Abstract

The article discusses judge-dependence of gradable adjectives and degree constructions. Take the positive construction: whether sentences ‘John is tall’, ‘This cake is tasty’, or ‘Mary is smart’ are true or false depends not only on a state of affairs but also on whose opinion is being expressed. At the same time, the adjectives appearing in these subjective statements differ in ways that are often considered crucial for shaping an analysis of judge-dependence. Consider embedding under ‘find’ and the ability to take overt ‘judge’-phrases: (1) I find {this cake tastier than that one / John smarter than Mary / *Mary taller than John}. (2) {Apples are tasty / ??John is tall / ??Mary is smart} (?)for / to me. To explain these patterns, I look at different classes of subjective lexical items, limiting myself to the domain of degree constructions. I take predicates of personal taste (PPTs) like ‘tasty’ or ‘fun’ as a starting point, as they have received the most attention in the literature and are usually taken to be representative of the whole class of subjective predicates. A closer cross-linguistic look at more items like positive dimensional adjectives (‘tall’), evaluative adjectives (‘smart’), extreme adjectives (‘gigantic’) and modal degree morphemes like ‘too’ shows that, for most cases, there is no evidence for a judge argument, quite often associated with subjectivity across the board. At the same time, I argue that postulating a judge argument for ‘tasty’ or ‘fun’ and capturing judge-dependence of the other classes with a judge index of evaluation—‘two types of subjectivity’ view—is not a good solution as well. Taking an intuition that the extra argument of PPTs is an Experiencer rather than the judge seriously, I observe that, indeed, the presence of an extra argument correlates with reference to an experience event being part of the predicate semantics. On top of that, I show that the PPT Experiencer argument does not show any special judge-dependent behaviour that is not observed for other Experiencers in different constructions. In a nutshell, I adopt a judge index of evaluation account of subjectivity ( Lasersohn 2005 , 2009 ), but the role that I assume for the overt ‘judge’-phrases is different from what Lasersohn suggests. For me, these ‘judge’ PPs are not index shifters, but are Experiencer arguments of PPT predicates. Finally, I motivate and formulate a general principle regulating who can make a direct assertion about someone's internal state (‘Judge=Experiencer’ principle).

1 Introduction

This article discusses judge-dependence of gradable adjectives and degree constructions. Take the positive construction as an example: whether sentences like John is tall , This book is interesting , or Mary is smart are true or false depend not only on a state of affairs but also on whose opinion is being expressed. In this sense, such sentences are subjective . Indeed, these examples pass one of the classic tests for subjectivity—the ability to be embedded under ‘subjective attitude verbs’ like find (unlike, say, a predicate 4 years old ): 1 As subjectivity is an indispensable factor in the breakdown of the domain of individuals into the positive and the negative extension of tall , interesting and smart , an obvious question to ask is what is the right way of analysing this dependency between an opinion-holder, or judge, and the breakdown into P and not - P .

  • (1) a.  I find this book interesting.

  •   b.  I find John tall.

  •   c.  I find Mary smart.

  •   d.  *I find Vera 4 years old.

A related question concerns the semantics of overt expressions introducing the opinion-holder—so-called ‘judge PPs’, illustrated in (2): What is the contribution of these phrases to the semantics of subjective statements? I believe that a better understanding of judge PPs can help in understanding the ‘bare’ case, where no overt judge PP is present—and, consequently, result in a more linguistically motivated theory of subjectivity in natural language in general.

  • (2) a. This book is interesting for / to John .

  •   b. The ride was fun for John .

  •   c. The news was pleasing to John .

  •   d. This cake is tasty (?)for / to John .

The goal of the article is twofold: (i) clarify the semantics and the compositional impact of overt judge PPs found with subjective predicates; (ii) contribute to an understanding of how judge-dependence is encoded in language, focusing on degree constructions. To approach these goals, I address the data on a variety of judge-dependent, or subjective, predicates with different linguistic properties, and try to find a system in this diversity.

To see that there is indeed a puzzle to solve, observe the fact that gradable predicates and degree constructions that show judge-dependency form quite a diverse landscape with respect to their subjective properties.

First, at least according to the find -test, interesting and smart , but not tall , are subjective both in the positive and the comparative form: Secondly, interesting , but not smart or tall , can take overt ‘judge’ PPs – to - or for -phrases introducing an opinion-holder: Thus by only looking at three gradable adjectives we already see three classes of judge-dependent expressions:

  • (3) a.  I find this book more interesting than that one.

  •   b.  I find John smarter than Mary.

  •   c. *I find Mary taller than John.

  • (4) a.  This book is interesting for / to me .

  •   b. ??John is tall for / to me .

  •   c. ??Mary is smart for / to me .

  1. Interesting -class: subjective both in positive and comparative form, take judge PPs;

  2. Smart -class: subjective both in positive and comparative form, no judge PPs;

  3. Tall -class: subjective in positive but not in comparative form, no judge PPs.

How should this diversity be explained? Does this diversity force one to say that there are many ‘types of subjectivity’ in any theoretical sense or can these differences be factored out and accounted for independently? Could there be a fourth class—subjective only in the positive form, but taking a judge-PP? What regulates the subjectivity in the comparative form and the acceptability of judge PPs?

While subjectivity in the positive v. comparative form has received some attention in the literature ( Richard 2004 ; Saebø 2009 ; Paenen 2011 ; Kennedy 2013 ; Fleisher 2013 , a.o.), the distribution of judge PPs and what it tells us about the semantics of judge-dependence in general has basically never been studied. The major existing theories of subjectivity predict ‘judge PPs’ to be available in all subjective environments ( Lasersohn 2005 , 2009 ; Stephenson 2007a , b ; Stojanovic 2007 ). This is not correct, as the examples above show—judge PPs are only available in a subset of subjective contexts.

In this article, I will look for a reason for such a limited distribution of ‘judge PPs’. This task will require close examination of different kinds of predicates and constructions showing subjectivity—the types 1–3 illustrated in the schematic typology above, and more.

I will argue that the differences between these ‘types’ should not be encoded in the linguistic representation of judge-dependence. Instead, I suggest that the linguistic ‘device’ for judge-dependence remains constant across different classes of subjective items, despite the visible variability. Factoring out these complications that obscure the picture of linguistic encoding of subjectivity, or judge-dependence, I argue for an analysis of the classification in 1–3 above that involves one kind of subjectivity.

To give the reader a preview of my analysis: I will propose that the presence of judge PPs with certain subjective predicates is due to the experiential semantics of such predicates, which is an issue independent (to a certain extent) of the subjectivity of these predicates. The apparent extra argument that we see with some subjective predicates is the experiencer rather than the judge and thus it does not have to be treated as an essential part of subjectivity per se . The intuitive relation between the experiencer introduced by such predicates and the contextual judge is regulated by a principle that restricts the value of the judge parameter for the statements about one's non-externalisable experience.

This accounts for part of the sketchy typology of subjective items outlined above—namely, for the distinction between predicates that can co-occur with ‘judge PPs’ and the ones that cannot. Subjective predicates that come with a ‘judge PP’ are experiential predicates projecting an experiencer argument that previous studies have been referring to as the ‘judge’ argument; other subjective predicates are not experiencer predicates, and therefore do not project an extra argument. One more ingredient is needed to derive the rest of the typology—namely, understanding why some predicates are subjective in the comparative, while others are not. The answer is that those predicates that are subjective in the comparative form are lexically subjective, while adjectives like tall show subjectivity only in positive form because it is the positive morpheme POS that is subjective rather than the adjective itself. These decisions, together with a more articulated relation between experiencers of predicates like tasty or interesting and judges, is the core of a unified analysis of subjectivity taken up here.

This article has two main content blocks, or lines of investigation: one that establishes the structure and semantics of sentences with overt ‘judge PPs’—and one that looks for the right way to analyse judge-dependence in general. These two lines are intertwined logically and will be developed together linearly throughout the article.

The structure of the article is the following: first, in section 2 , I introduce the necessary background on subjectivity in general and the core of the theoretical debate on the nature of subjectivity. I sum up the existing motivation for postulating a ‘judge’ argument for English predicates of personal taste (PPTs) in section 3 . Then, in section 4 , I broaden the class of adjectives under consideration, discussing positive forms of adjectives like tall and the fact that they seem to lack an extra argument—an observation supported cross-linguistically. I consider the ‘two kinds of subjectivity’ analysis, in which subjectivity comes with a ‘judge’ argument for tasty and the like, and without a ‘judge’ argument for tall and the like. I argue that it is not explanatory enough in that it misses an important similarity between these alleged ‘judge’ arguments and experiencers found with experiencer predicates in general. I broaden the empirical domain further and include more predicates showing subjective behaviour: I consider evaluative adjectives in section 5 and extreme adjectives in section 6 . I show that the cross-linguistically stable correlate of an extra argument is a reference to an experience event in the semantics of a predicate. I address an apparent counter-example to this generalisation—the too -construction—in section 7 , showing that it, in fact, confirms rather than falsifies it. I close the article with a discussion of a relation between the experiencer argument and the judge, proposing a principle restricting the judge value to the experiencer in a description of a non-externalisable experience event. I will cash out this idea in terms of presupposition.

2 Subjectivity: The Background

2.1 Relative truth

Intuitively, there are statements whose truth is a matter of opinion (5-a) rather than a matter of fact (5-b) ( Lasersohn 2005 , 2009 ; Stephenson 2007a , b ; Anand 2009 ; Moltmann 2010 ; Pearson 2013 , a.m.o.): As simple as this intuition looks, sentences like (5-a) and the puzzles they give rise to are serious challenges for classic truth-conditional semantics. As one famous case, consider the following dialogue: (6) has a special status: on the one hand, A and B seem to disagree with each other, but at the same time neither of them appears to be at fault. This kind of exchange is known as subjective (or faultless ) disagreement ( Lasersohn 2005 , 2009 ; Stojanovic 2007 ; Stephenson 2007a , b ). Compare the dialogue in (6) with the one in (7) ( Paenen 2011 : 6): In (7), either Chris or Paul are saying something false (which could potentially be checked), while (6) is compatible with both of the speakers speaking truthfully, which is puzzling as they seem to contradict each other at the same time. A definition of faultless disagreement common in the linguistic literature is from Stephenson (2007a) : The cases described in (8), at least on the face of it, seem immune to a simple analysis in terms of truth or falsity. At the same time, the existing theories of truth-conditional meaning assume that all assertive sentences are amenable to such an analysis. This is the core of the puzzle posed by subjective disagreement— Kölbel (2002) calls it the problem of excessive objectivity of the existing theories of truth-conditional meaning. Accounts treating the excessive objectivity problem propose ways of introducing the opinion-holder into the picture and look for theoretical and empirical arguments that would help decide which option fits the data better. I next discuss two major classes of theories of subjectivity.

  • (5) a. Roller coasters are fun.

  •   b. Leo Tolstoy wrote ‘War and Peace’.

  • (6)  A : That ride was fun.

  •    B : No it wasn't!

  • (7) Paul just entered Chris' room. Chris is wondering how he came in.

  •    Chris : The door was closed.

  •    Paul : No, it wasn't.

  • (8) S ubjectivedisagreement :

  •   a. Intuitively, the interlocutors disagree with one another.

  •   b. There is a sense in which both speakers have said something true, so long as each was sincere in her expression of her opinion.

  •   c. For this reason, the disagreement does not seem to be one that can be resolved.

2.2 Two theories of subjectivity

Predicates that systematically give rise to subjective disagreement are known as predicates of personal taste (PPTs). I return to the question of which predicates exactly are the members of that class in more detail below, but for now it would be enough to refer to some defining properties that are discussed in the literature: (i) PPTs trigger subjective disagreement (when no overt ‘judge PP’ is present); (ii) PPTs can take overt ‘judge PPs’; 3) PPTs can be embedded under subjective attitude verbs like find .

There are two major classes of theories of semantics of PPTs, differing with respect to the way they introduce the judge, and, as a consequence, how they see the contribution of the ‘judge PP’: what I will call relativist theories introduce a judge parameter as part of the index of evaluation ( Lasersohn 2005 , 2009 ); contextualist theories make the judge an argument of a PPT that can be filled in contextually in the absence of an overt judge-phrase ( Stojanovic 2007 ; Stephenson 2007a , b ) (although Stephenson uses both the judge index and a null pronoun). The labels ‘contextualist’ and ‘relativist’ come from the philosophical tradition, but have recently been adopted in the linguistic literature (see Saebø 2009 ; Bouchard 2012 ; Fleisher 2013 , a.o.). Not everyone may agree that the use of these terms in linguistic discussions which this article is part of is accurate. I will use them as handy labels, making sure I am explicit enough about what I mean by these terms in what follows. I now describe both theories and list the linguistic facts that have led many researchers to choose contextualism over relativism for English PPTs.

2.2.1 A relativist semantics for PPTs

A relativist semantics of a PPT like tasty would include a judge index of evaluation along with the familiar world and time indices ( Lasersohn 2005 , 2009 ) (these denotations ignore the fact that tasty is gradable; I will return to this in section 9 ): Lasersohn assumes a Kaplanian (1989) approach to meaning, which distinguishes two kinds of meaning— character and content —and can be represented schematically in the following way: Kaplanian ‘character’ can be seen as a function from contexts to contents: when the values for indexicals contained in the context are supplied and the indexicality in the character is resolved, the result is what Kaplan calls content—which, in turn, can get evaluated with respect to truth value.

  • (9) a. 〚tasty〛 c ; w , t , j  = λ xe . x tastes good to j in w at t

  •   b. 〚This cake is tasty〛 c ; w , t , j  = 1 iff this cake is tasty to j in w at t .

  • (10)  Character  → Resolve indexicality →  Content  → Evaluate truth value

If a sentence contains an indexical, say, a first person pronoun I , when uttered by different speakers it will have the same character but different content. See the following dialogue as an example: It sounds strange that a speaker B starts with a no —there is no sense of contradiction if the sentence expresses different content relative to the two speakers. Thus contradiction is a matter of content and not a matter of character (ignoring the potential complication with a metalinguistic no ).

  • (11)  A : I am a doctor.

  •      B : #No, I am not a doctor!

This is how Lasersohn explains the sense of contradiction in the classic subjective disagreement case: speaker A asserts and speaker B negates the same content (e.g. the ride was fun ). What is different between the ride was fun uttered by different speakers is a parameter of evaluation of truth value.

The difference between PPTs and non-subjective items is that PPTs have a semantic value that can vary across judges for a fixed world and time. In other words, PPTs lack the property of judge - rigidity ( Lasersohn 2005 ; Anand 2009 ): With these pieces in place, we can again formulate the explanation for the puzzle of faultless disagreement under a relativist perspective. The conflict between the ‘faultlessness’ part and the ‘disagreement’ part is an illusion because these parts operate at different ‘levels’ of meaning. In a Kaplanian framework, the content of The ride was fun is the same no matter with respect to which judge the truth is evaluated. Thus, the dialogue including The ride was fun and its negation form a natural case for disagreement, as the content of B's response is the negation of the content of A's utterance. At the same time, A's and B's utterances are compatible with each other in the sense that they can both be true when evaluated under different judge indices: the ride can be fun for A and be boring for B. In a nutshell, Lasersohn's account relies on an assumption that the intuition of disagreement is operative on the level of content, while the ‘faultlessness’ part concerns the stage when the truth values are assigned.

