a domain of individuals D, a set of times T, a set of possible worlds W, and a set I of interpretation functions, one for each predicate or constant term of L, which maps this predicate or constant to its value in the model. A context c in C specifies speaker sp(c), a time t(c) in T, a world w(c) in W, and a PT judge j(c) in D.² A predicate α is mapped by its interpretation

2 A fuller sketch would have the context specify addressees and places, to resolve the referents of a wider range of pronominals.

232 | Michael Hegarty

function I α to a subset of D which depends on a judge, world and time; we write [[α]]M,c,j,t,w = I α (j, t, w). Given a non-PT predicate such as contain-pork, for any judges j and j󸀠 , Icontain-pork (j, t, w) = Icontain-pork (j󸀠 , t, w) for all t ∈ T and w ∈ W, reflecting the judge-independence of such predicates. A sentence translated to a well-formed formula φ has a content in M and c, written as {φ}M,c , defined as a function on indices j ∈ D, t ∈ T, and w ∈ W, to truth values Truth and Falsehood, defined in (4). (4)

{φ} M,c (j, t, w) = Truth, iff [[φ]]M,c,j,t,w = 1

Then finally, a formula φ, and thereby any sentence that translates to it, can be specified as true in context c (relative to model M) as in (5), just in case its content evaluates to Truth at the judge, time and world indices given by c. (5)

{φ}M,c (j(c), t(c), w(c)) = Truth

So a sentence with an indexical pronoun will have a denotation, [[φ]]M,c,j,t,w, which depends on how the referent of the pronoun is resolved in c. But its content, {φ}M,c , effectively abstracts out the values of j, t and w, while keeping the resolution of the pronoun fixed. We can thus talk about the content {φ}M,c , where the referent of the indexical pronoun has been resolved, but whose Truth or Falsehood cannot be resolved without specification of the judge, time and world of evaluation. In this way, PT judgments have the same content from one experiencer judge to another, but need not have the same truth value from one experiencer judge to another, just as Max is smiling has the same content, but not necessarily the same truth value, from one time to another, and Max is a firefighter has the same content, but not necessarily the same truth value, from one possible world to another.³ Lasersohn proposes that the speaker would normally be chosen as the judge if the experiencer is not overtly specified, though the judge index wouldn’t have to be chosen as the speaker. As examples in which the speaker isn’t chosen, Laser-

3 The ordinary semantic evaluation of a sentence would be the truth conditions for the righthand side of (4), spelled out in terms of the semantic values of its component parts. These truth conditions allow most sentences to be evaluated at a time distinct from that in the context c, and at a world of evaluation distinct from the actual world. But Lasersohn’s account of the evaluation of a personal taste judgment given in (5) is notable in that it fixes the time and world of evaluation, as well as the judge. Many personal taste judgments have an immediacy about them that does indeed tie them closely to the context of utterance. But a sentence such as That is relaxing, or That could be relaxing, said of a hot mineral spring that is distal from speaker and addressee, could be a challenge for this aspect of Lasersohn’s analysis. This point will be taken up in connection with (8b) below.

Subjective meaning and modality | 233

sohn offers ones such as, This cat food is tasty, uttered by one cat-owner to another based on the behavior of the speaker’s cat, or, Was it fun?, asked by mother when father brings a child back home after riding a pony at the State Fair, meaning “Was it fun for the child?”. If the speaker is essentially a default for the judge, then we automatically explain the anomaly of examples such as This is not fun at all, although I’m having fun doing it (Lasersohn’s (22b)), and this is reinforced to the extent that the although-clause establishes the speaker as the salient judge for this clause and therefore (since the judge index applies to the interpretation of the full utterance) as the judge for the entire sentence. Most of semantic interpretation can be accomplished by evaluation relative to an index for context (including speaker, addressee, time and place of utterance) and an index for possible world of evaluation. Values for these parameters serve to fix the referents of indexical expressions that do not have content on their own; beyond that, the semantic value of an expression depends on its content. This might be lexical content, if the expression is word-level, or else it might be content determined through semantic composition, or by fixing values from the context—for example, the form of a definite nominal signals the cognitive status of the referent (Gundel et al., 1993), or its place in a referential hierarchy (Ariel, 1990). As Lasersohn admits, his account of the semantics of PT predicates and judgments is somewhat radical, in that it introduces a new judge index which is specially needed for PT predicates. A reflection of the extremity of the account is the assumption that, for non-PT predicates, the interpretation function is constant across judges, thereby neutralizing the judge index for all predicates except the class under study. The adoption of a special index for the interpretation of PT judgments is nevertheless justified, according to Lasersohn, to the extent to which there are no viable alternatives that cover the facts of PT interpretation equally well. The next section challenges the latter part of Lasersohn’s argument.

3 Community as a contextual parameter Lasersohn rejected the idea that PT judgments could generically quantify over an implicit argument. His examples are reproduced in (6) and (7). (6)

This is not fun at all, although I’m having fun doing it.


This is fun, but most people would hate it.

