Chapter 4 The Syntax of Personal Taste
-
Upload
eastanglia -
Category
Documents
-
view
1 -
download
0
Transcript of Chapter 4 The Syntax of Personal Taste
1
Chapter 4
The Syntax of Personal Taste
John Collins
1: Introduction: Predicates of personal taste
So-called predicates of personal taste (PPTs) have recently attracted a great deal of
philosophical and linguistic attention. Such predicates are exemplified below:
(1)a To ski is fun
b It is exciting to ski
c Skiing is enjoyable
d Liquorice is tasty
The truth or falsity of tokens of the sentences in (1), and innumerable other such cases,
depends upon a relevant experiencer or judge (someone who does or does not finds
something fun, exciting, etc.). In this sense, we may say that the cases in (1) express matters
of subjectivity or opinion or personal taste. To the best of my knowledge, so much is not
denied by anyone. The predicates are monadic, though, and their subjects refer to the stuff or
events that are experienced or judged without mention of the agents who are experiencing or
judging. A chief question to ask of PPTs, therefore, is how the requirement of an
experiencing or judging agent for the truth of a claim featuring a PPT can be squared with the
fact that no such agent needs to be, and most often isn’t, overtly linguistically specified in a
PPT construction. Broadly speaking, the problem has attracted two kinds of response. One
position has it that an experiencer/judge is a factor of the linguistic content of the utterances
that host PPTs (part of the literal meaning).1 Perhaps the strongest formulation of this
doctrine is that PPTs project a variable argument position (covertly or overtly) that is
mandatorily interpreted as an experiencer/judge. Such a position, note, expresses a syntactic
commitment and not merely a semantic one. As noted in the preface, nomenclature
2
hereabouts can be confusing, but let us refer to such a position as contextualism, otherwise
known by some as indexicalism when in the stronger flavour. I do not count this position as
pragmatist, for the valuation of the would-be experiencer/judge argument is intended to be
under a linguistic licence rather than being a matter of wider non-linguistic understanding.
Relativists think of the specification of a relevant agent as determining the assignment of
truth value to the PPT host, without the specified agent entering into the propositional content
of the utterances. According to this view, a PPT does not essentially involve an agent as a
matter of its linguistic identity. The arguments to follow are not designed to resolve this
dispute one way or another, but they do centrally bear upon it. My concern is wholly
linguistic: Do PPTs militate for syntactically realised variables, third-grade variable
involvement? I shall argue that they do not. As just indicated, this claim, if true, does refute a
strong formulation of the contextualist position that favours an experiencer/judge being part
of linguistic content. A contextualist is free to attempt to retreat to a position that views the
relevant variables as represented only semantically rather than syntactically, and seek a
defence of this thesis free of syntactic considerations. My interest in PPTs is restricted to their
offering a putative example of syntactic variable projection, and my present ambitions do not
go beyond establishing that the case is unsound.2 Still, it is of no little significance that, if I
am right, there is no syntactic basis for the presence of an experiencer or judge in the
linguistic content of a PPT. The best case for a variable being genuinely linguistic is precisely
that it is syntactically projected; if that move is unavailable, then the variable effect is liable
to a pragmatic explanation. So, for reasons already adduced in the previous chapters, I am
sceptical of seeking a non-syntactic semantic home for variables, for lexical content too has
an essential structural effect. I shall not repeat my arguments here. I shall broach this issue
directly in §6, but my chief concern is just to rebut the claim of third-grade variable
involvement vis-à-vis PPTs.
3
A good number of theorists, in somewhat different ways, assume or claim that syntax
realises experiencer/judge argument positions (overt and covert) for PPTs. The theorists
include Glanzberg (2007), Stephenson (2007), Stojanovic (2007), Moltmann (2010a, 2012),
Sæbø (2009), Schaffer (2011), and Pearson (2013). The considerations offered in support of
this syntactic hypothesis, however, are pretty thin. Explicit syntactic arguments are to be
found, though, especially from Shaffer (2011) and Glanzberg (2007), who offer what they
take to be persuasive, if not compelling, reasons for PPTs to project a covert
experiencer/judge argument, which may be free or bound. In this regard, they follow earlier
work of linguists exploring the syntax of evaluative predicates generally.3
Other
considerations along the same lines may also be offered, as we shall see. I aim to show that
the kind of syntactic argumentation Glanzberg and Schaffer present in fact indicates that the
predicates do not project experiencer arguments. The specification of an experiencer for the
predicates is a type of syntactic adjunction, and so optional―no covert or overt argument is
involved at all.4 Moltmann (2010a, 2012) and Pearson (2013) argue for a somewhat distinct
account, where the projected argument of the PPT is essentially bound by a higher operator. I
think this position inherits the problems that beset any position that fancies PPTs to project
arguments, and it suffers from some other problems unique to it. I shall first consider the
‘argument vs. adjunct’ question pertaining to PPTs and then look at a few other
considerations, including those from Moltmann and Pearson. Before we get to the detail, a
number of preliminary observations are in order.
2: Philosophical and linguistic problems of taste
Firstly, independent of the theoretical considerations that are taken to militate for
experiencer/judge argument projections, it is unclear if PPTs form a natural class, whether in
terms of syntactic behaviour, semantic properties, or conceptual content. First off, just what
4
predicates are apt to express matters of personal taste is subject to considerable dispute. I
shall turn to this issue presently. Yet even if we restrict ourselves to the paradigm cases of fun
and tasty, problems of classification persist. It is, for instance, obvious enough that there is no
surface syntactic condition for a predicate being a PPT. Tasty, for example, does not select
for non-finite complements (*It is tasty to eat meat) or gerund subjects (*Eating meat is
tasty). Fun, on the other hand, is happy in both environments. This difference is aligned with
a semantic difference: tasty attributes a property to an object, whereas fun attributes a
property to an event. Of course, Rollercoasters are fun is acceptable, but such constructions
appear to be elliptical for, or coercions of, something of the form Riding rollercoasters is fun.
None of this is to suggest that classing fun and tasty together is wholly erroneous. Both
predicates express a personal perspective on the relevant object or event, which does have its
syntactic correlate in the option of a to/for x phrase that may appear with PPTs. We may,
therefore, settle on a rough metaphysical condition for a predicate being a PPT:
(MPPT) P is a PPT only if the truth of a token of P(α) entails an agent who judges or
experiences α to be P.5
So much does not tell us anything definite as regards the semantic properties of PPTs, for the
content of the relevant phrases that may explicitly introduce the agent who judges or
experiences something to be so and so might not be constitutive of the content of PPTs, but
merely express additional information by way of adjunction. In effect, the question before us
is whether this perspectival feature of the predicates is syntactically realised as an argument
(as opposed to an adjunct) position, and so does unify the class of PPTs, notwithstanding
their other differences. For present purposes of argument, therefore, I shall mostly assume
that the differences between fun and tasty do not interfere pro or con with the considerations
5
militating for or against both predicates projecting an experiencer argument. I say ‘mostly’,
for as will become evident, the difference between the predicates does matter to the
applicability of some of the tests for distinguishing arguments from adjuncts.
Secondly, as just mentioned, arguments may be had about just what predicates should be
taken to express matters of opinion or subjectivity. As it is, I shall solely be concerned with
tasty and fun. This is a somewhat thin diet, but I take the choice of examples to be non-
prejudicial; indeed, the majority of discussions of PPTs are similarly narrow. If anything, the
predicates offer the best intuitive case for the presence of a covert experiencer argument. One
could broaden the class of cases to include aesthetic, moral, and even alethic predicates, but I
take there to be much less pretheoretical appeal to the thought that good, beautiful, and true
are essentially experiential or indexed to a judge. Perhaps I betray an intellectual bias (in
most moods, I am tempted to #, or even *, It is true for me), but I can scarcely imagine
anyone retreating from the claim that tasty projects an experiencer argument to the claim that
at least good and true do. At any rate, I relieve my opponents of the burden of unifying a far
wider class of predicate than our chosen examples. Besides: it is transparent how my
arguments in regard to fun and tasty generalise should my opponent be so befuddled as to
backtrack in the way just entertained.
Thirdly, the principal ground of dispute concerning PPTs in the contemporary literature
is not exactly linguistic, but more to do with propositional content. Perhaps the most useful
entering point is the notion of faultless disagreement.6 Let Judy assert (1d) and Jack assert its
negation (Liquorice isn’t tasty). On the face of it, Jack and Judy disagree; they contradict
each other. The problem is that we appear unable or unwilling (ceteris paribus) to lay blame
upon either of them; that is, we want to be able to say that Jack and Judy both assert
something true. After all, there is nothing in liquorice itself that will tell us that Jack or Judy
are mistaken, and it is hardly necessary that Jack or Judy be in a position to make such a
6
judgment about the other; perforce, a third party is not entitled to make a such a judgment
either. Thus, we appear to have a case of faultless disagreement. Of course, there are cases
and cases. Often, we might feel compelled to contradict someone’s judgement of personal
taste, perhaps on the grounds of the person being typically inattentive or lacking the relevant
experience (If you think that isn’t tasty, you should try the merlot. It is disgusting). Even so,
PPTs do seem to be peculiarly apt to give rise to conflicting intuitions, where we want to hold
that speakers contradict each other without either being mistaken.7 As indicated above, there
are two chief approaches to this puzzle: relativism and contextualism.8
The relativist takes faultless conflict to be genuine, but seeks to ameliorate the sense of
paradox. According to this position, Jack and Judy do disagree, for their attitudes are towards
the same propositional content that Liquorice is tasty: Judy asserts it, Jack denies it. Neither
need be at fault, though, for the truth-value of the one content is assessed relative to different
experiencers/judgers, and so, relative to Jack, the content is false, and relative to Judy it is
true. The point may be put as followers. If Judy were to assert Liquorice is tasty for me, and
Jack were to respond with No; liquorice is tasty for me, then we would conclude that there is
no disagreement at all, but mere confusion (at least on Jack’s part). In other words, the
utterances differ in content, so no disagreement is produced. What this appears to reveal is
that the explicit indexing of the PPT to a judge (for me) fixes a shared content for the PPT
construction itself (Liquorice is tasty) that is true just if the respective agents to which the one
content is evaluated find liquorice to be tasty (mutatis mutandis for other PPTs). Hence, there
is nothing to disagree about. Obviously enough, the difference in content between the cases
arises from the different content of the two tokens of for me, not a difference between the
tokens of the PPT itself. On the other hand, disagreement does arise in the absence of an
explicit for me phrase. If, though, we are happy to think of the PPT type to have a constant
content across tokens where for me is appended, then we should be equally happy about such
7
constancy where the phrase is not appended. Thus, tokens of Liquorice is tasty may vary in
truth value, but not content, relative to a determinate judge (Jack or Judy), whereas tokens of
Liquorice is tasty for me do vary in content, for a token’s content is fixed relative to the
speaker in the context (the value of me), which then determines a truth value for the
construction, which will be the same for, say, Jack and Judy in the above scenario where
there is just confusion, not disagreement. In the Kaplanesque terms Lasersohn (2005) favours,
we may say that for x is not part of the character of bare PPTs, and so PPTs with and without
an explicit judge/experiencer expression differ in content. The judge is a component of the
circumstances of evaluation, which, wedded to the shared content, takes a truth value.
The alternative contextualist position does not seek to save the (apparent) phenomenon of
faultless disagreement; the phenomenon is more a case of faultless misunderstanding. The
contextualist agrees with the relativist that, in our imagined scenario, neither Judy nor Jack
need be at fault, but this is not because the one content is assessed for truth-value relative to
each interlocutor; instead, the interlocutors simply judge distinct contents, or are confused
about which contents the other is judging. For example, we may depict Judy as assenting to
liquorice is tasty for her when she utters Liquorice is tasty, and Jack as denying that liquorice
is tasty for him, when he disputes with Judy. So, there is no contradiction. The
misunderstanding (the appearance of disagreement) might arise simply from one party
thinking the other party is speaking for them too, which may be revealed by, say, Judy
continuing with, Oh, I just meant that liquorice is tasty for me, not necessarily for anyone
else. So, the contextualist agrees with the relativist that no disagreement occurs where an
explicit for me is appended to both utterances. It doesn’t follow, for the contextualist, that the
content of bare occurrences of the PPT is tasty are constant, for semantically speaking, there
are no such occurrences, i.e., the explicit provision of a for x phrase simply makes explicit a
feature of the semantics of the PPT. Much of the dispute between relativism and
8
contextualism, therefore, turns on how to understand the very idea of disagreement and the
extent to which genuine disagreement may arise simply by dint of the semantic properties of
one’s utterances.
A mixed account is possible, where, say, one may be a relativist about some uses of a
PPT that give rise to ‘faultless disagreement’, but be a contextualist about other uses, where
no genuine disagreement does arise (cf., Stephenson, 2007; Lasersohn, 2008, 2011). I shall,
however, bracket such a possibility, because both accounts share a central feature that is my
concern, i.e., the determination of an experiencer/judge is essential to fixing a truth-value for
a personal taste utterance. The relativist holds that this determination occurs independently of
the fixing of propositional content, which may be invariant over differences of truth-value.
The contextualist claims that the determination of an experiencer/judge is an aspect of the
content expressed and so cannot occur independently of the fixing of content; without the
specification of a relevant individual, that liquorice is tasty is not truth-evaluable at all.
According to both positions, therefore, a relevant individual, even if generically understood,
is essential to the full assessment of personal taste utterances; the positions differ only in
regard to whether the individual enters into the linguistically licensed propositional content or
not. For my purposes, all other issues may be eschewed in order to focus on the semantic-
syntactic question of how, if at all, the experiencer/judge is linguistically represented.9
Further note that however this question gets answered does not decide the issue between the
relativist and contextualist, although it does affect the dialectical options. I am, therefore,
committed to neither position and my arguments affect the dispute only by way of knocking
out a contextualist argument, albeit, to my mind, a central one. Let me explain.
If one were to consider mere judgements of truth-value or their distribution over possible
utterances, it is difficult to distinguish relativism from contextualism, for the difference
between the positions arises only at a sub-truth-value level, as it were; that is, both parties can
9
agree about which utterances are true and which are false in exactly the same contexts.
Stojanovic (2007), for instance, argues that relativism and contextualism are notational
variants, if strictly construed as semantic claims, as opposed to meta-semantic claims about
how values of linguistic items get fixed. This claim is perhaps too strong.10
Still, to find
phenomena where the two positions diverge in their predictions is not easy.
The relativist and contextualist do disagree, it seems, on the nature of disagreement, but it
is not obvious how to settle that issue, let alone how to do so in linguistic terms. Where a
wedge can be inserted is precisely on the basis of the syntactic assumptions the contextualist
is liable or happy to make, which the relativist may demur without forfeit. That is to say,
leverage is to be found, if we focus on the PPT linguistic material free of truth-value
assessment. Effectively, the relativist says that we have a complete proposition (something
truth evaluable relative to a circumstance of evaluation) independent of the specification of
an experiencer/judge; for her part, the contextualist says that we don’t. Now assume that at
some level of syntactic organisation, all of the constituents of a sentence contribute to the
truth conditions of its tokens. Such a level is normally conceived of as logical form (LF), but
we may remain neutral on its precise identification. All that is important is that there is some
way of determining what constituents of a sentence contribute to truth conditions. So, if an
item is projected as an argument and theta-marked as an experiencer/judge, then the
interpretation of the item contributes to the truth conditions of its host structures. The right
level of syntactic organisation, therefore, may simply be the level where theta-assignment is
syntactically licensed. We now have the basis for a clear assessment of the relativist and
contextualist positions. If a PPT essentially projects an experiencer/judge argument position,
then contextualism is true, for if the position is not valued for some given utterance, then we
do not have complete truth conditions for the utterance, because truth conditions are
computed off the relevant level of organisation of linguistic material, which, ex hypothesi,
10
contains an experiencer/judge argument position. Hence it is that the strong formulation of
contextualism that expresses a syntactic commitment is of especial interest independent of its
being an instance of a general third-grade position.