  • (12) α is judge - rigid iff ∀ t , w , j , k 〚α〛 c ; t , w , j  = 〚α〛 c ; t , w , k

  •     where j and k are judge indices.

We can now return to the role of for - and to -phrases in the semantics of sentences with PPTs. If a judge is part of index of evaluation, then the judge for -phrase shifts the judge to the individual introduced by the for / to -phrase: Thus, in a relativist semantics, a for / to -phrase is a kind of a modifier. Let me now introduce an alternative existing analysis of PPTs and subjectivity.

  • (13) 〚α for / to β〛 c ; w , t , j  = 〚α〛 c ; w , t , b , where b = 〚β〛 c ; w , t , j

2.2.2 A contextualist semantics for PPTs

In a competing contextualist account, PPTs have a judge argument that can be filled by a silent pronoun ( Stojanovic 2007 ): The details of implementation differ: the pronoun might have different referential properties, or there might be a choice out of more than one silent pronoun to occupy the judge argument position, etc. As an example, in the analysis developed in Stephenson ( 2007a , b ), PPTs have a judge argument that can be filled by one of the two silent pronouns: a judge-denoting PRO j or a pro that denotes a contextually salient individual. Note that a pronoun denoting a judge directly is a departure from a strictly contextualist analysis as it involves a crucial reference to the judge index of evaluation: Let us turn again to the role of for / to -phrases in the semantics of sentences with PPTs. The semantics of a preposition introducing a judge phrase would be radically different under relativist and contextualist views. While for Lasersohn ( 2005 , 2009 ) for / to is a judge-shifter, under a contextualist account, in contrast, for / to would be semantically empty—it would be an identity function, as all it does is introduce a judge argument (the judge index is in brackets to indicate that it does not play a crucial role in this analysis but does not have to be removed): The motivation behind (strictly) contextualist analyses is based on reconsidering the status of faultless disagreement. Stojanovic (2007) argues that there is no such thing as faultless disagreement: all the apparent cases of faultless disagreement are real cases of misunderstanding—a misunderstanding concerning the referent of the judge argument in the opposing interlocutor's utterance. In a sentence like This ride was fun , the judge is left underspecified, and could be potentially the speaker him/herself, or the statement could be meant to be a more general, or generic one—a statement about the majority of the judges, or about the ‘normal’ judges according to the speaker etc. Judge genericity in statements with PPTs is discussed a lot in the recent literature, see in particular ( Anand 2009 ; Cohen 2010 ; Moltmann 2010 ; Pearson 2013 ).

  • (14) a. 〚tasty〛 c ; w , t ,( j )  = [λ xe . [λ ye . y tastes good to x in w at t ]]

  •    b. 〚This cake is tasty proc ; w , t ,( j )  = 1 iff this cake is tasty to 〚pro〛 in w at t .

  • (15) a. 〚PRO jc ; w , t , j  =  j

  •     b. 〚 pro 〛 is supplied by some salient individual in the discourse

  •     c. 〚This cake is tasty PRO jc ; w , t , j  = 1 iff this cake is tasty to j in w at t .

  • (16) 〚for / to〛 c ; w , t ,( j )  = λ ye . y

Thus, according to Stojanovic (2007) , statements involving PPTs with no overt judge-phrase are ambiguous between generic and non-generic readings. Following a dialogue like (6), A could recognize the misunderstanding and disambiguate his/her statement with something like (17-a) or (17-b): With this clarification, the disagreement seems to be taken away. In this way the core motivation for a relativist analysis of subjectivity developed in Lasersohn ( 2005 , 2009 ) would be removed.

  • (17) a.  A : I am only saying this ride was fun for me .

  •     b.  A : I found this ride fun, that's all I am saying.

Let me now turn to linguistic arguments that show that the status of the judge-PP with PPTs is more compatible—at least on the face of it—with the contextualist rather than with the relativist analysis of subjectivity.

3 ‘Judge PPs’ are Thematic

3.1 Arguments for ‘judge PP’ thematic status

There is evidence for PPTs having a ‘judge’ argument (in English). First of all, the very fact that judges can be expressed overtly opens a potential possibility of an analysis that reserves a slot for the judge in the PPT denotation, if these overt judge phrases can be shown not to have an adjunct status. Secondly, Stephenson ( 2007a , b ) argues that the choice of the preposition introducing the judge is idiosyncratic, does not have any semantic explanation, and has to be specified in the PPT lexical entry. This also strongly suggests the argumental status of the ‘judge PPs’: One more syntactic test shows that English ‘judge PPs’ pattern with arguments rather than with adjuncts: they are not separable from PPTs by an adjunct, in the same way as verbal arguments are not separable from their verb ( Paenen 2011 , based on Fults 2006 ): More motivating data come from stacking of arguments v. adjuncts: adjuncts are generally allowed to stack or iterate in a way arguments cannot (23). Glanzberg (2007) and Schaffer (2011) argue that ‘judge PPs’ pattern with arguments rather than adjuncts: While these contrasts look quite convincing, there have been debates in the literature on whether they force an argumental analysis of the ‘judge PPs’ ( Lasersohn 2005 , 2009 a.o.; for a summary, see Collins 2013 ). Moreover, there are facts that point in the opposite direction. Let us look at one of these facts—namely, the lack of crossover effect with PPTs.

  • (18) a.  fun for Sue / *fun to Sue

  •     b.  boring for Sue / ??boring to Sue

  •     c.  tedious for Sue / ??tedious to Sue

  •     d.  pleasurable for Sue / ??pleasurable to Sue

  • (19) a. ??pleasing for Sue / pleasing to Sue

  •     b.  *tastes good for Sue / tastes good to Sue

  •     c. ??tasteless for Sue / ?/OKtasteless to Sue

  •     d. ??funny for Sue / funny to Sue

  • (20) a. ??tasty for Sue / ?tasty to Sue

  •     b. ??delicious for Sue / *delicious to Sue

  • (21) a.  John ate [ arg an apple] [ adj yesterday].

  •     b. *John ate [ adj yesterday] [ arg an apple].

  • (22) a.  The apple was tasty [ judgePP to Eve] [ adj yesterday].

  •     b. *The apple was tasty [ adj yesterday] [ judgePP to Eve].

  • (23) a. *John kissed Mary Sally.

  •     b. John kissed Mary on the beach under the stars at midnight.

  • (24) a. *Sushi is tasty to me to Mary to everyone.

  •     b. *Roller coasters are fun for Ann for Ben for Claire.

Lasersohn ( 2005 : 681) observes that there is a contrast between an overt judge PP and the case when it is absent in the crossover environment—an environment where the moved element ‘crosses over’ a coreferential pronoun (see Stanley 2000 , a.m.o.). In the b-examples below a wh -item crosses over the position of an alleged judge argument (the a-examples are shown as a baseline as they do not involve crossover—the base position of the wh -word is higher than the judge): What is interesting here is the contrast between (25-b) and (26-b): they would both be expected to be degraded under the assumption that PPTs always have to project a judge argument, overt or covert. Under this view, the difference in the acceptability of crossover sentences between the overt and covert case is unexpected. Lasersohn uses this contrast to argue against the argumental analysis of PPT judges. 2

  • (25) a.  Who was upset that the ride wasn't fun?

  •     b.  Whom did the fact that the ride wasn't fun upset __ ?

  • (26) a.  Who was upset that the ride wasn't fun for him?

  •     b. ?Whom did the fact that the ride wasn't fun for him upset __ ?

Summing up, the tests for the argumental status of the ‘judge PPs’ are controversial and are subject to discussion. Interestingly, the same properties have been discussed in Rákosi (2006) for Dative experiencers on Hungarian data. Rákosi does not address the personal taste puzzle, but the observations and conclusion he makes about a class of constituents he studies is directly applicable to the ‘judge PPs’ with PPTs (furthermore, the class of ‘experiencer predicates’ that Rákosi discusses has a significant intersection with the PPT class I discuss here). The controversial properties of Hungarian Dative experiencers have led Rákosi (2006) to conclude that these phrases have a special status and fall within a category of ‘thematic adjuncts’—they are selected by a predicate lexically, quite like arguments, though have some syntactic properties similar to adjuncts, first of all, optionality.

Following Rákosi, I assume that these constituents are lexically introduced by a predicate, quite like canonical arguments are, despite their strange syntactic status that makes them different in some ways from the canonic arguments.

In what follows, I would not concentrate on the distinction between ‘thematic adjuncts’ and arguments, assuming that the former are not different from the latter in the relevant semantic sense, and their behaviour is exactly the same as the behaviour of certain classes of experiencers. This analogy is not a coincidence—in my analysis of PPTs that I will formulate in section 9 , I will treat ‘judge PPs’ as experiencer phrases.

For more details and tests of the status of such ‘mixed’ constituents with respect to the argument v. adjunct syntactic distinction, see Rákosi (2006) .

On the semantic side, it has been argued that there is a particular requirement holding between a PPT and its ‘judge’, which suggests a thematic relation between them. In particular, PPTs impose a direct experience requirement on their judge. Here is the relevant observation from ( Pearson 2013 ):

‘In order to assert that x is P for some taste predicate P, one typically must have direct sensory experience of the relevant kind on the basis of which to judge whether x is P . For tasty , for example, I must have tasted the object I am talking about …. If I have good reason to believe that shortbread is tasty, say because a reliable expert has told me so, I might say, Apparently, shortbread is tasty , but not, Shortbread is tasty ’.

Similar observations about constructions with PPTs have been discussed in the literature before, for example, Stephenson ( 2007a , b ) argues that in find  + PPT construction find introduces an inference that the subject is basing his/her opinion on direct evidence: (27) is odd as it suggests that Sam have tried cat food—an inference that disappears if find is replaced by think : Stephenson ( 2007a , b ) captures this fact by adding an evidential requirement to the semantics of find , otherwise similar to think : 〚find〛 = 〚think〛 + ‘and this is caused by x having a direct experience of p in w ’.

Anand (2009) observes the direct sensory experience requirement on the use of PPTs +  for/to -PP construction, which he analyses as part of semantics of the ‘judge PP’: These observations made in Stephenson ( 2007a , b ), Anand (2009) and Pearson (2013) uncover a similar requirement across PPT constructions—namely, a ‘direct experience’ requirement. At the same time, Stephenson and Anand encode this requirement not as part of PPT lexical semantics, but rather as the semantic contribution of other elements PPTs co-occur with—the verb find in find -construction, or for in the construction with the explicit ‘judge PP’. Importantly, what the find - and the for -PP-constructions have in common is an explicit judge (the subject of find and the complement of for ) that excludes the generic interpretation of the judge. I suggest that the direct experience requirement is the property of a PPT itself, but it can be lifted in certain environments, for example, if the statement is a generalisation (e.g. tasty for ‘normal’ perceivers in general).

  • (29)  Discussing a made to order entree at a much-favored restaurant :

  •   #Whatever she's making, it's tasty to me.

To these subtle facts in English I add a fact from Japanese which makes the direct experience requirement more visible. Japanese is known to have a first person constraint on the argument of direct perception predicates. This constraint is lifted when an evidential or a modal is used ( Kuroda 1965 ; Kuno 1973 ; Tenny 2006 ; McCready 2007 ): In contrast, predicates that do not describe direct perception, do not show any special behaviour of the 1st person v. 2/3-person subject—here illustrated with the predicate wakai ‘young’: This first-person constraint is usually explained in terms of privileged access to one's own sensations and emotions, about which no direct external assertion can be made.

  • (30) a. watasi-wa / *anata-wa / *kare-wa sabisii desu

  •      I- top         / you- top     / he- top    lonely cop

  •     ‘I'm/You're/He's lonely’

  •     b. Mary-wa  sabisii noda / ni tigainai        / mitaida / hazuda

  •      Mary- top lonely evid / there's.no.mistake / seems  / must

  •     ‘Mary must be lonely’

  • (31) a. watasi-wa / anata-wa / kare-wa wakai desu

  •      I- top      / you- top / he- top young cop

  •      ‘I'm/You're/He's young’

  •     b. Mary-wa  wakai noda / ni tigainai        / mitaida / hazuda

  •      Mary- top young evid / there's.no.mistake / seems  / must

  •      ‘Mary must be young’

I observe that PPTs in Japanese is another environment where the constraint is observed, restricting the judge of PPTs without evidentials to first person only (for the judge, which can be either a ni-wadat - top or ni-totte-(wa)dat-totte-(top) phrase – dative case marker plus totte morpheme with unknown etymology plus an optional topic marker): 3 The same holds for tanoshii ‘fun’, yoi ‘good’ and kimochiii ‘pleasant’. The first-person restriction on the judge of PPTs is the clearest possible illustration of the ‘direct experience’ requirement on PPT judges.

  • (32) a. watasi-ni-(totte)-wa /*John-ni-(totte)-wa   kono keeki-wa oisii

  •      I- dat - totte - top   /John- dat - totte - top this   cake- top tasty

  •      ‘This cake is tasty to me / to John’

  •     b. John-ni-(totte)-wa    kono keeki-wa oisii noda / ni tigainai

  •      John- dat - totte - top this   cake- top tasty evid / there's.no.mistake

  •      ‘This cake must be tasty to John’

I take the facts discussed above to be good enough reasons to postulate an extra argument for PPTs (in English as well as Japanese, and possibly cross-linguistically). The linguistic behaviour of judge phrases strongly suggests a thematic relation between the predicate and the judge phrase, which manifests itself in the direct experience requirement on the semantic side, and idiosyncricity of the choice of a preposition along with facts on stacking and separability from the predicate, on the syntactic side. The relativist analysis developed in Lasersohn ( 2005 , 2009 ) treats judge PPs as modifiers, which would make the whole range of properties described in this section unexpected. Therefore, I conclude that PPTs introduce an argument (or, more carefully, in Rákosi's terminology, a ‘thematic adjunct’) that presumably corresponds to the role of ‘judge’.

3.2 Complications

Let me now introduce some complications that enter the picture when more subjective adjectives and constructions outside of the PPT class are considered.

To count as a PPT, a predicate is usually taken to have the following properties: (i) it participates in statements whose truth is a matter of opinion; (ii) can embed under subjective attitude verbs like find ; (iii) it can take a ‘judge’-phrase; (iv) it possess a property of scalar variation —subjective ordering of objects along a scale (e.g. different people would order chocolate and apples on the scale of tastiness in different ways) ( Lasersohn 2009 ; Anand 2009 ).