If Sam is engaging in some activity which she finds fun, but recognizes that few people would share her interest in it, then under the generic quantifier theory,

234 | Michael Hegarty

the main clause of (6) evaluates as true, generically quantifying over people in general. But the although-clause is also true, being a report of Sam’s subjective experience. So the generic quantifier account predicts that (6) is true, while, in fact, it is at least infelicitous, and arguably false on the intuition that the main clause is falsified by the speaker’s enjoyment of the activity. In (7), given the content of the but-clause, the generic quantifier account predicts that the first clause is false since it is not generically true that people in general find the activity fun. Yet (7) seems perfectly natural and felicitous as an intuitively true characterization of Sam’s feelings about the activity in which she is engaged, and how other people would regard it. The problem lies with the generic truth of the main clause in (6), which is predicted to be true, but isn’t, and the first of the conjoined clauses in (7), which is predicted to be false, but isn’t. A salient feature of these counter-examples is that they quantify over all people, admitting exceptions as generics do, but in an impersonal way, disconnected from the speaker. This accords with the behavior of some standard generic examples, such as, Horses are smart, or Cats are playful. But this sort of impersonal generic isn’t the sort typically involved in PT judgments. We examine a wider range of PT examples, and some non-PT examples with which PT-judgments nevertheless have more in common than they have with standard generics. This will help to illuminate the particular form of genericity involved in the interpretation of PT judgments. Consider (8a), uttered by Sam while soaking in a hot mineral spring, or (8b), uttered by Sam while walking about in Yellowstone National Monument, pointing toward the hot spring. (8)

a. b.

This is relaxing. That is relaxing.

There are two levels of genericity that can be involved in these examples. First of all, (8a) could be interpreted with the speaker as implicit for-argument, “This is relaxing for me”. But it could have a more general interpretation, as asserting that it would be relaxing for most normal, healthy people to be sitting where the speaker is at the moment of utterance, in the hot mineral bath. The variant in (8b) has the second interpretation, but more generically over time or situations: that it would be relaxing for most normal, healthy people to sit in the hot mineral bath at any time. The generic quantification is distinguished by both its force and its domain; it is quantification over most (force) normal, healthy people (domain), recognizing that people who are not healthy enough to withstand the heat, or who are waterphobic, etc. would not find the activity relaxing. There is a normative element to

Subjective meaning and modality | 235

the domain of the generic operator; its domain contains people whose physical constitution falls within normal boundaries. The normative element in this case is “normative” in a descriptive, naturalistic sense, not a prescriptive sense, meaning, for example, that the domain contains people who fall within one standard deviation of the mean on a scale of health relative to preparedness to enjoy hot mineral springs. Consider (9a,b), uttered by Sam after watching a clip from The Daily Show with Jon Stewart with a group of friends. (9)

a. b.

That was funny. That is funny.

First of all, either can be interpreted with the speaker as implicit to-argument: that was, or is, funny to me. We put this aside. More typically, (9a) would mean that the clip would have been seen as funny by anyone with political sensibilities within a certain range. This has a descriptive dimension of normativity; it could also mean that the clip would be funny to any right-minded viewer, which gives the interpretation a prescriptive dimension of normativity. The present tense in (9b) can make the same statement, but abstracted from the time and occasion of viewing, asserting that the clip would be funny to any normal person with political sensibilities within a certain range, or to any right-minded person, who viewed it at any time. If (9a) were instead a response to a nerdy skit lampooning a school of macro-economic theory, the domain of the generic quantifier would be quite narrow, a community of scholars in that field, and could be further, and more prescriptively, narrowed to the “right-minded” among them. Finally, (9a) could be uttered by Sam to her friends after they observe something happen which is only funny because of a shared “in joke” within their peer group. In that case, (9a,b) means that the observed event would be funny for those who share the microculture of the peer group (descriptive normativity), or that it should be regarded as funny for anybody within that micro-culture (prescriptive normativity). The example in (10), uttered by Sam in reference to the BP oil spill in the Gulf of Mexico, is similar, implicitly invoking the shared cultural standards of a community at some level, asserting that people with normal sensibilities will find the spill outrageous. (10)

The oil spill is an outrage.

The normativity involved in this interpretation can be descriptive, if we suppose that Sam would assume that people’s sensibilities could be ranked on an objective scale (on environmental sensitivity, say), and that (10) would then be interpreted as asserting that people within a mathematically defined normal range on this

236 | Michael Hegarty

scale would find the spill outrageous. Or it could be intended in a more prescriptive sense: that any right-thinking person should find the spill to be outrageous. A community can be invoked in a descriptively normative way for the evaluation of some non-PT predicates, including predicates of causal efficacy. Regarding the efficacy of a cold remedy, Sam could utter the judgment in (11). (11)

This stuff is a good remedy; it worked for me.

This generalizes from Sam’s experience to that of other people, based on their observed or presumed similarity to Sam. The relevant “community” would consist of human beings, and the normativity condition would be that their physiology is sufficiently normal (or sufficiently similar to Sam’s) as to not produce significant variation in the efficacy of the remedy. For another domain of non-PT judgment where a community is invoked under normative conditions appropriate to that domain, consider the example in (12), uttered by Sam in doing trigonometry proofs.⁴ (12)

It was/is clear that the sum of the squares of the sine and the cosine is one.

This is not a PT judgment, but a judgment about cognition. But it has the same sort of normativity on an invoked community as the PT judgments above. It could, of course, be a statement just about Sam’s cognition, if Sam is merely describing her own experience at the culmination point of her working through the proof. But we put this aside. It could also mean that any person of normal intelligence who had the necessary background and followed the proof to its conclusion would find the conclusion to be clearly true (or would find that the proof is clearly valid and that its conclusion is clearly true). This is more likely to be descriptively than prescriptively normative. But the community invoked, against which the norm is computed, is very contextually specific. Each of these examples, whether PT or not, can be interpreted as invoking a community that includes the speaker, whether one which extends as broadly as all humanity, or as narrowly as a micro-culture. Each involves generic quantification over members of that community, where the generic quantification is implicitly restricted by normative conditions, whether descriptive or prescriptive, appropriate to the predicates involved and the judgments being made. And implicitly, in each case the speaker regards him- or herself as a normal member of the community in question, for whatever form of normativity is invoked.