I think the strong contextualist reasoning is valid, but the argument can be resisted by the
relativist or, indeed, rejected by a pusillanimous contextualist. Both parties might hold that
that truth conditions are not to be computed solely on the basis of syntactically projected
material. The relativist, without further ado, would remain unscathed by the argument. The
contextualist, for her part, may seek the putatively essential experiencer/judge constituent of a
PPT content at some level of conceptual structure not syntactically encoded. In short, the
‘relativism vs. contextualism’ debate is not just a debate about what argument positions PPTs
syntactically project.
The argument presented retains great interest, precisely because it does offer the strong-
willed contextualist a potentially decisive argument on the neutral basis of syntax: if PPTs do
project an experiencer/judge argument position, then the relativist would be obliged to deny
that it enters into the determination of truth conditions, which is absurd, or at best utterly ad
hoc. As it is, I think the evidence clearly tells against PPTs projecting the relevant argument
position. The prospects of contextualism are thus quite gloomy: a decisive argument against
relativism fails, and a semantic property with no syntactic profile must be posited. Glanzberg
(2007) and Hansen (2011) do make a good case that subjectivity in semantics generally
pertains to the satisfaction of a standard on a scalar assessment.11
This model might well
involve a notion of an implicit judge or experiencer realised as an argument, as Glanzberg
suggests, but other cases might not, such as the subjectivity of dimensionality (cf., Richard,
2004). In general, therefore, the very idea of catering for subjectivity within semantic
assessment might well be orthogonal to whether or not there are covert judge/experiencer
arguments (Lasersohn, 2008; Bylinina, 2013). As said, though, my current focus is the
11
plausibility of an experiencer/judge argument position for PPTs, not semantic subjectivity in
general.
All that said, I do not endorse relativism. Cappelen and Hawthorne (2011, p. 449) describe
the typical articulation of a relativist position as ‘sloppy, casual, obscure… confused’. I think
this assessment is unfair, certainly if we are meant to consider contextualists to be always
sharper in their formulations. My chariness over relativism, rather, is based upon the apparent
absence of a linguistic condition on where a relativist semantics does and does not apply;
unsurprisingly, there is much variation across speakers: some people tend to project personal
taste properties as objective, and whether they accept faultlessness in disputes or not is
contingent on a host of factors, but it is not the default. Relativism, in other words, appears to
be a specifically linguistic doctrine in a narrow sense, but more a claim about (typical)
judgements pertaining to a range of concepts. Of course, if the relativist core claim is that
truth turns out not to be a narrow linguistic matter, then well and good, but it remains unclear
when truth is relative, and, ex hypothesi, reflection upon language will not tell us.
In the terms of Chapter 3, borrowed from Recanati (2004, 2010), my arguments here go
against a saturational model of the role of the notion of an experiencer/judge in the semantic
evaluation of PPTs; instead, such an aspect of PPTs is pragmatically provided. Does it follow
that a ‘weatherman’-like scenario can be concocted for PPTs such that we can have a wholly
punkt reading of a PPT? I shall discuss this issue in §6. Pro tem, the short answer is ‘Yes’. If
a judge/experiencer is to be part of the lexical content of PPT adjectives, then it must have a
structural effect. On the face of it, though, no such effect is witnessed. So, while it is the case
that the truth of personal taste claims involves agency, the involvement appears not to be
linguistically licensed; instead, it is a metaphysical effect, that might well be core to our
concept of personal state notions, but need not, for all that, be linguistically encoded.
12
Fourthly, the considerations I shall be assessing for a syntactically realised covert
experiencer argument of PPTs are not treated as unassailable by either Glanzberg or Schaffer
(or anyone else, as far as I know). Glanzberg (2007, p. 11-12, n. 9) treats the claim as a
‘working hypothesis... the strongest position [of the options], and the one I would find most
congenial’. Still, the position ‘needs more investigation’. Similarly, Schaffer (2011, p. 192)
says the following:
It should be acknowledged from the start that all of these diagnostics [for covert
arguments as applied to PPTs] are defeasible, embody controversial theoretical
assumptions, and require judgments that may be contested. Such is unavoidable in
empirical inquiry. The best-case result is when all the diagnostics converge. I will be
claiming just such a best-case result for taste claims.
Such sentiments of caution properly reflect the complexity of the issues involved. My
conclusion, though, will be the exact opposite of the one Schaffer seeks: all of the diagnostics
he employs point towards PPTs not projecting an experiencer argument, and additional
considerations also point in the same direction.
3: On the putative evidence for syntactic variables
My approach will be as follows. Firstly, I shall go through four tests Schaffer (partially
endorsed by Glanzberg) presents for an experiencer/judge being a value of a syntactic
argument of PPTs. In each case, I shall show that the test does not deliver the result claimed
for it; on the contrary, the tests suggest that the relevant syntactic position is a site of
adjunction, which, as an optional projection, does not (apparently) carry information essential
to the semantic character of PPTs (more on this presently). Secondly, I shall look at some
further tests, not discussed by Schaffer, that clearly point in the same direction. Three
preliminary points are worth emphasis.
13
Firstly, I shall assume for the purposes of argument that there can be obligatory argument
positions that are mostly covertly realised and, more generally, that the broad syntactic
assumptions that admit covert material are perfectly in order. The crucial issue is solely
whether PPTs offer a case for covert variables, with all other syntactic and semantic
properties kept constant.
Secondly, I do not intend to employ a flat-footed adjunct/argument distinction. The
quandary of how (if at all) to distinguish arguments from adjuncts is highly fraught. The
present issue, however, is not really about the argument/adjunct distinction, but about the
semantic content of PPTs; the argument/adjunct distinction is being used as a probe for
content on the assumption that arguments reflect the constitutive content of predicates
whereas adjuncts tend not to do so, or at least need not do so. Thus, all parties should
recognise that the syntactic distinction between arguments and adjuncts may cross-classify
the semantic distinction between core and collateral information. As remarked above, it is not
implausible to think that certain adjuncts are obligatory (e.g., the PP complement of put).
Contrawise, perhaps some optional adjuncts encode core thematic information. As discussed
in Chapter 3 (§4.2), Neale (2007), for one, thinks of the locative content of weather reports as
realised by such a thematic adjunct. One might entertain the same idea for the
experiencer/judge aspect of PPTs. It is not clear, though, what is being intended by the notion
of a thematic adjunct. All parties agree that it is an aspect of the content of the employment of
the relevant predicates that they contain information about, as it might be, locations or
experiencers/judgers. The point of dispute is whether such information is linguistic proper
(syntactic or lexical) or more encyclopaedic or metaphysical. The question, that is to say, is
whether or not the relevant information has a structural effect on the possible syntactic
positions of the predicate. So, imagine a scenario in which a putative thematic adjunct
behaves just like a paradigmatic (optional) adjunct. In that case, describing the adjunct as
14
‘thematic’ is merely to record that we have some grounds, perhaps based on thought
experiments or a particular conception of disagreement, that the information the adjunct
records is somehow semantically essential to the predicate. This does not significantly
advance the issue, for everyone agrees that some information appears more core than other
information. The issue of contention is the explanation for such a difference, which will not
be linguistic/syntactic in the absence of a structural effect marking the adjunct out as not
merely another adjunct. Were the adjunct obligatory, some clear sense would be given to the
notion of a thematic adjunct. It is, therefore, beholden upon the defender of the claim that
weather reports or PPTs host thematic adjuncts that they identify some structural aspects of
the adjunct that mark it out as being thematic, or argument-like, rather than just another
optional adjunct. If this is not achieved, then no matter how otherwise compelling the
considerations might be that some information is core to a predicate, we do not have even
prima facie evidence that the property of x being core to y in this instance is a linguistic
matter. The moral for PPTs, then, is just this: even if one thinks that the argument/adjunct
distinction is essentially unclear, the tests for the distinction should give at least some
indication that experiencers/judgers are not expressed just as if collateral information by
paradigmatic adjuncts, for in that case there is no basis for the claim that being the subject of
an experience/judgement is a specifically linguistic property of the respective predicates
encoded in their lexical content. All that said, for the purposes of the present dialectic, the
claim that PPTs host thematic adjuncts may be eschewed, for, as explained, the very notion of
a thematic adjunct losses all linguistic content, if the relevant phrases don’t behave like
arguments at all. We may, therefore, just focus on the straight thesis that PPTs host
experiencer/judge arguments, which is the thesis entertained by the relevant parties.
Thirdly, and correlatively, regarding a phrase as an adjunct is not to claim that its
occurrence is nigh-on unconstrained, or even less constrained than an argument. The so-
15
called ‘cartography’ programme has made it at least plausible that all items of a structure find
a place within a functional hierarchy, which realises a template of how items may be
construed in relation to one another (Rizzi, 1997; Cinque, 1999, 2002; Haegeman, 2012). As
we shall see, this approach problematises some diagnostics for adjuncts insofar as one can’t
simply assume that restrictions on the permissible order of phrases entails that the phrases are
arguments, for adjuncts cannot be positioned willy-nilly either. Constraints on the occurrence
and interpretation of some phrases might not reflect their status as arguments, but indicate,
rather, that the phrases are adjuncts that have relative prominence as the SPECs of covert
functional heads. Be that as it may, cartography issues are mostly neutral in the present
dispute. My opponent claims, and I deny, that PPTs obligatorily project an experiencer
argument. The cartographic claim is that if and when items occur within a structure, they fit
into a hierarchical functional template; the claim is not that the template must always be
realised, albeit sometimes covertly. The point may be put like this. The cartography theorist
does not reject the argument/adjunct distinction, but holds that both arguments and adjuncts
occur within a functional hierarchy; the difference between the two is that arguments are
obligatorily selected by the lexical category and adjuncts are not (save for potential
exceptional cases). So, for present purposes, arguments are syntactically obligatory items that
reflect core semantic properties of the predicates, and adjuncts are non-obligatory items,
which may reflect core conceptual information, but, qua optional, do not reflect the narrow
linguistic content of the item that heads the projection.
3.1: The binding test
I shall begin with a methodological point, and then assess the crucial cases. Schaffer (2011, p.
192) notes that ‘[o]ne leading diagnostic for covert arguments is the possibility of binding
them’, and he cites Partee (1989) and Stanley (2000) as evidence for the acceptance of the
diagnostic (see discussion of quantification in Chapters 1, 2, and 3). In fact, though, binding
16
does not offer a test for syntactically realised covert arguments; indeed, Partee is explicit that
she does not think of the test in that way.12
For the test to work in the way Schaffer intends,
one needs the assumption that all binding is syntactically realised, but I know of no good
argument for such a grand claim. Schaffer (ibid., p. 193) says the following:
The standard view of binding is that it is a syntactic relation, requiring the right
syntactic environment (basically, co-indexing and c-command: Chomsky 1981). On
this view the possibility of bound readings is excellent evidence for the syntactic reality
of the argument in question… But alternative semantic views of binding have also been
proposed. For present purposes I remain neutral as between syntactic and semantic
views of binding. I only assume that bound readings require the existence of arguments
in logical form, whether or not these arguments are syntactically projected. In particular
I assume that bound readings reveal the pre-existence of the argument… In a slogan, I
assume that the quantifier reveals structure rather than creating it [sic].
I find this passage somewhat confusing. Firstly, I would wager that no-one in the field, and
certainly not Chomsky, endorses the classic GB account of binding in the full generality
Schaffer states.13
Yet even if classic GB binding is accepted, the inference Schaffer draws,
echoing Stanley (2000), is invalid. Classic binding theory pertains to elements that occur
within a syntactic projection, not to any and all binding intuitions one may have. Thus, bound
readings alone are not enough to attract a binding theoretic explanation and so do not suffice
to establish the ‘syntactic reality’ of the relevant items. For example, it might be that one has
independent reason to think that the relevant putative argument is not syntactically projected,
in which case binding theory does not apply to it as a syntactic condition. This is precisely the
position of Partee (1989) and Chomsky (1986) as regards ‘implicit arguments’. Schaffer does
concede that there are ‘semantic alternatives’ to his syntactic claim, but my present point
does not rest on the possibility of ‘alternatives’; it is, rather, that syntactic binding theory was
17
never intended as an account of any and all bound readings, so the occurrence of such a
reading is not, all by itself, evidence for a syntactic projection.14
Secondly, since Schaffer does concede ‘semantic alternatives’ and is ‘neutral’ about the
explanation of bound readings, the significance of the binding test is moot even by his own
lights. Before the binding test can be evaluated in terms of a ‘best-result’ convergence of
considerations on a syntactic explanation of the semantics of PPTs, it needs to be established
that the test even purports to reveal a (covert) syntactic position that is otherwise syntactically
coherent. Schaffer and others do not do this. Schaffer settles for the thought that if there is a
relevant bound reading, then that reveals a position in ‘logical form’ that is demonstrated by
the relevant quantifier reading, instead of created by it. The position Schaffer (2011, p. 193, n.
22) is here rejecting is Recanati’s (2002, 2007, 2010) variadic function account of implicit
object arguments (see Chapter 3, §5). Precisely what Schaffer wishes to claim in the passage,
though, is unclear. Obviously, ‘logical form’ cannot refer to the syntactic level of LF, if
positions in logical form are meant to be evidence for syntactic relations. Schaffer’s ‘logical
form’, therefore, must refer to some level of conceptual articulation that encodes truth
conditions and is somehow linguistically licensed, an aspect of literal meaning, one might say.
Such a conception, though, does not exclude a variadic account of the relevant readings,
which Recanati precisely intends to be an account of logical form in that sense. What the
account does claim, at least as Recanati presents it, is that the relevant argument positions are
not aspects of the lexical entries of the given predicates at issue, for punkt readings are
available. As things stand, therefore, the exact character of the neutral position Schaffer seeks
to adopt is unclear.
The above two points should give us serious pause in considering the binding test as even
bearing upon the issue at hand. As it is, the data Schaffer offers do even suggest a covert
experiencer argument for PPTs.
18
Consider (2):
(2) Everyone got something tasty
The unbound reading is one where everyone got the same thing (chocolate ice cream, say)
and found it tasty. The crucial bound reading is where what is tasty varies or depends upon
the range of the initial universal quantifier. So, Ma can utter (2) to her three children and
mean (and be understood as so meaning) that Mary got something tasty for her, Daisy got
something tasty for her, and Jimmy got something tasty for him, where what the three
children got may differ between them and, furthermore, the three children may disagree
between themselves about the tastiness of the thing the others got. Schaffer’s thought is that
the second reading is supported by a structure roughly corresponding to (3):
(3) (every person x)(something y)[x got y ˄ y is tasty for x]
The significant relation here is that the initial quantifier binds into a position in the scope of
tasty, rendering the value of ‘x’ an experiencer/judge of something being tasty. Without such
a position being there already, so the thought goes, such a construal would not be possible.