Many predicates, such as interesting , smart , tall etc., are not PPTs although they have some but not all of these properties. I will show that a focus on PPTs only misses out on potential data concerning properties normally associated with subjectivity. Thus, by studying a wider class of adjectives, we can study subjectivity rather than its subtype known as ‘personal taste’, and the relation between personal taste and subjectivity per se .

In what follows, I will show that if one looks at a wider class of subjective predicates, going beyond the classic PPTs, it turns out that (i) not all subjective predicates project a ‘judge’ phrase; (ii) the ones that do not also do not exhibit the semantic effects of thematic relation between the judge and the subjective predicate—in particular, the direct experience requirement. I will use this fact to severe the licensing of ‘judge PPs’ from subjectivity per se and relate it to the experiential semantics of PPTs instead.

In the next sections, I address different classes of adjectives showing subjectivity effects in different degree constructions, with special attention to the status of expressions allegedly introducing the overt judge. I focus on how the semantic differences between these subjective predicates and PPTs correlate with the licensing of overt judge or the lack thereof.

I start with dimensional adjectives and the problems they pose for the contextualist analysis that seemed like a reasonable analysis for PPTs given the discussion in section 3 .

4 The Positive Form of Dimensional Adjectives

This section discusses dimensional adjectives (DAs) like tall , wide or heavy and their subjectivity. The accepted characteristic property of DAs is that they measure the object's characteristics along the measurable scales (such as height, width, weight etc.) ( Kennedy 2007 , and references therein). This contrasts with typical PPTs that order objects along their scale subjectively and thus are predicates of scalar variation (predicates with subjective ordering of objects along a scale, see Lasersohn 2009 ; Anand 2009 ). In contrast, DAs order objects by their weight or height independently of the opinion-holder: if John's height exceeds Bill's height, this ordering would not change depending on who is observing their height or making the statement.

Nevertheless, DAs are known to be subjective. Richard (2004) points out that they can show faultless disagreement (see also Anand 2009 ; Moltmann 2010 ; Paenen 2011 ; Kennedy 2013 ): They pass the other familiar test for subjectivity as well—namely, they can be embedded under find (see Bouchard (2012) for a comprehensive discussion, but also see Kennedy (2013) and Fleisher (2013) for a judgement that DAs under find are slightly degraded—I will return to this issue later on): Observe that DAs are only subjective in the positive form: DAs in comparative constructions do not trigger subjective disagreement and are definitely bad with find ( Paenen 2011 ; Kennedy 2013 ): As it is the positive form of a DA (POS-DA) that is subjective, the existing analyses locate judge-dependence of POS-DAs in the silent morpheme POS (37-a) that is often postulated in the positive construction (see Kennedy 2007 a.m.o.). The existing implementations of POS subjectivity mostly opt for a contextualist analysis, totally parallel to the PPT case ( Saebø 2009 ; Paenen 2011 ): 4 (37-b) adds an extra argument ze to the denotation of POS, corresponding to the judge. Contextual standard dst is dependent on the judge. The notation dst I use in (37) can be understood as a shortcut for the standard-calculating function norm that would take a gradable predicate and the judge as arguments: In (38), the ‘judge’ argument is used as an input to the standard-calculating function norm , so that the standard of comparison used by POS is now judge-dependent.

  • (33)  A : Carla is rich/thin/heavy/old/young/short.

  •      B : No she's not!

  • (34) John finds this river deep.

  • (35) a.   A : Apples are tastier than bananas.     subjective disagreement

  •              B : No, they are not!

  •     b.   A : John is taller than Bill.     objective disagreement

  •              B : No, he's not!

  • (36) a.  I find apples tastier than bananas.

  •     b. *I find John taller than Bill.

  • (37) a. 〚POS〛 = λ Gd , et λ xe . maxd . G ( d )( x )) ≻  dst

  •     b. 〚POS subjective 〛 = λ Gd , et λ xe λ ze . maxd . G ( d )( x )) ≻  dSTz

  • (38) 〚POS subjective 〛 = λ Gd , et λ xe λ ze . maxd . G ( d )( x )) ≻  norm ( G )( z )

Let us see if the tests that motivated an extra argument for PPTs give the same result for POS-DAs as well. The next subsection presents the data that show a systematic difference between PPTs and POS-DAs.

4.1 The data

The motivation for the argumental status of ‘judge PPs’ in section 3 consisted of two main pieces of evidence—syntactic and semantic. The first and obvious ‘syntactic’ consideration was the possibility to introduce the overt judge—and the idiosyncratic choice of the preposition introducing it. This argument (and further ones testing the status of these PPs) is not valid with POS-DAs as they seem to not take judge PPs at all: 5 The only way to specify the POS-DA judge is (for most speakers) a sentence-initial for -phrase separated by comma-intonation, or other sentence-level expressions introducing perspective overtly: How would one explain this fact under the classic contextualist analysis of subjectivity that I sketched above? If judge-dependence by itself implies the presence of an extra ‘judge’ argument, why the difference between judge-dependent PPTs and judge-dependent POS-DAs with respect to linguistic evidence for an extra argument?

  • (39) a.  The ride was fun for John.

  •     b. *The Dom Tower is tall for John.

  • (40) a. For John, the Dom Tower is tall (though I disagree).

  •     b. For someone like me, this bag is heavy.

  •     c. In John's opinion / According to John, the Dom Tower is tall.

One option would be to say that incompatibility of English POS-DAs is somehow a matter of lexical idiosyncrasy of tall , wide or heavy in the same way as fun or tasty take an idiosyncratic PP judge argument. However, the difference in the way PPTs and POS-DAs treat their judges is found across languages systematically. Here are parallel data from Russian. In Russian, PPTs take a Dative argument corresponding to the ‘judge’ (41-a). But these Dative judge DPs are not compatible with POS-DAs (41-b), and one of the alternative ways to specify the opinion-holder must be used (41-c): Even in languages that can have Dative DPs both with PPTs and with POS-DAs, quite often a closer look reveals a crucial difference in the status of these phrases. One example is Hungarian, where Dative DPs are acceptable both with important and with tall (J. Bacskai-Atkari and G. Rákosi, personal communication, see Rákosi (2006) for a study of Dative experiencers in Hungarian): However, Dative DPs in these two sentences have different properties, suggesting that in the case of important we are dealing with an argument (or, as Rákosi argues, ‘thematic adjunct’, which I have used as basically synonymous to ‘argument’ for my purposes), while in the latter case with tall , the Dative DP is not a thematic element but an adjunct not introduced by the predicate lexically. One of the tests showing this is anaphor binding, possible with interesting and ungrammatical with tall (for more tests see Rákosi 2006 ): I conclude that there is a systematic contrast between PPTs and POS-DAs—in particular, POS-DAs in the languages that I have been discussing so far do not project an extra argument the way PPTs do. 6

  • (41) a. Mne  etot fil'm byl neinteresen    R ussian

  •      I. dat this film  pst not.interesting

  •      ‘This film wasn't interesting for me’

  •     b. *Mne Ejfeleva bashn'a vysokaya

  •      I. dat Eiffel   tower tall

  •      ‘For me, the Eiffel Tower is tall’

  •     c. Dl'a men'a / Po-moemu,   Ejfeleva bashn'a vysokaya

  •      For  me    / In.my.opinion Eiffel   tower   tall

  •      ‘For me, the Eiffel Tower is tall’

  • (42) a. János-nak fontos-ak     vagyunk    H ungarian

  •      John- dat important- pl be.1 pl

  •      ≈ ‘John finds us important’

  •     b. János-nak minden felnőtt magas

  •      John- dat every   adult tall

  •      ‘For John, all grown-ups are tall’

  • (43) a. Egymás-nak   fontos-ak     vagyunk    H ungarian

  •      each.other- dat important- pl be.1 pl

  •      ‘We are important to each other’

  •     b. ??Egymás-nak minden felnőtt magas

  •      each.other- dat every   adult   tall

  •      ≈‘All grown-ups find each other tall’

What can we say about the linguistic representation of subjectivity of POS-DAs? Intuitively, the judge-dependence of POS-DAs is a function of the particular way in which the standard of comparison is fixed ( Kennedy 2013 ; Barker 2002 , 2013 ; Fleisher 2013 ). Deciding what the cut-off point for a gradable predicate is depends on a plethora of parameters, each of which could potentially introduce judge-dependence. The potential candidates for the reason why POS-DAs are subjective are the following: either different judges have different comparison classes in mind, or the standard-calculating function norm differs for different judges, or different judges find different intervals between the standard and the subject significant enough for categorisation purposes ( Fara 2000 ; Richard 2004 ; Bouchard 2012 ; Bylinina 2014b , a.m.o.). The contrast between PPTs and POS-DAs suggests that, whatever reason for POS-DA subjectivity, it should be analysed in a way that reflects this contrast. Let us now turn to the theoretical implications of the systematic difference between POS-DAs and PPTs.

4.2 Theoretical consequences

Above I suggested a systematic contrast between PPT judges and POS-DA judges: PPTs project an extra thematic role, whereas POS-DAs in the languages I considered do not project a corresponding thematic role. If this contrast is indeed a systematic one, as the data above show, it does need a systematic explanation.

One explanation that has been suggested in the literature (though not based of the same set of data) is that there are ‘two types of subjectivity’ in natural language ( Kennedy 2013 , 2016 ; Fleisher 2013 ). This idea can take different forms—for example, ( Kennedy 2016 ) implements this idea as a type–theoretic difference between PPTs and POS-DAs. Kennedy (2013) and Fleisher (2013) do not build their analyses on a rigid type–theoretic difference between the two classes of subjective items, maintaining a weaker version of the ‘two types of subjectivity’ view.

Let us consider the extreme version of the ‘two types of subjectivity’ approach for the sake of illustration. The core of the idea is that language might in principle involve different devices for encoding different types of subjectivity linguistically, and we might find more than one way actually attested. For example, one might indeed build this difference into the semantic type of the subjective item, which would mean that both the relativist and the contextualist theories of judge-dependence in language might be correct, but each of them would have its own scope. Say, postulating a judge argument would be a natural move for tasty , but not for POS- tall . Thus, in the former but not in the latter case, subjectivity would have type–theoretic consequences. Summing up this take on the ‘two types of subjectivity’ view: As a core motivation for the ‘two kinds of subjectivity’ view, Kennedy (2013) and Fleisher (2013) discuss embedding under find and report a slight contrast between PPTs under find (perfectly acceptable) and POS-DAs (somewhat degraded): As already mentioned in footnote 1, not all speakers of English have this contrast (see e.g. Bouchard (2012) for different judgments). My own informal survey also gave a split in judgments among English speakers. I do not have an explanation for this split, nor do I have an account of the subcategorisation properties of find . One point worth noting here is that, contra some recent suggestions in the literature ( Fleisher, 2013 ), find does not subcategorise for predicates of scalar variation (i.e. scalar predicates with non-objective ordering of objects on their scale). First, adding very to dimensional adjectives improves the sentences for all speakers I have consulted, although very big / large / small are still predicates with objective ordering on their scale—the cut-off point for, say, very tall might be different for different speakers and in different contexts, but it still holds that if very tall is true for a certain individual, it has to be true for any individual (objectively) taller than that one as well. In this very tall is similar to plain tall . It is not clear so far what constitutes the relevant difference between tall and very tall such that find is sensitive to this difference. No existing theory, to the best of my knowledge, can account for this contrast.

  • PPTs are judge-dependent in a contextualist way:

  • (44) 〚tasty〛 c ; w , t ,( j )  = [λ xe . [λ ye . y tastes good to x in w at t ]]

  • POS is judge-dependent in a relativist way (I use a j -superscript to indicate the judge-dependence of the standard in the positive construction):

  • (45) 〚POS subjectivec ; w , t , j  = λ Gd , et λ xe . maxd . G ( d )( x )) ⪰  dSTj

  • (46) a.   Anna finds her bowl of pasta tasty/delicious/disgusting.

  •     b. ??Anna finds her bowl of pasta big/large/small.

The second data point is basically similar. As I will show in section 6 , the so-called ‘extreme adjectives’ ( gigantic , enormous , etc.) are also perfectly fine in the complement of find , although some of them correspond to objective scales such as scale of size, weight, etc.

To sum up, I agree with Kennedy (2013) and Fleisher (2013) that there is a systematic linguistic difference between POS-DAs and PPTs, but I do so on different data—the distribution of ‘judge PPs’ rather than subcategorisation properties of find —as I think the requirements of find have to be studied more to make a solid claim about what is going on there.

What is the relevant semantic difference between predicates like fun or tasty and predicates like POS-tall or POS-wide that results in different argument frames?

One possible explanation is that the relevant difference is between predicates encoding subjective v. objective measure functions, which is roughly equivalent to the distinction between predicates of scalar variation v. other gradable predicates. This explanation can be found in Kennedy (2013) and Fleisher (2013) . Indeed, the scale of height is a totally objective one and objects are ordered on this scale independently of any observer or opinion-holder. To the contrary, the order of objects on the scale of taste would be different depending on who is making the judgement about the taste of the object. This would motivate the presence of an extra argument in tasty , but not in POS-tall .

I propose an alternative explanation. I suggest that there is another difference between PPTs and POS-DAs, and that this other difference is responsible for the observed different pattern of argument projection. As discussed in some detail in section 3 above, PPTs show a ‘direct experience requirement’—a statement like This cake is tasty can be made only if the judge has had direct perceptual experience involving the subject (i.e. has tasted the cake). For me this means that PPTs lexically include a reference to an experience event as part of their semantics, while POS-DAs do not. As experience events involve an experiencer, an extra argument gets projected in the case of PPTs but not POS-DAs.

In other words, the subjectivity of PPTs has its source in ‘experience’ they semantically refer to. POS-DAs are subjective for an altogether different reason, that is because different judges calculate the standard of comparison in different ways (for a more detailed discussion, see Bylinina 2014b ). Once again, I leave out the discussion of the precise relation between judge-dependence and standard calculation. This is because I want to focus on my main claim: while the intrinsic reference to a non-externalisable experience that PPTs have implies the presence of an experiencer argument, there is no such implication to the judge-dependence of standard calculation.

As it stands, we have two potential theories of the observations made so far: (i) subjective predicates that project an extra argument are predicates of scalar variation (and thus the extra argument that they project is the ‘ordering judge’); (ii) such predicates have an experiencer semantics (and thus the extra argument is the experiencer). Dimensional adjectives fail to provide data that decide between these theories. The deciding evidence would come from predicates of scalar variation that do not refer to an experience event. The class of predicates with these properties is, I argue, the ‘evaluative adjectives’ as described in Bierwisch (1989) . I look at these predicates in section 5 , after a preliminary summary, to which I turn in section 5 .