4 Cohen (2010) also discusses sentences with such cognitive predicates and proposes that they be evaluated along the same lines as PT judgments.

Subjective meaning and modality | 237

Of course, as Lasersohn noted, there are examples in which the speaker is not a member of the invoked PT community, as in the example, This cat food is tasty, based on the behavior of the speaker’s cat, or Was it fun?, asked by mother to father about Billy’s pony ride, to mean, “Was it fun for Billy?”. We could simply take the speaker-centered description of the community given above and replace the speaker with a contextually determined variable. Then we would revise the account to say that PT judgments invoke a community centered about some individual, with generic quantification over that community. But, in fact, the examples of non-speaker centered PT judgments feel more like displacements, whereby the speaker-centered interpretation is computed as discussed above, and the speaker then replaced with the cat or the child. This impression is heightened by the fact that this sort of PT judgment seems most natural when centered about beings such as a cat or child who cannot be presumed to speak for themselves. For this reason, we will assume that the community is invariably speaker-centered, as the normal interpretation, but that the speaker can be replaced with another individual in a special interpretation exhibiting displacement of the center. Other types of sentences can invoke a community in the way described above, though not necessarily with the same restrictions. Consider a deontic modal statement such as Max must appear in court tomorrow. This expresses an obligation or expectation according to codified or understood standards within a community, or the imperatives of a recognized authority within the community. The community must include the person or people to whom the modal is intended to apply (Max, in the example given), but it need not include the speaker, and in this respect deontic modals differ from PT judgments. The community can be all of humanity, or defined by a particular political or legal entity (a nation, a club, a student body at a university), or it could be a culture, or a micro-culture (for example, in A surfer should not cut in front of another surfer). For the deontic modal, following Kratzer’s (1991) theory of modal interpretation, a conversational background consisting of a modal base and an ordering source is part of the semantic interpretation. The community for which the deontic modal statement has force is contextually supplied or understood, but not specified in the semantics. With this understanding, the generic quantification involved in the interpretation of PT judgments is not impersonal; it is speaker-centered, in that the judgment holds of the speaker and some wider group of people, a community of which the speaker is a member, other members of which share relevant characteristics with the speaker. They share, at the very least, human physical and cognitive nature; they typically involve at least one further shared feature, such as a shared profession, or age cohort, or peer group, or hobbies and enthusiasms. This type of PT judgment is typically based on personal experience, generic over people whom the speaker presumes to be such as would share that experience.

238 | Michael Hegarty

Another potential counter-example to the idea that PT judgments invoke a speaker-centered community and quantify over a normatively selected portion of it is the observation that PT judgments can be made according to codified criteria, rather than based on the speaker’s personal experience. Suppose that Sam, an expert on impressionist painting, has been enlisted as a judge for a contest of modern art, about which she knows little. Each of the judges is given a set of criteria for judging modern art. After examining a painting before her and consulting the criteria, Sam pronounces on it as in (13). (13)

That is a good painting.

Sam has not consulted her aesthetic experience for this judgment. But she invokes the judging criteria, and the criteria are based on the aesthetic judgments of experts who formulated them. These judgments involve a community of critics or modern art aficionados whom the experts presume to share the criteria (at least, they presume that the right-thinking among them do). This seems to be another case of displacement. The view developed here will be formalized below. But first, we make sure that it covers the empirical bases highlighted in Lasersohn’s article. In (14), John’s utterance invokes a community to which he belongs, which might be all of humanity or a micro-culture of chili enthusiasts, and asserts that normal members of the community should find the chili tasty. (14)

John: Mary: John: Mary:

This chili is tasty. No, the chili is not tasty. You’re crazy! No, you’re crazy!

If he addressed his remark to Mary, he would typically be presuming that she is a member of that community; her response, in any event, takes him to have done so. Mary’s reply presumes that she is a normal member of the community invoked by John, and denies that the chili is tasty for normal members of that community. Fixing these pragmatic parameters of interpretation, this challenges John’s PT judgment outright. In this case, Mary could suggest that John “re-think” his evaluation, or that he cleanse his palate and taste the chili again. Continued disagreement is likely to undermine the presumption of shared community, or of mutual normality of place within that community, implicating that John is not normal, or not a normal chili aficionado in good standing. This is how the dialogue in (14) plays out.

Subjective meaning and modality | 239

Next, consider Lasersohn’s example in (3), repeated here. (15)

Mary: This is not fun. John: Oh, yes it is!

