Long-Distance Degree Quantification and The Grammar of Subjectivity
David-Etienne´ Bouchard Department of Linguistics McGill University, Montr´eal
B.A., Universit´edeMontr´eal, 2006
December 2012
A thesis submitted to McGill University in partial fulfillment of the requirements of the degree of
DoctorofPhilosophy
c David-Etienne´ Bouchard 2012
Contents
Chapter 1: Introduction 5
1 Intensification At a Distance 5
2 Opinion Verbs 9
Chapter 2: Intensification at a Distance 12
1 Introduction 12
2 The Basic Distribution of IAD 15
3 The Lexicon of IAD 19 3.1ASyntacticCharacterizationoftheClassofIAD...... 20 3.2ASemanticCharacterizationoftheClassofIAD...... 21 3.2.1 GradableAdjectivesandComparatives...... 21 3.2.2 Too, Enough, and So...That ...... 29 3.2.3 Superlatives...... 33 3.2.4 Intensifiers...... 36
4 IAD as Overt DegP Movement 39 4.1BasicAnalysis...... 39 4.2GeneralPredictions...... 43 4.3DegPvs.DPscope...... 44 4.4IntensionalVerbs...... 51 4.5DegreeOperatorsandNegation...... 59 4.6SuperlativeScopeorContextualRestrictions?...... 67 4.7ScopeDataSummary...... 74 4.8 IAD and Ellipsis ...... 75 4.8.1 DegP Scope and Ellipsis ...... 75 4.8.2 FixingtheScope...... 77 4.8.3 Does DegP Scope Only Limit Ellipsis Size? ...... 80 4.8.4 DirectAnalysis...... 84 4.9 DegP Movement and the Merger Position of Than-Phrases...... 92
i 5 Asymmetries between DegP Movement and IAD 94 5.1AbsenceofIslandEffects...... 95 5.2AbsenceofSourcePosition...... 100 5.3AbsenceofDifferentialsandFactorials...... 101 5.4TheClassofIADOperators...... 103
6 Towards an Situ Semantics for IAD 110 6.1IADisnotQuantificationOverEvents...... 110 6.1.1 ApplyingaQADSemanticstoIAD?...... 118 6.2 Reconciling IAD with an In Situ Syntax ...... 119 6.2.1 Pragmatically-DerivedScales...... 123 6.2.2 MetalinguisticComparatives...... 129 6.3ExplainingtheLackofDifferentialsinIAD...... 135
7 Summary and Conclusion 138
Chapter 3: Opinion Verbs 141
1 Introduction 141
2 Opinion Verbs: The Basic Facts 143
3 Modeling Subjectivity 147 3.1 Lasersohn (2005) ...... 147 3.2AnAttemptedExtensiontoEpistemicModals...... 150 3.3Cotextualism:DisagreementisNeverFaultless...... 158 3.4Summary...... 161
4 Summary of Saebø (2009) 162
5WhatFind Asserts and Presupposes 165 5.1APresuppositionAccountoftheSubjectivityRequirement...... 167 5.2DerivingthePresupposition...... 169 5.3NegationintheComplementClause...... 172 5.4 Negative Quantifiers below Find ...... 175 5.5ThePresuppositionAccountvs.theType-MismatchAccount.... 177
ii 5.5.1 Trivial Propositions under Find ...... 177 5.6Summary...... 180
6 In Favour of Radical Reductionism 180 6.1 No De Dicto Ambiguities under Find ...... 181 6.2 Find isn’tEvidential,PPTsare...... 182 6.2.1 TheJudge’sBeliefs...... 183 6.3ThinkingandFinding...... 184
7 Syntactic Properties of the Complement of Find 186 7.1PresuppositionsandtheNecessityforDRT...... 186 7.2 The Syntactic Non-Locality of Find andSemanticTheories...... 188
8 Coordination under Find 191
9 Accounting For Saebø’s Syntactic Observations 203 9.1 Natural Judge-Dependency under find ...... 203 9.2ClauseStructure...... 205 9.3Summary...... 208
10 The Judge-Dependency of Gradable Adjectives 210 10.1DefinitenessRestrictiononGradableAdjectives...... 215 10.1.1ProblemswithPlurals...... 216 10.1.2 An Alternative Analysis: Subjective Comparison Classes . . . 217
11 Conclusion 222
Appendix A: A Judge-Free Semantics for Subjectivity 224
Appendix B: IAD across Opinion Verbs 228
References 230
iii Abstract
This thesis is concerned with two little known constructions at the syntax-semantics interface, namely a type of apparent long-distance degree quantification in Qu´ebec French called Intensification At a Distance (IAD), and a class of verbs that I simply refer to as opinion verbs and which include English to find and French trouver,for example. I examine two competing analyses of IAD, namely one where the surface word order is derived by overt DegP movement and one where it is base-generated. The former approach is a natural extension of the view in degree semantics that degree operators need a QR-type of operation to be interpretable. If it is right, then nothing needs to be added to the semantic component, and IAD can be treated as a distributional argument in favour of this semantics. Furthermore, it can be used to examine various proposals in this field, since if this hypothesis is right, then we can read the scope of degree operators right off surface syntax in this dialect. While this hypothesis is very successful at providing an interpretation to IAD sentences, it makes a number of incorrect syntactic predictions. I thus turn to an in situ analysis of IAD, which shares almost none of the syntactic problems of the movement analysis, but requires an entirely novel semantics to be interpretable. I thus suggest an interpretive mechanism for IAD sentences whereby the degree operator does not relate to any gradable predicate lower in the structure, but rather quantifies over degrees of appropriateness of its entire complement, in a manner very similar to how Morzycki (2011) analyses metalinguistic comparatives. The scale of the lower gradable adjective only comes to play a role in the pragmatics. I tentatively conclude in favour of the in situ analysis. Concerning opinion verbs, I present some novel data that show that sentences in find contain much semantic material that is presupposed, and I propose to formal- ize this in the form of what I call the Subjective Contingency Presupposition. This approach gives us an adequate way of describing what is asserted and what is presup- posed in such sentences, including in many problematic cases involving negation, and also gives us for free the fact that such verbs can only take subjective complements. I also suggest that a careful examination of the syntactic and semantic properties of their complement clause argues in favour of Lasersohn (2005)’s proposal that sub- jectivity in grammar is represented by a judge index on the interpretation function, rather than by null pronouns in the syntax. This is exactly contrary to Saebø (2009)’s conclusions, who recently proposed one of the first analyses of the verb find.This conclusion follows from the way that various kinds of subjective and non-subjective constituents may be conjoined under find.
1 Resum´ e´
Cette th`ese examine deux constructions peu connues `a l’interface de la syntaxe et la s´emantique, notamment un type de quantification de degr´es `a distance en fran¸cais qu´eb´ecois appel´e l’Intensification A` Distance (IAD), ainsi que la classe des verbes d’opinion, tels que trouver et to find en anglais. J’examine deux analyses possibles de l’IAD: une selon laquelle l’ordre des mots est d´eriv´e par une version visible du mouvement de DegP et une o`u l’ordre de surface est g´en´er´e tel quel. La premi`ere analyse est naturellement rattach´ee `a l’approche dans la s´emantique de degr´es selon laquelle les op´erateurs de degr´en´ecessitent une op´eration semblablealamont´ ` ee des quantificateurs pour ˆetre interpr´etables. Si cela est correct, il n’est pas n´ecessaire de modifier la s´emantique que nous donnonsa ` ces ´el´ements et l’IAD peut ˆetre consid´er´ee comme un argument distributionnel en faveur de cette s´emantique. De plus, l’IAD peut ainsi servirav´ ` erifier plusieurs hypoth`eses dans ce domaine, puisque si cette analyse est valide, il est possible d’´evaluer la port´ee des op´erateurs de degr´e`amˆeme la syntaxe de surface dans ce dialecte. Malgr´e que cette analyse arrive `a fournir une interpr´etation aux phrases `aIAD, elle fait plusieurs pr´edictions incorrectes sur le plan syntaxique. J’examinerai ainsi une analyse in situ de l’IAD, ce qui permet d’´eviter un grand nombre des probl`emes de l’analyse par mouvement, mais n´ecessite toutefois une s´emantique enti`erement nou- velle. Je proposerai donc un nouveau syst`eme d’interpr´etation pour les phrasesaIAD ` selon lequel l’op´erateur de degr´e n’a aucune relation directe avec un pr´edicat scalaire plus bas dans la phrase, mais quantifie plutˆot sur les degr´es d’ad´equation de son compl´ement entier. Cette analyse et largement inspir´ee de celle de Morzycki (2011) des comparatifs m´etalinguistiques. L’´echelle introduite par le pr´edicat scalaire ne de- vient pertinente pour l’interpr´etation de l’op´erateur qu’au niveau de la pragmatique. Je conclus provisoirement en faveur de l’analyse in situ. En ce qui concerne les verbes d’opinion, je pr´esente de nouvelles donn´ees d´emontrant que les phrases construites avec find contiennent beaucoup d’information pr´esuppos´ee. Je formalise cette information `a l’aide de ce que j’appelle la Pr´esupposition de Contingence Subjective. Cette formulation rend compte de ce que ces phrases affirment et pr´esupposent de mani`ere ad´equate, incluant dans de nombreux cas impliquant des n´egations enchˆass´ees. Le fait que les verbes d’opinion n’admettent que des compl´ements subjectifs est aussi expliqu´e par cette formulation. De plus, je sugg`ere qu’une analyse attentive des propri´et´es syntaxiques et s´emantiques des compl´ements des verbes d’opinion tenda ` confirmer l’analyse de Laser- sohn (2005) selon laquelle la subjectivit´eestrepr´esent´ee dans la grammaire `a l’aide d’un indice sur la fonction d’interpr´etation plutˆot que par des pronoms nuls dans la syntaxe, un r´esultat directement oppos´e`a celui de Saebø (2009), qui a r´ecemment propos´e une des premi`eres analyse des verbes d’opinion. Cela est justifi´e par les possibilit´es de conjonction de divers types de pr´edicats subjectifs et non-subjectifsa ` l’int´erieur du compl´ement de find.
2 acknowledgements
I would first of all like to thank my supervisor Bernhard Schwarz, without whom this thesis would never have seen the light of day. I have very often scheduled meetings with Bernhard with the firm impression that I was stuck for good and would not have more than five minute’s worth of things to say, only to come out over an hour later with a dozen interesting questions to work on and the clear feeling that I could answer them. I could not have hoped for a better supervisor. The other members of my committee, Junko Shimoyama and Michael Wagner, have also offered constant support and contributed many ideas that have made their way into the final version of this work. I am especially indebted to Michael, who would simply not be convinced by some parts of my proposals, and who in the end convinced me to explore an entirely different avenue, thus making this work so much more interesting. Brendan Gillon first, and Alan Bale later, helped me get a better grasp of what semantics really is, and got me to understand what all this formalism is all about instead of just pushing symbols around. A special thank goes to Luis Alonso-Ovalle, who finally made me understand something of modal semantics, a field that I could not afford to stay away from any longer. Lisa Travis’ syntax classes are fascinating enough that they almost got me to turn away from semantics completely. She has also been incredibly helpful to me in getting funding. My many discussions with her about how one writes a successful scholarship application are clearly behind the positive responses I have received from both SSHRC and FQRSC. Both Jon Nissenbaum and Maire Noonan have been great help in getting my evaluation papers together, and have left a lasting influence on how I think about, in the case of Jon, the syntax-semantics interface, and in the case of Maire, the syntax- phonology interface. During his short stay at McGill, Dan Silverman almost converted me to phonology with what was for me a drastically new outlook on the field. I will never be able to see phonology as I used to, even though I ended up resisting the call. A number of professors from the Universit´edeMontr´eal, where I did my under- graduate studies in linguistics, must also be mentioned. Daniel Valois’ encyclopedic knowledge of everything related to the grammar of French and his ability to uncover just the right pieces of data to further theoretical investigation has been an important inspiration for me. I have learned from him how to find a balance between letting the data that I find inform theory, and using theory to seek out new data. Both topics that I examine in this thesis were originally brought to my attention by Daniel, for which I am thankful. Rajendra Singh is probably the reason I have decided to pursue graduate studies in linguistics. The way he would hold us poor undergraduate students to the stan- dards usually reserved for established professors was both utterly unreasonable, and
3 incomparably motivating. I have never since had a professor so difficult to convince, which only made the few times that I did worth so much more. This is how I first got my first taste of what research in linguistics was all about, and I knew then that this is what I wanted to do. M. Singh passed away just as I was finishing writing this thesis, and I can only hope that it would have satisfied him. His standards are still the ones I hope to reach, however hard this may be. I thank Mireille Tremblay both for her great moral support and for giving me the opportunity to teach a number of linguistic classes, both at Queen’s and at the Universit´edeMontr´eal. I have discovered through this just how much I like sharing what I have learned about linguistics, and just how much one learns from teaching introductory material. My office mates have always helped create a fun and interesting working environ- ment. I am particularly thankful to Sacha Simonenko for patiently listening to me while I interrupted her work with half-baked hypotheses that I felt I could only get right if I spoke them out loud, as well as for the very enriching ensuing discussions. Jozina Vander Klok, Bethany Lochbihler and I have been nearly at the same stage of completion for a long while, and it has been an enormous confort to share with them these same last difficult steps, including applying for postdoctoral funding (suc- cessfully for all three of us, amazingly) and rushing to finish our respective theses. Bethany won the race to Jozina by a few weeks, who beat me by a mere ten days. I could not have asked for better companions in these tough times. A number of other students have made my experience at McGill very enjoyable. For all the good times, I thank Mina Sugimura, Walter Pedersen, Oner¨ Oz¸¨ celik, Eva Dobler, Eva Melkonian, Emily Sadlier-Brown, Edwin Howard, Michelle Saint- Amour, Gwendolyn Gillingham, Alanah Mckillen, Brian Buccola, Mike Hamilton, Moti Lieberman, and everyone else I may be forgetting. Tobin Skinner has been not only an eager participant in many discussions about syntax, but a great help in constructing this thesis in LATEX. Indeed, it is his template .tex files that I am typing these words. If this work looks any good, it is thanks to him. Heather Burnett is the one who pushed me to submit my first abstract to a con- ference, and then became my most important collaborator. Heather’s understanding of linguistic theory is phenomenal, and her ability with formal systems is something I can only hope to reach one day. Working with her has always been a pleasure, and I can hardly wait to collaborate again. Finally, I have to thank my wife Caroline Chevalier, who supported me through all these years of working nights and week-ends and travelling to conferences and summer schools. My daughter Alice deserves thanks too, both for turning my attention away from work once in a while to keep me sane, and for sleeping so well ever since she was only three-months old.
4 Chapter 1: Introduction
This thesis is concerned with two rather little-known phenomena at the syntax- semantics interface. I will examine a construction known as Intensification At a Distance (IAD), which as far as I know is unique to the dialect of French spoken in Qu´ebec, as well as the syntactic and semantic properties of opinion verbs, a class of verbs which has received very little attention in the past, but from which I believe numerous lessons about the properties of the grammar can be learned.
1 Intensification At a Distance
Let us start with IAD. Its name is somewhat of a misnomer since it does not necessarily involve “intensification”. Qu´ebec French allows degree operators like tellement ‘so’, plus ‘more’, and assez ‘enough’, for example, to appear either in the same position as they appear in English and in the standard dialect, or higher in the structure, usually in the adverbial position, in which case they are not string-adjacent to their associated gradable predicate. This is illustrated in (1-2).
(1) Roger a tellement donn´e un bon show hier. Roger has so given a good show yesterday (2) Roger a donn´e un tellement bon show hier. Roger has given a so good show yesterday ‘Roger gave such a good show yesterday.’
Since this construction almost always contains a degree operator relatively high in the structure and a gradable predicate lower, the obvious question to ask is whether IAD sentences are derived by movement of the degree operator from the position where it appears in canonical sentences, in French and in English, namely just before the gradable predicate. We will see that it is very difficult to tell whether this is correct, or whether an situ derivation is more appropriate. In general, the in situ analysis is more adequate to account for the syntax of IAD, while its semantics is more easily accounted for by a movement analysis. More easy does not mean more correct, however, and I will tentatively conclude that IAD sentences cannot be derived by movement, and
5 thus must be given a completely novel semantics, according to which such sentences are interpreted in a manner closer to that of metalinguistic comparatives than to the ordinary use of degree operators. I will argue that there is no compositional relation between, for example, tellement and bon in (1). This is not to say that the case is so clear-cut, however, and we will explore in detail what a potential movement analysis should involve. Arguing in favour of the movement analysis is the extensive literature in the field of degree semantics which suggests that degree operators are quantificational in nature and require movement of a constituent for interpretation, namely a DegP, just like the Quantifier-Raising school of thought requires quantified DPs to move for interpretation. Such theories should always raise suspicion for scholars interested in natural language, and independent evidence in favour of covert operations should always be sought out before we accept logical forms that differ too wildly from the observed surface structure. If IAD can be treated as an overt instance of DegP movement, then it would serve to strengthen the argument that the semantics associated with the LF involving movement is essentially correct. Degree operators in general, and comparatives in particular, have been given anal- yses that require LFs that differ from the surface structure in many respects. Of course, much work has already been done to give legitimacy to the claims that these constructions involve ellipsis, movement of some null operator, extraposition, and DegP movement. Still, cross-linguistic evidence that can be interpreted as showing that DegP movement can take place overtly would be a very strong argument in favour of the view that it is truly in effect even in languages where we do not see it. Supposing that we can indeed equate this alleged overt movement of a degree operator in Qu´ebec French with DegP movement, then we can use IAD to test various claims and compare various theories in degree semantics. DegP movement is normally a covert operation, and thus its properties can ordinarily only be glimpsed indirectly, via its role in licensing ellipsis, ACD, and scopal relations, among other things. This means that it is usually quite hard to tell exactly how it behaves, and many claims have been made that make specific assumptions about the details of DegP movement, despite the fact that these specific details have not been observed directly. I take up many of these claims in chapter 2 and use IAD to verify whether DegPs do exactly what these proposals assume that they do. This is a vast field, of course, and it is not possible to take up every proposal for which IAD may be of help. Here
6 are some of the issues I take up and the conclusion that treating IAD as overt DegP movement leads us to. First, I examine how scope ambiguities and lack thereof between DegPs and quan- tified noun phrases, then between DegPs and intensional verbs, carry over in Qu´ebec French. What we find is that IAD sentences where the DegP has apparently crossed over a quantificational element over which it has been claimed not to be able to take scope in English, then the result is ungrammatical. That is, the unavailability of wide-scope interpretations in English translates into judgements of ungrammaticality in exactly the way that is expected. Furthermore, when the semantics of the degree operator and other scope-bearing element are such that wide-scope readings are in- distinguishable from narrow scope readings, IAD sentences where the degree operator appears higher than that element are also ungrammatical, suggesting that quantified DPs and some class of intensional verbs are syntactic barriers to DegP movement. One apparent exception to this clean pattern concerns negation, since IAD is pos- sible across sentential negation, despite the fact that Heim (2001) specifically claims that the LF associated with such a movement should produce an uninterpretable re- sult. However, we will see that the way in which such IAD sentences are interpreted suggests that they do not involve true sentential negation, despite appearances, but rather constituent negation. As such, the possibility of degree operators crossing over pas in IAD sentences can be explained. If IAD is truly an overt instance of DegP movement, then this means that it allows us to tract the scope of degree operators with great precision, since its scope is given by its surface position. I use this feature of IAD to examine the relation between ellipsis and scope. Ellipsis in than-phrases in comparatives is usually assumed to be subject to a parallelism condition according to which the elided material may be interpreted as a subconstituent of the scope of the degree operator. Basically, the constituent that is derived by DegP movement sets an upper bound on the ellipsis size, and the elided material is allowed to be interpreted as parallel to the full scope of the degree operator, or to some part of it. DegP scope is hard to tell from English examples, however, so we turn to IAD sentences to examine how well this claim holds. What IAD allows us to conclude is that this theory is too permissive: the scope of the degree operator does not set an upper bound on the elided material in than-phrases, rather, it defines it exactly.
7 Finally, I show that Bhatt and Pancheva (2004)’s proposal concerning the precise mechanism involved in the merger of than-phrases in comparatives is incompatible with treating IAD as DegP movement. Simply put, their analysis assumes that DegP movement right-adjoins to its landing site, a fact that is directly contradicted by the observed word-order in French. In this case, IAD is a particularly useful tool, since their suggestion that DegP movement right-adjoins to its landing site is otherwise almost impossible to verify. Since the movement is covert, it has no effect on the surface structure, and since the authors assume a semantic framework where linear order plays no role in the interpretation, its direction of attachment is also semantically undetectable. Only by finding an overt version of DegP movement can we see that it is, in fact, left-adjoined. We then turn to an array of syntactic facts that cast doubt not just on treating IAD as overt DegP movement, but on any analysis where the degree operator is raised from besides its associated gradable predicate. We will see that IAD is not quite as sensitive to island constraints as it should be, and as previous work had assumed that it is. Also, comparatives in IAD are incompatible with differential and factorial arguments, a fact that receives no syntactic or semantic explanation under the overt DegP hypothesis. Additionally, some sentences clearly fall in the category of IAD, but they do not have a corresponding grammatical sentence with a canonical word order, suggesting that there is no acceptable ‘source’ structure for these sentences. Finally, we have no explanation for why only some degree operators whose semantics requires DegP movement participate in IAD, while others do not, such as the equative. Put together, these facts argue that IAD sentences are base-generated structures, but this is incompatible with standard assumptions about the semantics of these elements. I then turn to an examination of various options to provide an in situ semantics to IAD sentences. Drawing a comparison to another French construction, Quantification At a Distance, turns out to be a dead-end, since IAD has neither the syntactic nor the semantic properties to make such a semantics work, and even adding free lambda abstraction over non-local degree variables to the grammar turns out to get only part of the facts right (especially in not providing an explanation for the absence of differentials in IAD), in addition to being a major innovation to propose, in dire need of independent evidence. Finally, I settle on a different approach, likening the interpretation of IAD sen- tences to that of metalinguistic comparatives, using a modified version of Morzycki
8 (2011)’s analysis. Under this view, degree operators in IAD do not quantify over degrees on the scale introduced by some gradable predicate, but on the scale of ap- propriateness. That is, such sentences are used to make the claim that the property expressed by the VP is, informally, a more or less good fit for the matrix subject. The scale of the adjective only enters the interpretation at the level of pragmatics, where it provides the most salient way of ordering individuals with respect to the property expressed by the VP. A strong point in favour of this analysis is that it provides a straightforward explanation for why differentials like five feet are incompatible with IAD: it is simply because regardless what gradable adjective there may be below the degree operator, it is only the scale of appropriateness that matters in the semantics per se, and there is simply no measure word, in French or in English, that can be used to count degrees of appropriateness the way that feet can count length, or pounds can count weight.
2 Opinion Verbs
The second construction that I will examine is the English verb to find,asusedin sentences where an opinion is attributed to someone. It is an old problem of formal semantics that it assumes that every sentences, or at least every assertion, has a meaning that is reducible to its truth-conditions, and yet it seems that subjective statements are exactly those that do not share this property. The obvious problem is that it seems like there is no objectively satisfying way to define what a predicate like tasty could denote, under the assumption that it is just a set of individuals, or even if we treat it as gradable. One approach is to give up this assumption and to suggest that subjective predicates are all relational and generally involve covert pronominal arguments. That is, tasty relates individuals to the things they find tasty. Another approach is to take the much more drastic step of claiming that the interpretation function itself is relativized to some distinguished individual, the judge, and that what counts as tasty is different at different judges, in the same way that what the set denoted by the adjective dead varies across possible worlds. These two options are difficult to tell apart on the basis of the straightforward data made up of simple sentences like chili is tasty, perhaps even impossible if Stojanovic (2007) is to be believed. If we are to get a better grasp on how to analyse subjectivity in natural language, we need to look at more data concerning subjective statements,
9 and verbs like to find give us just that. First, however, we need to provide a thorough description of the meaning of this verb. Find is only compatible with complements that express subjective statements, which is the basic fact that any analysis of opinion verbs must account for. This is illustrated in (3-4).
(3) John finds Bill clever. (4) #John finds Bill dead.
On the basis of the observation that all the non-subjective information expressed in the complement of find is presupposed, as well as the behaviour of sentential negation in the complement clause, I propose that all sentences in find involve what I call the Subjective Contingency Presupposition, which states informally that keeping all non- subjective facts constant, it must be possible to judge the complement clause true, and it must be possible to judge it false. This presupposition adequately accounts for what we treat as asserted and what we treat as presupposed when we hear a sentence with find, including in cases where the complement clause is negated, and it also provides an explanation for why sentences like (4) are judged infelicitous: it is because their presupposition is a contradiction. I also show that this way of ruling out infelicitous sentences like (4) has the added advantage that it correctly predicts that tautological and contradictory statements below find should also be infelicitous, an observation that had not been made before, but that the Subjective Contingency Presupposition gets for free. Returning to the question of how best to represent judge-dependency in the gram- mar, Sæbø (2009) recently proposed that a careful study of the distribution of predi- cates of personal taste in its complement clause argues in favour of the pronoun theory of judge-dependency. I will argue for the exact opposite position. A more careful ex- amination of the syntactic, semantic and pragmatic properties of the complement of find argues in favour of relativizing the interpretation function to judges. The most crucial data concerns conjunction of subjective and non-subjective pred- icates below find. Saebø convincingly argues that this is predicted to be unavailable given the pronoun theory of judge-dependency, but not given the index theory. On the basis of the infelicitous nature of (5), then, he concludes that the pronoun theory is correct.
(5) #John is handsome and under 45.
10 However, while he is right that some cases are infelicitous, he is wrong in claiming that all are. The felicitous nature of the following example shows that at least some such conjunctions are actually felicitous.
(6) John finds that nobody at the party was handsome and under 45.
I thus propose that what we need is a semantic theory that admits conjunction of subjective and non-subjective predicates below find across the board, and rather suggest that the explanation for the infelicitous nature of (5) is due to problem in the pragmatics, not to any clash in the compositional semantics. Finally, I take a look at the behaviour of various classes of adjectives below find. I propose a definition for predicates of personal taste to distinguish them from pred- icates that are merely gradable and receive vague interpretations when used in the bare form. This definition states that true predicates of personal taste are those who remain subjective, and thus maintain their ability to license find,evenwhenused in the comparative. I also suggest that the ability of ordinary gradable adjectives to license find when used in the bare form is a consequence of the fact that their interpretation makes reference to a set called a comparison class, and that the value of this set is allowed to vary across judges.
11 Chapter 2: Intensification At a Distance
1 Introduction
Intensification At a Distance (IAD) is a little known construction that, to my knowl- edge, only exists in the dialect of French spoken in Qu´ebec. This construction is mainly characterized by the possibility of placing a degree operator higher in the sen- tence than where it usually appears, in addition to its canonical position just before a gradable predicate, which is where it appears in Standard French and English. (1) is an example of the canonical order, while (2) is an example of an IAD sentence.
(1) Jean a eu une tellement bonne note. Jean has had a so good grade
(2) Jean a tellement eu une bonne note. Jean has so had a good grade ‘Jean had such a good grade.’
IAD sentences are common enough in everyday speech, but they are restricted to a fairly low register. Furthermore, I have been much more successful at eliciting judgement on such sentences with speakers of a relatively young age (with best results with speakers below 20), while older consultants often flat out refuse every instance of the construction. As we will discuss at length in this chapter, there are a number of lexical items that behave in the way that tellement does in the examples above, but they are not always exactly the same across speakers. In order to provide a fuller description of IAD, the data that I present in this chapter concern the more permissive idiolects (which happen to include my own). Some of the data that I present is only acceptable to some speakers of Qu´ebec French, and I will be careful to note this each time. The first goal of this chapter is to give a more thorough description of the con- struction itself than it has received so far. The second goal is to use IAD to examine the syntax and semantics of degree operators more generally. We will examine two possible analyses of IAD. The first approach we will consider is that this construction is derived by an overt instance of DegP movement, an LF operation often assumed to be necessary for the interpretation of degree operators. The second approach that we consider is that IAD sentences are base-generated with their surface word-order, which
12 will make it necessary to produce a new semantics for degree operators, since the LF associated with a base-generated structure is incompatible with standard assumptions about the meaning of degree operators. The tension between these two approaches is as follows. On the one hand, treating IAD as an overt instance of DegP movement does a very good job of providing an interpretation to IAD sentences without any additional stipulation in most cases, in- cluding for sentences where very subtle scope effects are involved. On the other hand, there are strong syntactic arguments that seem to show that the degree operator could not have been moved from its alleged source position, suggesting that the word-order is base-generated, thus making the standard semantics of degree operators unusable. In other words, the former approach gets the semantics just right, but it does so at the price of a flawed syntax, while the latter approach gets the syntax just right, at the price of having to come up with a new semantics for degree operators. If we analyse IAD as being derived by movement, then surely the most interesting hypothesis is that we can equate this movement with raising of DegPs, since this movement has been independently proposed to be necessary for the interpretation of degree operators. Of course, an interesting feature of this approach is that it allows us to avoid adding anything substantial to the grammar of French to accommodate IAD, since DegP movement is assumed to be part of the grammar of this language anyway, only covertly, by those who adopt this view of the semantics of degree operators. Since this is a covert operation, however, it has thus far only been indirectly observed through its effect on scopal relations, ellipsis licensing, etc. If IAD is really an overt instance of this movement, then it can be used to examine many claims that have been made about DegP movement, confirming some and contradicting others. Under this view, we will see that the possibility of scope ambiguities and lack thereof that have been observed in English when degree operators appear on the sur- face below intensional operators and quantified DPs are transformed into grammatical and ungrammatical instances of IAD with surprising regularity. Furthermore, we can show that the source of the restriction against DegPs scoping over some elements is syntactic in nature since even when no truth-conditional effect would arise from this movement, it is still blocked in IAD. This is exactly the type of observation that could not be made by looking only at English and Standard French sentences: it is impossible to know whether a covert movement that has no effect on the semantics
13 can take place, since there is no way to observe it directly1. We will also examine the relation between DegP scope and ellipsis. It has been claimed by many that ellipsis in the than-phrase of comparatives is restricted in that the elided constituent can only be construed as a subpart of the LF scope of the degree operator. IAD will show us that this generalization should be strengthened to the claim that the ellipsis site in comparatives must be interpreted as exactly the scope of DegP. While the weaker generalization could be made to follow from standard views about parallelism in syntactic ellipsis, the stronger generalization that I propose suggests that we should adopt an analysis in terms of semantic ellipsis, rather than syntactic ellipsis, which comes out naturally under the Direct analysis of phrasal comparatives. We will also see that IAD provides straightforward evidence against Bhatt and Pancheva (2004)’s proposal concerning the mechanisms involved in merging the than- phrase in comparatives. Their proposal is entirely dependent on the claim that DegP movement right-adjoins to its landing site, an innocuous proposal if the operation is covert as it is in English. Since it can be overt in Qu´ebec French, we can observe directly where it attaches, and this position is not right-adjoined, but rather left- adjoined, as is usually (implicitly) assumed. These are all interesting results that we reach on the hypothesis that IAD can be equated with DegP movement. However, a number of problems emerge in any movement analysis of IAD: it can be shown to undergenerate, in cases where the target gradable predicate is embedded in a syntactic island, and when no source position is available for the degree operators. It also overgenerates since it predicts that it should be possible to produce IAD comparatives with differentials and factorials, and yet none of these are acceptable sentences. Finally, movement of the operator is unnecessary from a purely distributional point of view, since the operators involved in IAD sentences can all be generated in their IAD position independently in non-IAD sentences. This will lead us to compare IAD to another well-known quantification construction of French, namely Quantification At a Distance, which has generally received an in situ analysis. However we will argue that no such analysis can be extended to IAD. We then examine an alternative approach to the semantics of IAD sentences, which likens such sentences to metalinguistic comparatives. We will propose
1It is nevertheless possible to come up with ingenious designs such as those found in Fox (2000) that provide indirect evidence for the existence of such operations, although this evidence is not as strong.
14 that the interpretation of IAD sentences does not involve the scale introduced by some gradable predicate lower in the structure, but rather a scale of appropriateness. The scale of the adjective may come to play a role in how we understand the sentence, but only at the level of pragmatics. To put it bluntly, there is no compositional relation between degree operators and gradable predicates in IAD sentences. It is not easy to pick an analysis out of those two. In the end, I will reach the conclusion that a movement analysis in impossible given the strength of the syntactic arguments against it, and thus tentatively conclude in favour of the in situ analysis. A methodological point is in order before we begin. The very colloquial nature of this phenomenon forces a certain number of adjustments in constructing examples. IAD sentences quickly become unacceptable when they appear in a sentence with any element which is not typical of the more informal register of the spoken dialect of Qu´ebec. For this reason, I will avoid using the standard conjunction et et rather use the colloquial pis, I will also avoid the polite pronoun vous and systematically replace the first person plural nominative nous by on,etc.Aswewillseelater, I believe that not taking these precautions very seriously has led some authors to incorrect conclusions, since some sentences were judged ungrammatical by informants, but for reasons pertaining to dialect clashes, not to syntactic structure or semantic composition.
2 The Basic Distribution of IAD
Intensification At a Distance is a phenomenon brought to the attention of the lin- guistic literature as recently as Cyr (1991). In this section, we examine the basic distributional facts of this construction. A typical IAD sentence has a degree operator in a position higher than where it would appear in English or Standard French, and a gradable predicate lower in the structure. The gradable predicate may be an adjective as in (2) above, an adverb as in (3), or a gradable noun (4). Given a somewhat more complex structure, they can even relate to verbs lower in the structure (5).
(3) Monique a tellement travaill´e fort. Monique has so worked hard ‘Monique worked so hard.’
15 (4) Roger a tellement engag´e un g´enie. Roger has so hired a genius ‘Roger hired such a genius.’
(5) Paul a tellement trouv´e que ton chien puait. Paul has so found that your dog stank ‘Paul though your dog stank so much.’
Degree operators in IAD sentences have two positions that they can appear in. In addition to being to appear in the adverbial position between the auxiliary and the participle, the operator may also appear in the left periphery of indefinite DPs, before the article, as in (6) (Boivin and Valois (2009); Bouchard et al. (2011)). I call this construction “short IAD”, and distinguish it from “long IAD”, in which the operator is in the adverbial position already noted by Cyr. While I find short IAD sentences just as acceptable as long IAD, some speakers that I have consulted do not accept these, and only allow long IAD. As discussed in the introduction to this chapter, I will focus on the more permissive dialect here.
(6) On a vu tellement un bon film. We have seen so a good movie ‘We saw such a good movie.’
The position of the operator in this sentence is not unambiguously inside the DP, since it is well known that some adverbs can appear after the past participle and, furthermore, some may appear optionally before or after the participle (see Cinque (1999) for their full range of distributional possibilities). Souvent ‘often’ is one such example:
(7) Je suis souvent all´e au jardin botanique. I have often gone to-the garden botanical ‘I have often been to the botanical garden.’
(8) Je suis all´e souvent au jardin botanique. I have gone often to-the garden botanical Still, it is easy to show that the operator can appear inside DP. We can show this for example by conjoining the DP with another DP, as in (9). We can also insert the DP inside a PP to make sure that the operator is not in any low adverbial position, as in (10):
16 (9) J’ ai vu Jean pis tellement une belle fille. I have seen John and so a pretty girl. ‘I saw John and such a pretty girl.’
(10) J’ ai parl´e `a tellement une belle fille hier. I have talked to so a pretty girl yesterday ‘I talked to such a pretty girl yesterday.’
To avoid any confusion, henceforth examples of short IAD will always be contained inside PP to make sure that they cannot be confused with long IAD. The examples so far all look very much like exclamative, and this is also the case with other operators than tellement, which explains why the construction is called Intensification At a Distance. IAD sentences with assez as the IAD operator also naturally give rise such interpretations, which has led previous researchers to conclude that IAD sentences are always exclamatives (see Boivin and Valois (2009)):
(11) Les ´etudiants ont assez vu un bon film! The students have enough seen a good movie ‘The students saw such a good movie!’
However, this is not a necessary feature of IAD sentences, and I believe it is merely a consequence of focusing on a subset of the IAD operators, namely tellement,which lends itself easily to such a reading, and assez, which seems to be ambiguous between an operator that translates adequately as enough, and another that just serves to express a high degree2. It is not hard to come up with examples of IAD sentences that are not exclamative at all by using operators that do not lend themselves to such readings, such as the comparative:
(12) Jean a plus travaill´e fort que moi. John has more worked hard than me ‘John worked harder than I did.’ 2There may be more to this behaviour of assez than mere lexical ambiguity, since there is at least another item that is apparently ambiguous between a more ordinary meaning a a seemingly exclamative one, namely trop ‘too’.
(1) C’est trop un bon beat. ‘It’s such a great beat.’
The fact that trop and assez form a semantic natural class (cf. Meier (2003)) makes the claim that they are both ambiguous in the way just described seem unprincipled. I will not pursue this issue further here.
17 I will still use the name “Intensification At a Distance”, since this is how the construction is known already, but it should be clear that it is a misnomer: IAD could be better described as long-distance degree quantification. IAD can relate an operator higher in the structure to an adjective inside a DP, as we have seen, but only if that DP is an indefinite headed by either un(e)ordes.
(13) *Marc a tellement achet´e les/ deux/ plusieurs/ toutes les belles Marc has so bought the two many all the beautiful voitures. cars Faced with (a subset of) these distributional facts, Cyr’s work focuses on establish- ing that IAD sentences are derived by movement of the quantifier from the position where it appears in canonical sentences, constructing the non-canonical order on the basis of the canonical one. The way she shows this is mainly by placing the hypothe- sized source position inside a syntactic island of some sorts and showing that IAD is incompatible with this structure. For example, (14) is meant to show that IAD is im- possible when the source position of the quantifier is embedded inside a complement clause to a noun3:
(14) *On a trop ni´e le fait qu’ elle ait con¸cu de t grands projets. We have too denied the fact that she has conceived of great projects Intended: We denied the fact that she conceived too great projects.
Similarly, IAD cannot relate a degree operator in the matrix adverbial position to a gradable element inside an adjunct clause. This is shown in (15):
(15) *J’ai trop voyag´e en apportant des tgrosbagages. I-have too traveled while bringing of big luggage. Intended: ‘I traveled with luggage that was too big.’
Cyr attributes the ungrammaticality of these examples to a violation of the Complex-NP constraint, and a violation of the Adjunct-Island Constraint, respec- tively. In both cases, the movement of the quantifier trop from the trace position would have had to escape a constituent from which it is generally agreed that move- ment is disallowed. She provides additional examples to show that a similar sensitivity can be found with respect to other constraints on movement, such as the Wh-Islands
3We will see in Section 5.1 that this restriction does not actually hold and that syntactic islands do not always block IAD.
18 Constraint and the Internal Islands Constraint. Since embedding the base position of the quantifier inside any of those islands results in IAD sentences being ungrammati- cal, and since these constraints are considered to be strictly restrictions on movement, Cyr concludes that IAD sentences are always derived by movement of the quantifier from a position directly next to the predicate it is related to up to the adverbial position. Cyr’s analysis is strictly concerned with word-order and makes no attempt at pro- viding a formal semantic analysis of this phenomenon. However, it is interesting to note that the set of quantifiers that may participate in this construction is largely made up of items that are assumed to require covert movement in order to be inter- pretable in an analysis which treats comparatives and other degree constructions as quantificational, along the lines proposed in Heim (2001) and many others. A natural question to ask then is whether the degree operator movement that Cyr concludes to is in fact the overt counterpart of this semantically-driven movement independently proposed in the degree semantics literature. Most of this chapter is devoted to the examination of this hypothesis. I believe that it is an important question to answer since many claims have been made about DegP movement and, if an overt counter- part of it can really be found in IAD, it will be possible to use this construction as a powerful testing ground for these proposals. This corresponds to one way of picking out the class of IAD operators. In the next section, we examine this class in detail and propose two ways of defining this class. As suggested above, one is semantic and argues in favour of treating IAD as an overt instance of DegP movement. The other is syntactic and argues for an in situ analysis.
3TheLexiconofIAD
The quantifiers that can participate in IAD are very few. Below is what I believe to be an exhaustive list :
(16) plus, moins, tellement, assez, trop, le plus, le moins, more, less, so, enough, too, the most, the least, pas-diable, full not-a-hell-of-a-lot, quite There are two different ways we can pick out this class of items, one syntactic and one semantic. We begin by examining the syntactic generalization.
19 3.1 A Syntactic Characterization of the Class of IAD
The syntactic definition of the class of IAD operators is very simple: they are ex- actly those elements that can appear independently (that is, in non-IAD sentences) in preadjectival position, in the determiner position before de in nominals, and in the adverbial position between the auxiliary and the participle. Any element that can appear in all three positions can also participate in IAD. Take tellement for exam- ple. This operator can appear in all three positions, yielding interpretations where the length of the event, the number of individuals, or the degree of the adjective is intensified:
(17) On a tellement couru. We have so ran ‘We ran so much.’
(18) J’ ai rencontr´e tellement d’ ´etudiants. I have met so-many of students ‘I met so many students.’
(19) Je suis tellement fatigu´e. I am so tired ‘I am so tired.’
Every item in the list in (16) can replace tellement in these sentences and yield a grammatical sentence, provided that complement clauses that may be required by the degree operators, like than-phrases for comparatives, are added. What this shows is that they can participate in IAD precisely because the position that they occupy in such sentences is one that they can be generated in independently, in sentences where they do not relate to any non-local constituent. This is a very straightforward definition of the class of IAD operators, and very appealing, since the prenominal and preverbal positions are exactly those positions that the operators occupy in short and long IAD, respectively. As far as I know, this correlation is perfect: every single lexical item that can replace tellement in all three examples above can participate in IAD. It also explains why an obvious intensifier like tr`es, ‘very’ cannot participate in IAD. It is simply because tr`es only appears before adjectives and adverbs, not before verbs and nouns:
(20) * On a tr`es couru we have very ran
20 (21) * J’ ai rencontr´e tr`es d’ ´etudiants. I have met very of students This is one way of characterizing the class of IAD, and it is at first glance very successful. It picks out the class perfectly, and this observation means that we do not need to enrich the syntax to account for IAD sentences, since they can all be base- generated anyway. The problem we find is semantic. There seems to be a semantic relation between the degree operator and the gradable predicate below, but if IAD sentences are base-generated structures, then no such relation is expected. We can see that there is such a relation in the English paraphrases where the operator is always directly next to the gradable predicate. This is also clear from the readings that we get: while degree operators in adverbial position are normally used to make claims that concern quantifying over events, this is not the case in IAD sentences, where the degree operator clearly involve the scale of the adjective. On the other hand, there is another way of characterizing the class of IAD oper- ators, and it provides a way out of this semantic puzzle. Many, and possibly all the operators in (16) have been given a semantics that requires them to raise at LF for interpretation. If we can claim that IAD sentences are derived by an overt instance of this movement, then the semantic puzzle goes away, since this would mean that the surface structure of IAD sentences is just what is often assumed to be the LF of English sentences in the degree semantics literature. Better yet, if this case is strong, then we have in IAD an argument in favour of semantic analyses that require this movement. In the next sections, we review the semantic proposals that have been made about the elements in the list in (16).
3.2 A Semantic Characterization of the Class of IAD
3.2.1 Gradable Adjectives and Comparatives
The semantics of degree operators, and most notably comparatives, has received enor- mous attention in the literature, and it will not be possible to do justice to the entire set of proposals here. I will restrict myself to two popular recent approaches, namely that proposed in Kennedy (1997) and subsequent work, and the quantificational ap- proach, for which I take Heim (2001) as model. In order to have a theory of the comparative and other degree operators, we
21 first need a theory of gradable adjectives. Gradable adjectives are a somewhat fuzzy class, but they are typically defined as those adjectives that combine naturally with intensifiers like very, as well as with comparative morphology. Tall and heavy are clear examples of gradable adjectives, while dead is not gradable at all.
(22) John is very tall/heavy than Bill. (23) John is taller/heavier than Bill. (24) *John is very dead. (25) *John is deader/more dead that Bill.
The boundary between gradable and non-gradable adjectives is notoriously soft, however, and it is often possible to interpret adjectives in a gradable manner even if we would naturally be tempted to treat them as non-gradable. Adjectives for nationalities like American and Chinese seem like they should not be gradable, since one either has or does not have citizenship in a country, yet they are acceptable with very and the comparative, yielding a different meaning which can be paraphrased loosely as “having typical properties associated with people from this country”:
(26) John is very American. (27) John is more American than Bill.
This sort of metaphorical use is more or less easy with different adjectives, and it is quite unnatural with dead, so we will stick to this example as our benchmark non-gradable adjective. Both the quantificational theory and Kennedy’s theory agree that the semantics of gradable adjectives and the constructions in which they participate forces us to introduce the notion of degrees in our ontology. Degrees are the points that make up various scales, which correspond to various gradable properties expressed by adjec- tives. For example, tall is associated with a scale, namely that of height, and that scale is made up of an infinite number of ordered points, or degrees, each correspond- ing to some measure of height. Strictly speaking then, scales are just totally ordered sets of degrees4. In order to refer to them easily in our description of semantic deno-
4An additional factor to take into account is that degrees in a scale are also ordered along some dimension, which is what allows famous examples like (1) to be accounted for:
(1) This table is longer than it is wide.
22 tations below, we will follow the general convention and give degrees their own basic type, d, that we add to the standard basic types e and t. Gradable adjectives are used to relate individuals in their domain to such points. Of course, we have to assume that gradable adjectives have domains, since it is obvious that while an ordinary individual, say John, must have some weight and as such, must be in the domain of the adjective heavy, another proper name like England will denote something that clearly does not have any weight, and should not be in the domain of heavy.Thatis,heavy relates John to some degree of weight, but it does not relate England to anything at all. Kennedy’s theory and the quantificational theory differ in how this relation is cashed out compositionally. Kennedy claims that gradable adjectives are nothing more that measure functions, i.e. functions from individuals to degrees (type
(28) [[ tall ]] = λx. μheight(x) (Kennedy’s semantics)
In Heim’s quantificational approach, gradable adjectives are of type
(29) [[ tall ]] = λdλx. μheight(x) ≥ d (Quantificational semantics)
Note that the specific approach that I assume here involves what we can call an “at least” semantics for gradable adjectives, since for any gradable adjective P, it maps a degree d to the set of individuals who are P at least to degree d. For example, assume that an expression like six feet denotes an individual degree. Then, the following sentence means that John is at least six feet tall, rather than exactly six feet.
(30) John is six feet tall.
(31) [[ (30) ]] = μheight(j) ≥ 6’
We can make sense of such examples if we assume that the adjectives long and wide both involve the same scales, say “spatial dimension”, but different dimensions. Since degrees of length and degrees of width are on the same scale, they can be compared with the formal apparatus that I present in this section. For simplicity, we will disregard this refinement and treat each gradable predicate as if it involved its own unique measure function.
23 The motivation behind this “at least” semantics, rather than an “exactly” se- mantics, is the same as that behind the at least semantics of numeral determiners originally proposed by Horn (1972). We can show that the exactly interpretation that is the most natural in sentences like (30) can be denied without contradiction. This is demonstrated in (32):
(32) A: Is John six feet tall? B: Of course he is! In fact, he’s six feet three.
Under both theories, the adjective is not interpretable directly in the bare form in ordinary predicative sentences. In a simple sentence like (33), all the adjective has to combine with in the surface is an individual-denoting expression John, which is not sufficient to yield a truth-value with either denotation.
(33) John is tall.
In addition to the type problem, sentences like (33) have a meaning that is depen- dent on some contextual factors that serve to distinguish what counts as tall and what does not, i.e. to provide a standard of tallness in the context. This piece of meaning is not present in all uses of the word tall, say in the comparative or with measure phrases. Whether John is taller than Bill or not has nothing to do with any standard of height. In order to fix the type problem as well as introduce this context-dependent standard, we will assume that a covert positive operator POS is present in apparently bare uses of gradable adjectives (cf. Bartsch (1972)). In both theories, we can give a simplified5 version of the POS operator which simply turns the gradable adjective into a set of individuals, namely those who exceed some contextually-specified standard on the scale introduced by the adjective. This standard is fixed by the context-sensitive function norm. (33) becomes interpretable in either approach as long as the adjective is accompanied by this silent POS operator. Given the “at least” semantics of the quantificational theory, it will be useful to refer to the maximal degree in a set, which is exactly what the Max operator, defined below, will do.
(34) [[ POS ]] = λP
5These denotations are especially simplified in that they do not take into account comparison classes, i.e. those individuals on the basis of which the standard is derived, which would typically be taken to be arguments of POS (except by Kennedy (2007)). Much complexity is also hidden in the cover expression norm, an issue that we will take up in the next chapter.
24 (35) [[ POS ]] = λP
(38) [[ more ]] = λG
(39) t
e
Chicago
<
more large Rome large
6The suffix -er and the separate lexical item more are taken to be mere morphological variants of the same item with the meaning given above.
25 This is a somewhat simplified picture. More complex cases require some additional material in the than-phrase, including movement of a covert operator from inside the elided part of the than-phrase (originally proposed by Chomsky (1977); see also Bhatt and Pancheva (2004)), in order to ensure that the than-phrase ends up denoting a single degree. For example, according to Kennedy (2002), (40) would have the LF in
(41), where Deg0 is meant to make sure that the than-phrase denotes an individual degree. Its denotation, borrowed from Kennedy (2002) again, is provided in (42).
(40) John ate a larger piece of pizza than Bill.
(41) John ate a larger than [[Deg0 large] Bill ate a t piece of pizza]
(42) [[ Deg0 ]] = λG
I do not go in the details of examples of this sort here. Movement inside the than-phrase is a fairly standard assumption, independently of its necessity to derive the right meaning, since it has been known since Ross (1964) that the gap below than cannot be embedded inside a syntactic island:
(43) *John has better grades than Bill is a guy who has.
The way comparatives are interpreted in the quantificational approach is quite different. According to this view, the comparative morpheme denotes a relation be- tween sets of degrees. It is an operator of type <
(44) [[ more ]] = λP
The LF that this semantics requires for a sentence like (37) to be interpretable is quite different from the observed surface structure. Here is the LF that is needed for this example:
26 (45) IP
DegPi IP
Deg CP DP VP
more than Opk Rome Chicago V AP is tk-large
is ti A
large
Let us see how this is distinct from the surface. First, just as in Kennedy’s analysis, the than-phrase must be assumed to contain some elided material, in order to make sure that it gets to denote the set of degrees d such that Rome is d-large. This is done by postulating two covert things: an elided clause after the proper noun Rome, as well as a covert null operator moving from inside the elided clause to a position just above Rome. This null operator leaves a trace of type d and triggers lambda abstraction over this trace, but it is itself semantically vacuous. Also, despite the fact that the comparative morpheme sometimes appears as an affix on the gradable adjective on the surface, these two elements do not form a constituent at LF. This means that there is no sense in which the item larger actually gets any interpretation at all according to this approach, since it is not a genuine object at LF. Rather, the comparative morpheme forms a constituent with the than- phrase, which we label DegP, and together they move to the top of the clause. Since the than-phrase denotes a set of degrees and the comparative morpheme denotes a relation between sets of degrees, then together they form a DegP that denotes a generalized quantifier over degrees. The position that this constituent is generated in is where the gradable adjective expects to find a single degree-denoting argument (i.e. where the measure phrase six feet appears in (30)). The DegP would not be interpretable in situ, so it moves to the lowest clausal node, leaving a trace of type d again, in a manner very similar to how quantified object NPs are interpreted inside a Quantifier-Raising theory. Finally, we must assume that the than-phrase is subject to an obligatory extrapo- sition rule that does not feed LF, since it is neither pronounced in its base-generated
27 position (before large), nor its LF landing site7, but rather at the very end of the sentence. Once all this has taken place, the LF is finally interpretable. The than-phrase denotes the set of degree to which Rome is large, and the lower IP up to which DegP movement attaches denotes the set of degrees to which Chicago is large. The semantics that we have given says that the whole sentence will be true iff the maximal element in the latter set is higher than the maximal element in the former set. Both theories that I have presented here give adequate truth-conditions for com- parative sentences, but they do so in different ways, corresponding to different syntac- tic representations at LF. Kennedy’s semantics is fairly simple: it does not require LF movement of the degree operator or the than-phrase, and it provides an interpretation for words like larger, since the adjective and the degree operator form a constituent at LF. Some covert material must be assumed below than, but this is easily justifiable given the visibly truncated structure in (37). The quantificational approach appears at first glance very unwieldy in comparison. The LF constituency is very different from that at the surface. It requires the than- phrase to form a constituent with the comparative operator at LF, despite the fact that they are not adjacent in the surface, and conversely, the adjective and the degree operator do not form a constituent, despite the fact that they are syntactically close enough to take on a synthetic form when phonology allows it. It also requires an additional step compared to Kennedy’s semantics, namely the covert raising of the alleged DegP constituent, to yield an interpretable structure. All of these features of the analysis have received their share of attention and justification in the literature, and the analysis should certainly not be rejected wholesale because of its complexity. Nevertheless, the extent to which the LF that this theory requires differs from the surface structure means that we should have important evidence for this LF before we can accept the corresponding semantics, and I believe that IAD can contribute to this question in the following way. If the analysis according to which IAD is an overt instance of DegP movement can be maintained, then we would have direct evidence in favour of the reality of this LF movement, which would go a long way in strengthening the case for the corresponding quantificational semantics of degree operators.