4.3 Summary of the section and an outlook

We have observed in this section that subjectivity does not entail the projection of an extra argument. While both PPTs and POS-DAs are subjective, I suggest that they differ in two correlating ways: POS-DAs do not refer to experience and do not project an extra argument (in Russian, English and Hungarian); PPTs, on the contrary, refer to experience and do project an extra argument. Thus the presence or absence of this extra argument does not come with the subjectivity per se , but rather is an orthogonal issue (to an extent).

However, the reason for the observable differences between PPTs and POS-DAs could not be settled looking at these two classes of subjective predicates alone. One could argue that PPTs project an extra argument because they are predicates or scalar variation, while POS-DAs are not. I propose instead that the reason has to do with experiential semantics of PPTs that POS-DAs lack. In the rest of this article, I argue for this view.

To argue for this point, in the next sections I look at more classes of adjectives and degree constructions. One obvious class of predicates to include in the discussion is the class of evaluative adjectives like smart or lazy . This is a class of predicates that, like tasty , are predicates of scalar variation, but do not refer to an experience event. Then I consider a class of extreme adjectives like gigantic for further support. Then I talk about modal degree constructions (using the too -construction as an example) that sometimes appear with for -phrases that superficially look like judge/experiencer PPs. I argue that these for -phrases have to be given a different analysis. Finally, I elaborate on the relation between the experiencer and the judge.

I now move on to evaluative adjectives.

5 Evaluative Adjectives

Evaluative adjectives are a class of gradable predicates described in Bierwisch (1989) . They are associated with scales that are not objective physical measures (unlike dimensional adjectives), and in this they resemble PPTs. Indeed, these classes of predicates are similar in many respects, as we will see shortly. Here are some evaluative adjectives found in Bierwisch's paper:

charming, lazy, ugly, stupid, smart, brave, fearful, timid, clever, intelligent, brilliant, idiotic, foolish, pretty, beautiful, gorgeous, handsome, hideous

Evaluative adjectives are no doubt subjective. Again, the usual tests are faultless disagreement and embedding under find , which evaluative adjectives pass both in positive and comparative form: Subjectivity in comparative form signals that evaluative adjectives are predicates of scalar variation, just like classic PPTs like tasty . Do PPTs and evaluative adjectives form one group with respect to the semantics of their subjectivity, the way it is linguistically encoded, and the way it is syntactically manifested? On the face of it, the answer to this question is positive, but the data that follow will show that it is not the case. Let us turn to the data.

  • (47) a.  A: Mary is pretty!     subjective disagreement

  •       B: No, she's not!

  •     b. John finds Mary pretty / smart / lazy.

  • (48) a.  A: Mary is prettier than Anna!     subjective disagreement

  •       B: No, she's not!

  •     b. John finds Mary prettier / smarter / lazier than Anna.

5.1 The data

First, quite like POS-DAs and unlike PPTs, evaluative adjectives in English do not take overt judge/experiencer arguments. The only way to express a judge with evaluative adjectives are sentence-level for - or to -phrases, or other perspective-setting expressions, separated from the rest of the sentence by comma intonation: In Russian, the dative argument is not licensed with evaluative adjectives either: As a further illustration, recall the contrast between Dative ‘judge’ DPs with PPTs and POS-DAs in Hungarian: the former are thematic and thus if they contain an anaphor, it can be bound by the subject, while POS-DAs do not project a thematic Dative DP and thus anaphors inside it are out. The same holds for evaluative adjectives: they pattern with POS-DAs rather than with PPTs and thus do not project an extra argument ( Rákosi 2006 ): Turning to Japanese data, recall the two ways a judge can get expressed: it can be either a topicalised Dative DP ( John-ni-wa ‘John- dat - top ’) or a ni-totte-(wa) phrase (Dative case marker plus totte morpheme with unknown etymology plus an optional topic marker). So far—with PPTs like tasty and with POS-DAs—the two ways to express a judge behaved in the same way, being basically interchangeable. Interestingly, and unlike with PPTs/POS-DAs, a ni-wa judge phrase is not possible with evaluative adjectives in Japanese at all, no matter if the judge is first, second or third person: The only way to express a judge is the ni-totte-(wa) phrase, which can be used with all other judge-dependent expressions as well; in this case, an evidential is not necessary even with a third-person judge: This pattern is quite surprising under the assumption that the presence of an extra argument follows from subjectivity—the unavailability of ni-wa as a way to express the judge would remain unexplained. Evaluative adjectives are the first point where the distributions of ni-wa and ni-totte(-wa) phrases differ.

  • (49) a. ??Mary is pretty / smart / lazy for / to John.

  •     b.  In John's opinion / For John / (?)To John, Mary is pretty / smart / lazy.

  • (50) *Mne / (?)Dl'a men'a / Po-moemu    Vas'a umnyj

  •    I. dat / for     me    / in.my.opinion Vasya smart

  •    ‘For me, Vasya is smart’

  • (51) a. János-nak szép-ek      vagyunk

  •      John- dat beautiful- pl be.1 pl

  •      ‘For John, we are beautiful’

  •     b. ??Egymás-nak  szép-ek      vagyunk

  •      each.other- dat beautiful- pl be.1 pl

  •      ‘≈ We find each other beautiful’

  • (52) a. *watasi-ni-wa / *anata-ni-wa  / *John-ni-wa   Mary-wa  namakemono

  •      I- dat - top     / you- dat - top / John- dat - top Mary- top lazy

  •      da / darou

  •      is / evid

  •      ‘≈ In John's opinion, Mary is / must be lazy’

  •     b. *watasi-ni-wa / *anata-ni-wa / ?*John-ni-wa  Mary-wa kasikoi (darou)

  •      I- dat - top     / you- dat - top / John- dat - top Mary- top    smart evid

  •      ‘≈ In John's opinion, Mary is / must be smart’

  • (53) a. John-ni-totte(-wa)      Mary-wa  namakemono da  / darou

  •      John- dat-totte-(top) Mary- top lazy        cop / evid

  •      ‘≈ In John's opinion, Mary is / must be lazy’

  •     b. John-ni-totte(-wa)      Mary-wa  kasikoi (darou)

  •      John- dat-totte-(top) Mary- top smart   evid

  •      ‘≈ In John's opinion, Mary is / must be smart’

5.2 Theoretical consequences

The data in section 5.1 show a systematic contrast between PPTs and evaluative adjectives in all the languages we have considered so far. How does this fact fit the picture we have had so far? Recall from the previous section that we have already encountered the contrast between subjective predicates that project a ‘judge’ argument and the ones that do not. The issue that could not be settled with the help of the data we had by the end of the last section is the exact factor that is responsible for a subjective predicate projecting or not projecting an extra argument. We had two hypotheses: one was that predicates of scalar variation project an extra argument (the ‘ordering’ judge), while other subjective predicates do not; my alternative hypothesis was that the relevant factor was reference to an experience event. Evaluative adjectives discussed in the current section are a crucial case to make the choice between these two factors, as they are non-experience predicates of scalar variation.

I suggest that subjectivity in evaluative adjectives is, most prominently, related to the multidimensionality of such predicates (see also Sassoon 2012 ). There are different parameters or dimensions of, for example, being smart (maths skills, good memory etc.), and the impact of these dimensions in the overall evaluation can vary from judge to judge. This results in a subjective ordering along the overall scale associated with an evaluative adjective. Thus, predicates like smart or lazy are predicates of scalar variation, but they do not refer to internalized experience as part of their semantics (unlike tasty ) (see Bylinina (2014b) for a more detailed discussion).

The fact that evaluative adjectives still do not license overt judge/experiencer phrases falsifies the theory of scalar variation as the source of the judge-argument licensing. The idea that the relevant licensing condition is reference to an experience event thus gets further support.

A few words need to be said about my take on the Japanese data presented in this section and the difference between two ways of introducing the judge that Japanese shows. I suggest that Japanese ni-wa phrases are available in the environments where an argumental DP is expected, while ni-totte(-wa) phrases are available in all judge-dependent environments across the board. Thus I believe that ni-wa phrases should be seen as an analogue of English thematic for/to phrases, Hungarian thematic Dative DPs and Russian Dative DPs, while ni-totte(-wa) phrases are more like sentence-level expressions introducing a perspective-holder—expressions like in John's opinion or sentence-initial for -phrases with comma intonation. I will provide an explicit semantics for these expressions when I formulate my analysis. I conclude that Japanese patterns with English, Russian and Hungarian in not introducing an extra argument is associated with judge-dependence of evaluative adjectives.

To sum up the contribution of this section, crucially, the core of the subjectivity of evaluative adjectives has nothing to do with an experience of any sort. The contrast between PPTs and evaluative adjectives in their argument-licensing properties is naturally explained under this view, and again supports the claim that subjectivity per se or ‘subjective assessment’ does not have to come with an extra argument that we find with PPTs.

6 Extreme Adjectives

A further case to consider is the class of the so-called ‘extreme adjectives’ described in ( Morzycki 2012 ):

gigantic, excellent, huge, monstrous, hideous, microscopic, gorgeous, fantastic

These adjectives pick a very high value on a certain scale: excellent arguably uses the same scale as good , but refers to very high values on this scale. Some of the extreme adjectives are associated with the same objective scales as dimensional adjectives (height, weight etc.), but even those ones are clearly subjective, although the scale they are associated with refer to an objective, measurable scale: Extreme adjectives do not readily combine with comparative morphology, which makes it hard to check if their comparative forms are subjective (for possible reasons and ways to somewhat improve these combinations, see Morzycki 2012 ): The question to ask about these adjectives is—again—if their subjectivity entails the presence of an extra argument, and why.

  • (54) a.  A: The Dom Tower is gigantic!     subjective disagreement

  •        B: No, it's not!

  •     b. Mary finds John huge / gigantic.

  • (55) ??more gigantic / huge / microscopic …

6.1 The data

Unlike PPTs, in English these adjectives do not take judge PPs: In Japanese, extreme adjectives do not go well with ni-wa judges; preferably, they take a ni-totte(-wa) judge, without the first-person constraint (in this they are similar to Japanese evaluative adjectives, in contrast with Japanese PPTs): 7 Russian patterns with Japanese and English in not allowing Dative DPs as arguments of extreme adjectives: Hungarian behaves in a parallel way to English, Russian and Japanese in not introducing thematic judge for extreme adjectives. The Dative DP found with predicates like gigantic can be shown to be a non-thematic adjunct, for example, according to anaphoric binding test that I borrow from Rákosi (2006) (see also the section on POS-DAs): Thus all the four languages we use for our limited data about cross-linguistic behaviour of ‘judge arguments’ pattern in not having an extra argument associated with judge-dependence of extreme adjectives.

  • (56) ??John is gigantic / huge for / to Mary.

  • (57) a. ?John-ni-wa   Mary-wa gokuhin darou

  •      John- dat - top Mary- top destitute evid

  •      ‘Mary must be extremely poor for John’

  •     b. John-ni-totte(-wa)      Mary-wa gokuhin da

  •      John- dat-totte(-top) Mary- top destitute cop

  •      ‘Mary is extremely poor for John’

  • (58)  *Mne  / (?)Dlja menja / Po-moemu     Ejfeleva bashn'a ogromnaja.

  •    I.- dat  / for     me    / in.my.opinion Eiffel    tower   huge

  •    ‘For me, the Eiffel tower is huge’

  • (59) a. János-nak minden felnőtt hatalmas

  •      John- dat every   adult   gigantic

  •      ‘For John, all grown-ups are gigantic’

  •     b. ??Egymás-nak ez   a   két  baseball játékos hatalmas

  •      each.other- dat this the two baseball player  gigantic

  •      ≈‘These two baseball players find each other gigantic’

6.2 Theoretical consequences

Let us concentrate on the semantics of extreme adjectives and the source of its subjectivity.

At least some of the extreme adjectives are not predicates of scalar variation, as they use objective dimensional scales ( size , height etc.). According to Morzycki's (2012) analysis, they involve a requirement of having gone ‘off the scale’ of contextually provided degrees Cgigantic means something like ‘so big that this size is not in the size range one normally talks about’. So the denotation of extreme adjectives involves exceeding the greatest degree in contextual degree interval C : This account is built on an idea of contextual domain restrictions discussed in Westerståhl (1984) , Schwarz (2009) and von Fintel (1994) , illustrated here with a quantifier every restricted to the objects that we own: Domain restrictions can be thought of as mechanisms similar to comparison classes—a salient set of objects relevant for evaluating the truth of a sentence, the former playing a role in a range of quantificational constructions, while the latter being specific to degree constructions of a certain kind. Domain restrictions / comparison classes as a point where subjectivity enters the picture have been discussed in Bouchard (2012) for positive dimensional adjectives (see also section 4 and Bylinina 2014b ). Typically, in a positive construction with dimensional adjectives like John is tall , John's height is evaluated with respect to the height of men in general ( Bale 2008 , 2011 ). The speaker's perspective comes into play together with the background of the speaker making the statement and the context of the conversation the interlocutors are in: a salient set of individuals can be the men that the speaker has known in his/her life, the average height of men in the region of the world he/she grew up in, the men that are now in the room etc. Thus the information about the relevant ‘reference set’ is dependent on who is making the assertion.

  • (60) 〚gigantic〛 = λ x . maxd . big ( d )( x )) ≻  max ( C )

  • (61) a. 〚every〛 = λ P λ Q λ R . P  ∩  Q  ⊆  R

  •     b. Every-C cat left the room.

  •      ≈ Every cat that we own left the room.

I assume that a similar mechanism is at work in the case of extreme adjectives as well: The absence of an extra argument associated with the source of variation of the contextual interval C between judges is natural under assumption that this extra argument appears only when an experience event is involved—which is not the case for the domain restriction as the source of subjectivity of extreme adjectives. Therefore, the extreme adjectives data provide further support for the connection between the availability of ‘judge PPs’ and experiential semantics of the predicate. Let me summarise.

  • (62) 〚gigantic〛 c ; w , t , j  = λ x . maxd . big ( d )( x )) ≻  max ( Cj )

7 Intermediate Summary

The question why I started my investigation of subjectivity in degree constructions with was the following: what is the correct way to linguistically capture judge-dependency? And, as a sub-question: is there only one kind of subjectivity out there, or do we need different analyses for different cases? I observed throughout this article that there might be many sources and reasons for subjectivity of different lexical classes and different constructions, but there is not much variation in how this subjectivity is linguistically encoded. However, one parameter that we have seen seems to correlate with the presence or absence of an extra argument—it is the reference to an experience event (and thus the presence of an experiencer argument) as part of PPT semantics, but not anywhere else. See the summary of the cross-linguistic data on different classes of subjective predicates in Table 1 .