If PT judgments universally quantified over a group to which speaker and addressee both belonged, John’s response is bizarre since Mary has already, in effect falsified this interpretation of John’s utterance. But on the present account, John and Mary are disagreeing, once again, on the evaluation of a mutual experience for normal members of a relevant community (which in this case could be as small as their peer group, or just the two of them). And persistent disagreement could cast doubt on the normality of one another within this community. For the reasons given in this section, we propose that there is an implicit argument analysis which Lasersohn did not consider, and which handles the data perfectly well, without introducing relativity to a judge or assessor index. The implicit argument is an experiencer argument, and it is speaker-centered in that it consists of either the speaker, or a group to which the speaker belongs, where the group shares with the speaker his or her basic physical and cognitive nature (for we would hesitate to declare that aliens from another star system should share our tastes and preferences), and cultural background, which can extend down to the level of micro-culture. Furthermore, where the group is larger than just the speaker, the PT judgment is understood to hold generically over members of that group. To hold generically means to hold for normal members of the group, members falling within normative parameters by virtue of their shared nature and culture. We take a PT judgment which is intended to hold just for the speaker as a limiting case of this. To formalize this idea, we will interpret syntactic tree structures for natural language expressions directly into the model using English combined with elementary set theory as the metalanguage in which to express conditions in the model. While natural language expressions are italicized (as usual), their corresponding syntactic tree structures will be indicated in bold type. We assume that the context c includes a time, along with specification of the speaker and addressee(s). The indices for evaluation therefore include just a context and a possible world. In line with the discussion above, a PT predicate such as fun, used by a speaker in a context, invokes a community C of potential experiencers to which the speaker belongs (assuming a typical example, and not one involving displacement), and generically quantifies over members of C. The generic quantifier Gen over a domain X in a possible world v is defined as universal quantification

240 | Michael Hegarty

over members of X that fall within normal bounds, as in (16), where φ(y) is a metalanguage statement that is open in the variable y. (16)

(Gen v y : y ∈ X in v) : φ(y) in v iff (∀z : z ∈ X in v ∧ z is normal in X with respect to φ) : φ(z) in v

This is a strict construal of Gen, realized as a universal quantifier over normal members of X. We could define a less strict Gen operator using the generalized quantifier MOST in place of the universal quantifier. Then the adjective phrase fun is interpreted as in (17), which we can abbreviate as in (18), where GEN w abbreviates the quantification in (17) over possible worlds accessible from w. (17)

[[fun]]M,c,w = λx[for all worlds v accessible from w : [(Gen v y : y ∈ C and y experiences x in v): x is fun for y in v]]


[[fun]]M,c,w = λx[(GEN w y : y ∈ C ∧ Exp(y, x)) : fun(x, y)]

The context c is an ordered list which includes the speaker as (we will presume) its first element: c = ⟨speaker, addressee, time, place⟩. This supports the ordinary indexicality of pronouns I, we, you, as well as here and now references, used to determine the content of the proposition which is evaluated in model M at world w. The lexical semantics of the AP fun given in (18) triggers the identification of a community C, of people who share some of the properties in the following ordered list with the speaker, including the initial property and some subset of the subsequent properties: human nature (physical and cognitive), broad cultural background, profession, peer group, hobbies and enthusiasms. Profession, peer group and hobbies can delineate a micro-culture, such as the culture of World of Warfare game enthusiasts, weightlifters, joggers, Jane Austen enthusiasts, or the like. C would not have to be a world-wide community; it could be as small as a group of four or five friends with shared background and interests. In an extreme case, it could be just the speaker and addressee, or in the limiting case, it could be the speaker alone. With this understanding, the choice is then not one of speaker versus community, for the evaluation of predicates of personal taste; it is simply a choice of various communities including the speaker, of whatever size, with the speaker alone as the limiting case. On this account, the community invoked for the interpretation of PT predicates is an inferred property of the speaker. It bears much in common with what goes on in the resolution of pronoun reference when a speaker says, with regard to an upcoming election, We are going to win. The referent of the pronoun we in this case could be a political party, a class of society, members of a profession, the entire nation (for example, if the speaker is uttering a post-partisan bromide), or even all of humanity (if the speaker is very idealistic). In the example just given,

Subjective meaning and modality |


the pronoun we conventionally signals that it’s referent is a highly salient group to which the speaker belongs. Similarly, we propose that a PT predicate conventionally signals that it invokes a contextually determined group to which the speaker belongs. The specification of this PT community is part of the determination of the content of a PT judgment. On this view, the interpretation of a PT predicate is relative to a context, and thus indexical to the speaker, but not relative to a judge. If we were to write out something analogous to Lasersohn’s content reproduced in (4) above, it would depend on a possible world index, but not a judge index.⁵

4 Epistemic modal evaluation Lasersohn’s proposal that PT judgments are evaluated relative to a judge index is parallel to proposals made in the literature on epistemic modality, that epistemic modal statements are evaluated relative to an assessor index, where the assessor may be different from the speaker, and cannot uniformly be identified with any group to which the speaker belongs. Interestingly, the proposal made in Section 3 above is parallel to proposals made by other researchers on epistemic modality, who argue that epistemic modal statements invariably invoke the knowledge of a group to which the speaker belongs. Without making a case for either of the two approaches to epistemic modal interpretation, we briefly review this literature here in order to place the present proposal in the theoretical context of this parallel across the two empirical domains. This section recapitulates part of von Fintel & Gillies (2008), differing by a new example, a slightly more particular view of the interpretive procedures involved, and ongoing comparison with the discussion and results of Section 3 above. In a model theory for modal logic, a statement of modal necessity universally quantifies over accessible possible worlds, and a possibility modal existentially quantifies over accessible worlds. For the interpretation of modal statements in natural language, Kratzer (1991) argued that a binary valued (true or false) accessibility relation is too blunt an instrument, and proposed that modal statements be evaluated with respect to a two-part conversational background consisting of a modal base and an ordering source. The modal base of an epistemic modal is a set of propositions that confines the modal generalization to worlds consistent with what the speaker knows. The ordering source is a set of propositions that partially order possible worlds according to which propositions in the ordering source are

5 Unlike Lasersohn, we have put time in the context since we take the context to include everything that can be indexically specified.