Assume that (2) is ambiguous in the way indicated. One might now wonder what this tells
us about tasty. Perhaps the ambiguity indicated is general and so does not turn on the
particular argument structure of tasty. To test for this, Schaffer (ibid., p. 194) presents three
other examples that he claims are ‘without bound readings’, unlike (2):
(4)a Everyone got something frozen
b Everyone got something round
c Everyone got something artificially sweetened
Schaffer offers no discussion of these cases, so I am not sure exactly what readings are meant
to be ruled out. It is obvious enough that none of the cases in (4) admit the kind of structural
paraphrase presented in (3), if for no other reason then that frozen/round/artificially
sweetened for x are semantically deviant. Yet this difference cannot be presently presupposed
19
as demonstrative of a covert argument position for tasty that is apt to be bound, for the
binding test is intended to reveal an independent phenomenon that explains that difference.
That is to say, the test is supposed to reveal a difference between (2) and (4) that is best
explained by a syntactic projection covertly holding for (2) but not (4), such that the
‘syntactic reality’ of the projection explains the experiencer reading of (2) and the
unavailability of such readings for (4). But what is the independent phenomenon? Think
again about Ma and her three children Mary, Daisy, and Jimmy. Clearly, all of the cases in (4)
support readings where the three children got different things that happened to be frozen
(Mary got a pizza, Daisy got a bag of peas, etc.), round (Mary got a ball, Daisy got a mirror,
etc.), or artificially sweetened (Mary got a juice drink, Daisy got a sorbet, etc.), just as (2)
supports the case where the children got different things that happened to be tasty. In short, as
far as I can see, there just is no quantifier dependence relation that distinguishes the cases in
(4) from (2): in both cases, the dependent readings can be captured by having the existential
take narrow scope; there is no requirement to treat the monadic-seeming PPT as dyadic. If
that is right, then (3) is, in fact, too rich a paraphrase of the relevant reading of (2), at least as
far as the binding test goes, for one can capture the scope interaction of the two DPs without
rendering tasty as a dyadic predicate. What all the cases share is a scope ambiguity between
the existential taking wide or narrow scope with respect to the initial universal quantifier. The
narrow scope readings are true of the cases where the things got are dependent on the values
of the universally bound quantifier:
(5)a (every person x)(something y)[x got y ˄ y is tasty]
b (every person x)(something y)[x got y ˄ y is frozen/round/artificially sweetened]
The natural riposte at this point is that I have missed the crucial aspect of the relevant
reading of (2) depicted in (3), viz., that the things the people got are tasty for them, not
necessarily for anyone else. (5a) does not reflect such relativity or person-dependence of
20
tastiness, which does not, indeed, hold for being frozen/round/artificially sweetened. The
riposte is exactly right, but simply begs the question. Let us assume, as Schaffer no doubt
does, that people prefer the reading of (2) indicated by (3) to the one indicated by (5a). It
remains utterly unobvious that this is a consequence of a syntactic binding relation rather than
simply a preference for relativising the tastiness of the objects the children got to the
respective children, which is perfectly consistent with (5a). In effect, I am claiming that (2)
permits two readings, even on the assumption that tasty is monadic. A weak reading
corresponding to (5a) pairs a tasty thing with each person and leaves it open for whom the
thing is tasty. A stronger reading settles for whom the things are tasty by pairing each person
with something that was tasty at least for them. The binding test, however, gives no
indication that the stronger reading is syntactically mandatory or even requires tasty to be
dyadic. Of course, if we assume that tasty is dyadic, its second argument may be bound,
which produces the relevant reading, but the binding test itself is silent on whether tasty is
dyadic. The test just tells us that we can have the scope dependence exhibited in (5a), which
is consistent with the stronger reading, of course, as well as the weak reading. Again, if we
assume that all binding readings must be syntactically licensed, rather than merely be
consistent with the syntactic relations, then (3) will, indeed, correspond to the relevant
syntactic form. The assumption here, however, is doing all of the work, not any independent
syntactic principle or phenomenon.15
Schaffer (ibid., pp. 194-5) offers the same kind of argument for fun as he offers for tasty.
Again imagine Ma and her three children. At the fair, Mary went on the rollercoaster, Daisy
went on the helter-skelter, and Jimmy went on the waltzers. On the way home, Ma utters (6):
(6) Everyone did something fun
(6) is ambiguous. One reading is where everyone did the same thing. Again, the relevant
reading is the bound one, where what the children did that was fun potentially differs between
21
them, so what was fun is dependent on the range of values of the initial universal quantifier. I
am not sure how Schaffer would wish to analyse (6), but I take (7) to be the kind of
structure/paraphrase he has in mind on the model of (3):
(7) (every person x)(some event e)[ACT(x, e) and fun(e, x)]
Here I take ‘ACT(x, e)’ to express x being an actor or participant in the event e, and ‘fun(e,
x)’ to express e being an event that was fun for x (fun is a dyadic relation between an event
and an actor in the event). We can be neutral about detail. As before, we need to ask whether
such a reading tells us anything about fun in particular, or whether the ambiguity is perfectly
general. Schaffer offers the following cases parallel to those in (4) that are supposed not to
have bound readings, and so demonstrate that the bound reading of (6) does have something
to do with the predicate fun:
(8)a Everyone did something legal
b Everyone did something sedentary
c Everyone did something environmental
I shall not rehearse the relevant reasoning that pertains to these cases as pertains to the cases
in (4). Suffice it to note that all of them support readings where the existential is in the scope
of the initial universal quantifier, and so support truth conditions under which the events that
are legal, sedentary, and environmental vary with the values of the initial quantifier. Of
course, the cases are different in that legal, sedentary, and environmental do not have dyadic
construals. The crucial point, however, is that (7) is not syntactically licensed by (6) without
further ado. As before, we have a weak reading, which leaves it open for whom the event was
fun, and a strong reading, where each person did something that was at least fun for them.
Both readings are consistent with fun being syntactically monadic. What is in dispute,
therefore, is whether the preference for the stronger reading is syntactically licensed or is an
effect of our wider conception of fun that seeks out a plausible experiencer whether such a
22
role is syntactically licensed or not. The binding test gives us no steer on the matter, unless
we simply assume that overtly monadic fun is covertly dyadic, which is the point of
contention.
There are other cases. Glanzberg (2007, p. 10) offers (9a) as a ‘possibility of binding into
the experiencer position’ and the bare (9b) would also serve:
(9)a Everyone had a fun vacation
b Everyone had fun
The additional interest of these cases is that they do not feature two quantifier phrases at the
surface and the constructions are restricted to a fairly narrow class of predicate.16
It seems to
me that the cases here are not ambiguous at all, i.e., everyone takes wide scope. As it stands,
though, this bound reading does not militate for the projection of an experiencer argument.
That is, there is nothing in the available readings of (9) that oblige us to favour (10a-b) over
(10c-d):
(10)a (every person x)(some event e)[ACT(x, e) ˄ fun-vacation(e, x)]
b (every person x)(some event e)[ACT(x, e) ˄ fun(e, x)]
c (every person x)(some event e)[ACT(x, e) ˄ fun-vacation(e)]
d (every person x)(some event e)[ACT(x, e) ˄ fun(e)]
(10c-d) clearly express at least the weak readings, which leave it open for whom the event is
considered fun. As before, the stronger reading is preferred, especially so in these cases, I
think, but it is unclear why this should be due to (10a-b) reflecting the syntactic structure. If
we accept (10c-d) as the minimal truth conditions, we are obliged to account for the
apparently disprefered weak reading that (10a-b) do apparently rule out as a matter of logical
form. I shall say something about this presently. For the moment, at least, the moral is that
binding considerations alone appear not to mandate PPTs being dyadic; minimally, they only
tell us that PPTs can consistently be construed as dyadic by being monadic predicates whose
23
sole argument position is bound within the scope of a higher quantifier that provides potential
experiencers to be semantically interpreted as such.17
What should we say about the dispreferred monadic readings? The issue turns, I think, on
the possibility of a free experiencer variable. A basic motivation for contextualism about
PPTs is that the experiencer of the tasty object or fun event is variable, it may be the speaker,
the addressee, or some other salient person or group, with the overt linguistic material kept
constant. This is supposed to be revealed by the availability of all manner of overt
experiencer arguments (…tasty/fun for Bill/me/you/everyone, etc.). If this is so, then one
should wonder if the kind of cases exemplified in (2), (6), and (9) admit free construals along
with the bound cases. Such construals are somewhat odd or dispreferred as already noted. It
is, anyway, hard to read Everyone had a fun vacation as saying that everyone had a fun
vacation for me, without the entailment that the relevant people have had a fun vacation too. I
shall return to this issue, but for the moment, consider the options. Assume that the
contextualist claims that the free construals are impossible so that the putative second
argument of tasty or fun cannot be free in the presence of a higher quantifier as in (2), (6),
and (9). The truth of this claim cannot follow from the syntax as so far imagined, for
assuming that fun and tasty project an argument independently of any higher quantifier,
which is an explicit assumption of Schaffer’s (see above), then there is no syntactic basis for
claiming that some such quantifier has to be in place to bind the otherwise free variable; after
all, the variable is precisely supposed to be free where there is no antecedent quantifier.
Furthermore, the quantification when present would not be vacuous even were it not to bind
into the would-be dyadic predicate, for the quantifier binds a position in the subject
restriction (witness (3a), (7), and (10a-b)). So, if the free construals are impossible, then it
appears not to be a syntactic effect, but a wider semantic effect, perhaps due to experiences
being subject-centred such that one cannot experience or gainsay others’ experiences (see §5).
24
Whatever the explanation might be, one who does not countenance an experiencer/judge
argument is in a stronger dialectical position than the contextualist. The contextualist appears
to be lumbered with being obliged to explain an unrealised reading that is apparently
syntactically licensed by their own assumptions. The opponent, on the other hand, has no
syntactic assumptions that license such a reading in the first place and may freely help herself
to whatever the best explanation happens to be.18
The other option for the contextualist is to admit the free construals as kosher, but
explain away their oddness in terms of bizarre truth conditions, say. Again, though, this does
not leave the contextualist in a stronger dialectical position than the denier of a syntactically
realised experiencer argument. Firstly, the opponent does not so much deny the relevant
readings, as merely claim that they are not syntactically licensed in terms of the saturation of
an argument position. So, both parties can think that the sentences may attract bizarre truth
conditions. Secondly, as it happens, I think the untoward readings are not so unacceptable,
and this, as we shall see, undermines another consideration of Schaffer and Glanzberg.
3.2: The licensing test
Schaffer (2011, pp. 195-8) and Glanzberg (2007, p. 11, n. 9) argue that the behaviour of overt
experiencer/judge PPs appended to PPTs indicates that the position is argument-like, or at
least not simply adjunctional. If this is so, then we may infer the presence of a covert
experiencer/judge position, on the assumption that argument structure is always syntactically
projected. Of course, the mere fact that we may append for/to x onto a PPT does not indicate
that the phrase is an argument as opposed to an adjunct, which we are assuming to be always
optional, as the relevant PPs patently are in the present case. Schaffer and Glanzberg appeal
to two tests in order to show that the PPs, although optional, are argument-like.
25
First consider the constraint that arguments cannot be separated from their predicate (the
item that confers a semantic role upon them) by an adjunct or other material (an adverbial or
a relative clause, say). So, to use Glanzberg’s example:
(11)a John kissed Mary on the beach
b *John kissed on the beach Mary
The point here is that kiss takes two arguments, and on the beach is a paradigmatic locative
adjunct, so (11a) is OK, because the adjunct does not intervene between the verb and its
object argument, whereas (11b) is not OK because such an intervention does occur. Schaffer
(2011, p. 197) applies the test to tasty and fun with the following examples:
(12)a Liquorice is tasty to me when sober
b *Liquorice is tasty when sober to me
c Roller coasters are fun for me when sober
d *Roller coasters are fun when sober for me
For the moment, let us agree that the judgements indicated are correct. The general claim,
however, is not supported, for some adjuncts can intervene between PPTs and their putative
argument with much more acceptable results:
(13)a Liquorice is tasty on Tuesdays for me (as opposed to Wednesdays after the dentist)
b Roller coasters are fun on the beach for me (as opposed to in town centres)
c Football is fun in the park for me (as opposed to indoors)
d Ice-cream is tasty in a cone for me (as opposed to in a dish)
I admit that placing the locative adjunct last is probably the preferred option, but none of
these cases are anywhere near as bad as (11b) and innumerable other such cases one may
readily formulate. There is, suffice it to say, great variation in the naturalness of the choice of
which adjunct goes with which predicate. The bottom line, though, is that such variation does
not pertain to (11b), which is supposed to be the paradigm of illicit intervention to which the
26
PPT cases correspond. As for (12), the confounding factor appears to be when sober, which
implies an understood individual, for the subject of the sentence (liquorice or roller coasters)
can’t be sober. By contrast, the cases in (13) feature pure locative adjuncts that do not imply
agency; consequently, they are much more acceptable. I suggest, therefore, that the oddness
of (12b) and (12d) is due to the initial interpretation of the subject of sobriety being
understood to be the speaker (or at least a class including the speaker), which information the
adjunct then duplicates. The cases in (14) are fine and are naturally understood with the
speaker (or a class including her) as the subject of sobriety:
(14)a Liquorice is tasty when sober
b Roller coasters are fun when sober
Of course, the understood subject of sobriety can be the addressee or the generic person, but
one gets essentially the same effect: a following experiencer adjunct will either conflict with
or duplicate what one initially understands to be the subject of sobriety. In sum, the
intervention test gives at best a variable judgement about the argument status of overt
experiencers, and not a result that would significantly contribute to an all-round evaluation of
the syntactic claim being the best explanation of the various phenomena. What one here
witnesses, I think, is an effect of the interaction of complex semantic and lexical factors that
determine the acceptability of predicate-adjunct combinations, which are absent from parade
cases such as those exhibited in (11).
The other constraint Schaffer (2011, p. 198) and Glanzberg (2007, p. 11, n. 9) present is
one that allows adjuncts to stack or iterate in a way arguments cannot do. Consider the
following judgements, drawn from Schaffer and Glanzberg:
(15)a *John kissed Mary Sally
b John kissed Mary on the beach under the stars at midnight
(16)a *Sushi is tasty to me to Mary to everyone
27
b *Roller coasters are fun for Ann for Ben for Claire
(15) exemplifies what we expect, where a predicate (the verb kiss, in this case) has a strict
number of arguments, and so cannot support iterated would-be arguments, but can support
iterated adjuncts, which offer additional optional information. (16) is supposed to present the
interesting case, where we find the absence of iteration of experiencers. The significance of
this reasoning is wholly vitiated by a lack of attention to other factors.
Arguments do not stack because predicates tend to have a strict number of them, or at
least a small upper limit.19
Adjuncts can stack, but it just doesn’t follow that any sequence of
adjuncts is OK or semantically equivalent to any rearrangement into another sequence. In
particular, iteration of adjuncts (PPs, adjectives, adverbials) appears to be constrained.20
For
present purposes, we need not take a definite stand on these subtle matters; it suffices to point
out that adjuncts of the same type cannot stack in a semantically unconstrained way. Consider:
(17)a John kissed Mary on the beach on the grass on the car’s roof
b John kissed Mary in Manchester in London in Birmingham
The examples in (17) have two readings. The most natural readings are the ones where there
are three distinct kissing events identifiable by their distinct locations. On these readings, the
sentences appear to be elliptical lists or conjunctions. We may wish to rule out such cases as
genuine adjunction, for the adjuncts do not modify the preceding phrases. This is significant,
for when the adjuncts are of the same type, an order of inclusion seems mandatory; when the
adjuncts are distinct, no such inclusive ordering can apply. Thus, the second available
readings are the bizarre ones, where a beach is on grass on top of a car and Birmingham
contains London as a part, which in turn contains Manchester. So, the only natural non-
bizarre readings of (17) render the cases as perhaps non-adjunctional, because of the
constraint on how iterated phrases of the same type must be interpreted. If this is right, then
28
the cases in (16) give way to a straightforward explanation that treats for/to x as nothing other
than an adjunct.