Table 1

Cross-linguistic properties of subjective predicates

 PPTs POS-DAs EVAL Extreme 
English ‘judge’ PPs ✓ × × × 
Japanese 1P restriction ✓ × × 
Hungarian anaphors in Dative DPs ✓ × × × 
Russian Dative ‘judges’ ✓ × × × 
 PPTs POS-DAs EVAL Extreme 
English ‘judge’ PPs ✓ × × × 
Japanese 1P restriction ✓ × × 
Hungarian anaphors in Dative DPs ✓ × × × 
Russian Dative ‘judges’ ✓ × × × 

Do these data mean that we need ‘two types of subjectivity’, or can we do with only one? In what follows, I suggest an analysis of PPTs that would answer this question. But before doing so, I will quickly address one apparent counterexample to the generalisation described here—the too -construction that seem to host a ‘judge PP’ in the absence of a reference to experience.

8 An Apparent Counterexample: The Too -Construction

In section 4.1 , footnote 3, the judge-dependence of the too -construction was discussed a bit in passing. This construction can include for -phrases that one could take to be ‘judge PPs’: In this section, I point out that the too -construction indeed shows subjectivity, but does not have a judge argument. 8

  • (63)  This river is a bit too deep for John .

The too -construction has been extensively argued to have a modal semantics. The modal component is clear in a paraphrase suggested in Meier (2003) and von Stechow et al . (2004) : The type of modality involved in the too -construction is sometimes called ‘normative’ ( Saebø 2009 ). It is root modality with a ‘normative’ ordering source (see Kratzer 1981 ). A normative ordering source can just be preferences of the judge in the world of evaluation, and the ‘purpose’ or a ‘goal’ can be taken to restrict the modal base ( Saebø 2009 ), though there might be other ways to implement the intuition that something is better or worse from someone's point of view with respect to a particular purpose. One example of a modal verb which is clearly lexically ‘normative’ is English ought .

  • (64) The book is longer than it may/can be (given a particular goal or purpose).

The too -construction is subjective, according to the familiar tests (it is easy to construct a faultless disagreement dialogue): I propose that it is the modal component that makes the too -construction subjective.

  • (65) John finds this book too long (for me) to read in 1 day.

‘Normative’ modals are independently known to be judge-dependent ( Saebø 2009 ). Saebø (2009) introduces the judge as an argument of a normative modal, say, ought ( f for modal base, g for ordering source; z for judge)—in exactly the same way as a contextualist would do for PPTs: The question is if this move of adding a judge argument to the denotation of a modal justified linguistically for the too -construction, which hosts overt for -phrases. Do these for -phrases have the same argumental status as the ‘judge PPs’ with PPTs? If the answer is positive, it will be a counterargument against my generalisation that ‘judge PPs’ are only licensed with experiencer predicates—the too -construction does not describe experience in any reasonable sense, therefore, I do not predict it to host ‘judge’-phrases. However, I believe that the answer to this question is negative.

  • (66) 〚ought〛 f , g  =  λφλz.Ovf,g(z)(φ)

An easy way to see this has to do with animacy. Judges have to be animate or sentient, while the for -phrase in the too -construction does not have to be: Note that the sentence-initial for -phrase, which gets unambiguously interpreted as an opinion-holder, cannot be inanimate: Compare this to a grammatical example with an animate for -complement: The same holds for other unambiguously perspective-related expressions ( in X's opinion ; subject of find etc.).

  • (67) This hotel is too expensive for this conference to take place in.

  • (68) ??For this conference, this hotel is too expensive to take place in.

  • (69)  For John, this hotel is too expensive to stay at.

A detailed study of the semantics of for -phrases in the too -construction is beyond the focus of the current discussion (see Bylinina 2014a , b : 123–125, 139–141, in particular, on the of the infinitival clause in (69)). What is important for the current discussion though is that it provides one more case where a subjective element turned out not to project a judge/experiencer argument, despite appearances. This allows us to keep the generalisation about the factors affecting the licensing of ‘judge’ for/to -phrases or the lack thereof.

9 Taking Stock: Experiencer, Judge, and How They are Related

9.1 The ‘judge=experiencer’ requirement

The claim I make in this article is that the presence or absence of an experiencer argument is orthogonal to judge-dependence, and that the experiencer argument of PPTs is not the way PPTs linguistically encode their judge-dependence.

The experiential status of the ‘judge PPs’ that appear with PPTs in English is supported by a number of considerations discussed in previous sections: first, syntactically they pattern with Dative experiencers according to all known diagnostics ( Lasersohn 2005 ; Rákosi 2006 ); secondly, they pattern with experiencers semantically, in particular, in the ‘direct experience requirement’ (the most easily observable form of this requirement is the first-person constraint in Japanese).

The difference between experientiality and subjectivity can be further clarified with the help of the following parallel. Take a random transitive gradable adjective—say, proud . Proud has two individual arguments: the first one, introduced by preposition of ( proud of X ); and the subject ( Y is proud of X ). Proud is also gradable—it participates in comparative and other degree constructions: The order of individuals denoted by the subject of proud on the proud of X scale depends on the first argument X . Say, proud of John and proud of Mary could give two totally different orderings of people's level of ‘pride of John’ and ‘pride of Mary’, respectively. Let us assume there are three individuals that are potential subjects of the predicate proud of Xa , b , and c . Then the measure function that is part of the semantics of proud of John could give the following ordering: 〈 a , c , b 〉, and proud of Mary would give the ordering 〈 b , c , a 〉.

  • (70) a. John is more proud of his students than he should be.

  •     b. John is too proud of his students.

Basically, this is the same as saying that proud is a predicate of scalar variation dependent on its first argument—this is exactly what the literature says about PPTs like tasty and fun and their ‘judge PP’ ( Anand 2009 ; Lasersohn 2009 ). However, this does not automatically entail that this argument is a judge argument—in fact, to my knowledge, a suggestion like that has never been made for the first argument of proud .

A similar observation is made in Anand (2009) . He observes the ‘troubling issue’ that the definition of scalar variation he uses (following Lasersohn 2009 ) overgenerates, predicting that any transitive scalar predicate should trigger subjectivity effects, which is not what we find with transitive adjectives in general. 9

What ruins—or maybe complicates—the parallel between proud and tasty is a very intuitive difference between the roles of their internal argument. Unlike proud , tasty seems to express an opinion, or a judgement, by the individual denoted by the internal argument. And in fact, in a sentence like This cake is tasty to John the position of the cake on the taste scale is decided according to John's perspective—that is, John's judgement of the cake's taste is relevant for the truth of the whole statement. In this sense, the experiencer has a certain connection to the judge (however it is implemented) which the first argument of proud lacks.

A further illustration of the special relation between the judge and the experiencer argument of PPTs comes from the interaction of PPT experiencers and elements that—at least according to the existing analyses—shift the judge index as part of their lexical semantics. For example, given that find shifts the judge to the individual denoted by its subject ( Saebø 2009 ; Bouchard 2012 ), it comes as a surprise that the subject of find restricts the experiencer of, say, tasty , if the judge index and the experiencer argument are totally independent: In a similar way, assuming a judge-shifting analysis for (Japanese) evidentials ( McCready 2007 ), the view that the judge and the experiencer of PPTs are not related in any way is not compatible with the fact that the presence of non-first-person PPT experiencer requires an evidential—that is, requires the judge to be shifted: These facts strongly suggest that the analysis of PPTs setting their experiential argument and their judge-dependence apart has to be supplemented with a mechanism regulating how the experiencer argument of PPTs and the judge index are related. The first take on this kind of mechanism would be something like the following: However, in the form stated in (73), the requirement is too open-ended. It is not clear if it should be treated as a linguistic universal rooted in facts about human cognition, or it is a lexical requirement of some sort—a presupposition, or maybe a meaning postulate—relevant for a certain class of predicates.

  • (71) a.  〚find〛 c ; w , t , j  = λϕλ x . ϕ c ; w , t , x     ( Saebø 2009 )

  •     b. *John finds this cake tasty to Mary.

  • (72) a. 〚 evid 〛(ϕ c ; w , t , j ) =  φc;w,t,j     ( McCready, 2007 )

  •     b. John-ni-wa     kono keeki-wa oishii *(noda)

  •      John- dat - top this   cake- top tasty  evid

  •      ‘This cake is tasty to me / to John’

  • (73) J udge =E xperiencer R equirement

  •    A statement about someone's internal state can be made only if the judge parameter is set to the same value as the experiencer of this internal state.

I will formulate it as a presupposition—part of the felicity conditions imposed by PPTs, with a potential deeper explanation from human cognition. In the following subsection, I present an explicit proposal built on this decision, and after that I discuss some issues concerning it.

9.2 Implementation

Let us make things more precise. To start, I need a semantics for PPTs that would explicitly refer to an experience event. Only one of the existing analyses for PPTs has such semantics: Anand (2009) proposes a semantics for PPTs that explicitly refers to a ‘perceptual’ situation. As I do not share the semantic division of labour between the for -phrase and the PPT itself that Anand proposes, I will not adopt these details and will modify Anand's semantics. Importantly, the ‘direct experience’ requirement is made explicit in Anand's entry for tasty : it states that the experiencer had a perceptual experience of the subject of tasty and that this gave rise to a percept on the tasty scale greater than some degree d . The degree relation tasty holds between a tasting situation and a degree of tastiness experienced in this situation. I use a neo-Davidsonian semantics as it is very transparent (see Kratzer 1996 ; Landman 2000 ). The desired result would be something like the following, first take (the ‘judge=experiencer’ requirement phrased as a presupposition; I also include the assignment function g as a parameter of the interpretation function 〚〛, as it will be relevant for some of the cases discussed below): (74) says that the semantics of tasty amounts to the following: s is a situation such that it is a tasting situation, z is an experiencer in this situation, x is a stimulus, and the degree of the stimulus on the taste scale exceeds a contextual standard, according to the judge j (at time t and in world w ). The felicity condition states that j and z —the judge and the experiencer—refer to the same individual.

  • (74) 〚tasty〛 c ; g ; w , t , j ( ze )( xe )( ss )    P reliminaryversion

  •     a. is defined iff j  =  z ;

  •     b. whenever defined, denotes 1 iff [ taste ( s ) & Experiencer( s , z ) &

  •          Stimulus( s , x ) & maxd . taste ( d )( s )) ≻  dstj for j at t in w ]

Thus, a simple sentence This cake is tasty would require that the speaker has tried the cake because the judge parameter is set to the speaker by default, and the judge=experiencer requirement would force the covert experiencer to take the same value—I assume that this position is occupied by a silent pronoun ( Sp in the index of evaluation indicates that the judge is set to the speaker; I assume that the situation variable gets existentially closed): The semantics in (75) explicitly states that the experiencer of tasting and the judge of the taste have to be one and the same individual—in the default case where no judge-shifts happens, it would be the speaker.

  • (75) 〚This cake is tasty pro8c ; g ; w , t , Sp

  •     a. is defined iff g (8) =  Sp ;

  •     b. whenever defined, denotes 1 iff ∃ s [ taste ( s ) & Experiencer( s , Sp ) &

  •      Stimulus( s , this.cake ) & maxd . taste ( d )( s )) ≻  dstSp for Sp at t in w ]

In the case the judge is shifted—for example, in the complement of find —the result will be as follows: In (76), pro and the ‘local’ judge, set to John, denote the same individual. (76) allows us to make it more precise why the sentences like (77) are deviant—the deviance is located in the felicity condition that is part of the semantics of (77): (77) is not semantically well formed because there is no way for Mary to denote the same individual the judge is set to—namely, John. This correctly predicts ill-formedness of such combinations of judge and experiencer.

  • (76) 〚John finds this cake tasty proc ; g ; w , t , Sp  = 〚This cake is tasty pro8c ; g ; w , t , John

  •     a. is defined iff g (8) =  John ;

  •     b. whenever defined, denotes 1 iff ∃ s [ taste ( s ) & Experiencer( s , John ) & Stimulus( s , this.cake ) & maxd . taste ( d )( s )) ≻  dstJohn for John at t in w ]

  • (77) 〚John finds this cake tasty to Mary〛 c ; g ; w , t , Sp  = 〚This cake is tasty to Mary〛 c ; g ; w , t , John

  •     a. is defined iff 〚Mary〛 c ; g ; w , t , John  =  John ;

  •     b. whenever defined, denotes 1 iff ∃ s [ taste ( s ) & Experiencer( s , Mary ) & Stimulus ( s , this.cake ) & maxd . taste ( d )( s )) ≻  dstJohn for John at t in w ]

The impact of the sentence-initial ‘context-setting’ phrases introducing an opinion-holder (such as English for -phrases with comma intonation or Japanese ni-totte(-wa)dat - totte - top phrase) would be the same as of find : A similar semantics can be shown to work for Japanese sentences with evidential markers, which are judge-shifters, according to the analysis by McCready (2007) —and with overt ni-wadat - top experiencers: Let us now turn to cases that seem to be problematic for the picture I have outlined so far. The first problematic case would be English sentences with overt non-first-person experiencer and without any overt judge-shifter such as a ‘perspective’-phrase or find . What is expected in this case would be a clash between a (default) judge set to the speaker and the non-first-person experiencer: The semantics in (80) as it is now predicts such sentences to be ill-formed because the ‘judge=experiencer’ requirement is not met. There are several ways one could treat such examples. First, I want to point out that for many English speakers examples like (80) are somewhat marked. They are definitely not ungrammatical, and they make perfect sense with respect to the meaning that they convey, but there is something odd about them. This oddness, I believe, has to do precisely with the mismatch that my semantics makes explicit—these sentences are an attempt to make an external assertion about something one cannot make a direct external assertion about. But, as the overall acceptability of (80) shows, language has mechanisms to allow us to make such assertions after all. What are these mechanisms? In Japanese, this mechanism is the system of evidential marking, as we have seen above. English does not have grammatical evidentiality, so there is something else going on in English that allows for (80). One option is that the default setting of the judge parameter to the speaker is weaker in English than the requirement that the judge is the experiencer. Then, in a sentence like (80), the judge index would be set to John rather than to the speaker, because John is the experiencer in this sentence. This would make English and Japanese quite different in how the value of the judge index is set and shifted—apparently, Japanese cannot shift the judge in a similar sentence without overt evidential or modal morphology (the restriction known as 1P-restriction).