242 | Michael Hegarty

true of them. A necessity modal statement can then be interpreted so that, rather than unrealistically stating that the prejacent (the statement without its modal) is true in all possible worlds in the modal base, it states that the prejacent is true for worlds which rank above a certain threshold determined by the satisfaction of propositions within the ordering source. Kratzer took the epistemic modal base to be a set of propositions expressing knowledge available to the speaker. This view is problematic for examples such as the constructed discourse in (19).⁶ (19)

[Monday. Mary, Sally and George at their office in New York City] Mary: Sally:

Where’s Joe this week? He might be in Boston. The draft duty roster for this week listed him there. George: No, he can’t be in Boston. I just saw him in the hallway five minutes ago. The problem in (19) is how we can make sense of the fact that George can truthfully contradict Sally’s modal statement. If Sally’s modal statement is evaluated with respect to her modal base, what she knew at the moment of her utterance, then it is true and George’s denial of it is false. If George’s modal statement is evaluated with respect to his own modal base, what he knew at the moment of his utterance, then George doesn’t contradict Sally since Sally’s statement was not made based on George’s evidence. And it doesn’t help if Sally’s modal statement is interpreted as accommodating a shared modal base. The shared base would be the minimum of knowledge shared by Sally and George, and given that George knows more in this case (or what counts as more for the purpose of determining where Joe is), the shared modal base would therefore be what Sally knows, so Sally’s modal statement would evaluate as true, and George’s contradictory statement as false. The problem is compounded by the observation of MacFarlane (2003) that (20a) is an appropriate follow-up for Sally, but not (20b). (20)

Sally: a. Oh, then I guess I was wrong. b. #Oh, then I guess he can’t be in Boston. But I nevertheless uttered a true statement when I said “He might be in Boston.”

6 This is adapted slightly from von Fintel & Gillies (2008), and was discussed previously by MacFarlane (2003) and Hawthorne (2004).

Subjective meaning and modality | 243

If Sally’s statement is interpreted with respect to her modal base, then the response in (20b) should be a natural follow-up for her. The fact that it isn’t shows that something else is going on. Further compounding the problem, there are examples in which epistemic modals seem to be evaluated relative to the knowledge of an eavesdropper, as in the example in (21), also adapted from MacFarlane (2003). (21)

Bond has infiltrated SPECTRE’s Zürich headquarters, planted evidence indicating that he might be in Zürich, and hidden a bug in the conference room. He has then escaped to London, where he and Leiter monitor transmissions from the bug. Blofeld convenes a meeting in the conference room in Zürich and announces, “Bond might be in Zürich.” Eavesdropping on this, Leiter says, “Ha! Blofeld is wrong again.”

It would be possible in this example for Leiter to make allowances for what Blofeld knows, and to respond with something like, “Well, Blofeld is right, for all he knows.” But the fact remains that Leiter’s less generous response in (21), evaluating Blofeld’s modal statement according to what Bond and Leiter know, is equally felicitous, showing that Blofeld’s modal base need not be interpreted as narrowly speaker-centered. MacFarlane offers this example as evidence that the truth value of Blofeld’s modal statement is relative to an assessor, which must be specified for modal interpretation.⁷ Choosing Leiter as the assessor, Blofeld’s modal statement is false. Returning to (19) and (20) with the assessor approach in mind, George knows more about Bill’s location than anybody in the discourse at the moment of his utterance (since he has already incorporated Sally’s observations and weighed them against what he knows), so George has salience as an assessor for the evaluation of modal statements in this discourse. Based on what George knows, Sally’s modal claim in (19) is false, and George has grounds on which to contradict her. Upon recognizing this, Sally can sensibly retract her prior modal claim in (20). This can be implemented by adding an assessor index to Kratzer’s modeltheoretic interpretations, and enforcing the condition that the epistemic modal base reflects the knowledge of the assessor. Stephenson (2005, 2007) combines the assessor approach to epistemic modal interpretation with Lasersohn’s judge index, arguing for a general model-theoretic relativism to world-time-individual indices, where the individual is realized as a judge or assessor in the domain of PT judgments and epistemic modal statements.

7 See also discussion in Egan (2005), Egan et al. (2005), von Fintel & Gillies (2008).

244 | Michael Hegarty

There is a curious fact though, brought out most clearly by von Fintel & Gillies (2008). The theory of the assessor index should allow that the assessor could be anybody who might have an interest in the truth of the prejacent of the modal statement. But our intuitions do not seem to allow the assessor to be as flexible as this. In (19), it would be odd to choose Sally as assessor for the evaluation of modal statements in the discourse; the more knowledgeable discourse participant seems to be a better assessor. Furthermore, although an assessor can be at the periphery of a discourse, as peripheral as an eavesdropper, it seems that the assessor cannot be outside the discourse altogether. The example in (22), from von Fintel & Gillies (2008), illustrates this. (22)

Detective Parker has been going over some old transcripts from Al Capone’s court case in the 1920s. Capone is being asked whether his underlings, might, without his knowledge, have stashed money from a bank heist in a particular safe. Capone [on tape]: The loot might be in the safe. Parker [85 years later]: ??Capone was wrong. / ??What Capone said is false. The safe was cracked in the 1980s, and there was nothing there.