7Although Bhatt and Pancheva (2004) propose an analysis of the than-phrase according to which its pronounced position is actually the same as its LF position, thanks to a rightward movement analysis of DegP movement. I will show in section 4.9 that such an analysis is incompatible with the view that IAD sentences are derived by overt DegP movement.
28 It is of course a trivial matter to adapt either theory to cover the operator less as well. It suffices to claim that sentences with less follow the exact same derivation as those with more, and the denotation of less differs from the one we have given above in (44) only in changing the “greater than” relation for the “smaller than” relation:
(46) [[ more ]] = λG
So far, we have seen that Kennedy’s semantics as well as the quantificational approach can both provide a suitable interpretation for sentences with comparative operators, but that the latter requires additional syntactic manipulations at LF to get off the ground. Among these, raising of the DegP constituent is the one that we will be concerned, since we will be examining the hypothesis that Qu´ebec French IAD is the overt counterpart of this movement. We now turn to the other operators involved in IAD and show that they also require a similar LF under a quantificational approach.
3.2.2 Too, Enough, and So...That
Meier (2003) specifically examines the English “modal comparatives” too, enough and so...that inside the quantificational theory, which I take to be the adequate transla- tions of French trop, assez and tellement...que, respectively. I will be taking it for granted here that the English lexical items are exactly synonymous to the French ones8. All three of these expressions take complement clauses, either a to-infinitival in the case of too and enough or a tensed that-clause with so.
(48) Billy is too old to drive. (49) Mark is smart enough to pass the test. (50) Jim is so tall that he can touch the ceiling.
We call these operators modal comparatives because, in the terms of the quantifi- cational theory, they involve comparison between a set of degrees in the real world to a hypothetical set of degrees. For example, (48) is analysed as expressing a relation
8See Hacquard (2011) for a discussion of how implicative readings emerge in such constructions in French but not in English. She claims that this difference is not a result of the two languages having degree operators with different meanings, but rather in the different way that the two languages encode aspect.
29 between the set of degrees d such that Billy is d-old, and another hypothetical set of degrees d’ to which Billy could be d’-old and still drive. Too is used to make the claim that the maximal degree in the first set is greater than the maximal degree in the second set. Meier (2003) and Heim (2001) disagree as to where existential quan- tification over worlds in this paraphrase comes from: it is inherent in the meaning of too for Heim, while it is introduced by a covert possibility modal for Meier. I will not attempt to choose between these two options here, rather taking Heim’s approach for ease of exposition9. Under this analysis, the LF of a too sentence is exactly parallel to that of a comparative, except for the content of the complement clause introduced by to.The complement clause forms a constituent with the degree operator, which we will still call DegP, and it has the denotation of a generalized quantifier over degrees. As such, since it is generated where the gradable adjective expects a degree-denoting expression, it must raise for interpretation to a type-t node, leaving behind a trace of type d, and the movement operation triggers lambda-abstraction over this variable. Some covert material must be postulated in the complement clause, but we will not be concerned with it here, except for assuming that the PRO subject of the infinitival verb is co-indexed with Billy.
(51) IP
DegPi IP
Deg CP NP IP
too PROk to drive N IVP
Billyk V AP
is ti old
Working from the top down, I give the denotation for this entire sentence, then, for the DegP only, then for too by itself. In these denotations, I do not specify what the
9These two approaches differ in the internal structure of the complement clause, but not so much in the meaning of the DegP constituent resulting from combining the degree operator and the complement clause. For the purposes of examining the behaviour of DegP as a whole, it is thus not necessary to decide strictly between Heim and Meier’s semantics.
30 accessibility relation is, since (48) is actually ambiguous as to whether is expresses that Billy is so old that he is no longer allowed to drive according to the law, or whether he is so old that he is no longer capable of driving, for example.
(52) [[ (51) ]]w =Max{d: oldw(d, b)} > Max{d: ∃w’∈Accw [oldw’(d, b) &
drivew’(b)]}
(53) [[ DegP ]]w = λP
(54) [[ too ]]w = λp
The main difference between a too-sentence and comparatives is the presence of modality in the complement clause. Fundamentally, too compares two sets of degrees just as more does. Crucially for our purposes, this analysis requires that the degree operator combines with the complement clause first to form a DegP which raises to the top of the clause, just like the comparative with its than-phrase above. The other operators enough and so can be given a parallel interpretation. (49) can be understood as making the claim that the maximal degree to which Mark is actually smart is higher than the minimal degree to which Mark could be smart and still pass the test. That is, the only difference between this operator and too is that enough refers to the minimal degree in the complement clause, while too refers to the maximal degree. Again, I give the denotation for the whole sentence, then the DegP enough to pass the test,thenenough by itself.
(55) [[ (49) ]]w =Max{d: smartw(d, m)} > Min{d: ∃w∈Accw [smartw’(d, m) &
pass-the-testw’(m)]}
(56) [[ DegP ]]w = λP
testw’(b) ] }
(57) [[ enough ]]w = λp
∃w’∈Accw[Qw’(d) & pw’]} (Quantificational semantics)
10This denotation is, in fact, neither Heim’s nor Meier’s. I avoid Meier’s complete semantics since it involves a major incursion into the meaning of conditionals, and Heim does not take into account the contribution of the complement clause. It is, however, very close in spirit to Heim’s proposal. See Nissenbaum and Schwarz (2011) for a denotation for too very similar to the one I use, except in that they do not make use of the maximality operator.
31 Again, the DegP is of type <
(58) Billy is too old to drive. (59) Billy is so old that he can’t drive.
While the modality is implicitly realized in (58), it is overt in (59). In both sentences, Meier claims that the operator relates the maximal degree expressed by the main clause at LF, i.e. the maximal degree to which Billy is old, to the maximal degree to which Billy can be old and still be capable or allowed to drive. Both operators serve to assert that the former is higher than the latter. So while there might be some difference in the content of the complement clause, Meier claims that the DegPs tootodriveand so that he can’t drive are exactly synonymous. Thus the fact that the latter must QR for interpretation follows from the same reasoning according to which the former must do so. What we have shown here is that the interpretation of sentences with the operators too, enough and so...that can reasonably be assumed to follow a similar syntactic derivation to that postulated for comparative sentences above, including movement of a constituent made up of the degree operator along with its complement clause. Much as in the case of comparatives, it is possible to analyse modal compara- tives inside Kennedy’s semantics, and it will not be necessary to involve any DegP movement to do this, although Kennedy does not do this himself. In the spirit of his denotation for the comparative that we used in section 3.1.1, and keeping constant the assumptions that the PRO subject in the complement clause is coindexed with Billy and that the modality is inherent in the degree operator, we can give too a meaning that will make it interpretable in situ. According to the following denotation, too combines with a gradable adjective, then with a proposition, and returns a set of individuals11:
11Even in an ‘exactly’ semantics of gradable adjectives, it is necessary to make use of the Max op- erator here, since the adjective appears in an intensional context (cf. Von Stechow (1984), Rullmann (1995)), Kennedy (2002). Von Stechow shows this with the following example:
(1) Kim can jump as far as Lee can jump.
There are multiple degrees d to which lee can jump d-far, so the semantics of this sentence can
32 (60) [[ too ]] = λG
pw’]}
This makes too interpretable in situ. The string too old to drive can now be claimed to be a constituent, both at LF and PF, and it gets interpreted as the following set of individuals:
(61) [[ too old PROb to drive ]] = λx. old(x) > Max{d: ∃w’∈ Accw [oldw’(x)=d
&drivew’(b) ] } (Kennedy’s semantics)
Of course, this kind of semantics can again be adapted to enough simply by chang- ing the maximality operator for the minimality operator:
(62) [[ enough ]] = λG
&pw’]}
The assumption that too and so are synonymous can be maintained in a Kennedy semantics just as well as in Meier’s. What we have seen here is that so-called modal comparatives can be (and have been) given a quantificational semantics according to which they involve comparison between sets of degrees, just as regular comparatives. This semantics requires the same LF operation that we focused on before, namely movement of the DegP to a clausal node. Pushing the parallel one step further, I have shown that it is possible to give to these operators a semantics which does not require such movement. So the situation is the same as before: there is at least one respect in which the quantificational semantics is, apparently needlessly, more complicated that the semantics that I have constructed on the basis of Kennedy’s proposal, and such complication should receive independent support if we are to admit the quantificational approach. Still moving down the list of the operators involved in IAD, we now turn to the interpretation of the superlatives le plus and le moins.
3.2.3 Superlatives
Superlatives le plus and le moins are a very different type of operator from those we have seen so far. According to Heim (1999), these items do not involve quantification only be captured by referring to the maximal such degree. The situation is the same for too, since it also involves existential quantification over worlds. There is no unique degree d such that Billy can be d-old and still drive, and it is the maximal such degree that interests us here.
33 over degrees, but rather relations between individuals and gradable properties. In order to maintain a consistent semantics for all degree operators, we can give a deno- tation to the superlative morpheme which is again based on maximization, which is the one used in Heim (2001)12:
(63) [[ -est ]] = λP
In its basic predicative use, a superlative operator attached to an adjective like what we see in (64) does not need to move to be interpreted, since it is already in a position to takes its adjective argument and its individual argument.
(64) (Of all my friends), Iza is the tallest. (65) [[ (64) ]] = Max{d: tall(d, i)} > Max{d: ∃y[y=i & tall(d, y)]}
However, when used attributively, it is clear that -est cannot be interpreted di- rectly next to its adjective. (67) gives the denotation that would result from inter- preting -est where it appears. This requires Iza to be a woman, and to be the tallest individual in the domain.
(66) (Of all my friends), Iza is the tallest woman. (67) [[ (66) ]] =Max {d: tall(d, i)} > Max{d: ∃y[y=i & tall(d, y)]} & woman(i)
In terms of the type calculus, this sentence is interpretable, but it will not yield the right interpretation if we do not change anything to our LF or our semantics. (66) makes an assertion concerning Iza’s height when compared to every other woman in the relevant domain, but makes no claim concerning the height of anyone who is not a woman. This sentence can still be true if some male friend of mine is taller than Iza. If -est were to combine directly with the adjective tall, it would yield the singleton set of the tallest individual in the domain. This would then be intersected with the noun woman, and the sentence would be true iff Iza is the tallest individual in the domain, as well as a woman. Of course, this is not the correct meaning as Iza need only be taller than other relevant woman for (64) to be true.
12Just as with more and -er, I assume that the suffix -est and the separate lexical item most are morphologically distinct but semantically identical. This denotation is meant to apply equally to both.
34 The correct meaning follows naturally if the superlative morpheme is merged next to the adjective, but moves higher in the structure, suppose just below the determiner in this example. Again, let us assume that that this movement leaves as a trace a variable of type-d and that it triggers lambda-abstraction over this variable. The resulting structure is one in which the superlative operator does not combine directly with the gradable adjective, but rather it combines with a complex gradable property, namely a relation between degrees of height and individuals that are women. The structure below the definite determiner should have the following form:
(68) NP
-est NP
AP NP
tAN
tall woman
This structure will give us the following truth-conditions for (66), which is the intuitively available meaning of this sentence since it only requires Iza to be the tallest woman, not the tallest individual period13.
(69) [[ (66) ]] = Max{d: tall(d, i) & woman(i)} > Max{d: ∃y[y=i & tall(d, y) & woman(i)]}
It is unclear what forces the superlative morpheme to move away from the ad- jective, since the whole structure would be interpretable without any type-mismatch in situ, yielding the meaning that Iza is the tallest in the context, and she is also a woman, which is a perfectly sensible thing to say. Since (64) says nothing about Iza’s height relative to non-women, however, we can at least be sure that whatever
13Admittedly, movement of the superlative morpheme is not the only way to get the correct meaning. The domain of tall could be narrowed down by the context to only include women, and we would get the same result. However, this solution is highly unsatisfying as it requires the overt nominal woman to play no part in the interpretation of the sentence, while a contextually specified parameter is responsible for restricting the comparison set to only women. Such an uneconomical mechanism seems highly suspicious to me. We will come back to the issue of contextual restrictions in superlatives when we discuss absolute/comparative ambiguities and superlative scope in section 4.6.
35 the reason, the superlative morpheme must obligatorily move at least as high as the top of DP14. This makes the superlative somewhat different from the regular compar- ative and the modal comparatives in that there is no type-mismatch to be resolved by movement of the DegP. Still, this makes it similar to these other constructions in that we have to postulate that DegP movement does take place, regardless why. There is really nothing interesting to say about a Kennedy semantics for superla- tives, since even Heim’s approach does not in principle require movement to yield an interpretable LF. We can just adapt the denotation in (70) to Kennedy’s types to provideaninterpretationinthisframework:
(70) [[ -est ]] = λP
Although I am not aware of any explicit discussion of this in the literature, such an approach would also require some changes to accommodate cases where the adjective is in attributive position, although it is unclear to me how a movement solution parallel to what I have outlined above would help, since I am unsure how one would need to interpret the trace of a degree operator in this semantics. I will not pursue this issue here, simply noting that attributive uses of superlative adjectives are problematic for the same reason in Kennedy’s semantics as in a quantificational approach.
3.2.4 Intensifiers
The last lexical items remaining on the list are pas diable and full. Pas diable is diffi- cult to translate. It is essentially a very colloquial way of expressing the same thought as not very in English. Conversely, full translates adequately as very or quite. Unlike the items we have examined so far, these lexical items are not the kind of element that one would expect to require any kind of LF movement in order to be interpretable, even inside the quantificational theory of degree operators. Matushansky (2002a) dis- tinguishes between degree operators, which is the category that contains every other
14For reasons unrelated to the issue here, Hackl (2000) suggests that predicative and attributive adjectives are actually different creatures entirely. He does this by enriching the entry for attributive adjectives by adding an
36 lexical item from the list of elements that participate in IAD and which require DegP movement in order to be interpretable, and degree modifiers, which she claims can be interpreted in situ. She specifically claims that very falls in the category of degree modifiers, and should not trigger movement for interpretation. She gives to the modi- fiers the type <
(71) Jean est full/pas diable grand. John is very/not devil tall ‘John is very/ not very tall.’
The only contribution of full and pas diable here is to make the standard of comparison involved in the interpretation of the sentence more or less stringent. The bar is set higher for one to be full grand ‘very tall’ than for one to be grand period, while it is lower for one to be pas diable grand ‘not very tall’. This contribution is akin to that performed by the POS operator introduced in section 2.1, which simply serves to provide a standard of evaluation for bare positive adjectives. There was no reason there to posit any sort of movement of POS for interpretation, just as there is no reason here to postulate movement of full and pas diable. For simplicity, we can give full and very the exact same semantics, which we model on the meaning we have given to POS. This denotation differs from that of POS only in that it makes use of the norm’ function, a stricter function than norm. Finally, we can give pas diable a similar denotation, now changing the function to norm”, which is less strict that norm15.
(72) [[ very/full ]] = λP
(73) [[ pas diable ]] = λP
Nothing in the semantics of these elements forces them to move for interpretation
15It is tempting to avoid introducing a third standard-fixing function norm” by making use of the presence of negation pas in the expression pas diable. I do not see any good reason to attempt this, however, since the full expression pas diable cannot be broken down into negation and diable.This latter word without negation loses its colloquial meaning and just means the same as the English word devil. As such, it cannot serve to modify a gradable adjective in any way.
(1) *Jean est diable grand.
37 at LF16. So far, they stand as the only counterexamples to the generalization that every item that participates in IAD must involve some version of DegP movement in Heim’s type of semantics for degree constructions. This is not to say that there is no precedent for claiming that intensifiers must move for interpretation. It has been claimed by Schwarz (2010) that the POS operator must be scopally mobile in order to account for fact that in a sentence like (74), reference is made to a standard price for hats belonging to three-year olds, and the claim is made that Mia’s hat is more expensive than that standard.
(74) Mia has an expensive hat for a three-year old.
Schwarz proposes that the POS operator must be able to move at LF, in this case up to the top of VP, in order to combine with a gradable property relating individuals who are three-year old to the price of their hat17.
(75) Mia POSk has [an tk expensive hat for a three-year old].
We can maintain the interpretation that we have given above for the POS operator, be it in Kennedy or Heim’s semantics, as both will produce the right interpretation here. On the assumption that the intensifiers like full and pas diable are essentially syn- onymous with the POS operator except in involving a more or less strict standard, then the fact that the following sentences also involve a standard price for hats be- longing to a three-year old means that they must have the same scopal mobility as POS. (76) makes the claim that Mia’s hat is a lot more expensive than that standard, while (77) says that it is somewhat below this standard.
(76) Mia a un chapeau full cher pour une fille de trois ans. Mia has a hat very expensive for a girl of three years ‘Mia has a very expensive hat for a three year old.’ 16Matushansky (2002b) makes the claim that POS must raise for interpretation in exactly the same way as DegPs like comparatives, since she makes it into an existential quantifier over degrees. She does not offer any evidence for this, however, and simply proposes it as a means to explain why sentences like the following are grammatical, given her proposal that seem is licensed only by complements that involve DegP movement.
(1) John seems sick
For reasons that I do not discuss here, however, I do not agree with her analysis of seem, and thus I do not adopt her view that POS must raise for interpretation unless independent evidence can be given for it. 17Some technical details about how this is derived compositionally are omitted for simplicity.
38 (77) Mia a un chapeau pas diable cher pour une fille de trois ans. Mia has a hat not devil expensive for a girl of three years ‘Mia has a not very expensive hat for a three year old.’
This puts intensifiers in the same category as the superlative operator. Their basic semantics should in principle make them interpretable in situ, but the interpretations that they actually produce is only consistent with an LF where they have raised higher in the structure. Again, it is somewhat unclear what forces or allows POS and the intensifiers to move at LF, but it will suffice for our purposes to observe that they do move. Note that there is no difference in prediction between Heim or Kennedy’s semantics here. Under both approaches, intensifiers are interpretable in situ, but we must nevertheless posit covert movement to account for the actual readings that are produced. What we have seen in the previous sections is that many operators that are allowed to participate in IAD must involve movement in a quantificational analysis of degree operators, but not in Kennedy’s. This makes an analysis of IAD in terms of overt degree QR appealing, although the fact that the superlatives and intensifiers do not actually require DegP movement, yet they also participate in IAD makes the case weaker. In the next section, we first show what such an analysis would look like, and then examine what we can learn about DegP movement under the hypothesis that IAD is an overt instance of it.
4 IAD as Overt DegP Movement
4.1 Basic Analysis
Raising of the [degree operator + complement clause] constituent for type reasons is the movement which we uniformly call DegP movement and that I hypothesize here takes place overtly in Qu´ebec French IAD sentences. In recent theories of generative syntax, especially in the Minimalist Program, the distinction between overt and covert movement is taken to result from the movement taking place after or before Spell- Out: if takes place before, it is overt; if it takes place after, it is covert. The choice is often taken to be an arbitrary fact about a given language, with the same syntactic operation taking place overtly in one language and covertly in another. Chinese wh- movement is covert, for example, while it is overt in English (Huang (1982)). Closer to
39 the subject at hand, Kiss (1995) claims that Hungarian quantified NPs are subject to an overt application of QR, providing direct distributional evidence for the existence of a universal rule of quantifier raising. Quite generally, in fact, finding an overt counterpart in a language to a hypothesized covert operation in another serves as strong evidence for the validity of the covert movement analysis. As far as I know, there is no such evidence in favour of DegP movement in the syntactic and semantic literature18.IfQu´ebec French IAD is indeed analysable as overt DegP movement, then it will strengthen the position that comparatives and similar constructions are indeed derived by covert DegP movement in other languages, as Hungarian overt QR favoured the movement analysis of QPs. Since there are alternative theories like Kennedy’s that do not involve this additional step of covert movement, such evidence is particularly important. The idea is the following. Syntactically, degree quantification constructions in- volve movement of the degree quantifier to a type-t node and extraposition of the complement of the degree operator to the right of the clause. The hypothesis that I want to follow below is that Qu´ebec French allows DegP movement to be covert, yielding essentially the same word order as in English and in Standard French, or overt, yielding the IAD sentences specific to Qu´ebec French. Let us examine the derivation of a simple IAD sentence such as (78) under this hypothesis.
(78) Anne a plus ´ecrit un bon article que moi. Anne has more written a good article than me ‘Anne wrote a better article than I did.’
Plus ‘more’ is generated directly to the left of the gradable adjective bon ‘good’, along with its sentential complement, forming a DegP interpreted as a generalized quantifier over degrees. This constituent raises to the lowest type-t node, namely the top of VP (assuming that the subjects are generated inside VP; cf. Koopman and Sportiche (1991)). Extraposition then moves the clausal complement to the right end
18Although Matushansky (2002a) suggest that Degree Fronting is movement to the escape hatch of DP in preparation for further covert DegP movement.
(1) John is [DP [more intelligent]j atj manthanhisfather].
That is, Degree Fronting is indirectly related to DegP movement, without being exactly its overt counterpart. Since this movement is not actually equated with DegP movement, it does not provide any genuine argument in favour of its existence.
40 of the clause, stranding plus to the left edge of VP. Finally, subject raising and verb raising take place, putting both elements before the moved degree determiner. The following tree illustrates in some detail the constituency involved in such a derivation:
(79) IP
NPj I
Anne a VP
DegPk VP
Deg CP tj VP
plus que moi V DP
´ecrit un NP
AP N
tk A article
bon
These broad lines of analysis should be common to all IAD sentences where the degree operator is in adverbial position. As we have seen, however, when the gradable adjective is embedded inside a DP, it is also acceptable for many speakers for the degree operator to appear in the left periphery of this DP.
(80) Jean a travaill´e dans tellement un bureau prestigieux. Jean has worked in so a cabinet prestigious ‘Jean worked in such a prestigious cabinet.’
This is only consistent with the current analysis to the extent that we admit that DPs contain a type-t node to which DegP movement can attach. This is not a particularly controversial claim, however. Heim and Kratzer (1998) argue that examples such as (81) show that QPs must be able to take scope from a position
41 internal to other DPs. Under the assumption that QPs are only interpretable if they move to a type-t node, then there must be one such node inside DP.
(81) No owner of an espresso machine drinks tea.
The interpretation of this sentence does not require that there be a single espresso machine such that no owner of it drinks tea, corresponding to the wide-scope of the indefinite over ‘no’. Yet, this should be the only available interpretation if the embedded indefinite were forced to escape the subject DP to reach a type-t node. Since the “no > ∃” reading is readily available, and even very strongly favoured, then the QP ‘an espresso machine’ must be able to find an interpretable position below ‘no owner’. This is accomplished by postulating a semantically null PRO subject in the structure and allowing it to move inside DP. The exact structural position that this element occupies is of little importance here, as long as it stands somewhere between theheadnounandD.
(82) DP
no NP
PRO1 NP*
t1 NP
N PP
owner of an espresso machine
With this derivation, we now have a type-t node inside DP, namely the NP node marked with an asterisk above. As such, it is an interpretable position for a generalized quantifier, so the DegP can move to this position to yield an interpretable structure. Extending the reasoning from the cases where the degree operator is in adverbial position, we can now claim that short IAD is the result of overt movement of the degree operator to this lower DP-internal position. That is, in (80), the degree operator has moved from its base position directly next to the adjective prestigieux up to its surface position just above the merger position of the DP subject. Note that I am assuming
42 that the indefinite article un is entirely vacuous here, following Heim (1982) and many others since then.
(83) PP
P DP
dans DegP DP
tellement D NP
(un) N NP
PROk tk NP
N AP
bureau tj A
prestigieux
An immediate advantage of treating IAD as overt DegP movement is that it pro- vides us with a way of explaining why IAD is only possible with indefinite DPs. This restriction can now simply be viewed as another instance of the more general re- striction against extraction from ‘specific’ or ‘definite’ DPs, exemplified in (84), from Davies and Dubinsky (2003):
(84) *Who did Tom read his/their/my book about t?
Regardless of the analysis that we give to this restriction, my point is simply that the DPs that block IAD are the same as those that block overt wh-movement. We can thus safely assume that this is not a specific property of IAD, but rather an instance of a more general phenomenon of conditions on extraction.
4.2 General Predictions
If we analyse IAD sentences as involving overt DegP movement, then we expect it to behave in all respects in the same way as covert DegP movement does. Since covert
43 DegP movement has been argued to be blocked in some cases, yielding only some plausible meanings but not others, then we expect these judgements of unavailable meanings in English to translate into grammaticality judgements in Qu´ebec French. The arguments that we will be looking at mainly concern the scope of compara- tive quantifiers. We will be looking at a number of restrictions on DegP movement, focusing in particular on the claims made in Heim (2001), where she explicitly con- trasts Kennedy’s semantics with a quantificational one, showing that we find some genuine scope ambiguities with degree operators if we look hard enough, despite the claims made in Kennedy (1997). First, we examine Kennedy’s Generalization, which disallows degree quantifiers from moving above quantified NPs. We then turn to DegP movement over intensional verbs, which is sometimes allowed and sometimes not, depending on the identity of the intensional verb. We then turn to negation and downward entailing environments in general, which never allow DegP movement to cross over it in English. Finally, we examine the absolute/comparative ambiguity in superlatives, which has been claimed before to be the result of multiple possible landing sites for DegP movement.
4.3 DegP vs. DP scope
As discussed extensively in Kennedy (1997), any quantificational theory of the com- parative construction according to which DegPs denote generalized quantifiers over degrees is faced with the problem that DegPs can never outscope c-commanding quantificational DPs. Heim (2001) calls this constraint ”Kennedy’s Generalization”:
(85) Kennedy’s Generalization If the scope of a quantificational DP contains the trace of a DegP, it also contains that DegP itself.
This is illustrated in the following example from Heim (2001)19:
(86) Every girl is less tall than that. (where that is equal to five feet)
19The than-phrase only contains a demonstrative to avoid any potential confound that could arise from having an ellipsis site in the sentence, or any quantificational material that could scopally interact with the rest of the sentence. I assume that this than-phrase denotes a singleton set of degrees, namely the degree which is equal to five feet tall.
44 (87) IP
QP IP
Q N DegP IP
Every girl Deg CP IVP
less than that V AP
is t tall (88) ∀x[ girl(x) → max{d: tall(d, x) } < {d: d=5’ } ]
(89) IP
DegP IP
Deg CP QP IP
less than that Q N IVP
every girl V AP
is t tall (90) #max{d: ∀x girl(x) → tall(d, x)} < max{d: d=5’}
(88) says that every girl is such that the highest degree d such that she is d-tall is lower than five feet, adequately representing the sentence’s truth-conditions. (90), however, makes a weaker claim. Since the maximal degree of height that every girl reaches ends up being the maximal height of the smallest girl, this simply means that the shortest girl is shorter than five feet, allowing for any number of other girls to exceed this height. This is not an available interpretation for (86). Kennedy’s Generalization is meant to exclude such unavailable readings. It is only if we adopt the quantificational theory of degree operators that this issue comes up at all. Kennedy (1997) claims, precisely on the basis of the absence of readings of this sort, that we must give up this approach, and adopt his analysis of degree operators,
45 which predicts that there should be no scope ambiguities of any sort between degree operators and other quantifiers, since degree operators are not quantificational at all. In contrast, the quantificational theory is forced to analyse this data as involving an ad hoc constraint of DegP movement, excluding LFs like (89) from being generated by syntactic principles. There is an important methodological point to be made here. This example from Heim (2001) is carefully chosen to have two potential LFs that are predicted to have distinct truth-conditions, given the meaning that we have given to the comparative operator less. This is easier said than done. For many sentences, no ambiguity is pre- dicted to ever be possible, with or without Kennedy’s Generalization. Generally, sen- tences containing the upward monotone more are such that whether the DegP scopes over or under another quantificational element will not make any truth-conditional difference. Let us look at the example (86) that showed us that the DegP could not scope over DP, but changing less for more, and the semantics associated with both its plausible LFs:
(91) Every girl is taller than that. (where that is equal to five feet) (92) max{d: ∀x girl(x) → tall(d, x)} > max{d: d=5’} (93) ∀x[ girl(x) → max{d: tall(d, x) } > max{d: d=5’} ]
(93) is quite transparent and adequately represents the meaning of this sentence. (92) is not quite as transparent, but actually says the exact same thing as (93). It says that the maximal degree d such that every girl is d-tall is greater than five feet. Again, the maximal degree such that every girl is that tall is the same as the actual height of the shortest girl, given the ‘at-least’ semantics we have been adopting for gradable adjectives. But if the shortest girl exceeds five feet tall, then all girls do. So both LFs can be claimed to be in principle possible, since they both yield the same truth-conditions, and those truth-conditions are satisfactory20. It is only when we look at examples like (86), or others with exactly-differentials that the potential for truth-
20This is somewhat of a controversial statement, since principles have been suggested to block exactly this kind of derivation. In particular, Takahashi (2006), following Fox (2000)’s more general principle, claims that some unavailable scope possibilities of comparative quantifiers like more than five can be correctly ruled out if we assume that scope shifting operations like DegP movement can only take place if they have truth-conditional consequences. The example that we are looking at is precisely a case where movement would be blocked by such a principle. We will discuss Takahashi’s proposal some more at the outset of this section.
46 conditional effects arise, and those sentences show us that Kennedy’s Generalization holds. Standard French does not behave any differently from English with respect to this constraint and respects Kennedy’s Generalization in sentences with the canonical word-order. That is, the direct translation of (86), given in (94), has the exact same interpretation and does not have the wide-scope reading of the comparative that would allow for any girl to exceed five feet, as long as one doesn’t.
(94) Toutes les filles sont moins grandes que ¸ca. All the girls are less tall than that ‘Every girl is less tall than that.’
What we want to see now is whether IAD, which we are analysing as an overt instance of DegP movement, actually allows moving a comparative quantifier over a QP. If our characterization of Kennedy’s constraint is correct, then the result should be ungrammatical. We will see below that this is indeed the case. In order to determine whether IAD is capable of moving a degree operator above a QP, we have to construct examples using ditransitive verbs, since the highest overt position of the degree operator is in the adverbial position, between the subject and auxiliary. It cannot be overtly placed before the subject under any circumstances. The canonical version of the sentence that we will use is the following:
(95) Jean a pr´esent´e tous les ´etudiants `a un moins grand professeur que Jean has presented all the students to a less great professor than ¸ca. that ‘Jean presented all the students to a less great professor than that.’
What we want to see then is whether the degree quantifier in the second object can be raised up to the adverbial position, above the first object which is a QP. As it turns out, the result is ungrammatical:
(96) *Jean a moins pr´esent´e tous les ´etudiants `a un grand professeur que Jean has less presented all the students to a great professor than ¸ca. that ‘Jean presented all the students to a less great professor than that.’
47 Before we can draw any conclusion from this, we must first show that this is not due to any restriction disallowing IAD to target a gradable predicate inside a second object. This is shown in (97), where IAD is perfectly acceptable, as long as the first object is a simple proper name or a definite:
(97) Jean a moins pr´esent´e Paul `a un grand professeur que ¸ca. Jean has less presented Paul to a great professor than that ‘Jean presented Paul to a less great professor than that.’
(98) Jean a moins pr´esent´e les ´etudiants `a un grand professeur que ¸ca. Jean has less presented the students to a great professor than that ‘Jean presented the students to a less great professor than that.’
This sentence also shows that IAD can target a gradable predicate inside a PP, which is not a trivial matter since French generally disallows overt movement out of PPs. The ungrammaticality of (96) exemplifies precisely the expected behaviour of overt DegP movement. English sentences like (86) show that DegPs cannot move over QPs at LF, since the interpretation associated with high DegP scope is unavailable. If we accept that IAD is an overt instance of DegP movement, then we expect that this movement should produce ungrammatical sentences when it crosses over a QP. This is exactly what is happening in (96). The degree operator moins, ‘less’, has raised from inside the second object up to the adverbial position, crossing over the QP tous les ´etudiants, and the result is ungrammatical, while this same construction is fine if we replace the offending QP by a proper name (97) or a definite (98). IAD sentences tell us more than their English counterparts, however. English examples can only show us that, semantically, DegPs cannot outscope QPs. On the one hand, this can be taken to show that our semantics is all wrong, as Kennedy claims, and that we need interpretive mechanisms that do not predict that such scope ambiguities can arise at all. On the other hand, it can be taken to show that the movement operation involved in deriving the unavailable reading is blocked for some syntactic reason. The ungrammaticality of (96) seems to argue in favour of the latter view. It is somewhat difficult to imagine exactly what interpretation we could give to an IAD sentence inside Kennedy’s semantics, apart from assuming that the degree operator is reconstructed at LF for the purpose of interpretation. Generally speaking, the fact that his theory is constructed so as to make sure that degree operators are
48 not scope-bearing elements predicts that the relative LF position of QPs and degree operators should have no effect on the interpretation of the sentence. This means that moving a degree operator above a QP in Kennedy’s semantics is not expected to have any effect at all, and the resulting sentence should not be rejected because of this21. Under the hypothesis that IAD sentences are derived by movement of the degree operator, what these sentences really show us is that the presence of a QP literally blocks movement of the degree quantifier. This is not a restriction on possible inter- pretations, since it is quite possible to imagine that the derivation of an IAD sentence that involves movement of a degree operator over a QP could be followed by sub- sequent QR of this QP over the DegP, thus yielding an LF which obeys Kennedy’s Generalization and produces the same interpretation as a canonical sentence. Since this option is not available for IAD sentences, then we are forced to conclude that Kennedy’s Generalization translates directly into a restriction on movement22. Above, we have seen that given the semantics provided by the quantificational theory of comparatives, in English, it is only possible to tell whether a DegP has moved above a quantified DP if the DegP is headed by a non-upward monotone de- gree quantifier, since otherwise the truth-conditions of the wide and narrow scope of DegP are indistinguishable. Because of this, it is impossible to tell from the English data whether Kennedy’s Generalization is in place across the board, for all degree quantifiers, or whether it only concerns those that would create an undesirable in- terpretation. IAD can help answer this question. Although English can only verify whether DegP movement has taken place when truth-conditional effects are triggered, we can simply build an IAD sentence where this movement takes place overtly and verify if it is grammatical or not. The result is that it is not, as shown by (99):
(99) *Jean a plus pr´esent´e tous les ´etudiants `a un grand professeur que Jean has more presented all the students to a great professor than ¸ca. that
21Again, this depends on the view that we adopt concerning the role of economy in the grammar. However, since degree operators in Kennedy’s semantics are not quantificational, moving them in an IAD sentence should not count as a scope-shifting operation, and it is thus not clear that principles of Economy like Fox (2000)’s and Takahashi (2006) would really be relevant. 22This restriction is highly reminiscent of Intervention Effects (cf. Beck (2006)), in which ’split’ constructions are disallowed when some sort of quantifier appears between the two parts of the split construction. I do not pursue this parallel in any detail here, but it is interesting to note that intervention effects are usually considered to be restrictions on LF movement only, while the effect here is on surface structure.
49 ‘Jean presented all the students to a greater professor than that.’
This sentence is the same as our example (96) above, except that moins ‘less’ has been replaced by plus ‘more’. The interesting feature of this example is that its predicted truth-conditions, were it an acceptable sentence, are the exact same as those of the corresponding sentence in the canonical word-order, which is the same as (95) with again plus substituted for moins. Yet the sentence is still unacceptable, strengthening the claim that Kennedy’s Generalization is actually a genuine syntactic restriction, and not a restriction based on excluded interpretations. Even when the semantics of the wide and narrow scope of the DegP are the same, it is still blocked from moving up above a QP. From this, I tentatively conclude that if I am correct in analysing IAD as overt DegP movement, then Kennedy’s Generalization is concretely realized in language as a syntactic restriction on DegP movement prohibiting it from crossing over quantifi- cational elements. On a final note to this section, the reader who is familiar with the literature on the scope of degree operators may wonder whether IAD can be of any help in explaining a well-known interpretive restriction concerning complex determiners such as more than three and less than five. QNPs headed by such determiners in object position can never outscope any other quantifier in subject position, nor can they outscope sentential negation (Beghelli (1995), Szabolcsi (1997) and Takahashi (2006), among others).
(100) Some student read more than five books. (101) John didn’t read more than five books.
The existential force in the subject DP must take scope over the object DP in (100). The inverse scope interpretation, stating that there there are more than five books that have been read by some (possibly distinct) student, is unavailable. Similarly, (101) does not have a reading that could be paraphrased as ‘There are more than five books that John did not read’, which would correspond to the wide scope of DP over negation. Similar facts can be shown to obtain when comparative determiners are used with basically any type of quantified noun phrase in subject position as well. (100) in particular is quite similar to the examples examined in this section in that they seem to show that comparative operators cannot raise above QPs.
50 Takahashi (2006) argues that this restriction obtains because in every example containing a complex comparative determiner, the reverse scope interpretation would follow from an illegitimate instance of DegP movement. The details of his analysis will not concern us here, however, since IAD will not be capable of helping us eval- uate it. The reason why this is so is that IAD is simply incompatible with complex comparative determiners. As shown in the following examples, there is no way to put the comparative operator anywhere in the sentence except inside the determiner, where it appears in canonical sentences. There is simply no variability in word order in comparative determiners.
(102) Jean a ´ecrit plus que trois livres. Jean has written more than three books ‘John wrote more than three books.’
(103) *Jean a plus que trois ´ecrit livres. Jean has more than three written books
(104) *Jean a plus ´ecrit que trois livres. Jean has more written than three books Why IAD is impossible at all in these constructions is an interesting question in itself, and we will return to it in section 5.1.2. For the moment, however, the fact that IAD is not compatible with complex determiners makes it impossible to use it to investigate their scopal possibilities.
4.4 Intensional Verbs
As noted by Stateva (1999) for superlatives and Heim (2001) for other degree con- structions, genuine scope ambiguities can be found when downward monotone degree quantifiers like less or non-monotone ones like exactly-differentials appear in inten- sional contexts. The simplest examples which show this involve inserting less under the modal required :
(105) The paper is required to be less long than that. (where that = 10 pages)
(106) ∀w∈Acc [max{d: paperw(x) → longw(d, x)} < max{d: d=10 pages}]
(107) max{d: ∀w∈Acc [paperw(x) → longw(d, x)]} < max{d: d=10 pages}
51 This sentence could be interpreted as stating that in all accessible worlds, the paper is shorter than 10 pages. This is equivalent to saying that it has a upper limit of 10 pages, and it corresponds to the narrow scope reading of the comparative with respect to the modal, with the semantics given in (106). The wide scope reading of the comparative states that the maximal length that the paper reaches in every accessible world is below ten pages. Given the at least semantics that we use for gradable adjectives, this is equivalent to saying that the required minimal length of the paper is somewhere below 10 pages. That is, no maximum is imposed on the length of the paper under this reading, but it says that there is some minimum whose exact value is not specified, but it should stand somewhere below 10 pages. This is the reading shown in (107). Both of these readings are available, providing us with the first case of a scope ambiguity involving DegP movement. As we will see, the wide-scope interpretation of the DegP over an intensional verb is possible for some but not all intensional verbs. As with the relative scope of DegP and DP, the potential for truth-conditional effects only arise with non-upward monotone quantifiers, so we will again stick to less in our examples. As a first step, we must establish whether the French equivalent of the English verbs that produce DegP scope ambiguities behave in the same way as their English counterparts. The following examples show that they do. Comparatives in moins produce ambiguities of the same sort as the English example above (105). This is shown with the possibility modal pouvoir and the necessity avoira `23. In all these examples, assume that the demonstrative ¸ca below que refers to a single degree which isequalto10pages.
(108) Paul a eu `a ´ecrire un article moins long que ¸ca. Paul has had to write a paper less long than that ‘Paul was required to write a less long paper than that.’
(109) ∀w∈Acc [max{d: paperw(x) & wrotew(p)(x) & longw(d, x)} < max{d: d=10 pages}]
(110) max{d: ∀w∈Acc [paperw(x) & wrotew(p)(x) & longw(d, x)]} < max{d: d=10 pages}
23I use the expression avoir `a, ‘to have to’, rather than the more standard devoir, since the latter is fairly rare in everyday spoken French. This is in line with the methodological discussion in the introduction: In order for judgements to be as easy as possible, I try to keep my examples in the same register that IAD sentences are ordinarily part of, namely a fairly colloquial register.
52 (111) Paul a pu ´ecrire un article moins long que ¸ca. Paul has can write a paper less long than that ‘Paul was allowed to write a less long paper than that.’
(112) ∃w∈Acc [max{d: paperw(x) & wrotew(p)(x) & longw(d, x)} < max{d: d=10 pages}]
(113) max{d: ∃w∈Acc [paperw(x) & wrotew(p)(x) & longw(d, x)]} < max{d: d=10 pages}
(108) has the same two scope possibilities as (105), as described above. (111) also has two possible readings. If the modal pu has scope over the DegP, the resulting reading is that Paul was allowed to write a paper that was shorter than 10 pages, by far the most obvious reading. The wide-scope reading of the DegP says that the maximal degree d for which there is an accessible world in which Paul wrote a paper that is d-long is 10 pages, meaning that it was not allowed for him to write a paper that exceeded 10 pages. This reading is also available, though not quite as readily. Turning back to IAD sentences, the reasoning that we will follow here is similar to the one used in the preceding section. That is, we expect that IAD sentences will be capable of placing a degree quantifier above an intensional verb in the same set of contexts in which covert movement is capable of yielding wide-scope interpretations for the DegP. In the contexts where such wide-scope interpretations are unavailable, then we expect IAD to yield ungrammatical sentences. Again, interpretive unavailability in English should translate into ungrammaticality in Qu´ebec French. In the following examples concerning the possibility modal pouvoir and the neces- sity modal avoira ` that we have just shown to yield scope ambiguities in canonical sentences, we can see that IAD is indeed free to place the degree quantifier either in the adverbial position embedded under the modal, yielding only the narrow scope interpretation, or above it, yielding the wide-scope interpretation.
(114) Jean a eu `a travailler moins longtemps que ¸ca. Jean has had to work less long than that ‘Jean had to work less than that.’ (less > modal) (modal > less)
(115) Jean a eu `a moins travailler longtemps que ¸ca. Jean has had to less work long than that (*less > modal) (modal > less)
53 (116) Jean a moins eu `a travailler longtemps que ¸ca. Jean has less had to work long than that (less > modal) (*modal > less)
(117) Jean a pu travailler moins longtemps que ¸ca. Jean has could work less long than that ‘Jean had to work less than that.’ (less > modal) (modal > less)
(118) Jean a pu moins travailler longtemps que ¸ca. Jean has could less work long than that (*less > modal) (modal > less)
(119) Jean a moins pu travailler longtemps que ¸ca. Jean has less could work long than that (less > modal) (*modal > less)
We use examples in the canonical word-order (114) and (117) as benchmark. Both of these sentences are ambiguous in the expected manner. In (115) and (118), the degree operator appears at the top of the embedded clause, while in (116) and (119), it appears above the modal in the matrix24. All of the IAD sentences are unambiguous and only allow the readings corresponding to their surface scope relations. The degree operator is interpreted below the modal in (115) and (118), and it is interpreted above it in (116) and (119). These examples show that in at least some cases, IAD sentences perfectly mirror what we expect the LF of the English and French canonical counterparts of these sentences to be. Covert DegP movement must be available to derive the correct inter- pretations in (114) and (117), and we see that the exact same movement operations can take place overtly in IAD sentences. We now turn to other modal contexts, starting with the intensional verbs want and need. Heim (2001) claims that of these two verbs, only need allows an interpretation corresponding to the wide-scope reading of the the DegP. I have however been unable to reproduce this judgement with speakers I have consulted, who uniformly accept this reading equally well with want and need.
(120) John wants to be less rich than that (where that=1 million dollars).
24The pass´ecompos´e tense is used in the matrix to create a position, in between the auxiliary and the modal, which is unambiguously in the matrix. In the absence of an auxiliary, the degree operator would necessarily follow the modal, and it would be difficult to tell whether it is part of the matrix clause or the embedded clause.
54 (121) John needs to be less rich than that.
The wide scope reading of want over DegP says that there is an upper limit to how rich John wants to be: he does not want his wealth to exceed 1 million dollars. This could be the case for example if John is convinced that you become a bad person if you have too much money, and so he wants to make sure to stay below some threshold, which for him is set at one million dollars. The reverse scope reading says that his minimal desired wealth, although unspecified here, is somewhere below 1 million dollars. This reading would arise naturally in a context in which John and Bill have discussed with me how much money they are trying to make. Bill says that he wants to be a millionaire, otherwise he will not be happy. On the other hand, John says that just half a million would be fine. In this case, I can also say (120), this time referring to John’s minimal desired wealth instead of the maximum. The readings corresponding to the same sentence with need are similar, with John’s desires replaced with his needs. Since these two verbs are almost only distinguished in the identity of the accessibility relation that they involve, it would be very strange to find that one of the two would disallow the wide-scope reading of the DegP, as Heim claims25. Regardless of this possible irregularity in English, the French pattern here is quite clear. Vouloir ‘want’ easily allows high scope of the degree operator, just as avoir besoin ‘need’ does.
(122) Jean veut ˆetre moins riche que ¸ca. Jean want to-be less rich than that ‘John wants to be less rich than that.’
(123) Jean a besoin d’ˆetre moins riche que ¸ca. Jean has need to-be less rich than that ‘Johnneedstobelessrichthanthat.’
Since Heim’s data on English want and my own differ, I will not be overly con- cerned with a comparison between French and English here. In any case, what con- cerns us more directly is the comparison between the French canonical order and the corresponding IAD sentence. The canonical sentences with the French equivalent of
25In a early version of (Heim (2001)), she suggest that the difference between these two verbs with respect to DegP scope is related to the fact that want is a neg-raising verb, while need is not, but this is abandoned in later versions.
55 want and need both allow the wide-scope reading of the DegP, and IAD behaves as expected in allowing movement of the degree operator over both verbs with equal ease. Again, long movement of the degree operator all the way above the intensional verb produces an unambiguous sentence where the DegP has wide scope, while a shorter movement that leaves the degree operator inside the embedded clause forces a narrow scope reading of DegP. (124) Jean a voulu moins ˆetre riche que ¸ca. Jean has wanted less to-be rich than that (*less > want) (want > less) (125) Jean a moins voulu ˆetre riche que ¸ca. Jean has less wanted to-be rich than that (less > want) (*want > less) (126) Jean a eu besoin de moins ˆetre riche que ¸ca. Jean has has need of less to-be rich than that (*less > need) (need > less) (127) Jean a moins eu besoin d’ˆetre riche que ¸ca. Jean has less had need of-to-be rich than that (less > need) (*need > less) While it may be the case that French and (some dialects of) English differ in that the former allows DegP to scope over want more readily, the important thing to see here is that the scope possibilities of the sentences in the canonical word order in French are the same as the possibilities allowed for IAD. This is in accordance with the view that IAD corresponds to an overt instance of DegP movement. Other classes of verbs that Heim does not examine provide us with examples where DegPs cannot be interpreted above an intensional verb in canonical sentences, and IAD is correspondingly impossible. Epistemic verbs like penser, ‘to think’, show this26. Take the following sentence (128), and the formulas associated with its two
26As discussed extensively in Von Stechow (1984), it has been claimed that the Russell ambiguity exemplified below is due to the possibility for DegP to scope over or under think (cf. Russell (1905) and Postal (1974), for example), which if true would show that at least in English, DegPs can move over think.
(1) John thinks the yacht is longer than it is.
This analysis is incorrect, however. Von Stechow shows that this is a matter of whether the DegP is interpreted de re or de dicto, which is not quite the same as whether it is interpreted inside or outside the scope of think. Russell ambiguities do not show that DegP can cross over such verbs. We will see below that there are reasons to think that they actually cannot.
56 potential LFs, corresponding to the narrow scope (129) and wide scope (130) of the DegP relative to the intensional verb.
(128) Alex a pens´e que Paul a ´ecrit une moins longue lettre que ¸ca. Alex has thought that Paul has written a less long letter than that (¸ca=10 pages)
‘Alex thought that Paul wrote a less long letter than that.’
(129) ∀w∈Acc [ max{d: letterw(x) & wrotew(p)(x) & longw(d, x)} < max{d: d=10 pages}]
(130) #max{d: ∀w∈Acc [letterw(x) & wrotew(p)(x) & longw(d, x)]} < max{d: d=10 pages}
The surface scope interpretation represented in (129) says that according to what Alex thinks, Paul’s letter didn’t exceed 10 pages. This is the only available meaning of (128). The formula in (130) corresponds to what would be the high scope interpretation of DegP over the intensional verb. It says that the longest that Paul’s letter reaches in every one of Alex’ doxastic alternatives is lower than 10 pages. That is, the shortest that Alex thought Paul’s letter might be was somewhere below 10 pages. This just sets a lower bound to how short Alex thought the letter might be: it does not specify any maximal length that Alex thought the letter might be, just that there is some minimal length, and that it is below 10 pages. This is not an available meaning, however, since this sentence does require that Alex thought that the letter’s length did not exceed 10 pages. Since this latter meaning is not available, then the natural conclusion to reach is that covert DegP movement is genuinely blocked by penser. The prediction that we make is that IAD sentences where this movement takes place overtly should not be acceptable. This is indeed the case.
(131) *Alex a moins pens´e que Paul a ´ecrit une longue lettre que ¸ca. Alex has more thought that Paul has written a long letter than that (¸ca=10 pages)
(131) is not grammatical, and the meaning that it would have if it were according to the semantic calculus is exactly the one that its canonical counterpart does not
57 have. This does not explain why it is that DegP movement is not acceptable over epistemic verbs, but it shows that where the covert movement is blocked, IAD also is, confirming the parallel. We find a similar pattern with verbs of reported speech, such as assurer ‘to assure’, garantir ‘to guarantee’, certifier ‘to certify’, etc.
(132) *Paul nous a tellement assur´e qu’ il avait une bonne job. Paul us-CL has so assured that he has a good job Also ungrammatical in the context of IAD is the use of verbs of perception.
(133) *J’ai trop vu qu’ elle avait un beau char. I-have too seen that she had a nice car
One class of verbs that does allow IAD across it are opinion verbs27,suchastrouver ‘find’ and consid´erer ‘consider’:
(134) J’ ai tellement trouv´e que Paul a donn´e un bon show. I have so found that Paul has given a good show ‘I found that Paul gave such a great show.’