  • (78) 〚For John, this cake is tasty pro8c ; g ; w , t , Sp  = 〚This cake is tasty pro8c ; g ; w , t , John

  •     a. is defined iff g (8) =  John ;

  •     b. whenever defined, denotes 1 iff ∃ s [ taste ( s ) & Experiencer( s , John ) &

  •     Stimulus( s , this.cake ) & maxd . taste ( d )( s )) ≻  dstJohn for John at t in w ];

  • (79) 〚 evid This cake is tasty to John〛 c ; g ; w , t , Sp  = 〚This cake is tasty to John〛 c ; g ; w , t , John

  •     a. is defined iff 〚John〛 c ; g ; w , t , John  =  John ;

  •     b. whenever defined, denotes 1 iff ∃ s [ taste ( s ) & Experiencer( s , John ) &

  •     Stimulus( s , this.cake ) & maxd . taste ( d )( s )) ≻  dstJohn for John at t in w ]

  • (80) 〚This cake is tasty to John〛 c ; g ; w , t , Sp

  •     a. is defined iff 〚John〛 c ; g ; w , t , Sp  =  Sp ;

  •     b. whenever defined, denotes 1 iff ∃ s [ taste ( s ) & Experiencer( s , John )

  •     & Stimulus( s , this.cake ) & maxd . taste ( d )( s )) ≻  dstSp for Sp at t in w ]

Alternatively, one could assume that English sentence in (80) contains a silent evidential or a silent modal that can be inserted as a last resort to resolve the conflict between the judge and experiencer taking different values. The role of this silent element would be to mediate between the non-externaliseable situation and the external individual making a statement about it, specifying that it is an inference on the basis of external evidence. Quite like Japanese evidential markers, this silent element shifts the contextual judge, so that the local judge would be the overtly specified experiencer: The insertion of silent epistemic modals as a ‘last resort’ to satisfy the requirements associated with some elements in the sentence is attested outside of PPT constructions. In particular, Nouwen (2010) argues for precisely the same mechanism in sentences with modified numerals such as at least 10 that require a range of values as its input. When this range of values is not provided in a certain configuration, a silent epistemic modal is inserted, and the values for the range gets picked from different epistemically accessible worlds.

  • (81) 〚 epist/evid This cake is tasty to John〛 c ; g ; w , t , Sp  = 〚This cake is tasty to John〛 c ; g ; w , t , John

  •     a. is defined iff 〚John〛 c ; g ; w , t , John  =  John ;

  •     b. whenever defined, denotes 1 iff ∃ s [ taste ( s ) & Experiencer( s , John ) &

  •     Stimulus( s , this.cake ) & maxd . taste ( d )( s )) ≻  dstJohn for John at t in w ]

I cannot decide here between the two options outlined above for the way the correct interpretation of sentences like This cake is tasty to John is derived. 10

The question immediately arises why this silent modal, if it is real, is not available to save the derivation in (77), where there is a conflict between the judge in the subject of find and the experiencer in the embedded clause. I do not have a good answer to this question rather than blaming this ban on the lexical selectional properties of find itself. As a related matter, consider one more problematic property of sentences with find —namely, the ban on the experiencer phrase in the embedded clause even if it coincides with the subject of find : Given the formulation of the ‘judge=experiencer’ requirement I have used so far, there's nothing that would preclude such a sentence from being acceptable: However, this sentence is clearly deviant. The reason could be Condition C and the fact that the proper name John appears twice in this sentence. However, if the second occurrence of John is substituted by a pronoun, the sentence does not improve: * John finds this cake tasty to him . The reason for the degraded status of such find- sentences with overt experiences denoting the same individual as the subject of find could be due to redundancy, as they specify that John is the judge and the experiencer twice—once in the matrix sentence and once in the embedded clause. This could be thought of as an instance of a general ban on redundant sentences, or as a lexical property of the verb find , which in general is known to bear all sorts of contingency / non-triviality presuppositions (see Bouchard (2012) for a detailed discussion—and also Saebø (2009) and Moulton (2009) for the state-of-the-art discussion).

  • (82) *John finds this cake tasty to John.

  • (83) 〚John finds this cake tasty to John〛 c ; g ; w , t , Sp  = 〚This cake is tasty to John〛 c ; g ; w , t , John

  •     a. is defined iff 〚John〛 c ; g ; w , t , John  =  John ;

  •     b. whenever defined, denotes 1 iff ∃ s [ taste ( s ) & Experiencer( s , John ) & Stimulus ( s , this. cake ) & maxd . taste ( d )( s )) ≻  dstJohn for John at t in w ]

One more problematic case to mention is the ill-formedness of sentences where the `perspective'-phrase and the overt experiencer phrase are present in the sentence: (84-a) is, again, redundant, while (84-b) is contradictory—but I do not have an explanation why the silent existential modal is not available to save (84-b). I leave this issue for future study of the availability of such rescuing mechanisms.

  • (84) a. *For John, the ride was fun for John.

  •     b. *For John, the ride was fun for Mary.

So far I have ignored the fact that tasty in the examples I discussed is a gradable predicate in the positive form—and thus, presumably, it appears in combination with the silent positive morpheme POS, which is itself judge-dependent, under the view that I take here. The question might arise, what is the relation between the judge that comes from the lexical meaning of tasty and the one that comes from POS? Do they always have to be aligned, or can there be ‘mixed’ interpretations? An example due to Chris Kennedy (personal communication), on the face of it, suggests that these two judges do not have to be aligned: On the one hand, it is clear that the speaker is making a guess about the emotional experience of the dog, and in this sense, the ‘judge’ is set to the dog. On the other hand, there is a judge that is responsible for setting the cut-off point for the extension of this predicate. Do we want to be committed to the position that dogs can judge the extensions of vague predicates? There are two possible ways to proceed from here. One way would be assuming that the ‘mixed’ interpretations are possible (with one of the ‘judges’ set to the speaker and the other one set to the experiencer). The other option would be, indeed, committing to the position that when someone makes a statement like (85), they put themselves in the dog's shoes in a sense—they are making a guess about the quality and intensity of the dog's emotions, which might actually bring in certain ‘anthropomorphization’ of the dog's cognitive state. The analysis I formulate here supports the latter option, which would result in a sort of ‘shift-together’ effect for perspective—sensitive items (see Anand and Nevins (2004) for ‘shift-together’ in the domain of indexicality). The data on ‘shift-together’ and the lack of ‘mixed readings’ in the domain of perspective—sensitivity are not clear (for discussion see Bylinina et al . 2014 ), so I cannot give a decisive answer here.

  • (85) Playing fetch is fun for my dog.

Whatever the correct understanding of the example in (85) is, the semantics of tasty in the positive construction needs to be spelled out as a combination of the meaning of the gradable adjective tasty (that can also appear in other degree constructions such as comparative, superlative, etc.) and the POS morpheme that brings in the positive degree semantics.

In order to derive the meaning of sentences like (85) compositionally, I need to make several assumptions. Let us start with the lexical entry for gradable tasty . It should now have a degree argument in addition to the other arguments that were listed before (I do not have conclusive evidence concerning the relative order of the degree argument and the other arguments of PPTs, the choice I make here is just one of the possible ways of introducing the degree argument): Secondly, we will need a slightly modified semantics for the positive morpheme POS—this modified morpheme POS s will take a gradable predicate over situations (type 〈 d , st 〉) as its first argument, while the usual positive morpheme POS takes a gradable predicate over individuals (type 〈 d , et 〉) as its input. Compare the regular (87-a) and the modified version (87-b): Thirdly, I will make a common assumption that the situation argument is bound by existential closure at the clausal level, thus existential closure will have the following meaning: These ingredients combine in the following way giving us the LF that needs to be interpreted: To take care of the correct presupposition projection—that is, to make sure that the judge=experiencer requirement gets passed up the tree and still relates the experiencer to the local value of the judge parameter in sentences involving judge shifting, I will assume the following version of the Function Application (FA) rule (see Heim & Kratzer 1998 : 49): These building blocks will give us the correct semantics for (89). I will not spell out every step of the derivation here, but I will show the crucial point where for the result to be defined (= in the domain of the interpretation function), it has to be the case that the assignment function returns the Speaker for the index of the null pronoun. For a detailed step-by-step derivation of the semantics for (89), see Appendix. Let us now move to sentences that involve judge-shifting. We had a number of expressions that force judge-shifting at their level of attachment. I will give a semantics for a judge-shifting operator Opi that is a generalisation of the judge-shifting expressions I mentioned so far (evidential morphology, epistemic modals, perspective-shifting expressions like in John's opinion etc.). I assume that on top of the judge-shifting effect Opi has some other semantic contribution (indicated in (92) as ω) which I will ignore for the purposes of this article: 11 I assume Opi appears above existential closure: The derivation of (93) would proceed in the same way as before, except for the upper layer where the judge gets shifted. Note that the experiencer argument pro8 has to denote the same individual as the local (shifted) judge rather than the global judge, as desired. For a step-by-step derivation of (93) see, again, Appendix: To sum up, I believe that the ‘judge=experiencer’ requirement that I propose gives good results in covering the data that have constituted the core of the debates around the semantics of PPTs and judge-phrases.

  • (86) 〚tasty〛 c ; g ; w , t , j ( ze )( xe )( dd )( ss )

  •     a. is defined iff j  =  z ;

  •     b. whenever defined, denotes 1 iff [ taste ( s ) & Experiencer( s , z ) &

  •               Stimulus( s , x ) & taste ( s , d ) for j at t in w ]

  • (87) a. 〚POS〛 c ; g ; w , t , j  = λ Gdet λ xe . maxd . G ( d )( x )) ≻  dstj

  •     b. 〚POS sc ; g ; w , t , j  = λ Sdst λ ss . maxd . S ( d )( s )) ≻  dstj

  • (88) 〚∃ Cc ; g ; w , t , j  = λ Pst .∃ s [ P ( s )]

  • (89) [ ∃ C [POS s [the cake is [tasty pro8 ]]]]

  • (90) F unction A pplication

  •     Let α be a branching node with two daughters β and γ. For any context c, assignment g , for any world w , for any time t , and for any judge j , α  ∈  dom (〚〛 c ; g ; w , t , j ) iff

  •     a.  β  ∈  dom (〚〛 c ; g ; w , t , j ) and

  •     b.  γ  ∈  dom (〚〛 c ; g ; w , t , j ) and

  •     c. 〚γ〛 c ; g ; w , t , j  ∈  dom (〚β〛 c ; g ; w , t , j ).

  •       Whenever defined, 〚α〛 c ; g ; w , t , j  = 〚β〛 c ; g ; w , t , j (〚γ〛 c ; g ; w , t , j ).

  • (91)  C POS s the cake tasty pro8  ∈  dom (〚〛 c ; g , w , t , j )

  •     a. only if 〚 pro8c ; g , w , t , j  ∈  dom (〚tasty〛 c ; g , w , t , j ) ⇔ 〚 pro8c ; g , w , t , j  =  g (8) =  j .

  •     b. Whenever defined, 〚∃ C POS s the cake tasty pro8c ; g , w , t , j  = 

  •      〚∃ Cc ; g , w , t , j (〚POS sc ; g , w , t , j (〚tasty〛 c ; g , w , t , j (〚 pro8c ; g , w , t , j )(〚the cake〛 c ; g , w , t , j ))) = 1 iff ∃ s [ taste ( s ) & Experiencer( s , j ) & Stimulus( s , the.cake ) & maxd .  taste ( d )( s )) ≻ dstj for j at t in w ]

  • (92) For any c, g, w, t, j, [Op i A] is in dom (〚〛 c , g , w , t , j ) iff A is in dom (〚〛 c , g , w , t , g ( i ) ). If so, 〚[Op i A]〛 c , g , w , t , j  = 〚ωA〛 c , g , w , t , g ( i )

  • (93) [Op i [∃ C [POS s [the cake is [tasty pro8 ]]]]]

  • (94)  Op i POS s the cake is tasty pro8  ∈  dom (〚〛 c ; g , w , t , j )

  •     a. only if 〚 pro8c ; g , w , t , g ( i )  ∈  dom (〚tasty〛 c ; g , w , t , g ( i ) ) ⇔

  •      〚 pro8c ; g , w , t , g ( i )  =  g (8) =  g ( i )

  •     b. Whenever defined, 〚Op iC POS s the cake tasty pro8c ; g , w , t , j  = 1 iff

  •      ∃ s [ taste ( s ) & Experiencer( s , g ( i )) & Stimulus( s , the.cake ) & maxd . taste  ( d )( s )) ≻ dstg ( i ) for g ( i ) at t in w ]

Concerning the alleged presuppositional status of the ‘judge=experiencer’ requirement—on the one hand, the inference that the judge has been involved in an experience event survives in certain contexts where well-behaved presuppositions project as well ( The cake is not tasty still strongly suggests the speaker has tried the cake). On the other hand, the behaviour of this inference differs from the behaviour of known presuppositions as far as presuppositional plugs are concerned: there is no ‘judge=experience’ parallel to examples like I have a sister. My sister is 30 years old , where the content of the presupposition is explicitly introduced prior to the presupposition-inducing possessive DP. I believe that it has to do with the particular kind of content that the ‘judge=experiencer’ requirement has—in particular, this content is meta-linguistic in that it refers to language-internal facts rather than non-linguistic facts of the world (cf. Barker 2013 ), which might block its explication in a way the non-meta-lingusitic presuppositions are compatible with.

The remaining question is the scope of this requirement—that is, which experience predicates it applies to.

9.3 The scope of the ‘judge=experiencer’ requirement

By scope here I mean the class of experiencers and experiencer predicates that show the same kind of interaction with the judge. In particular, is it only experiencer arguments of PPTs that are related to the judge in the specified way, or is it experiencers in general?

The facts fall somewhat in between these two options. On the one hand, the phenomenon in question is clearly not restricted to PPTs—but on the other hand, it does not systematically hold for any argument that would be classified as an experiencer.

Experiencers of adjectives like lonely or cold are clearly distinct from PPT ‘judge-phrases’, but in Japanese they interact with the judge parameter quite like PPT ‘judge’—PPs: they obey the first-person restriction and cannot be used without an evidential with the second and third person: On the other hand, a number of other experiencer predicates do not show the same behaviour—say, okotteiru ‘angry’ or haradatashii ‘irritating’ (both with a watop experiencer) do not require an evidential in Japanese: It is not clear what is responsible for the difference between predicates in (95) and (96). The corresponding English predicates that are (rough) translations of Japanese predicates from (95) and (96)— lonely , cold , irritating , angry —do not really help clarify this point. Irritating in English patterns with PPTs with all known respects, such as optionality of the experiencer PP and embeddability under find : The other three predicates— lonely , cold , angry —have subject experiencers that cannot be omitted and thus show different behaviour. They are out in the small-clause complement of find : 12 However, there is a methodological point here that precludes me from drawing conclusions from (98). As an overt experiencer is obligatory with these predicates, the sentences in (98) do not constitute a minimal pair with sentences like I find this cake tasty , where the experiencer is not overtly present—rather, they are parallel to equally bad I find this cake tasty to John . Thus the degraded status of (98) can mean several things: (i) these predicates are just not subjective in English at all; (ii) these predicates are quite like PPTs semantically, but the obligatoriness of the overt experiencer does not allow us to show this with the find -test.

  • (95) a. watasi-wa / *anata-wa / *kare-wa sabisii desu

  •      I- top      / you- top     / he- top     lonely cop

  •      ‘I'm/You're/He's lonely’

  •     b. watasi-wa / *anata-wa / *kare-wa samui desu

  •      I- top        / you- top     / he- top     cold   cop

  •      ‘I'm/You're/He's cold’

  • (96) a. Mary-wa  John-ga   haradatashii

  •      Mary- top John- nom irritating

  •      ‘John is irritating to Mary’

  •     b. John-wa  Mary-ni   okotteiru

  •      John- top Mary- dat angry

  •      ‘John is angry/mad at Mary’

  • (97) a. John is irritating (to Mary).