Consider (16) again. If the iterated phrases are arguments, then the sentences should be as
bad as (15a), for arguments don’t iterate at all. As it is, the cases in (16) should not be *-ed
for they are ambiguous, much like those in (17). On one marginal reading, they claim that a
list of people find sushi tasty and another list of people find roller coasters fun. This is the
elliptical conjunction reading. The interesting case is the reading where adjunction of the
same type imposes an inclusion ordering. Both sentences of (16) are rendered bizarre, much
like those of (17) are, although the bizarreness differs, of course. For (16a), the reading would
have everyone having the experience of Mary having the experience of oneself finding sushi
tasty, with oneself and Mary potentially not in fact having the specified experiences (mutatis
mutandis for (16b)).21
Such truth conditions are suitably absurd, perhaps more so than
Birmingham containing Manchester, which in turn contains London. I shall leave it to the
reader to indulge in the relevant sci-fi. The crucial point is that (16) perfectly patterns with
(17), both of which can be explained in terms of the constraints on stacked adjuncts of the
same type without appeal to experiencer arguments. There is no dispute that the adjunct for/to
x does receive an experiencer interpretation, much as on x receives a spatial interpretation and
in x receives a container interpretation. It is such interpretations that enter into the
explanations offered. None of this, though, indicates that the phrases as they occur in the
above examples are arguments, at least if we are taking (15) as the benchmark. To be sure,
(15a) is also susceptible to a list reading, although somewhat more marginal than (16) or (17).
The crucial thing about (15), though, is that it has no other reading, not even a bizarre one.22
I shall discuss below a number of other tests that distinguish arguments from adjuncts,
which point in the same direction as the tests just discussed. It seems to me that taking the
29
for/to x phrase to be an argument does not even begin to explain the relevant phenomena, let
alone offer a best explanation of them.
3.3: The Control test
The phenomenon of control is a contested topic in syntax and semantics (see Chapter 2, §3.2).
For present purposes, though, we may side-line much of the controversy, although I shall
return to some aspects of the disputes below. Pro tem, assume that a range of structures
project a covert pronominal item, which is designated ‘PRO’. In particular, PRO occurs as
the subject of non-finite clauses and can be obligatorily bound or controlled by an antecedent,
or can be free or arbitrary, in which case PRO has an indefinite/generic reading as opposed to
a contextually determined reading (see below).23
So:
(18)a Billi tried [PROi to leave] (obligatory subject control)
b Bill asked Maryi [PROi to leave] (obligatory object control)
c PRO smoking is a bad habit (arbitrary)
d PRO to swim regularly is a good exercise (arbitrary)
There are other environments in which PRO arguably occurs, but for our purposes, the above
cases suffice. Further, where control is obligatory, it appears that an antecedent controller
must be in a local position of c-command over PRO.24
Assume that control is a well-understand mechanism. It can now seem as if a control
reading entails the syntactic projection of an antecedent; after all, if there is no antecedent to
be a controller, then there is no relation of control. In (18a) and (18b), such reasoning clearly
holds, for the respective controllers are overt and one cannot read (18a), say, as Bill tried to
get someone or other to leave. In such cases, though, the controller is present anyway, and is
not posited or explained just in order to capture the control construal, i.e., Bill is required to
serve as the subject of try/ask independently of its serving as a controller into the subordinate
clause. Still, the presence of a control reading may lead one to posit a covert controller
30
precisely because control requires an antecedent, even if a potential antecedent is not overt.
The standard example of such reasoning, and the one Schaffer (2011, p. 198) cites, is the
contrast between passive and inchoative sink:
(19)a *The ship sank to collect the insurance
b The ship was sunk <by αi> [PROi to collect the insurance]
The thought is that passive sink in (19b) projects a covert agentive by-phrase that may serve
to control the PRO in the subordinate clause. Hence it is that (19b) is OK. Inchoative sink in
(19a) does not project an argument, save for the theme/patient subject (the ship), which
cannot control PRO without producing a bizarre reading. This reasoning is not sound as an
argument for the positing of an implicit (covert) agentive argument (cf., Williams, 1985,
1987; Landau, 2000; Bhatt and Pancheva, 2006; Collins, 2007).25
For the moment, though,
assume that the argument does go through. Schaffer thinks that such reasoning applies to
PPTs too and so gives us reason to think that they project an experiencer, much as passive
sink projects an agent.
Schaffer (ibid., p. 199) asks us to consider the pair in (20):
(20)a It is fun [PRO to dance]
b It is popular [PRO to dance]
He then writes:
To see that “fun” is doing something special―namely projecting an experiencer
argument―which is generating the control relation, contrast [(20a) with (20b)]…
[(20b)] simply comments on the social regard of the practise of dancing. There is no
control reading of [(20b)], presumably because “popular” does not project any
argument that could control PRO. Hence I conclude that the best explanation for the
control reading of [(20a)], and the contrast between [(20a)] and [(20b)], is that taste
predicates project syntactically real experiencer arguments.
31
Such reasoning, of course, can’t apply to tasty, which doesn’t take non-finite complements.
Schaffer (ibid., p. 199, n. 32) acknowledges this restriction, but makes nothing of it.26
Still, if
the case offered is a good one for fun, then that should be grounds to posit a syntactic
projection for fun regardless. As it is, the reasoning is awry for a number of reasons. Perhaps
most centrally, fun and popular differ independently of a control construal, so the difference
between the cases in (20) does not by itself militate for an experiencer projection for fun. I
shall get to this, but even waiving such an independent difference, Schaffer’s reasoning is
incondite.
It is simply not the case that contrasts such as (19) together with a standard
understanding of obligatory control entail that an implicit argument is syntactically projected.
The control reading might be lexically or semantically licensed without the projection of a
controller. For one thing, the agent(s) sinking the ship need not be the agent(s) collecting the
insurance. In other words, a control reading is not obligatory as it is with standard cases of
subject and object control. This is further evidenced by the subordinate clause of (19b) being
able to occur higher than the passive clause, without an effect on the content, which is
impossible with (18a) and (18b), for example:
(21)a To collect the insurance the boat was sunk
b *To leave Bill tried
c *To leave Bill asked Mary
On the standard theory of control, PRO in (21a) cannot be controlled, and the structure does
not, anyway, have an essential control reading (Collins, 2007; Lasersohn, 2009, p. 369). In
(21a) the would-be covert controller associated with the passive predicate is not in a local c-
command position relative to PRO (we shall return to this point presently). The contrast
between (19a) and (19b) remains unaffected, for the undisputed reason that ships don’t
collect insurance ((19a) would be perfectly fine were we to treat the ship as animate, as in a
32
somewhat dark children’s story). So, there is no inference to the best explanation in the offing
here: the best explanation of the contrast in (19) does not translate into an argument for a
covert experiencer controller for fun in order to explain the contrast in (20). For such an
inference to go through, it would have to be established (i) that a syntactic projection is
required to realise an ‘implicit argument’ in (19b) but not (19a) and (ii) that the apparent lack
of control in the passive cases as described is explicable in a way that doesn’t undermine the
putative syntactic projection. As things stand, one finds no good reason from the theory of
control to posit syntactic arguments that are otherwise unmotivated.
Saying so much, however, leaves the contrast in (20) unexplained. Note, though, that the
syntactically based control reading of (20a) is problematic in exactly the same way as (19b) is
in light of (21a). Thus:
(22) To dance is fun
(22) has a control reading alright, at least in the sense that whoever is having fun is the one
dancing, which distinguishes this case from the passive sink case, where the agents of the
sinking may be disjoint form the agents of the collecting. Crucially, however, it is as unclear
how PRO can be controlled by an experiencer argument in (22) as it is how PRO can be
controlled in (21a).27
If, then, (20a) and (22) are not cases of syntactic control, then what is the right account?
Perhaps the correct account is simply that PRO is arbitrary in both cases, which does not
mean that it is a variable that takes an arbitrary value, but rather that it is indefinite or generic.
So, where for one interprets PRO, we have:
(23)a It is fun [for one to dance]
b [For one to dance] is fun
It follows that the person having fun is the person dancing precisely because fun is predicated
of the dancing event involving the indefinite agent of the dancing, and no other object. This is
33
achieved without the positing of a syntactically realised experiencer, but follows from the
semantics of fun: for there to be fun, there must be an experiencer, but so much doesn’t
amount to an argument for a syntactic relation of control. What, then, of the contrast with the
popular cases?
(24)a It is popular to dance
b To dance is popular
Well, there isn’t the relevant contrast, i.e., fun and popular differ in their semantic type
independently of any experiencer feature. The cases in (24) have an arbitrary PRO
interpretation just as in (23). The difference is that popular when taking an infinitival
complement is a predicate to a type or kind of event, which does not entail that any specific
event of that type is popular; no particular event can be popular in the intended sense. The
same goes for rare, if substituted for popular. Patently, no particular event can be rare,
although a type of event might be. So, the contrast between (23) and (24) is not a control
contrast; rather, some predicates (popular, rare, common, widespread) impose a kind reading
on the event that doesn’t license an existential inference to any particular event having the
given property, and so no-one can have an experience of the property holding for an instance
of the event. In the terms of Krifka, et al. (1995, pp. 95-8), popular in the control
construction is a higher-order quantificational predicate that expresses something about the
distribution of the instances of its subject within the domain of discourse, e.g., dancing is
ranked high within the activity preferences of the relevant population.28
Other predicates,
such as fun, also have a generic reading, but do license the existential inference, and so have
the control-type reading, because they are predicates not of kinds or types, but of collections
of instances of events. Arbitrary PRO is generic in both cases, but gets either an existential or
a kind-like reading depending on the predicate. It may be noted that the notion of a
quantificational predicate was introduced by Krifka, et al. as not essentially kind-selecting
34
(Elephants with blue eyes are rare/common/widespread). For present purposes, though, all
that is important is that the predicates do not license the existential inferences, regardless of
precisely how the predicates relate to uncontroversial kind-selecting predicates, such as
extinct (cf., Cohen, 1999, pp. 92-4).29
So, what is required for Schaffer’s argument to go through, all else being equal, is a
predicate that does hold of instances of events picked out by non-finite clauses that does not,
in contrast to fun, support a control reading. There are no such predicates, though, for PRO
requires an interpretation, and if a predicate does hold over instances of the events, the
predicate will, ipso facto, hold for the interpretation of PRO, regardless of any additional
appeal to a would-be controller projected by the predicate. Consider: It is impolite to smoke.
If true, it follows, whether the sentence is construed episodically or generically, that at least
one person who smokes (the value of PRO) would be impolite, but we don’t require impolite
to project a of x argument that is valued as the person being impolite to secure this inference.
It suffices that impoliteness is a property of the event, i.e., a person’s smoking, not the person
punkt.30
In sum, the contrast between (23) and (24) is not explained by a difference in control
but by the independent difference in semantic type of the predicates.31
One needn’t accept the above proposal to see that Schaffer’s account is non-explanatory.
Even if one accepted that fun does syntactically project an experiencer argument that controls
PRO, the difference between (23) and (24) would remain to be explained, which the theory of
control does not do.
I have so far considered the ‘control test’ from a traditional perspective on PRO and
control, following Schaffer himself. It is worthwhile to pause, though, to see if alternative
accounts give more favourable results on the test.
A serious complaint against the applicability of the control test is that (22) supports the
same reading as (20a), even though the would-be controller in (22) is not in a local c-
35
command position relative to PRO. It might be, however, that such a stricture needs to be
relaxed for independent reasons. Polinsky and Potsdam (2002) present evidence for backward
control in Tsez (a Caucasus language), and further work has found evidence for it in other
languages, which this initial work suggested (Alexiadou, et al., 2011). Control in this case is,
again, between a covert item and an overt item, but the covert item is in a higher position,
allowing for a structure akin to
(25) ei began [Billi to leave],
where Bill is understand to designate the value of both thematic positions in the matrix and
subordinate clauses. If backward control is a real phenomenon, might it ameliorate or even
dissolve the problem with (22), for that problem was precisely one of the apparent controller
being lower than PRO?
One immediate problem is that backward control strongly suggests that control is in fact
a species of raising as per the so-called movement theory of control.32
The crucial point here
is that the dual predication of control structures is theorised not in terms of an additional
pronominal item (PRO), but rather in terms of a single item and its copy being assigned
distinct theta-roles, one in the launch site and one in the landing site, where the initial item is
copied. In effect, therefore, backward control is a species of forward raising, as it were. If this
is the right way of viewing the matter, then the problem of (22) remains at least as severe as it
was. Backward control, like normal obligatory control, requires an overt controller, which (22)
lacks. So, even if backward control were perfectly kosher, it would not offer a basis from
which to understand the intended analysis of (22), where both controller and controlee are
understood to be covert. This conclusion is obvious, if backward control is a species of
raising, which involves an item being spelt-out at, or overtly displaced to, a position distinct
from its first merge position.
36
All that said, the possibility of a backward control relation, however realised, offers some
succour to Schaffer as regards (22). The problem remains, of course, of how to understand
the control relation syntactically. One idea would be to think of the control relation being
established with the non-finite clause in its low position, which is retained when the clause as
a whole moves up, with the expletive subject somehow deleted:
(26)a It is fun <for αi> [PROi to dance]
b [PROi to dance] is fun <for αi> <[PROi to dance]>
Such a proposal, however, constitutes a whole new analysis of control rather than an instance
of the traditional model with a modification. Obligatory control is supposed to be established
on the basis of configuration (i.e., the presence of a controller in the right position) and the
constituent lexical items, which determine whether the structure is object or subject control.
Hence it is that (21b) and (21c) are unacceptable. On the proposal just mooted, both
conditions are disregarded, in favour of, in effect, the retention of a structural relation in a
distinct kind of structure. The mistake is encouraged by an uncritical view of indexes.
Correctly viewed, indexes are not elements of a structure that can be retained across
movement, but theoretical artefacts to indicate relations of construal that are fixed
independently of indexes, such as via the configurational and lexical conditions on control as
indicated. From this perspective, the indexes in (26) only indicate the intended readings, but
go no way to show how such readings are syntactically supported.
An alternative analysis along backward control lines might attempt to do without PRO,
and view control as a forwards or backwards movement relation of the arguments, without
any movement of the clauses themselves:
(27)a It is fun <for αi >[<αi> to dance]
b [<αi > to dance] is fun <for αi>
37
It is not obvious how to understand such relations in a wholly covert way as is here required.
The best conclusion we can reach, I think, for a backward control analysis of (22) and similar
structures is that it is presently unclear even how to formulate the would-be proposal, but the
bare idea of backward control (that the controllee might be higher than the controller) does
suggest that a syntactic analysis of (22) that supports the intended covert control reading is
not impossible. Exploring this possibility would involve ditching the traditional syntactic
assumptions that were supposed to make the control test a good indicator of covert arguments,
i.e., the presence of a local c-commanding controller.