Since the semantics of opinion verbs is not very well understood, and since they are the subject of the next chapter of this thesis, I will postpone the analysis of such examples until we have a working semantics for these verbs. A discussion of such examples can be found in Appendix B. For the moment at least, we can use these examples to show that the grammaticality of IAD examples has nothing to do with the tensed/infinitival character of the complement clause. The observant reader may have noticed that except for opinion verbs, every grammatical example of IAD across a clause boundary so far involved an embedded infinitival clause, while the ungrammatical examples all involved tensed clauses. It would be tempting to make use of this fact to explain the ungrammaticality of (133) and (132), since tensed clauses are notoriously more opaque to extraction than their infinitival counterparts. However, the grammaticality of examples like (134) show that this cannot be the case. It is a task beyond the scope of this thesis to go through every possible embedding verb in French to see whether we find a correlation between their ability to allow DegP to scope over them and whether they allow IAD to cross them. So far, we have seen that for all the verbs that Heim (2001) has examined as well as some others,
27I thank Daniel Valois and Marie-Claude Boivin for pointing this out to me.
58 the correlation has held very well. Although this does not tell us why it is that certain verbs allow DegP movement to cross over them and others do not, it helps confirm the view that IAD is grammatical exactly where covert DegP movement is also grammatical. Especially given the very fine nature of the judgements involved in such examples, I believe that the fact that unavailability of readings in English and French canonical sentences turn into grammaticality judgements in IAD sentences is a strong argument in favour of treating IAD as overt DegP movement.
4.5 Degree Operators and Negation
Degree operators appearing in sentences that also contain sentential negation do not produce any truth-conditional ambiguities. For example, sentence (135) has a single interpretation, corresponding to low scope of the degree operator with respect to the negation:
(135) John is not taller than Bill. (136) ¬ Max{d: tall(d, j)} > Max{d: tall(d, b)}
Given the maximality-based version of the quantificational analysis of compara- tives that we have been using so far, this should not be surprising. As Heim (2001) shows, if DegP were to move outside the scope of negation, it would yield an invalid semantic formula, because it contains at least one instance of the Max operator where it applies to a set that does not have a unique maximal element. This formula is given in (137):
(137) Max{d: ¬tall(d, j)} > Max{d: ¬tall(d, b)}28
The problem with this formula is that there is no such a thing as the maximal degree to which John (or Bill) is not tall. Since the scale of height has no upper bound, for any degree d that we find such that John is not d-tall, we can always find another degree d’ which is higher than d and such that John is not d’-tall29. We can explain
28Since the value of the ellipsis site below than is provided by the matrix clause, the fact that the matrix contains the negation forces negation to also appear in the than-phrase. Details concerning the ellipsis will be discussed in Section 4.8. 29This is perhaps not the best explanation for the absence of any ambiguity in this case since other adjectives actually do have scales that are bounded. According to Kennedy and McNally (2005) and Kennedy (2007), an adjective like full for example is bounded, and so for some individual x which is not perfectly full, we could say that max{d: ¬full(d, x)} is defined, since the maximal degree d
59 why (135) is not ambiguous because the LF in which the DegP raises over negation does not produce a consistent semantics interpretation. That is, such sentences tell us nothing about DegP movement because we cannot tell if it is syntactically possible, since it would be semantically undetectable. If this explanation for the unavailability of any detectable meaning corresponding to the DegP over negation scope option is correct, then we expect IAD sentences to simply disallow overt placement of the comparative operator before sentential nega- tion, because the resulting sentence should not be associated with a meaningful in- terpretation. However, as shown in (138), it is possible to construct grammatical examples of this kind, although the result is somewhat marginal.
(138) ?Jean a plus pas travaill´e fort que Paul. Jean has more not worked hard than Paul. ‘Jean worked (even) less hard than Paul did.’
This sentence says a bit more than simply comparing how hard Paul and Jean worked. It also gets an evaluative reading and, as such, says both that Jean and Paul did not work much, and also that Jean worked even less that Paul did. Thus there are two things to explain about (138), namely why it is grammatical at all, and why it has an evaluative reading. To make things more concrete, I borrow Bierwisch (1989)’s test, according to which a sentence with a gradable adjective is evaluative iff it entails the same sentence without degree morphology. According to this test, (138) is evaluative because it entails (139):
(139) Jean a travaill´e pas fort Jean has worked hard ‘Jean didn’t work hard.’ such that x is not d-full is the degree d corresponding to being absolutely empty. This means that a similar sentence to (135) with full instead of tall should be ambiguous, and yet it is not:
(1) This glass is not more full than that cup.
In addition to its obvious available meaning, this sentence should have an additional meaning according to which the maximal degree to which this glass is not full, i.e. complete emptiness, is higher than the maximal degree to which that cup is full. That is, it should compare complete emptiness to how full that cup is, in practice ignoring the glass completely. Since this is not an available meaning for this sentence, I am not certain that excluding the wide-scope reading of DegP over negation on the basis of an undefined maximum in the semantic calculus is entirely correct. This counterargument is entire dependent on accepting that adjectives like full genuinely have maximal degrees, which is not entirely obvious in itself. More work on the typology of adjectives would be necessary to see how bounded adjectives can be integrated in this analysis.
60 Note that this evaluativity is not present in the canonical version of this sentence, as we can see from the fact that (140) does not entail (139).
(140) Jean a pas travaill´e plus fort que Paul. Jean has not worked more hard than Paul. ‘JeanworkedlesshardthanPauldid.’
We could answer the first question, namely why IAD is possible over negation, by modifying the semantics that we give to the comparative operator. Heim claims that the absence of scope ambiguity between degree operators and negation does not allow us to tell whether DegP can or cannot move over negation, since in this case, we can make the case that its associated semantics is ill-formed anyway, which would make a wide-scope LF undetectable independently. In other words, either syntax or semantics may be responsible for the absence of a wide-scope interpretation of DegP. Either of these two explanations predicts that IAD should be impossible across sentential negation, since it should be either blocked syntactically, or associated with an ill-formed semantic formula. However, it could also logically be the case that the lack of ambiguity results from the two LFs being truth-conditionally equivalent. Heim’s analysis of the lack of ambiguity is dependent on the assumption that the semantics of plus involves maximization in both the than-phrase and the matrix, which is not the standard view, as Heim herself notes. Suppose we adopt a slightly different view of the meaning of the comparative morpheme, and rather give it the meaning in (141):
(141) [[ plus ]] = λP
In effect, given the “at least” semantics we have provided for gradable adjectives, this will yield the same result as the previous, maximality-based entry in most cases, as the reader can verify30. Below are the predicted truth-conditions of the LFs corresponding to the narrow and wide scope of this new plus with respect to negation in (135):
(142) ¬({d: tall(d, b)}⊂{d: tall(d, j)}) (143) {d: ¬tall(d, b)}⊂{d: ¬d: tall(d, j)}
30Although see Heim (2006) for arguments in favor of this latter view in terms of the proper subset relation.
61 (142) gives the meaning for the LF where DegP scopes below negation, and it is transparently the only attested interpretation of this sentence. What about (143) then? It says that the set of degrees to which Bill is not tall is a proper subset of the set of degrees to which John is not tall. This is almost an adequate description of the truth-conditions of this sentence, except that it does not allow for John and Bill to be of the same height, while intuitions about (135) do. However since the two readings are in an entailment relation ((143) entails (142)), it is possible that the reading in (143) is actually available, but difficult to detect in such examples. Suppose this is correct, and we import it to the Qu´ebec French case, treating plus in this sentence as having the same denotation as more in (141). Then we expect that IAD should be possible across sentential negation, thus correctly predicting that (138) is a grammatical sentence. Furthermore, this sentence cannot be used truthfully if Jean and Paul both worked equally hard. As the following formula representing the wide-scope reading of the comparative operator shows, this is correct:
(144) {d: ¬(Paul worked d-hard)}⊂{d: ¬(John worked d-hard)}31
Parallel to what we saw in (143), it is the proper subset relation here that is responsible for ruling out the possibility that both participants worked equally hard. This is a striking point in favour of this analysis. By switching to a semantics based on the proper subset relation rather than on maximality, we predict not only that IAD should be possible across negation, but also that it triggers a slight reinforcing of the truth-conditions, adequately ruling out that the set of degrees expressed by the matrix could be the same as that expressed by the than-phrase. However, a major point against this is that it has nothing to say about the emer- gence of an evaluative reading when IAD crosses over negation. That is, the change in the semantics of the degree operator only answers one of the two questions raised by (135), and I see no plausible way to amend it to answer the second question. As such, I examine an alternative approach which focuses specifically on accounting for the evaluative reading. In order to account for the appearance of evaluativity in (138), I would like to suggest that despite the surface word order, the negation that we see in this sentence, namely pas, is not genuine sentential negation, but is interpreted as constituent nega- tion on the adverb, and that the constituent made up of the adverb and negation
31Exceptionally here, I use English as my metalanguage instead of predicate logic to avoid com- plications concerning the representation of an adverb like fort ‘hard’.
62 is interpreted as a negative gradable adverb, basically what would be the antonym of fort if French had one. In French, the negative particle pas sometimes appears directly before gradable adjectives, as well as many other constituents. In an example like (145) from (Vinet (1995)) or (146), the negation unambiguously combines with the adjective in the surface syntax32
(145) J’ ai conclu une pas si vilaine affaire I have concluded a not so nasty deal ‘I have reached not such a bad deal’
(146) Elle a un pas gros chien. She has a not big dog. ‘She has a dog that’s not big.’
What these examples show is that we need to interpret constituent negation as operating directly on the adjective itself in examples like (146), and thus that we have to give it a different meaning from that of the logical connective. I propose that what this pas does is turn a gradable predicate into its negative antonym. Since natural language does not always provide an antonym for every gradable predicate (indeed, thereisnonefortheadverbfort), for any positive gradable adjective P, I will simply mark its antonym as P*.
(147) [[ pasDEG ]] = λP
As I have defined degree negation in (147), the denotation of any constituent of the form pasNEG A will have the exact same denotation as the antonym of A, if any. This accounts for the fact that pas grand ‘not tall’ is synonymous with petit ‘small’. Of course, this means that we now need a theory of the relation between positive and negative gradable adjectives. It has been claimed by many authors (cf. Rett (2011) and references within) that pairs of gradable adjective antonyms like tall and
32Admittedly, these examples are harder to find than they should be. Most cases of pas appearing with adjectives involve postnominal adjectives, such as in the following example:
(1) Jean s’ est achet´e un ordinateur pas cher. ‘John bought himself a computer that’s not expensive.’
However, such examples can be reanalysed as involving a reduced relative clause (cf. Cinque (2010) and references therein), in which case the negation may turn out to be be part of the sentential syntax of that clause, and not be constituent negation at all. I do not have an explanation for the scarcity of unambiguous cases like (145) and (146).
63 short differ in involving reversed scales. That is, if John is 6’, then short relates John to all the degrees of height from 6’ and up, while tall maps him to all the degrees of height from zero to 6’. Furthermore, 6’ is the maximal degree in both these sets, since the scale in short is the reverse of that in tall.Ontheshort scale, 6’ is higher than 7’. This is intuitively appealing, and we can show that it is on the right track to explain the surprising grammaticality of IAD over negation (138), as well as its evaluative interpretation. Evaluativity is a typical feature of comparison using negative gradable adjectives, especially with the equative as and with how-questions, which is not shared with comparison using positive adjectives (at least, not obligatorily)33.
(148) John is as tall as Bill. John is tall. (149) John is as short as Bill. → John is short. (150) How tall is John? John is tall. (151) How short is John? → John is short.
While we cannot conclude from (148) that John or Bill are actually tall, we can safely conclude from (149) that they are both short. Similar judgements arise from the degree questions in (150) and (151). So we know that negative gradable adjectives give rise to evaluative readings more easily than positive gradable adjectives, and I have suggested that negation in the problematic IAD sentence (138) is actually interpreted as turning the gradable adjec- tive into its negative antonym. How does this help make this sentence grammatical? Negative gradable adjectives have a scale that is reversed compared to their positive antonym. This means that while there is no such a thing as the maximal degree to which John didn’t work d-hard (with sentential negation), there is a maximal degree to which John worked d-hard*.
(152) max{d: Paul worked d-hard*} > max{d: Jean worked d-hard*}
33As Rett (2011) discusses, evaluative readings are obligatory with negative adjectives in the equative, but they are usually optional with comparatives. I do not know why the evaluative reading is obligatory in the example that concerns us (138), despite the fact that the degree operator is clearly the comparative, and not the equative. Nevertheless, the fact that an evaluative reading does emerge is consistent with treating the negation as reversing the scale of the adverb and producing a negative gradable adverb. Note that as Rett herself admits, some gradable adjectives like dirty trigger evaluative interpretations even in the comparative:
(1) This glass is dirtier than that glass. (Rett (2011)’s (40b))
64 The important feature of this formula is that it is interpretable because it does not contain any undefined maximum. The truth-conditions that this represents are actually almost equivalent to those corresponding to the low scope of the degree operator if pas is interpreted as sentential negation, except in that the latter makes the situation where the two individuals worked equally hard true. This reading is given in (153):
(153) ¬(max{d: Paul worked d-hard} > max{d: Jean worked d-hard})
This analysis thus makes it possible to account for the fact that IAD is possible across apparent sentential negation, and provides at least a direction of analysis for why such sentences get an evaluative reading. This view is also supported by the fact that we can actually alter our example (138) by putting the negation before the adverb, and the result if, as far as I can tell, exactly synonymous with the original sentence:
(154) ?Jean a plus travaill´e pas fort que moi. Jean has more worked not hard than than me. ‘Jean worked (even) less than I did.’
I do not have an explanation for why it is that apparent sentential negation has the option to be interpreted as constituent negation in IAD sentences, but I believe that the fact that evaluative readings result from such examples, coupled with the assumption that constituent negation on a gradable adjective reverses its scale, serves to show that it is right, even if we do not know why. IAD sentences with intensifiers like tellement ‘so’, rather than comparatives, also tend to confirm this. Assuming as I have done so far that the interpretation of intensifiers does not involve maximization, they are expected to be fine with negation, and indeed they are:
(155) Paul a tellement pas travaill´e fort aujourd’hui. Paul has so not worked hard today ‘Paul worked so little today.’
The intensifying effect that tellement has in this sentence clearly bears on the reverse of the adverb fort, constructing the meaning that Paul worked very little, and not that it is not the case that he worked very hard.
65 This analysis of the phenomenon makes it possible to maintain the view that IAD is overt DegP movement, despite the fact that it can cross over negation, an operation that it is not expected to be allowed to do, since it actually only crosses over something that is interpreted as constituent negation, not sentential negation. Heim (2001) claims that either syntax or semantics must block the emergence of readings corresponding to an LF where a DegP takes scope over sentential negation, as we have discussed, but also inherently negative verbs such as refuser, ‘to refuse’. The explanation that she suggests here is the same as with negation, namely that the resulting semantics would contain reference to an undefined maximum. Using a semantics for refuse that states that its complement clause is false in all accessible worlds, she provides the following example, whose only available meaning is given in (157), since the meaning associated with the high scope of DegP over refuse contains an undefined maximum (158):
(156) (Mary works 60 hours a week, and) she refuses to work harder than that.
(157) ∀w∈Acc [ ¬max{d: Mary worksw d-hard} > max{d: d=60 hours a week}]
(158) #max{d: ∀w∈Acc [¬Mary worksw d-hard ]} > max{d: d=60 hours a week}
Corresponding IAD sentences are expected to be unacceptable if the degree oper- ator has raised above the negative verb, and the structure should not have the option of being saved in any way parallel to the case of sentential negation above. This prediction is correct, and IAD is simply disallowed over refuser, as shown in (159)34:
(159) *Marc a plus refus´e de travailler fort que ¸ca. Marc has more refused to work hard than that. ‘John refused to work harder than that.’
What we have seen in this section is that while IAD is possible over negation, we can show that this negation obligatorily gets interpreted as constituent negation,
34This sentence does have a marginally available reading under which it is grammatical, namely if plus is understood as applying on the intensity of the refusal. It could be used to describe a situation where Marc has refused very vehemently to work hard, say. This reading does not correspond to that of an IAD sentence, however, but merely to a derivation where more relates to a degree variable associated with the verb itself, just as it would in a sentence like I loved her more than you did.I have avoided this issue so far by choosing verbs that do not lend themselves very well to a gradable interpretation, but this factor becomes harder to control for when we restrict ourselves to inherently negative verbs like we do in this section.
66 rather than sentential negation, and in this case the resulting semantics does not contain reference to any undefined maximum. The data in (138) thus does not argue against viewing IAD as overt DegP movement. When we turn to negative verbs, in contrast, the result is quite straightforward and IAD is simply blocked from placing a degree operator over it. This is exactly as expected since the resulting semantic formula would be ill-formed and no alternative derivation can save the structure, unlike in the case of sentential/constituent negation.
4.6 Superlative Scope or Contextual Restrictions?
There is a well-known ambiguity in superlative sentences between the so-called ‘ab- solute’ and ‘comparative’ readings35 (Heim (1999), Szabolcsi (1986), Farkas and Kiss (2000), Sharvit and Stateva (2002)).
(160) John climbed the highest mountain.
The sentence in (160) could be used to make the claim that John climbed the mountain that is simply the highest in the world, Mount Everest, corresponding to the absolute reading. It could also be used to make the claim that compared to a relevant set of mountain-climbers, John climbed a mountain that is higher than the ones climbed by any one of them, corresponding to the comparative reading. Heim (1999) focuses on this ambiguity and shows that it could be accounted for either by appealing to a context-dependent restriction on the comparison set implied by the use of the superlative, or by an ambiguity at the level of LF scope. Under the first approach, the two readings result from either considering the set of all mountains in the world, or simply the set of all mountains climbed by a relevant climber. This is a very natural assumption to make, since it is obvious that universal quantification in natural language is most naturally interpreted as restricted to some relevant domain. This is a largely recognized phenomenon, as discussed in Westerstahl (1985), Von Fin- tel (1994), Stanley and Szab´o (2000), among others. For simplicity, we can represent contextual restrictions as an argument in the semantics of universal quantifiers and saturate it with an unpronounced argument present in the syntax, as suggested by Von Fintel (1994). Modifying the denotation for the superlative morpheme given in section 3.3 in this way, we get the following denotation (adapted from Heim (1999)’s example (10)):
35Heim (1999) credits this observation to Ross (1964).
67 (161) [[ -est ]] = λC
68 (164) John slept the most.
Additionally, it would be very surprising if such DPs were true definites, since the scope solution requires the superlative morpheme to escape the DP, and definite DPs are well-known syntactic islands and should not allow extraction. Both the scope and comparison class analyses of the comparative/superlative am- biguity are adequate for (160) and other simple sentences. I propose to use IAD to help tease the two approaches using IAD, since the two analyses differ in that only the scope solution involves a crucial use of DegP movement. As we have seen in the list in (16), IAD is possible with the superlative expressions le plus and le moins. Superlatives in canonical position in Standard French give rise to absolute/comparative ambiguities just as in English.
(165) Paul a grimp´e la plus grosse montagne. Paul has climbed the most big mountain ‘Paul climbed the biggest mountain’
In a manner very similar to previous examples, the IAD version of this sentence is disambiguated compared to the possibilities of its canonical counterpart. The only reading that is possible here is the comparative reading.
(166) C’est Paul qui a le plus grimp´e une grosse montagne. It-is Paul that has the most climbed a big mountain. ‘Paul climbed the biggest mountain.’ (comparative reading only)
There are a number of things to say about this structure, since it is not directly parallel to (165). However the first and most important observation is that if we are right in treating IAD as the overt counterpart of the DegP movement that takes place at LF in English, and if comparative readings of superlatives are derived by long DegP movement as in the scope solution presented above, then we expect exactly this movement to be possible, and we expect it to disambiguate the sentence in exactly this way. If le plus has moved from its canonical position and raised to the top of VP, then the gradable property that is formed by lambda-abstraction over its trace should be exactly the same as that hypothesized by Heim (1999) for the comparative reading under a scope analysis, which is given above in (162). Admittedly, this is only circumstantial evidence in favour of the scope analysis of the superlative ambiguity. The superlative morpheme moves exactly where we
69 would expect it to in order to yield the comparative meaning under this approach, but there is nothing in the contextual restriction approach that should forbid the morpheme from moving. Indeed, the revised denotation with a contextual parameter (161) would produce the comparative reading if it moved this way. It should just not have to. So far we have not demonstrated with any certainty that either analysis is superior to the other in accounting for IAD sentences. Below we examine the other peculiarities of (166) and find some additional evidence in favour of the movement solution. An odd fact about (166) is that it is only grammatical with the cleft construction shown in (166). If we try to simply alter (165) by moving the superlative morpheme and leave the subject where it is, the result is degraded:
(167) ?? Paul a le plus grimp´e une grosse montagne. Paul has the most climbed a big mountain. This is interesting given the observation that focus correlates with the comparative readings in a truth-conditional way. Heim (1999) gives the following example:
(168) John put the tallest plant on the table.
This sentence could have different interpretations depending on where focus lands. If the subject John is focused, then the sentence gets a comparative reading that corresponds to comparing people with respect to the height of the plant that they put on the table. The sentence would be true iff of all the people who put plants on the table, John put the highest one there. If the focus lands on table, then the comparative reading concerns various places where John put plants. This sentence would be true iff of all the places where John put plants, it is on the table that he put the tallest one. In the superlative IAD sentence, cleft is obligatory, and the interpretation of clefts is very similar to that of focus. Indeed, the English sentence (168) can be disam- biguated by focusing some constituent as Heim discusses, or alternatively by con- structing cleft sentences. That is, the disambiguating effect is the same if we focus a constituent, or if we put it in a cleft at the beginning of the sentence.
(169) It’s on the table that John put the tallest plant. (170) It’s John who put the tallest plant on the table.
This effect is the same in French, as the following examples show:
70 (171) C’est sur la table que Paul a mis la plus grosse plante. it’s on the table that Paul has put the most big plant ‘It’s on the table that Paul put the tallest plant.’
(172) C’est Paul qui a mis la plus grosse plante sur la table. it’s Paul that has put the most big plant on the table ‘It’s Paul who put the tallest plant on the table.’
It is easy to construct a derivation that accounts for this effect of the cleft. Assum- ing an LF where the superlative morpheme is interpreted below the cleft and above the subject, and assuming that the clefted element is moved from the position where it appears in (168). Then the constituent with which the superlative morpheme will merge will be a different gradable property depending on which element was clefted. In (169) the second object of put is clefted, so the relevant gradable property is (173)36. This is a relation between places where John has put a plant and the height of the plant that he put there. If the subject is clefted as in (170), then the result is the gradable property in (174), a relation between individuals who put a plant on the table, and the height of the plant they put there.
(173) λdλx. ∃y [plant(y) & put(j, y, x) & tall(d, y) ]. (174) λdλx. x put a d-tall plant on the table.
What this shows is that clefting can be used to create gradable properties that would otherwise not correspond to any constituent of the sentence, at LF or at the sur- face. Such constructions of course always correspond to various comparative readings since, in a scope analysis, the absolute reading results from the superlative morpheme combining with a gradable property that only contains the gradable adjective and the head noun. The cleft that we see in the superlative IAD sentences (166) is in the right po- sition to yield the comparative reading that we do get (in fact, this sentence is not ambiguous and doesn’t have any other comparative readings). What is puzzling is the obligatory character of this structure. If all that mattered for the derivation was that the superlative morpheme get to combine with a node whose denotation is a relation between individuals and the height of the mountain they climbed, then the
36I simplify the denotation somewhat by entirely ignoring the preposition and treating the loca- tive as an individual argument of put. I believe this is inconsequential here, and it simplifies the presentation.
71 non-clefted IAD construction where the morpheme is between the subject and the verb (167) should be acceptable. This is because the full VP without the subject should denote precisely that. A very technical solution would be to appeal to the VP-internal Subject Hypothesis (Koopman and Sportiche (1991)), which states that subjects are generated inside the VP and raised to the specifier of (say) TP at the surface. Assuming as we have done so far that movement of a DP leaves a type-e trace and triggers lambda-abstraction over this variable directly below the landing site, `a la Heim and Kratzer (1998), the superlative morpheme would need to insert itself be- tween the moved subject and the abstractor directly below it. However, the abstractor is generally assumed to be directly tied to the moved constituent. Indeed, it is the movement operation itself that triggers the appearance of the abstractor. I believe it is not unreasonable to assume that it would not be possible for the superlative mor- pheme to “tuck in”, as it were, between the raised subject and its lambda-abstractor. That is, adding the abstractor nodes that we have been omitting so far, (167) would require the following LF to derive the comparative reading:
72 (175) IP
NP2 I
Paul a VP
DegP1 VP
le plus 1 VP
2VP
t2 VP
V DP
grimp´e une NP
AP N
t1 Amontagne
grosse The problem with this structure is that the DegP has moved in between the raised subject and its abstractor, marked as 2 in the tree. On the reasonable assumption that this is not an available position for movement (say, given the Extension Condition, Chomsky (1995)), this is not an available LF, and the sentence is judged unacceptable because of this37. If the cleft construction is base-generated, in contrast, then we expect that it should be possible for the superlative morpheme to come in between the clefted subject and the rest of the clause, making the clefted sentence in (166) acceptable. One last interesting thing to note here is that, assuming that the use of the superla- tive in IAD is a true marker of the raising of the morpheme, then we have confirmation that the superlative DP is indeed interpreted as an indefinite, since we literally see 37Of course, this begs the question as to why the corresponding English sentence is grammatical. For the moment, I have no explanation for why this would be different in French than in English.
73 the indefinite determiner emerge in (166). Since this sentence is synonymous to the comparative reading of its English counterpart, and it is the result of doing in overt syntax what is done at LF in English, then it is evidence of an additional kind for the scope solution, since the shift from definite to indefinite is otherwise somewhat surprising.
4.7 Scope Data Summary
Here, I summarize the similarities and discrepancies that were found between the scope possibilities of degree operators as described in the literature and overt distributional options of degree operators in IAD.
• The behaviour of IAD quantifiers with respect to quantified DPs is exactly as we would expect if Kennedy’s Generalization is a syntactic phenomenon. As far as I know, there is no known case where DegP and DP scope interact in a way that creates scope ambiguities and allows inverse scope interpretations. Similarly, IAD never allows a degree operator to relate to a gradable predicate if a QP is in the way.
• IAD can raise a degree operator over a modal or intensional verb in exactly the cases where scope ambiguities emerge in the canonical order, for a large number of verbs that we have examined.
• Grammatical cases of IAD over apparent sentential negation seem to contradict this hyopthesis, but it can be shown that such sentences actually involve hidden constituent negation, which is expected to be compatible with DegP movement. Inherently negative verbs, in contrast, behave as expected and block IAD.
• IAD offers distributional support for the scope analysis of the comparative read- ing of superlatives, since the superlative morpheme can be placed overtly at the beginning of the VP in QF and forces the comparative reading when it is. Its overt behaviour thus mimics the behaviour that this approach suggests it has at LF in canonical sentences.
74 4.8 IAD and Ellipsis
4.8.1 DegP Scope and Ellipsis
We have mentioned in Section 2 that the derivation of a than-constituent in the comparative involves some amount of ellipsis. For example, in (176), some clausal material as well as movement of an empty operator must be present at LF for the semantics we have provided to go smoothly, even though it is absent at the surface.
(176) John is heavier than Opk Mary is tk-heavy.
The exact nature of this ellipsis phenomenon is still a debated subject and it is unclear even whether all comparative constructions involve the same type of ellipsis, or even if they all involve ellipsis at all. Lechner (2004) distinguishes between three types of comparatives when it comes to their ellipsis properties: phrasal comparatives, which only contain a DP under than, which is then called the Remnant DP, clausal comparatives, in which only the VP is absent, and subcomparatives,whereonly the comparative morpheme is missing. These three types are exemplified respectively in (177), (178) and (179).
(177) Mark is crazier than Bill. (178) Mark is crazier than Bill is. (179) Mark is crazier than Bill is useful.
As shown at least as early as in Williams (1977), comparative ellipsis cannot look outside of the sentence to fill in the missing material, unlike the better known cases of VP-deletion. In (180), the missing material after the verb saw cannot be filled in by the object cows from the previous sentence. It can only be construed as the object from the main clause of the B-sentence, i.e. horses, and this is true whether the sentence is a phrasal comparative (without the verb saw)oraclausalone(with the verb).
(180) A: Did John see cows? B: Yes, but Sam saw more horses than John (saw).
In contrast, in a case of VP-ellipsis like (181), B’s reply contains an ellipsis site whose value is provided by material in the discourse, but outside of the sentence itself.
75 (181) A: Did John see cows? B: Yes, he did.
What this tells us is that comparative ellipsis is sentence-level ellipsis, rather than discourse-level. As such, it must find its antecedent structurally close to the gap. The remaining question is how exactly is this constituent identified. Most current analyses of ellipsis accept some version of Sag (1976) and Williams (1977)’s Parallelism Condition, which states that a constituent may be elided iff a semantically parallel antecedent constituent can be found. This requirement is usually taken to be a restriction on LF. As pointed out in Wold (1995), DPs that contain comparatives license Antecedent- Contained-Deletion (ACD). In an example such as (182), the material that must be reconstructed after Bill can only be found in the matrix VP, which also contains the ellipsis site. This creates the familiar situation where the ellipsis apparently cannot be resolved because the constituent that should serve as the antecedent contains the ellipsis site itself.
(182) John climbed more trees than Bill did.38
This would only a problem if we were to assume that the comparative operator and the than-phrase are actually part of the VP at the level of representation where the Parallelism Condition is verified. Given our assumptions so far that this condition is verified at LF, and that the degree operator along with the than-phrase move to the top of the clause for interpretation, then the VP can now be claimed to contain the right information to serve as antecedent for the ellipsis.
(183) [-er than Bill e ]k John [VP climbed tk-many trees].
Infinite regress is avoided because the DegP [-er than Bill e] is moved out of the way at LF, creating the VP [climbed t-many trees] that serves as the antecedent for the ellipsis site in the than-phrase. Wold takes this to be straightforward evidence in favour of the existence of DegP movement.
38For simplicity, I will treat comparatives that combine with nominals as involving a silent many (cf. Bresnan (1973)). Silent many is simply an ordinary gradable adjective that denotes a relation between (plural) individuals and the degree that represents their number of subparts. This makes the structure of more trees exactly parallel to that of taller trees, for example. See Hackl (2000) for a more refined version of this idea according to which silent many is rather a “gradable determiner”.
76 This establishes a relation between DegP movement and ellipsis. More precisely, since the antecedent must not contain the DegP itself in order to avoid infinite regress, then this means that the ellipsis site should only be allowed to be interpreted as parallel to a constituent that is smaller or equal to the scope of the moved DegP. Bhatt and Pancheva (2004) call this the Sag-Williams Ellipsis-Scope Generalization and define it in the following way:
(184) Sag-Williams Ellipsis-Scope Generalization The scope of a DegP containing elided material must contain the antecedent of the ellipsis.
Once again, we will use IAD to test this claim directly. What we expect to find is that the overt position of the degree operator should impose an upper bound on the elided material. The higher the operator is found, the more material should be allowed in the ellipsis site. What we will find instead is that the relation is even more stringent: the elided material must be exactly equal to the scope of DegP.
4.8.2 Fixing the Scope
In order to test this, we will use examples which are ambiguous in the amount of silent material that must be posited to the right of the remnant, such as (185):
(185) Jean a parl´e `a un plus grand professeur que moi. Jean has talked to a more great professor than me ‘Jean talked to a greater professor than I am/than I did.’
This sentence has two readings. On the first reading, the sentence says that Jean and me both talked to some professor, but the one that Jean talked to was greater than the one that I talked to. Under the second reading, it says that the professor that Jean talked to is a greater professor than I am. The fact that I must also be a professor is presupposed. Given our assumptions so far, we can say that these two interpretations correspond to different possibilities in the amount of material present in the ellipsis site. We will see later that appealing to ellipsis size is not the only way to get this result, and consider an alternative using the so-called Direct Analysis of phrasal comparatives. The first reading obtains if the ellipsis is interpreted as the same as the matrix VP, while the second reading results from the ellipsis site being only interpreted as the
77 matrix object DP. The following paraphrases illustrate this (the elided constituent is bolded):
(186) Jean a parl´e`a un plus grand professeur que Opk moi [VP ai parl´e`aun
tk-grand professeur].
(187) Jean a parl´e`a un plus grand professeur que Opk moi [NP un tk-grand professeur]39. These two readings are distinguished by the amount of covert material that must be present at LF in the ellipsis site, and this elided material tells us something about DegP movement. In order to license the ellipsis in (187), movement must have taken the DegP at least as high as the top of the matrix object DP (recall that we have claimed in section 3.1 that this is an interpretable position for the DegP, in order to accommodate short IAD). In order to license the larger ellipsis in (186), however, the DegP must have moved higher, at least as high as the top of the matrix VP. If it has moved this high, then it should also be possible to license the smaller ellipsis in (187), since the parallelism required for ellipsis is satisfied in this case also. In fact, given these assumptions, there is no way to tell if covert DegP movement has stopped at the top of DP or if it has gone farther up to the top of VP (or whether both derivations are possible) when the ellipsis corresponds to only the matrix DP. We can now use IAD to fix the LF position of the degree operator by moving it overtly and forming a pair of sentences, one short IAD sentence and one long IAD sentence. In (188), the degree operator has been moved to the left edge of DP, while in (189), it has moved all the way up to the top of VP40. (188) Jean a parl´e `a plus un grand professeur que moi. Jean has talked to more a great professor than me
39The than-phrase is not a syntactically well-formed sentence the way I have represented it here, since it would require a copula verb to make it acceptable, and yet none is available in the antecedent. I will assume that this copula is not necessary at LF, since its absence does not affect the truth- conditions of the sentence. I thus treat the requirement for there to be a copula in ordinary predicative sentences in French as a purely surface phenomenon that does not carry over at LF. This seems difficult to avoid, since there is clearly no copula in the antecedent (or indeed anywhere in the sentence), and one would clearly be necessary to yield a grammatical clause. I see no alternative to generate this meaning inside the ellipsis theory we are contemplating here. 40These sentences are not judged as equally acceptable by all my consultants, the short IAD sentence being unacceptable to a surprisingly large number of speakers, although I find them fine myself. This is at least consistent with the observation made earlier that short IAD is a more marginal phenomenon than long IAD, although I have no explanation for this fact. What is consistent, however, is that for all those who accept both sentences, they are unambiguous and only get the reading that corresponds to the translation I give below them.
78 ‘Jean talked to a greater professor than I am.’
(189) Jean a plus parl´e `a un grand professeur que moi. Jean has more talked to a great professor than me ‘Jean talked to a greater professor than I did.’
Both sentences are disambiguated compared to the canonical sentence in (185) in the way expressed in the translations. Starting with (188), we can see that when the degree operator is found at the left edge of DP, only the interpretation corresponding to the short ellipsis is available. This is expected under the assumptions that IAD is overt DegP movement and that DegP movement sets an upper bound on the size of the ellipsis site, since the other candidate interpretation can only be obtained if the full matrix VP serves as the antecedent for the ellipsis, and this VP still contains the DegP in (188), creating a structure of antecedent-containment which makes this ellipsis impossible. As such, the single interpretation available to (188) is consistent with the expectations laid out so far. Turning to (189), the fact that it allows the interpretation corresponding to the large ellipsis is unsurprising. This interpretation obtains when the matrix VP parl´e `a un t-grand professeur serves as the antecedent for the ellipsis site. This is not problematic since the DegP has moved out of this constituent, creating precisely the right kind of antecedent. The more interesting question is why this sentence does not also allow the short ellipsis interpretation. Looking at (189), we can see that the matrix clause does contain a suitable antecedent DP to license the small ellipsis, in fact exactly the same DP as in (188). A priori, there seems to be no reason why this interpretation should not be available, since according to the Generalization laid out in (184), DegP movement should impose an upper bound on the amount of covert material found in the ellipsis site, but it should not require that the ellipsis site be interpreted as exactly the scope of DegP, as the data we have just seen seems to suggest. Before moving to an analysis of the mismatch between the predicted results and the interpretations available to French speakers, it is important to note that the correlation between DegP scope and ellipsis size that we expected to find given our theoretical assumptions, but did find not for IAD sentences, has been claimed to be empirically adequate, at least for English sentences. I will show that the data that has been claimed to show this can be reinterpreted in a manner that is consistent with the IAD data.
79 4.8.3 Does DegP Scope Only Limit Ellipsis Size?
Bhatt and Pancheva (2004) examine the following example, borrowed from Sag (1976):
(190) Mary’s father tells her to work harder than her boss does.
This sentence is claimed to have three available meanings.
(191) Mary’s father tells her: “Work harder than your boss.”
(192) Mary’s father tells her to work d1-hard, Mary’s boss tells her to work d2-hard,
and d1>d2
(193) Mary’s father tells Mary: work d1-hard; Mary’s boss works d2-hard; d1 > d2
These three readings correspond to the three following logical options at LF:
1. the DegP [more than her boss does e] raises to the top of the embedded clause and the ellipsis site corresponds to the embedded VP, i.e. work d-hard.
2. the DegP raises to the top of the matrix clause, and the ellipsis site corresponds to the full matrix VP tells her to work d-hard
3. the DegP raises to the top of the matrix clause, but the ellipsis site is still only parallel to the embedded VP.
The derivation in 3., claimed to be available in English by Bhatt & Pancheva, corresponds exactly to the derivation which appears to be unavailable in Quebec French IAD sentences, cf. (189), since here the ellipsis site is claimed to correspond to a subpart of the scope of DegP. Since my purpose in analysing IAD sentences is to make a contribution to our understanding of DegP movement in general and not only in French, I believe this discrepancy is important and must be addressed. It is not clear to me that this reading must necessarily result from the derivation that is at issue here. This interpretation corresponds descriptively to a comparison between the set of degrees defined as the opaque reading of Mary works d-hard and the transparent reading of Mary’s boss works d-hard 41. Putting this in possible-world semantics terms, this sentence compares the set of degrees to which Mary works d- hard in the worlds that are compatible with what her father tells her, with the set of
41I thank Keir Moulton for leading me in this research direction.
80 degrees such that her boss works d-hard in the real world. It is possible to model this interpretation without appealing to the derivation that Bhatt & Pancheva propose by making use of explicit world variables in the syntax. I am assuming a system which is essentially what is proposed in Percus (2000), which makes it possible for a constituent at LF to scope below an intensional operator and still get a de re interpretation. Under this view, intensional verbs like think and tell introduce world-pronouns (or situation pronouns, in Percus’ terms), which may be co-indexed with world variables lower in the structure to derive opaque readings. Transparent readings obtain when these variables are not co-indexed with the world-pronoun introduced by the inten- sional verb, but by the real-world-pronoun which is assumed to always be projected at the top of the main clause. For example, in the following sentence, Bill must believe that someone is a spy, but he does not have to think that that person is my brother.
42 (194) @ Bill believes-w1 that my brother-@ is a spy-w1.
If the world-variable contained in the DP my brother is co-indexed with the real- world-pronoun, then we get the interpretation that Bill believes that someone, who actually is my brother, is a spy, regardless of whether he knows that this person is my brother. If that world-variable is rather co-indexed with the one introduced by the intensional verb believe, then the reading is that Bill holds the belief that someone that he believes to be my brother is a spy. Under this reading, this sentence can be true even if that person is not actually my brother, and even if I have no brother, without resulting in a presupposition failure.
(195) @ Bill believes-w1 that my brother-w1 is a spy-w1.
This makes the relation between the scope of a constituent and its transpar- ent/opaque interpretation weaker: a constituent can only receive an opaque inter- pretation if it appears at LF below an intensional operator, but it may also receive a transparent interpretation in this position, as long as its world-pronoun is coindexed with the real-world pronoun. Turning back to our example, we can now use this mechanism to derive the read- ing that interests us in (193) without making reference to the derivation that seems impossible in French. Since the verb tell is intensional, we know that it introduces a world-pronoun. Additionally, we must assume that the embedded verb work contains some world-variable w’, just as the indefinite DP in the example above. 42The symbol @ stands for the actual world.
81 Suppose that the DegP raises only as high as the top of the embedded clause. Then we know that is can only have the form [-er than her boss works t-hard] because a larger ellipsis could not be licensed from this position. Since there are two instances of the verb work in this LF, and since both of them appear in the structure lower than the intensional verb and its world-pronoun, both of them can in principle be co-indexed either with the real-world-pronoun, or by the one introduced by tell.The tree in (197) gives the structure for the embedded clause given short DegP movement.
(196) @ Mary’s father tell-w her that IP
(197) IP
DegP IP
Deg CP NP IP
more than her boss PROMary IVP works-w t-hard V AdvP
work-w t Adv
hard
The reading that we are trying to derive corresponds to the assignments where the world-variable of the main verb work in Mary works d-hard is bound by the intensional verb world pronoun and the variable of work in Mary’s boss works d-hard is bound by the real-world-pronoun.
(198) @ Mary’s father tells-w1 her to work-w1 harder than her boss works-@ t- hard
Thus, the possible readings of (190) do not provide any evidence in favour of the view that comparative ellipsis may correspond to a piece of structure smaller than the scope of the degree operator, since the same reading can be constructed using other means which are independently motivated. This discussion leads me to believe that the Sag-Williams Ellipsis-Scope general- ization as formulated by Bhatt and Pancheva (2004) is too weak. It states that the
82 scope of DegP sets an upper bound of the possible size of the ellipsis in the than- clause, while the data from the IAD examples show that it should rather state that the size of the ellipsis size must be exactly equal to the scope of the DegP. Since I have shown that the kind of examples that have previously been claimed to show that a small ellipsis could correspond to a wide DegP scope can be analysed with other independently motivated means, I propose to strengthen the generalization:
(199) The scope of a DegP containing elided material must be exactly equal to the antecedent of the ellipsis.
How does this fit in with our previous IAD data? We have seen that the overt scope of the DegP in IAD sentences fixes the size of the ellipsis site unambiguously, rather than simply giving an upper bound to it. So far, this is consistent with the stronger generalization in (199). If the story laid out above within Percus’ semantics to explain the availability of an additional reading is right, then we expect that if we translate the sentence in (190) into French, and make in into an IAD sentence where DegP has scope only over the embedded clause, then it should be ambiguous between the two readings that I claim both correspond to a low scope of DegP.
(200) Le p`ere de Marie lui dit de plus travailler fort que son boss. the father of Marie her say to more work hard than her boss This sentence should have two readings incommonwithitsEnglishcounterpart, and only lack the reading corresponding to the high scope of DegP. These two readings are expected to be available for (200) because they both correspond to a low scope interpretation of the DegP.
(201) Mary’s father tells her: “Work harder than your boss.”
(202) Mary’s father tells Mary: work d1-hard; Mary’s boss works d2-hard; d1 > d2.
(203) #Mary’s father tells her to work d1-hard, Mary’s boss tells her to work d2-
hard, and d1>d2
The respective LFs of the meanings predicted to be available are only distinguished in the indexing of the world pronoun on the elided verb travaille in the than-phrase.
(204) @ Le p`ere de Marie lui dit-w1 de plus travailler-w1 fort que son boss travaille-
w1 t-fort.
83 (205) @ Le p`ere de Marie lui dit-w1 de plus travailler-w1 fort que son boss travaille- @ t-fort.
Surprisingly, this prediction is borne out. (200) does allow for both of these read- ings, and does not allow for the third reading corresponding to the high DegP scope. This shows that the reading that Bhatt and Pancheva (2004) claim results from a derivation whereby the ellipsis site is interpreted as a subconstituent of the scope of DegP can be derived even if we disallow precisely such derivations. The unique available meaning of the long IAD sentence (189) provided us with evidence that we should strengthen the Sag-Williams Ellipsis Generalization to the claim that the ellipsis site in comparatives should be interpreted as exactly the scope of DegP, rather than only being limited by it. This faced the problem that some readings had been claimed in the past to require a derivation involving an ellipsis site interpreted as a subconstituent of the scope of DegP, but (200) shows that this interpretation can be obtained even if DegP scopes low. The Sag-Williams Ellipsis-Scope Generalization as stated by Bhatt and Pancheva (2004) followed naturally given some fairly standard assumptions given how ellipsis and scope are usually understood, and given the semantics we have given the compar- ative operator. Its stronger version given in (199), however, is rather more mysterious. I believe the most straightforward way to derive the stronger generalization is to appeal to the so-called Direct Analysis of phrasal comparatives, which does not involve genuine syntactic ellipsis in the than-phrase, but rather what we will call “semantic ellipsis”, and from which the strong generalization follows directly.
4.8.4 Direct Analysis
Let us start with the question of whether we want to claim that phrasal comparatives involve any kind of ellipsis at all. So far, we have treated phrasal comparatives and clausal comparatives as semantically identical, being only distinguished by the amount of material deleted at PF. That is, (207) and (206) are claimed to have the exact same LF, and the only difference between the two is that the VP has been deleted in (206), while a larger constituent has been deleted in (207) (say, I’):
(206) John is taller than Bill is d-tall.
(207) John is taller than Bill is d-tall.
84 As discussed by Heim (1985) and Hoeksema (1983), there are reasons to believe that perhaps the difference between the two constructions go farther than that. Some notable differences between the two include an asymmetry in the possibility of ex- tracting from the than-phrase. This is entirely impossible in clausal comparatives, while the result is only somewhat marginal in phrasal comparatives (cf. Hankamer (1973)):
(208) *Who is John taller than t is? (209) ?Who is John taller than?
Similarly, reflexives bound by an element in the matrix are acceptable comple- ments to than in a phrasal comparative construction, but not in a clausal comparative (examples from Hoeksema (1983)):
(210) No man is stronger than himself. (211) *No man is stronger than himself is.
This leads Hankamer to claim that phrasal comparatives actually involve a differ- ent than from that of clausal comparatives. The former is the head of an ordinary prepositional phrase, followed by its ordinary nominal complement and no ellipsis at all, while the latter is of category C and introduces a reduced clause. Of course, if this is correct, then the quantificational semantics for the comparative that we have been defending in this chapter is not compatible with phrasal comparatives, only with clausal ones. One way to resolve this is to claim that English possesses two distinct lexical items pronounced as more, but with functionally very different semantics. One is the determiner-type more that we have discussed already, which we will call Quantifier -er, while a second homophonous more is available in the lexicon, which we will call the ‘Direct -er’. The analysis according to which there is only the Quantifier -er in English and even phrasal comparatives involve ellipsis is usually known as the Reduction Analysis, while the “Direct Analysis” makes use of the two homophonous degree operators -er 1 and -er 2. Direct -er, also known as 3-place -er, is an operator which takes as its arguments a gradable property and two individuals, and returns true iff the two individuals are mapped to (sets of) degrees that stand in the proper relation. Below is a candidate denotation, again using a maximality-based semantics43. 43This denotation is very similar in spirit to that used in Bhatt and Takahashi (2011). I use the maximality operator in this semantics to highlight the resemblance with the version of Quantifier -er
85 (212) [[ -er2 ]] = λP
Under most circumstances, Direct -er and Quantifier -er yield exactly the same truth-conditions, although they do so in very different ways and fit with very different LFs. In a simple comparative sentence like (213), 3-place -er is interpretable in situ and without any covert material after the remnant Bill.
(213) Lisa is taller than Bill. (214) Max{d: tall(d, b)} < Max{d: tall(d, l)}
Direct -er does not involve any syntactic ellipsis per se, but it does involve some sort of ‘semantic ellipsis’, in the words of Bhatt and Takahashi (2011) and Heim (2001), in the sense that the adjective-type argument is used twice in the semantics. That is, P appears twice in the function description in (212), applying once to one individual and once to the other. This is how we obtain the fact that both Lisa and Bill are compared with respect to the same gradable property, i.e. tallness in (213). In contrast, the Reduction analysis gets this same effect by having the gradable property appear twice in the syntax, hence syntactic ellipsis. A major difference between these two approaches is that semantic ellipsis absolutely forces the two individuals to be compared along the same property, while the Reduction analysis could allow more flexibility, provided that the right context to license different instances of ellipsis is present. To put this in different words, the Direct Analysis predicts that the scope of the degree operator fixes the interpretation of the Remnant by providing the gradable property, while in the Reduction Analysis, its scope puts a upper limit on the size of the gradable property, given the Parallelism Condition, but does not strictly define it. This property of Direct -er makes it perfectly well-adapted to derive the meanings of the IAD sentences that were problematic. These examples are repeated here:
(215) Jean a parl´e `a plus un grand professeur que moi. Jean has talked to more a great professor than me ‘Jean talked to a greater professor than I am.’
(216) Jean a plus parl´e `a un grand professeur que moi. Jean has more talked to a great professor than me that I have been using so far, just as they use existential quantification over degrees in both their direct and quantifier denotations. I also altered the order in which Direct -er takes its arguments to make the basic predicative cases interpretable in situ.
86 ‘Jean talked to a greater professor than I did.’
Suppose that French plus has the denotation of Direct -er given in (212). I will show that this assumption, coupled with the hypothesis that IAD is overt DegP movement, predicts exactly the available and unavailable interpretations for these sentences. Let us start with (215). The constituent that plus moves above is the following:
(217) DP
(un) NP
AP NP
tAN
grandprofesseur
Assuming as usual that this movement leaves a type-d trace that saturates the adjective’s degree argument, the NP can then be interpreted intersectively with the AP to create the set of things that are both professors and have some measure of greatness (i.e., are in the domain of great). In order for this semantics to work out compositionally, I have to assume that the indefinite determiner is semantically vacuous, `a la Heim (1982). Movement to the top of this constituent triggers lambda- abstraction over the degree variable denoted by the trace, producing type
(218) [[ NP ]] = λdλx. professor(x) & great(d, x)
This constituent is of the appropriate type to serve as the first argument for plus. Adding the first individual argument which is provided by the remnant moi,weget the whole constituent to denote the set of all individuals who are have the property in (218) to a greater degree than I do. This indefinite object can then be interpreted as the object of the verb parler by whatever means one prefers to interpret indefinites in object positions. So far, the Direct and Reduction analyses yield the same results, which correspond well to speakers’ intuitions. It is only when we examine the sentence in (216), where
87 the degree operator is in the left edge of VP, that we see a difference emerge. Let us examine this example in detail. Again, this is the constituent to which plus attaches.
(219) VP
V PP
parl´e(`a) DP
(un) NP
AP NP
tAN
grandprofesseur
The entire VP is interpreted exactly the same way as any other transitive VP and ends up denoting a simple set of individuals, i.e. those who have talked to a professor with some measure of greatness. Lambda-abstraction over the degree variable kicks in and produces a constituent of the proper type to serve as input for the degree operator, with the following denotation:
(220) [[ VP ]] = λdλx. ∃y.met(x, y) & professor(y) & great(d, y)
This serves as the P argument for the degree operator, while the Remnant moi and the subject Jean fill the two individual arguments. The resulting interpretation corresponds to the reading where both Jean and me met a professor and the one that Jean met was greater than the one I met. This is an adequate representation of this sentence’s meaning. Unlike the Reduction Analysis that we have been assuming so far, the Direct Analysis offers no way to derive any other meaning for (216), which is a welcome result since the sentence is unambiguous. The syntactic mechanisms used in the Direct Analysis produce an LF which is distinguished from the surface only by the application of one (relevant) operation, namely DegP movement. The Reduction analysis requires both degree QR and some amount of ellipsis (and also a null operator movement in the than-phrase). As such,
88 there is in a sense ‘more room’ for ambiguity since the input to semantic interpretation may vary in multiple ways for a single surface string. The data from IAD does not corroborate this, however, as it appears that if the scope of the degree operator can be fixed, i.e. by applying degree QR in overt syntax, then all ambiguity disappears. The data from IAD thus favours a Direct Analysis of Qu´ebec French phrasal comparatives over a Reduction analysis. In fact, we must adopt the stronger position that in the sentences we are exam- ining, the only possible derivation is the one just outlined, and that a Reduction analysis is not available there. This is not a trivial issue since French does not only allow phrasal comparatives, which can be interpreted using Direct -er, but also sub- comparatives, which are naturally analysed as involving Quantifier -er,andsimply cannot be interpreted with Direct -er in any way:
(221) Marc est plus gros qu’ il est grand. Marc is more big than he is tall ‘Marc is bigger than he is tall.’