  •     b. I find John irritating.

  • (98) a. ??I find John lonely.

  •     b. ??I find John angry at Mary.

  •     c. ??I find John cold.

A more careful look at syntactic and semantic landscape of experiencers and experiencer predicates is needed to find out what is at play in the English and the Japanese patterns discussed above—it might be the syntactic status of the experiencer (see Belletti and Rizzi (1988) and the discussion this work gave start to), or its semantic properties (see Reinhart's work on semantic classification of thematic roles in general and experiencer subtypes in particular based on binary features, and further developments— Reinhart 2002 ; Marelj 2004 ; Siloni et al . 2012 ; other relevant work is Pesetsky 1995 and Landau 2010 ), or the combination of syntax and semantics.

I leave further study on which experiencers or experiential predicates interact with the judge parameter and how this interaction is restricted for future work. What is clear so far, crucially for the point I want to make here, is that this interaction is most probably not a property of the PPT argument exclusively, and can be found for other kinds of experiencer arguments as well (with further semantic and/or syntactic constraints, probably subject to cross-linguistic variation). Thus the PPT data do not undermine the unification account of subjectivity effects found across degree constructions I have looked at. The extra argument of PPTs does not show any particular judge-related behaviour that is not observed for atleast some other classes of experiencers.

The conclusion thus is that the judge-dependent properties of PPTs should be seen as a combination of the following ingredients: (i) judge-dependence per se , (ii) experiential semantics that is responsible for the presence of the experiencer argument, and (iii) the relation between the judge and the experiencer that is not particular to PPTs.

10 Conclusion and Further Issues

I looked at different classes of subjective lexical items, limiting myself to the domain of degree constructions. I took predicates of personal taste like tasty or fun as a starting point, as they have received the most attention in the literature and are usually taken to be representative of the whole class of subjective predicates, but a closer look at more items like positive dimensional adjectives ( tall ), evaluative adjectives ( smart ), extreme adjectives ( gigantic ), and modal degree morphemes like too shows that there is much less evidence for an extra argument for subjective terms in general than have been argued for based on predicates of personal taste data. At the same time, I argue that postulating a judge argument for tasty or fun and capturing judge-dependence of the other classes with a judge index of evaluation—‘two types of subjectivity’ view—is not a good solution either. Taking an intuition that the extra argument of PPTs is an experiencer seriously, I observe that, indeed, the presence of an extra argument correlates with reference to an experience event as part of the predicate semantics, and, on top of that, I show that the PPT experiencer argument does not exhibit any special judge-dependent behaviour that is not observed for other experiencers in different constructions.

The task formulated in the beginning of the article was to clarify the role judge-dependence plays in the semantics of certain gradable predicates and degree constructions, how this judge-dependence is encoded linguistically, and what the semantic contribution of ‘judge PPs’ is. Now I can give my solution to the puzzle formulated as a starting point of this discussion: what accounts for the observable diversity of subjectivity ‘types’ that we found only looking at three adjectives— tasty , smart and tall :

  1. Tasty -class: subjective both in positive and comparative form, take judge PPs;

  2. Smart -class: subjective both in positive and comparative form, no judge PPs;

  3. Tall -class: subjective in positive but not in comparative form, no judge PPs.

I argue that the overt ‘judge PPs’ are experiencers licensed when the predicate refers to an experience event as part of its semantics, the ‘judge’ flavour being due to the interaction between experiencers in general and the judge. The lack or presence of subjectivity effects in degree constructions other than the positive construction depends on which element exactly is subjective—in the case of POS-tall , the source of subjectivity is the positive morpheme POS, while tall in the absence of POS in a comparative construction does not show subjectivity because tall itself lexicalises a perfectly objective measure that cannot be the source of subjectivity. On the other hand, smart is lexically subjective (it is a predicate of scalar variation), hence subjectivity effects in non-positive constructions.

This said, there are still a number of loose ends and problems remaining. The first issue that needs further work concerns the particular formulation of the dependency between the judge and the experiencer argument that I have observed in the previous section: As noted in the previous section, this formulation is probably too strong, and I leave its refinements for future work.

  • (99) J udge =E xperiencerrequirement

  •    A statement about someone's internal state can be made only if the judge parameter is set to the same value as the experiencer of this internal state.

Another concern was pointed out to me by an anonymous reviewer: how would the semantics I propose account for cases with quantificational experiencers, such as The book was interesting to everybody ? What value would the judge index take? Indeed, it is not straightforward under the relativist account of subjectivity that I have been using throughout the article. At the same time, I do not have to be committed to the relativist approach—the core observation that I made was that judge-dependent items in general do not have the experiencer argument that PPTs have. This does not itself rule out a possibility that the ‘real’ judge argument still exists and takes the form of a null pronoun co-indexed with the experiencer argument. In other words, the contextualist approach is viable—it is just that the judge argument cannot be the same argument as the experiencer argument of PPTs. In sentences like The book was interesting to John , PPTs would have two internal arguments: one would be the experiencer, and the other one would be the judge (null pronoun). If this is so, one could think of a mechanism making sure that in a sentence like The book was interesting to everybody the QR'ed quantifier would bind both its trace and the experiencer pro . The ‘judge=experience’ requirement would be an essential part of that story.

Finally, one more problematic issue that I have not touched in this article concerns the differences within the class of PPTs with respect to their argument structure. For example, while tasty takes an DP subject denoting a kind of food, fun can take either an event-denoting DP subject ( The ride was fun), or an infinitive clause: While both fun and tasty seem to semantically refer to an experience event, they do it in different ways. Developing a semantics that would account for these differences is a task for future work. As a first schematic attempt, I propose that the first argument of fun denotes a predicate over events rather than the ‘stimulus’ individual. This would amount to the following different denotations for tasty and fun : I believe that this take on the semantics of predicates like fun has potential for substantial improvement, but I leave it here as a starting point of future research.

  • (100) a.  The cake is tasty.

  •       b. *To eat the cake was tasty.

  • (101) a. The ride was fun.

  •      b. To ski is fun.

  • (102) 〚This cake is tasty pro8c ; g ; w , t , Sp

  •     a. is defined iff g(8) =  Sp ;

  •      b. whenever defined, denotes 1 iff ∃ s [ taste ( s ) & Experiencer( s , Sp ) &

  •       Stimulus( s , this.cake ) & taste ( s ) ≻  dSTSp for Sp at t in w ]

  • (103) 〚The ride was fun pro8c ; g ; w , t , Sp

  •     a. is defined iff g(8) =  Sp ;

  •     b. whenever defined, denotes 1 iff ∃ s [ ride ( s ) & Experiencer( s , Sp ) &

  •        fun ( s ) ≻  dSTSp for Sp at t in w ]

1 As discussed in ( Kennedy 2013 , 2016 ) and Fleisher (2013) , some English speakers find sentences like (1-b) somewhat degraded. Other speakers find these combinations acceptable, see, for instance, Bouchard (2012) , who bases his analysis on perfect acceptability of such examples. I will discuss the requirements of find more in section 4.1 .
One more potential ‘subjective attitude verb’ is consider . It is uncontroversial that it can combine with all the items in (1), including tall . However, I will not use it as a test for subjectivity due to the lack of understanding of the semantics and subcategorisation properties of consider . Although the requirements of find are not absolutely clear either, as the reader will notice as the discussion proceeds, there are serious reasons to think that find is the closest we get in English to a pure ‘subjective attitude’ verb requiring of its complement to be subjective. Look at the following contrast:
  •   (i) a. ??I find the Earth flat.

  •      b.  I consider the Earth flat.

The shape of the Earth is an objective fact, and thus it is unacceptable as a complement of find . Consider , to the contrary, is fine with the same complement, for most speakers, and describes the epistemic state of the subject. See the discussion in Stephenson ( 2007a , b ), Saebø (2009) and Bouchard (2012) .
2 As an anonymous reviewer points out, this argument only holds under an assumption that the covert judge argument is co-indexed with the wh -phrase that crosses over it. Many analyses would dispute precisely this claim.
3 Thanks to Yasutada Sudo for translations and judgements and to the audiences of FAJL6 in Berlin and ICL19 in Geneva for discussion.
4 Bouchard (2012) dismisses subjectivity of POS altogether, reducing it to the variation of comparison classes. Kennedy (2013) does not present an explicit semantics for subjective POS, but one can speculate that he would treat it along contextualist lines for some cases, and along relativist lines for other cases.
5 Sometimes for -phrases do appear with POS-DAs:
  •   (i) This river is a bit (too) deep for John.

Imporantly, these for -PPs have a different semantics—they are not judge or experiencer PPs. I address this issue in section 8 .
6 There is a potential counterexample to this generalisation: on the face of it, it is not very clear if the first-person restriction holds for POS-DAs in Japanese. It seems to depend on the kind of interpretation the sentence gets:
  •  (i) a. watasi-ni-wa kono kaban-wa omoi    J apanese

  •     I- dat - top       this bag- top    heavy

  •     ‘For me, this bag is heavy’

  •     b. John-ni-wa   kono kaban-wa omoi

  •     John- dat - top this   bag- top   heavy

  •     ‘For John, this bag is heavy’

  •      OK : Considering John as a candidate for carrying the bag.

  •      ?? : Looking at John suffering trying to carry the heavy bag.

For a discussion of the reasons for this behaviour of Japanese dative arguments with POS-DAs, I refer the reader to Bylinina ( 2014b : 64–71).
7 The contrast between ni-wa and ni-totte(-wa) judges is for some reason weaker for extreme adjectives than for evaluative adjectives, but the sentence with John-ni-wa is still slightly degraded. I do not have an explanation for why the contrast is weak.
8 Exactly the same reasoning is applicable to other ‘modal degree constructions’, such as the attributive-with-infinitive construction (i-a) ( Fleisher 2008 , 2011 ; Bylinina 2013 , 2014b ) or the ‘functional standard’ construction (i-b) ( Kagan and Alexejenko 2010 ; Bylinina 2012 ):
  •   (i) a. This is a long book (for me) (to read in one day).

  •      b. This book is a bit long (for me) (to read in one day).

I do not discuss these constructions here.
9 Anand assumes that subjectivity arises iff the predicate possesses the property of scalar variation—which is not correct, as I have shown above. Some examples of subjective predicates that are not predicates of scalar variation include POS-DAs like tall and extreme adjectives like gigantic . Still the point Anand makes is valid—how does one tell when the individual that has an effect on truth value of a statement is a judge, and when it is not?
10 The judgements on the acceptability of unembedded clauses with non-speaker judges are quite shaky. While some English speakers find sentences like This cake is tasty to John perfectly normal (or at least the version with taste as a verb— This cake tastes good to John ), some of the speakers I consulted said these sentences did not sound natural. On a related note, Orin Percus (personal communication) reports a split in judgments concerning sentences like John talked to a beautiful lady. #I mean, beautiful for him. Here the intended interpretation of beautiful is with respect to the subject of the clause the adjective appears in. Some speakers find this interpretation possible though. Interestingly, subjective items contrast with non-subjective perspective–sensitive items such as foreigner , which seem to allow such interpretations more easily: John talked to a foreigner. I mean, a foreigner for him. Possible reasons for this asymmetry are discussed in Bylinina et al . (2014) .
11 One could go for a decompositional analysis of judge-shifters assuming that they introduce a judge-shifting operator in syntax, and this operator shifts the judge, while the item itself does not. Both options are compatible with my analysis.
12 I discuss only small clause complements of find . For many speakers, full CPs with that are somewhat degraded as a complement of find in general, so I exclude it from consideration:
  •    (i) (??)I find that John is smart.

See Stephenson ( 2007a , b ) for a judgement that find only embeds small clauses and Bouchard (2012) for a judgement that find takes that-CPs without any problem.
There is also an option for an infinitive (Acc-cum-Inf) complement of find, but this combination has certain semantic effects that I try to avoid (see Moulton 2009 ).

Appendix: The Derivations

The analysis suggested in this paper relies on the following denotations for predicates of personal taste (104) and of the ‘situational’ positive morpheme POS s (105): I am assuming that existential closure closes the situation argument at a certain level in the derivation: The version of Function Application (FA) rule I am using takes care of semantic well-definedness of the constituents that are being combined (see Heim and Kratzer 1998 : 49): This is all we need to derive the meaning of a simple statement with a predicate of personal taste. Each sub-example in (109) unwraps the previous one, so (109-a) ⇔ (109-b) ⇔ (109-c) ⇔ (109-d): The derivation above gives the desired result for the baseline subjective statement when no judge-shifting is involved. To derive the meaning of a statement with judge-shifting, we need to spell out the semantics of a judge-shifting element. Here I define a judge-shifting operator Opi that has a judge-shifting effect and, in addition, some kind of other semantic impact that I indicate here with ω added to the input. I will ignore this bit of semantic contribution as it is orthogonal to the main task of this article. The semantics in (110) is a generalisation of meanings of various judge-shifting expressions that I discuss in the article (evidential morphology, epistemic modals, perspective-shifting expressions like in John's opinion etc.): Now we can derive the meaning of a subjective statement involving judge-shifting. It will be basically the same as the no-shifting example except for the upper layer in the structure, where the judge-shifting operator resides: The semantics for other items and constructions I mention (comparative / superlative constructions with predicates of personal taste and evaluative adjectives; the positive construction for dimensional adjectives, evaluative adjectives, and extreme adjectives) can be derived in basically the same way.

  • (104) 〚tasty〛 c ; g ; w , t , j ( ze )( xe )( dd )( ss )

  •     a. is defined iff j  =  z ;

  •     b. whenever defined, denotes 1 iff [ taste ( s ) & Experiencer( s , z ) &

  •              Stimulus( s , x ) & taste ( s , d ) for j at t in w ]

  • (105) 〚POS sc ; g ; w , t , j  = λ Sdst λ ss . maxd . S ( d )( s )) ≻  dstj

  • (106) 〚∃ Cc ; g ; w , t , j  = λ Pst .∃ s [ P ( s )]

  • (107) F unction A pplication

  •    Let α be a branching node with two daughters β and γ. For any context c, assignment g , for any world w , for any time t , and for any judge j , α  ∈  dom (〚〛 c ; g ; w , t , j ) iff

  •     a.  β  ∈  dom (〚〛 c ; g ; w , t , j ) and

  •     b.  γ  ∈  dom (〚〛 c ; g ; w , t , j ) and

  •     c. 〚γ〛 c ; g ; w , t , j  ∈  dom (〚β〛 c ; g ; w , t , j ).

  •    Whenever defined, 〚α〛 c ; g ; w , t , j  = 〚β〛 c ; g ; w , t , j (〚γ〛 c ; g ; w , t , j ).