It would be natural for Detective Parker, living in the present, to say that Capone and the 1920s prosecutor were misguided in thinking the money might be in the safe. But the judgment that their modal statements on the matter were false seems to be a particularly uncharitable evaluation of their modal claims. It seems, in fact, to be an extreme point on a range of interpretive options. The felicity judgments in (22), given by von Fintel & Gillies, suggest that the assessor for Capone’s modal statements cannot be too far removed from the context of Capone’s utterance, and Detective Parker, speaking nearly a century later, is too far removed to be accommodated into a shared modal base with Capone. A series of examples explored by DeRose (1991) strongly support an analysis of epistemic might statements as based on knowledge of a group to which the speaker belongs. And as pointed out by von Fintel & Gillies (2008), examples with eavesdroppers, such as (21), can be accommodated by this view, provided that membership in the group can be counted quite loosely, encompassing eavesdroppers at the periphery of the group. In support of this idea, there are examples in which knowledge on which an epistemic modal statement is based is distributed throughout members of a group in such a way that several members of the group hold bits of this knowledge, but no single member has all of it to put together and serve as assessor for the modal statement. Such an example is constructed in (23),

Subjective meaning and modality

| 245

a discourse among members of a Wall Street investment group, as they monitor trading on the floor of the stock exchange. (23)


Look at the numbers for the past hour! It’s possible that the market is crashing. George: No, I talked to all the traders this morning, and they made it clear that there are no major sell-offs in the works today. That’s just a temporary dip. Molly: Actually, somebody changed their mind at the last minute: 2G stock is being sold off today. That’s what you’re seeing. But here’s a list of indexings, and you can see that 2G stock is not indexed to any others, so this is clearly an isolated event. Nothing to worry about. Sam: Oh, no! There’s a mistake in that list. 2G stock is indexed to Intel, Google, GE, and other heavyweights. So Tom was right: it is possible that the market is crashing.

There is a set of worlds compatible with what Tom, George, Molly and Sam know in which the traders reported that there were no major sell-offs, but they were wrong this once; in which the indexing list said that 2G was not indexed with others, but it was wrong this once; in which 2G stock is actually selling off and is indexed to several major stocks, and thus, in which it is indeed possible that the market is crashing. The set of worlds consistent with this distributed knowledge is obtained by taking, for each discourse participant Tom, George, Molly and Sam, the set of worlds consistent with what that participant knows, and then taking the intersection of these across all four discourse participants. The examples from the literature reviewed here, together with (23), suggest that epistemic modal statements invoke a group G containing the speaker, which can extend as wide as eavesdroppers, but reduces in the limit to the speaker alone. The epistemic modal base is the knowledge of the group, defined as follows. For each x ∈ G, let Kx (w) be the set of propositions expressing what x knows at the world of evaluation w, where each of these propositions is a set of possible worlds. Then ∩Kx (w) is the set of possible worlds consistent with everything that x knows in w. So, finally, ∩{∩Kx (w): x ∈G} is the set of possible worlds consistent with what each member of G knows. The most knowledgeable member has the smallest set of worlds consistent with what they know, since greater knowledge excludes progressively more possible worlds from consideration. But as seen in (23), the intersection need not correspond to ∩Kx (w) for any single x ∈ G. In this way, the epistemic modal base reflects the best knowledge available within the group.

246 | Michael Hegarty

Among researchers in epistemic modality, the view given above is nearly articulated by von Fintel & Gillies (2008). However, their view is actually slightly broader. Although acknowledging that the epistemic modal base can reflect the knowledge of any group to which the speaker belongs, and showing that many examples can be accommodated by construing the notion of group broadly enough, von Fintel & Gillies do not propose that the epistemic modal base reflects the best knowledge available within the group; it could reflect the knowledge of a particular person within the group, even if that person is not the most knowledgeable. In (24), cited by von Fintel & Gillies (2008, 83), an epistemic modal statement seems to be evaluated relative to the addressee’s knowledge or evidence. (24)

Mordecai and Pascal are playing Mastermind, a board game with colored pegs which are inserted into holes in a board. During each turn, one player, the codemaker, places a row of pegs behind a screen, and the other player, the codebreaker, tries to figure out the sequence of colors in the hidden row, using feedback from the codemaker. Pascal, playing codebreaker, hesitates over his next move to guess the sequence which has been set by Mordecai, and the following dialogue ensues. Pascal: There might be two reds. Mordecai: That’s right. There might be two reds.

In this discourse, Mordecai’s utterance serves as an affirmation of Pascal’s thought processes, specifically affirming Pascal’s modal statement. The modal base for this affirmation can only reflect Pascal’s knowledge, not Mordecai’s. In this case, the knowledge reflected by the epistemic modal base is still knowledge within a group containing the speaker, but it isn’t the best knowledge of the group, since Mordecai knows more than Pascal. As a result of including this example, von Fintel & Gillies (2008) propose that there is a set of alternative contexts available for epistemic modal evaluation, each based somehow on knowledge of a group to which the speaker belongs, but not necessarily reflecting the best knowledge of the group. We could analyze Mordecai’s modal statement in (24) as involving a form of displacement, akin to the PT judgment The cat food is tasty, expressing the speaker’s report that the speaker’s cat (not the speaker himself) finds the cat food tasty, or The pony ride was fun, expressing the parent’s judgment that Billy enjoyed the pony ride. On this approach to (25), the speaker-centered group would be displaced to put the addressee at the center for pedagogical purposes, and then chosen to be as narrow as possible, consisting of the center alone, which after displacement, leaves the addressee as the sole member of the group. The main reason for adopting the displacement approach to such non-speaker-centered statements

Subjective meaning and modality | 247

in both domains is purely intuitive—that many speakers report an intuition that these cases are somehow special, derivative of the normal processes of interpretation of PT judgments and modal statements. For present purposes, no argument will be given for this over the broader range of alternative contexts proposed by von Fintel & Gillies for epistemic modal evaluation. Acceptance of this view of the example in (24) would not change the present thesis much, since the epistemic modal in (24), like the others considered here, does invoke a group to which the speaker belongs. Taking this example at face value merely prevents us from adopting the view that the modal base is ∩{∩Kx (w): x ∈G} , the best knowledge available within the group G.