The existence of such sentences in French is important because it shows that we cannot claim that this language only makes use of Direct -er. It is only possible to show this with subcomparatives because the other, more common type of comparative in English naturally analysed with Quantifier -er are clausal comparatives, but these do not exist in French. That is, there is no way for a than-constituent to have the form of a full clause without its VP. The closest grammatical construction one can build is one where the VP is replaced by a clitic, and a more or less dummy verb faire is used:
(222) Jean a vot´e lib´eral plus souvent que Marc ne l’ a fait. Jean has voted liberal more often than Marc ne it has done ‘Jean voted liberal more often than Marc did.’
(223) *Jean a vot´e lib´eral plus souvent que Marc (ne) a. Jean has voted liberal more often than Marc ne has Thus, while French does have a Quantifier -er in its lexicon, it cannot be used in phrasal comparatives, but only in subcomparative constructions. Presumably, this is because of a more general property of this language which disallows ellipsis in many cases where English allows it44. 44There is a broader issue at hand concerning the different ellipsis possibilities offered by French and English grammars. In addition to comparatives, other constructions such as before-phrases
89 This is a striking conclusion, since it has been claimed explicitly in Lechner (2004) and Bhatt and Takahashi (2011) that English cannot have anything like Direct -er, the exact opposite conclusion to what I have reached for Qu´ebec French. They make this claim on the basis of condition C effects that seem to show that even in phrasal comparatives, covert material parallel to the matrix clause must be present at LF, which is only required by Quantifier -er, not Direct -er. The following example is meant to show this:
(224) *More people introduced himk to Sally than to Peterk’s sister.
Coreference between the pronoun him and Peter is not possible here, which Bhatt & Takahashi attribute to a Condition C violation. We can show that the overt copy of him is not responsible for triggering this effect. The following example shows that the direct object position does not necessarily c-command the than-clause.
(225) Mary gave himk more presents than Johnk’s mother.
Here, coreference between him and John is possible, so the pronoun clearly does not c-command the proper name. So what could be blocking coreference in (224), then? Bhatt & Takahashi claim that it can only be some covert copy of him merged inside the than-phrase. They claim that the LF for (224) is (226)45, where the unpronounced copy of him does c-command the proper name:
(226) [ morej [Opi than ti-many people introduced himk to Peterk’s sister]] [tj-many
people introduced himk to Sally]
involve obviously elided constituents in English, but only allow DP complements or full clauses with a clitic in lieu of a gap in French.
(1) John met the president before me/before I did.
(2) Jean a rencontr´elepr´esident avant moi. ‘John met the president before me.’
(3) Jean a rencontr´elepr´esident avant que je (ne) le fasse. ‘John met the president before I did.’
I do not explore this broader issue here, but I believe that it shows that French has generally more restricted options than English concerning ellipsis. 45For consistency, I present this LF in the same way that I have done so far in this chapter, rather than how Bhatt & Takahashi do.
90 The configuration that causes a Condition-C violation is entirely inside the than- phrase. While the overt copy of him is not in a c-command relation with Peter, the covert copy inside the than-phrase is. If this is the actual LF, then we have an explanation for why (224) is bad. But if this is the right LF, then it must be that English only has Quantifier -er in its lexicon, otherwise Direct -er could have been merged instead, and then no covert material would have to be postulated, and no Condition C effect would have been expected to be triggered. Bhatt & Takahashi claim that this shows that a derivation involving Direct -er is impossible, since it would have rescued the sentence, contrary to fact. It is difficult to construct French examples that mirror this, since direct and indi- rect object pronouns get spelled out as clitics in this language, and their LF position is not as obvious as that of the ordinary pronouns in English. In the remainder of this section, I will show that this argument is not very strong in English either, thus making the claim that Direct -er is unavailable difficult to maintain for either language. First, as the authors themselves admit, these judgements are not very solid (I have been unable to replicate them myself). This is an important point, since it is the only kind of data that they claim actually rules out a Direct -er in English. Still, let us accept them here for the sake of argument. Even assuming that the data is what they say, more must be said in order for their claim that a Reduction analysis does account for the Condition C effect to hold. If we examine the LF that Bhatt & Takahashi postulate for (224) in some detail, we can see that it cannot be right, since the elided material in the than-phrase, namely the string t-many people introduced him, does not form a constituent. The standard assumption in a Reduction analysis of phrasal comparatives (see Merchant (2009) and references within) is that the Remnant must be raised to the top of the than-phrase to allow the ellipsis to take place, in a manner very reminiscent of sluicing. That this movement must take place is fairly uncontroversial, since it can be shown to trigger island effects. In both the examples below, respectively from Merchant (2009) and Heim (1985), the Remnant would have had to be extracted from a relative clause, and the result is correctly predicted to be ungrammatical:
(227) *More people live in the country that Putin governs than Bush. (228) *I spent more time with a woman who played the clarinet than the lute.
91 In (224), this means that the PP to Peter’s sister must have raised to the top of the than-phrase, crucially above the offending pronoun, in order for the elided material to actually form a constituent.
(229) ... Opj than [to Peterk’s sister]i tj-many people introduced himk ti
Once we take into account the fact that the remnant must move to license the ellipsis, then we lose the explanation for why a Condition C effect arises, since as can be seen in (229), the pronoun and proper name are actually not in a c-command relation. To summarize, it is difficult to decide whether English actually has or does not have a Direct -er. There are reasons to believe that phrasal and clausal comparatives involve very different structures and that a genuine Direct -er is necessary, as the facts concerning extraction from the than-phrase and Condition A licensing show. On the other hand, Bhatt & Takahashi’s data offers some inconclusive support in favour of the view that Quantifier -er is the only available comparative operator in English. Since their conclusion is exclusively supported by this Condition C effect, which speakers do not agree on, and the explanation for which does not appear to actually hold, I conclude that the evidence is not strong enough to rule out that English and French both do not make use of Direct -er.
4.9 DegP Movement and the Merger Position of Than- Phrases
So far, we have assumed that the rightmost position of the than-phrases results from a rightward extraposition rule that does not feed LF. This view has recently been chal- lenged on both conceptual and empirical grounds by Bhatt and Pancheva (2004). In order to provide a more explicit account of the syntax of the comparative construction, they borrow the analysis of extraposition proposed by Fox and Nissenbaum (1999). According to this view, extraposition is not (always) a type of rightward movement, but rather late merger of the extraposed constituent to a covertly raised element that right-attaches to its landing site. Fox & Nissenbaum thus analyse extraposition of adjuncts such as from the museum in (230) as involving an intricate sequence of overt and covert operations.
(230) We saw a painting yesterday from the museum.
92 First, the DP apaintingis merged in its normal argument position, without the adjunct PP. Since it is a quantified DP, it QRs to some type t node in order to be interpreted. Crucially, the authors claim that this movement adjoins to the right of its landing site, in contrast to the usual null-hypothesis that it left-adjoins to it. Finally, the adjunct PP is merged next to the covertly raised copy of the indefinite. The QRed DP is pronounced in its base position, while the adjunct PP is pronounced next to this DP’s moved position. The analysis entirely hinges on the proposal that QR adjoins to the right of its landing site. Typically, this would be a fairly inconsequential statement since it has no effect on word order, given that the movement is covert, or on interpretation, since semantic calculation is not usually assumed to be sensitive to branching direction, only to constituency, at least in the framework borrowed from Heim and Kratzer (1998) that we have been assuming. It is only when something is late-merged to the moved constituent that it has an effect on word order, leaving so-called extraposed constituents at the right edge of the clause and their host DP in their base position. Bhatt & Pancheva borrow this analysis and apply it directly to the comparative construction. That is, they claim that a typical comparative such as (231) is derived following the same steps as (230).
(231) John is more intelligent than Bill.
First, the degree operator more is merged in its surface position before the adjec- tive. Since it must move in order to be interpreted, it QRs to the nearest type t-node and right-adjoins to it. From this position, the than-constituent can be merged as the complement to the raised copy of more. The result of this is that the than-constituent ends up to the right edge of the clause, while still being in the complement position of the degree operator. Bhatt & Pancheva then provide arguments to tie up the loose ends of the analysis, such as the fact that it must allow for late-merger of a complement, in direct opposition to Lebeaux (1991)’s original formulation, and why late-merger should be obligatory in the comparative. We do not go into these details here (although see Grosu and Horvath (2006) for comments). Just as in Fox & Nissenbaum’s proposal, the entire validity of the analysis depends on the assumption that degree QR attaches the degree operator to the right of its landing site. As mentioned above, this is difficult to verify because this movement is
93 covert and this parameter is semantically irrelevant. Since we are assuming that IAD is an overt instance of exactly this movement, then we can use this construction to verify their claim directly. The conclusion that we will reach is that if we are correct in assuming that IAD is overt DegP movement, then DegP movement targets the leftward position of its landing site. This means that Bhatt & Pancheva’s analysis of the rightward position of the than-argument cannot be maintained. If the position of the degree operator in IAD is the result of overt degree QR and the position of the than-phrase is the result of late-merging it as the complement of the raised copy of the operator, then we would expect the degree quantifier to appear overtly to the right of the clause, directly before the than-phrase. However, this is not what we get. In an IAD sentence, the degree quantifier appears to the left of its canonical position. The result predicted by extending Bhatt & Pancheva’s analysis to French is not grammatical.
(232) *Jean a lu un long livre plus que moi. Jean has read a long book more than me Intended: ‘Jean read a longer book than I did.’
The fact that Qu´ebec French makes use of overt QR, or some QR-like operation, shows us that it targets a leftward position in this language, as is usually assumed. This conclusion can be imported to English by virtue of the fact that the same array of scope facts that drove the authors so propose their analysis for English are exactly parallel in French. Since the empirical ground to explain is the same in both languages, and we know that rightward covert movement of the degree quantifier cannot be right for French, then I conclude that it cannot either be right for English.
5 Asymmetries between DegP Movement and IAD
What we have done so far in this chapter is the following: we have found a striking resemblance between DegP movement and IAD, and showed that we can reasonably believe that they are one and the same phenomenon: IAD is simply the overt counter- part to covert DegP movement. We have then used this to test for various properties of DegP movement that could not as easily be observed, if at all, from its covert English counterpart.
94 However there are facts that we have not as yet examined and that cast doubts on the validity of the claim that the surface position of the degree operator in IAD sentences is the result of movement from a position next to some target gradable predicate. In this section, we examine those arguments in some detail. In section 5.1, we will see that despite Cyr’s claim, IAD actually does not obey island constraints quite as strictly as it should. Section 5.2 shows that is some cases, there is simply no likely source position for the degree operator, and 5.3 shows that differentials and factorials are excluded from IAD sentences, for no obvious reason. Finally, we will see that the surface syntactic component of the grammar does not require movement at all to derive IAD sentences, since all such sentences can be generated independently, as a careful observation of the identity of the class of IAD operators in section 5.4 will show. This characterization of the class of IAD operators is in fact more adequate than the semantic characterization based on the operators’ necessity for DegP movement. The difficulty lies in providing the right interpretation for these sentences, not in generating the strings themselves.
5.1 Absence of Island Effects
As I discussed at the beginning of this chapter, Cyr (1991)’s original motivation for deriving IAD sentences by overt movement of the degree operator is her claim that IAD is sensitive to syntactic islands, yielding ungrammatical sentences when the hypothesised source position appears in an environment from which movement should not be allowed to escape. However, it is not clear that this is entirely empirically adequate. As discussed in Bouchard et al. (2011), there are syntactic environments where IAD is possible even though syntactic movement would normally be prohibited. This is for example the case with some cases of adjunct islands:
(233) On a tellement travaill´e l`a-dessus pendant une belle journ´ee. We have so worked this-on during a nice day. ‘We worked on this during such a nice day.’
The PP pendant une belle journ´ee ‘during a nice day’ is an adjunct to the matrix clause, which should make it opaque to movement. It is shown in (235) that this PP does block wh-movement.
95 (234) On a travaill´e l`a-dessus pendant une belle journ´ee du mois de We have worked this-on during a nice day of-the month of juin. June ‘We worked on this during such a nice June day.’ (235) *De quel mois on a travaill´e l`a-dessus pendant une belle journ´ee? Of what month we have worked this-on during a nice day. Intended: ‘During a day of what month did we work on this?’ (233) should be ungrammatical given our assumptions so far. First, it is clear from the meaning of the sentence that tellement relates to the gradable adjective in the adjunct PP. We can show this by contrasting this sentence with another where the target adjective is removed. The result is not ungrammatical, but it is interpreted much differently, with tellement relating to the verb travailler ‘to work’. The glosses for (233) and (236) show the resulting difference in meaning. (236) On a tellement travaill´e l`a-dessus pendant une journ´ee de Juin. We have so worked this-on during a day of June. ‘We worked on this so much during some June day.’ If IAD is derived by DegP movement, then (233) should be derived by extraction of the degree operator from the adjunct and the sentence should not be grammatical. This is problematic for the view that IAD is derived by movement at all. What this shows is that if IAD is derived by movement, it is a type of movement that is freer than wh-movement. We have not yet shown that it is freer than DegP movement, since we have not seen any direct contrast where DegP movement as observed by scopal interpretations differs from the overt distribution of the degree operator in IAD sentences46. To do this, we examine the English sentence in (237), which contains a modal and the degree operator less in order to make it possible to tell the scope of DegP, and the degree operator is embedded during an adjunct during- PP. If the wide-scope reading of DegP over the modal is available, then we must conclude that it is DegP movement in general that is unusual in allowing extraction from adjuncts. If it is unavailable, then we must conclude that English covert DegP movement behaves as expected, but IAD is different. The latter turns out to be true, as the unavailability of (239) shows47: 46I thank Rick Nouwen for urging me to show this. 47As before, I turn to English as my metalanguage here to avoid issues about the formal repre- sentation of events.
96 (237) John is required to work during days (that are) less hot than this. (this = 25 degrees)
(238) ∀w∈Acc [max{d: John worksw during a d-hot day} < max{d: d=25 degrees}]
(239) #max{d: ∀w∈Acc [John worksw during a d-hot day]} < max{d: d=25 de- grees}
These two formulas are truth-conditionally distinct: On the narrow scope reading of DegP (238), the sentence says that John is not allowed to work on days where it is more than 25 degrees out. This is the only available meaning. On the wide scope reading described in (239), the sentence would say that John has some requirement that he must work on days that are hotter than some temperature, and that temper- ature, although unspecified here, is somewhere below 25 degrees. It would be made true for example if John had to work if it was 20 or more out. The last necessary step is to show that French and English do not differ here. This can be done simply by taking the French translation of (237) and examining its available interpretations. As it turns out, its only possible meaning is the one given above in (238), corresponding to the narrow scope reading of moins.
(240) Jean est oblig´e de travailler pendant des journ´ees qui sont moins Jean is required to work during of days that are less chaudes que ¸ca. (¸ca = 25 degrees) hot than this ‘John is required to work during days (that are) less hot than this.’
What we have found is a difference in behaviour between the observed scopal behaviour of DegP and the overt distribution of degree operators in IAD sentences. We find that sentences with the canonical word order in French and English have the same scope options, but that it is possible to find cases where IAD is possible even though neither language allows a wide-scope interpretation of the degree operator in the canonical word order. In other words, while covert DegP movement is restricted in a consistent manner in both languages, this movement is not quite as perfectly parallel to IAD as we have seen so far, casting doubt on the view that they are indeed one and the same thing. The grammaticality of IAD sentences like (233) goes contrary to Cyr’s claim that IAD is island-sensitive, the original motivation for her claim that it is derived by
97 movement of the degree operator. However, the data is trickier than it appears at first glance. She provides the sentence in (14), repeated here as (241), to illustrate her claim:
(241) *On a trop ni´e le fait qu’ elle ait con¸cu de grands projets. We have too denied the fact that she had conceived of great projects However, it is not obvious that this example alone suffices to establish that IAD is sensitive to syntactic islands. On the contrary, it is possible to construct examples with a similar structure (in the relevant sense) and yet the result becomes grammatical. In (242), the gradable adjective gros is embedded in a relative clause island, and the IAD constructions is grammatical nonetheless.
(242) Paul a tellement mari´e une fille qui avait un gros nez! Paul has so married a girl who had a big nose ‘Paul married a girl who had such a big nose.’
This contrast cannot be accounted for by appealing to any special property of the relative clause in (242) that would make it transparent to movement for some reason, since we can show that regular wh-movement is still blocked:
(243) *Comment est-ce que Paul a mari´e une fille qui avait un nez? How did that Paul has married a girl who had a nose
So what is the difference between (241) and (242) that makes the former ungram- matical and the later fine? I argue that that what makes (241) ungrammatical is not a violation of an island constraint, but rather a clash of registers in a single sentence. As Cyr herself notes, IAD is a construction that is only found in spoken Qu´ebec French and in very informal registers. In contrast, the direct object in the relative clause in (241) is introduced by the element de. Direct objects in de that are not licensed by negation or other relevant operators (see Heyd (2003) for details) are only possible in a very restricted set of contexts which mainly involve the necessity for a prenominal adjective and are unique to a very high register. An example of this use is given in (244):
(244) Nous avons vu de grands films. We have seen of great movies ‘We saw great movies.’
98 This sentence, although grammatical in Standard French, is extremely stilted, and would normally never be pronounced in everyday speech, since such DPs are normally only found in written language. It should not be controversial to argue that IAD and objects introduced by de are quite simply not part of the same dialect of French, explaining the ungrammaticality of (241)48. Basically, (241) does not show what it is supposed to show, namely that IAD is impossible through a relative clause boundary, since there are other confounding factors at play which contribute to its ungrammaticality, while other better controlled examples are grammatical. Above, we used scope data to show that DegP movement is constrained by adjunct- islands, while IAD is not. Below, we will use ellipsis size in the than-phrase to show that DegP movement is also constrained by relative clause islands, again unlike IAD. The following sentence contains a degree operator embedded inside a relative clause, and I will show that DegP movement cannot escape it.
(245) John married a girl who has a bigger family than Bill.
As discussed in Section 4.8, the size of the elided material that can fill the gap after Bill in the than-clause is limited by the scope of the degree operator. If the movement of the degree operator can escape the relative clause, then we expect (245) to be ambiguous. It should be able to compare either the number of siblings that Bill and the girl that John married have, as well as compare the number of siblings of the girl that John married and the girl that Bill married, respectively, corresponding to the two representations of the elided material below:
(246) John married a girl who has a bigger family than Bill has a d-big family. (247) John married a girl who has a bigger family than Bill married a girl who has a d-big family.
If the movement is restricted to the relative clause, then only the former should be acceptable, while both should be possible readings if DegP movement can escape
48This entire discussion about register may turn out to be superfluous because it is not the only thing that is wrong with this sentence. Indeed, in this case, even changing the object to one without the offending de does not make the sentence grammatical. One reasonable hypothesis about what makes it bad is that IAD is blocked in (241) because the verb nier ‘to deny’ is an inherently negative verb. As discussed in Section 4.5, such verbs have been argued by Heim (2001) to disallow inverse scope of degree operators, and shown to consequently block IAD. It is quite possible that there are multiple problems with this sentence, both a clash in register and a semantic incompatibility caused by movement over a negative verb.
99 the relative clause. My informants agree that only the former is available, from which I conclude that DegP movement in English comparatives is indeed bound by relative clause islands. What this discussion shows is that while some sentences where the alleged source position of the degree operator is inside an island are indeed ungrammatical, this can generally be explained by other properties of the sentence. The fact that at least some other sentences are fine despite the source position being embedded inside an island, in contrast, suffices to establish that IAD is not a perfectly parallel overt counterpart to covert DegP movement.
5.2 Absence of Source Position
A very simple argument against deriving IAD sentences by overt movement of the degree operator is that such sentences can be grammatical even though the alleged source position of the operator is not in fact a legitimate position for degree operators. One such example is (248):
(248) Marc a tellement fait une connerie! Marc has so did a foolishness ‘Marc did such a stupid thing!’
The target gradable predicate in (248) is the noun connerie (see Morzycki (2009) on gradable nouns), but nouns are not compatible with degree operators in prenominal or postnominal position (cf. (249)).
(249) *Marc a fait une (tellement) connerie (tellement)! Marc has did a so foolishness so Intended: ‘Marc did such a stupid thing!’
In fact, there is simply no position except for the IAD position where we can put the degree operator to yield a grammatical sentence. In order to express this without appealing to IAD, we must turn to size adjectives like big, which can themselves be modified by degree operators in the normal way.
(250) Marc a fait une tellement grosse connerie! Marc has did a so big foolishness ‘Marc did such a stupid thing!’
100 (248) should not be a grammatical sentence since there is no way to generate its source structure in the first place, unless we posit that movement should be obligatory when this structure is created. I see no reason to suppose that this should be the case.
5.3 Absence of Differentials and Factorials
One piece of data concerning the comparative operators that participate in IAD sen- tences that we have mostly ignored so far is their ability to introduce differential and factorial arguments. What we call differentials are the measure phrases that can ap- pear in comparatives and serve to quantify the difference between the two compared elements. The measure phrase four inches in (251) is an example of a differential.
(251) John is four inches taller than Bill.
Factorial arguments are similar to differentials, except that they essentially involve multiplication rather than addition. An example of this is the phrase three times in (252). While factorials can appear either with the equative as or the comparative in English, they can only appear with the comparatives in French:
(252) Your television is three times as/more expensive as/than mine.
(253) Ta t´el´evision est trois fois plus ch`ere que la mienne. Your television is three times more expensive than the mine ‘Your television is three times as expensive as mine.’
What does the overt DegP Movement hypothesis predict should happen if an IAD sentence is derived from a base structure that contains a differential argument? The predictions depend on the constituency that we assume for the base structure. If the differential is generated ”close enough” to -er, namely inside DegP, then we expect that QR could take it along to its higher position. If it is generated too far, then it should be forced to stay in its base position after DegP has moved. As it turns out, however, it is not necessary to pursue this in any detail since both predictions are wrong. Differentials are never acceptable in IAD sentences, neither stranded in the operator’s base position nor raised up together with the operator:
(254) Jean a ´ecrit un travail cinq pages (de) plus long que Al. Jean has written a paper five pages of more long than Al ‘Jean wrote a paper five pages longer than Al did.’
101 (255) *Jean a cinq pages (de) plus ´ecrit un long travail que Al. Jean has five pages of more written a long paper than Al
(256) *Jean a plus ´ecrit un (de) cinq pages long travail que Al. Jean has more written a of five pages long paper than Al At least one of those options should be grammatical. This is because DegP move- ment is just as necessary in comparative sentences with differentials as in those with- out. Consider Von Stechow (1984)’s analysis of differentials. Under his approach, comparatives always involve a differential argument, although it does not need to be pronounced. The semantics he gives to a simple comparative clause that does not contain an overt differential argument is something like (258). This simply says that there is some non-null degree of difference between John’s height and Bill’s.
(257) John is taller than Bill.
(258) For some degree d1>0, John is d1+d2-tall, where d2 is the d s.t. Bill is d-tall
We can easily recast this in the terms of the maximality-based semantics we have been using so far in the following way:
(259) ∃d: d=0. max{d: tall(d, b)} +d=max{d: tall(d, j)}
If the value of d1 is provided overtly, as in (251) by the measure phrase four inches, then we get the following truth-conditions, which really only differs from (259) in that the value of d is specified to be exactly four inches, rather than simply said to be non- null.
(260) max{d: tall(d, b)} +4’=max{d: tall(d, j)}
Sentences with differentials and those without them are structurally isomorphic at LF. When no overt differential is present, a covert one must be assumed to be present, which means that sentences with differentials and those without it should be subject to exactly the same rule of degree QR introduced in section 2. Since differentials do not interfere with DegP movement in general, there is no reason why they should not be compatible with IAD. The overt DegP movement hypothesis thus overgenerates in allowing the derivation of sentences such as (255) or (256). I see no non-ad hoc way to block the generation of such sentences.
102 The same situation obtains with factorial arguments. There is still no possible position for this constituent in IAD sentences. The result is ungrammatical if we try to pied-pipe it along with the degree operator, as well as if we attempt to leave it lower in its source position:
(261) J’ai achet´e une t´el´evision trois fois plus ch`ere que toi. I-have bought a television three times more expensive than you ‘I bought a three times more expensive television than you did.’
(262) *J’ai trois fois plus achet´e une t´el´evision ch`ere que toi. I-have three times more bought a television expensive than you
(263) *J’ai plus achet´e une t´el´evision trois fois ch`ere que toi. I-have more bought a television three times expensive than you Again, there is no reason why a factor argument should have this effect. Von Ste- chow (1984) and Bhatt and Pancheva (2007) both agree that the absence of an overt factor argument is simply synonymous with the presence of a covert one time argu- ment. The overall semantics is the same in both cases. If we assume that equatives involve DegP movement that is necessary for interpretation, then it will be neces- sary regardless of the presence or absence of an overt functor argument. The overt DegP movement hypothesis for IAD thus also overgenerates in the case of functor arguments, and no natural explanation for this follows either from the syntax or semantics of these constructions.
5.4 The Class of IAD Operators
So far, we have seen reasons why deriving IAD by overt DegP movement undergener- ates in that it wrongly predicts that some good sentences should be ungrammatical, namely when the source position is embedded inside an island and when the source position is not a legitimate position for the degree operator. We have also seen that this approach overgenerates in not providing us any reason to block those examples that contain differential and factorial arguments. As mentioned already in section 3.1, we also know that the position that IAD operators appear in is one that is indepen- dently available for the lexical items anyway, showing that movement is not necessary to generate the word order in IAD sentences. If the previous arguments are taken seriously and IAD sentences are not derived by overt movement of the degree operator, then we are forced to accept that they are
103 simply inserted in this position in the first place. This does not require any extension of the syntactic component since every single lexical item that participates in IAD can independently be generated in the adverbial position, where they appear in long IAD sentences, as well as in the left edge of DP, where they appear in short IAD sentences. In the former case, they get interpreted as quantifying over the event variable of the VP and in the latter, they get interpreted as quantifying over individuals. Of course, they also need to be allowed besides adjectives, where they get interpreted as quantifying over degrees. For example, for those speakers of Qu´ebec French who have the word full in their idiolect, it is allowed to appear with adjectives (264), nouns (265), and verbs (266), and therefore participates in IAD (267). In all cases, in brings about a meaning of high intensity, large amounts, etc.
(264) Le film qu’ on est all´e voir ´etait full bon. The movie that we is went see was full good ‘The movie we went to see was really good.’
(265) J’ ai bu full de caf´e ce matin. I have drunk full of coffee this morning ‘I drank a lot of coffee this morning.’
(266) Alice a full dormi la nuit pass´ee. Alice has full slept the night past ‘Alice slept a lot last night.’
(267) Maxime est full rendu riche depuis qu’ il a ´et´e engag´e. Maxime is full become rich since that he has been hired ‘Maxime has become very rich now that he has been hired.’
This test provides us with a perfect characterization of the class of IAD operators. A degree operator can participate in IAD iff it can appear independently in pre- adjectival position, before de in the left edge of DP, and in adverbial position. The correlation is strikingly perfect. Every single item that participates in IAD and given in the list in (16) and repeated here as (268) obeys this.
(268) plus, moins, tellement, assez, trop, le plus, le moins, more, less, so, enough, too, the most, the least, pas-diable, full not-a-hell-of-a-lot, quite
104 I believe that the absence of the equative from this list is particularly telling. Vir- tually every account of the semantics of degree operators recognizes that comparatives and equatives are syntactically and semantically very similar. In order to adequately describe the meaning of English as, it suffices to claim that it has the same denotation as more, except that it involves the “greater or equal than” relation, rather than the strict “greater than” relation49.
(269) [[ as ]] = λP
Under the hypothesis that this dialect has in its grammar a rule of overt DegP movement, we would naturally be led to expect that every operator that requires DegP movement should participate in IAD. Since the equative clearly falls in this category (it has the exact same type as the comparative), then the equative aussi should also participate in IAD. This is not the case, however, as the ungrammaticality of the following sentence shows:
(270) *Jean a aussi achet´e une grosse maison que Paul. Jean has as bought a big house as Paul What this shows is that not all operators that require DegP movement for inter- pretation can participate in IAD, so this property does not pick out the class of IAD operators perfectly. In contrast, it is easy to show that aussi does not pass the distri- butional test presented in this section, and is thus not expected to participate in IAD. This is because a different lexical item, namely autant, must be used to express equa- tives in the nominal and verbal domain. While aussi is fine before gradable adjectives (cf. (271)) it is not acceptable in the left periphery of DP or VP (272)-(273).
(271) Le film qu’ on a vu ´etait aussi bon que celui d’ avant. The movie that we have seen was as good as that of before ‘The movie we went to see was as good as the one before.’
(272) *J’ ai bu aussi de caf´e que mon p`ere ce matin. I have drunk as of coffee as my father this morning Intended: ‘I drank as much coffee as my father this morning.’
(273) *Alice a aussi dormi que nous la nuit pass´ee. Alice has as slept as us the night past Intended: ‘Alice slept as much as we did last night.’
49Or, according to Rett (2011), the equality relation, although Rett’s position goes contrary to much received wisdom about the issue.
105 Autant fares no better than aussi in IAD, since it is only capable of appearing independently in DP and VP, but not in AP. It is in complementary distribution with aussi, so the result of replacing aussi with autant in (271)-(273) yields the exact opposite judgements:
(274) *Le film qu’ on a vu ´etait autant bon que celui d’ avant. The movie that we is have seen as good as that of before Intended: ‘The movie we went to see was as good as the one before.’
(275) J’ ai bu autant de caf´e que mon p`ere ce matin. I have drunk as of coffee as my father this morning ‘I drank as much coffee as my father this morning.’
(276) Alice a autant dormi que nous la nuit pass´ee. Alice has as slept as us the night past ‘Alice slept as much as we did last night.’
That is, since there is no single lexical item that can express the equative in the adjectival, nominal and verbal domain, then neither aussi nor autant can participate in IAD. There is also an interesting asymmetry in how the comparative plus is pronounced depending on its syntactic position. When it appears in preadjectival position, the final /s/50 is obligatory only if the adjective is vowel-initial. When the adjective is consonant-initial, the final /s/ is possible, but not obligatory.
(277) Jean a engag´e un gars /plys/ intelligent que moi. Jean has hired a guy more intelligent than me ‘Jean hired a smarter man than I am.’
(278) Jean a engag´e un gars /ply(s)/ qualifi´e que moi. Jean has hired a guy more qualified than me ‘Jean hired a guy who is more qualified than I am. ’
In its adverbial use, plus has no such sensitivity to the initial segment of the following word. Whether the verb is consonant-initial or vowel-initial, the final /s/ is always pronounced.
50The consonant that emerges in Standard French is usually /z/ in this position, but it is more frequently /s/ in the Qu´ebec dialect.
106 (279) Jean a /plys/ ´ecrit que moi en fin de semaine. Jean has more written than me in end of week ‘Jean wrote more than I did this week-end.’
(280) Jean a /plys/ dormi que moi. Jean has more slept than me ‘John slept more than I did.’
This begs the question how plus is pronounced when it appears in the adverbial position, but it relates to a gradable predicate lower in the structure, forming an IAD sentence. The result is that the adverbial pattern is maintained. In other words, even if plus relates to a non-local gradable predicate, its pronunciation still remains insensitive to the initial segment of the following word:
(281) Jean a /plys/ eu une bonne note que moi. Jean has more has a good grade than me ‘John has a better grade than I did.’
(282) Jean a /plys/ gagn´e un gros prix que moi. Jean has more won a big prize than me ‘John won a bigger prize than I did.’
I see two ways of explaining this pattern. The first option is that there are two lexical items plus: plusA that is syntactically a degree modifier and that is merged before adjectives only (perhaps adverbs also), and another, plusV that is syntactically an adverb. PlusA is lexically specified to have an /s/ that only appears in liaison 51 contexts, say because it is unsyllabified . PlusV, in contrast, has a fully syllabified final consonant, which will then be pronounced in all contexts, regardless of its neigh- bouring segments. If this is correct, then we must conclude that the lexical item that we find in IAD sentences like (281) and (282) is not the same as the one that appears in preadjectival position, say in (277) and (278). This, in turn, means IAD cannot be derived by movement of the degree operator, since the phonological properties of the plus that appears in IAD are those of plusV, not those of plusA. There is an alternative approach that could be proposed, however, according to which there is only one plus involved in all those examples. According to this view, the final consonant of the comparative operator is unsyllabified, so it is dependent
51Nothing in what I say here hinges on any particular theory of liaison. I speak of unsyllabified segments only to make the theoretical options more concrete.
107 on the following segment in the preadjectival cases, but when it appears in preverbal position, it is not in the right syntactic configuration to yield the short form. This is an unusual situation, however. In most if not all cases of liaison, the ’else- where’ form is the short form, i.e. the one with a missing final consonant. Various theories of liaison propose various explanations for what makes the underlying conso- nant emerge, but they usually agree that what is needed is some degree of closeness between the host and the following word to allow the consonant to appear. This does not seem right for plus. When it is used as a single word answer, for example, the final /s/ is pronounced:
(283) A (While pointing a glass half full of juice): Would you like more or less juice than that? B: /plys/.
This shows that the final consonant does not need a host to attach to in order to emerge, and it is only in special circumstances that it is not pronounced, i.e. before consonant-initial adjectives. While it is perhaps not impossible to explain the fall of /s/ before consonant-initial adjectives by phonological means (say, in order to break up a consonant cluster), it seems that it would necessarily involve phonological mechanisms that are only at play in the pronunciation of the word plus,sinceIcan think of no other case where a similar phonological phenomenon occurs in French. This does not make any phonological explanation of the fall of /s/ impossible to maintain, but it makes it less plausible. What all of this shows us is that there is really no syntactic puzzle in IAD. Words like tellement can be generated in adverbial position outside of IAD sentences, so there is no need for the syntactic component to be enriched with overt movement of the degree operator to generate these sentences. Rather, the motivation for such a derivation is semantic and comes from the apparent synonymy between IAD sentences and their canonical counterparts. As we have seen in this chapter, deriving IAD sentences by movement of the degree operator provides us with the right interpretation for these sentences in just about every case, as long as we allow that some restrictions limit the scopal possibilities of degree operators. So far, we have taken the analysis of IAD from the semantic side and we have been successful at providing an interpretation for every IAD sentence. However,
108 the previous sections and this one have shown that this is only consistent with a syntax that is problematic in many respects. What we will now try to do is take the syntactic generalization concerning the class of IAD operators as our starting point instead. That is, the syntax seems to be telling us that the position occupied by degree operators in IAD sentences is not the result of movement, so our task should be to provide an adequate semantics for such sentences assuming an LF representation where the degree operator is base-generated in its surface position. Our first task then is to distinguish IAD readings from the ordinary adverbial use of degree operators. Usually, in non-IAD sentences, degree operators appearing in adverbial position are interpreted as operating on the degree variable of the verb52 when it has one, or on the event variable if it does not. If the verb has no event-variable and isn’t gradable, then degree operators are simply incompatible with them. A verb like aimer ‘to love’ is an example of the first kind of verb, while sauter ‘to jump’ exemplifies the second. Reposer ‘to lie’ (as in “to lie down”) is a reasonable candidate for the third category:
(284) J’ ai tellement saut´e hier! I have so jumped yesterday ‘I jumped so much yesterday!’
(285) On a assez aim´e votre film pour l’ acheter. We have enough liked your movie for it buy ‘We liked your movie enough to buy it.’
(286) #Le roi a tellement repos´e dans cette crypte pendant plusieurs the king has so lied in this crypt during multiple si`ecles. centuries It is possible to construct examples that are ambiguous between an IAD reading and an ordinary adverbial reading. Here is one way to construct such an example Suppose we take a simple sentence with the verb sauter like (284) and add something gradable like haut ‘high’ below the verb. The resulting structure has a degree operator
52I treat verbs that yield an intensified interpretation when they combine with a degree operator or modifier as taking a degree argument. Love, for example, gets such an interpretation (cf. John loves Mary a lot), and would then be of type
109 in adverbial position next to a verb with an event variable, as well as a potential gradable target in VP.
(287) J’ai tellement saut´ehauthier! ‘I jumped so high yesterday!’ OR ‘I jumped high so many times yesterday!’
This sentence is ambiguous between an IAD reading where tellement relates to the non-local gradable adverb high, and one in which it quantifies over jumping events53. The frequency interpretation is the only one that we expect to find if we believe that IAD sentences are simply derived in the same way as every other example where degree operators appear before the verb. It has been observed before that French has another construction in which an op- erator quantifies over a non-local constituent and where the ability of these operators to affect the aspectual properties of the sentence is relevant. This construction is known as Quantification At a Distance (QAD). QAD has specifically been argued not to be derived by movement of the quantifier, following a similar reasoning to what we have been following in this section, namely because every quantifier involved could be generated there in the first place. Since such a syntax seems well justified, various authors have proposed semantic analyses that correspond to such a representation. Although they have been very successful for QAD, we will see that it will not be possible to construct any similar solution for IAD.
6 Towards an Situ Semantics for IAD
6.1 IAD is not Quantification Over Events
In this section, we compare IAQ with what is arguably its most similar construction, Quantification At a Distance (QAD), which is common to all varieties of French. This construction has some structural features in common with IAD, but a quick glance at its interpretation will demonstrate that it presents important aspectual restrictions, requiring the events denoted by the VP to have certain properties, that IAD does not
53The IAD reading is much more salient that the event quantification reading. Nevertheless, if we can imagine a context where ‘jumping high’ is distinguished from ‘jumping low’, say if I might have either practised my high jumps or my low jumps, then it becomes easy to understand (287) as claiming that I jumped high a large number of times.
110 share. This, in turn, warrants an analysis of QAD where the quantificational element combines compositionally with a VP constituent denoting a set of events in order to explain these restrictions, rather than genuinely combining with the object “at a distance”. In addition to this, QAD has a much more restricted distribution than IAD. Such an analysis will be shown to be inapplicable to IAD. There is an obvious intuitive resemblance between QAD and IAD. In both cases, a quantificational element appears in the adverbial position and relates to another item lower, in the VP. A typical example of QAD is given in (288), which contrasts with the structure with the canonical word-order given in (289).
(288) Paul a beaucoup rencontr´e de ministres. Paul has a-lot meet of ministers ‘Paul has met with a lot of ministers.’
(289) Paul a rencontr´e beaucoup de ministres. Paul has meet a-lot of ministers In simple cases like these, QAD sentences are truth-conditionally indistinguishable from sentences with canonical quantification. This synonymy, in addition to the fact that these two word-order possibilities exist in parallel, as well as the dependency between the determiner and the form of the object (the fact that it starts in de) has lead Milner (1978) to conclude that QAD sentences are base generated with the determiner in adnominal position, and then derived by movement of the determiner. Basically, (289) would then be the underlying form of (288). This is not the generally accepted view, however. Kayne (1975) first observed that the set of lexical items that can participate in QAD is exactly the same as the set of those that can equally well be generated as adverbials and as prenominal determiners. That is, the position where beaucoup appears in (288) is a position where it is allowed to be generated anyway, even in the absence of any direct object:
(290) Jean a beaucoup dormi. Jean has a-lot slept ‘John slept a lot.’
Since the class of IAD operators is exactly those that can be generated as adver- bials, prenominal determiners as well as with gradable adjectives, then the class of IAD operators repeated in (291) from (16), is a subset of the class of QAD operators,
111 given here in (292)54. In other words, IAD operators are simply QAD operators who also have the ability to modify adjectives.
(291) plus, moins, tellement, assez, trop, le plus, le moins, pas-diable, full (292) beaucoup, peu, autant, `a peine, davantage, ´enorm´ement, gu`ere, pas mal, rudement, suffisamment, tant, un peu, vachement, plus, moins, tellement, assez, trop, le plus, le moins, pas-diable, full
Much as with IAD then, this shows that the syntactic component of the grammar actually does not need any movement rule to get the determiner in the adverbial po- sition, since QAD sentences can be generated anyway, if we put aside for the moment the problem of object DPs that start in de. We can easily show that peu ‘few’, for example, is predicted to participate in QAD, since it can appear as the head of DP, in adnominal position (293), as well as before intransitive verbs (294), where it yields a frequency interpretation. Thus, it is acceptable in QAD (295):
(293) Peu de gens ont abandonn´e la course. few of people have given-up the race ‘Few people gave up in the race.’
(294) Marc a peu dormi cette nuit. Marc has little sleep this night ‘Marc slept a little last night.’
(295) Jean a peu vu de films. Jean has few seen of movies ‘John saw few movies.’
That is, any determiner that can replace peu in (293) and (294) can participate in QAD, which therefore means that it can also replace it in (295). Just as with IAD, I am not aware of any exception to this generalization. Many authors since then have adopted this position and provided various explana- tions concerning the licensing of the object in de. Objects in de have a very complex distribution and I will not do justice to the entire literature on it here. They are not acceptable by themselves, if no quantificational element is present in the sentence to license it. 54This list is built on the basis of Doetjes (1997)’s list of “Degree Quantifiers” (p. 92), with the addition of the superlatives and operators unique to the Qu´ebec French dialect like pas diable.
112 (296) * J’ ai vu de films hier I have seen of movies yesterday DPs headed by any of the quantifiers that participate in QAD are all of the form Q-de-N, which makes it appealing to claim that QAD sentences are derived on the basis of this word order (Milner (1978)). Nevertheless, it is clear that objects in de can be licensed by higher quantifiers even in cases where no plausible derivation involves the licenser appearing inside the de-DP. For example, a negative DP like personne ‘no one’ appearing in subject position can license a de-DP in object position, despite the fact that it is highly unlikely that personne could have been moved from next to de (Moritz and Valois (1994)):
(297) Personne n’ a vu d’ ´etoile filante cette nuit. Nobody ne has seen de stars flying this night ‘No one saw any shooting stars tonight.’
What this shows is that while the form of the object in QAD sentences like (288) is compatible with an analysis whereby the quantifier is generated inside the object DP, it is by no means strong evidence in favour of it, since such objects can be licensed by other means independently. An additional observation in favour of the base-generation analysis of QAD is provided by Obenauer (1983; 1994), who observes that the QAD sentences are not in fact exactly synonymous with their canonical counterparts in Standard French. They only appear to be so in examples that are too simple to show the full meaning of QAD sentences properly. More to the point, Obenauer claims that QAD sentences obey what he calls the ’Multiplicity of Events Requirement’, which states that QAD is only possible if the VP next to it denotes a set of events which is larger than a singleton set. For example, take the grammatical and felicitous QAD sentence (298).
(298) Il a beaucoup trouv´e de pi`eces d’ or. He has a-lot found of pieces of gold ‘He found a lot of gold pieces.’
If we add the PP modifier en soulevant le couvercle, ’as he lifted the lid’ to it, we force the VP to denote a single event and the resulting QAD sentence is judged contradictory by native speakers of European varieties of French (299). Speakers who attempt to judge it report that it seems to be trying to say that there were multiple
113 events of him finding gold pieces in the single event of opening the lid, which they judge as contradictory.
(299) * En soulevant le couvercle, il a beaucoup trouv´e de pi`eces d’ or. In lifting the lid he has a-lot found of pieces of gold The sentence becomes meaningful again when we put the quantifier back in canon- ical position, since the resulting sentence is not an instance of QAD, and thus does not produce a reading that involves quantification over events, but only over individuals.
(300) En soulevant le couvercle, il a trouv´e beaucoup de pi`eces d’ or. In lifting the lid he has a-lot found of pieces of gold ‘As he lifted the lid, he found a lot of gold pieces.’
As discussed in Cyr (1991), Bouchard and Burnett (2010) and Burnett (2012), these aspectual restrictions do not carry over to Qu´ebec French QAD. While (299) is judged contradictory by speakers from France, for example, it is fine for me as well as for the majority of the speakers interviewed by Cyr. We will see that there is a simple way in which we can parametrize this difference between the two dialects, following Burnett (2012). For the moment, however, we focus on Standard French, where the aspectual restrictions do hold. The ungrammaticality of examples like (299) in Standard French establishes that QAD sentences are not simply synonymous with their canonical counterparts in that dialect. However, as discussed by Doetjes (2007) and Burnett (2012), some subtle interpretive properties of QAD sentences also show that they are not simply the result of semantically combining beaucoup with the VP in the same way as when they are used strictly as adverbials. Burnett (2012) shows that QAD is also sensitive to a Multiplicity of Objects Requirement, which she states as follows:
(301) QAD sentences are only true in contexts involving multiple objects.
For example, (302) is not made true in a situation where I have called my own mother many times, but I have not called any other mothers. If quantification over the event variable were all that was needed, then this reading should be available since the situation does involve a large number of calling-a-mother events by me.
(302) J’ ai beaucoup appel´e de m`eres I have a-lot called of mothers ‘I called mothers a lot.’
114 Given this new fact, Burnett claims that that Standard French QAD sentences involve quantification over
(303) λxλy. feed(y, x) & dog(y).
The theme of the verb is now restricted to dogs only, and it has been pushed to the last argument position, rather than the first. As far as I know, such a rule is of no use for English (Chung & Ladusaw propose it for Chamorro). Let us see how making use of this rule can help provide an interpretation for an ordinary QAD sentence. I repeat (288) here as (304) for convenience:
(304) Paul a beaucoup rencontr´e de ministres. Paul has a-lot meet of ministers ‘Paul has met with a lot of ministers.’
Burnett assumes the following LF, in which the subject is interpreted below beau- coup:
55Alternatively, Burnett suggest that we might view de in this case as the morphological spell-out of the application of Restrict.
115 (305) IP2
Q IP1
beaucoup N VP
Paul V DP
rencontr´ede N
ministres
The important step is the combination of the verb with its property-denoting object.
(306) [[ VP ]] = λx. λe. λy. meet(x, y, e) & minister(y)
The order of the arguments is crucial here. After the VP has been formed with Re- strict, the argument corresponding to the theme of the verb is now the last argument. The subject may now be added, and the resulting IP has the following denotation. This is the characteristic function of a set of
(307) [[ IP1 ]] = λe. λx. meet(p, x, e) & minister(x)
Finally, Standard French beaucoup is interpreted as a binary quantifier, i.e. a quantifier that takes a set of ordered pairs and returns a truth-value. It is ‘binary’ in the sense that it closes off both the theme argument as well as the event argument. Its denotation is given below, where I give events the type ‘v’. Context-dependent standards for what counts as a lot are introduced independently for individuals and events as s and s’, respectively56.
(308) [[ beaucoupSF ]] = λP
Putting all of this together, the truth-conditions of the full sentence in (304) are the following:
56This is not quite Burnett’s formulation, but it is sufficient here.
116 (309) [[ IP2 ]] = |{x: ∃e [meet(p, x, e) & minister(x)]}| > s&|{e: ∃x [meet(p, x, e) & minister(x)]}| > s’
This formula states that the sentence is true if the number of ministers that Paul met is above some standard, and if the number of events of Paul meeting ministers is also above some standard. It is easy to adapt this semantics to Qu´ebec French, where QAD only involves a multiplicity of objects, not of events. Burnett assumes that Qu´ebec French QAD sentences differ from their Standard French counterparts in two ways. First, it in- volves a different lexical item beaucoup that only quantifies over individuals and not events, and second, it has a slightly different LF, since existential closure over the event variable is assumed to take place just below the position where the quantifier is merged. That is, semantic composition follows exactly the same steps as in the
Standard French example up to IP1. At this point, existential closure kicks in, and the resulting constituent, call it IP1’, denotes an ordinary set of individuals. They are the individuals x for which there exists an event of Paul meeting x and where x is a minister.
(310) [[ IP1’]]=λx. ∃e [meet(p, x, e) & minister(x)]
Qu´ebec French beaucoup can now be given the denotation of a unary quantifier, which simply maps a set of individuals to true iff its cardinality is higher than some contextually provided standard:
(311) [[ beaucoupQF ]] = λP
The full sentence thus gets the following truth-conditions in Qu´ebec French:
(312) [[ IP2’]]=|{x: ∃e [meet(p, x, e) & minister(x)]}| > s
We now have an interpretation for QAD sentences both in Standard French and Qu´ebec French. We will now try to apply the same logic to IAD. However, we will see that IAD lacks one crucial property that QAD has that make this analysis work: the position of the target gradable predicate in not constant, while that of the object relevant for QAD is.
117 6.1.1 Applying a QAD Semantics to IAD?
Relating this back to IAD, the fact that both constructions involve a class of operators that could be generated in this position in the first place makes it very tempting to give a unified analysis of both phenomena. Furthermore, the relevant classes of operators are largely overlapping: specifically, all IAD operators are also QAD operators. Since we have seen that we have reasons to believe that a movement analysis of IAD is problematic in multiple counts, it is tempting to ask whether it may be possible to borrow (some features of) the in situ analysis of QAD to the case of IAD. First, just like Qu´ebec French QAD, IAD does not involve any aspectual restrictions. (313), for example, shows that unlike QAD, IAD is compatible with sentences describing single events.
(313) En soulevant le couvercle, il a tellement trouv´e des vieilles pi`eces In lifting the lid he has so found of old pieces d’or. of-gold ‘As he lifted the lid, he found such old pieces of gold.’
Additionally, statives (314) and existential sentences (315) are also compatible with IAD, despite the fact that they presumably do not introduce events at all.
(314) Ton p`ere connait tellement des belles chansons. Your father knows so of nice songs ‘Your father knows such beautiful songs.’
(315) Il y avait tellement des bons groupes hier. It there had so of good bands yesterday ‘There were such great bands yesterday.’
This data shows that quantification over
118 does not have a double object construction. Second, it is semantically special, in that it denotes a bare property, which in turn means that it is not interpreted with the ordinary mechanism that are sufficient for individual-denoting expressions and quantified objects. Since it is interpreted via Restrict, the verb’s internal argument is not actually closed off, just ‘delayed’. QAD quantifiers are always interpreted at the top of the clause, hence after every other argument of the verb has been filled. Since the internal argument always gets pushed to the last position, and since the quantifier always appears at LF above the position where every other argument is filled, the quantifier will always combine with a constituent of the right type to yield correct truth-conditions. Does IAD have the relevant properties for a similar analysis to hold? By anal- ogy, a parallel analysis should involve pushing back the gradable predicate’s degree argument, rather than closing it off, in such a way that the constituent with which the degree operator combines denotes a set of degrees. However, given the fact that the gradable adjective target of IAD operators is usually a modifier, rather than an argument of the verb, it is difficult to see how such an analysis could get off the ground. The fact that Restrict pushes the verbs innermost argument to the last posi- tion makes it possible for the verb to combine with its other arguments just as well as if the internal argument actually had been closed off. In contrast, pushing a gradable adjective’s degree argument to be its last argument does not make it any easier to interpret it when it appears as a modifier inside NP, as it does in most of the examples we have examined so far. Tweaking the adjective’s argument structure in a manner akin to what Restrict does is of no help to interpret the constituent made up of the noun and adjective. Broadly speaking, this is because the target gradable predicate can appear in various positions in the VP, and it can be embedded fairly deep. Although this analysis cannot be naturally adapted to IAD, we will see that this is not the only way to relate a degree operator to an non-local gradable predicate.