Another alternative to traditional theories of control is to review the relation as (more or
less) semantically licensed. At least in the hands of Culicover and Jackendoff (2006), this
kind of approach is motivated by the apparent latitude there is to control structures, which
appear to lack any underlying syntactic unity. In particular, control of some kind seems
possible even without a controller:
(28)a The French agreement to raise the steel tariffs was met with outrage
b The US attempt to dominate oil supply is no longer to be doubted
Patently, the agent of the raising of the tariffs and the dominating of the oil supply is non-
arbitrary, but there is no apparent antecedent for PRO in these cases. For present purposes, I
remain perfectly neutral on how best to understand such phenomena. There are, though, two
broad options available, neither of which may be happily endorsed by one who thinks that
control signals a projected experiencer in PPTs. One option that retains the traditional
analysis of control as far as possible is to view such cases as highly ‘marked’; that is, they are
acceptable structures, but do not reflect the core machinery of syntax, being in some sense
figural, marginal for many speakers, subject to late acquisition, and so on. If that is so, then,
without further ado, one might say exactly the same thing about the apparent control
properties of fun. Fun realises a control reading, but it has no syntactic license; it is merely a
38
lexical implication of some kind, much as is witnessed in (28). Another option would be to
embrace the apparent syntactic latitude of control relations. To take that option, however, is
just to give up on using control as a test for covert syntactic projections.
I hope enough has been said to establish that employing control to test for covert syntactic
arguments is by no means straightforward and does not in any clear way contribute to a ‘best-
result’ convergence on PPTs projecting an experiencer.
3.4: The Sluicing test
Sluicing is a complex form of clausal ellipsis. On the standard analysis, it consists of the
elided (covert) clause being the TP complement of an overt wh-item in SPEC-CP, which has
moved from a lower position in the TP clause.33
(29) offers an example, where the
underlining indicate covertness:
(29) Bill kissed someone, but I can’t remember [CP who [TP Bill kissed <who>]]34
Assume, as is standard, that ellipsis is licensed under an identity condition with the
antecedent. In principle, therefore, the nature of the ellipsis will indicate the constituents of
the antecedent, even the covert ones. It is this possibility that Schaffer (2011, p. 199-201)
seeks to exploit to test for an experiencer argument. Thus, compare the cases in (30):
(30)a The cheese is tasty, but for whom?
b *The cheese is round, but for whom?
(30a) appears to be a case of sluicing, in which case it receives the (partial) analysis:
(31) The cheese is tasty <for α>, but [CP for whom [TP the cheese is tasty <for whom>]]
So, the experiencer argument must be projected in the antecedent, for otherwise it wouldn’t
occur under ellipsis, which it must do, given that a corresponding phrase (for whom) has
moved into the higher CP position. Such reasoning is no good.
As Schaffer (ibid., p. 201) readily concedes, sluicing allows for adjuncts to move into CP
position:
39
(32)a Pam ate, but [CP when [TP Pam ate <when>]]
b Pam ate, but [CP where [TP Pam ate <where>]]
c Pam ate, but [CP why [TP Pam ate <why>]]
If sluicing were a test that confirmed the presence of an experiencer/judge argument for PPTs,
it would equally confirm that every predicate projects arguments for time, place, purpose, and
manner. The test is therefore useless to distinguish arguments from adjuncts. Still, Schaffer
(op cit.) concludes:
Overall, sluicing constructions seem to diagnose the prospect of either arguments or
optional adjuncts. So the same confound recurs here from the licensing test, in that…
[(30a) may be regarded as a case of sluicing] targeting an optional adjunct. But, given
the argument above that experiencer phrases are not adjuncts, this confound is already
resolved.
The reasoning here is that since the binding and control tests are positive for the experiencer
phrase being an argument, the licensing and sluicing tests can be resolved positively too: they
test for arguments or adjuncts; since the latter disjunct does not hold, the first does. This
argument is not invalid, but it is hardly compelling, since all the tests are equivocal, at best,
and none of them serve to establish the experiencer phrase as possessing the same status as
the paradigmatic arguments occurring in the examples employed to establish that the tests are
good ones in the first place. One might equally well throw out the licensing and sluicing tests,
for it is a tautology that the experiencer phrase is either an argument or an adjunct.
Further, consider Schaffer’s (op cit.) general conclusion:
All four of the diagnostics I have invoked—binding, licensing, control, and sluicing—
have converged on the claim that taste predicates project experiencer arguments. This is
a best-case result… Anytime there is a taste predicate, there is an experiencer specified,
even if only covertly.
40
This conclusion is at best overstated. The tests do not converge at all. The licensing and
sluicing tests might as well be thrown out, the latter being entirely empty. The control test
should really be thrown out too: the crucial data submit to an independent explanation, and
the assumption of a projected experiencer/judge is not fully explanatory of the contrasts
between control-like readings and generic-type readings. As for the binding test, it does not
offer a test for a covert position, for all of the relevant readings can be explained by quantifier
interaction. The matter there reduces to whether to treat the PPTs as monadic or dyadic,
which is just the initial problem. The binding test offers a way of restating the options, but
not of resolving them. The standing of Schaffer’s conclusion considerably worsens when
other tests he neglects to discuss are assessed. We find here a convergence in the other
direction.
4: Other considerations for a syntactic variable
In this section, I shall consider four further conditions that we might apply to test whether
experiencer phrases are adjuncts or arguments. Just one of these tests is equivocal; the others
give clear results in the negative. I’ll discuss the negative ones first.
4.1: Floating
Adjuncts can float within their clause in a way arguments cannot; in particular, adjuncts can
extend maximal projections:
(33)a Billy plays football on Tuesdays
b Billy on Tuesdays plays football
c On Tuesdays, Billy plays football
d *Billy plays on Tuesdays football
Arguments can topicalise, of course (Football, Billy plays), but I assume that this is a case of
movement to a Topic projection, which is only witnessed in (33c). Experiencer phrases
patently float, even when the PPT takes a complement:
41
(34)a Liquorice is tasty for me
b Liquorice for me is tasty
c For me, liquorice is tasty
(35)a Roller coasters are fun for me to ride on
b Roller coasters for me are fun to ride on
c For me, roller coasters are fun to ride on
d Roller coasters are fun to ride on for me
I cannot readily imagine an explanation for why a would-be covert argument of a predicate
may be overtly realised upon any maximal projection within the clause that contains the
predicate. There is no mystery here at all, if the experiencer phrases are adjuncts.
If we depart from these relatively simple cases to more complex constructions, the
situation does not fundamentally change. It is widely assumed, after Rizzi (1997), that the
fronting of adjuncts and arguments are both cases of movement to the SPEC of a Top(ic)
head (potentially other positions too, such as a Foc(us) position). Asymmetries still obtain,
however. Haegeman (2012) has identified a series of such differences that may serve as
criteria for the identification of arguments and adjuncts. One key example will serve our
present purposes.35
Adjuncts can take a higher position than moved wh-items within a root clause:
(37)a Yesterday, at what time did you meet Bill?
b At what time yesterday did you meet Bill?
Here, in both cases, yesterday and at what time have been fronted. Arguments can be fronted
(topicalised) too, of course, but not as in (37):
(38)a *Kant at what time did you read?
b *At what time Kant did you read?
42
It is worth noting too that (37) exhibits the fact that adjuncts can stack in the left periphery in
the way arguments cannot:
(39)a *Flowers to Mary Bill gave
b *To Mary flowers Bill gave36
In order to apply the test to PPTs, consider (40):
(40)a For Bill, when is ice cream tasty/football fun?
b When for Bill is ice cream tasty/football fun?
(40) patterns with (37), suggesting that the experiencer of the PPT is an adjunct. At any rate,
(38) and (39) are severely degraded in a way (40) patently isn’t.
As indicated, Haegeman (2012) has a battery of diagnostics for the differentiation of
arguments and adjuncts in the left periphery. I leave to the reader to see if any of the further
tests pattern PPT experiencers with arguments. They do not, insofar as I can reconstruct the
tests to apply to PPTs. It might be, of course, that this is simply a result of experiencer
phrases being expressed by adjuncts within PPT predicates. Yet that is exactly the point: we
have no syntactic reason to think of experiencer phrases of PPTs as arguments. The position
that, somehow, experiencers are arguments albeit in adjunct form is not a clear hypothesis as
things stand.
4.2: Ellipsis
In Chapter 3 (§4.2), an ellipsis test for arguments was offered, where a VP including its
(internal) arguments may undergo ellipsis, but a VP without its (internal) arguments may not.
Adjuncts, on the other hand, do not need to be elided along with the verb; they can be left as
remnants or substituted within the VP containing the ellipsis:
(41)a Billy drinks wine on Wednesdays, but he doesn’t on Thursdays
b Billy drinks wine on Wednesdays, and Sally does on Thursdays too.
c *Billy drinks wine on Wednesdays, but he doesn’t beer on Thursdays37
43
d *Billy drinks wine on Wednesdays, and Sally does wine on Thursdays too38
The data here clearly tell us that wine is an argument of drink and on Wednesdays/Thursdays
are adjuncts. So, if an experiencer phrase is an adjunct, we predict that it may be left as a
remnant (i.e., only the VP is elided) or substituted (i.e., the verb or adjective is elided and the
elliptical clause is supplied with a distinct adjunct) without producing unacceptability;
otherwise, the phrase appears to be an argument. It is not straightforward to apply this test to
PPTs because they are adjectives, and it is difficult to separate VP-ellipsis proper from
gapping or pseudogapping, where an argument or adjunct remnant can be left, rendering the
construction neutral as a test case. We can, however, consider the ellipsis of seem (and other
experiencer verbs) that take a PPT as a complement including the preposition for x. As
predicted, the for/to x phrase can be left as a remnant or substituted, indicating that it is an
adjunct:
(42)a Liquorice seems tasty for Mary, but aniseed doesn’t for her
b Liquorice seems tasty for Mary and does so for Sally too
c Roller coasters seem fun for Mary, but the waltzers don’t for her
d Roller coasters seem fun for Mary and do so for Sally too
For x behaves here as an adjunct. 39
Something else merits notice. The elision of the experiencer phrase in an antecedent
produces a certain semantic effect, almost as if a contradiction is being made:
(43)a #Liquorice is tasty, but isn’t for Sally
b #Roller coasters are fun to ride on, but aren’t for Sally
It is not clear what, if anything, one should conclude from this. What is suggested, though, is
that if there is an experiencer argument, its value is not default-set to the speaker or addressee
or some salient person, for if it were, there would be no anomalous effect with (43). It
appears, therefore, as if the understood experiencer is generic or indefinite, which is predicted
44
by the account offered above that the PPTs are simple predicates of the events semantically
implying indefinite or generic participants of events.
A further bit of evidence in favour of this reading is a similarity between PPTs and
generics in relation to specified exceptions.
(44)a Mosquitos carry malaria
b Tigers have four legs
c Paperbacks are cheap
The cases in (44) are construed as generic predications in part because they appear to accept
exceptions; that is, all may remain true, even if not every value of the subject satisfies the
predicate. Such is why generics are thought not to be equivalent to universals. The
specification of an exception produces a certain oddity:
(45)a #Mosquitos carry malaria, but this one doesn’t
b #Tigers have four legs, but this one hasn’t
c #Paperbacks are cheap, but this one isn’t
Put otherwise, the actual specification of an exception to a generic appears to invite a
qualification or retraction of the original generic claim, even if the claim by itself is not
construed as a universal.40
This effect patterns PPTs with generic predicates, and so
dissociates PPTs with predicates whose argument values are fixed contextually to the speaker,
addressee, or an otherwise salient individual (see below).
4.3: Movement
As discussed in Chapter 3 (§4.2), an insight of the GB framework is that adjuncts are more
restricted than arguments in their movement (more ‘minimal’ in the distance they may move),
liable to contravene both island constraints (for our purposes, subjancency), and the ECP
(Empty Category Principle), if construed long. So, the distinction between the availability of
45
a long and a short construal might offer a test for distinguishing arguments from adjuncts.
Glanzberg (2007, p. 11, n. 9) suggests such an extraction test for PPTs. Consider the
following pair:
(46)a ?Who do you wonder whether John kissed?
b *Where do you wonder whether John kissed Mary?
Glanzberg’s judgements (as indicated) are as follows. (46a) is marginal, but acceptable to
some, with who being construed as questioning the object position of kiss. Who, therefore, is
an argument moved from the theme position of kiss.41
(46b) is unacceptable, with where
being moved from an adjunct position in the complement, for the object position of kiss is
occupied, so can’t be the launch site of where.42
The basic thought, therefore, is that adjuncts
and arguments differ in terms of their movement potential. As we shall see, the detail matters,
though. Applying the test to an overt experiencer, we have the following (again, Glanzberg’s
judgements are indicated):
(47)a ??Who/to whom do you wonder whether sushi is tasty (to)?43
b ??Who/for whom did Mary wonder whether the ride was fun (for)?
Glanzberg’s thought is that if the higher wh-items are adjuncts, then the pair in (47) should be
unacceptable to the same degree as (46b), for in all three cases, ex hypothesi, there are no
unoccupied argument positions from which the wh-items might have moved and so they are
obliged to have a long construal modifying the complement of wonder, which is ruled out in
the case of (46b). If, on the other hand, who is an argument in (47), then the pair should be at
worst marginal along the lines of (46a), with who being moved from the experiencer position
of tasty/fun, much as who is moved from the theme object position of kiss in (46a). As it is,
Glanzberg (op cit.) notes:
Some informants find cases like these [(47)] outright unacceptable…, while some find
them marginal. The test is difficult to apply in practice, as it calls for differential
46
judgements between argument and adjunct extractions, but predicts that both should be
at least somewhat degraded. And of course, there may be any number of factors
involved beyond argument/adjunct asymmetry in these cases.
The trouble to which Glanzberg’s is here alluding is that the constructions presented exhibit
‘weak island’ effects brought about by wonder selecting an interrogative complement headed
by a wh-item. The difficulty with making sense of the significance of (47), however, is worse
than that, I think, for Glanzberg has mispresented the data.
The island effect is clear in the case of (46b), which readily has a reading, albeit an odd
one, where it is the location of the wondering that is being asked for, rather than the location
of the kissing. The former reading doesn’t involve an island violation, for the adjunct is
moved from a position adjoined to wonder outside of the complement. Pace Glanzberg,
therefore, the relevant datum here is not that (46b) is unacceptable, but that it does not
possess a long construal where it is the location of the kissing that is being asked for. The *
for (46b) is undeserved and irrelevant, anyway. The real contrast is that a long construal in
(46a) is more acceptable than in (46b), and it is such a difference that identifies the adjunct in
(46b) by way of its acceptable short construal. I think, therefore, that the kind of data
appealed to in order to motivate minimality or ECP violations holding for adjuncts but not
(object) arguments is clear enough despite the ‘differential judgements’ between not wholly
acceptable constructions: (46) offers a case in point. So, Glanzberg is wrong: it is not so
much that (47) is confusing because it probes for graded judgements between unaccepatable
sentences, for (46) does that too, but that (47) is more confusing than (46), even though both
pairs are supposed to exhibit the same effect.
Controlling for the island effect can be achieved by simply substituting a declarative
complement with a corresponding change of verb for the interrogative complement, say,
replacing wonder whether with think. Making such a change, however, erases all differences
47
between the cases, as one removes the ECP-effect too that pertains to the adjuncts. (46a)
becomes perfectly OK, for its only problem was the island effect. (46b), per the
reformulation, strikes me as ambiguous:
(48) Where did you think John kissed Mary?
