Gradable adjectives: A defence of pluralism

Abstract

This paper attacks the Implicit Reference Class Theory of gradable adjectives and proposes instead a ‘pluralist’ approach to the semantics of those terms, according to which they can be governed by a variety of different types of standards, one, but only one, of which is the group-indexed standards utilized by the Implicit Reference Class Theory.

Notes

¹First, it's important to note that I am not using the label ‘IRT’ to cover just any treatment of gradable adjectives that somehow or other makes use of the idea of implicit comparison classes. Much of the most important literature on these adjectives uses syntactic evidence to yield conclusions about the role that implicit comparison classes play in at least some uses of gradable adjectives. Such conclusions do not get you to what I'm calling ‘IRT,’ according to which, where ‘A’ is a gradable adjective, ‘X is A’ quite generally means that X is A for an F.

So, how complicit in the myth is the literature that is devoted to gradable adjectives and that uses the notion of comparison classes? Here I'll confine myself to one very prominent example of the literature in question: [Ludlow 1989—though much of what I say about it applies to other work as well. First, there seems to be no hint at all here of what in Section III I will call the ‘strong’ form of IRT, according to which it's not only the case that gradable adjectives are quite generally governed by what I'll call group-indexed standards, but also that supplying the relevant comparison class is all context ever does in fixing the content of a use of a gradable adjective.

However, there does seem to be suggestion of what I call the ‘weak’ form of IRT, which claims only that ‘X is A’ quite generally means that ‘X is A for an F,’ without also insisting that all context does in fixing the content of such a claim is supply a comparison class. Ludlow's paper opens with these words: ‘A common view of attributive adjectives like ‘tall’, ‘fast’, ‘large’, and ‘heavy’ is that they express relations between objects and either some comparison class (hereafter c-class) or an attribute’1989: 519]. This ‘common view’, then, is not presented as one on which such adjectives sometimes or often express relations between objects and c-classes, but one on which such relations simply are what the adjectives express. Now, it also is not explicitly specified that on the ‘common view’ the adjectives always express such relations to c-classes, but in this context, the simple statement that such adjectives express such relations (as opposed to, say, that they often express such relations) strongly suggests something at least approximating that they always express such relations. (Perhaps the best reading is something along the lines that the adjectives at least normally express such relations.) So, that ‘common view’ seems to be at least very close to what I call the weak form of IRT. After a little more explanation, Ludlow continues: ‘If we adopt this view, an interesting question arises’ 1989: 519], and he proceeds to address that question, and others, and the ‘common view’ is never withdrawn. It seems that the ‘common view’ is being assumed—and perhaps even to some extent defended, since Ludlow seeks to show how the questions that arise can be answered.

However, Ludlow's focus is not on defending the ‘common view,’ but on other conclusions. He seems willing to assume the ‘common view’ and to work within it to get at the questions that are his focus. But the supposition that many uses of gradable adjectives are not governed by group-indexed standards would seem to do little to upset the conclusions that are Ludlow's real focus. Thus, those who assume IRT, even in its weak form, would seem to be making a mistake if they appealed to what's shown in [Ludlow 1989 in support of their assumption; Ludlow is setting his sights on quite different targets.

²A few examples:

‘All I was claiming was that I know it…

 -quite well

 -beyond any reasonable doubt

 -by ordinary standards

 -by any reasonable standard

 -with a high degree of precision.’

‘I never meant to be claiming to know it…

 -for certain

 -with absolute certainty

 -beyond all possible doubt

 -perfectly

 -as God would know it.’

These seem to be perfectly fine pieces of English, not cases of mere ‘philosopher-speak’—though philosophers, too, have occasion to say such things. The challenger might not like these responses, and might have objections to them, but the responder doesn't seem to be abusing the English language in using these as devices of clarification. For many more such examples, see Ludlow 2005. It's important that these are not just modifiers that ‘S knows that P’ can take, but function as devices of clarification: In a context in which one's meaning or consistency are questioned, these devices at least arguably can function to clarify the content of one's claim. By contrast, while ‘quite well’, ‘with a high degree of precision’, ‘as God would play it’, and ‘perfectly’ can be used to fill in the blank in ‘She played the piano piece _____,’ the resulting sentences don't seem to function as devices of clarification.