6.2 Reconciling IAD with an In Situ Syntax
Suppose that we take the syntax of IAD sentences at face-value, and accept that degree operators in IAD sentences are base-generated in their surface position. Then, as we have discussed, we get the distributional facts right, but we must now explain how IAD sentences get interpreted the way they do.
119 One way to do this would be to appeal to a derivation according to which the result- ing semantics is essentially indistinguishable from what was derived by the movement analysis. This can be done if we accept free lambda abstraction over a non-local degree variable as part of the semantic apparatus. The derivation of any IAD sentence would then involve merging a degree variable as the argument of the target gradable predicate, and merging an element that triggers lambda abstraction over that variable just below the position of the degree operator. Given the way that we have been treating the interpretation of syntactic structures with movement so far, the meaning resulting from such an LF will be the same as what would result from movement. The only difference is that no movement relation is established between the position of the degree operator and that of the degree variable. Let us see what advantages there are to this approach, compared to treating IAD as overt DegP movement. This will allow us to see what problems in the previous analysis were specifically attributable to syntactic movement. First, island constraints are no longer expected to be relevant for IAD. A more thorough examination of the conditions under which IAD can relate an operator with some adjective across various kinds of boundary may reveal as yet unknown restric- tions on this relation, but we know that ordinary syntactic islands do not block this construction in the way that would be expected given an analysis in terms of overt DegP movement. It would appear that the base generated analysis is too lenient, however, because we have seen in section 4.3 that some instances of IAD across quan- tified NPs were ungrammatical despite the fact that no plausible semantic reason for this could be found. This has led me to claim in this section that the restriction, expressed as Kennedy’s Generalization, must be syntactic, and related to movement. Still, there is no principled reason why this restriction should hold, and it is no more costly to claim that the rule that allows abstraction over a non-local degree variable is forbidden from taking place in exactly the same environments as those that were claimed to prevent movement before. As such, the base-generated analysis solves the undergeneration problem concerning IAD across island boundaries. Second, the problematic cases where no source position could be found for the allegedly moved degree operator no longer raise any difficulty. In order to derive a sentence like (248), repeated here as (316), it is no longer necessary to assume that tellement should be allowed to be generated beside the gradable noun, which we know
120 independently it cannot.
(316) Marc a tellement fait une connerie! Marc has so did a foolishness ‘Marc did such a stupid thing!’
Tellement need only be generated in the adverbial position along with the abstrac- tor, while the gradable noun has its degree argument saturated locally by a degree variable. The abstractor will re-open the degree argument of the gradable noun, mak- ing the whole constituent a suitable argument for the degree operator. This solves the problem of undergeneration of the overt DegP movement analysis. Third, assuming that lambda abstraction over a degree variable is free to take place anywhere in Qu´ebec French, we have an explanation for the generalization concerning the class of IAD operators. If an operator can be generated independently in adverbial position, and can be interpreted as quantifying over degrees, then it can participate in IAD since this interpretation will naturally follow if the operator has been generated as an adverb, and the abstractor is generated just below the operator. An item like tr`es, ‘very’ cannot participate in IAD because although it operates on degrees, it cannot appear in adverbial position independently. The ungrammaticality of (318) follows from that of (317). That is, while (318) is semantically well-formed, it is ruled out by the syntax.
(317) * J’ ai tr`es dormi. I have very slept
(318) * J’ ai tr`es vu un bon film. I have very seen a good movie This analysis thus provides an explanation for the generalization on the class of IAD operators, which the movement analysis failed to account for. What about the issue concerning factorial and differential arguments? The move- ment analysis could not explain why such constituents could neither be pied-piped, nor left in the operator’s alleged source position. Looking at the bad examples with differentials again, it is at least clear why the ‘stranding’ example repeated here as (319) is unacceptable, since under the base-generated analysis, plus was never next to long in the first place, so no stranding could take place.
(319) * Jean a plus ´ecrit un cinq pages long travail que Marc. Jean has more written a five pages long paper than Marc
121 However, it is not quite as clear how to explain the ungrammaticality of (320), where the differential is next to the degree operator in adverbial position:
(320) * Jean a cinq pages plus ´ecrit un long travail que Marc. Jean has five pages more written a long paper than Marc One promising sketch of a solution that I see would be to claim that there is no placeforthedifferentialinthesyntax.Thiscouldbedoneifweadopttheviewthat DegPs in canonical position are part of the extended projection of AP, rather than appearing in their specifier (cf. Abney (1987) Corver (1997)). If this is so, then it is reasonable to assume that differentials and factorial arguments are also generated in this extended projection. When a degree operator is generated in the adverbial position, it is clearly not in the extended projection of AP, but more likely that of VP (or some relevant aspectual projection). If the case can be made that this extended projection does not contain a suitable position for factorials and differentials, then it would be possible to explain why these are not compatible with IAD. Unfortunately, this case is very much weakened given that even in its true adverbial use, plus can combine with factorials and differentials.
(321) Jean a trois fois plus travaill´e que Marc. Jean has three times more worked than Marc ‘JohnworkedthreetimesasmuchasMarcdid.’
(322) Jean a travaill´e trois heures de plus que Marc. Jean has worked three hours of more than Marc ‘John worked three hours more than Marc did.’
This shows that whatever the structure involved in differentials and factorials, the base-generated analysis of IAD is of no help in ruling out (320), since these constituents are compatible with the adverbial use of more. Thus, the incompatibility of factorials and differentials with IAD remains a some- what mysterious issue. As such, the base-generation analysis by itself of IAD does not entirely resolve the overgeneration issue of the movement analysis57. A final advantage of this sort of solution is that is makes it relatively easy to give an adequate description of why IAD is possible in Qu´ebec French, but not in English 57I personally find factorial arguments with the adverbial use of plus to be better when both the factorial and the degree operator follow the participle:
(1) Jean a travaill´e trois fois plus que Marc. ‘Jean worked three times as much as Marc did.’
122 or other varieties of French, since we can just say that only the grammar of the former makes use of an item that triggers free lambda abstraction over some non-local degree variable. This solution produces the same interpretations as the overt DegP movement analysis, since the effect of the abstractor is ultimately identical to lambda-abstraction triggered by movement, but it does not predict the same restrictions, which we have shown in this section to be largely for the better. Still, we do not have an explanation for why factorial and differential arguments are ruled out in IAD sentences. Furthermore, this assumes that the grammar of Qu´ebec French makes use of a mechanism that basically mimics the effect of move- ment, without actual movement. This is certainly not an uncontroversial claim to make, and I do not believe that such a major enrichment of the formal apparatus of semantics is called for here. Nevertheless, it is important to note that for the scholar who believes independently that free lambda abstraction over non-local variables is part of the grammar, then the analysis of IAD is burdened with much fewer problems than for the scholar who does without this type of operation. Rather than enrich the semanticist’s toolbox in this way, we will explore the possi- bility that there is actually no direct compositional link between the degree operator and the (degree argument of) the target gradable predicate lower in the structure in IAD sentences. Rather, the degree operator combines locally with its VP argument, and the scale that is necessary for the interpretation of the operator is provided by the pragmatics, not by the semantics of any lexical item in the VP58. We will see that this seems to be necessary anyway in order to account for the interpretation of other constructions, both in French and in English.
6.2.1 Pragmatically-Derived Scales
Throughout this chapter, we have followed the very standard assumption in the liter- ature on degree semantics that the interpretation of degree operators always involves reference to a scale provided by some lexical item, which are always gradable predi- cates. In this section, we explore the possibility that this is not how IAD sentences are
Nevertheless, IAD sentences are not made any better by pushing these items to the right of the participle, as the following example shows:
(2) *Jean a ´ecrit cinq pages (de) plus un long travail que Marc. 58I thanks Michael Wagner for urging me to explore this possibility, as well as for much help in constructing this argument.
123 interpreted. Rather, the scale that the degree operator makes use of in such sentences is how appropriate the VP is to describe the subject. For example, this means that a typical IAD sentence like (323) could be adequately paraphrased as (324):
(323) Jean a plus ´ecrit un bon article que Paul. Jean has more written a good paper than Paul (324) It is more appropriate to say of John that he wrote a good paper than it is appropriate to say it of Paul.
Let us focus on this paraphrase as the actual meaning of the sentence, putting aside for the moment what exactly we mean by “appropriate”. Compositionally, the degree operator combines directly with the constituent that is its sister in the surface structure, namely the VP, to yield a relation between individuals, and it is used to make the claim that it is more appropriate, more adequate, or more true, to use this VP to describe the main clause subject Jean than it is to use it as a property of the individual introduced in the than-clause, Paul. As it happens, the VP contains the gradable predicate bon ‘good’. How can it be more true of one person than another that he or she wrote a good paper? The most obvious answer is that it will be truer of x than of y if the paper that x wrote is better than the paper that y wrote. In this example, it is difficult to see how else we could rank x and y relative to the appropriateness of using the VP to describe them. In other examples, it is easier to find such alternative orderings:
(325) Jean a plus mari´e une fille avec un gros nez que Paul. John has more married a girl with a big nose than Paul (326) It is more appropriate to say of Jean that he married a girl with a big nose than it is to say it of Paul.
Of course, this can be understood to compare the relative nose-length of the girls that Jean and Paul married, but it is not the only way that Jean and Paul can be ranked relative to the appropriateness of being described by this VP. This could also be used to describe a situation in which Jean was officially wedded, in front of the mayor and with all the appropriate paperwork, while Paul got married in a Vegas church. Under this reading, it is not how long the noses are that matter, but rather just how right it is to describe the individuals as “married”. The fact that both girls
124 have big noses is accidental here, and it does not matter for the truth-conditions of the sentence whose nose is bigger. This reading is much less salient than that were nose-lengths are compared, to be sure. Still, provided with the appropriate context, the sentence will be judged true and felicitous with that meaning. I believe this shows that degree operators in IAD sentences need a way of ordering individuals relative to the property expressed by the VP, and the presence of a gradable predicate in the VP will strongly encourage using the scale introduced by the gradable predicate to do so, but it crucially does not need to. Before we examine this proposal in detail and see exactly what the relevant scale is and how it comes into the truth-conditions, I believe we can convince ourselves that we need to appeal to comparing appropriateness of use by looking at other constructions, including some constructions of English. Take the following sentence, for example.
(327) John married a girl who has a big nose, more so than Bill.
I believe this is the closest to an IAD sentence that we can construct in English. The reading that interests us is the one where it is claimed that John married a girl who has a bigger nose than the girl that Bill married59. How is this interpretation derived? We have already shown that DegP movement in English cannot escape relative clauses (cf. section 5.1), so it cannot be the case that this interpretation is derived by having more escape from a covert copy of the main clause merged after so. That is, the following LF for (327) should be unavailable:
(328) # [ more [ than Opj Bill married a girl who has a tj big nose ]]k [ John married
a girl with a tk big nose ]
Since the gradable predicate is embedded inside a relative clause, there is no way for the degree operator to have access to it, even if we posit enough covert material after so. Just as in what I am proposing here for IAD sentences, the degree operator does not relate compositionally with the gradable adjective in its scope. Rather, it is the property married a girl who has a big nose that serves as the argument of the degree operator, and it is the appropriateness of using it to describe Bill or John that
59This sentence also has a reading where the girl’s nose is compared to Bill’s nose. This reading does not concern us here.
125 is compared. Again, the most salient reading by far in one where nose-lengths are compared, but it is also possible to get a comparison of how adequate it is to call the two individuals “married”. The situation is similar with the following sentence, where comparison of degrees of truth is explicitly used:
(329) It is more true of John than of Paul that he has married a girl with a big nose.
Regardless what the formal meaning of “more true” is, it is clear that the degree operator more does not relate in any direct compositional way to the degree argument of big, since it is instead related with that of true, which is here coerced into gradability. This sentence is also ambiguous in the same way as the previous examples. Finally, man-initial sentences also often involve a non-local relation between a degree operator, man, and some gradable predicate:
(330) Man, this salsa is spicy.
As discussed in McCready (2005) and in more detail in McCready (2008), man- initial sentences can receive two types of interpretations depending on their prosodic contour. If there is a fairly large break after the particle, then its semantic contribu- tion will be to add that the speaker is either very happy or very unhappy about what follows. We will not be concerned with this reading. If there is no prosodic break or almost none, however, and the sentence rather exhibits what McCready calls an “integrated intonation”, then the result is an interpretation where the particle con- tributes an intensification over a gradable element lower in the structure. In order to make it clear what intonation I refer to, I will henceforth omit the comma after man in examples that should be read with integrated intonation. We can see that man-initial sentences are just like IAD is allowing the gradable predicate to appear inside a relative clause60.
(331) Man that guy tells stories that have evil villains.
60McCready (2008) claims that sentences with this configuration are degraded and makes a point of attempting to derive this alleged restriction, but I have found that it is possible to construct examples that are natural enough to be acceptable by all native speakers that I have consulted. I thus disagree with McCready that relative clauses boundaries block man from relating to a gradable predicate inside.
126 Again, this puts the operator in a position from which it cannot have a direct compositional link with the gradable predicate in its scope, despite the fact that the obvious interpretation of this sentence is that the villains in the stories are very evil indeed. For some reason, it is not quite as easy to use man to comment on the appropri- ateness of using the word “married” if a gradable predicate is also present. That is, while we can get such a reading from (332), it is harder, if not impossible, for (333):
(332) Man is this guy married. (333) Man is this guy married to a girl who has a big nose.
(332) can be understood to make the claim that this guy has all the stereotypical properties of being married: he and his wife share a bank account, own a nice house together, have a child, etc, corresponding to a reading where it is the “married” property that is intensified. In (333), intensification seems to be only available on the adjective big. The most likely explanation that I find for this is that while it is in principle available for man to concern the appropriateness of describing this guy as married even in (333), this is harder to get with man than with the comparative because it is not as easy to imagine what counts as very high on the scale of being married. In the absence of any gradable predicate, we have no choice but to come up with an interpretation, but it is by no means obvious what this can be. Indeed, different speakers I have consulted came up with different scenarios which would make (332) true. This is because there is no obvious way to rank individuals with respect to how appropriate it is to say that they are married, while it is very easy to come up with a way of ordering individuals with respect to how appropriate it is to say that they are married to a girl with a big nose, since the gradable predicate then provides the ordering. In all the English examples (327), (329) and (331), syntax precludes that there could be a direct relation between the degree operator and the gradable adjective, yet in all three cases the most salient interpretation of the sentence makes use of the scale which is part of the meaning of the non-local gradable adjective. What I want to propose is that in IAD sentences, just as in English sentences of the sort we have just examined, the degree operator relates to a scale that is pragmatically built on the basis of the appropriateness of use of the full VP or clausal constituent, and whenever a gradable adjective is present in the VP, then pragmatics will naturally
127 default to using the scale which is part of the meaning of the adjective. If no gradable element is present to provide a scale that pragmatics can appeal to directly, then some other salient ordering must be used. If nothing in the VP can reasonably be taken to provide a scale, then the result is infelicitous. We will focus on comparing man-initial sentences with IAD below, showing that the relation between the operator and the gradable predicate is not as direct as we originally thought. Some predicates, although not gradable in the typical sense, are adequate licensers for both IAD and man, since they involve scales in an obvious way. The PPs like crap and its Qu´ebec French equivalent comme de la bouette are good examples of this.
(334) Man John tells stories that end like crap.
(335) Jean a tellement racont´e des histoires qui finissent comme de la Jean has so told of stories that end like of the bouette. mud ‘John told stories that end very badly.’
Similarly, mass nouns and plurals are typically assumed to denote join semi-lattices (cf. Link (1983) and many authors since then), and hence contain a natural ordering, provided by the part-of relation61 The following examples show that mass nouns can provide the scale that man and IAD operators are looking for.
(336) Man we drank beer yesterday.
(337) On a tellement bu de la bi`ere hier. We have so drank of the beer yesterday. ‘We drank so much beer yesterday’
These examples are particularly interesting in that the case here is stronger than in the PP cases that there is truly no gradable predicate in the sentence. It has been claimed at least in Matushansky (2002b) that PPs can be gradable in the ordinary sense, while I believe it is more controversial to claim that all mass nouns involve a silent gradable predicate, say a silent much, although this view is of course not unheard of (cf. Hackl (2000) and references therein). If the constituent that the degree operator combines with has no predicate that is either gradable in the ordinary way or can be treated as gradable by pragmatics,
61See Hackl (2000) for details on how to integrate this ordering in the semantics of gradability.
128 or if it contains only a gradable predicate whose degree argument is saturated by a measure phrase, then IAD and sentence-initial man are both out:
(338) #Man this table is four feet long.
(339) #Man the king laid in the crypt for three centuries.
(340) # On a tellement achet´e une table de quatre pieds de long. We have so bought a table of four feet of long
(341) # Le roi a tellement repos´e dans cette crypte pendant plusieurs the king has so lied in this crypt during multiple si`ecles. centuries If we accept all the evidence we have seen so far that there is no direct semantic relation between the degree operator and the gradable predicate in IAD sentences and related English constructions, then we must come up with a semantics for these sentences that makes more precise the vague claim made at the beginning of this section that such constructions involve degrees on the scale of appropriateness, not on the scale introduced by the gradable predicate. This is of course highly reminiscent of the way that metalinguistic comparatives are interpreted. We now turn to an examination of how such constructions are interpreted, and suggest a slight adaptation to make this semantics applicable to IAD.
6.2.2 Metalinguistic Comparatives
Metalinguistic comparatives are sentences where, loosely speaking, the speaker makes the claim that a word or expression is better suited than another to describe an indi- vidual or situation. Although they have mostly received only passing comments, there have recently been explicit proposals concerning their interpretation by Giannakidou and Stavrou (2008), Giannakidou and Yoon (2009) and Morzycki (2011). Morzycki provides the following example:
(342) I am more machine now than man.
Here, the speaker claims that it is more appropriate, more adequate, or more true, to describe him as a machine than as a man. Morzycki cashes out the intuition that metalinguistic comparatives involve a claim of being “more or less close to the truth”
129 by appealing to a modified version of Lasersohn (1999)’s notion of pragmatic halos. In Lasersohn’s original proposal, the pragmatic halo of a predicate P in a context C is just the ordinary denotation of P along with things that are similar enough in C to those things to count as P. For example, while one must ordinarily hold an MD to count as a doctor, in some contexts it is sufficient to have some medical training to qualify as ‘doctor’. Similarly, when the degree phrase six feet is used to describe someone, there is a bit of slack in how precise we expect this measurement to be. We will not typically treat it as false if the person is five feet eleven and three quarters, for example. The pragmatic halo of an expression is basically the extension of its imprecise use, and it is naturally understood as being derived from the strict extension of the predicate and the context of utterance, which provided the level of precision that is expected in the discourse. In order to account for metalinguistic comparatives, Morzycki requires a gradable notion of imprecision, not present in Lasersohn’s work62. The intuition here is that a metalinguistic comparative like (342) can be properly analysed as claiming that it is a more precise statement to say that the speaker is a machine than to say that he is aman. The way Morzycki does this is to introduce a degree of precision on the interpre- tation function. Degrees of precisions are simply numerals ranging from 0 to 1. The interpretation of a predicate at degree of precision 1 has no pragmatic slack at all. In other words, at degree of precision 1, the pragmatic halo of a predicate P is no larger than P itself. At degree of precision 0, the pragmatic halo of P would be so large as to encompass everything with the same type. For example, (343) at degree of precision 1 is only true if John is absolutely, exactly six feet tall, an almost impossible state of affairs since just about everyone is off from six feet from some microscopic measure at least, and any slight variation from this would make it false.
(343) John is six feet tall.
We can imagine, say, some laboratory settings that could involve an utterance context with degree of precision at or near 1. At degree 0, the sentence would be made true regardless of John’s height. This makes the sentence absolutely uninformative, and I can think of no context that would make this the natural interpretation.
62He also recasts Lasersohn’s proposal in alternative semantics, a move that I do not take into account here to simplify the presentation.
130 It is at values between the extremes that we find ordinary uses. We can imagine that at degree of precision 0.9, for example, (343) would be true as long as John’s height does not deviate from six feet by more than a half inch. Degrees of precision are monotonic, in the special sense here that for any denota- tions [[α]]n and [[α]]n’ where n > n’, [[α]]n ⊆ [[α]]n’. In words, this just means that if an individual is a member of the precise interpretation of a predicate, it is a member of every less precise interpretation. Although Morzycki does not discuss this explicitly, this property will actually be crucial for us. We now have everything we need to analyse metalinguistic comparatives. Let us treat the more in such sentences as distinct from the ordinary more,andrefertoitas moreML. Its semantic contribution is to assert that there is a degree of precision for which what is expressed by the main clause is true, while what is expressed by the than-phrase is false. In example (342), the claim is that there is a degree of precision such that the claim Iamamachineis true, while Iamamanis not:
(344) [[ (342) ]] = T iff ∃d. [[ I am a machine ]]d =T&[[Iamaman]]d =F
This does not require that the speaker be perfectly adequately described with the predicate machine. All that is required is that at some level of precision, even if quite low, this predicate be true of the speaker, while the predicate man is not, since man would only be a true description of the individual at an even lower degree of precision. So for example, this could be true if the speaker is adequately described as a machine at degrees of precision 0.7 and lower, while he is only adequately described as a man at degree of precision 0.5 or lower. This means that there is a degree (in fact, an infinity of degrees), say 0.6, for which it is true that he is a machine, but false that he is a man. Since the syntax of English metalinguistic comparatives is not similar to IAD in any interesting way, we will not be interested in the specific compositional seman- tics, but rather with the overall interpretation of sentences involving metalinguistic comparison. I thus give a simplified version of a syncategorematic interpretation for moreML:
(345) [[ x is moreML P than Q ]] = ∃d. [Pd(x)=1 & Qd(x)=0]
How can we adapt this to cover IAD sentences? It seems like precision is not exactly the notion that we need. Suppose we take Morzycki’s analysis as is and apply
131 it to an ordinary IAD sentence. What we get for (346) is the claim that there is a degree of precision for which it is true that John married a girl with a big nose, and it is false that Paul did.
(346) Jean a plus mari´e une fille avec un gros nez que Paul. John has more married a girl with a big nose than Paul
(347) ∃d[[Jeanamari´e une fille avec un gros nez ]]d =T&[[Paulamari´e une
fille avec un gros nez ]]d =F
First of all, this begs the question as to what is the meaning of the VP. Since the gradable predicate gros ‘big’ does not locally combine with any degree operator or measure phrase, it is reasonable to assume that POS provides its degree argument.
This is what Morzycki assumes takes place when moreML combines with gradable predicates. So what this tells us is that there is a degree of precision for which it is true that Jean married a girl who has a nose bigger than the standard size for noses, and at that degree of precision, it is not true that Paul did the same. This doesn’t seem quite right, although it is somewhat difficult to evaluate. The problem here is that the constituent that IAD operators combine with contain bare gradable adjectives, which, notoriously, are vague. Morzycki is careful to distin- guish the notions of vagueness and imprecision. Predicates like six feet tall can have an imprecise interpretation exactly because they have a precise one: there is such a thing as exactly, uncontroversially, six feet tall, even if we do not have the instruments to measure it with sufficient precision. The bare use of gradable adjectives however do not have any precise interpretation: there is no uncontroversial interpretation of what counts as tall, regardless of method of measurement. Because of this, there is no such a thing as a more or less precise interpretation of a vague statement like John is tall. We can see this more clearly given the fact that we cannot use precision regulators like precisely, which normally force a more precise interpretation of the sentence, in sentences with vague predicates. Precision regulators are flat out incompatible with vague predicates like tall:
(348) *John is precisely tall.
Yet there is an obvious sense in which it is closer to the truth to say that John is tall if he is seven feet tall than if he is only six feet tall. It is this measure of “being
132 close to the truth” which seems to matter here, more than that of precision. For our purposes, then, we will borrow Morzycki’s mechanisms, except that we will not treat the numeral on the interpretation function as providing the degree of precision to which the sentence should be understood, but rather something less clearly defined, which I will simply refer to as appropriateness63. How can it be more or less appropriate to say of an individual that he or she has some vague property? We will simply assume that appropriateness subsumes the notion of precision, except that it is also defined with predicates that involve contextually specified standards. We will assume that for any two individuals x and y and a gradable property P interpreted relative to a contextually-specified standard s, if x has property P to a higher degree than y, then the maximal degree of appropriateness d such that Pd(s, x) is greater than the maximal degree d such that Pd(s, y). Basically, if John is taller than Mary, it will be more appropriate to say John is tall than it would have been to say Mary is tall. How does this reinterpretation of the numeral on the interpretation function help us interpret IAD sentences? The advantage is that we can now make sense of the claim made by the formula in (347). If we understand the numeral on the interpretation function as a degree of precision, we cannot make sense of this because there is no such a thing as a more or less precise interpretation of a vague statement. If we rather understand it as a degree of appropriateness instead, in the manner defined above, then this can be made true simply by virtue of the fact that one individual’s nose is larger than the other’s. Of course, it is still possible to get the alternative interpretation according to which it is the degree to which it is appropriate to say that each individual counts as “married” that matters. If both relevant girls have a nose that counts as big in the context, regardless whose is bigger or if they are of the same size, the sentence could be judged true as long as the circumstances described above stand, say that John is truly married while Paul’s marriage is for some reason less genuine. The semantics that I am proposing leaves a lot of work to the pragmatics. That is, the output of the semantic component does not specify which way of ordering the
63Morzycki argues specifically against using appropriateness as the relevant property in metalin- guistic comparatives on the basis of the fact that the English word “appropriate” can be used to mean something like “appropriate given social conventions” or “appropriate given aesthetic princi- ples”, etc. We will avoid this issue entirely by treating “appropriate” here as a technical term, not as the ambiguous English term.
133 individuals relative to the property expressed by the VP the speaker has in mind. The output formula just says that one proposition is more appropriate than another, and pragmatics has the responsibility of providing the missing details. It is in spirit very similar to the way Kaplan (1989) treats indexicals, which do not actually get their referent in the semantics itself. Another precedent for this way of dividing the labour between semantics and pragmatics is the way Schwarzschild (1992) interprets sentences like (349)64:
(349) The cats and the dogs were separated.
While the most salient way of understanding this is that the animals were separated by species, cats on one side and dogs on the other, sufficient context can also make this sentence true if some other property than the species was used, say if they were separated by colour, black on one side, and white on the other. What this shows is that while the conjunction of plurals in the subject seems to provide the partition that the verb separate requires, it does not do so in the semantics per se. The semantic calculus just tells us that the animals were separated given some contextually specified partition, and the conjunction of plurals makes the partition in terms of species highly salient, but this is not encoded in the semantics. That is, this sentence is not ambiguous, no more than a sentence with an indexical like IamDavidis, despite the fact that it does not make the same claim in all contexts. I suggest that IAD sentences are semantically unspecified in the same way as Schwarzschild claims sentences like (349) are. The semantic calculus applied to IAD sentences only says that one proposition is more appropriate than another, and it is left to the context to specify exactly what it is that makes this so. The fact that the scale provided by the gradable predicate is almost always used, yielding interpretations that are very close to what we get from an analysis where there is long-distance binding of a degree variable, with or without movement, is not taken to have any semantic import, in the strictest sense of the term, since it is only a fact of pragmatics that this scale is relevant. Again, the fact that this interpretation is highly preferred should not be taken as evidence that the scale of the adjective interacts with the degree operator in the semantics, no more than partitioning the animals in terms of species is semantically encoded by the conjunction of plurals in (349).
64Again, I thank Bernhard Schwarz for providing me with this comparison.
134 I believe this approach gives us a way of adequately analysing the semantics of IAD sentences, in a way that does not require the problematic assumptions that were involved in a movement analysis65.
6.3 Explaining the Lack of Differentials in IAD
One advantage of the approach in terms of degrees of appropriateness is that it pro- vides a clear explanation for why differentials are excluded from IAD sentences. Obvi- ously, in ordinary comparatives, measure phrases that can serve as differentials must express values that are on the same scale as that introduced by the gradable adjec- tive. The measure phrase four pounds is fine in a comparative construction where the adjective is heavy,andfour inches is fine with the adjective tall, but if we switch the measure phrases, the result in semantically ill-formed.
(350) John is four inches taller than Bill. (351) John is four pounds heavier than Bill. (352) *John is four pounds taller than Bill. (353) *John is four inches heavier than Bill.
This is a basic fact of the interpretation of differentials and any analysis of this construction should account for it. Clearly, Von Stechow (1984)’s analysis and its altered version in a maximality-based semantics that I provide in section 5.3 do so. There, we have said that in a sentence like (350), it is claimed that adding the value of the measure phrase four inches to Bill’s maximal height gives us John’s maximal height. Similarly, the addition of four pounds to Bill’s maximal weight is equal to John’s maximal weight in (351). The problem with (352) and (353) is that adding four inches to Bill’s maximal weight, or four pounds to his maximal height, is not well-formed. This is because the adjectives heavy and tall are incommensurable, i.e. they involve distinct scales. Given this basic fact, there is no need to say anything
65In this section, I have focused on the comparative operator only. The next logical step will be to generalize this analysis to the entire class degree operators that participate in IAD. I am hopeful that this can be done in a fairly straightforward manner for all of them. For example, a superlative IAD sentence like (1) should get the interpretation that it is more appropriate to say of John that he climbed a high mountain than it is appropriate to say it of anyone else.
(1) C’est Jean qui a le plus grimp´e une gross montagne. ’It’s John who climbed the highest mountain.’
135 more to exclude the unacceptable (352) and (353), since their interpretation involves an addition that the semantics cannot process. In the unacceptable IAD examples, however, the differentials were degrees on the same scale as that introduced as the adjective:
(354) *Jean a cinq pages plus ´ecrit un long travail que Marc. Jean has five pages more written a long paper than Marc What I want to say here is that the interpretation of the degree operator in this sentence does not involve the scale of length of a paper, for which 5 pages would be an adequate differential, but rather it involves the scale of appropriateness, for which a number of pages is no more adequate as a differential than pounds were for tall or inches for heavy. Here the analogy with metalinguistic comparatives is useful again. It is a well- known fact that comparatives where an object is compared with adjectives that involve distinct scales is infelicitous:
(355) *John is taller than he is heavy.
Metalinguistic comparatives, on the other hand, are not sensitive to this. As long as we can find a context that makes the comparison relevant, these are fine with gradable predicates that involve distinct scales. Morzycki suggests a scenario where someoneisunsuccessfullytryingtogetachair through a doorway. The following dialogue is felicitous in this context:
(356) A: Are you having trouble carrying this chair into the living room? I guess it’s kind of heavy. B: Well, it’s not really heavy as such. It’s more wide than heavy. I can’t get it through the door.
This is expected, since the interpretation of B’s statement is not predicted to involve a comparison of the chair’s width to its weight, but rather, in Morzycki’s analysis, the precision to which we can say that it is wide to the precision to which we can say that it is heavy66.
66Again, precision does not seem to be the adequate property here, especially as the predicates heavy and wide are vague, and it seems that comparison of appropriateness is what this sentence involves. However, since my concerns are not with metalinguistic comparatives in general, but only with borrowing the mechanisms involved in their interpretation for IAD, I do not explore this any further here.
136 Differentials are not compatible with such constructions, for the obvious reason that neither the scale of width or the scale of weight are actually used by the com- parative here, only that of precision. This is why neither four pounds nor four inches can be added here to produce a grammatical IAD sentence67:
(357) *It’s four inches/four pounds wider than it is heavy.
Interestingly, metalinguistic comparatives are not entirely incompatible with dif- ferentials. To the extent that modifiers like alotcan be treated as such, then we must 68 admit that some differentials are acceptable with moreML :
(358) John is a lot more of a semanticist that a syntactician.69
What seems to make this fine is that the modifier alotdoes not make reference to any scale at all. For this reason, it is compatible with ordinary comparatives just as well as with metalinguistic comparatives.
67A related matter pointed out to me by various members of my committee is that many examples in sections 4.3 and 4.4 involve than-phrases whose entire content is the demonstrative ¸ca.Thisis meant to avoid complications regarding the internal structure of than-phrases and potential de dicto interpretations in intensional contexts, borrowing the method from Heim (2001). For Heim, the demonstrative rigidly denotes a singleton set containing one degree, either of length, width, etc., depending on what the appropriate scale is. It is not obvious how to analyse the demonstrative in the approach that I advocate here. Take the following example:
(1) Jean a plus travaill´efortque¸ca. ‘John worked harder than that.’
In an ordinary analysis of the comparative, we could say that the demonstrative picks up a value on the appropriate scale from the context, say thirty hours a week, which the comparative can take as an argument. Under the appropriateness-based analysis, I need to claim that ¸ca picks out a degree of appropriateness from the context. The question is where that degree comes from exactly. A plausible context in which (1) is adequate is after someone has asserted that Mary worked thirty hours a week. In order to get the right meaning for (1), it seems to be necessary for me to say that ¸ca must pick out the (maximal) degree of appropriateness of something like Mary worked hard in that context. (1) then says that it is more appropriate to say that John worked hard than it is to say that Mary worked hard. The main problem with this is that nowhere in the discourse is there a claim that Mary worked hard, which could serve as an antecedent for ¸ca. I leave it an open question here how exactly the demonstrative can get this interpretation. 68I thank Bernhard Schwarz for this observation. 69Oddly, many speakers whose judgement I have asked reported that the preposition of becomes optional when the modifier much is used:
(1) John is much more a semanticist than a syntactician.
I have no explanation for this fact, especially given the fact that I have implicitly treated this preposition as semantically vacuous so far.
137 Interestingly, IAD does allow this kind of modifiers to serve as differentials. As the following examples show, beaucoup can serve as a differential in metalinguistic comparatives, as well as in IAD:
(359) Jean est beaucoup plus un s´emanticien qu’ un syntacticien. John is a-lot more a semanticist than a syntactician ‘John is a lot more of a semanticist that a syntactician.’
(360) Jean a beaucoup plus ´ecrit un gros travail que Paul. John has a-lot more written a big paper than Paul ‘John wrote a much bigger paper than Paul.’
I believe this is major evidence in favour of our analysis according to which IAD sentences are interpreted in a manner very close to metalinguistic comparatives, cru- cially in involving a scale which is distinct from that of any gradable predicate in the sentence70.
7 Summary and Conclusion
The purpose of this chapter has been to carefully examine IAD, contrasting two can- didate analyses. First, I have examined the possibility that IAD sentences are derived by an overt instance of DegP movement. The overt distribution and interpretation of degree operators in IAD mimics the expected behaviour of these same operators in many ways. The structures in which IAD is possible vastly overlap with those in which we have reason to believe that DegP movement is also possible. Following
70As pointed out to me by Luis Alonso-Ovalle, factorial arguments are not as easy to rule out with this analysis. Factorials are excluded from IAD sentences, as discussed in section 5.3, and they are also impossible with metalinguistic comparatives in English:
(1) #John is four times as much of a semanticist as a syntactician.
This strengthens the claim that such constituents are excluded from IAD and metalinguistic com- paratives for the same reason, but it is not obvious how the discussion in this section can extend to them. In principle, factorials like four times could be good with both constructions, say if John is a semanticist is true at degree of appropriateness .8 and John is a syntactician is true at degree of appropriateness 0.2. A reasonable explanation for the oddity of such examples is simply that speakers do not actually quantify degrees of appropriateness at all, meaning that they are not ex- pected to be subject to arithmetic operations. What matters for the analysis is simply that speakers have access to an ordering in appropriateness, not that propositions be related to numerical values corresponding to the degree to which they are appropriate. That is, appealing to actual numbers in the presentation should be seen only as an artifact of the formalism, not as an integral part of the analysis.
138 this trail has led to many interesting observations, in particular concerning Kennedy’s Generalization, the interpretation of French negation with adjectives, the compara- tive/absolute ambiguity in superlatives, the nature of comparative ellipsis, and the mechanisms for merging than-phrases. There is a necessarily programmatic character to this approach, as it is impossible to review every proposal made in the degree semantics literature and weight it against the IAD data. We have also seen that treating IAD as being derived by movement of the degree operator faces a number of challenges. Indeed, the syntax that we have to assume under this view makes a large number of incorrect predictions. Following this trail instead, I have suggested that IAD sentences are interpreted in a manner close to that of metalinguistic comparatives, without involving any compositional link between the degree operator and the gradable predicate. Treating IAD operators as making use of a scale of appropriateness makes it possible to give a convincing explanation for the absence of differential arguments in IAD. This sort of mechanism is not an innovation that we make for IAD sentences only. I have shown that there are a number of English constructions that are in fact very similar to IAD, and require the same type of semantics. It will in time be necessary to be more specific about the compositional mechanisms involved in the interpretation of IAD sentences under this view. In particular, I have not been explicit about how it is possible for the degree operator to access the degree of appropriateness of its complement. This could be done with the help of a type- shifter, much as Morzycki does to give metalinguistic more access to the degree of precision of its complement in his analysis, for example. One could also take this to be a specific lexical property of some items that they are inherently capable of accessing this variable. Although this is not so appealing at first sight, it makes it possible to give an analysis of why IAD is possible with some degree operators for some speakers and not for others. No such variation is expected to be possible across speakers if degree operators in IAD have nothing special in their lexical semantics. In other words, the variation in acceptability of IAD sentences depending on the degree operator points to an analysis whereby IAD is made possible by some special semantic property of individual degree operators. This is still not precise enough, obviously, and future work work will hopefully be able to fill this gap. There is still much to do at this point, especially in returning to the large amount
139 of data uncovered in the examination of an overt DegP movement analysis of IAD in the light of the in situ analysed sketched in section 6. In particular, the apparent intervention effect of quantified DPs from section 4.3 seem particularly problematic. For the moment, I find it difficult to see why intervening DPs should have the effect of blocking IAD under an in situ analysis, while a movement analysis at least points to some type of intervention effect. Admittedly, a failing of this analysis is that it cannot account for the apparent parallel between Quantification At a Distance and Intensification At a Distance. Both constructions involve much the same class of operators, and in both cases, what seems crucial is their semantic underspecification: QAD operators need to be able to quantify over individuals and events to participate in this construction, and IAD operators must also be able to quantify over degrees. Still, the semantics that I have provided for IAD tells us nothing about QAD, and I see no reasonable way that any analysis could cover both sets of data. This may be for the best, since as I discussed in section 6.1.1, QAD has some very specific syntactic properties that make it very different from IAD, and which make an analysis like Burnett (2012) possible, while it is clearly inapplicable to IAD.
140 Chapter 3: Opinion Verbs
1 Introduction
This chapter examines the syntactic and semantic properties of what we will sim- ply call opinion verbs, focusing on the specific verbs to find in English and, more marginally, trouver in French. These verbs are dependent on the presence of an ele- ment with the right kind of semantics somewhere in their complement clause, namely a predicate whose semantic value is subjective. This class of verbs has received very little attention in the semantics literature, except in Sæbø (2009) and Stephenson (2007b), who discuss it in some detail in the light of discussion surrounding Lasersohn (2005)’s influential contribution to the debate concerning subjective predicates. A number of the observations that I make in this chapter are also pre-empted by Ducrot (1975), who focuses on French trouver 1 The proximate goal of this chapter is to provide a semantic, syntactic, and prag- matic analysis of these verbs. The broader aim, however, is to use these verbs as a testing ground for the study of subjectivity in the grammar of natural languages. Sentences that express subjective thoughts stand as obvious challenges for the truth- conditional semantics that we have largely inherited from Tarski’s work, since this type of approach is based on the presupposition that all (assertoric) sentences are amenable to an analysis in terms of their potential for truth and falsity. Yet, this seems to be exactly the property that is not shared by subjective predicates. Simply stated, sentences like (1) just don’t seem to be analysable in terms of their truth- conditions, since whether or not licorice is tasty is not a matter of fact.
(1) Licorice is tasty.
This is what K¨olbel (2002) calls the problem of excessive objectivity. He suggests that it can be resolved if we accept that truth, or at least that notion of truth that we use when we discuss truth-conditional semantics, is not entirely connected with the notion of objectivity. Rather, truth is relativized to a perspective, which we will simply equate with an individual. This position is known as Relativism, and
1I thank Fabienne Martin for bringing my attention to this important reference, which had escaped me until a very late stage of this research.
141 this chapter will largely be concerned with defending one specific incarnation of this theory, namely the one proposed in Lasersohn (2005). Such an approach is particularly appealing for the semanticist since if we accept it, the problem of excessive objectivity goes away without interfering with anything inside of semantic theory. That is, we can maintain a truth-conditional semantics for natural language despite problematic examples like (1) as long as we can say that there is a sense of truth according to which such examples are true or false, and this is what relative truth gives us. Note that this does not mean that there is only one notion of truth, and that it must be relative. It may very well be that relative truth is a useful concept for linguists but not for philosophers, for example. I propose to make use of opinion verbs to help fill in the specifics of a relativist theory. As we will discuss in some detail in this chapter, these predicates are sensitive to the subjective/objective property of their complement and thus allow us to see in more detail how the grammar deals with this feature. Although this observation has been made before, I believe its importance has not been fully appreciated. The fact that the grammar is actually sensitive to this feature of language forces us to make use of some device in the formal apparatus of semantics to represent it, which we will argue are individual perspectives but could in principle be something else at this point. There is no sense in which this had to be the case. The grammar does not mark whether a proposition is morally reprehensible or not, or any similar irrelevant feature of meaning. I believe it is thus crucial to examine this data in much detail, since it will guide us in constructing an adequate model of subjectivity in grammar. This chapter is organized as follows: First, we examine the basic features of opin- ion verbs and use their properties as a guide to see how better to formalize semantic relativism. This is the main goal of this chapter. We do this first by painting a broad picture of the syntactic and semantic properties of opinion verbs (section 2), focusing on what makes them different from other verbs like familiar epistemic verbs whose meaning can appear to be very similar in some circumstances. We then review recent proposals on adapting formal semantics to include subjective predicates (section 3), concluding that it is not possible to choose decidedly in favour of treating individ- ual perspectives as indices on the interpretation function, or as hidden pronominals present at (some level of) syntax. Section 4 reviews Sæbø (2009)’s proposal that opin- ion verbs argue in favour of the pronoun approach. However, his proposal is based on an incomplete picture of the meaning of sentences with opinion verbs, since it does not
142 take into account important presuppositions introduced by such sentences. I present this new data in section 5 and suggest a way of formalizing the presupposition, in the form of what I will call the Subjective Contingency Presupposition. This approach makes opinion verbs completely different formally from epistemic verbs like think, despite the fact that they can sometimes basically be used inter- changeably. I will claim in section 6 that this is actually correct, despite appearances, and that opinion verbs really do warrant, in Saebø’s terms, a fully reductionist anal- ysis. In section 7, I review the special syntactic assumptions that are necessary for Saebø’s proposal to hold. Section 8 takes up one of Seabø’s syntactic arguments in favour of a pronoun theory. There, I will show that a more complete examination of the phenomenon of conjunction under find actually argues precisely against the view that individual perspectives are pronominal elements in the syntax, while an index theory combined with the Subjective Contingency Presupposition can adequately deal with most or all of it. Section 9 examines some remaining problematic data introduced by Saebø. Once we have a clear picture of the meaning contribution of opinion verbs and how they relate to subjective predicates in their complement, we turn to a brief examination of various types of predicates that also license the use of opinion verbs, yet are not ordinarily classified in the category of subjective predicates, at least not as obviously as predicates like tasty are. There we will mostly be concerned with ordinary gradable adjectives (Section 10). Of course, this should not be taken as a complete investigation of subjectivity in natural language. Rather, this approach should be seen as programmatic: it is a demonstration of how we can use the special environment that is the complement clause of opinion verbs to see which predicates are subjective and which are not, and to see exactly what it is in them precisely that makes them subjective.
2 Opinion Verbs: The Basic Facts
Opinion verbs like find and trouver are embedding verbs. They can take as comple- ment either full clauses introduced by that or que, or small clauses. (2) John finds Bill clever. (3) John finds that Bill has a nice car.
143 (4) Jean trouve Paul intelligent. John finds Paul intelligent ‘John finds Paul intelligent.’
(5) Jean trouve que Paul a une belle voiture. John finds that Paul has a nice car. ‘John finds that Paul has a nice car.’
Of course, we can express very similar thoughts with ordinary epistemic verbs like think. In fact, this may be the more natural way of expressing one’s opinions, at least in English:
(6) John thinks that Bill is clever.
(7) Jean pense que Paul est intelligent. John thinks that Paul is intelligent ‘John thinks that Paul is intelligent.’
There are important differences between think and find, however. Ducrot (1975) observed that the French verb trouver is only felicitous when followed by a clause that expresses a subjective statement. This observation was independently made more recently by Sæbø (2009), who makes a similar claim concerning the English verb to find as well as its correlates in many languages, including trouver in French, finden in German, and synes in Norwegian. Lasersohn (2009) also makes a similar set of observations about consider 2 (I use the hashmark symbol # to indicate that an example is infelicitous.).
(8) John finds Bill boring. (9) #John finds Bill dead.
(10) Je trouve les dinosaures terrifiants. I find the dinosaurs terrifying ‘I find dinosaurs terrifying.’
2For some reason, consider is somewhat less difficult to satisfy than find. For example, consider is felicitous with a predicate expressing an occupation, like doctor, while find is not:
(1) #John finds Bill a doctor. (2) John considers Bill a doctor.
I leave it open for future research exactly how various opinion verbs are licensed by various classes of more or less subjective expressions.
144 (11) # Je trouve les dinosaures disparus. I find the dinosaurs extinct ‘#I find dinosaurs extinct.’
Boring and terrifying are highly subjective concepts and serve as appropriate licensers for the opinion verbs, while dead and extinct are in principle verifiable, objective facts, which then cannot license the use of opinion verbs. I will reserve the term “opinion verb” for exactly those verbs that are only compatible with subjective complements. Ordinary epistemic verbs are of course insensitive to this. They may take as complements clauses expressing non-subjective thoughts without any infelicity.
(12) John thinks that Bill is dead.
(13) Jean pense que les dinosaures sont disparus. John thinks that the dinosaurs are extinct ‘John thinks that dinosaurs are extinct.’
This is also the case with many other expressions which are naturally used to express one’s opinions. Even phrases like in my opinion can naturally be used with non-subjective statements:
(14) In my opinion, John is in his office.
(15) A` mon avis, Jean est dans son bureau. In my opinion John is in his office ‘In my opinion John is in his office.’
This shows that we have to distinguish between things that one can have an opinion on, broadly speaking, and things that are genuinely subjective. One can have an opinion on whether or not John is having an affair, and on whether or not Guinness is a good beer, but only the former is actually a truth-evaluable proposition. The fact that some lexical items like find and trouver are only compatible with the latter kind of proposition shows that the grammar is sensitive to this difference. I believe that the fact that grammar actually cares about this property is a very important observation. As we will see in the following section, adapting our seman- tic model to take this property into account is no trivial task and it should only be undertaken if we have serious reasons to believe that it is necessary. To put it
145 simply, opinion verbs justify modifying the grammar to distinguish subjective propo- sitions from the more familiar truth-evaluable statements, while the mere existence of predicates like tasty could have been accommodated in less drastic ways. In this chapter, I will mostly be concerned with the verb to find, and sometimes with its French equivalent trouver, but it is of course not the only lexical item that is sensitive to the subjective character of the sentence. While the phrase in my opinion above was shown not to be sensitive to this property, the minimally different phrase in my books actually is. The French phrase tant qu’`amoiis an adequate translation for it, and it also shows the same restriction3:
(16) In my books, John is a nice guy. (17) # In my books, John is in his office.
(18) Tant qu’`a moi, Jean est un bon gars. As that-to me John is a nice guy ‘Inmybooks,Johnisaniceguy.’
(19) # Tant qu’`a moi, Jean est dans son bureau. As that-to me John is in his office ‘#In my books, John is in his office.’
This special behaviour is the most crucial data point that this chapter will in- vestigate. We will try to come up with an explanation for why verbs like find and trouver are sensitive to the subjective character of their complement clause, and see what broader conclusions about the grammar we can derive from this analysis. Inordertoeasilyrefertoiteasily later on, we can simply state the basic observa- tion to be explained in the following way:
(20) Subjectivity Requirement: The complement of an opinion verb must be a subjective statement.
In order to account for this generalization inside a formal framework, we first need to get an idea of how to formally distinguish between subjective statements and non- subjective ones. In the next section, we examine some recent proposals that aim to do just that.
3Some speakers of French I have consulted have in their grammar an almost word-for-word translation of the English phrase, namely dans mon livre (`amoi). For those who have it, this PP is also only compatible with subjective propositions.
146 3 Modeling Subjectivity
Subjective statements are not typically included in the set of sentences for which formal semantics has something to say. Lasersohn (2005) recently rekindled interest in this question and much discussion has taken place since. In this section, we will examine some recent proposals concerning how to inte- grate subjective predicates in the study of formal semantics, starting with Lasersohn’s proposal. Some counterarguments and competing theories will be examined, but ulti- mately we will only be left with two contenders, namely Lasersohn’s original proposal, and an alternative making use of judge pronouns proposed by Stojanovic (2007).
3.1 Lasersohn (2005)
It is an obvious problem of model-theoretic semantics that it is tailored specifically for statements that concern simple matters of fact, not subjective statements. This is of course a natural consequence of extending to natural language the interpretive mech- anisms that were originally proposed for artificial logical languages, which contained no such subjectivity as a matter of design. Lasersohn (2005) made a very influential proposal to integrate such statements into linguistic theory. His claim mosty concerns adjectives which he dubs “predicates of personal taste”4,suchasfun and tasty 5, but naturally extends to other categories like nouns such as success and idiot,aswellas to verbs like stinks or rocks. The basic problem posed by such predicates for ordinary semantic theory is that they give rise to an intuition of “faultless disagreement” (K¨olbel (2004)). That is, if speaker A says that something has property P, where P is a predicate of personal taste, and speaker B answers that no, it is not P, we have the intuition that both speakers can in fact be right, as long as they are being honest about their opinions. The following short dialogue illustrates this property:
4Lasersohn explicitly rejects the simpler expression “subjective predicates” to avoid committing himself to the idea that all forms of subjectivity are to be encompassed by his analysis. In particular, he avoids predicates like beautiful, which are subject to study in other fields of inquiry like aesthetics. I will have no such qualms in this chapter, as I do not believe that such considerations are of any rele- vance for the linguist interested in natural language. I will thus use the terms “subjective predicate” and “predicate of personal taste” interchangeably to pick out the relevant class of predicates. 5There is no actual test proposed in the literature for strictly defining predicates of personal taste. The notion of faultless disagreement seems to be an important property, but it fails to distinguish these predicates from ordinary gradable predicates. A better test will be proposed in section 10, namely the persistence of faultless disagreement in comparative constructions.
147 (21) A: This ride was fun! B: No, it wasn’t!
This intuition of faultless disagreement breaks down into two parts:
1. Both A and B’s statements are true, as long as they are being honest about their opinion.
2. A and B are expressing a disagreement.
We could account for intuition 1. only in a very simple manner, which would be to claim that B’s statement is not actually the perfect negation of A’s statement because it also differs in the value of a hidden pronoun which refers to what we will call the judge of the sentence. This pronoun would be an argument of the predicate of personal taste, and can even be expressed overtly:
(22) A: This ride was fun for me! B: No, it wasn’t for me!