5 Conclusion Judgments of personal taste and preference and epistemic modal statements are similar in that both have prototypical examples in which the interpretation is speaker-determined—based exclusively on the speaker’s tastes or the speaker’s knowledge. In both domains, as a broader inventory of examples was accumulated, it became clear that semantic evaluation relative to the speaker is not a general property of sentences of either of these two types. In both domains, the variety of examples available motivated some researchers to propose general relativity to a model-theoretic index, an assessor whose knowledge is invoked as the modal base for epistemic modal statements, and a judge whose personal tastes and preferences are invoked for evaluation of PT judgments. The present work, following examples in the literature in both domains, argues that relativity to a judge or assessor index would lead us to expect a greater robustness of arbitrariness in the choice of judge or assessor for the interpretation of statements in these two domains than we actually observe. In contrast to what the judge and assessor indices would lead us to expect, judgments in both domains tend to be typically very speaker-centered (though not speaker-determined), involving or invoking a group or community of which the speaker is a member. Examples that diverge sharply from this by evaluating relative to an individual who lies outside the most salient groups or communities at issue, seem to have a special status, as “off center” versions of normal statements in these domains, in which the normal center has been displaced. This rejects the view that PT judgments and epistemic modal statements are alike in relativity to an index for a specified individual (judge or assessor), and asserts instead that they are alike in that they contextually invoke a community which is prototypically speaker-centered.

248 | Michael Hegarty

Acknowledgment: The author expresses appreciation for helpful feedback from participants in the Subjective Meaning Workshop (Berlin, 2010), where an early version of this paper was presented, and especially to Cecile Meier for more recent discussion of the analysis presented in Section 3.

References Ariel, Mira. 1990. Accessing Noun-Phrase antecedents. London: Routledge. Bhatt, Rajesh & Roumyana Izvorski. 1995. Genericity, implicit arguments and control. Presented at the Seventh Annual Student Conference in Linguistics, University of Connecticut, Storrs, 8–9 April, 1995. Cohen, Ariel. 2010. A tasty mixture. Unpublished manuscript. Ben-Gurion University of the Negev. DeRose, Keith. 1991. Epistemic possibilities. The Philosophical Review 100. 581–605. Egan, Andy. 2005. Epistemic modals, relativism, and assertion. In Jon Gajewski, Valentine Hacquard, Bernard Nickel & Seth Yalcin (eds.), New work on modality, vol. 51 MIT Working Papers in Linguistics, 35–62. Cambridge, MA: MITWPL. Egan, Andy, John Hawthorne & Brian Weatherson. 2005. Epistemic modals in context. In Gerhard Preyer & Georg Peter (eds.), Contextualism in philosophy: Knowledge, meaning and truth, 131–168. Oxford: Oxford University Press. von Fintel, Kai & Anthony Gillies. 2008. CIA leaks. The Philosophical Review 117(1). 77–98. Gundel, Jeanette K., Nancy Hedberg & Ron Zacharski. 1993. Cognitive status and the form of referring expressions. Language 69. 274–307. Gutzmann, Daniel. 2016. If expressivism is fun, go for it! In Cécile Meier & Janneke van Wijnbergen-Huitink (eds.), Subjective meaning: Alternatives to relativism, 21–46. Berlin: De Gruyter. [This volume]. Hawthorne, John. 2004. Knowledge and lotteries. Oxford: Clarendon Press. Kratzer, Angelika. 1991. Modality/Conditionals. In Arnim von Stechow & Dieter Wunderlich (eds.), Semantik: Ein internationales Handbuch der zeitgenössischen Forschung/Semantics: An international Handbook of Contemporary Research, 639–656. Berlin: De Gruyter. Lasersohn, Peter. 2005. Context dependence, disagreement, and predicates of personal taste. Linguistics and Philosophy 28(6). 643–686. MacFarlane, John. 2003. Epistemic modalities and relative truth. Unpublished manuscript, now superseded by MacFarlane (2011). MacFarlane, John. 2011. Epistemic modals are assessment-sensitive. In Andy Egan & Brian Weatherson (eds.), Epistemic modality, 144–178. Oxford: Oxford University Press. Stephenson, Tamina. 2005. Assessor sensitivity. In Jon Gajewski, Valentine Hacquard, Bernard Nickel & Seth Yalcin (eds.), New work on modality, vol. 51 MIT Working Papers in Linguistics, 179–206. Cambridge, MA: MITWPL. Stephenson, Tamina. 2007. Towards a theory of subjective meaning. Cambridge, MA: MIT dissertation.