The location of the thinking or the kissing can be here being questioned given the removal of
the island. That indicates, of course, that where is an adjunct, which we knew, anyway. The
suitable reformulations of (47) tell us nothing, I fear, for arguments and adjuncts can move
equally under such conditions. To be sure, the movement creates no ambiguity here but that
is for independent reasons (prepositional stranding indicates the launch site of the movement
and verbs like think don’t support the intended experiencer modification). One might seek to
control the cases further by removing the stranding and using a verb that does support an
experiencer modification:
(49)a To whom does it seem that sushi is tasty?
b For whom does it seem that the ride is fun?
The question here is whether both cases support readings where the experiencer of the
seeming or the experiencer of the tasty sushi/fun ride is being questioned. If so, then the
moved item behaves like an adjunct. I think the pair in (49) do support such readings,
although, of course, the natural reading for both is one where there is one experiencer for
whom it only seems that sushi is tasty or the ride is fun, i.e., the movement would not be from
the would-be argument position of tasty/fun, but from an adjunct position on the verb seem,
with the complement providing the content of the experience. Still, imagine a parent
anxiously observing their toddler on a ride, trying to judge whether the ride is fun for their
precious offspring. In such a scenario, one could use (49b) and be coherently answered with
‘the parent’ or ‘the toddler’ (I’ll leave it to the reader to formulate the relevant scenario for
(49a)). So much indicates that the moved phrases are adjuncts, even where we take the
48
movement as being from the complement predicate, precisely because such a launch site for
the phrase is not necessary, which it would be, were it an argument position.
In sum, then, the right diagnosis of the confusion engendered by (47) is that the moved
wh-items are adjuncts, but they lack a short construal, for wonder does not support the
relevant modification. Such is why they are pretty marginal at best. If we select a matrix verb
that does support the relevant modification and so supports a short construal, then the wh-
items can be happily read as adjuncts, with the long construal being blocked by ECP or
whatever generalisation holds in this area. So, once we cut away various confounds, it seems
to me that the extraction test, if pointing any way, points against PPTs possessing covert
arguments.
4.4: Crossover effects
Stanley (2000) appeals to crossover effects as a diagnostic for overt items. Such effects are
when a joint construal of a moved item (wh or quantifier) and a pronoun it moves over
(hence, crossover) are excluded (the effect is ‘weak’ if the pronoun does not c-command the
launch site, and ‘strong’ if it does) (see Chapter 5, §4.2 for further discussion). Lasersohn
(2005, p. 681) offers the following cases to test for the effect, where a wh-item crosses over
the position of a putative experiencer argument (parallel cases can easily be constructed for
tasty):
(50)a Who was upset that the ride wasn’t fun?
b Whom did the fact that the ride wasn’t fun upset?
(51)a Who was upset that the ride wasn’t fun for him?
b ? Whom did the fact that the ride wasn’t fun for him upset?
Lasersohn argues as follows. There is no crossover in the a-cases, so their acceptability is OK
for everyone. (50b) is equally predicted to be OK, if we hypothesise fun not to project an
argument, for even though the wh-item launches from the object position of upset, it does not,
49
ex hypothesi, crossover an item jointly construed with the wh-item. So, the acceptability of
(50b) would suggest that there is no crossover effect. If we hypothesise that fun does project
an argument, then we should predict at least an effect of the kind witnessed in (51b), where
the experiencer is overt, holding for (50b) too.
Lasersohn (2008, p. 327) offers a further crossover reading involving quantifiers rather
than wh-items:
(52)a A tasty dish awaited every diner
b (every diner x)(a dish y ˄ tasty(y, x)) [awaited(y, x)]
c [TP every dineri [TP a tasty dish xi [VP awaited <every diner>i]]]
The wide-scope reading of every diner is clearly available, as depicted in (52b). If, however,
we assume that tasty projects a covert argument x, then the reading is a (weak) crossover
violation, with the quantifier object moving over the non-c-commanding variable, but the
raised operator, the variable, and the launch site copy all share an index. The violation
disappears, of course, if we simply eliminate the variable within the syntax.
I share Lasersohn’s judgements on these cases. It seems to me, though, that a defender of
covert experiencer/judge arguments of PPTs shouldn’t appeal to crossover effects. The status
and explanation of crossover effects is somewhat fraught, and, as classically understood,
pertains to linearity effects, which would be impossible to detect for covert cases.
Furthermore, the supposedly illicit reading of (51b) is not so bad, or as bad as (50b), whereas
if we were appealing to the overt case to tell us about the covert case, we should want a
stronger indicator of illicitness. Indeed, if we move away from wh-cases to quantifiers, it is
hard to detect any effect:
(53)a Every boy was upset that the ride wasn’t fun
b The fact that the ride wasn’t fun upset every boy
(54)a Every boy was upset that the ride wasn’t fun for him/them
50
b The fact that the ride wasn’t fun for him/them upset every boy
In both of the b-cases, one can construe the object quantifier phrase to take wide scope over
the position of the experiencer/judge (would-be covert in (53b), overt in (54b)), such that
every boy didn’t have fun and was upset.44
So, if there were some definite crossover effect for PPTs, then it would be good news for
the advocate of covert experiencer/judge arguments, but the test in general is somewhat
unreliable, being marginal in the wh-case and non-existent in the quantifier case. The absence
of the effect is of no great import.
5: Still other considerations: clausal selection and generic operator binding
The moral of the forgoing considerations is that there is no good reason for thinking that
PPTs syntactically project an experiencer/judge argument. The provision of such a participant
in events or states PPTs describe appears to be delivered by an optional syntactic adjunct.
Still, there may be reasons independent of general diagnostics for argument-hood to think that
PPTs essentially carry an experiencer/judge projection. I shall consider two arguments to
such an effect. The first bears on the selection properties of certain verbs. The second
concerns generic binding.45
(a) Verb selection.46
Verbs that take clausal complements select for certain clause types.
Know takes declarative and interrogative complements, both finite and non-finite. Believe and
other related ‘truth-aiming’ verbs are far more restrictive, only taking declarative finite
clauses. Wonder and other interrogative verbs take finite and non-finite interrogative clauses.
Other verbs, such as find, think, reckon, and consider take finite and non-finite declarative
clauses as well as small clauses, as in Bill found/considered Sam stupid. Let’s say that this
last kind of verb, the ones that select small clauses, express opinion or at least support such a
construal.47
Given such a categorisation, one may reason as follows. If verbs select for kinds
51
of complements, and there are opinion-expressing verbs, then we may predict that the clauses
the verbs select are expressions of opinion or personal taste. If such verbs select PPT-
featuring clauses, then the prediction appears confirmed, and a natural explanation of the
basis of the opinion-character of the clause is that it features an experiencer/judge projection
(Stephenson, 2007; Lasersohn, 2005, 2009; Sæbø, 2009; Moltmann, 2010a, 2012; and
Pearson, 2013). The crucial issue, of course, is whether the relevant verbs do select for
expressions of opinion, which is somehow linguistically encoded independent of the verb.
Some data appear to indicate this. Just consider find, which offers perhaps the best case (if the
argument doesn’t hold for find, we should hardly expect it to hold for think or consider,
where analogous data are much less striking):
(55)a Bill finds ice cream tasty/rollercoasters fun
b ?#Bill finds Sam tall/10st/a man48
The thought here is that the difference between such cases, and so what explains the anomaly,
is that find only takes clauses that express matters of opinion. A person’s dimensions or
gender are not matters of opinion, whereas ice cream being tasty or rollercoasters being fun is
mere opinion. Thus, it seems as if some clauses, qua selected by certain verbs, are inherently
expressions of opinion, which would be explained were PPTs that form such clauses to
project an experiencer. I think this argument is no good at all.49
I shall first show that the data
do not suggest that there is any subjective or opinion-expressing clause. A more modest
explanation is ready to hand. Furthermore, the idea is theoretically incoherent, at least as
presented above. This will lead to a consideration of the account of Moltmann (2010a, 2012)
and Pearson (2013), which has the theoretical resources to overcome the problem, but the
solution is problematic for independent reasons.
Firstly, the deviance of (55b) is much less severe than other cases of clause selection
violation:
52
(56)a *Bill knows Sam stupid
b *Sam wonders that the car is fixed
c *Sally finds if ice cream is tasty
So much tells us that clause selection is a syntactic condition, i.e., the clauses selected are not
merely constrained to be the kind of clause that can express what is known, wondered, or
found, but that this condition is encoded in the head of the clause as expressed by the
functional C projection, or at least some definite functional projection in the left periphery of
the clause. When we reflect on the kind of cases in (55b), it is relatively straightforward to
imagine contexts where the deviance of the cases is ameliorated or just disappears. All we
need do is think of a scenario where people are employing different standards or devices of
measurement, and here we can find a kind of faultless disagreement or at least justified
difference of opinion pretty much about the application of any predicate above and beyond
tasty or fun. What this tells us is that it is the relevant verbs that impose a ‘matter of opinion’
construal on the clauses; the clauses themselves are neutral. This is further corroborated by
the context-sensitivity of the construal of certain verbs in relation to their objects. Pederson
(2013) offers the following case:
(57)a #Bill found that Sam won the race
b Bill found that Obama won the debate
The point here is that winning a race is not a mere matter of opinion in the way winning a
debate plausibly is. On the face of it, though, there are no features that distinguish the pair of
embedded clauses in (57), either by way of the interpretation of the lexical items (win is not
ambiguous or relevantly polysemous) or a difference in functional/clausal projection.
If the above reasoning is anywhere near right, the argument we are entertaining is back-
to-front. The behaviour of find and related verbs in no way show that clauses come marked as
expressions of opinion in the way other clauses come marked as declarative or interrogative.
53
Rather, it is the constural of the verbs that indicate that the embedded clause is a matter of
opinion. This is confirmed by the fact that more or less any clause of the right syntactic kind
may be taken by the verbs. The resulting oddity of many cases is simply due to the oddity
involved in the expression of an opinion about a matter that is not normally up for judgement
or is otherwise definite. We can easily finagle cases, though, to make them much less
degraded. The situation is very different where the selection is syntactic; no amount of
contextual scene setting or substitution of predicates can rescue (56) from flat-out
unacceptability.
A complicating factor here is the polysemy of find. It has a stative construal and an
achievement construal akin to discover; mixing the two perhaps leads to anomaly.50
Clearly,
given the right choice of a complement featuring a PPT, the stative construal of find is
entailed, for one cannot find the liquorice to be tasty and not take the liquorice to be tasty for
oneself—that is just what the stative construal demands. On the achievement construal, find
is also factive, but does not entail any kind of experiential state. None of this, though,
suggests that find under either construal selects a certain kind of subjective complement; on
the contrary, the polysemy of find is resolved as a function of the complement, i.e., whether
or not the complement features a predicate that entails experiential states. So, the selection
features of find do not so much as suggest a syntactically realised experiencer projection. The
complements find selects are syntactically neutral, but the verb does acquire an experiential
construal dependent on its complement, rather than it syntactically selecting such a
complement. This reading of the situation gains considerable support when we consider the
theoretical basis of selection.
The argument we are considering holds that certain clauses are typed as expressions of
opinion or experience, such is why only certain verbs select them. Contra the evidence,
assume, that the facts were to support this claim. It would remain to be discovered how such
54
putative clauses are typed as relevantly subjective. The projection of an experiencer argument
cannot possibly determine such a type, for clauses are not typed on the basis of their
argument positions. Equally, therefore, even if the relevant verbs did only select opinions,
whatever they might be, the case for experiencer argument projections would not be
corroborated, for no such projection could identify a clause as subjective. Clauses are
identified as of their kind, and so selected or not by clause-taking items, in virtue of their
head, which is a complementiser (covert or overt), either occupying C position (or its SPEC)
just above the TP or the highest position in the left periphery. Arguments cannot play such a
role. The bottom line here is that even if we might semantically type a predicate in terms of
its holding or not being a matter of opinion, the clause that hosts such a predicate cannot be
typed as an opinion clause in a way that will determine its selection, at least not if we are
imagining such a determination to be akin to any known syntactic or semantic mechanism.
Such selection properties as we find, therefore, hardly militate for an experiencer projection.
As things stand, then, the argument under consideration is empirically mistaken and
theoretically incoherent. There might, however, be other considerations that weigh in favour
of the argument.
(b) Generic binding. Moltmann (2006, 2010a, 2010b, 2012) takes arbitrary PRO to express
‘first-person-based genericity’, which means that, like one, arbitrary PRO is interpreted as
being generically quantified over such that it designates everyone (admitting exceptions)
relevantly similar to the speaker, minimally, all conscious persons. The relevance of this
account to PPTs is twofold (cf. Pearson, 2013). Firstly, Moltmann (2010a, pp. 213-4; 2012,
pp. 170-2) offers an argument based upon her account that argues in favour of an experiencer
argument for PPTs. Secondly, the operator constitutive of Moltmann’s and Pearson’s analysis
could perhaps serve as the head of a subjective/opinion clause. Let’s consider the argument
first and then assess the position of the head.
55
The argument will be familiar from the above discussion of control phenomena. Consider
the case in (58):
(58) It is fun to play football
Moltmann notes that the construction expresses a covariation between whoever has fun and
whoever plays football. She proposes that this condition can be established as follows:
(59) [CP Genx [It is fun x [PROx to play football]]].
‘Genx’ is a generic quantifier that binds the experiencer position of fun and the PRO position
of the non-finite clause (see Krifka, et al. 1995). In effect, for Moltmann, arbitrary PRO is
always bound by such an operator in virtue of its equivalence to one, and it is precisely the
presence of the operator that explains the covariation reading of (58), i.e., the one operator
binds the variables in the matrix and subordinate clauses. If all of this is right, then we also
appear to get a nice consequence: the subjective/opinion clause might be headed by the kind
of operator Moltmann and Pearson posit. Notwithstanding the attractiveness of this thesis, it
is problematic.
First off, all by itself, the argument from covariation is not a good reason to posit an
experiencer argument projection. It is persuasive only if one has already accepted the
presence of the higher operator, whose occurrence explains the covariation in terms of
operator scope. As explained above, though, the covariation can be accounted for without
appeal to joint binding, i.e., without the positing of an argument position for the matrix PPT.
It suffices if we merely let the PPT be a predicate of the non-finite clause. For the fun to be
disjoint from the playing would, therefore, involve another predicate such that it is fun to be
related to the playing of football by that predicate. As previously discussed, predication to the
non-finite clause does not by itself entail covariation, but the problem cases are independently
explicable. For instance, It is popular to dance does not have a covariation reading. The
explanation for that, however, follows from what it takes to be popular, which blocks the
56
entailment to a token of the type (dancing in this case) being popular. That explanation does
not require the positing of an extra argument for PPTs. Moreover, there is no independent
syntactic reason to posit a Gen operator (more on this below). Since, in other words, we are
not obliged to posit the operator and can explain the covariation phenomenon without it, that
we can explain covariation phenomenon with the operator plus the experiencer argument
hardly constitutes a good reason for positing the experiencer argumenta case of inference
to the worst explanation.