However this leaves little room to account for intuition 2. If A and B’s statements actually differ in the value of some covert pronoun, and if A and B know this, there should be no disagreement at all. Indeed, this is obvious from the fact that if we do pronounce the PPs, as in (22) then we no longer get an intuition of disagreement. Rather, this dialogue as it is represented is not felicitous. We can conclude from this difference that (21) and (22) are not merely distinguished in whether the PP is covert or overt. Lasersohn (2005) rejects many other potential analyses which we do not discuss here and finally proposes a way to resolve this paradox. His solution is to treat judges not as arguments of predicates of personal taste, but rather as an index analogous to world and time indices. This means that, using Kaplan (1989)’s vocabulary, the content of a sentence will now be said to be a set of
148 any judge indices j and j’, [[ P ]]w, t, j =[[P]]w, t, j’. This means that for example, a non-subjective predicate like dead will always have the exact same extension across judges for a fixed world. Of course, this is necessary to restrict the relativism that we are introducing to subjective statements only. Without it, the truth-value of any proposition could be treated as judge-dependent. To make things simple, we will say that predicates of personal taste like tasty simply denote sets of individuals, whose value can vary across judges because they are subjective. In doing this, we ignore the very obvious fact that these adjectives are gradable, witness the fact that they can modified by adverbs like very, and can very naturally participate in comparative constructions.
(23) This ride was very fun. (24) My chili is tastier than yours.
This is a useful simplification and it follows more closely what Lasersohn proposes, but it will have to be refined, a task that we will undertake in section 10, where we will contrast ordinary gradable adjectives and genuine predicates of personal taste. Formally, judges are provided by the context, much as context provides various other pieces of information, such as the time, world, speaker, etc. In most uses, the judge will be set to the speaker, yielding what Lasersohn calls an ‘autocentric’ perspective, but this is not necessary, just common. We can just as well take an ‘exocentric’ perspective which means that we speak while referring to someone else’s opinions. An example of this from Stephenson (2007b) is (25):
(25) This catfood is tasty.
The most natural way of understanding this sentence is that it claims that the catfood is tasty for the cat, not for the speaker, since this is the most plausible situation given our knowledge of the fact that catfood is usually not eaten by people. This exocentric reading thus simply results from the fact that context provided some specific cat or cats in general as the judge for this sentence, rather than the speaker, a universally available option for Lasersohn. This system helps us account for intuitions 1 and 2 together in the following way. Both A and B’s statement in (21) can receive different truth-values because although they are interpreted at the same world and time, the value of their judge index is different. Since the lexical item fun is subjective, its value varies across judges, so its
149 extension is potentially different in the interpretation of A’s sentence and that of B. So in the two formulas below, while the interpretation of the NP this ride is constant, that of fun is not. The two claims that these sentences make can thus both be true at the same time, since fun denotes potentially different sets at the two indices.
(26) [[ (21A) ]] = T iff [[ this ride ]]w, t, A ∈ [[ fun ]]w, t, A
(27) [[ (21B) ]] = T iff [[ this ride ]]w, t, B ∈/ [[ fun ]]w, t, B
The last necessary innovation is Lasersohn’s characterization of the intuition of disagreement. He claims that there is disagreement between two speakers A and B if the content of what A says is the complement of the content of what B says. In our example, the presence of sentential negation in the B sentence will ensure that the content of A’s sentence will be the complement of the content of B’s sentence. Any
3.2 An Attempted Extension to Epistemic Modals
Stephenson (2007a) notices a similarity between the behaviour of predicates of per- sonal taste and that of epistemic modals. Epistemic modals have a meaning that is often represented as involving phrases like “according to what is known” and similar somewhat imprecise expressions (cf. Kratzer (1977)):
150 (28) John might be in his office. = In some world compatible with what is known in the actual world, John is in his office.
There are some parallels between the understood “thinker” in sentences with epis- temic modals and judges in sentences with predicates of personal taste. First, both can apparently be shifted by embedding epistemic verbs like think. For example, in (29), it is John’s opinion concerning Mary’s beauty that matters. The sentence is not used to make the claim that John thinks that Mary is beautiful for me, the speaker.
(29) John thinks that Mary is beautiful.
Similarly, (30) contains an epistemic modal embedded under think, and the one whose knowledge matters for evaluating the modal claim is the subject of the matrix verb John, not the speaker.
(30) John thinks that Bill might be in his office.
Secondly, the identity of the relevant individual, the thinker and the judge, can be shifted by an appropriate PP in both cases. Predicates of personal taste usually require a for-PP or a to-PP, while epistemic modals need an according to-PP.
(31) This ride was fun for me. (32) According to Bill, John might be in his office.
A small detail not present in Stephenson (2007b) is that for and to PPs are arguments of the predicates of personal taste, but according to PPs are adjuncts to the full clause. We can see that the latter can be used to change not only the identity of the thinker of an epistemic modal, but it can also shift the judge for a subjective predicate, strengthening the hypothesis that shifting of the judge and of the thinker are one and the same thing.
(33) According to John, Mary is beautiful.
Finally, in a case where multiple embeddings are present, the only individual whose tastes or knowledge matter is the subject of the closest embedding verb. In the examples below, the subjective predicate or epistemic modal is bolded, and so is the individual whose taste or beliefs is relevant:
151 (34) Mary thinks that Sam thinks the dip is tasty. (35) Mary thinks that Sam thinks it might be raining.
Stephenson proposes to take these similarities seriously and suggests that epistemic modals simply involve the beliefs of the judge. An initially plausible way to cash this out would be to simply make might sensitive to the judge index of the sentence by replacing “what is known” in our original denotation with “what the judge knows”6:
(36) [[ might ]]w, t, j = λp
Although this is a tempting proposal, especially as it fills a gap in the meaning of epistemic modals, it cannot be the entire story. This is because while the shifting of the judge parameter below think is optional in the case of predicates of personal tastes, it is obligatory for epistemic modals. To show this, Stephenson presents the following scenario and sentence (37) used to describe it. Scenario 1: A cat is eating his food ravenously, from which John concludes that the cat must like it.
(37) John thinks that the catfood is tasty.
This sentence can be used felicitously in this context, since it can be used to make the claim that John thinks that the catfood tastes good to the cat, not to himself. Of course it could also be used to make the claim that John finds the catfood tasty himself, but it crucially does not need to mean this. Now consider the following scenario: Scenario 2: A dog is looking intensely at a box in which he obviously expects to find the cat hidden, while John knows that the cat is in fact elsewhere.
(38) #John thinks that the cat might be in the box.
(38) cannot be used to express that John thinks that it may be compatible with what the dog knows that the cat is in the box, which would be the result of shifting the judge, hence the thinker for might. Worse yet, we can construct examples where both an epistemic modal and a pred- icate of personal taste appear below find and see that the thinker and the judge can
6The type i stands for intervals in time.
152 be distinct individuals. In (39), we see that the thinker for might can be understood as the matrix subject John, while the understood judge for tasty is the cat.
(39) John thinks that the catfood might be tasty.
So far, this approach is thus too permissive. It predicts that shifting of the knower for epistemic modals and the judge for predicates of personal taste by verbs like think should be identical, yet it is obligatory for the former, and only optional for the latter. On the face of it, there is no reason to believe that might and tasty should behave any differently in this respect. In order to account for this difference and still maintain some parallel between epistemic modals and subjective predicates, Stephenson suggests a radical change from Lasersohn’s theory and suggests that we need both covert pronominal complements to predicates of personal taste, as well as a judge index. Subjectivity can then enter the picture in two different ways: an element may be inherently judge-dependent and thus get its value directly from the judge-index, or it can combine in the syntax with a special covert pronoun PROj whose value is provided by the judge index, rather than the ordinary variable-assignment function.
(40) [[ PROj ]]w, t, j =j
In order to account for their distinct behaviours, Stephenson claims that epistemic modals are inherently judge-dependent, essentially in the way presented in (36), while predicates of personal taste are treated as two-place predicates, whose internal (judge) argument can either be PROj, or ordinary pro, whose value is provided in the ordinary way, namely by the variable-assignment function. This means that while the identity of the thinker for an epistemic modal is predicted to always be fixed by the judge index, the identity of the judge for a subjective predicate should be free to pick out a different individual. In other words, predicates of personal taste can end up being independent from the judge index, since they are only judge-dependent when they combine with PROj, but epistemic modals are always judge-dependent, period. Crucially, this allows Stephenson to explain the difference between (38) and (37). In both cases she argues, the judge index is switched to John by the matrix verb think. Since she claims that the thinker is set directly by the judge index, and that think obligatorily shifts the judge index, then the sentence can only be understood with John as the thinker. There is simply no alternative, and the sentence is judged infelicitous
153 in the scenario which requires the dog to be the thinker. In (37), the judge index is also switched to John by the matrix verb think, but the subjective predicate has the option of disregarding the judge index by making use of the ordinary pro as its internal argument. Thus Stephenson claims that altering the judge index obligatorily alters the knower for epistemic modals because these modals are obligatorily dependent on the judge index, but it only optionally affects subjective predicates because they only optionally depend on the judge index. In practice, it will be impossible to see the effect of switching the judge index on subjective predicates under Stephenson’s analysis, since this will always have the possibility of being made irrelevant by merging pro instead of PROj, thus effectively insulating the subjective predicate from the judge index. This analysis also comfortably extends to the cases like (39) where an epistemic modal and a subjective predicate both appear together below think,witha different individual serving as judge and thinker. Again, the identity of the thinker will obligatorily be affected by think switching the identity of the judge index, while tasty is free to avoid depending on the judge index by making use of ordinary pro.
The addition of this element PROj in the theory ensures that we can still account for the intuition of faultless disagreement, since the syntactic constituent made up of a predicate of personal taste along with PROj will have the exact same interpretation as the subjective predicate by itself in Lasersohn (2005)’s analysis. His characterization of faultless disagreement can thus be imported to Stephenson’s approach without change. This analysis is built to preserve Lasersohn’s results while also extending the explanatory power of the theory to epistemic modals. Since epistemic modals and predicates of personal taste do not get their distinguished individual, judge or thinker, shifted in quite the same way, and the individual for the two can be distinct inside a single clause, then we must conclude either that think does not actually shift the judge- index, or that their subjectivity can come in two distinct forms at LF. Stephenson takes the second option. It should be clear that this entire discussion is heavily dependent on the hypothesis that think obligatorily shifts the judge index of its complement clause, a claim which is made without much justification. Allegedly judge-dependent elements below it that fail to have their subjectivity shifted to the matrix subject must then be explained in some other manner, opening the way to her suggestion that there are two ways to be judge-dependent. However, it is unclear to me that this is a justified position to
154 take. A comparison between think and find reveals that the latter verb is a much more credible candidate for obligatorily shifting the judge index of its complement. If we replace think with find in (37), yielding (41), the resulting sentence no longer has the option to refer to the tastes of some other individual (i.e. the cat), rather than John.
(41) John finds that the catfood is tasty7.
Why should this be? Under Stephenson’s theory, even if find obligatorily shifts the judge index of its complement, it should be possible for tasty to have as its internal argument pro, resulting in a reading where some salient individual, say the speaker, serves as the judge rather than John. We would need extra stipulation to require that this pronoun cannot be used below find, while it can be used below think. This problem only arises because Stephenson opened the possibility that subjec- tive predicates can have their judge-argument saturated by some covert pronominal whose value is independent of the judge-index, though, which she does to distin- guish epistemic modals from subjective predicates. Putting epistemic modals aside for the moment, in Lasersohn’s original approach, it would suffice to claim that find obligatorily shifts the judge index, and that subjective predicates are inherently judge- dependent, period. Since subjective predicates below think are not affected in this way, there is no reason to believe that this verb truly shifts the judge-index. The necessity for allowing them to not depend on the judge index only arises if we claim that think does this. In other words, contrary to Stephenson, I believe that data like (37) do not show that subjective predicates are not dependent on the judge index, but rather that think does not shift this index. It is only if we want to claim that the identity of the thinker for might is fixed by the judge-index that this simple picture must be changed.
7This sentence sounds quite a bit better with a small clause complement:
(1) John finds the catfood tasty.
Generally, speakers whose judgements I have asked are uncomfortable with sentences where find take as its complement a full clause whose VP simply contains a predicative adjective after a copula. I take this to be the result of a blocking effect that makes a competing sentence with less structure more felicitous, since a small clause complement could always have been used in such cases. As pointed out to me by Junko Shimoyama, these sentences get better when the complement clause contains tense and aspect information:
(2) John finds that the catfood used to be tasty.
155 It is not clear to me at all that this is something that we do want to claim. As we will discuss in ample detail over the course of this chapter, find requires its comple- ment clause to have a semantic value that can vary across judges, a straightforward rewording of the Subjectivity Requirement in (20) using the notion of judges. In principle, this verb can thus be used as a simple test to evaluate whether a clause is subjective or not. Trusting this test without justification for the moment, we can see that while some modals do seem to have a judge argument (as discussed already by Sæbø (2009)), might is not one of them. This is illustrated in the contrast between (42) and (43). Skipping most details, in (42), the modal should intuitively refers to an accessibility relation related to John, while no corresponding reading that makes John the thinker for might is available for (43). Instead, this sentence sounds degraded to most of my informants.
(42) John finds that Bill should quit. (43) #John finds that Bill might quit.
Another problem with claiming that might is judge-dependent is that it predicts that intuitions of faultless disagreement should be just as strong with epistemic modals as with subjective predicates. This is not the case, as the following example shows (closely resembling a similar example from MacFarlane (2004)):
(44) A: John might be in his office. B: No, you’re wrong/he can’t be. I just saw him outside.
Although it is clear that A and B are expressing a disagreement, it is not quite so clear that this disagreement is faultless. If A’s claim were really only expressing a possibility relative to A’s knowledge, then no amount of new information should be able to sway A into believing that her original claim was wrong, at the moment where it was made. Yet, the more natural natural reaction on the part of A following B’s answer is to back down, not to maintain the truth of the first claim in its original context.
(45) A: OK then, I guess I was wrong. A: #Oh, OK. So he can’t be in his office. Nonetheless, when I said “John might be in his office,” what I said was true, and I stand by that claim.
Judgements concerning disagreement in modal contexts are notoriously fickle, and they clearly allow more options than is expected on the simple view that individual
156 whose knowledge is relevant in epistemic modals can be equated with the judge-index (see Von Fintel and Gillies (2008) for much recent discussion of this issue). Thus, epistemic modals do not present the crucial tell-tale symptoms of judge-dependency that are the intuition of faultless disagreement and the licensing of find. There seems to be an interesting relation between, on the one hand, the subject of think and the thinker involved in might, and on the other, between the subject of find and the judge of subjective predicates. However, since think does not obligatorily shift the judge argument of subjective predicates and find does not “pick up” the thinker of might, the argument for claiming that judge-dependency and dependency on a relevant thinker can receive a unified explanation seems to me weak at best. Note also that according to PPs, which were presented above as shifting both the knower and the judge, do so obligatorily for epistemic modals, but optionally for predicates of personal taste, mirroring the behaviour of think. In (46) the catfood can be claimed to be tasty for cats rather than for the veterinarian while in (47), the person whose knowledge is relevant cannot be anyone else than John.
(46) According to my veterinarian, this catfood is tasty (= for cats). (47) According to John, Bill might be in his office (= given what Mary knows.).
Lasersohn’s approach includes the possibility of exocentric readings, in which the context provides a different judge than the speaker. We can thus simply treat the apparent cases of optional shifting of the judge index of predicates of personal taste as cases where the context provides as the judge the individual denoted by the subject of think or the complement of according to. In examples like (29), where the understood judge is the subject of think and (47), where it is introduced by according to,the individual John and his mental state are made salient by the sentence, so it should come as no surprise that context would naturally ascribe him as the judge under the most accessible reading. There is no need to insist that think shifts the judge-index to get this. For these reasons, I reject Stephenson’s claims that 1- subjective predicates are indirectly judge-dependent through the value of their PROj argument, and 2- that might is judge-dependent at all. For the moment, then, I see no reason to change Lasersohn’s original proposal, and I thus maintain his purely indexical view of judges, since it is entirely satisfactory to deal with the behaviour of subjective predicates in embedded and unembedded contexts. The behaviour of might, especially in embedded
157 contexts, is simply too different from that of subjective predicates to justify a unified analysis and the complications that it raises for a theory of subjectivity in natural language.
3.3 Cotextualism: Disagreement is Never Faultless
Lasersohn (2005) explicitly rejects the covert pronoun analysis of judge-dependency because it can only account for the intuition that both speakers can simultaneously be right in a faultless disagreement dialogue like (21), not the intuition that there is disagreement between them. Despite this, Stojanovic (2007) argues that we cannot choose between a theory which represents judges as indices and one which represents them as arguments to predicates of personal taste based on simple semantic data. She argues that the major piece of evidence that Lasersohn (2005) provides against the implicit argument approach, namely the intuition of faultless disagreement, is not valid because there is simply no such thing as faultless disagreement. That is, disagreement is always a matter of verifiable fact and, in the cases where it seems not to be, we are rather dealing with a case of misunderstanding, not disagreement. Once this piece of data that is crucial for Lasersohn is put out of the way, it is no longer possible to choose between an index theory and a pronoun theory of judge-dependency. To show this, Stojanovic proceeds to build two toy languages, one where judges are introduced as syntactic arguments to predicates of personal taste and one where they are introduced as parameters of the interpretation function, and then shows that we can come up with a translating procedure to derive one from the other, thus concluding that the two languages are equivalent, and that arguments in favour of one or another must come from other sources than such simple semantic data. Her argument makes heavy use of the fact that if a speaker utters a simple sub- jective statement like those we have been looking at, such as (21A) for example, the hearer cannot know (without any additional contextual clues) whether the speaker meant this statement to be taken as the expression purely of her own opinion on the ride, or if the statement was meant to express a more general claim, say that the ride was fun for most people who were on it, which one could presumably establish by taking a survey of those people. Since A’s statement did not explicitly state who this ride was claimed to be fun for, B may infer either that the former or the latter was A’s intention. However, B can only reasonably disagree with A’s statement if he believes
158 that what A meant was that the ride was fun for most people on it. That is, she claims that subjective statements without any overtly expressed judge are ambiguous between a generic and an personal reading, and that disagreement is only possible in the former case, not the latter. Given that we can generally admit that predicates of personal taste can be de- scribed as exhibiting the property of first-person authority, which simply means that speakers are always well-informed about their own states of mind, including their beliefs, opinions. etc., B would be foolish to try to contradict A’s statement if it were meant to express A’s opinion only. If this were the thought that B expressed, we should get the intuition that B claimed that A lied, as is the case if a speaker contradicts another person’s belief-report:
(48) I believe that John is in his office. (49) No, you don’t! You know very well that he is with his mistress!
In contrast, if we make this ambiguity disappear by clearly expressing that we are only talking about our own opinion, any trace of disagreement disappears, as she shows with the following example in response to the claim “Paris is lovely”, similar in spirit to the examples with overt judge arguments for the predicate of personal taste used above (cf. (22)).
(50) Tarek: OK. I find Paris lovely, that’s all I’m saying. (51) Inma: OK, and I don’t find Paris lovely at all; that’s all I’m saying.
Stojanovic thus agrees with Lasersohn that expressions of personal taste are fault- less, at least when they are not used generically, a property which she assimilates to the more general phenomenon of first-person authority, but she denies that there can be such a thing as faultless disagreement. If a speaker A disagrees with another speaker B’s subjective claim, there are only two logical possibilities that make for a sensible conversation. The first possibility is that B believed A’s claim to be meant to be general, in which case the disagreement concerns a matter of fact. One could take a survey of the people on the ride to see if they found it fun. If most of them did, A is right. If not, then B is right. This disagreement is not faultless at all, since it can be established who is right and who is not. On the other hand, if B believes that A’s statement was meant to express her own personal opinion and he nevertheless contradicts it, then he is literally saying
159 that A’s statement does not reflect A’s own opinion, say that she is lying or does not remember the situation well. Using find as an unambiguous expression of a personal opinion, (52) shows this kind of meaning more clearly.
(52) A: I found this ride fun. B: No you didn’t, you complained the whole time that you were going to be sick!
This should be again a matter of fact (even if it is a difficult one to establish), and disagreement should still not be faultless. The last logical option is for both A and B to be making personal claims about their own taste. This is exactly the situation that Lasersohn takes for granted is the standard interpretation for dialogues like (21). Stojanovic, in contrast, claims that this is not a possible dialogue, or at least not a sensible one. If A meant her claim to be personal, B recognizes this, and B also meant his claim to be personal, then, she claims, the dialogue would be infelicitous. Lasersohn and Stojanovic do not give the same description of how we understand dialogues like (21). For Lasersohn, it clearly expresses a disagreement where both speakers are nevertheless making true claims about themselves. For Stojanovic, it is ambiguous and normally can only give rise to disagreement if the initial claim was meant to be generic, and if the second one was meant to challenge it. The fact that disagreement disappears when the ambiguity is removed by overtly stating who the relevant judge is meant to show that the apparent disagreement was, in fact, only a misunderstanding. In principle, only one of these descriptions should fit native speaker intuitions. As Stojanovic herself recognizes, however, it is extremely difficult to see which of these descriptions is the best fit, since the judgements are too subtle. Must A have meant that the ride was fun for most people on it for B’s statement to be a felicitous reply (Stojanovic), or is it felicitous even if A was only making a personal claim (Lasersohn)? If it were possible to answer this question with any degree of certainty, we could establish if there is such a thing as faultless disagreement, and thus whether it can serve as an argument for the index theory of judges against the pronoun theory. Since these judgements are too unclear or too weak, however, we cannot use faultless disagreement as an argument for either view and we are led to look elsewhere in the language for clues in favour of either.
160 Sæbø (2009) means to use the behaviour of find to do exactly that. He claims that we can only account for the behaviour of this verb and the restrictions that we find inside its complement if we accept the pronoun theory. In section 4, we will see what his arguments are for this view, and the rest of this chapter will largely be concerned with providing a counter analysis in terms of an index-theory of judge-dependency.
3.4 Summary
Lasersohn (2005)’s theory is very straightforward. It simply says that sentences are interpreted relative to a distinguished individual, the judge, in addition to the less controversial world and time parameters, and that subjective predicates are those whose extension can vary across judges. Stojanovic’s contribution is to claim that there is no way to tell this story apart from a theory without judge-indices but with judge pronouns, at least from the simple data that we usually look at, which mainly consist of so-called dialogues of faultless disagreement. She poses the challenge to linguists to find some new arguments and more fine-grained data to help us distinguish the two approaches, and it is exactly this that Sæbø (2009) attempts to do. His paper aims to take up this challenge and show that the syntactic and semantic properties of the verb find and its correlates in various languages argue in favour of the pronoun theory. The rest of this chapter is devoted to carefully reexamining the properties of find. Much new date will be presented and we will see that Saebø’s conclusion is unwarranted. This chapter is thus meant to be a comprehensive analysis of the syntax and semantics of find, but I also mean it to be a contribution to our understanding of how best to model subjectivity in a formal framework. That is, although I disagree with Saebø’s conclusions, I agree with him that opinion verbs constitute a rich testing ground for any theory of subjectivity in natural lan- guage. It is, as far as I can tell, the best place to start looking to take up Stojanovic’s challenge. Given our similarities in purpose, I will be making heavy use of Sæbø (2009) and use his analyses as the starting point for mine.
161 4 Summary of Saebø (2009)
The most important data point that we must explain concerning find is the Subjec- tivity Requirement that it imposes on its complement clause, as described in (20). As Saebø shows, both the index theory and the pronoun theory of judge-dependency can account for it. In order to see this, we can take the two following denotations to be reasonable candidates for the contribution of this verb. (53) gives its denotation inside an index theory, while (54) gives its equivalent in a pronoun theory (the latter denotation is precisely the one used by Saebø):
(53) [[ find ]]w, t, j = λφ. λx. φw, t, x (Index theory)
(54) [[ find ]]w, t = λφ
(53) is straightforward. According to it, find takes as its first argument a propo- sition φ, then an individual x, and returns true iff the proposition φ is true relative to x, i.e. if the judge of φ is set to x. (54) requires that the complement clause denote a set of individuals. As long as that condition is satisfied (we will see how below), find can be assumed to be semantically almost null: it takes as arguments a property and an individual and returns true if the individual is a member of this set. In Saebø’s terms, it is nothing more than “semantic glue”. Both these denotations make no place for any epistemic component to the meaning of opinion verbs, making it “radically reductionist”, in Saebø’s terms. That is, the sole function of find is to impose the matrix subject as the judge for the embedded clause, contra Stephenson (2007a), as we will discuss in more detail in section 6. How can we use these denotations to get us the Subjectivity Requirement? In an index theory, the problem with a sentence in which a non-subjective clause follows find (such as (9), repeated here as (55)), is pragmatic.
(55) #John finds Bill dead.
While Saebø is not explicit as to what pragmatic principle would be used to exclude such sentences, we can at least make the case that shifting the judge for the embedded proposition, which is all that find is taken to contribute according to the denotation in (53), is entirely vacuous in this sentence, since there is no item in the complement clause whose value varies across judges. (55) could then be claimed to be infelicitous
162 because the meaning of the full sentence reduces to the meaning of the material in its embedded clause, an uneconomical situation to be sure. Things are quite different in a pronoun theory of judge-dependency. In such an approach, sentences like (55) are ruled out as instances of type-mismatches. Predicates of personal taste have an extra judge argument, which Saebø argues for independent reasons is actually its second argument, coming in after the external argument, making them predicates of type
(56) [[ beautiful ]] = λx.λy. x is beautiful for y
According to the denotation for find in a pronoun theory in (54), find is looking for a complement whose denotation is of the type
(57) Anne finds Mary beautiful. (58) [[(57)]] = T iff [[ beautiful ]]([[ Mary ]])([[ Anne ]]) = T
The small clause Mary beautiful denotes a set of individuals which corresponds to the judge argument of beautiful, when applied to the object Mary.Thissmallclause then serves as a suitable argument for find. Given the denotation in (54), the VP then combines with the subject and returns true iff Anne is a member of the set denoted
8In order to show clearly the important features of the analysis, I remove references to times and worlds, present in Saebø’s paper, from the formalism, since they make no relevant contribution to the question at hand.
163 by the small clause, i.e. those judges for whom Mary is beautiful. The types line up correctly and the truth-conditions are satisfactory. Notice that there is no mention of find itself in (58). This is what Saebø means when he says that the analysis is fully reductionist. For all that semantics cares, the opinion verb might just as well not have been there. In contrast, in an example like (55), the adjective dead only takes a single indi- vidual argument, since it is not judge-dependent.
(59) [[ dead ]] = λx. x is dead
This means that the small clause Bill dead has the denotation of a regular propo- sition, and thus cannot serve as argument for find. The derivation cannot proceed any further and the sentence is judged unacceptable for this reason9. These are the two candidate explanations that Saebø considers for the Subjectivity Requirement, and it is the second approach, where the Requirement falls out as a way to avoid a type-mismatch, that he adopts. It is important to note that he does not claim that the index theory cannot explain the subjectivity requirement on the complement of find, only that it requires additional work to do so. While the derivation of a sentence with this property is literally blocked under Saebø’s semantics, it is only redundant according to the semantics of find in the index theory. It would be possible to claim that non-subjective clauses under find are infelicitous because they reduce semantically to their embedded clause. The matrix verb and subject are simply superfluous, a situation which presumably could be ruled out by pragmatic principles. We would thus need a theory of how redundancy can make sentences infelicitous, under which conditions it does, etc., before we can use this property to exclude sentences like (55). The main arguments that he puts forward for a pronoun theory of judge- dependency concern some special syntactic facts about the possible positions of the predicate of personal taste. We will examine those in section 7-9. For the moment, however, we turn to a third possible analysis of the Subjectivity Requirement that Saebø does not consider. This explanation is based on data con- cerning the presuppositional content of find sentences that are absent from Saebø’s
9Presumably, the sentence is judged infelicitous rather than purely ungrammatical because as we hear such a sentence, we attempt to coerce the predicate dead into something judge-dependent. It is the failure to make any sense of such a coerced predicate that accounts for the judgement of infelicity, since a bare, unresolvable type-mismatch should produce a much stronger intuition of ungrammaticality.
164 paper, and we will see that it makes the type-based approach that he suggests unnec- essary. In fact, the explanation that we will propose is compatible with either view of predicates of personal taste.
5WhatFind Asserts and Presupposes
The radically reductionist character of Saebø’s proposal predicts that using an opinion verb or a for-PP to ascribe a judge to a judge-dependent predicate should yield perfectly synonymous propositions, since both constructions do nothing else than provide the value for the judge argument. This is not quite right:
(60) John finds that Mike gave a great class yesterday. (61) Mike gave a class that was great for John yesterday.10
Although these sentences do mean roughly the same thing, they are distinct in that in (60), the fact that Mike gave a class at all, great or not, is presupposed, while this information is asserted in (61). Saebø hints at this in his section 4.3, claiming that everything in the complement clause but the predicate of personal taste “can be presupposed” (p. 340), although he does not examine this idea further. The observation is also present in Ducrot (1975) concerning the complement clause of the french verb trouver. He gives the following contrast as proof:
(62) Je trouve qu’il a eu tort de faire cela. I find that-he has had wrong to do this ‘I find that he was wrong to do this.’
(63) # Je trouve qu’il a eu le tort de faire cela. I find that-he has had the wrong to do this Sentences with the phrases avoir tort de and avoir le tort de both include a claim that the subject did whatever is said in the complement clause, and a claim that the speaker judges this action to be morally wrong. They differ in that with the former phrase, the negative judgement is asserted, and the truth of the complement clause de faire cela is presupposed, while the exact opposite is true of the latter
10The for-PP in this sentence sounds a little odd to some speakers. My informants are nevertheless consistent in interpreting this sentence in a way that is expected if the PP provides the judge for the subjective predicate great.
165 expression. The complement clause of avoir le tort de is asserted while the negative judgement is presupposed. What the infelicitous nature of (63) shows is that the subjectivity requirement concerns its asserted content. find,ortrouver in this case, requires its complement clause to assert something subjective. It is not satisfied if it only presupposes something subjective. I want to make the stronger claim that not only can the non-subjective material below find be presupposed, it must be presupposed. This is difficult to see when the complement is a small clause, but is more readily visible when we examine larger complements like the one in (60). It is clear that at least part of the asserted material in this sentence is the attri- bution of an opinion to John. It would be made false if John’s opinion concerning this class were different. In contrast, the fact that Mike gave a class at all yesterday seems to follow indisputably from (60), regardless whether the sentence is true or false. The sentence would not be made false, but rather unusable if this fact is not granted. Indeed, if it is not a shared assumption that Mike gave a class yesterday, it seems appropriate to respond to (60) by saying (64), a reaction typical of not sharing a presupposition (cf. Shanon (1976) and, more recently, Von Fintel (2004)).
(64) Wait a minute, I didn’t know that Mike was teaching yesterday.
This intuition seems to correspond to that of a presupposition of the sentence, as Seabø suggests, and we can confirm this by applying ordinary presupposition projec- tion tests:
(65) John doesn’t find that Mike gave a great class yesterday. (66) Did John find that Mike gave a great class yesterday? (67) If John finds that Mike gave a great class yesterday, you should get him to come to Bill’s.
All three of these sentences entail that Mike gave a class yesterday. It also seems unlikely that this should be an implicature of some sort, since it does not appear to be cancellable:
(68) #John finds that Mike gave a great class yesterday, but Mike didn’t teach at all yesterday.
166 We can make sure that the absence of any presupposition when the complement is a small clause is not a direct consequence of the small clause structure by constructing an example where the complement is a full clause, and yet we find no presupposition, simply because there is no non-subjective matter to speak of11. This can be accom- plished for example by using a judge-dependent intransitive verb like stinks as the licensing subjective predicate. We find that the resulting sentence does not involve any detectable presupposition (other than the existence presupposition introduced independently by the proper name Bill).
(69) John finds that Bill stinks.
How can we characterize the presupposition introduced by find-sentences? A simple but imprecise way to put it is to say that such sentences presuppose the non-subjective component of its complement. Before committing to any specific for- mulation of the presupposition, we will thus focus on examples where the predicate of personal taste is an attributive adjective. In such examples, we can simply construct the presupposition by taking the complement clause and removing the subjective predicate.
(70) John finds that Mike gave a great class yesterday. (71) Presupposition: Mike gave a class yesterday.
This is simply a helpful algorithm and should not be taken as a serious proposal for constructing the presupposition. We will use it to explain why non-subjective sentences below find are not acceptable, and the demonstration will be that much easier from using this simplified algorithm. It will of course be necessary to make this more precise later on, a task that we take up in section 5.2.
5.1 A Presupposition Account of the Subjectivity Require- ment
Here is how we will be describing the meaning contribution of sentences with find. Their meaning is divided in two parts: All the non-subjective material is presupposed, and the assertion is that the complement clause is true with respect to the main
11I thank Luis Alonso-Ovalle for bringing this point to my attention and providing the example that shows it.
167 clause subject as the judge. This assertion is exactly what we had already in our two candidate denotations in (53) and (54). We can take this proposal to be agnostic on the question whether judges are introduced by null pronouns or through the index, although we will later pick a side in favour of the index theory. Since this is the conclusion that we will end up reaching, I will be presenting this section using the formalism of the index theory. I believe that we can use this observation to derive the requirement that the complement of find be a subjective statement without appealing to Saebø’s analysis in terms of a type-mismatch. The idea in a nutshell is that every sentence where the complement of find is not a subjective statement will have as its assertion exactly the same proposition as its presupposition. Pragmatic principles will then exclude this because the sentence expresses no information that is not presupposed by the same sentence. Let us see how this works with an infelicitous example:
(72) #John finds that Bill bought a BMW.
We now construct the presupposition by taking the complement of find and re- moving any predicate of personal taste. In this case, the result is exactly equal to the embedded clause itself, since there is no predicate of personal taste at all:
(73) Presupposition: Bill bought a BMW.
The asserted material is the claim that the embedded clause is true if we assign John as the judge. Since there is no judge-dependent element in the complement, then assigning John or anyone else as the judge is vacuous, and the assertion is thus also semantically equivalent to the embedded clause.
(74) Assertion: (Relative to John as the judge,) Bill bought a BMW.
The problem lies precisely in the equality between (73) and (74). This sentence can only be used if the proposition Bill bought a BMW is already part of the common ground, to put it in the words of Stalnaker (1978), since it is presupposed, but its sole contribution would be to add exactly this proposition to it. (72) is thus ruled out because it contributes nothing to the common ground. We will refine this below, but we can already see that however we formalize the presupposition, the mere observation that the presupposition does exist and that it
168 includes all the non-subjective information in the complement clause makes sentences that do not obey the Subjectivity Requirement at best uninformative. We will now make this analysis more precise by spelling out exactly how the presupposition comes about, and what form it takes precisely.
5.2 Deriving the Presupposition
Of course, the algorithm to construct the presupposition that simply consists in re- moving the predicate of personal taste from the complement clause cannot be right. It would obviously be a strange rule for the grammar to make use of, but it also has the deeper problem of yielding incorrect presuppositions in some cases, and entirely ungrammatical sentences in others:
(75) John finds that Bill behaved strangely yesterday evening. Presupposition = Bill behaved yesterday evening.
(76) John finds that Billy became nervous around 5pm. Presupposition = *Billy became around 5pm.
The most obvious way of constructing the presupposition that I see would be to simply to state that find requires that its complement be true for some judge:
(77) Presupposition : The complement of find mustbetruerelativetosomejudge. (First attempt)
The judge only comes into play in fixing the extension of predicates of personal taste (and some related semantic roles concerning gradable adjectives; cf. section 10), so the only way that a proposition p can be true relative to some judge j is if everything non-subjective about p is true. This is especially obvious in an example like (60), repeated here as (78). The denotation of the embedded clause is given in (79), and if it is true for any judge, then it entails the weaker claim (80), which just says that Mike gave a class:
(78) John finds that Mike gave a great class yesterday.
(79) ∃x. [[ class ]]w, j(x) & [[ gave ]]w, j([[ Mike ]]w, j, x) & [[ great ]]w, j(x)
(80) ∃x. [[ class ]]w, j(x) & [[ gave ]]w, j([[ Mike ]]w, j,x)
169 Since shifting the judge can only affect the extension of great in this sentence, then if the embedded proposition is true of at least one judge, there must be at least one thing that is in the extension of class, something that Mike gave, and is great for that judge. If anything is in the intersection of all three of these properties, then it is in the intersection of the two first properties, which just says that Mike gave a class. The fact that some judge makes the sentence true is a sufficient condition to ensure that the non-subjective part of the complement must be true, but it is not a necessary one. Some sentences with find can be felicitous even if no known or relevant judge yields true for the embedded clause. This is because existentially quantifying over judges limits us to the set of individuals provided by the context, and who serve as potential judges. To see this clearly, consider the following sentence: (81) No one finds Billy smart. The asserted material of this sentence is in direct contradiction with the presuppo- sition as stated in (77). The presupposition says that some judge must yield true for Billy smart, and the assertion says that there is no such judge. The sentence should thus only allow for two judgements: It should be judged false if someone does find Billy smart, since then the presupposition is respected, but the assertion is contra- dicted. Otherwise, if it really is true that no one at all finds Billy smart, it should be judged infelicitous since the presupposition is not true, despite the fact that the assertion is. This does not correspond to speakers’ judgements, who have no problem judging this sentence true if no one relevant in the context or even no one in existence finds Billy smart. It should be obvious now what the problem is. This way of forming the presup- position is limited to using actual people as judges in a way that it has no reason to be. I will thus not adopt this definition of the presupposition, despite its simplicity and intuitive appeal. The problem with the way we have stated the presupposition here is ontological: there isn’t in general a judge for every possible judgement. I believe that what we really want to say is that the presupposition is an existential claim on possible judgements, rather than on actual judges. Broadly speaking, it must be possible to judge the complement clause true. This is the view that I will adopt. Of course, now the challenge is to come up with a formalization of the notion of “possible judgement”. In order to do this, I propose to make use of an accessibility
170 relation Accopinion which, for any given world w, gives a set of worlds such that all the objective facts of w are kept constant, while opinions are allowed to vary freely, which just means that the extension of subjective predicates are allowed to vary, but nothing else is allowed to vary. This corresponds to what we can call the opinion space on w. A possible judgement is simply a pairing of the judge with such a world12.
(82)
I stipulate that the worlds in the opinion space exhaust every possible way of assigning an extension to a subjective predicate that conforms to its domain restric- tions. That is, given a subjective predicate like tasty with a domain restriction that says that only edible things can be part of its extension, for every set S in the power set of its domain, there is a world in the opinion space where the extension of tasty is equal to S. So there is world in Acc(w) where the judge finds every kind of food tasty, and there is a world where the judge finds nothing tasty, and everything in between. Once we have this definition, we can provide the following preliminary definition for the presupposition introduced by find-sentences:
(83) Presupposition for x finds p = ∃w’ [w’ ∈ Acc(w) & pw’, j =T] (To be revised)
12This can be a confusing issue. In Lasersohn (2005)’s fragment grammar, denotations for lexical items are defined at a world, time and judge index. A specific clause of the interpretation function states a given subset of lexical items, those that we call subjective, can have a different value at different judges in a same world and time, while ordinary lexical items do not. Nothing is said however about how subjective items can or cannot vary across worlds for a fixed judge. I see no reason to assume that this should be restricted in any way, meaning that the extension of tasty, for example, should be able to vary across worlds for a given judge. This variation is what my interpretation for Accopinion makes use of. I believe this is basically correct, and even necessary to interpret sentences like the following in a manner consistent with general assumptions about the interpretation of conditionals.
(1) If John found liver tasty, then I would make it every day.
Assuming that conditionals involve quantifying over possible worlds, say for simplicity that they involve a subset relation between the set of worlds denoted by the antecedent and that denoted by the consequent (cf. von Fintel and Heim (2007); § 4.2), then allowing the extension of tasty to vary across worlds for a constant judge makes it possible to interpret (1) as stating that for all those worlds where liver is tasty for John as the judge, I make liver every day. It would not be possible to do this if judges fixed the extension of subjective predicates consistently across worlds. 13 From this point on, I will simply be referring to the special accessibility relation Accopinion as “Acc” to remove clutter from the formalism.
171 Much as with the previous definition that involved quantification over (actual) judges, the presupposition that the non-subjective material in the complement clause must be true follows from the fact that if a proposition can be true for a judge, then it must be the case that every piece of information whose semantic value is not affected by shifting the judge (i.e. the non-subjective component) is true. What we have gained at this point is that this presupposition no longer entails that any actual judge makes the complement clause true. The presupposition corresponding to (81) is now formalized as follows:
(84) Presupposition for (81) = ∃w’ [w’ ∈ Acc(w) & [[ Billy smart ]]w’, j =T]
This no longer entails that any real person in the context is required to judge Billy smart for the presupposition to hold, and the presupposition is thus no longer in conflict with the assertion. The presupposition only says that it would be possible for the judge to find Billy smart, since it is part of the opinion space, while the assertion says that there is no such actual judge (at least in the context). Since every possible opinion is realized in the opinion space, there is necessarily at least one world in the accessibility relation where Billy is considered smart by the judge. One such world is the one where the judge finds absolutely everyone smart, for example. We will further refine this way of building presupposition introduced by find in the next section in order to fix an empirical challenge concerning embedded negation, and this will turn out to make the presupposition also enforce the Subjectivity Re- quirement in a more convincing way than what we have suggested so far, in addition to making sure that the non-subjective material is taken for granted.
5.3 Negation in the Complement Clause
An unusual effect emerges when the complement clause of find is negated. We still find a presupposition, but it turns out to be exactly the same as that of the corresponding sentence without embedded negation.
(85) John finds that Billy didn’t write a great book. (86) Presupposition = Billy didn’t write a book. (87) Presupposition: Billy wrote a book.
This is not expected given the way we have defined the presupposition so far. If we blindly apply the method outlined in (83), we get the following presupposition:
172 (88) ∃w’. [w’ ∈ Acc(w) & [[ Billy didn’t write a great book ]]w’, j =T]
This formula turns out to be a tautology because of the way we have constructed the accessibility relation. It would be made true in any situation where Billy didn’t write a book at all, and if Billy actually has written a book, then there will always be a possible judgement that makes it not great, making the proposition true in this case too. If this were the correct way of formalizing the presupposition, then this sentence should be understood as having no (detectable) presupposition at all, since a tautology does not restrict the common ground in any way. This is wrong, however. (85) does have a detectable presupposition, and it states that Billy wrote a book. This effect is consistent with embedded sentential negation, regardless where the subjective predicate appears:
(89) John finds that Mary didn’t sing well yesterday. (90) Presupposition: Mary sang yesterday. (91) John finds that Billy didn’t behave strangely at the party. (92) Presupposition: Billy behaved in some manner at the party.
Complement clauses with negative quantifiers give rise to more complex patterns of presuppositions and we will return to them in the next section. The most straightforward way of modifying the presupposition in a way that makes sure that embedded negation is ignored is to modify the definition in (83) by adding a conjunct that says that there must also be a possible judgement that yields false for the complement clause14. Since this definition involves a special kind of contingency, namely that it it is not the case that every judge makes the complement true or that every judge makes it false, I will call this the Subjective Contingency Presupposition.
(93) Subjective Contingency Presupposition (Final version):
Presupposition for x finds p =(∃w’. [w’ ∈ Acc(w) & pw’, j =T])&(∃w”.
[w” ∈ Acc(w) & pw”, j =F])
This new definition states that it is possible for a judge to yield true for the complement clause, and it is also possible for a judge to yield false. An immediate
14I thank Bernhard Schwarz for suggesting this to me.
173 advantage of the Subjective Contingency Presupposition is its intuitive appeal. Basi- cally, it states that both judgements concerning the complement clause, positive and negative, are in principle available. This can be understood as a minimal usability condition: if only one judgement were possible for a proposition p, then it would be pointless to ascribe a judge for this proposition, since it would yield the same value as any other judge. Indeed, this is intuitively exactly what being subjective is all about. Another advantage of this approach is that it gives us a better way of excluding non-subjective propositions under find. So far, we had to be satisfied with claiming that (72), repeated below as (94), was judged infelicitous because it contributed noth- ing to the common ground. In our earlier discussion, I claimed that the problem with it was that its presupposition was the same as its assertion.
(94) #John finds that Bill bought a BMW.
Clearly, this makes the sentence uninformative, but should we assume that this is enough to make it infelicitous? As observed by many authors and discussed explicitly in Gajewski (2002), among many others, it is not sufficient for a sentence to be truth-conditionally vacuous to become unacceptable. Many standard tautologies and contradictions are obviously acceptable, if perhaps not communicatively very useful.
(95) A bachelor is a married man. (96) It’s raining and it’s not raining.
Our earlier definition of the presupposition faced the same challenge for excluding (94). If all that was wrong with this sentence was that it contributed nothing to the common ground, then it would be predicted to be exactly as bad as (95), which is to say not very bad at all. Yet there is an obvious contrast between the two, with the former being the only one that we really would want the grammar to rule out. Now let us see what the Subjective Contingency Presupposition yields as the presupposition for (94):
(97) (∃w’. [w’ ∈ Acc(w) & [[ Bill bought a BMW ]]w’, j =T])&(∃w”. [w”∈
Acc(w) & [[ Bill bought a BMW ]]w”, j =F])
This formula says that there is a possible judgement that makes Bill bought a BMW true, and another judgement that makes it false. Since every item in the embedded proposition is judge-independent, then this cannot be the case because every item will
174 have the same extension at w’ and w”, and indeed the same as in the actual world w, since the accessibility relation is precisely meant to include only possible worlds in which items like BMW and buy have the same value. From this we can conclude that the presupposition is itself a contradiction, which is of course not an acceptable pragmatic situation. It cannot be part of the common ground, since this would entail that the common ground is essentially reduced to the empty set15.Thisisastronger case for making the sentence infelicitous. There is no situation where this could be a useful contribution to any discourse. Given these advantages, we will maintain this line of thinking, and we now turn to the more complex case of negative quantifiers below find.
5.4 Negative Quantifiers below Find
Negated embedded propositions and their unnegated correlates discussed in the pre- vious section could be described as involving the same presupposition on the basis of native speaker intuition. Things are a little more complicated with negative quanti- fiers. Intuitively, what is the presupposition introduced by (98)?
(98) John finds that nobody gave an interesting talk.
Does this sentence have the weak presupposition that somebody gave a talk, or a stronger presupposition that everyone of some relevant set of people gave a talk? Judgements I have asked on similar examples are somewhat inconclusive. Let us first see what the Subjective Contingency Presupposition predicts. Applying the formalism gives us the following formula as the presupposition for (98):
(99) (∃w’[w’ ∈ Acc(w) & [[ nobody gave an interesting talk ]]w’, j =T])&(∃w”[w”
∈ Acc(w) & [[ nobody gave an interesting talk ]]w”, j =F])
The first half of this formula is a tautology for the same reason that (88) was a tautology: it would be made true if nobody gave a talk at all, or if the standard for great were set high enough to exclude every talk, which is always part of the opinion space. As such, it doesn’t tell us anything. The second half tells us that it is a
15I am assuming, following Stalnaker (1978) and many others since, that the common ground is modeled as a set of worlds, and that adding a new sentence to it simply means intersecting the set of worlds denoted by this new sentence to the original set. Since a contradiction in possible worlds terms always denotes the empty set, adding it to the common ground means intersecting the common ground with the empty set, which results in the empty set.
175 possible judgement that nobody gave an interesting talk is false, which only entails that somebody gave a talk, not that everyone did. So the prediction that we make is clear: only the weak existential presupposition should be found. So now we have a clear prediction from the theory, but the actual facts are murky. Let us look at examples where the restrictor of the negated universal quantifier is explicitly provided:
(100) John finds that none of his roommates gave an interesting talk.
Judgements on this sentence lean more towards a universal presupposition than a mere existential one. Yet this sentence is almost structurally identical to the previous example (98), whose status was not as clear. This issue is reminiscent of a well-known debate in the literature concerning pre- supposition triggers that appear in the scope of a quantifier. Chemla (2009) discusses the following example:
(101) No student knows that he is lucky.
There is disagreement concerning whether this sentence has the universal presup- position that every student is lucky, or only the weaker presupposition that some student is lucky. The interesting thing to note is that judgements become notice- ably crisper in such examples too when the restrictor of the quantifier is made more specific:
(102) None of my Semantics 1 students knows that they are lucky.
Again, making the restrictor more precise favours the universal presupposition reading. This is an interesting parallel and it strongly suggests that whatever is re- sponsible for strengthening and making clearer the presupposition in the better-known environment of presupposition triggers like know is also responsible for the same effect that appears with find. I will not go in detail into this issue here since it would take us too far afield16, and because I believe that some additional work, perhaps exper- imental, would be necessary to establish exactly when it is that the presupposition
16An obvious difference between the data that I introduce in this section and those discussed in Chemla (2009) and others is that my problematic examples involve a quantifier that stands inside the complement clause of the presupposition-introducing verb, while the others involve a quantifier in the matrix binding a pronoun in the complement clause. Nevertheless, the similarity between examples like (98) and (100) on the one hand, and (101) and (102) on the other, makes a unified analysis of both phenomena highly desirable.
176 is strengthened into a universal claim before any analysis of the phenomenon can be proposed.
5.5 The Presupposition Account vs. the Type-Mismatch Ac- count
The analysis of the Subjectivity Requirement that I provided in the previous sections contrasts heavily with that made by Saebø. So far, I have not shown any way in which my proposal is superior to his. We can minimally make the conceptual argument that since we have to account for the presence of a presupposition concerning the non-subjective component of the complement of find, the fact that we can make the Subjectivity Requirement follow from it makes it an economical solution. Of course, if Seabø is right that judges are not introduced by indices but rather as complements to predicates of personal taste and if this can be justified independently, then non- subjective statements under find will always result in a type-mismatch and mere economy will not be able to tease the two accounts apart, since both explanations will essentially “come for free”. In fact, nothing precludes that the sentences that we want to exclude may be bad for more than one reason, both because the types do not line up correctly and because their presupposition is a contradiction. Below, I will argue that the presupposition account is superior in at least one way, namely in correctly ruling out trivial propositions under find.
5.5.1 Trivial Propositions under Find
Here is an initial expected contrast: the presupposition account predicts that comple- ment clauses with contradictory truth-conditions should be unacceptable under find even if they contain a subjective predicate, for the same reason that non-subjective clauses are out. This is because in both cases, the presupposition will be be a con- tradiction, a situation that we have argued is judged infelicitous when it arises. This prediction is borne out, as the infelicitous nature of (103) shows.
(103) #John finds that Mary is pretty and (that) she isn’t pretty.
(104) (∃w’. [w’∈ Acc(w) & [[ Mary is pretty and she isn’t pretty ]]w’, j = T]) &
(∃w”. [w”∈ Acc(w) & [[ Mary is pretty and she isn’t pretty ]]w”, j =F])
177 Before analysing this example in detail, we can rule out the hypothesis that this sentence is infelicitous because the subjective predicates appear in conjoined clauses under find. The following example shows this, since here the two conjuncts are not in any entailment relation, and the sentence is judged felicitous, despite the fact that the predicates of personal taste are still in conjoined clauses.
(105) John finds that he saw an interesting talk and had an enjoyable dinner.
(104) gives the presupposition for (103). What this formula says is that there is a possible judgement according to which Mary is pretty and she isn’t pretty is true, and one for which it is false. Of course, this cannot be true for a single judge in a constant world. The first part of this claim, that there is such a judge, is a contradiction, so the fact that the second half is a tautology is inconsequential. The entire formula in (104) is a contradiction. (103) is thus ruled out by exactly the same mechanisms that enforced the Subjectivity Requirement. Basically, it is infelicitous because the complement clause is not something for which both a positive and a negative judgement are possible. The type-mismatch account does not extend to such cases. The presence of a sub- jective predicate should be enough to satisfy the subjectivity requirement, regardless of the fact that it appears inside a contradiction17. To be fair, though, it would still be possible to attribute the infelicity of such examples directly to the fact that one is not expected to hold contradictory opinions, so the sentence is judged odd because of this, quite independently of any presupposition properties. Although this is a logical possibility, it would be a strange analysis given the well-known fact that sentences where contradictory beliefs and desires are expressed, for example. We may think less of John if he contradicts himself in this way, or assume that he is confused, but still, the following sentences are not judged infelicitous by most speakers:
(106) John thinks the Bill is dead and that he is alive.