³This is most vividly displayed by ‘here’—though similar problems can arise with the temporal extent of ‘now’, and, as very strange imagined cases can show, with the issue of just how far out, temporally and/or spatially, ‘I’ reaches. ‘Here’ is used to refer to the location of utterance, all right. (Well, this already seems questionable, because of cases like those in which a speaker says, while pointing to a spot on a road map, ‘The exit we want is here’. How and whether to handle such cases within a general theory of the workings of ‘here’ can get to be fairly complicated business.) But what are the boundaries of the region designated by a given use of ‘here’? How far out does a given use of ‘here’ reach? Sometimes we have simple little phrases that can explain fairly precisely just what we mean: To echo an example that Stephen Schiffer uses, we might mean ‘in the building’, ‘on campus’, ‘in the States’, etc.—though on some occasions it might be unclear exactly which we mean [Schiffer 1995: 112 – 13]. Often, when I use such a simple phrase, I feel that I could just about as well have used another—a bit broader or narrower—instead: Instead of ‘in the building’, I could just as well have said I meant ‘somewhere in the philosophy department’, when I said that Larry wasn't ‘here’. It at least feels as if, while I am somehow giving my permission to let my earlier use of ‘here’ be understood in the specified quite precise way, what I meant at the time I said it was vague enough, even in my own mind, that a different precisification would have been just as faithful to what I meant at the time of utterance. Schiffer presents a case in which it isn't clear which of many possible specifications of the meaning of ‘here’ is right, and advocates a supervaluational approach to such situations. Schiffer does seem to assume, though, that each of the candidate specifications for the use of ‘here’ in question can be given by means of a handy little phrase. But there are cases in which such handy phrases don't just fail to precisely provide the meaning in context of ‘here’, but fail to even specify the candidate meanings among which the given use is indeterminate. Sometimes handy little phrases simply fail us, as we can see by moving to a different example. ‘Is Michael still here?’, a speaker might ask, while looking about for Michael, in a conversation taking place at one of the ‘smokers’ held in giant ballrooms at APA conventions. Well, obviously, Michael isn't right here, within just a few feet of us. What could the speaker mean? ‘Here in the ballroom,’ perhaps? That's the only simple and fairly precise phrase that seems available. But, no, Michael himself was taking part in our conversation just a minute or so ago, before he drifted a few feet over to say a few words to someone else who was standing nearby. Now he's apparently drifted even further away, but we're all quite certain he's still in the ballroom, so the speaker wouldn't be asking that. But what, then? Given the way the speaker casts her eyes about as she asks the question, she seems to mean, however vaguely, to be designating a region of the ballroom in which there's a relatively high percentage of people known to us. (Has Michael just moved a little further away, to a spot around here, or has he perhaps slipped out of ‘the region’ to go to the bar and get a new beer or to talk to a whole new group of people in some other area of the ballroom?) But if you were to ask the speaker to specify exactly what area she meant to be designating by her use of ‘here,’ she'll just laugh and explain that she didn't mean anything very exact: ‘You know, right around here’, she might say, perhaps making vague, sweeping gestures with her hands. No simple phrase is available, and, for certain spots at some distance, but not a great distance, from the speaker, even the speaker has no idea whether they would or would not count as being in the region she meant to be designating with her use of ‘here’. What's more, there seems to be no way, at least by means of simple little phrases, to precisely identify even various candidate meanings that the speaker's use of ‘here’ is indeterminate among.

⁴Standards that can be specified simply by citing the threshold for what percentage of the Fs must be A-er than to count as being ‘A’ are a sub-type of what we can call ‘roster standards’. Roster standards are functions from ‘rosters’—specifications of how A each of the Fs is—to what, starting in Section IV below, we will call ‘tape-measure standards’. (The idea is that if a standard is a roster standard, then the only information one needs to generate a tape-measure standard from it is the information from the relevant ‘roster’—specifications of how A each of the Fs is.) So, for instance, the group-indexed, taller than at least 98% of 10-year-old boys, is a roster standard, because, to generate a tape-measure standard for tallness from it, the only information one needs is how tall the various 10-year-old boys are. (Since I have little idea how tall 10 year old boys tend to be, I have little idea what tape measure standard is generated by the roster standard in question.) Roster standards need not be matters of simple thresholds of percentages, but those that work by simple percentages in that way are the only ones that seem easy to specify and explain. In fact, it's possible that we resort to simple percentages to explain our standards simply because those are the only easily specifiable standards in the vicinity, and that many of our ‘roster’ uses of adjectives are in fact more complex, despite the simple way we would specify them if asked. I say, ‘Ralph is rich’, and I'm asked what I mean. ‘Rich for an American’, I reply. ‘How rich for an American?’ I'm asked. Meaning to employ fairly demanding standards, I reply, ‘Oh, in the upper 5% of Americans’. But how would my use of ‘rich’ apply to a counterfactual situation in which the distribution of wealth in America were even more unequal than it is? What if well over 99% of America's wealth were held by the upper 2% of Americans, and the remaining 98% of Americans were all desperately poor, and all very close to one another in how wealthy they were? Is my actual use of ‘rich’ such that it would count someone as rich for an American if they were in the upper 5%, but not the upper 2%, of Americans under those circumstances? If Ralph were in that position—in the upper 5%, but only marginally wealthier than the very poorest Americans, and, like almost everyone else, much poorer than the upper 2%—it seems that one wouldn't have to be employing particularly demanding standards to count him as not being rich. But he would be in the upper 5%, and so would count as ‘rich’ if the standard really were simply that of being in the upper 5%—which is the standard I had specified, thinking that I was employing a very demanding standard. Perhaps, despite what I'd say if asked what I mean, my use of ‘rich’ would be more accurately captured by a more subtle roster standard (a more subtle function from rosters to tape measure standards), but, since my listeners and I would naturally assume a more normal and even distribution of wealth than exists in our imagined situation, there's no harm done in specifying what I mean by simply saying ‘upper 5%,’ not worrying about whom would be counted as ‘rich’ under various extreme counterfactual conditions.