Let us assume, however, that there is a syntactically realised Gen operator as the head of
CP. Could this operator serve as the head that identifies clauses as expressions of opinion? In
constructions such as (58), where there is an embedded arbitrary PRO, the proposal seems
plausible. At least initially, the operator was posited to explain features of PRO tout court,
not PPTs. So, while the operator might explain features of PPTs in interaction with PRO, it
looks to be unavailable in the absence of PRO. Perhaps the proposal could be extended,
however, so that any occurrence of a PPT projects an argument bound by a higher operator
that gives a subjective generic kind of reading to the experiencer similar to that which
arbitrary PRO enjoys (cf., Pearson, 2013). There are problems with such a proposal.
Firstly, it is far from obvious where the Gen operator is supposed to sit in any of these
accounts. Moltmann and Pearson assume that it is a C head. That can’t be right, however, for
one can embed the relevant clauses under verbs with selection requirements that would
compete with the putative operator:
(60) Bill knows [that/Gen ice cream is tasty/it is fun to play football]
Here, Gen competes with that to be head of the clause. We might assume, then, as perhaps
should be assumed anyway, that there is no unitary CP projection, but a structured left
periphery that may host a range of functional elements. For example, a Topic projection also
competes with that, if CP is a unitary whole; it would be no cause for concern, if Gen were as
57
disreputable as Topic (Rizzi, 1997; Haegeman, 2012). Questions now arise concerning the
position of the operator in the left hierarchy and why it is essentially covert. However these
questions might be answered, it is clear that the assumption that the Gen operator is a C head
is mistaken.
Empirical problems also arise. The hypothesis that PPTs project first-person based
generically bound arguments makes good sense of the semantics when no other scope-taking
element is present. When some other such items are present, it is unclear how to understand
the hypothesis. Consider:
(61)a Everyone finds ice cream tasty
b (Every person x)[x finds ice cream tasty for x]
Here, it appears as if the quantifier, as displayed in the reading provided, binds the
experiencer position. Of course, I think the additional adjunct is not required, but my
opponent should hardly think so, it being their very thesis that PPTs project experiencer
arguments. The problem now, however, is that if experiencer positions are essentially
generically bound, then they can’t be universally bound too, as they appear to be in (61).
Obviously, the problem generalises to all bound readings that occur with PPTs.
I do not think the problems I have raised for the Moltmann-style position are decisive.
The position has resources to handle many subtle phenomena, but I do not see how it can
militate for experiencer projections, even if it gives a compelling reading of arbitrary PRO.
Again, we have found no reason to think that selection properties support experiencer
projections of PPTs.
6: Lexical content and personal taste adjectives
If we assume that all of the above is correct, then we appear to have a quandary. On the one
hand, the truth of a personal taste claim depends upon there being an experiencer/judge for
whom the predicate holds of the subject; on the other hand, no such agent is mandatorily
58
represented in the syntax of a PPT. Moreover, features of lexical content must have a
structural effect; otherwise, they are not linguistic features proper, aspects of linguistic
competence as such. This is no paradox, of course. The upshot is that the agency involved in
PPTs is not a linguistic matter, even if it is truth-determining. In this regard, PPTs are parallel
to weather reports. The conclusion we reached about It’s raining in Chapter 3 is that, plainly
enough, tokens of the sentence cannot be true unless it is raining somewhere or other―it
can’t be raining, yet raining nowhere. Still, the locative aspect of the meteorological predicate
is not linguistically encoded, for no locative item is structurally mandatory or otherwise has a
systematic effect. Semantically speaking, It’s raining does not entail a location, even if,
metaphysically, rain must have a location. The possibility of ‘weatherman’-like scenarios
exhibits the effect. One can imagine scenarios where It’s raining as uttered is true, even
though no location at all has been ruled-out as truth-conditionally irrelevant; hence, the report
is not about any particular location to the exclusion of some other location, notwithstanding
that there must be some location where it is raining, if the token is to be true (see discussion
of the ‘intergalactic weatherman’ in Chapter 3, §4.1). Semantically punkt readings are
available, but are always overridden, as it were, by the metaphysics of weather.51
Engineering a weatherman-like scenario for PPTs is as straightforward as it is for
weather reports. In Chapter 3 (§4.3), I (oxymoronically) described meteorological verbs as
agentless unergatives. In terms of argument structure, therefore, raining is not predicated of
anything; it simply picks out an event including rain. A punkt reading, therefore, should be
available from the bare syntax, albeit metaphysically overridden. PPTs are different. A PPT
does have an argument, viz., whatever is funny or tasty. The predication of the property to the
object or event, however, is not structurally mediated by a judge or experiencer; rather, it is
our understanding of what it would take for the property to hold that involves such an agent,
but no such understanding need be lexically or syntactically encoded. Imagine that humans
59
have lost their taste of liquorice, or, at any rate, can’t figure out if it is the liquorice that has
changed or their own taste. Fortunately, every human has implanted in their brains a device
that sends out a signal should some pleasure threshold be reached upon the ingesting of
liquorice. A chef monitors the signals, hoping that the day arrives when liquorice will be tasty
again. One bright morning, the console flashes and the chef exclaims, ‘Liquorice is tasty!’.
As with the weatherman scenario, the chef does not know who finds liquorice tasty and no-
one is excluded as truth-conditionally irrelevant. So, the PPT features in a true claim, let’s say,
but the putative experiencer/judge variable is unsatisfied. As with weather reports, the punkt
reading is overridden because, of course, there must be someone or other for whom the
liquorice is tasty.52
The crucial point, however, is no identification of such a person need be
made in order to value a would-be variable of the PPT. The chef receives evidence that
liquorice is tasty without any evidence about for whom it is tasty. Furthermore, as with the
weatherman scenario, it will not do to presume that a value for the variable is provided
implicitly. This move is unsound in the weatherman case, for no location is in fact truth-
conditionally irrelevant; so there is no location, no matter how expansive, that can value the
supposed locative variable of the meteorological predicate. In the present case, likewise, no
human is truth-conditionally excluded as an experciencer/judge that liquorice is tasty. The
chef merely has evidence that liquorice is tasty, and so claims that it is. He does not claim
that it is tasty just for the person whose brain-implanted device sent the relevant signal,
whoever that might be; after all, the chef’s claim would remain true, and be treated as true, if
the relevant signal was in fact a false alarm, but another person does in fact find liquorice
tasty.
In short, in both cases, the claims are perfectly general, and since there is no linguistic
basis for the posting of an argument position of the relevant respective kind, the generality
60
arises from a metaphysical overriding of a semantically punkt structure. At any rate, such a
position makes best sense of the linguistic phenomena.
7: Conclusion
PPTs present a range of philosophical problems beyond the question of whether or not they
select experiencer/judge arguments. If the forgoing considerations are on the right lines, then
they do not select such arguments. The core syntax of a PPT is much like one would expect
from the ‘surface’ properties: the predicate forms an adjectival phrase, which may be
extended by the adjunction of prepositional phrases such as for me. With a bit more
elaboration, one may think of the matrix subject being first merged with the PPT to form a
small clause, and then moved up to the Tense edge. So, in the simplest case, we have a
structure approximating the following:
(62)a [TP Liquorice is [SC/AP <liquorice> tasty]
b [TP Liquorice is [SC/AP [SC/AP <liquorice> tasty] [PP for me]]]
PPTs occur in other environments, of course, but I hope the reasoning above shows that no
such context requires the position for an experiencer/judge argument. That suffices for my
negative purposes of showing that the predicates do not indicate the occurrence of syntactic
variables.
Notes
1 For the most part, I shall simply refer to an ‘experiencer’, but will mean an agent who finds
something to be such and such on the basis of their experience. Thus, I do not mean to
assimilate PPTs to experiencer verbs. I shall occasionally revert to ‘experiencer/judge’ to
remind the reader.
61
2Lasersohn (2005, 2008) is the only other theorist I know who challenges contextualism
about PPTs on syntactic grounds. I share his concerns, as will be noted below. Unlike
Lasersohn, however, I do not seek to defend a relativist semantics.
3 See, for example, Epstein (1984), Safir (1991), Condoravdi and Gawon (1996), and Bhatt
and Pancheva (2006). I do not mean to suggest, however, that these authors endorse the
covert projection of experiencer/judge arguments in PPTs. The issue of covert projection
clearly bears upon the more general issue of the projection of implicit arguments, which I
cannot broach in the space available, but see Chapter 3, §5, and §3.3 below.
4 I assume that adjuncts are essentially optional. There are marginal or idiosyncratic cases of
obligatory adjuncts, but such considerations do not bear upon the predicates that concern us
here. See Grimshaw and Vikner (1993) and Baker (2003).
5 Other predicates, such as experiencer ones (frightening, scary, etc.), similarly entail agents.
Hence, for present purposes, I am happy to settle just for a necessary condition.
6 See Kölbel (2002, 2003). The notion of faultless disagreement is not essential to the dispute
between relativism and contextualism I am about to discuss, but it is useful nonetheless (cf.,
Stojanovic, 2007; MacFarlane, 2007a; Richard, 2008).
7 Stojanovic (2007) argues that genuine disagreement only arises where all misunderstanding
has been ruled out, where, say, both parties agree on who the relevant judge of the matter
should be. Be that as it may, my concern here is not to make ultimate sense of disagreement,
but only to inquiry into a potential linguistic basis for its realisation.
8 See Garcia-Carpintero and Kölbel (2009) for general discussion and articulation of varied
positions on the dispute. As remarked upon in the Preface, ‘contextualism’ is perhaps not the
happiest term for the position at issue, for it is also used for positions that view content as
pragmatically determined independent of any specific linguistic license. ‘Indexicalism’ might
be a better term, but that too is somewhat overdetermined. Moreover, ‘contextualism’ is often
62
used as a label for the ‘standard’ or ‘received’ view without a proper specification of who the
contextualists are; see Lasersohn (2005, 2009) for an example of such usage. As in the main
text, hereon I take the contextualist to be someone who holds that there is a specific linguistic
license in the form of a variable for the contextual determination of the linguistic content of
PPT predicate tokens. Of course, one may be neither a contextualist nor a relativist. Cappelen
and Hawthorne (2009) reject relativism in favour of a monadic conception of truth, but do not
endorse contextualism as I have defined it. On the other hand, MacFarlane (2007b, 2009)
rejects contextualism about content but endorses the doctrine for extension. These positions,
however, do not turn on arguments pro or con the presence of the relevant variables in
linguistic structure.
9 Both relativist and contextualist, for instance, must treat the experiencer/judge position as
variable, insofar as PPTs are useable from different perspectives. I take it that the fixation on
the value of the variable is an issue for both sides. My concern below will be, in effect, just
for a particular strong version of contextualism that treats the variable as syntactically
realised.
10 Lasersohn (2008, pp. 317-8) correctly notes that Stojanovic’s claim trades upon a failure to
distinguish between the determination of content relative to context and the determination of
truth value relative to a point of assessment or a circumstance of evaluation. If we assume
that the putative covert variable the contextualist posits is just like a pronoun, then it should
have its value fixed at a context, but according to the relativist, there is no such pronoun in a
bare PPT (the relevant judge is a parameter or index of assessment and so is not valued
relative to context). Thus, at the level of content fixed relative to context, the two positions
are not notational variants at all.
11 The basic idea here is to treat (gradable) predicates as expressing a function from objects to
degrees on a scale. A scale can be modelled as a triple: <D, >, >, where D is a set of points,
63
> is a total ordering, and is a dimension along which the points are ordered. In a non-
comparative case (e.g., tall rather than taller), the context of utterance settles a standard value
ds on the scale such that the predication holds, if the object to which the predicate applies
meets or exceeds ds. The nice feature of this account is that it incorporates the context-
sensitivity of property attribution and the divergent dimensions along which it may be
assessed into a systematic account of the compositional semantics of predicates in such a way
as potentially to capture the assessment-relative factors that pertain to many predicates
discussed by contextualists and relativists alike (see Kennedy (1999) and Kennedy and
McNally (2005) for the detail). For present purposes, I suggest this kind of account as a
means for the contextualist to account for the semantics of PPTs (and other predicates)
without engendering syntactic commitments to covert experiencer arguments. Of course, the
account might be flush with the idea that syntax does contain variable items that correspond
to the dimension of a scale () and the standard degree on the scale (ds). In the absence of
specifically syntactic arguments for the projection of such items, however, I shall assume that
syntax lacks such items.
12 Partee (1984, p. 171) does remark: ‘I don’t know of any cases of implicit arguments which
can be interpreted only as bound variables or only as indexicals’. That thought is right,
although indexicals are not free variables. The question at hand is whether it is also right to
claim, as Stanley and others do, that syntax needs to represent an item with such duality. For
her part, Partee (1989, p. 265) commends ‘a rather holistic’ account of the phenomena, one
that integrates different facets of linguistic understanding at the level of the sentence as used
rather than the simple valuation of a variable at ‘some syntactic level such as a level of deep
structure or a level of logical form’.
64
13
Binding theory has become an increasingly complex field since the development of the
classic position, with perhaps the central issue being whether binding is a mostly syntactic or
semantic phenomenon, or a single phenomenon at all. See Chapter 2 (§2).
14 Schaffer cites Schlenker (2005) as a semantic alternative to classic binding. He could also
have cited Jacobson (1999, 2000), Szabolcsi (1989, 1992), and others in the combinatory
categorical grammar tradition.
15 My point here coheres with Lasersohn’s (2008, pp. 323-6) claim that bound readings of the
kind exhibited in (3) are instances of ‘metalinguistic’ binding, i.e., the bound relation is
established in the semantic metalanguage, not in the object language (cf., Pagin, 2005).
Lasersohn’s underlying claim here is that the judge index that is part of the posited
circumstance of evaluation is represented as being bound in order to fix the right truth
conditions, but that nothing within the linguistic material is itself bound. One can accept the
import of this point without endorsing the accompanying relativist assumptions. One can, in
other words, accept the relevant bound reading as a product of our wider understanding of
tastiness (tastiness involves an experiencer, unlike being artificially sweetened) that is not
fully encoded within the syntax of tasty-predicates. At any rate, I think an argument is
required for such a minimal position not to be default.
16 For present purposes, we may simply assume that fun is unique in its grammatical and
semantic selection properties. At any rate, I can’t think of another English item that behaves
like fun with respect to had and other environments. Bare fun in (9b) appears to be mass
nominal such that everyone took part in some measure of the eventish stuff.
17 Lasersohn (2008, p. 325) claims that binding considerations in fact militate against
contextualism because the presence of a covert variable overgenerates readings. Consider:
(i) Every man gave some woman a fun ride and a tasty dish
65
Lasersohn correctly notes that the experiencer/judge is uniform across the conjunction: it is
either the speaker or each man or some woman making a judgement about both ride and dish.
If, however, fun and tasty are dyadic, with variables to be bound, then there would appear to
be no restriction upon a mixed quantification, where, say, the ride is fun for the man and the
dish tasty for the woman. That reading is unavailable, though. Clearly, this situation favours a
single variable being bound that creates a uniform assessment, which is just what Lasersohn’s
index binding provides. This is an interesting case, but the contextualist has some room to
wiggle. Co-ordinations generally favour a uniform treatment across the phrase, in terms of
movement and ambiguity resolution, for example. Indeed, it might even be that the DP
conjunction in (i) should be treated as a fused predicate.
18 Epstein (1984) posits a covert item pro as an axperiencer argument of the relevant
predicates. The item, however, is never free. It is either construed as a universal quantifier or
referentially dependent on an antecedent. For present purposes, I have no argument with the
kind of account Epstein offers, although see below for some critical comments.
19 One can drop arguments in non-subject positions and sometimes add cognate objects with
variable and complex results. The notion that lexical predicates have a precise addicity,
therefore, is moot. It remains clear, though, that arguments do not iterate.