17It is unclear to me how the type-based theory could function at all if two subjective predicates are present. Saebø seems to imply (in his section 3.3) that such a situation would be accounted for in his approach, but he does not give any technical detail. It would not be to his advantage to simply rule completly out multiple subjective predicates below find in his approach, since although it would correctly rule out (103), it would undergenerate in not accounting for the meaning of many acceptable cases such as the following sentence, where both adjectives nice and good are picked up by the opinion verb:
(1) John finds that he bought a nice car for a good price.
178 (107) I wanna go and I don’t wanna go.
The intuitive availability of the contradictory thought reading in Russell ambiguity sentences like (108) also argues against the fact that infelicity results from ascribing a contradictory thought to an individual, thus weakening the hypothesis that Saebø could make that the infelicity of (103) can be attributed to a general prescription against attributing contradictory mental states to individuals.
(108) John thought the yacht was longer than it was.
Similarly, if the complement clause is a tautology, a situation that we can get simply by replacing the conjunction with a disjunction in (103), the result is still infelicitous, and still for the same reason: the complement clause does not express something for which both a positive and a negative judgement are possible. This is illustrated in (109) and (110):
(109) #John finds that Mary is pretty or she isn’t pretty.
(110) (∃w’. [w’∈ Acc(w) & [[ Mary is pretty or she isn’t pretty ]]w’, j =T])&(∃w”.
[w”∈ Acc(w) & [[ Mary is pretty or she isn’t pretty ]]w”, j =F])
What these examples show is that the Subjectivity Requirement is not a mere matter of find needing to get a subjective predicate. The unacceptable examples (103) and (109) both have such predicates. We know already that it would not be possible to make the case that the problem with these examples is that predicates of personal taste appearing in conjoined propositions are incapable of licensing find for some reason, since such a configuration is fine in (105). The contrast between (103) and (109), on the one hand, and (105) on the other argues strongly in favour of viewing the source of the infelicity of (103) and (109) as not coming from its syntactic structure, but rather from the contradictory nature of their presupposition. This is a correct prediction of the presupposition account of the Subjectivity Requirement, and it is an important weakness of Saebø’s type-based approach that it cannot account for these facts. I want to emphasize again that none of this discussion shows that there is no type-mismatch in examples that do not obey the Subjectivity Requirement, and that the pronoun theory of judge-dependency is wrong. However what these examples do show is that the presupposition account is necessary since it covers more data than the type-mismatch account.
179 5.6 Summary
In this section I have suggested an alternative to Sæbø (2009)’s analysis of the Sub- jectivity Requirement which builds on previously unnoticed presupposition properties of opinion verbs, namely the fact that the non-subjective part of the complement clause is assumed to be true by the speaker. This analysis rests on very minimal assumptions, and it is not tied directly to any specific view of the nature of predicates of personal taste. I have proposed that the specific way in which this presupposition comes about is in the form of the Subjective Contingency Presupposition. This makes a stronger case for enforcing the Subjectivity Requirement, as well as accounting for the otherwise puzzling behaviour of negation in the embedded clause. It also has the advantage that it predicts that all tautologies and contradictions below find should be judged infelicitous, a prediction that is not shared by Saebø’s type-based approach. As such, we can claim that the presupposition-based analysis of the Subjectivity Requirement is necessary anyway, since it predicts that every case that is ruled out by the Saebø’s analysis should also be ruled out by mine, but there are additional cases that are only ruled out by the presupposition-based approach.
6 In Favour of Radical Reductionism
There is an important disconnect between the way in which Lasersohn (2005)’s theory of judge-dependency models opinions and the long-standing tradition in intensional semantics of modeling attitude verbs as quantifying over possible worlds restricted by some accessibility relation starting since Hintikka (1969). Beliefs, desires and needs, for example, can all be modeled as sets of accessible worlds, while an opinion seems to be a completely different object. This is most evident in the way in which the matrix subject of find and that of think relate to their embedded proposition. In the present account as well as in Sæbø (2009)’s, the subject of find provides the value of the judge, either index or pronoun, of its complement clause, while for most authors, the subject of think contributes to fix the accessibility relation to the right set of epistemically accessible worlds. This is in contrast with many authors who claim that opinion verbs are ordinary epistemic verbs with the added role of shifting the identity of the judge of the embedded clause (cf. Lasersohn (2005), Nouwen (2007) and Stephenson (2007b)). Another piece of meaning that has been suggested to be
180 part of the semantic contribution of find is evidential: Stephenson as well as Pearson (2012) both claim that find presupposes that the judge has some basis for holding the opinion given in the complement clause. Saebø calls radical reductionism the position according to which find does nothing else than assign the matrix subject as the judge for the embedded clause, and it is the one that I will be defending in this section. I will offer support for this view and show that find has no epistemic component, and imposes no particular evidential constraint that is not already present in the use of predicates of personal taste in the first place.
6.1 No De Dicto Ambiguities under Find
The most obvious place to look for signs that find could involve quantification over epistemically possible worlds is to look at the interpretation of DPs in its complement. The prediction is that if opinion verbs do not make reference to other possible worlds at all, then indefinites in their complement should not have opaque readings. This prediction is borne out. It is well-known that we can accept a sentence like (111) even if John has no particular person in mind. It would be felicitous (and true) to say this in a situation where John has found some pieces of clever machinery somewhere, and concludes that whoever built them is a genius. The identity of that person is allowed to vary across John’s epistemically accessible worlds.
(111) John thinks that someone is a genius.
The following sentence with find, in contrast, is not so permissive.
(112) John found that someone is a genius.
(112) can only be used to express John’s opinion regarding an actual person. It could not be used in the context described above, for example, while it would be felicitous in a context where John actually know who this person is. It is perhaps not impossible to come up with an explanation for the absence of de dicto readings under find that does not make it into such a different object from think, but I do not see any principled way of doing so. The most obvious technical solution to this problem would be to claim that DPs must scope outside of the complement of think for some reason, in a manner similar to how de re readings are often assumed to come about with epistemic verbs. This cannot be right, however, and we can show
181 this by constructing an example in which an indefinite DP contains a predicate of personal taste. The Subjectivity Requirement can only be respected if the predicate of personal taste remains inside the scope of find, but an analysis of the absence of de dicto readings under find that involves obligatory movement would force the DP, along with the predicate of personal taste, out of the complement clause. (113) shows that this is incorrect. The DP someone with a strange name can only be interpreted de re, and yet the predicate of personal taste is clearly still capable of licensing find.
(113) John finds that you married someone with a strange name.
If the movement analysis were correct, this sentence would not be acceptable, since it would either violate the Subjectivity Requirement because the predicate of personal taste, along with the DP, is outside the scope of find, or it would violate whatever forces the DP to move out of the complement clause. Examples like (113) stand as a difficult challenge to anyone who wants to claim that find involves an epistemic component. Under the fully reductionist approach, however, no de dicto reading is ever expected to arise with find.Unlessanysuch reading can be found, I see no reason to believe that the matrix subject’s epistemically available worlds should play any role in the interpretation of find sentences.
6.2 Find isn’t Evidential, PPTs are
This discussion is reminiscent of Stephenson (2007b)’s and Pearson (2012)’s claim that find is distinguished from think in that the former requires that the subject have direct evidence for supposing that the complement clause holds, a feature of interpretation that both authors integrate as a presupposition directly into the verb’s denotation. This is justified for example by the contrast between the two following sentences:
(114) John thinks that the cat food is tasty. (115) John finds the cat food tasty.
(115) triggers the inference that John has tasted the cat food, while (114) does not. As Stephenson remarks, (114) can be understood as claiming that the cat food is tasty for the cat, or for John himself. Even in the second case, this can be based on
182 indirect evidence, say looking at the list of ingredients and concluding that he would probably like it. However, the source of the evidence in (115) should not be treated as part of the lexical entry of find, since it can be shown to follow purely from the fact that John is interpreted as the judge of the embedded clause. Indeed, this inference even holds if the claim that cat food is tasty is made in an unembedded context:
(116) Cat food is tasty.
This sentence allows the hearer to infer that the speaker tasted the cat food just as much as (115) allows us to infer that John tasted it. It seems to be simply the case that we cannot imagine any other way for the one to be a competent judge of something’s tastiness, rather than some special property of find. Indeed, it is unclear to me that this inference is even linguistically-based at all. Since this inference that the judge has direct evidence of the relevant sort is an integral part of any attribution of an opinion to a judge, then the fact that this presupposition is found in find sentences follows naturally from the fact that these sentences always involve attributing an opinion to a judge. In fact, this is all they do. What I conclude from this discussion is that the necessity for direct evidence in find sentences does not constitute an argument in favour of adding an evidential component to the denotation of opinion verbs, since it can be traced back to the way we understand any judge-dependent sentence at all.
6.2.1 The Judge’s Beliefs
As we have discussed in some detail in section 5, sentences with find presuppose the non-subjective part of their complement. There is an additional piece of information that appears to be inferable from a find-sentence, namely the fact that the matrix subject also believes that the non-subjective part of the complement is true. For example, it seems to be true that the following inference holds:
(117) John finds that Mike gave a great class yesterday. → John believes that Mike gave a class yesterday.
Since this inference holds here, then it is tempting to generalize this by stating that the truth of the non-subjective part of the complement clause must not only be accepted by the speaker, but also by the judge. This is too strong, however. Indeed, we
183 can construct scenarios where the judge does not believe that the non-subjective part of the complement clause is true, and still the sentence is judged felicitous. Examine for example the following scenario:
(118) John comes to see Bill and notices a brand new car in front of Bill’s house. This car was actually bought by Bill’s wife, not by Bill. John comments on this by saying to Bill: “You bought a very nice car.”. For whatever reason, Bill doesn’t correct John, who maintains his false belief that Bill bought the car. Bill later reports this exchange to his wife and says: (119) John finds that you bought a very nice car.
The fact that (119) is judged felicitous in this scenario despite the fact that John does not believe that Bill’s wife bought a car shows that the presupposition does not concern John’s beliefs. I believe the inference that we do find in (117) is accidental. That is, for John to have any opinion on a class that Mike gave, it must necessarily be the case that John has seen it, otherwise it would be impossible for him to have any such opinion. In contrast, for John to have an opinion regarding the car in (119), he needs to have seen it, but he does not need to know who bought it. That is, as we have just discussed in the previous section, the inference that is triggered by ascribing a judge to a proposition is simply that the judge has sufficient information to have the opinion expressed by the complement clause. In (117), this entails seeing the class, and thus knowing (and believing) that Mike gave a class, while in (119), this entails seeing the car which does not in turn entail knowing who bought it. Thus the belief restriction in (117) is a mere epiphenomenon. It is merely a consequence of the more general requirement that judges have some basis for their opinion.
6.3 Thinking and Finding
Given the observation made earlier that thoughts and opinions are formalized using unrelated formal objects, we do not a priori expect to see any relevant entailment re- lationship between sentences with find and corresponding sentence where the opinion verb is replaced by think. Yet in some simple cases, we seem to find just that. The two sentences below seem to entail each other, at least if we understand (120) to say that John thinks the chili is tasty for himself.
(120) John thinks that this chili is tasty.
184 (121) John finds this chili tasty.
This is not a generalized phenomenon, however. If we look at examples where the complement clause is large enough to involve a non-trivial presupposition, then we find that neither entailment goes through.
(122) John thinks that Bill bought a nice car. (123) John finds that Bill bought a nice car.
It is not entirely clear that (123) entails (122), since (123) does not require that John hold any belief on whether Bill has bought a car at all, which (122) does require, as we have discussed in the previous section18. The fact that the reverse entailment does not go through either is harder to show. We know already from our discussion of Stephenson (2007b) that predicates of per- sonal taste under find are only optionally interpreted relative to the subject of think as the judge. The most salient interpretation of (122) is one in which John does serve asthejudgefornice. This is not necessary, even if it may require a convoluted context to make any other possible reading available19. Under the other reading, where John is not the judge for nice, it is clear the (122) does not entail (123). In the more salient reading where John is indeed the judge, then the entailment still does not go through because although both sentences assert that the car is nice with respect to John as
18Marcin Morzycki (p.c.) provided the following sentence as a counterexample:
(1) John finds that Bill bought a nice car, but he doesn’t think so.
Under my analysis, it should at least be possible to get a non-contradictory reading from this sentence, but this does not seem to be the case. This is a strong case against the fully reductionist view that I adopt here, and it shows that the claim that find is not a propositional attitude in any sense whatsoever is perhaps overstated here. The fact that subjective predicates come with some evidentiality of their own muddles the issue somewhat, but it may be the case that find also has an evidential component. Such a claim also seems to be a necessary first step to analyse a different use of find from the one discussed in this thesis, namely where it is used to describe a discovery:
(2) Copernicus found that the earth revolves around the sun.
In such examples, it seems quite obvious that find is evidential in at least some way. I hope to examine this use of find further in the future, and a unified analysis of both uses would surely shed some light on the role of evidentiality in the opinion use of this verb that we have focused on. 19For example, imagine a situation where John is very rich, and he thinks that Bill bought a new Honda. This is not a car that John would find nice himself, him being rich, but Bill and myself, the speaker, would find it nice. In this situation, I can say (122) to describe the situation, and it is clear then that John cannot truthfully serve as the judge for nice.
185 the judge, the find sentence also has a presupposition that Bill bought a car that is not shared by (122). Thus, I believe that despite appearances, we are essentially correct in giving to find a semantics that makes it entirely distinct from epistemic verbs like think. Apparent cases of inferences between think and find sentences are merely the result of the fact that find does nothing but shift the judge of its embedded clause, while think may sometimes make an individual salient enough to serve as the judge.
7 Syntactic Properties of the Complement of Find
Saebø frames his paper as an argument in favour of the pronoun theory of judge- dependency and he claims that novel data concerning the complement clause of find argue in favour of this. I believe that a more thorough examination of the facts casts doubts on this conclusion, and in fact rather argues in favour of an index theory. We will examine in turn the major arguments that he makes and provide some novel data and analyses concerning each of them. In each case, we will see that despite Saebø’s claims, the index theory is capable of providing an explanation, and better yet, it can explain a broader set of facts than the competing pronoun approach. We will first examine how Saebø deals with full clauses under find, rather than only small clauses, and show that his system is only compatible with some theories of the syntax-semantics interface, such as DRT, which is the one that he uses. We then discuss the fact that the subjective predicate that must be present under find to license it does not have to be in a local relation with it. It can, in fact, be quite deeply embedded. Next, we discuss cases of coordination of subjective and non-subjective complements under find, which will turn out to be the most crucial point in favour of the index theory. Finally, we discuss cases where subjective predicates seem to fail to license find and attempt to provide an explanation for such cases. Much of this discussion is based on observations from Saebø, but some new data will be presented and the conclusions that I will reach will be quite different from his.
7.1 Presuppositions and the Necessity for DRT
We begin with the following contrast that Saebø examines since it introduces what I believe to be a crucial piece of his analysis. In the two Norwegian sentences below,
186 we see that synes ‘find’ can be licensed by a subjective attributive adjective on the complement of gift ‘married’, but not on the complement of kjenner ‘know’.
(124) Jeg synes du er gift med en vakker mann. I seem you are married with a beautiful man ‘I find the man you are married to beautiful.’
(125) ?? Jeg synes du kjenner en vakker mann. I seem you know a beautiful man Saebø claims that (124) is felicitous because “everything in the complement clause but the taste predicate vakker ‘beautiful’ can be interpreted as presupposed” (p. 340). He makes use of this fact by using a DRT-type notation in which semantic material that is presupposed is separated from the assertion. This leaves nothing but the unsaturated subjective predicate in the assertion, thus permitting the opinion verb to access it. In this system, the meaning of the complement clause is represented as follows. Note that we have to assume that the variable y above the line can be bound by the existential quantifier below it.
(126) λx. beautiful(x)(y) ∃y. man(y) & married-to(y)(z)
What this formal mechanism does is allow the opinion verb to reach the subjective predicate despite not being in a close syntactic relationship to it. As we will discuss below, this DRT mechanism is actually crucial to the analysis, which otherwise cannot get off the ground and fails to give an interpretation to sentences with find even in the most simple cases. In contrast, the main verb of the complement clause in (125) is kjenner ‘know’, which in Saebø’s words, is not ‘functional’ and thus does not lend itself well to a read- ing in which it is presupposed. Because of this, the semantic information contained in the complement clause must be part of the assertion. For the complement clause to be able to get an interpretation at all, it must be that the judge-argument of the predicate of personal taste is filled (otherwise, being of type
187 to give the opinion verb access to the predicate of personal taste. Whether or not some part of the complement clause is presupposed or not makes no difference to this approach, since it simply involves changing the judge index of the entire complement. In principle, the Subjectivity Requirement should be satisfied by (124) just as much as by (125), since all it requires is that it not have the same value at all judges. The compositionality problem that (125) raises for the pronoun-based approach does not arise in the index theory, which leaves it without any way of ruling out this sentence and accounting for the observed contrast between verbs like to marry and to know. However, I do not believe that this contrast has anything to do with the way that judges are represented in the grammar. We have observed as an independent fact that all non-subjective material in the complement clause of find is always presupposed. We have formalized this in terms of the Subjective Contingency Presupposition, which I have assumed is hardwired in the meaning of find 20. If Saebø is right that a verb like to know in the sense of “to be familiar with” resists being part of the presupposition content of a sentence, then it is expected that it would be infelicitous under find, since this environment would precisely have the effect of making it presupposed. It seems to me unnecessary to appeal to the DRT mechanism presented above, since the mere observation that find forces the non-subjective material in its complement to be presupposed is enough to rule out material that ‘refuses’ to be presupposed from appearing in this position. As such, I do not consider the contrast between (124) and (125) to argue for any specific way of representing subjectivity in the grammar, but rather as confirmation of the fact that non-subjective complements under find must obligatorily be presupposed.
7.2 The Syntactic Non-Locality of Find and Semantic The- ories
This discussion on how find can or cannot access a subjective predicate inside its com- plement clause brings up a general issue with the interpretation that Saebø provides
20This is perhaps an oversimplification. As observed in section 2, there are other elements in the grammar that require their complement clause to be subjective, such as the verb consider and the PP in my books. If it can be shown that these and similar predicates can all successfully be analysed as introducing a Subjective Contingency Presupposition, then it would be more economical to state it as a general feature of the grammar that one can only overtly use an element that shifts the judge-index of some constituent if the value of that constituent can be true for some judge and false for another.
188 for find, which is that this approach makes it somewhat unclear how to consistently match up the syntactic representations with the intended meaning. Let us repeat Saebø’s denotation for find:
(127) [[ find ]] = λφ
What find takes as a complement must be a constituent that denotes the set of individuals corresponding to the judge argument of the subjective predicate for the appropriate internal argument. In the most simple cases, where find is followed by a small clause consisting of a proper name and a subjective predicate like the example (57) we examined in some detail in section 4, the syntax provides the right configuration to yield this meaning without any alteration. Repeating the example here, it is clear that if beautiful is taken to be a relation between some individuals and those who find them beautiful, then the small clause will simply denote the set of individuals who find Mary beautiful, as desired.
(128) Anne finds Mary beautiful.
Putting aside for a moment the DRT mechanism mentioned in the previous section, things get quite a bit more complicated once full clause complements are introduced in the picture. In (129), for example, it is hard to see exactly how find is supposed to relate to the predicate of personal taste nice.
(129) John finds that Mary is a nice girl.
Clearly, this kind of example can be made to work out inside a DRT approach, as demonstrated for a similar example using the nominator/denominator notation in (126). What I want to show here is that it is in fact only possible to apply Saebø’s analysis of find inside a framework where presupposed material can be pushed out of the way to allow semantic composition between two non-local constituents, as there must be between find and nice in (129). In simple cases like (129), there are other options that one could appeal to to make the semantic composition proceed smoothly inside a pronoun theory of judge- dependency. The most obvious way that I see of doing this is to appeal to a mechanism whereby a null PRO is generated in the position of the judge-argument of nice and raised to the top of the clause. While this PRO is completely vacuous, its trace denotes a variable that will be bound by the lambda-abstractor introduced just below
189 the landing position of PRO. This mechanism is the same as that proposed in (Heim and Kratzer 1998: pp. 225-230) to create a type-
(130) PRO λx [Mary is a [ nice tx ]girl]
Indeed, as far as I can tell, inside the semantic model proposed in Heim and Kratzer (1998), this is the only way to get such a clause to get the denotation that is needed here. This predicts that the relation between the verb find and the predicate of personal taste should be restricted in the same way that syntactic movement is. Of course, movement is not entirely free, and we know very well that some syntactic structures prevent movement from escaping them, which should in turn entail that the relation between the opinion verb and its licensing predicate of personal taste should be blocked by islands. This prediction is clearly not borne out, however. We can see that relative clause islands are no obstacle to the relation between find and its licensing predicate of personal taste. This can be seen in (131) and (132), where the subjective predicates weird and cher ‘expensive’ are inside a relative clause attached to the direct object of the complement clause. Despite this, these sentences are perfectly fine. It is also perfectly clear from their meaning that the opinion verb in the matrix relates to the predicate of personal taste inside the relative clause
(131) John finds that Anna married a guy who has weird tastes in music.
(132) Je trouve que Jean veut s’acheter une voiture qui coˆute beaucoup trop I find that Jean wants to-buy a car that costs a-lot too cher. expensive. ‘I find that John wants to buy a car that’s much too expensive.’
Because of this, we cannot adopt Saebø’s analysis inside the Heim and Kratzer (1998) framework, since then there would be no way to relate the opinion verb and its predicate of personal taste in some cases. What this shows is that Saebø’s analysis of find and how it relates to its licensing predicate of personal taste is only available if we adopt a semantic framework in
190 which structurally non-local elements can semantically combine directly despite the impossibility of relating them by movement. This is not the place to compare semantic theories more generally, but it should be made clear that this analysis requires that find have a very close semantic relation with its predicate of personal taste, while the facts just outlined show that they do not need to have a corresponding close syntactic relation. For the researcher who believes that all semantic composition must be syntactically local, even if only at LF and allowing for covert movement to alter the structure which is the input to semantic interpretation, the non-local character of the licensing relationship between the predicate of personal taste and find argues straightforwardly against Saebø’s proposal. If judges are rather represented as indices rather than pronouns, then find can be assumed to simply shift the index of its complement clause and no special syntactic relation is expected to have to hold between this verb and its subjective predicate, except that the subjective predicate must be inside the complement clause to satisfy the Subjectivity Requirement. Regardless how deeply a subjective predicate is buried in the clause, changing the index of the clause itself is sufficient to ensure that the judge for that predicate will also be changed, which allows us to explain easily why a subjective predicate could appear inside an island like a relative clause and still be picked up by find. Since no special syntactic relation is assumed to hold between the two elements, this analysis is not restricted to any specific theory of the syntax- semantics interface. The meaning of find in an index theory of judge-dependency imposes no restriction whatsoever on where a subjective predicate can appear, but Saebø shows that there actually are some such restrictions. We will see that the index theory still fares better in allowing the compositional semantics to provide even the infelicitous sentences with an interpretation, and letting additional pragmatic principles take care of ruling out the bad cases. We now turn to those cases where find cannot find its subjective predicate.
8 Coordination under Find
Saebø claims that whenever a predicate of personal taste below find is coordinated with another predicate whose value does not depend on the judge, the resulting sen- tence is infelicitous. An example of this is provided in (133):
191 (133) #She finds him handsome and under 45.
Saebø claims that this is expected if we adopt the pronoun theory of judge depen- dency, but it is surprising if we adopt the index theory, and he consequently considers such examples to support the pronoun theory. However, as we will see in this sec- tion, some cases of coordination of subjective and non-subjective predicates under find are actually felicitous, which the pronoun theory predicts should never be the case. A pragmatic explanation for the unacceptable cases will be proposed. Since Saebø makes a strong case that the pronoun theory does indeed predict the unavailability of such coordinations under find in all cases, the fact that we can construct felicitous examples with this property actually turns out to be a strong argument against this approach. Saebø attributes the unacceptability of sentences like (133) to the fact that the predicates handsome and under 45 are not of the same type, making it impossible to coordinate them. That is, handsome is a predicate of personal taste and if the pronoun theory of judge-dependency is right, then it denotes a relation between individuals (type
(134) ??
handsome under 45
The only way in which the coordination could be successful is if the predicate of personal taste were interpreted with a covert pronoun filling the judge position, but this would make the entire conjunct no longer judge-dependent, and it would then no longer be an adequate licenser for find. As discussed in section 7.2, we could imagine a slightly more complex derivation whereby a null PRO is generated in the judge-argument position of handsome and raised to the top of the embedded clause, thus making the two conjuncts symmetrical in type, while still making the clause denote a set of judges. The following tree illus- trates how the types would line up in this hypothetical derivation, and the denotation of this constituent is given in (136).
192 (135)
(PRO)
λ
him
handsome tPRO (136) λx. under-45(h) & handsome(x, h)
Even this solution is not predicted to be available, however, as it would involve moving the null PRO out of a conjunction, in violation on the Coordinate Structure Constraint. The grammar is thus caught in a situation where the only potential derivations lead to some problem, and the fact that the sentence is judged infelicitous is accounted for in the pronoun theory of judge-dependency. The relativist view predicts no such interference of conjoining a non-subjective predicate with a subjective one, in contrast. Under this view, both conjuncts in (133) are of the same type (
193 embedded contexts, which is correct:
(137) He is handsome and under 45.
The view that Saebø advocates predicts that such a conjunction should be bad under find, since it is either a type-mismatch at the level of coordination, or at the level where find combines with its complement clause, but it is not obvious how acceptable examples like (137) where the conjunction is not under find should be derived. Recall from Section 4 that judge-arguments must be the second argument of PPTs for simple cases like (57) above to work out. Faced with this situation, Saebø concludes that it must be possible for predicates of personal taste to have their judge 21 argument “pre-saturated”, coming from the lexicon with a designated variable xT in the position of the judge-argument, or unsaturated, with its judge argument open. That is, handsome must be ambiguous between an
(138) Everyone has a hobby which is fun. (Lasersohn (2005)’s (41))
I will not go further in trying to fix Saebø’s account here. While it is clear that his theory rules out all cases of conjunction of subjective and non-subjective predi- cates under find, it also forces us to introduce some complications to allow for such conjunctions in unembedded contexts. Rather, we turn to some additional data that shows that Saebø’s generalization is actually wrong: it is not always infelicitous to conjoin a subjective and non-subjective predicate under find. An example of this is provided in (139):
21This is the notation that Stojanovic (2007) uses for the variable that provides the value for judge.
194 (139) John finds that nobody at the party was handsome and under 45.
This is a felicitous sentence, despite the fact that the exact same two predicates whose conjunction is impossible in (133) are conjoined under find here also. This is clearly not predicted to be possible for Saebø. The asymmetry in type between handsome and under 45 that was claimed to make (133) infelicitous should block (139) as well. Of course, since the index theory has no reason to disallow the type of conjunctions we are concerned with, (133) is correctly predicted to be acceptable under this theory. In the remainder of this section, I provide a more precise characterization of when conjunction of subjective and non-subjective predicates is possible and when it is impossible under find. We will find that all the infelicitous cases can be claimed to pragmatically odd in a manner to be defined below, suggesting that the correct analy- sis of this phenomenon must be that there is nothing wrong in the syntactic derivation or in the semantic composition in combining subjective and non-subjective predicates. The grammar should allow this in general, thus accounting for the acceptability of (139), while independent pragmatic principles will be responsible for blocking cases like (133). So what is the relevant difference between (133) and (139) that makes the latter better? I believe it is simply the fact that adding the conjunct and under 45 in (139) actually contributes to the truth-conditions of the sentence, not only to its presupposition. To see this, let us compare this sentence with a corresponding sentence without the non-subjective conjunct:
(140) John finds that nobody at the party was handsome.
(140) has stronger truth-conditions than (139), where the conjunct is present. It should be clear that when the PP is added, John only needs to have a negative opinion of the looks of a subset of the people at the party for the sentence to be true, namely those people who are under 45. For (140) to be true, John must have that opinion of everyone at the party, regardless their age. In fact, intuitions tell us that (140) entails (139). Adding the conjunct to (140) weakens its truth-conditions, but it also alters its associated presupposition. Let us see what the Subjective Contingency Presupposition predicts for the variant without the conjunct:
195 (141) (∃w’[w’ ∈ Acc(w) & [[ nobody at the party was handsome ]]w’, j = T]) &
(∃w”[w” ∈ Acc(w) & [[ nobody at the party was handsome ]]w”, j =F])
The first half of this formula tells us nothing interesting, since it is necessarily true. This is because the accessibility relation will always contain worlds where the judge finds nobody handsome by virtue of how it is defined. The second half says that it is a possible judgement to say that nobody at the party is handsome is false. For that to be false, it must be that someone at the party was handsome, hence that there was someone at the party. This is all that this sentence presupposes. Compare this to the predicted presupposition of the variant with the added PP conjunct:
(142) (∃w’[w’ ∈ Acc(w) & [[ nobody at the party was handsome and under 45 ]]w’, j =T])&(∃w”[w” ∈ Acc(w) & [[ nobody at the party was handsome and
under 45 ]]w”, j =F])
Again, the first half of this formula tells us nothing. This time, however, the second half tells us it is a possible judgement that it is false that nobody at the party was both handsome and under 45. If that is false, then it must be that someone at the party was handsome and under 45, hence that someone at the party was under 45. This is a slightly stronger presupposition from the variant without the conjunct, which only required that there be someone at the party, while this sentence requires that there be someone at the party who is under 45. To summarize, (139) and (140) differ both in their presupposition and in their truth-conditions, and I will argue that this is a necessary condition for the non- subjective conjunct to be allowed. We will see that this pattern is not what we find with similar sentences where nobody is replaced with every, and that the resulting sentence is correspondingly judged infelicitous. Let us construct just such an example by replacing nobody in (139) with every:
(143) #John finds that everyone at the party was handsome and under 45.
This sentence is infelicitous. Once again, we will compare its truth-conditions and presupposition with that of a variant without the offending conjunct. Unsurprisingly, this variant is a felicitous sentence:
(144) John finds that everyone at the party was handsome.
196 It will be easier to start with the presuppositions. The felicitous sentence (144) is predicted to have the following presupposition.
(145) (∃w’[w’ ∈ Acc(w) & [[ everyone at the party was handsome ]]w’, j = T]) &
(∃w”[w” ∈ Acc(w) & [[ everyone at the party was handsome ]]w”, j =F])
This says very little. At best, we can conclude from this formula that there was someone at the party (perhaps that there were more than one person), no more. This stands in stark contrast to the presupposition associated with (143):
(146) (∃w’[w’ ∈ Acc(w) & [[ everyone at the party was handsome and under 45
]]w’, j =T])&(∃w”[w” ∈ Acc(w) & [[ everyone at the party was handsome
and under 45 ]]w”, j =F])
The first half of this formula says that it must be possible to judge as true that everyone at the party was handsome and under 45, from which we can conclude that everyone at the party was under 45. This is a much stronger presupposition than that associated with the variant without the PP conjunct. What about the truth-conditions of these two sentences? The felicitous sentence (144) is quite simple: it asserts that everyone at the party is judged handsome by John. The variant with the added conjunct has stronger truth-conditions: according to John’s judgement, everyone at the party is handsome and under 45. Of course, the fact that everyone at the party was under 45 is unaffected by attributing John as the judge, but the point is that this fact is also part of the assertion. How do the variants with and without the PP conjunct differ? The only difference is that the variant with the PP both asserts and presupposes that everyone at the party was under 45, while the variant without the PP neither asserts nor presupposes that. This means that in all worlds where the presupposition of the sentence with the conjunct is satisfied, the two sentences are interchangeable. So there is a sense in which the assertion is strengthened by adding the PP, but this is not strengthening in the usual sense. The fact that everyone at the party was 45 is also presupposed by (143), which means that this sentence is only usable in situations where this piece of information is true. To put it a bit more formally, although adding the PP conjunct to (144) narrows the set of possible worlds in which this sentence would be true, it does not broaden the set of possible world in which it would be false. It rather broadens the truth-value gap, making any possible world where it is not the case that everyone
197 at the party was under 45 neither true nor false. I argue that it is this that makes (143) infelicitous: it differs from an alternative sentence without the conjunct only in that its truth-value gap is larger. We can easily show that this is not the case with our earlier acceptable case with nobody. While adding the conjunct under 45 to (144) adds the presupposition that someone at the party was under 45, it also genuinely alters the assertion in the normal sense, in this case by weakening it. While (144) can only be true if John did not find anyone at all at the party handsome, adding the PP makes it possible to judge the sentence true if he did find someone handsome, as long as that person was over 45. This shows that adding the conjunct to (139) does more that extend the truth-value gap: it also makes the sentence true in more possible worlds than its variant without the conjunct, and false in fewer. What we have seen so far is that adding a non-subjective conjunct below find in a sentence where it will only extend the truth-value gap of the sentence is not acceptable, but adding a conjunct that makes it true in more cases, i.e. that extends the set of possible worlds that it denotes, is acceptable. We can show that adding a non- subjective PP with or rather than and is subject to the same restriction. Whenever the result is infelicitous, we can show that the problematic sentence differs from an alternative sentence without the disjunct only in having a larger truth-value gap. In order to see this, we will look at examples like our previous cases with every and nobody, but switching and for or. Let us start with nobody and or,whichwecompare to the alternative without the disjunct:
(147) #John finds that nobody at the party was handsome or under 45. (148) John finds that nobody at the party was handsome.
How do these sentences differ in terms of their predicted truth-conditions and pre- suppositions? Once more, we can apply the Subjective Contingency Presupposition to (147):
(149) (∃w’[w’ ∈ Acc(w) & [[ nobody at the party was handsome or under 45 ]]w’, j =T])&(∃w”[w” ∈ Acc(w) & [[ nobody at the party was handsome or under
45 ]]w”, j =F])
The first half of this formula tells us that on at least one judgement, nobody at the party was either handsome or under 45, from which we can conclude that nobody
198 at the party was handsome. This is a very strong presupposition. Just like in the case with every and and (143), (147) and (148) differ only in the size of the truth-value gap. While (148) is true as long as no one at the party is considered handsome by John, (147) is would only be true if no one is considered handsome by John, and if no one is under 45. Thus there are fewer possible worlds where it is true, but not more worlds where it is false, since all those worlds where at least one person at the party was under 45 will contradict the presupposition expressed in (149), and those worlds will thus not have a truth-value at all. The final logical possibility is to examine every with or. Again, we compare a sentence with the disjunct to an alternative without it, and we give the predicted presupposition of the variant with the disjunct.
(150) John finds that everyone at the party was handsome or under 45. (151) John finds that everyone at the party was handsome.
(152) (∃w’[w’ ∈ Acc(w) & [[ everyone at the party was handsome or under 45 ]]w’, j =T])&(∃w”[w” ∈ Acc(w) & [[ everyone at the party was handsome or
under 45 ]]w”, j =F])
This is a weak presupposition. The first half tells us nothing about who is under 45, since it would be made true by a judgement which makes everyone at the party handsome, regardless their age. The second half tells us that at least one person at the party was neither handsome nor under 45, from which we can conclude that at least one person was over 45. In this case, then, adding the conjunct genuinely affects the truth-conditions, by weakening them in this case, making the resulting sentence true if there are people that John did not judge handsome at the party, as long as they were under 45. Finally, we return to Saebø’s original example (133), repeated here as (153). It is easy to see that it patterns with our example with every and and (143). Once more, let us contrast it with an alternative sentence without the offending conjunct:
(153) #She finds him handsome and under 45. (154) She finds him handsome.
(154) just tells us that the individual picked up by the pronoun him falls into the extension of handsome if the judge is switched to the individual referred to by she.
199 The Subjective Contingency Presupposition does not produce anything interesting here, as this presupposition is met simply by virtue of the fact that handsome is a predicate of personal taste.
(155) (∃w’[w’ ∈ Acc(w) & [[ him handsome ]]w’, j =T])&(∃w”[w” ∈ Acc(w) & [[
him handsome ]]w”, j =F])
How does this differ from the predicted truth-conditions of the infelicitous (153)? Its presupposition says the following:
(156) (∃w’[w’ ∈ Acc(w) & [[ him handsome and under 45 ]]w’, j =T])&(∃w”[w”
∈ Acc(w) & [[ him handsome and under 45 ]]w”, j =F])
Since the complement clause must be true for some possible judgement, and since under 45 does not vary across judgements, then it must be the case that the individual picked up by him is under 45. Even if this information is also technically part of the assertion, the fact that it is presupposed means that (153) differs from (154) only in having a larger truth-value gap. So we can add a non-subjective conjunct (or disjunct) to a subjective constituent below find only if doing so alters the truth-conditions of the sentence in more than just broadening its truth-value gap. That is, if adding a conjunct makes the sentence false for fewer possible worlds, it should also make it true in more. If this correlation is correct and actually covers all the possible and impossible cases of conjunction and disjunction below find, as I believe it does, then this suggests that a pragmatic explanation for this phenomenon is in order, rather than a compositional semantics one. What does it mean for something to be presupposed by a speaker? In the words of Stalnaker (1978):
“The presuppositions of a speaker are the propositions whose truth he takes for granted as part of the background of the conversation. A propo- sition is presupposed if the speaker is disposed to act as if he assumes or believes that the proposition is true, and as if he assumes or believes that his audience assumes or believes that it is true as well.”
A presupposition is a piece of information that the speaker takes for granted and normally assumes that the hearer also takes for granted. One solution to the problem at hand would be to appeal to a general pragmatics principle that states:
200 (157) Avoid optional syntactic constituents whose sole contribution is to strengthen the presupposition of the sentence.
We can see this principle as an ordinary economy restriction: If a presupposition is something that is taken for granted by both participants in a conversation, then adding a stand-alone constituent like a conjunction whose only purpose is to add to the presupposed content should be a waste of time, since it will just repeat something that must have already been in the common ground to begin with. This is reminiscent of another pragmatic principle that has recently been proposed, Minimize Restrictors! (Schlenker (2004))22, which is meant to exclude DPs like the small president if it is known already that there is just one president. The idea is that the restrictor of a definite article should not contain more restrictions than what is required to ensure that it refers successfully, unless these restrictions serve other pragmatic purposes. The specific formulation of this principle is tailored to only apply to definite DPs since the actual purpose of this paper is to derive Binding Condition C from independent principles, but it is in spirit very similar to what I propose. I leave it as an open question here exactly what relation we should understand exists between Schlenker’s principle and mine, and between these and Levinson (1987)’s Maxim of Minimization, which states: “Produce the minimal linguistic clues sufficient to achieve your communicational ends”. On this view, what is wrong with the infelicitous examples in this section is that the complement of find is a syntactic environment where everything that is non-subjective is presupposed, so adding this conjunct does not serve to strengthen or weaken the truth-conditions of the sentence, but only to strengthen the presupposition of the sentence. On the face of it, the pragmatic principle that I suggest appears to go exactly contrary to Heim (1991)’s principle of Maximize Presupposition. This is only apparent, however. Heim’s principle is clearly not meant to enforce that a speaker should, for example, include in a definite description every predicate that is known to hold of the individual it refers to. Rather, it is meant to force the choice between two determiners to go in favour of the one whose presuppositions are strongest. That is, her principle imposes a choice on which item to use when one has to be used anyway. As far as I know, it is never used to force the insertion of any extra material, and it would make incorrect predictions if it did. This is why I specify in my definition of the
22I thank Sacha Simonenko for pointing out this similarity to me.
201 pragmatic principle that only optional syntactic constituents are affected. Maximize Presupposition says that where a determiner is necessary, then one should insert the one with the strongest presupposition. My principle says that one should not add constituents that are not syntactically necessary if all they do is strengthen the presupposition. Thus there is no contradiction between the two principles. What we have done in this section is give a new explanation for why sentences like (133) are judged infelicitous: it is because they minimally differ from the corre- sponding sentence without the non-subjective conjunct only in the strength of their presupposition. Together with a pragmatic principle that encourages speakers to avoid restating as a separate constituent what is already part of the common ground, we have a reason for excluding these sentences. This is a purely pragmatic explanation for the phenomenon and it is, at least in principle, compatible with any semantic theory of judge-dependency, but only in as much as that theory allows conjunction of subjective and non-subjective predicates under find. Since it only predicts that some such conjunctions should be bad under find, it is a weaker restriction than Saebø’s, so if his pronoun theory of judge-dependency were correct, then we should expect to see all such conjunctions ruled out. Since this is not what we find, as demonstrated by the fact that (140) is actually felicitous, then we must conclude that the pronoun theory makes incorrect predictions concerning the availability of conjunction under find and should thus be dispreferred on the basis of such data, which is exactly the reverse conclusion that Saebø reaches. This is the first genuine argument against the pronoun theory of judge-dependency that we give. Note that the index-theory by itself has absolutely nothing to say about examples of this sort. It simply provides an interpretation for every sentence that we examined in this section, both felicitous and infelicitous. This is a good thing if we want pragmatics to take care of the problematic examples. From a purely compositional standpoint, this approach simply allows conjunction of subjective and non-subjective constituents across the board, both under find and elsewhere indiffer- ently. If the pragmatic principle advocated above is indeed in effect, then this is the situation that we want. The compositional mechanism will provide an interpretation for all sentences that involve such conjunctions, but pragmatics will filter some of them out, ruling in correct examples like (139), while ruling out the bad examples (133) and (143).
202 9 Accounting For Saebø’s Syntactic Observations
9.1 Natural Judge-Dependency under find
In cases where more than one predicate of personal taste appears under find,itis not always the case that the matrix subject serves as the judge for all of them. For example, while Mary does serve as the judge for both nice and good in (158), the referent of the DP the mother snipe in (159) clearly only serves as the judge for the adjective beautiful. The other predicate of personal taste, the adjective ugly in the superlative DP, unambiguously makes reference to the speaker’s judgement.
(158) Mary finds that she bought a nice car for a good price. (159) The mother snipe finds the ugliest baby birds beautiful.23
Saebø’s proposal can account for this again using the fact that predicates of per- sonal taste can enter the derivation with their judge-argument either unsaturated or filled in with some variable which will end up being interpreted as the speaker. This approach would allow us to claim that both nice and good are unsaturated in (158) and hence get picked up by find, while in (159), the adjective ugly enters the deriva- tion with its judge-argument filled in and beautiful with its judge-argument open. The subjectivity requirement is satisfied by virtue of the fact that at least one predicate is judge-dependent. According to the index theory, we have simply stated that find alters the judge index of the embedded clause and sets it to the referent of the matrix subject. For concreteness, we can simply use the denotation in (53), repeated here:
(160) [[ find ]]w, t, j = λφ. λx. φw, t, x
Given this meaning, find does not relate directly to any subjective predicate in the complement clause, but rather changes the index for the entire clause in one fell swoop. As such, it is not expected to be able to leave any element in the complement clause dependent on the contextually provided judge and predicts that every piece of subjective information in it should be interpreted relative to the matrix subject as
23This example is modified from Saebø’s example (23). His example uses the verb think,whichis distinct from find in a number of ways, most crucially in not involving the Subjectivity Requirement that find exhibits, as discussed in section 2. Since this chapter, just as Saebø’s paper, is only concerned with those predicates who impose a the Subjectivity Requirement on their complement, then I do not see the relevance of think here.
203 judge. According to Saebø, this makes it unable to account for the interpretation of ugly in (159) without special stipulation. This could involve moving the DP outside of the scope of the opinion verb, for example. This is an unattractive solution, however, since we have no reason to believe that a DP like the ugliest baby bird would be able to move out, while one like a anicecarin (158) would not. Standard assumptions about the scope possibilities would rather lead us to predict that if anything could scope out of find, it would rather be the indefinite. Note in passing that we cannot simply claim that it is the superlative morphology that interferes here, say by making all adjectives with superlative morphology relative to the normal judge for some reason, since examples like (161) show that a predicate of personal taste in the superlative can very well license the use of find,andbe interpreted relative to the matrix subject as the judge. (161) John finds his daughter the prettiest. Again, the situation is not quite as simple as it is presented. The cotextualist view as it is described here predicts that if there are more than one predicate of personal taste below find these should be able to be freely interpreted as dependent on the matrix subject or the contextually-specified judge, just as long as one of them depends on the matrix subject. This is incorrect. In (158), it is not possible for either adjective to be interpreted relative to anyone else but Mary. For example, it is not possible to understand this sentence as claiming that Mary finds that she bought a car for a good price, and the car is nice according to me, the speaker. This is the meaning that would result from inserting the lexical item nice with its judge-argument filled in with a variable whose value is set to the speaker, as is the case with the adjective ugly in (159). The reverse situation where nice is interpreted relative to Mary and good relative to the speaker is equally impossible. Why should this be so? Nothing in Saebø’s analysis predicts that dependence on the contextually provided judge should ever be blocked, as long as the Subjectivity Requirement is satisfied by some other subjective predicate. While the relativist view is too restrictive in never allowing subjective predicates to escape being caught in the judgement ascription, the cotextualist view is too per- missive in always allowing it as long as some other subjective predicate fulfils the Subjectivity Requirement. Note that even in the problematic example that Saebø provides, the situation is not that the DP the ugliest baby bird optionally gets inter- preted relative to the speaker as the judge, but that it obligatorily gets interpreted
204 this way. In fact, I have not been able to construct an example where a predicate of personal taste below find has the option of being interpreted either relative to the subject or relative to the speaker. I believe that what (159) shows is that something about the DP or the rest of the embedded clause structure insulates the subjective predicate from being relativized to the matrix subject. This can be made more obvi- ous if we change the sentence so as to take out the other predicate of personal taste beautiful. In (162), ugly does not get interpreted relative to the subject. Rather, the sentence is simply judged infelicitous:
(162) #John finds the ugliest baby bird dead.
We cannot conclude in favour of either theory until we establish more firmly exactly when a predicate of personal taste must be interpreted relative to the subject and when it must be interpreted relative to the speaker (or another contextually provided judge). In the next section, we examine some more data that Saebø discusses which shows other environments where a predicate of personal taste is unable to license the use of find.
9.2 Clause Structure
Saebø provides the following Norwegian example whose ungrammaticality he at- tributes to the fact that the predicate of personal taste is embedded, in this case in a relative clause on the subject of the complement clause:
(163) #Hun synes alle som er sympatiske, er ikkerøykere. she seems all that are pleasant are non-smokers ‘#She finds everyone who is pleasant non-smoker.’
Note that this is not unique to Norwegian, but also carries over to French and English:
(164) #I find that everyone who is nice is a non-smoker.
(165) #Je trouve que tous les gens qui sont gentils sont non-fumeurs I find that all the people who are nice are non-smoker ‘#I finds everyone who is nice non-smoker.’
Saebø’s explanation for the infelicitous nature of these examples is that the pred- icate of personal taste is embedded inside the restrictor of a quantified DP. For this
205 DP to receive an interpretation at all, the predicate of personal taste must have its judge-argument filled in, since otherwise the relative clause would be interpreted as a type
(166) John finds that he talked to every pretty girl at the party. (167) I find that some weird people came to your party.
(168) Jean trouve qu’il a parl´e `a toutes les belles filles `a ta fˆete. John finds that-he has talked to all the pretty girls at your party ‘John finds that he talked to every pretty girl at your party.’
(169) Je trouve que des gens bizarre sont venus `a ta fˆete. I find that some people weird have come to your party ‘I find that some weird people came to your party.’
From the acceptable nature of these examples, we must conclude that the problem cannot simply be that semantic composition inside the restrictor of the quantifier is impossible in examples (163-165). Putting together the facts from the previous section and those we have just seen, it would appear the the only cases where a predicate of personal taste fails to be
24I do not have access to Norwegian data to contrast with Saebø’s, but there is in principle no reason why the facts should be different in this language. Saebø makes it quite clear that his analysis is not meant to apply to only Norwegian.
206 picked up by find is when it appears inside the subject, and then only with strong determiners. Let us repeat the bad English examples together here for convenience, together with another bad example that I add, in which the predicate of personal taste in inside a subject definite DP which isn’t a superlative:
(170) #John finds the ugliest baby bird dead. (171) #John finds that everyone who is nice is a non-smoker. (172) #John finds that the pretty girl has left.
The Subjective Contingency Presupposition will not help us here, since there is nothing wrong with the presuppositions that these sentences yield. Taking the definite example (172), we can easily show that its presupposition is consistent:
(173) (∃w’[w’∈Acc(w) & [[ the pretty girl left ]]w’, j =T])&(∃w”[w”∈Acc(w) &
[[ the pretty girl left ]]w”, j =F])
This formula is neither a tautology nor a contradiction. It just says that there is a possible judgement that makes the pretty girl left true and one that makes it false. Since the predicate pretty is subjective, the subject DP is allowed to denote different individuals at different judges, thus potentially yielding different truth-conditions for different judges. The fact that it can be true for some possible judge entails that at least one girl left. A possible hypothesis to explain what is going on here is that the problem with these infelicitous sentences concerns information structure, namely that some DPs are obligatorily interpreted as given, which prevents them from having a different value at different indices. From a purely descriptive point of view, in all the problematic sentences, the DP containing the predicate of personal taste is in subject position, and is headed by a strong determiner. Subjecthood by itself cannot not the problem, since we have shown above that a predicate of personal taste in an indefinite subject is fine (cf. (167)). Strong determiners alone are not the issue either, since (166) shows that subjective predicates in such DPs are fine with find. So it is the conjunction of those two properties that makes sentences (170-172) infelicitous. The relation between givenness and structure is a complex affair. In general, there is some consensus that given information tends to linearly precede new information (cf. Skopeteas and Fanselow (2007), who attribute the original claim to Clark and Clark (1977)). This is most visible in languages that allow scrambling, since then the
207 word-order can be seen to be directly altered by givenness, with given constituents being scrambled to the front of new elements in Czech (Kucerov´a (2008)), as well as in German and Georgian (cf. Skopeteas and Fanselow (2007)), for example. Languages with more rigid word orders like English and French will tend to use passivization to move to the front of the clause a given constituent which would not have emerged as subject in the active voice. In rigidly ordered SVO languages, then, subjecthood has a tendency to correlate with a given interpretation. So we expect that definite DPs like the tall girl in subject position as in (172) would strongly tend to receive a given interpretation. The problem is that a DP cannot both be given and subjective. That is, for a definite DP to be subjective, it must be able to refer to different individuals for different judges. This is the only way that the entire clause could end up being true for a possible judgement and false for another, as expected by the Subjective Contingency Presupposition. But this is precisely what cannot be the case if the referent of the DP is given, since then its identity is completely fixed. In the end, we can thus rule out sentences where the only predicate of personal taste in the complement of find is in a definite subject, since then the referent of this DP will be fixed, and the Subjectivity Requirement will not be respected. Outside of the subject position, even definites have a weaker tendency to be interpreted as given, making it possible for the DP the right girl to get a different interpretation at different possible judges
(174) John finds that he married the right girl.
The cases involving universal quantification remain to be explained. The logic of the argument should essentially be the same. If the embedded clause everyone who is nice is a non-smoker is not subjective, then it must be that the generalized quantifier everyone who is nice gets the same denotation across judges.