Index accessible possible worlds, 241 acquisition, 204, 207–215 adjectives – color words, 121–123, 158 – denominal adjectives, 120 – dimension vs. evaluation, 109–115, 121–123, 138 – dimensional adjectives, 11–13, 109, 113, 138 – evaluative adjectives, 11–14, 109, 112, 130, 138 – positive form, 112, 115–118, 125 – vague adjectives, see vagueness adjectives dimensional, 150 agreement, see also disagreement – faultless agreement, 183–185 – faulty agreement, 183–185 assertion operator, 80, 161 assessment, 133, 175, 185 – qualitative assessment, 113, 115, 119–123 assessor, 228, 243–244, 247, see also judge beautiful, 136, 138, 151, 153 cold, 191 commissive speech act, 30 community, 233, 241 comparative, 112, 138 comparison class, 109 consider, 111 context, 178, 236, 239–241 contextualism, 1–4, 8, 14, 70, 91–93, 97, 102, 106, 124, 206–207 – epistemic contextualism, 64 – presuppositional contextualism, 102 control, 203, 204, 208–212 damn, 31, 37 denial, 5, 14, 26, 128, 130, 136–139, 155, 159–165 deontic force, 41 deontic modals, 155, 237 dimensional predicates, see adjectives, dimensional directive speech acts, 29

disagreement, 47–56 – faultless disagreement, 4–9, 12–14, 16, 21, 23–71, 78, 82–83, 87, 91, 93–94, 105–106, 112–118, 124, 127, 170, 176, 188–192, 196 – faulty disagreement, 171, 174, 175, 177, 189–193, 196 – ordinary disagreement, 171, 175 – simple disagreement, 94–100 displacement, 237, 238, 246 epistemic modals, 5, 8–12, 14–15, 91, 132, 186–188, 228, 241–247 epistemic step, 70, 79–80 error theory, 48, 61–65, 67 evaluation, 2 – objective evaluation, 75–77 – objectivized evaluation, 77–78 – subjective evaluation (degree), 109 – subjective evaluation (truth value), 70–75 evaluative predicates, see adjectives, evaluative evidence – direct evidence, 175–177 – indirect, 145 – indirect evidence, 175–177 evidential step, 80–81 evidentiality, 174, 186, 196 experiencer, 170–171, see also judge, 175, 180, 183, 185, 186, 189–196, 228 expressive content, 29, 31, 34, 36 expressive speech act, 26 expressive speech acts, 25 expressives, 26, 31, 212, 215 expressivism, 25–34 find, 5, 9–12, 111, 155, see also German finden folk theory, 48, 53 for-phrase, 190–193, 213 fun, 194–196, 240 generic sentences, 152 generic statements, 229–230, 234–235, 239–240

250 | Index

genericity, 6, 9, 17 German finden, 155–160, see also find gradable adjectives, 5, 11, 58, 116–117, 121, 153 green, 121 hard, 191 hybrid semantics, 35–38 implicit argument, 2, 128, 131, 132, 204, 207–213, 223, 228 indexicals, 23, 24, 32, 36, 38, 228, 232, 240–241 interjections, 25, 28, 34, 45 judge, 1–9, 170, see also assessor judge argument, 108, 118, 119, 122 judge index, 228, 231–233, 239, 241, 243 judge-dependent, 16, 131, 134, 137, 156 knowledge, 244–247 large, 150 logical reasoning, 26, 41 measure function, 122 – generalized measure function, 150 – subjective measure function, 120, 122 measurement, 113, 117, 122, see also assessment – qualitative measurement, 123, 150 – quantitative measurement, 123, 150 metalinguistic usage, 130 mixture model, 77–78, 81–84 modal base, 237, 241–247 modal logic, 241 model, 231 modus ponens, 27, 29 normative meaning, 35, 39, 41 normativity, 137, 235–236 objectivism, 47–48, 52–56, 60–61 oops, 35 ordering source, 237, 241–242 personal taste judgments, 228–233, 247 point of view, 202, 203, 216 positive morpheme, 117

pragmatics, 14–17, 169, 171, 173, 178, 180, 185, 189, 193, 196–198 predicates of personal taste, 91, 109, 122–124, 169, 197, 240 – acentric use, 131 – egocentric/autocentric use, 3, 6, 71, 97, 131 – exocentric use, 97, 122, 131 – generic use, 97 – group use, 97 prejacent, 242 presupposition of commonality, 100, 102 principle of charity, 64–65 principle of semantic competence, 48, 62, 134 PROarb, 203, 210, 212, 213, 223 propositions – descriptive usage, 130, 134, 138–150 – interpetational usage, 134, 138–150 – metalinguistic usage, 134 questions, 177–182, 185, 192 relativism, 1–4, 8, 12–14, 17, 70, 91–93, 97–98, 102, 106, 206–207, 228–233, 243 – moral relativism, 61 retraction, 15–16, 48, 50, 163, 242, 243 rigid designators, 231 salty, 122 scalar predicates, see adjectives semantic blindness, 62, 64, 67, 92, 99–100 source principle, 178 subjective attitude verbs, 107, 109–112, 118, 155 subjectivism, 55, 56, 61, 63 subjectivity, 105–109, 202, 204, 206 – kinds of subjectivity, 5, 9–14, 105, 124 tall, 140 tasty, 1, 2, 38, 40, 83, 194–196 theory of mind, 216 update – descriptive update, 153, 154 – interpretational update, 153, 154 use conditions, 35–38 use-conditional content, 37, 42, 43 vagueness, 78, 109, 111–112, 115–119