  • (108) ∃ C POS s the cake is tasty pro8 .

  • graphic

  • (109)  C POS s the cake tasty pro8  ∈  dom (〚〛 c ; g , w , t , j )

  •     a. iff C  ∈  dom (〚〛 c ; g , w , t , j ) and

  •       POS s the cake tasty pro8  ∈  dom (〚〛 c ; g , w , t , j ) and

  •      〚POS s the cake tasty pro8c ; g , w , t , j  ∈  dom (〚∃ Cc ; g , w , t , j )

  •     b. iff C  ∈  dom (〚〛 c ; g , w , t , j ) and

  •       POS s  ∈  dom (〚〛 c ; g , w , t , j ) and

  •       the cake tasty pro8  ∈  dom (〚〛 c ; g , w , t , j ) and

  •      〚POS s the cake tasty pro8c ; g , w , t , j  ∈  dom (〚∃ Cc ; g , w , t , j ) and

  •      〚the cake tasty pro8c ; g , w , t , j  ∈  dom (〚POS sc ; g , w , t , j )

  •     c. iff C  ∈  dom (〚〛 c ; g , w , t , j ) and

  •       POS s  ∈  dom (〚〛 c ; g , w , t , j ) and

  •       the cake  ∈  dom (〚〛 c ; g , w , t , j ) and

  •       tasty pro8  ∈  dom (〚〛 c ; g , w , t , j ) and

  •      〚POS s the cake tasty pro8c ; g , w , t , j  ∈  dom (〚∃ Cc ; g , w , t , j ) and

  •      〚the cake tasty pro8c ; g , w , t , j  ∈  dom (〚POS sc ; g , w , t , j ) and

  •      〚the cake〛 c ; g , w , t , j  ∈  dom (〚tasty pro8c ; g , w , t , j )

  •     d. iff C  ∈  dom (〚〛 c ; g , w , t , j ) and

  •       POS s  ∈  dom (〚〛 c ; g , w , t , j ) and

  •       the cake  ∈  dom (〚〛 c ; g , w , t , j ) and

  •       tasty  ∈  dom (〚〛 c ; g , w , t , j ) and

  •       pro8  ∈  dom (〚〛 c ; g , w , t , j ) and

  •      〚POS s the cake tasty pro8c ; g , w , t , j  ∈  dom (〚∃ Cc ; g , w , t , j ) and

  •      〚the cake tasty pro8c ; g , w , t , j  ∈  dom (〚POS sc ; g , w , t , j ) and

  •      〚the cake〛 c ; g , w , t , j  ∈  dom (〚tasty pro8c ; g , w , t , j ) and

  •      〚 pro8c ; g , w , t , j  ∈  dom (〚tasty〛 c ; g , w , t , j ) ⇔ 〚 pro8c ; g , w , t , j  =  g (8) =  j .

  •    Whenever defined, 〚∃ C POS s the cake tasty pro8c ; g , w , t , j  =

  •    〚∃ Cc ; g , w , t , j (〚POS sc ; g , w , t , j (〚tasty〛 c ; g , w , t , j (〚 pro8c ; g , w , t , j )(〚the cake〛 c ; g , w , t , j ))) = 1 iff

  •    ∃ s [ taste ( s ) & Experiencer( s , j ) & Stimulus( s , the.cake ) & maxd . taste ( d )( s )) ≻  dstj for j at t in w ]

  • (110) For any c, g, w, t, j, [Op i A] is in dom (〚〛 c , g , w , t , j ) iff A is in dom (〚〛 c , g , w , t , g ( i ) ). If so, 〚[Op i A]〛 c , g , w , t , j  = 〚ωA〛 c , g , w , t , g ( i )

  • (111) Op iC POS s the cake is tasty pro8 .

  • graphic

  • (112)  Op i POS s the cake is tasty pro8  ∈  dom (〚〛 c ; g , w , t , j )

  •     a. iff C POS s the cake is tasty pro8  ∈  dom (〚〛 c ; g , w , t , g ( i ) )

  •     b. iff C  ∈  dom (〚〛 c ; g , w , t , g ( i ) ) and

  •      POS s  ∈  dom (〚〛 c ; g , w , t , g ( i ) ) and

  •      the cake  ∈  dom (〚〛 c ; g , w , t , g ( i ) ) and

  •      tasty  ∈  dom (〚〛 c ; g , w , t , g ( i ) ) and

  •      pro8  ∈  dom (〚〛 c ; g , w , t , g ( i ) ) and

  •     〚POS s the cake tasty pro8c ; g , w , t , g ( i )  ∈  dom (〚∃ Cc ; g , w , t , g ( i ) ) and

  •     〚the cake tasty pro8c ; g , w , t , g ( i )  ∈  dom (〚POS sc ; g , w , t , g ( i ) ) and

  •     〚the cake〛 c ; g , w , t , g ( i )  ∈  dom (〚tasty pro8c ; g , w , t , g ( i ) ) and

  •     〚 pro8c ; g , w , t , g ( i )  ∈  dom (〚tasty〛 c ; g , w , t , g ( i ) ) ⇔〚 pro8c ; g , w , t , g ( i )  =  g (8) =  g ( i ).

  •  Whenever defined, 〚Op iC POS s the cake tasty pro8c ; g , w , t , j  = 1 iff

  •  ∃ s [ taste ( s ) & Experiencer( s , g ( i )) & Stimulus( s , the.cake ) & maxd . taste ( d )( s )) ≻  dstg ( i ) for g ( i ) at t in w ]

References

Anand, P. (2009), Kinds of taste. Ms., UCSC.
Anand, P. & A. Nevins (2004), ‘Shifty operators in changing contexts’. In Proceedings of SALT14 , pp. 20–37.
Bale
A.
(
2008
), ‘
A universal scale of comparison
’.
Linguistics and Philosophy
 
31
:
1
55
.
Bale
A.
(
2011
), ‘
Scales and comparison classes
’.
Natural Language Semantics
 
19
:
169
90
.
Barker
C.
(
2002
), ‘
The dynamics of vagueness
’.
Linguistics and Philosophy
 
25
:
1
36
.
Barker
C.
(
2013
), ‘
Negotiating taste
’.
Inquiry
 
56
:
240
57
.
Belletti
A.
Rizzi
L.
(
1988
), ‘
Psych-verbs and θ-theory
’.
Natural Language and Linguistic Theory
 
6
:
291
352
.
Bierwisch
M.
(
1989
), ‘
The semantics of gradation
’. In
Bierwisch
M.
Lang
E.
(eds.),
Dimensional Adjectives: Grammatical Structure and Conceptual Interpretation
  ,
Springer-Verlag
.
Berlin
.
71
261
.
Bouchard, D.-E. (2012), Long-distance Degree Quantification and the Grammar of Subjectivity . Ph.D. thesis, McGill.
Bylinina
L.
(
2012
), ‘
Functional standards and the absolute / relative distinction
’. In
Sinn und Bedeutung (SuB) 16
  .
141
57
.
Bylinina
L.
(
2013
), ‘
Degree infinitival clauses
’. In
Snider
T.
(ed.),
Proceedings of SALT23
  .
394
411
.
Bylinina
L.
(
2014a
), ‘
Distributivity and definiteness in comparison classes
’. In
Proceedings of Sinn und Bedeutung 18
  .
Bylinina, L. (2014b), The Grammar of Standards: Judge-dependence, Purpose-relativity, and Comparison Classes in Degree Constructions . Ph.D. thesis, Utrecht Institute of Linguistics OTS, Utrecht.
Bylinina, L., E. McCready, & Y. Sudo (2014, November), The Landscape of Perspective Shifting . Talk given at “Pronouns in Embedded Contexts at the Syntax-Semantics Interface”, University of Tübingen.
Cohen, A. (2010), A tasty mixture. MS., Ben-Gurion University of the Negev.
Collins, J. (2013), The syntax of taste. MS., University of East Anglia.
Fara
D. G.
(
2000
), ‘
Shifting sands: An interest-relative theory of vagueness
’.
Philosophical Topics
 
28
:
45
81
.
Originally published under the name ‘Delia Graff’
.
Fleisher, N. (2008), Adjectives and Infinitives in Composition . Ph.D. thesis, University of California, Berkeley.
Fleisher
N.
(
2011
), ‘
Attributive adjectives, infinitival relatives, and the semantics of inappropriateness
’.
Journal of Linguistics
 
47
:
341
80
.
Fleisher
N.
(
2013
), ‘
The dynamics of subjectivity
’. In
Proceedings of SALT23
  .
276
94
.
Fults, S. (2006), The Structure of Comparison: An Investigation of Gradable Adjectives . Ph.D. thesis, University of Maryland.
Glanzberg
M.
(
2007
), ‘
Context, content, and relativism
’.
Philosophical Studies
 
136
:
1
29
.
Heim
I.
Kratzer
A.
(
1998
),
Semantics in Generative Grammar
  .
Blackwell
.
Oxford
.
Kagan, O. & S. Alexejenko (2010), ‘Degree modification in Russian morphology: The case of the suffix “-ovat”’. In Israeli Association for Theoretical Linguistics (IATL) 2010 .
Kennedy
C.
(
2007
), ‘
Vagueness and grammar: the semantics of relative and absolute gradable adjectives
’.
Linguistics and Philosophy
 
30
:
1
45
.
Kennedy
C.
(
2013
), ‘
Two sources of subjectivity: Qualitative assessment and dimensional uncertainty
’.
Inquiry
 
56
:
258
77
.
Kennedy
C.
(
2016
), ‘
Two kinds of subjectivity
’. In
Huitnik
J.
Meier
C.
(eds.),
Subjective Meaning
  .
Alternatives to Relativism
  .
Mouton de Gruyter
.
Berlin, Boston
.
105
26
.
Kölbel
M.
(
2002
),
Truth Without Objectivity
  .
Routledge
.
London
.
Kratzer
A.
(
1981
),
Words, Worlds, and Contexts
  ,
Chapter The Notional Category of Modality
,
de Gruyter
.
Berlin
.
38
74
.
Kratzer
A.
(
1996
), ‘
Severing the external argument from its verb
’. In
Roorych
J.
Zaring
L.
(eds.),
Phrase structure and the lexicon
  .
Kluwer Academic Publishers
.
Dordrecht
.
Kuno
S.
(
1973
),
The Structure of the Japanese Language
  .
MIT Press
.
Cambridge, MA
.
Kuroda, S.-Y. (1965), Generative Grammatical Studies in the Japanese Language . Ph.D. thesis, MIT.
Landau
I.
(
2010
),
The Locative Syntax of Experiencers
  .
MIT Press
.
Cambridge, MA
.
Landman
F.
(
2000
),
Events and Plurality: The Jerusalem Lectures
  .
Kluwer Academic Publishers
.
Dordrecht
.
Lasersohn
P.
(
2005
), ‘
Context dependence, disagreement, and predicates of personal taste
’.
Linguistics and Philosophy
 
28
:
643
86
.
Lasersohn
P.
(
2009
), ‘
Relative truth, speaker commitment, and control of implicit arguments
’.
Synthese
 
166
:
359
74
.
Marelj, M. (2004), Middles and Argument Structure across Languages . Ph.D. thesis, UiL-OTS.
McCready
E.
(
2007
), ‘
Context shifting in questions and elsewhere
’. In Estela Puig-Waldmüller (ed.),
Proceedings of Sinn und Bedeutung 11
  .
Universitat Pompeu Fabra
.
Barcelona
.
433
47
.
Meier
C.
(
2003
), ‘
The meaning of “too”, “enough”, and “so…that”
’.
Natural Language Semantics
 
10
:
69
107
.
Moltmann
F.
(
2010
), ‘
Relative truth and the first person
’.
Philosophical Studies
 
150
:
187
220
.
Morzycki
M.
(
2012
), ‘
Adjectival extremeness: Degree modification and contextually restricted scales
’.
Natural Language and Linguistic Theory
 
30
:
567
609
.
Moulton, K. (2009), Natural Selection and the Syntax of Clausal Complementation . Ph.D. thesis, UMass.
Nouwen
R.
(
2010
), ‘
Two kinds of modified numerals
’.
Semantics and Pragmatics
 
3
:
1
41
.
Paenen, M. (2011), Big fun! Master's thesis, UiL OTS.
Pearson
H.
(
2013
), ‘
A judge-free semantics for predicates of personal taste
’.
Journal of Semantics
 
30
:
103
54
.
Pesetsky
D.
(
1995
),
Zero syntax. Experiencers and Cascades
  .
The MIT Press
.
Cambridge, MA
.
Rákosi, G. (2006), Dative Experiencer Predicates in Hungarian . Ph.D. thesis, UiL-OTS.
Reinhart
T.
(
2002
), ‘
The theta system: an overview
’.
Theoretical Linguistics
 
28
:
229
90
.
Richard
M.
(
2004
), ‘
Contextualism and relativism
’.
Philosophical Studies
 
119
:
215
42
.
Saebø
K.-J.
(
2009
), ‘
Judgment ascriptions
’.
Linguistics and Philosophy
 
34
:
327
52
.
Sassoon
G.
(
2012
), ‘
A typology of multidimensional adjectives
’.
Journal of Semantics
  .
To appear
.
Schaffer
J.
(
2011
), ‘
Perspective in taste predicates and epistemic modals
’. In
Epistemic Modality
  ,
Oxford University Press
.
Oxford
.
179
226
.
Schwarz, F. (2009), Two Types of Definites in Natural Language . Ph.D. thesis, UMass.
Siloni
T.
Everaert
M.
Marelj
M.
(eds.) (
2012
),
The Theta System: Argument Structure at the Interface
  .
Oxford Studies in Theoretical Linguistics
.
OUP
.
Oxford
.
Stanley
J.
(
2000
), ‘
Context and logical form
’.
Linguistics and Philosophy
 
23
:
391
434
.
Stephenson
T.
(
2007a
), ‘
Judge dependence, epistemic modals, and predicates of personal taste
’.
Linguistics and Philosophy
 
30
:
487
525
.
Stephenson, T. (2007b), Towards a Theory of Subjective Meaning . Ph.D. thesis, MIT.
Stojanovic
I.
(
2007
), ‘
Talking about taste: Disagreement, implicit arguments, and relative truth
’.
Linguistics and Philosophy
 
30
:
691
706
.
Tenny
C. L.
(
2006
), ‘
Evidentiality, experiencers, and the syntax of sentience in Japanese
’.
Journal of East Asian Linguistics
 
15
:
245
88
.
von Fintel, K. (1994), Restrictions on Quantifier Domains . Ph.D. thesis, MIT.
von Stechow, A., S. Krasikova, and D. Penka (2004), The meaning of German um zu : Necessary condition and enough/too . Tubingen workshop on modal verbs and modality hand-out.
Westerståhl
D.
(
1984
), ‘
Some results on quantifiers
’.
Notre Dame Journal of Formal Logic
 
25
:
152
69
.