9.3 Summary
In this section, we have examined the distribution of predicates of personal taste under find, guided largely by the data that Sæbø (2009) presents as arguments for the pronoun theory of judge-dependency, and showed that these are either actually arguments against this view, or are at least just as compatible with an index theory. First, we have seen that the relation between find and its associated subjective predicate can be quite non-local. This means that some special mechanisms must
208 be available to allow semantic composition to proceed across a long-distance. Saebø adopts a DRT framework that allows this, given that it allows presupposed infor- mation to be pushed out of the way, and that the non-subjective material in the complement clause is presupposed. In a framework like Heim and Kratzer (1998) where covert movement would be the only way to establish this link, incorrect pre- dictions concerning the availability of placing the subjective predicate in a syntactic island are made. This is not a counterargument to Saebø’s proposal as much as estab- lishing its limitations: His analysis can only get off the ground so long as we adopt a semantic framework which, like DRT, allows semantic composition between non-local constituents when the intervening material is presupposed. We have also seen that the infelicitous nature of some conjunctions between a subjective and non-subjective predicate below find cannot serve as an argument for this view, since we can also come up with acceptable examples that have this structure, in direct contradiction with the pronoun theory. I rather propose an alternative analysis to explain away the infelicitous cases in the form of a pragmatic principle excluding standalone constituents whose sole semantic contribution is to strengthen the presupposition. Similarly, infelicitous examples where the subjective predicate appears inside the restrictor of a determiner do not suffice to conclude that find cannot be licensed by a subjective predicate in this position. Again, I provide examples of this sort that are acceptable to show this. This is in direct contradiction with the predictions made by the pronoun theory. This is a trickier case to explain, however. The generalization seems to be that subjective predicates in the restrictor of strong quantifiers in the sub- ject position of the complement clause cannot license find. The fact that the subject position in particular is problematic leads me to believe that information structure is involved, an approach that I cash out by claiming that a subjective predicate can only license find from inside a definite DP if this DP does not receive a given interpreta- tion, since otherwise its value cannot vary across judges. To the extent that a definite DP in subject position must be interpreted as given, this behaviour is expected. Ad- mittedly, this does not carry over very well to the case of universally quantified DPs, a problem that still begs an adequate analysis at this point. Although this last explanation is somewhat inconclusive, it should be clear that it is at least possible to come up with one such explanation for the infelicitous cases that still allows the subjective predicates inside the restrictor of determiners in object
209 position to be ruled in. Under a pronoun theory, it should be impossible for any sub- jective predicate to relate to find from inside the restrictor of a determiner, incorrectly ruling out many sentences that we have shown to be acceptable.
10 The Judge-Dependency of Gradable Adjectives
It would be overstating the generalization about opinion verbs to claim that they always require a predicate of personal taste in their scope. After all, it is easy to show that adjectives with perfectly measurable properties like the prototypical gradable adjective tall are sufficient to make the use of find felicitous:
(175) John finds his son tall.
Of course, height is not a subjective property, yet the small clause his son tall does seem to have a value that varies across judges. Saebø attributes this judge-dependency to the standard-fixing function POS, which many authors assume is responsible for providing the derivation with a standard of evaluation, which is simply a value for what counts as P for whatever gradable predicate P in a given context (see Kennedy (2007) for a an overview), in this case, providing a standard for tallness. Basically, he adds a judge argument to POS and makes it return a possibly different standard of tallness for different judges. Below is an initial candidate denotation for POS that incorporates this idea using judge-indices by making the function called norm subjective. POS takes as an argument a gradable property P and returns the set of individuals that have this property at least to some degree d which counts as the standard as fixed by the judge.
(176) [[ POS ]]w, j = λP
Saebø does not provide any explicit argument for this, other than rightly pointing out that POS is very often assumed to be a context-dependent operator, and that judges are simply one feature of the context. The formulation in (176) simply serves to make the contribution of the judge visible. I take this as a first step in what will be a more thorough investigation of the relation between subjectivity and gradability, which is the real purpose of this section. We have established earlier in this chapter that find is sensitive to the subjective character of its complement clause, and we now use this property to examine to what
210 extent more complex structures like comparatives are subjective, as well as various subclasses of adjectives under find. A well-known property of gradable adjectives is that they lose their context depen- dency when they appear in comparative constructions. That is, while (177) is context- dependent, (178) is not, and we do not get an intuition of faultless disagreement from the dialogue in (179). One of the two speakers must be wrong, and establishing which of the two this must be is subject to objective verification.
(177) John is tall.
(178) John is taller than Bill.
(179) A: John is taller than Bill. B: No, he isn’t!
This is all expected under standard assumptions in degree semantics. Since the locus of context-dependency for bare gradable adjectives is in the POS operator and not in the adjective itself, we expect this dependency to go away when a different, non-context-dependent, operator is used, such as the comparative. The facts are quite different with predicates of personal taste. Properly subjective predicates like fun do not lose their judge-dependency when they appear in compar- atives. Indeed, even the comparative constructions give rise to faultless disagreement with predicates of personal taste:
(180) A: Water slides are more fun than roller coasters. B: No, roller coasters are more fun that water slides.
In fact, as mentioned in footnote 4, I would like to suggest that it is this property which defines true predicates of personal taste:
(181) A predicate P is a predicate of personal taste iff it gives rise to faultless disagreement under comparative morphology25.
This definition is helpful in constructing a typology of adjectives, and relating it to the possibility of using an adjective to license opinion verbs. Predicates of personal
25This is implicitly recognized in Saebø’s paper since some of his examples of find contain predi- cates of personal taste in the comparative. Glanzberg (2007) also draws a parallel between predicates of personal taste and gradable adjectives, although neither author takes the further step of taking faultless disagreement in the comparative to be the defining property of predicates of personal taste.
211 taste are inherently judge-dependent in that the judge is instrumental not only in specifying a cut-off point for what is P and what is not P, but also in specifying the ordering on the domain of P. Because of this, predicates of personal taste remain judge- dependent even in the comparative and thus maintain their ability to license find in this construction. Ordinary gradable adjectives, in contrast, are only judge-dependent to the extent that they are accompanied by a POS operator, which is the actual locus of judge-dependency. Since no POS or any other judge-dependent operator is present in the comparative, then ordinary gradable adjectives in the comparative do not give rise to faultless disagreement and do not license find. Of course, other operators than POS turn out to be judge-dependent themselves, with inherently modalized degree operators too, enough,andso...that (cf. Meier (2003)) being the most obvious examples. Note that even when the equative as, which is clearly not judge-dependent in itself, is used with a predicate of personal taste, the result is still a subjective proposition.
(182) I find Mary as pretty as Lily.
We can refine this further by including Kennedy and McNally (2005)’s subdivision of gradable adjectives according to which they are classified depending on the structure of the scale that they involve, namely whether they are closed or open at the upper end and at the lower end. As discussed in more detail in Kennedy (2007), closed scales tend to make context-dependency disappear since they provide distinguished degrees on the scale, which the POS operator will be forced to refer to by some pragmatic principles. Upward and downward boundedness is tested for by the compatibility of an adjec- tive with adverbials that refer to these boundaries. The adverb slightly refers to the lower end of a scale, while absolutely refers to the higher end (Rotstein and Winter (2004); Kennedy and McNally (2005); Kennedy (2007)). If an adjective is compatible with slightly, then the scale that it involves is lower-bound, if it is compatible with absolutely, then it is upper bound. An adjective may be compatible with both, in which case it has a fully-closed scale, or neither, in which case it is fully open. All four possible options exist in natural language and are exemplified below26:
26There is a further complication if we take into account negative adjectives like short, empty, flat and pure, since the tests will then give us information about the opposite end of the scale then it would for the positive counterpart. That is, the fact that absolutely flat is acceptable shows that flat, a negative adjective, is lower-closed, while the fact that absolutely pure is acceptable shows that
212 (183) This television is perfectly/??slightly flat. (=upper closed) (184) This water is ??perfectly/slightly pure. (=lower closed) (185) This door is perfectly/slightly open. (=fully closed) (186) This man is ??perfectly/??slightly tall. (=fully open)
Adjectives used in the bare positive have a strong tendency to be used in a way that refers to one of those limit degrees whenever they have them. For example, a sentence like (187) is made true in a situation in which the door is only slightly ajar, as long as it is not completely closed, in other words that the degree to which it is open is higher than the minimal degree on the ‘open’ scale.
(187) The door is open.
In contrast, since tall is neither upper-closed or lower-closed, the sentence in (188) does not mean that John is maximally tall or just above minimally tall, just that he is taller than some contextually-specified standard. This is because there is no such degree as ‘minimally tall’ or ‘maximally tall’ to which the grammar could refer.
(188) John is tall.
The reason for this is that the grammar will make use of limit degrees that are directly provided by the scale as much as possible, rather than refer to some degree which is pragmatically provided. Kennedy (2007) attributes this to the work of a pragmatic principle which asks of the interpretive mechanisms that they only refer to contextually specified values when elements present in the compositional semantics do not provide any alternative. This is what he calls the principle of Interpretive Economy, which I quote below in full:
(189) Interpretive Economy Maximize the contribution of the conventional meanings of the elements of a sentence to the computation of its truth conditions.
There is some individual variation in the interpretation of bare gradable adjectives, but the strong tendency is that adjectives involving scales with salient minimal or pure, a positive adjective, is upper-closed. We will keep to positive adjectives for the rest of this chapter to avoid having to take this into account.
213 maximal degrees will tend to be context-independent, while those without any such distinguished degrees will be heavily dependent on context. To put this in compositional terms, the POS operator does not always introduce context-dependency and thus judge-dependency. It only does so if the adjective does not provide a scale that has natural distinguished degrees like a maximum or a mini- mum. How do we expect this to relate to the distribution of gradable adjectives under find? For our purposes, this means that gradable adjectives with open scales should be perfect in this environment, while those with closed scales should get worse the more salient their minimal or maximal degrees are. This correlation holds very well for fully open-scale adjectives (cf. (190)) and fully closed-scale adjectives (193), but the acceptability of (191) and (192) is somewhat surprising:
(190) John finds Mary tall/short. (191) John finds his appartment dirty/clean. (192) John finds this neighbourhood safe/dangerous. (193) ??/*John finds the door closed/closed.
Adjectives with open scales are predicted to be acceptable (because they are neces- sarily context-dependent) and those with fully-closed are predicted not to be (because they are not context-dependent at all27), so those two categories behave as predicted. However, those in between are also predicted to be bad under find since they do involve a distinguished degree and so should not be context-dependent. Their acceptability thus requires explanation. What this shows is simply that the pragmatic principle that Kennedy proposes should not be taken as an unbreakable rule, but rather as a tendency to favour con- ventional meaning whenever possible. If we let POS refer to a maximal or minimal element in sentences like (191) or (192), the result will be in contradiction with the Subjectivity Requirement, which should make these sentences infelicitous. Since they are not while (193) is, then we must conclude that bare adjectives with lower or upper bound scales must be able to refer to contextually-specified values as standards, while fully-closed adjectives simply do not have this option.
27Of course, this is an overstatement since even these so-called absolute adjectives can have a ‘loose’ interpretation, as is especially well-known of the famous example of the philosophy literature bald. All I am claiming is that the degree that they normally refer to when use in the positive form is not given by context.
214 10.1 Definiteness Restriction on Gradable Adjectives
While we have noted in section 9.1 that predicates of personal taste inside definite DPs are fine under find as long as they are not in subject position, this is not the case with ordinary gradable adjectives. In the examples below, the genuine subjective predicate right is capable of licensing find even in a definite DP, while the gradable adjective big in a similar context is not.
(194) John finds that you married the right girl. (195) #Peter finds that you wrote the big paper.
It seems that while a subjective predicate in a definite DP remain subjective, bare gradable adjectives do not. We will use this fact to establish more clearly what it is exactly about bare gradable adjectives that is subjective. Gradable adjectives appearing inside definite DPs get readings which strongly dif- fer from their counterparts in indefinite DPs (Kyburg and Morreau (2000), Kennedy (2007), Foppolo and Panzeri (2009), Syrett et al. (2010)). Importantly for our pur- poses, while gradable adjectives in indefinites require that this property hold of some individual in the context according to the default judge (the speaker, usually), grad- able adjectives in definite DPs have no such requirement. For example, suppose that a schoolteacher assigns a choice of two books to read to the children: one which is 50 pages long and one which is 75 pages long. Both books would be considered short by most adults, so even if Billy chose to read the 75-page book, I cannot truthfully say (196). I can, however, say (197):
(196) Billy read a long book. (197) Billy read the long book.
The uniqueness presupposition of the definite means that the context can only contain one individual which satisfies the description [POS long book]. This can be made true if there is a single book in the context (making the sentence infelicitous, since the adjective does not restrict anything, cf. Schlenker (2004)), or, more plausibly, if there is a single book which counts as long. For this to be the case, it must be that the POS operator returns a standard which stands exactly between the longest and the second longest book. This will leave only one individual in the positive extension of long and ensure that the uniqueness presupposition is satisfied. Any other value for
215 the standard would either leave more than one individual in the positive extension, or none at all, in either case failing to satisfy the requirements of the definite article. The most important observation is that this value for the standard need not cor- respond to the speaker’s standard for book length, nor anyone else’s for that matter. The POS operator is forced by the definite determiner to yield a value which does not correspond to the judgement of any contextually relevant judge. That is, the standard-fixing function in POS is not judge-dependent when it is under the definite, since it always returns as the standard a value which stands between the highest and second highest individual on the scale. Truth-conditionally, it is synonymous to the corresponding superlative form (in this case, the longest book). We can now see why gradable adjectives no longer license find when they are embedded in definite DPs. The explanation is simply that the POS operator that they come with is no longer judge-dependent in this environment. The clause Billy read a long book can be true or false depending on who the judge is, but not Billy read the long book. Hence (198) is felicitous while (199) is not:
(198) John finds that Billy read a long book. (199) #John finds that Billy read the long book.
10.1.1 Problems with Plurals
The explanation of the indefiniteness restriction just given rests heavily on the fact definite DPs may only pick out one individual in the context. However, consider what happens if a gradable adjective appears in a plural definite DP28, rather than a singular definite.
(200) Billy read the long books.
The result of turning the DP into a plural is that it should now be judge-dependent again. While the singular DP thelongbookcould only refer to a single object, namely the longest book in the context, the plural DP in (200) can refer to various plural individuals corresponding to different groups containing the longest and some of the shorter books. For example, if the context contains books that have a 100, 200, 300 and 400 pages, this plural DP can refer minimally to the 300 and 400 page books together, or
28I thank Alan Bale for pointing out this problem with the account laid out in the previous section.
216 to the 200, 300 and 400 page books. The more books there are in the context, the more this variability will be visible. The plural means that the DP could not pick out only the longest one, and the use of the attributive adjective would make this DP infelicitous if it referred to all the books in the context. Whether it picks out the two longest or the three longest, however, seems to be a matter of what one judges to be actually long. Thus the prediction seems to be that such DPs should be able to license opinion verbs since their value varies across judges. This is not the case, however. Ordinary gradable adjectives inside plural definites fare no better that those in singular definites in this context:
(201) #John finds that Billy read the long books.
We now consider revising our earlier assumptions and displace the source of the judge-dependency in gradable adjectives from the POS operator to the comparison class.
10.1.2 An Alternative Analysis: Subjective Comparison Classes
Although the explanation we have given so far is intuitively appealing, there is an alternative that covers both plural and singular examples29. Suppose that what is subjective about positive gradable adjectives is not the standard-fixing function POS, but rather the comparison class, that we will take to be expressed by a covert set- denoting pronominal. This is an intuitively appealing view. When I say that Mary is tall, my personal viewpoint comes into play in determining who counts as a tall woman for me by determining the women that I have known in my life, the average height of women in the region of the world I grew up in, i.e. those women that I compare Mary to in assessing whether she is tall or not. It does not come into play so much in how I derive a standard of tallness from a set of individuals, as we implicitly assume in making the norm function in POS subjective. In fact, it has been suggested by Fern´andez (2009) and others that it may actually be possible to come up with a reasonable algorithm that derives a standard of, say, tallness, from a set of objects in a way that speakers consistently agree with. If this proposal is correct, then Saebø’s suggestion that we make the standard-fixing function POS judge-dependent in the
29I thank Sacha Simonenko for convincing me of the correctness of this solution, despite initial resistance on my part.
217 way we did in (176) must be wrong, since POS should then be deterministic, so to speak. It would always return the same standard value from a given comparison class. One way to include comparison classes in the semantics of gradable adjectives is simply to treat them as sets of individuals30 which appear at LF as covert pronominal arguments to POS. We will note this pronominal Cw, j and let its value vary across judges in a given world, just like any predicate of personal taste.
(202) John is POS-Cw, j tall.
Now we can treat POS as objective and give it the following semantics, leaving to the function norm the task of deriving a standard on the basis of the comparison class and the adjective, in a manner to be specified, possibly as suggested by Fern´andez (2009)31.
(203) [[ POS ]]w, j = λC
Contextual restrictions of this sort are commonly assumed to play a role in re- stricting the claims that are made by sentences with universal quantification. In fact, comparison class pronominals are more naturally seen as an extension of the better-known notion of quantifier restriction pronominals such as those proposed in Von Fintel (1994). Following Schwarz (2009), I will assume that such restrictions appear at LF as arguments to the determiner, and not of the nominal head as Stanley and Szab´o (2000) propose. For example, we can just alter the semantics for every to make it take an additional set argument, whose value is assumed to be provided by context, and it is the intersection of this set with the restrictor that is claimed to be
30I am aware that complications arise from treating comparison classes as sets of individuals, even relativized to contexts or events as in Stanley and Szab´o (2000) and Stanley (2002) (cf. Schwarz (2009), who proposes a analysis in terms of situation semantics.). My purpose here is not to claim that comparison classes are not relativized to contextual factors other than the identity of the judge, just that the judge is an important factor. The general effect of context is obvious when we consider examples like the much cited example (1), originally from Kamp (1975), in which what counts as a remarkable violinist may vary depending on whether the context of utterance is a dinner party or a professional concert.
(1) Smith is a remarkable violinist. 31To be somewhat more specific, Fernandez proposes an algorithm which seeks out, for a given set C, the most prominent differences between subsets of C. The standard for a gradable property P given a comparison class C is one that divides this set into two maximally distinct subsets. It is actually of no importance to my proposal whether this procedure is truly a part of the meaning of POS and thus a genuine part of the faculty of language, or whether it is dependent on other cognitive faculties relevant for constructing categories. My point simply hinges on the proposal that POS is not subjective, although it may be more imprecise than Fernandez’ proposal makes it.
218 a subset of the scope argument. In (205), C could refer to the things that we own to make sure that the sentence only says anything about our cats, not all the cats in the universe.
(204) [[ every ]] = λPλQλR. P∩Q ⊆ R (205) Every-C cat left the room. ≈ Every cat that we own left the room.
Of course this can be generalized to other determiners. Much as with universal quantification, which would only very rarely yield true sentences if it were not re- stricted, the definite determiner can be assumed to introduce a contextual restriction on its NP argument, otherwise it would rarely be able to pick out a single individual and, consequently, be able to refer.
(206) [[ the ]] = λCλQ. ιx[x∈C∩Q] (207) The-C cat came back in. ≈ The unique cat that we own came back in.
Contextual restrictions introduced by determiners and by bare gradable adjectives (or rather, their associated POS operator) are not independent objects. Attributive adjectives inside definite DPs really involve just one contextual restriction that serves to restrict both the comparison class for the adjective as well as the domain for the definite determiner.
(208) The-C POS-C big cat came back in. ≈ The unique cat that we own and that is big for a cat we own came back in32.
32To be precise, the comparison class is not equal to the determiner’s domain restriction argument, but rather to the intersection of that set with the denotation of the head noun. That is, the cat is presupposed to be big for a cat we own, but not big more generally for anything we own. We can treat this as resulting from a syntax where the POS operator moves out of its base position next to the adjective to a position just below the determiner. From its moved position, POS thus combines with the following constituent at LF:
(1) λd. [ [ tPOS big ] cat ]
The constituent below the lambda abstractor is interpreted as the intersection of [[ cat ]] and [[ t big ]], which is simply the set of things that are both a cat and have some measure of bigness, i.e. are in the domain of big. With the lambda, we get a derived gradable function that maps any degree d to the set of individuals such that they are both at least d-big and a cat.
219 It is important at this point to point out that while the comparison class argument for gradable adjectives must be assumed to be judge-dependent to account for the fact that bare gradable adjectives give rise to intuitions of faultless disagreement, the otherwise very similar domain restriction argument for determiners must not be judge-dependent, precisely because it does not trigger faultless disagreement (cf. (209), where we get the intuition that either A or B must be wrong), nor does a determiner with such an argument suffice to license the use of find in the absence of any other subjective predicate in the complement clause, as shown by (210).
(209) A: Every cat came back in. B: That’s not true, Mittens is still outside. (210) #John finds that he saw every movie.
The interesting thing for our purposes is the observation that while the comparison class argument usually has a value that is subjective, this is not the case when it appears inside a DP whose head introduces contextual restrictions. This is what our example with the definite example shows: I can felicitously use a sentence like (208) in a situation where I own four cats, three of which are skinny and the other which is of average size, if the largest came in. Crucially, I do not have to find this cat big with respect to cats in general, or some with cats that I have known in my life. The comparison class that I use is only those cats as contextually restricted by the domain restriction argument on the definite determiner. This is now the explanation that we give for the non-subjective nature of examples like (208): the potential locus for subjectivity in this sentence, the comparison class argument, now has a value that is tied to the value of another element in the clause, namely the domain restriction pronominal C introduced by the determiner. The easiest way to spell this out is to claim that the comparison class argument is obligatorily bound by the domain restriction argument, forcing the latter to have the same value as the former whenever an attributive adjective modifies the head noun of a DP. The most straightforward advantage of this approach is that it allows us to explain our initial puzzle, namely why gradable adjectives in definite DPs cannot license find,
This is the gradable property that POS combines with after its comparison class argument. In the end, the head noun and the comparison class pronominal end up restricting the actual comparison class in two different way: the noun by being an integral part of the derived gradable property, and the latter as a direct argument of POS.
220 in a way that will cover singular and plural cases equally well. The challenge, to reiterate, it to explain why sentences like (211) are infelicitous, which in turn involves explaining why DPs like the long book and the long books are not subjective.
(211) #John finds that Billy read the long book(s).
We already have everything we need to explain this phenomenon now. Gradable adjectives are only as subjective as their comparison classes are. This is where we have located the source of judge-dependency for those predicates in this section. We know that attributive adjectives are forced to use as their comparison class the set that is provided by context as the restriction on the determiner. This will be the case regardless whether the DP is singular or plural. In both cases, the comparison class pronominal of the adjective will be bound by the domain restriction of the determiner. The explanatory power of this analysis is no longer tied to the uniqueness presup- position of the definite determiner, but rather to the fact that it introduces a domain restriction. This makes a straightforward prediction: We should find that every de- terminer that introduces such domain restrictions makes any attributive gradable adjective no longer subjective. I believe this is essentially correct. Determiners with universal force always require contextual restrictions to have any chance of making true claims, while indefinites do not. As shown below, gradable adjectives inside DPs headed by every, the and no all fail to license find.
(212) #John finds that Billy read every long book. (213) #John finds that Billy read no long book. (214) #John finds that Billy read two long books.
The major failing of this analysis is that it does not offer any explanation for why it is that the comparison class argument must be bound by the quantifier’s domain restriction argument. We have independent arguments for postulating both kinds of domain restriction, since comparison classes are relevant in sentences with no deter- miner, and quantifier domain restrictions are relevant in sentences with no gradable adjective. The conclusion that they interact in a meaningful way is inescapable given the observation that the comparison class is fixed by context in sentences like (208). Yet we cannot say either that they are simply both fixed by context in the same way, which would force them to co-vary simply by virtue of the fact that they get their meaning from the same source, since we have shown that comparison class arguments
221 are judge-dependent, while quantifier domain restrictions are not. I will have no more to say on this subject here, and I leave it open to further inquiry to explain why it is that these two types of domain restrictions are forced to co-vary in this structural configuration.
11 Conclusion
In this chapter, we have discussed how the properties of opinion verbs can inform us on how subjectivity should be modeled in semantic theory. We have seen that some previously unnoticed presupposition properties of such verbs, once properly formalized in the form of the Subjective Contingency Presupposition, provide us with an easy way of accounting for the fact that these verbs are only felicitous when followed by a subjective clause. In this sense, opinion verbs are remarkably well-behaved: they can only be used with a complement that one could judge true or false. It is unmistakable that it is really a grammatical notion of subjectivity that is at play here, since we can see that essentially synonymous expressions like in my opinion and in my books do not behave in the same way: only the latter is sensitive to this semantic property. Accounting for the Subjectivity Requirement by appealing to the presupposition properties of find does not make Saebø’s type-based approach wrong, however, just insufficient, since the account that we suggest correctly predicts that analytical state- ments below find should not be felicitous, a prediction that is not made by the type- approach. What does make it wrong however is the possibility of coordination be- tween subjective and non-subjective statements under find, which we have argued is restricted by pragmatic factors, not compositional ones. A similar argument comes from the fact that subjective predicates inside the restrictor of DPs are acceptable, at least in some syntactic contexts. Treating subjective predicates as different in type from non-subjective ones makes it hard to see how the types could line up properly in an environment like the complement of find, where the judge-argument of sub- jective predicates must be left open. Saebø finds some structures where subjective predicates appear uncomfortable when they appear below find, but once we examine them in more detail, we can see that they are not really quite as problematic as they appear and we can explain away many of these examples in the light of the Subjective Contingency Presupposition. Some problematic cases remain, but it appears to me a better methodology and a better match to the data to provide a basic semantics
222 that makes even the problematic sentences interpretable, and focusing on ruling the unacceptable cases by appealing to pragmatic principles. This verb’s requirement for a subjective complement makes it a formidable tool for investigating judge-dependency, since it allows us to always verify whether a clause is actually judge-dependent or not. We have used this to look more closely at grad- able adjectives, and concluded that the definiteness effect that emerges with these predicates can receive a comprehensive explanation if we assume that the locus of judge-dependency in bare gradable adjectives is actually in the comparison class ar- gument, not in POS itself. This analysis should be seen as an instance of the type of investigation that one can do using the complement of find to figure out exactly what it is in natural language that is truly and grammatically subjective.
223 Appendix A: A Judge-Free Semantics for Subjectivity
In examining the grammar of subjectivity, I have mostly been concerned with the two options for modeling subjectivity outlined in chapter 3: either judges are represented as pronominal arguments to predicates of personal taste, or they are introduced in the index. This is not to say that these options exhaust the possibilities for expressing subjectivity. A very recent proposal is made by Pearson (2012), who entirely rejects the concept of judges, hence her claim to a “judge-free semantics” for predicates of personal taste. What she proposes is that what we are dealing with when we find predicates of personal taste is an instance of first-person oriented genericity (Molt- mann (2006); Moltmann (2010)). That is, whenever we make an ordinary subjective statement like (1), we make the claim that borsch is tasty for people in general based on our own first person experience.
(1) Borsch is tasty.
Unlike Stojanovic, she does not claim that such sentences are ambiguous between a generic and personal reading, but rather they only receive a generic one. Predicates of personal taste are treated as transitive predicates whose internal argument, if covert, can only be a pronominal whose denotation is that of a variable that ranges over individuals that the speaker “identifies with”, and who have direct preceptual evidence of the right kind to find something tasty, fun, etc. Tasty thus gets the denotation in (2), and its internal argument gets the interpretation in (3), whereIisthe“`ıdentifies with” relation:
(2) [[ tasty ]]w = λxλy: x has direct perceptual experience of the relevant kind of y in w. y is tasty to x in w
(3) [[ pro1 ]]w = λx. I(speaker, x)
Pearson does not go into the details of who exactly it is that a speaker identifies with, except that we always identify with ourselves and, in normal circumstances, we identify with the person we are talking to. Presumably, what she means is that this variable should range over people that the speaker assumes have the same tastes as her.
224 Pearson draws a parallel with individual-level predicates (ILPs), noting that the paradigm predicates of personal taste fun and tasty pass all or nearly all the tests for individual level predicates1. We can see this from simple tests meant to show the presence or absence of an event variable in the adjective. The ungrammaticality of examples (4-8), and the unavailability of an existential reading for the indefinite in (9) are all arguments that show that tasty is an individual-level predicate. These tests are largely borrowed from Milsark (1979) and Carlson (1980).
(4) *There were cakes tasty. (5) *John has seen the cake tasty. (6) *When Nutella is tasty, Mary is happy. (7) *Nutella is always tasty. (8) *Nutella is tasty in the morning. (9) A cake is tasty. (gets a universal reading, not a existential one).
It has been proposed by Chierchia (1995) that ILPs always involve an LF where a genericity operator GEN is present. What GEN does is the same as always or usually: It says that a proposition is true in all accessible worlds where the subject is relevant.
(10) John (GEN) is tall. (11) ∀w’:[Acc(w, w’) & C(John, w’)]. [tall(John, w’)]
Understand C(x, w) to be true iff x exists and is relevant in w. The accessibility relation is meant to exclude counterfactual worlds, say worlds where John is shorter than he actually is. The idea now is to extend this to predicates of personal taste.
(12) This cake is tasty. (13) ∀x, w’:[Acc(w, w’) & C(this-cake, x, w’) & I(speaker, x)]. [tasty(this-cake, x, w’)]
1Some tests would seem to make fun stage-level, such as the acceptability of sentences like (i): (i) Monopoly is always fun. This is probably just a consequence of the fact that fun is allowed to take non-finite clauses as subjects, and (i) should just be interpreted as if the subject were not the proper name Monopoly, but the clause Playing Monopoly. It is the event variable of the verb playing which is bound by always, not some event variable which would be part of the meaning of fun.
225 The three-place C function is similar to the two-place version above: C(x, y, w) is true iff x and y are relevant and exist in w. The entire formula says that (12) is true iff this cake is tasty to every person that the speaker identifies with, in all accessible worlds where that person and this cake exist and are relevant. This analysis is meant to specifically capture the generic character of predicates of personal taste, as well as its speaker-oriented character. It does this by forcing the presence of the GEN operator at LF whenever a sentence contains a predicate of personal taste, and forcing the internal argument of PPTs, if covert, to be a variable ranging over individuals that the speaker identifies with. The main problem with this analysis and the reason why I do not adopt it in this chapter is that, contrary to what Pearson claims, it fails to account for the intuition of faultless disagreement. It also makes incorrect predictions concerning the availability of exocentric and personal interpretations. The explanatory weight of Pearson’s account of faultless disagreement rests on the I function. This intuition is what happens when two speakers identify with each other, but wrongly so. In an ordinary setting, two speakers naturally assume that they identify with each other. So when A says that the ride is fun, she includes B in the set of people for whom the ride is said to be fun. This characterization of the now familiar dialogue of disagreement does not predict it to differ from any ordinary, factual disagreement. If A said that the ride was fun and included B in the people that she identifies with, and B did not find the ride fun, then A has made a false statement, according to this theory. This is an obligatory consequence of the universal quantification involved in this semantics. As the reader can verify, the following formula yields false if I(A, B) is true and fun(this-ride, B, w’) is false for some appropriate, non-counterfactual world.
(14) This ride was fun. (15) ∀x, w’:[Acc(w, w’) & C(this-ride, x, w’) & I(A, x)]. [fun(this-ride, x, w’)]
There is simply no sense in which this semantics can make disagreement faultless. As long as a speaker identifies with the person she is talking to, but it turns out that they have different opinions, then this semantics will yield false for a subjective statement that they disagree with. I believe this is much too strong. Even supposing we accept Stojanovic’s claim that there is no such a thing as fault- less disagreement, then there is still an additional problem concerning the availability
226 of exocentric and personal readings. Exocentric readings are made possible in Pear- son’s system when the speaker identifies with some individuals who are apt judges for the subjective property, but where the speaker is not. The catfood examples we have seen in discussing Stephenson are clear cases of this. If John claims that the catfood is tasty and understands this to mean that it is tasty for the cat or cats in general, then John must identify with his cat or cats in general for the purposes of this sen- tence. Since tasty has a presupposition that the set of possible internal arguments can only contain individuals who have relevant experience, i.e. people who have tasted the catfood in this example, John is excluded from individuals that the variable that saturates the internal argument ranges over. As such, the sentence is made true as long as all the individuals that John identifies with for the purposes of this sentence, namely his cat or cats in general, find the catfood tasty, while John does not need to find the catfood tasty himself. Here, the explanatory weight in making exocentric readings possible rests on the domain restriction on PPTs that makes sure that they only concern apt judges. How- ever, this is clearly not permissive enough. As Lasersohn (2009) amply discusses, for example, I can very well go on a ride with a young child, find the ride uninteresting myself, and still come out and say that the ride was fun because the child enjoyed it. I am, in Lasersohn’s words, “putting myself in someone else’s shoes”. It is not necessary for the speaker to be unavailable as a judge for an exocentric reading to be possible, which is what is predicted by Pearson’s theory. The explanatory weight thus does not rest on the right place. Of course, it is also predicted that using predicates of personal taste to express nothing more than a personal opinion should never be possible, since the internal argument of such predicates can only be saturated by the special pronoun targeting those that the speaker identifies with. As most other authors who discuss this issue usually agree, such readings are certainly possible and forcing subjective predicates to be consistantly generic is surely too restrictive. Given these inadequacies, I exclude Pearson’s theory from this thesis and concen- trate on comparing the index and pronoun theories of judge-dependency.
227 Appendix B: IAD across Opinion Verbs
As mentionned in section 4.4 of chapter 2, IAD can take a degree operator over an opinion verb like trouver.
(1) J’ ai tellement trouv´e que Paul a donn´e un bon show. I have so found that Paul has given a good show ‘I found that Paul gave such a great show.’
Given that chapter 3 is devoted precisely to the exploration of the semantics of such verbs, I have postponed the analysis of such sentences. In this section, I will show that the fact that these sentences are acceptable at all and are not judged to be in violation of the Subjectivity Requirement argues in favor of the in situ analysis of IAD. The movement analysis would involve a representation whereby the complement clause of trouver would not have a value that can vary across possible judgements, since the degree argument of the gradable predicate would be saturated by a simple degree variable. First of all, notice that this configuration is grammatical even in cases where the licenser of the opinion verb is gradable, but it is not a predicate of personal taste:
(2) J’ ai tellement trouv´e que Paul a donn´e un long show. I have so found that Paul has given a long show ‘I found that Paul gave such a long show.’
Under the movement theory of IAD, this sentence is derived by raising the operator tellement from beside the gradable adjective long, leaving a type-d trace and triggering lambda abstraction over this trace just below the operator’s landing site. At LF, then, we have the following configuration:
(3) tellement [ λd. [ j’ai trouv´e [ que Paul a donn´e un d-gros show ]]]
Recall from section 10 of chapter 3 that ordinary gradable predicates are only subjective when used in the bare form, since while these adjectives have a constant meaning across judges, the standard-fixing function that they combine with in the bare form requires a comparison class argument, and the value of this element can
228 vary across judges. In (2), however, there is no standard-fixing function, and thus no comparison class argument, only a type-d trace. The complement clause Paul a donn´e un d-gros show does not have an extension whose value can vary across judges, and as such it should be associated with a contradictory presupposition, and thus be judged infelicitous. We do not even need to calculate the semantic contribution of the degree operator at this point because it will not be able to rescue the presupposition. Since this sentence is fine, however, I conclude that this derivation cannot be right. The in situ analysis of IAD does not have this problem, however. Under this view, IAD sentences involve bare gradable predicates in the normal sense, meaning that they combine locally with the POS operator, which in turn brings about a comparison class (CC) argument.
(4) tellement [ j’ai trouv´e[quePauladonn´e un POS-CC-gros show ]]
Since the value of CC is subjective, the Subjectivity Requirement is respected and the sentence is correctly predicted to be acceptable. Recall that according to the in situ analysis of IAD, the degree operator does not directly relate to the gradable predicate in its scope, so the subjective or non- subjective character of this predicate should not be affected by the presence or absence of a degree operator higher in the structure. This means that IAD sentences where the degree operator appears above an opinion verb will should always be as acceptable as their non-IAD counterpart, since they are, in fact, the same at LF up to the point where the degree operator is merged. Since it is merged above the opinion verb itself, then it will never be in a position to interfer with the Subjectivity Requirement and IAD is thus predicted to be possible across opinion verbs.
229 References
Abney, S.P. 1987. The English noun phrase in its sentential aspect. Doctoral Disser- tation, Massachusetts Institute of Technology.
Bartsch, R. 1972. The grammar of relative adjectives and comparison. Formal Aspects of Cognitive Processes 168–185.
Beck, S. 2006. Intervention effects follow from focus interpretation. Natural Language Semantics 14:1–56.
Beghelli, F. 1995. The phrase structure of quantifier scope. Doctoral Dissertation, Ph. D. thesis, UCLA.
Bhatt, R., and R. Pancheva. 2004. Late merger of degree clauses. Linguistic Inquiry 35:1–45.
Bhatt, R., and R. Pancheva. 2007. Degree quantifiers, position of merger effects with their restrictors, and conservativity. Direct compositionality 306–335.
Bhatt, R., and S. Takahashi. 2011. Direct comparisons: Resurrecting the direct analysis of phrasal comparatives. In Proceedings of SALT , volume 17, 19–36.
Bierwisch, M. 1989. The semantics of gradation. Dimensional adjectives 71:261.
Boivin, M.C., and D. Valois. 2009. L’intensification `a distance en fran¸cais qu´eb´ecois. Proceedings of the CLA 2009 Conference .
Bouchard, D.E., and H. Burnett. 2010. La quantification et intensification `a distance en fran¸cais qu´eb´ecois. In Vues sur les fran¸cais d’ici, ed. C. Leblanc, 31–48. Presses Universitaires de Laval.
Bouchard, D.E.,´ H. Burnett, and D. Valois. 2011. Degree fronting in Qu´ebec French and the syntactic structure of degree quantifier DPs. Romance Languages and Linguistic Theory 2009: Selected Papers from’going Romance’Nice 2009 39.
Bresnan, J.W. 1973. Syntax of the comparative clause construction in English. Lin- guistic inquiry 4:275–343.
Burnett, H. 2012. The role of microvariation in the study of semantic universals: Adverbial quantifiers in European and Qu´ebec French. Journal of Semantics 29:1– 38.
Carlson, G.N. 1980. Reference to kinds in English. Garland Pub.
230 Chemla, E. 2009. Presuppositions of quantified sentences: experimental data. Natural Language Semantics 17:299–340.
Chierchia, G. 1995. Individual-level predicates as inherent generics. The generic book 125.
Chomsky, N. 1977. On wh-movement. Formal Syntax 71–132.
Chomsky, N. 1995. The Minimalist Program, volume 28. Cambridge University Press.
Chung, S., and W.A. Ladusaw. 2004. Restriction and saturation, volume 42. the MIT Press.
Cinque, G. 1999. Adverbs and functional heads: A cross-linguistic perspective.Oxford University Press, USA.
Cinque, G. 2010. The syntax of adjectives: A comparative study, volume 57. MIT Press.
Clark, H.H., and E.V. Clark. 1977. Psychology and language. Cambridge Univ Press.
Corver, N. 1997. The internal syntax of the Dutch extended adjectival projection. Natural Language & Linguistic Theory 15:289–368.
Cyr, F. 1991. La quantification `a distance en fran¸cais qu´eb´ecois. Doctoral Dissertation, Universit´edeMontr´eal.
Davies, W.D., and S. Dubinsky. 2003. On extraction from NPs. Natural Language & Linguistic Theory 21:1–37.
Dekydtspotter, L., R.A. Sprouse, and R. Thyre. 2000. The interpretation of quan- tification at a distance in English-French interlanguage: Domain specificity and second-language acquisition. Language Acquisition 8:265–320.
Doetjes, J. 2007. Adverbs and quantification: degrees versus frequency. Lingua 117:685–720.
Doetjes, J., and M. Honcoop. 1997. The semantics of event-related readings: A case for pair-quantification. Studies in Linguistics and Philosophy 263–310.
Doetjes, J.S. 1997. Quantifiers and selection: On the distribution of quantifying ex- pressions in French, Dutch and English, volume 32. Holland Academic Graphics.
Ducrot, O. 1975. Je trouve que. Semantikos 1:63–88.
Farkas, D.F., and K.E. Kiss. 2000. On the comparative and absolute readings of superlatives. Natural Language & Linguistic Theory 18:417–455.
231 Fern´andez, R. 2009. Salience and feature variability in definite descriptions with positive-form vague adjectives. In Proceedings of COGSCI . von Fintel, K., and I. Heim. 2007. Intensional semantics lecture notes.
Foppolo, F., and F. Panzeri. 2009. Do children know when their room counts as “clean”? NELS 40 40.
Fox, D. 2000. Economy and semantic interpretation, volume 35. The MIT Press.
Fox, D., and J. Nissenbaum. 1999. Extraposition and scope: A case for overt QR. In Proceedings of the 18th West Coast Conference on Formal Linguistics, volume 18, 132–144.
Gajewski, J. 2002. L-analyticity and natural language. Manuscript .
Giannakidou, A., and M. Stavrou. 2008. On metalinguistic comparatives and nega- tion in Greek. Greek Syntax and Semantics. MIT Working Papers in Linguistics, Cambridge, MA: MIT Linguistics .
Giannakidou, A., and S. Yoon. 2009. Metalinguistic comparatives in Greek and Ko- rean: Attitude semantics, expressive content, and negative polarity items. In Pro- ceedings of Sinn und Bedeutung , 141–156.
Grosu, A., and J. Horvath. 2006. Reply to Bhatt and Pancheva’s ’late merger of degree clauses’: The irrelevance of (non) conservativity. Linguistic Inquiry 37:457–483.
Hackl, M. 2000. Comparative quantifiers. Doctoral Dissertation, MIT.
Hacquard, V. 2011. Aspects of” too” and” enough” constructions. In Proceedings of SALT , volume 15, 80–97.
Hankamer, J. 1973. Why there are two than’s in English. In Proceedings of the 9th Annual Meeting of the Chicago Linguistics Society. Chicago Linguistics Society, Chicago, IL.
Heim, I. 1982. The semantics of definite and indefinite noun phrases. University of Massachusetts.
Heim, I. 1985. Notes on comparatives and related matters. Unpublished ms., Univer- sity of Texas, Austin .
Heim, I. 1991. Artikel und definitheit. Semantik: ein internationales Handbuch der Zeitgen¨ossischen forschung 487–535.
Heim, I. 1999. Notes on superlatives. Unpublished ms. .
232 Heim, I. 2001. Degree operators and scope. In Proceedings of SALT , volume 10, 40–64.
Heim, I. 2006. Little. In Proceedings of SALT XVI , ed. J. Gibson and J. Howell. CLC Publications.
Heim, Irene, and Angelika Kratzer. 1998. Semantics in generative grammar,vol- ume 13. Wiley-Blackwell.
Heyd, S. 2003. L’interpr´etation des syntagmes nominaux en” des” et” de” en posi- tion sujet et objet: g´en´ericit´e, habitualit´e et incorporation s´emantique. Doctoral Dissertation, strasbourg 2.
Hintikka, J. 1969. Semantics for propositional attitudes. Models for modalities 87–112.
Hoeksema, J. 1983. Negative polarity and the comparative. Natural Language & Linguistic Theory 1:403–434.
Horn, L. R. 1972. On the semantic properties of the logical operators in English. Doctoral Dissertation, UCLA.
Huang, C.T.J. 1982. Move wh in a language without wh-movement. The linguistic review 1:369–416.
Kamp, H. 1975. Two theories about adjectives. Formal semantics of natural language 123–155.
Kaplan, D. 1989. Demonstratives. In Themes from kaplan, ed. John Perry Joseph Al- mog and Howard Wettstein.
Kayne, R.S. 1975. French syntax: The transformational cycle. MIT Press (Cambridge, Mass.).
Kennedy, C. 1997. Projecting the adjective: the syntax and semantics of gradability and comparison. University of California, Santa Cruz.
Kennedy, C. 2002. Comparative deletion and optimality in syntax. Natural Language & Linguistic Theory 20:553–621.
Kennedy, C. 2007. Vagueness and grammar: The semantics of relative and absolute gradable adjectives. Linguistics and Philosophy 30:1–45.
Kennedy, C., and L. McNally. 2005. Scale structure, degree modification, and the semantics of gradable predicates. Language 345–381.
Kiss, K.E.´ 1995. NP movement, operator movement, and scrambling in Hungarian. Discourse configurational languages 207:43.
233 K¨olbel, M. 2002. Truth without objectivity. Psychology Press.
K¨olbel, M. 2004. Faultless disagreement. In Proceedings of the Aristotelian Society , 53–73. Wiley Online Library.
Koopman, H., and D. Sportiche. 1991. The position of subjects. Lingua 85:211–258.
Kratzer, A. 1977. What ‘must’ and ‘can’ must and can mean. Linguistics and Phi- losophy 1:337–355.
Kucerov´a, I. 2008. Givenness and maximize presupposition. In Proceedings of Sinn und Bedeutung, volume 12, 353–366.
Kyburg, A., and M. Morreau. 2000. Fitting words: Vague language in context. Lin- guistics and Philosophy 23:577–597.
Lasersohn, P. 1999. Pragmatic halos. Language 522–551.
Lasersohn, P. 2005. Context dependence, disagreement, and predicates of personal taste. Linguistics and philosophy 28:643–686.
Lasersohn, P. 2009. Relative truth, speaker commitment, and control of implicit arguments. Synthese 166:359–374.
Lebeaux, D. 1991. Relative clauses, licensing, and the nature of the derivation. Syntax and semantics 25:209–239.
Lechner, W. 2004. Ellipsis in comparatives, volume 72. Mouton de Gruyter.
Levinson, S.C. 1987. Minimization and conversational inference. The pragmatic per- spective 61–129.
Link, G. 1983. The logical analysis of plurals and mass terms: A lattice-theoretical approach. Formal Semantics 127–146.
MacFarlane, J. 2004. Epistemic modalities and relative truth. Unpublished ms. .
Matushansky, O. 2002a. Movement of degree/degree of movement. Doctoral Disser- tation, Massachusetts Institute of Technology.
Matushansky, O. 2002b. Tipping the scales: The syntax of scalarity in the complement of seem. Syntax 5:219–276.
McCready, E. 2005. The dynamics of particles. Doctoral Dissertation, Ph. D. thesis, University of Texas at Austin.
McCready, E. 2008. What man does. Linguistics and philosophy 31:671–724.
234 Meier, C. 2003. The meaning of too, enough, and so... that. Natural Language Semantics 11:69–107.
Merchant, J. 2009. Phrasal and clausal comparatives in Greek and the abstractness of syntax. Journal of Greek Linguistics 9:134–164.
Milner, J.C. 1978. De la syntaxea ` l’interpr´etation: quantit´es, insultes, exclamations. Seuil.
Milsark, G.L. 1979. Existential sentences in English, volume 19. Garland New York and London.
Moltmann, F. 2006. Generic one, arbitrary pro, and the first person. Natural language semantics 14:257–281.
Moltmann, F. 2010. Generalizing detached self-reference and the semantics of generic one. Mind & Language 25:440–473.
Moritz, L., and D. Valois. 1994. Pied-piping and specifier-head agreement. Linguistic inquiry 667–707.
Morzycki, M. 2009. Degree modification of gradable nouns: size adjectives and ad- nominal degree morphemes. Natural Language Semantics 17:175–203.
Morzycki, M. 2011. Metalinguistic comparison in an alternative semantics for impre- cision. Natural Language Semantics 19:39–86.
Nissenbaum, J., and B. Schwarz. 2011. Parasitic degree phrases. Natural Language Semantics 19:1–38.
Nouwen, R. 2007. Predicates of (im)personal taste. Unpublished ms. .
Obenauer, H.G. 1983. Une quantification non canonique: la quantificationa ` distance. Langue fran¸caise 58:66–88.
Obenauer, H.G. 1994. Aspects de la syntaxe a-barre: Effets d’intervention et mouve- ments des quantifieurs. Doctoral Dissertation, ANRT Universit´e de Lille.
Pearson, H. 2012. A judge-free semantics for predicates of personal taste. Natural Language Semantics .
Percus, O. 2000. Constraints on some other variables in syntax. Natural Language Semantics 8:173–229.
Postal, P.M. 1974. On certain ambiguities. Linguistic Inquiry 5:367–424.
Rett, J. 2011. Antonymy and evaluativity. In Proceedings of SALT , volume 17, 210–227.
235 Ross, J.R. 1964. A partial grammar of English superlatives, MA thesis. Doctoral Dissertation, University of Pennsylvania.
Rotstein, C., and Y. Winter. 2004. Total adjectives vs. partial adjectives: Scale structure and higher-order modifiers. Natural Language Semantics 12:259–288.
Rullmann, H. 1995. Maximality in the semantics of wh-constructions. Doctoral Dis- sertation, University of Massachusetts, Amherst, MA.
Russell, B. 1905. On denoting. Mind 14:479–493.
Sæbø, K. J. 2009. Judgment ascriptions. Linguistics and philosophy 32:327–352.
Sag, I. 1976. Deletion and logical form, unpublished doctoral dissertation. Doctoral Dissertation.
Schlenker, P. 2004. Minimize restrictors! (notes on definite descriptions, condition C and epithets). In Proceedings of Sinn und Bedeutung , 385–416.
Schwarz, B. 2010. A note on for-phrases and derived scales. Handout of a talk given at Sinn und Bedeutung 15.
Schwarz, F. 2009. Two types of definites in natural language. Doctoral Dissertation, University of Massachusetts Amherst.
Schwarzschild, R. 1992. Types of plural individuals. Linguistics and Philosophy 15.
Schwarzschild, R. 2008. The semantics of comparatives and other degree construc- tions. Language and Linguistics Compass 2:308–331.
Shanon, B. 1976. On the two kinds of presuppositions in natural language. Founda- tions of Language 14:247–249.
Sharvit, Y., and P. Stateva. 2002. Superlative expressions, context, and focus. Lin- guistics and Philosophy 25:453–504.
Skopeteas, S., and G. Fanselow. 2007. Effects of givenness and constraints on free word order. Information Structure from different perspectives .
Stalnaker, R.C. 1978. Assertion. Formal Semantics 147–161.
Stanley, J. 2002. Nominal restriction. Logical form and language 365–388.
Stanley, J., and Z. Szab´o. 2000. On quantifier domain restriction. Mind & Language 15:219–261.
Stateva, P. 1999. In defense of the movement theory of superlatives. In Proceedings of ESCOL, volume 99.
236 Stephenson, T. 2007a. Judge dependence, epistemic modals, and predicates of per- sonal taste. Linguistics and Philosophy 30:487–525.
Stephenson, T. 2007b. Towards a subjective theory of meaning. Doctoral Dissertation, Ph. D. thesis, MIT.
Stojanovic, I. 2007. Talking about taste: disagreement, implicit arguments, and relative truth. Linguistics and Philosophy 30:691–706.
Syrett, K., C. Kennedy, and J. Lidz. 2010. Meaning and context in children’s under- standing of gradable adjectives. Journal of semantics 27:1–35.
Szabolcsi, A. 1986. Comparative superlatives. MIT Working papers in Linguistics 8:245–265.
Szabolcsi, A. 1997. Ways of scope taking, volume 65. Springer.
Takahashi, S. 2006. More than two quantifiers. Natural Language Semantics 14:57– 101.
Vinet, M.T. 1995. Adverbial quantifiers, negation and stress rules effects. University of Venice working papers in linguistics Vol. 5 115–139.
Von Fintel, K. 1994. Restrictions on quantifier domains. Doctoral Dissertation, Uni- versity of Massachusetts.
Von Fintel, K. 2004. Would you believe it? The king of France is back! Presup- positions and truth-value intuitions. Descriptions and Beyond. Oxford University Press, Oxford 315–341.
Von Fintel, K., and A.S. Gillies. 2008. CIA leaks. Philosophical Review 117:77–98.
Von Stechow, A. 1984. Comparing semantic theories of comparison. Journal of se- mantics 3:1–77.
Westerstahl, D. 1985. Determiners and context sets. Generalized quantifiers in natural language 1:45–71.
Williams, E.S. 1977. Discourse and logical form. Linguistic Inquiry 101–139.
Wold, D. 1995. Antecedent-contained deletion in comparative constructions.
237