20 This is the chief moral of the ‘cartography’ project. See citations above.
21 Recanati (p. c.) raises the concern that in the bizarre readings being entertained for the
cases in (16), the truth of the matrix clause does not follow from the truth of its host clause
(the matrix clause plus the adjuncts), making the cases different from those in (17). In itself,
this does not suggest that the readings are illicit, only that the for x adjunct renders the matrix
clause the content of the experience or judgement. In this respect, the same peculiarity holds
for (i):
(i) Liquorice is tasty for Mary
66
Does (i) entail that liquorice is tasty? No; not if we seek to read the clause in a punkt way,
without the implicit understanding that at least Mary finds liquorice tasty, which suffices for
it to be tasty. The cases in (16) are bizarre, but the logic is the same. It might now be thought,
however, that adjunct deletion should always be valid, for adjuncts are just conjuncts; that the
inference is not apparently valid in this case signals that the modification is not adjunction.
This argument is far from compelling. Firstly, the argument assumes that adjunction has a
uniform semantic interpretation, but this is doubtful; indeed, cases of non-intersective
adjunction have been extensively studied. Secondly, in the present case, it appears that for x
is interpreted as a kind of intensional modifier, which would explain the inference pattern,
both in the case of (i) and (16), i.e., the adjunct renders the modified clause a judged or
experienced content.
22 Epstein (1984, p. 23) offers cases such as (i) as evidence for the presence of a covert
experiencer argument:
(i) It is fun for Lucy for Joe to play baseball.
The idea here is that for Lucy overtly realises the argument role of fun, where what Lucy
finds fun is Joe’s playing baseball. As with (16), though, it seems to me that one can get a
reading where Joe finds it fun for Lucy to play baseball, with PRO arbitrary. That would
indicate the possibility of an adjunct construal. Again, stacking the phrases up quickly leads
to bizarreness.
23 Recall from Chapter 2: PRO occurring in an arbitrary position may also have a bound
construal. Consider:
(i) Bill thought that PRO smoking was damaging to his health
(i) is ambiguous between a bound reading, where the person smoking is Bill, or an arbitrary
reading, where Bill might have in mind so-called passive smoking. Hereon, by ‘bound’ I shall
mean ‘obligatorily bound’, unless otherwise stated. I assume, in other words, that the bound
67
reading in this case is a semantic effect of the favoured construal of the lower pronoun, for, in
general, arbitrary PRO cannot be bound. Consider:
(ii) Bill thinks that PRO smoking/to smoke is a bad habit
(ii) is not ambiguous between a reading where only Bill’s habit is at issue and a reading
where the habit as such is at issue. It only has the latter reading, although we might
independently think of Bill as a smoker.
24 For example, if we embed (18a), the control relation remains the same:
(i) Mary wondered why Bill tried to leave
Here, the controller of PRO can’t be Mary.
25 There might, of course, be other reasons to think that implicit arguments are syntactically
projected (cf., Landau, 2010). I shall argue, however, that PPTs do not provide such a case.
26 It might be thought that if tasty does project an experiencer, then it should take a non-finite
complement, much as fun, appear, frighten, etc. do. As it is, it does not, and (i) is pretty
marginal:
(i) ?Liquorice is tasty for Sam to eat
I assume, however, that the grammatical selection properties (sub-categorisation) of
predicates may diverge from their semantic properties.
27 Epstein (1984, p. 25, n. 5) correctly marks the kind of contrast between (22) and, say, *To
play baseball is certain, and claims that the contrast shows that PRO must be controlled, the
problem with the latter construction being that ‘there exists an uncontrolled, hence illicit,
PRO’. Yet none of this explains how PRO can be syntactically controlled by the putative
projected argument of the adjective fun in (22), which for Epstein is a pro item default
interpreted as a universal quantifier, i.e., the would-be operator is not in a c-commanding
position.
68
28
This is not to say that popular exactly patterns with rare, common, etc. One clear
difference is that popular takes singular subjects:
(i)a Ronaldo is popular
b #Ronaldo is rare/common
(ib) has its uses, but only figurally or with the subject serving as an abbreviation for an event
type involving Ronaldo. The crucial feature they share, which is what the argument of the
main text depends upon, is that the predicates are kind-selecting, when, in the case of
popular, at any rate, the subject/complement can be construed as a kind, i.e., popular
exceptionally (of the class) takes subjects that cannot be construed as kind-designating. Thus,
the two inferences below are bad:
(ii) Footballers are rare/common;
Therefore, some footballer is rare/common
(iii) Footballers are popular;
therefore, some footballer is popular.
In (iii), the inference is blocked, for the premise says that the kind of thing that is a footballer
is popular, not that all footballers are or even any individual footballer is. For instance, if
every person on earth had a single favourite footballer, but hated every other footballer, then
the premise would be true: footballers, as opposed to trombonists, say, would indeed be
popular. No particular footballer, however, would be popular; on the contrary, each
individual footballer would be the subject of nigh-on universal loathing.
29 Cappelen and Hawthorne (2011, p. 458-9) admonish that generic construals can interfere
with verdicts of possible disagreements; hence they favour examples such as That was fun in
place of Rollercoasters are fun. This advice seems sound to me, and in fact supports the
present point: fun can be construed generically or not, given one subject or another, but
69
popular can only be a predicate of a type (i.e., the existential inference is blocked), when the
argument can be construed as kind-designating.
30 The thought that the lexical semantics of the predicate affects the construal of PRO is akin
to the position of Dowty (1985) and Jackendoff and Culicover (2003), who take analogous
phenomena to militate against a syntactic theory of PRO. For present purposes, I am neutral
on the various disputes centred on the status of PRO. It bears emphasis, though, that the
treatment of PRO I am suggesting is not by itself an argument against a uniform syntactic
account of arbitrary PRO or an argument in favour of treating non-finite clauses as predicate
abstractions. The minimal consequence is merely that the interpretation of PRO is not
syntactically fixed.
31 Moltmann (2006, 2010a, 2010b, 2012) and Pearson (2013) also argue for a particular
generic construal of PRO that involves it being essentially bound. I shall discuss this position
below. Much earlier, Epstein (1984) argued that arbitrary PRO should be understood as a
bound item in the scope of a pro item projected by the experiencer adjective and construed as
a universal quantifier. Epstein’s principal consideration in favour of this thesis is that (i) has
the reading in (ii), not (iii):
(i) It is fun [PRO to play baseball]
(ii) (xi) [It is fun for xi [PROi to play baseball]]
(iii) It is fun [(xi) [PROi to play baseball]]
That is, (i) is a claim that everyone finds something fun, viz., playing baseball, not a claim
that it would be fun for everyone to play baseball. This observation is clearly correct, but the
distinction also holds on a generic reading of arbitrary PRO without the need to posit an extra
item pro. Besides which, it seems mistaken to construe arbitrary PRO as a universal. Like
other generic construals, the truth conditions of (i) readily admit exceptions. Bhatt and
70
Pancheva (2006, p. 578) also regard arbitrary PRO as generic, but suggest that it supports a
contextual reading too, where episodic. Thus:
(iv) This morning, it was difficult [PRO to dance the tango] since we were all hung over
The natural reading here, though, does not refute the generic construal of arbitrary PRO but
merely supplies the relevant dancers as we, which is not contextual, as far as I can see. This is
supported by the fact that (iv) does possess a fully generic reading. Imagine that the value of
we is some group who prepare the dance floor, but since they are all hung over, they have
done a bad job, rending the planned dancing difficult to perform.
32 See note 24, and Landau (2008), in particular, for scepticism of the significance of
backward control for the movement theory of control.
33 See Ross (1967) for the original discussion of sluicing and Merchant (2001) for an
excellent contemporary discussion of the phenomenon and related matters.
34 I assume that the T head doesn’t move into the C head position in the lower clause.
35 Haegeman (2012) explains the asymmetries in terms of intervention effects in the ECP-like
tradition. The ultimate explanation of the phenomena is beyond my present scope.
36 Plausibly, (39a) improves with the dative argument being focused. That, however, is an
independent matter that does not affect the underlying asymmetry.
37 The gap construction is fine:
(i) Billy drinks wine on Wednesdays, but beer on Thursdays
Gapping, however, is not diagnostic of the argument/adjunct distinction precisely because
gapping allows for stranded complements.
38 Likewise, the gap constructon is OK:
(ii) Billy drinks wine on Wednesdays, and Sally beer on Thursday.
39 One might think that the cases here not VP ellipsis, but pseudogapping, where for sally is a
remnant argument of the elided tasty. A condition on ellipsis, however, is that it operates
71
backwards; pseudogapping doesn’t. The cases in (42) admit backwards constructions. For
example,
(i)a *Although Bill doesn’t wine, he does drink beer
b Liquorice doesn’t for Mary, but does seem tasty for Sally
Ultimately, however, even if a pseudogapping analysis is available, since it isn’t a diagnostic
on the argument/adjunct distinction, the ellipsis phenomenon is rendered merely neutral.
40 The phenomenon witnessed in (45) is hardly remarked upon in the literature on genericity.
It relates, however, to the slight shift in meaning that arises when a generic adverb is inserted
into a ‘characterising’ sentence (Krifka, et al., 1995, p. 9-10). Thus, contrast the change in
meaning between the cases in (i) and (ii); radical in the first pair, subtle in the second:
(i)a A mosquito bit my arm
b A mosquito usually/typically/generally bit my arm
(ii)a A mosquito carries malaria
b A mosquito usually/typically/generally carries malaria
That there is a shift at all suggests that the generic without the adverb is understood as
expressing a general rule, or perhaps carries the presupposition of universality. Hence it is
that the conjunction of the generic with an exception as in (45) produces a certain oddity,
which is wholly absent when the adverb is present:
(iii) Mosquitoes usually carry malaria, but this one doesn’t
The picture is complicated, however, by the fact that generics can be true even if most
instances of the generic nominal do not satisfy the predicate (most mosquitos do not in fact
carry malaria) (e.g., Leslie, 2008). Cohen (1999, 2004a) discusses a similar effect whereby,
according to him, the generic implies homogeneity across its domain, whereas the adverbial
construction does not. If that is right (it seems to be), it might be that (iii) is OK because the
adverbial licenses arbitrary exceptions, whereas (45) isn’t OK because a constraint of
72
homogeneity would demand that an arbitrarily chosen instance should satisfy the predicate
much as any subset of the domain of the generic should be homogenous with respect to
satisfaction of the predicate. Cohen doesn’t say as much, but the idea is consistent with his
notion that a generic works as a default rule of generalisation that allows inferences to
instances in the absence of other evidence (hence, homogeneity) (cf., Cohen, 1999, 2004a,
pp. 550-1). Similarly, Cohen (2004b) claims that adverbs of quantification associate with
focus in a way generics do not need to, i.e., with focus, a consideration of alternatives fixes
what satisfies the restrictor or scope of the quantification. Without focus, therefore, the
generic is construed as a kind, so should not admit arbitrary exceptions. However these and
other complications might be resolved does not appear to affect the basis of the comparison
between PPTs and generics pointed to in the main text. See Mari, et al. (2012) for an
overview of recent work on generics.
41 Following the standard GB explanation, we may take the marginal status of (41a) to be due
to a subjacency (‘weak island’ in this case) violation, where the wh-item has moved outside
of an indirect interrogative complement. Glanzberg does not offer an explanation.
42 Again, Glanzberg does not offer an explanation of the unacceptability, but I take him to be
presuming that it is due to an ECP-like violation in addition to a subjacency violation, with
where moving from the interrogative complement, with its lower trace or copy being blocked
(not ‘properly governed’) by the presence of whether, which intervenes between the
antecedent and the trace/copy.
43 There is an issue here pertaining to the difference between movement from adjuncts and
the movement of adjuncts, i.e., in the present case, whether or not the preposition to needs to
move along with the wh-items. I assume that the illicitness or not of extraction from adjuncts
is independent of the question whether the movement of adjuncts differs from that of
arguments (see Stepanov, 2007).
73
44
Lasersohn (2008, p. 327) considers a range of object deletion cases that appear to exhibit
the weak crossover effect. For example:
(i)a The runner who lost threw every race
b (every race x)[the runner y ˄ lost(y, x)](threw(y, x))
c [TP every racei [TP [the runner who lost xi [VP threw <every race>i]]]
According to Lasersohn, (ia) does not admit the reading exhibited in (ib), which would be
explicable as a weak crossover violation, given the structure in (ic). Firstly, it is not clear to
me that the reading in (ib) is so bad, but even if it is, the explanation might not devolve upon
the structure in (ic), but more on that we understand the object of lost to be independently
existentially bound or supplied by context, which precludes its being bound by the scoped-out
every race. This, at any rate, would save us from assuming that the object of lost is
syntactically projected.
45 Stephenson (2007) suggests a further line of argument. PPTs select for specific PPs:
(i)a fun for Mary/*fun to Mary
b ??pleasing for Mary/pleasing to Sue
c tastes good to Mary/ *tastes good for Mary
Such data suggest that the PPTs lexically select for particular PPs, given that the choice has
no clear semantic rationale. This observation, however, strikes me as incorrect. The
prepositions do differ. For x is more act/judgement-orientated, whereas to x is more
experience orientated. Thus, looks green to Mary is much preferred to looks green for Mary,
where looks green is picking out the content of the experience. If this is so, then it is little
wonder that fun, as a predicate of events, should select for rather than to (the point generalises
to the other cases exemplified, and more generally, I think).
46 I thank Kjell Sæbø, Barbara Partee, and Ede Zimmerman for conversations on this topic,
although I am not sure we agree.
74
47
Of course, this kind of categorisation has many exceptions and some cases cross-classify.
48 I have used the small clause construction, but the same affect holds with non-finite clauses,
as with Bill finds Sam to be tall/10st/a man.
49 Two caveats. Firstly, it is not obvious if the authors mentioned would wish for the
consideration of find and other verbs to be a pure matter of syntactic selection. I am,
however, presently interested in such a strong position. After all, we are after a syntactic basis
for the argument projection of PPTs. If find merely expresses a matter of opinion, with no
selection restriction, then the data only tell us that the predication of PPTs expresses a matter
of opinion, which no-one denies. The issue of contention is whether or not this fact is
syntactically underwritten. Similarly, no-one need deny that PPTs express a relation of direct
experience, to which (stative) find is sensitive, i.e., one can’t find x to be F unless one has
direct experience of x (Stephenson, 2007; Pearson, 2013). Again, the question is not whether
there is such a relation, but whether such a relation is syntactically realised, which it would be
were find only to select complements that possessed the relevant relational argument
structure, i.e., the projection of a direct experiencer.
50 For example, (i) is somewhat odd:
(i) #Gödel found the wine to be tasty and arithmetic to be incomplete
On the other hand, (ii) strikes me as OK, even though there are clearly two senses of have
involved:
(ii) Sam has a broken leg, but a nice car
The lighter the verb is, or the more regularly polysemous it is, the more acceptable the
elliptical conjunction is. For what it is worth, I find find to be fairly light.
51 By ‘overridden’ I mean that the speaker/hearer holds the following equivalence to be true:
It’s raining iff it’s raining somewhere. The truth of this equivalence, however, is not
75
linguistically generated (analytic), but follows instead from the agent’s conception of raining
as a locational event, i.e., the verb does not select a locative argument.
52 As with weather reports, this means that the chef and anyone listening to him, holds that
liquorice is tasty iff liquorice is tasty for someone.