⁵By ‘interestingly different’ here, I specifically mean that Fara's cases seem to involve group-indexed standards that are not what, in note 4, above, I call ‘roster standards’: One cannot determine a tape-measure standard for oldness for dogs from the group-relative standards Fara is pointing out, using only the ‘roster’ information of how old the various dogs are.

⁶Here I am not assuming that content cannot be vague. Of course, it can be. But the uses of the adjectives that we looked at have (fairly precise) contents that could not have been fixed simply by specifying the comparison classes relative to which the uses are to be understood.

⁷This conclusion is also shown by our brief look at the workings of ‘here’ in note 3, above. The region of the ballroom vaguely designated by our speaker's use of ‘here’, without the benefit of any handy little phrases to specify it, would also seem to be something supplied by context, though it's not the kind of thing that seems to make the assumed short-lists.

⁸Christopher Kennedy 1999 has argued, on different grounds, for a thesis complementary to my conclusion here. Kennedy concludes that ‘an empirically and explanatorily adequate semantics of gradable adjectives must introduce abstract representations of measurement—degree qua intervals—into the ontology’. Kennedy argues for the need for tape measures (not his term) in our semantics for gradable adjectives because accounts that work with comparison classes but without tape measures cannot account for a variety of comparative constructions. If Kennedy is right and tape measures are involved anyway in the semantics of gradable adjectives, one might then wonder: ‘Well, what's to stop us from at least sometimes simply using gradable adjectives governed by tape-measure standards, without any comparison class involved? Often, that would be very handy’. Well, if ‘A’ quite generally meant ‘A for an F’, the meaning of gradable adjectives is what would stop us from using them with tape-measure standards. But apparently nothing stops us from using tape-measure standards. We do it very often.

⁹More generally, this is the case whenever what I'm calling a ‘roster standard’ is used. See note 4, above.

¹⁰It's tempting to think that something must be an F before it can sensibly be said that it either is or isn't ‘A for an F’. If so, then that Joey isn't a 10-year-old boy would be by itself, regardless of Joey's height, sufficient to rule out that Joey is tall for a 10-year-old boy. I'm not assuming that here. However odd it may be to say ‘X is A for an F’ when X isn't an F, suppose the statement could still be true. (Suppose, then, that though I'm far more than 10 years old, ‘He is tall for a 10-year-old’ can still be true of me.) Still, from the grandfather's point of view, if the standard is set by 10-year-olds, and Joey may well be just 6, then Joey may well not be ‘tall’ by the relevant standard.

¹¹As mentioned in the previous note, there is something at least quite odd about saying ‘X is A for an F’ when A is not an F. Thus, even if a 14-year-old can count as being ‘tall for a 7-year-old’, there would be something quite unusual about applying such a standard to him.

¹²For some explanation of this presumption, as well as the methodology of using ‘flat-footed, all-in-one-breath’ conjunctions to test for ties (or what Kit Fine called ‘penumbral connections’ 1975: 270]) among potentially context-sensitive terms that the presumption underwrites, see DeRose 1998: 69 – 72].

¹³Stanley's (30), and similar examples, show that ‘X is A and/but Y is not A’ can be true where X is not A-er than Y. Recognizing group-relative standards shows that this can happen where there is no switching of the standard that governs the use of A in the conjunction. ‘X is A and/but Y is not A’ seems to entail ‘X is A-er than Y’ where the two occurrences of A in the conjunction are governed by the same standard that is of the tape-measure or group-indexed variety, but not where a group-relative standard is used.

¹⁴Stanley also relies on other examples, some of them quite interesting, to show that we appropriately change contents for context-sensitive terms over conjunctions 2004: 134 – 9]. Addressing these examples one by one would take us too far afield for our current purposes, since the other examples don't involve gradable adjectives. But it is worth looking at one of Stanley's examples here, because holding to our presumption in treating that example can make us recognize a kind of norm that is perhaps important to gradable adjectives as well as to the term, ‘many’, that's featured in Stanley's example. So, consider Stanley's

(32) In Syracuse, there are many serial killers and many unemployed men.

Do we have to suppose that the two instances of ‘many’ are governed by different standards to make sense of this claim? Well, how many xs do there have to be in Syracuse for ‘there are many xs in Syracuse’ to be true? If you're looking for a number, there seem to be many numbers that seem about as likely as any others to be the threshold that would make (32) true when that number sets the threshold for both uses of ‘many’. Yet there's some plausibility in Stanley's suggestion that (32) can be used in such a way that it takes more unemployed men to make its second part true than it takes serial killers to make its first part true. (I'd find the suggestion still more plausible if I could get myself to hear In Syracuse, there are many serial killers, but not many unemployed men as being potentially OK (appropriate/correct, as well as true) despite the fact that there are more unemployed men than serial killers there. But I can't.) But when I suppose that Stanley's right, that doesn't make me think that there is a change in the standards governing ‘many’ from one use to the other in (32), but rather inclines me to conclude that ‘many’ can be governed by standards that aren't simply a matter of the number of things in question—perhaps something like the ‘norms of expectation’ that Fara writes about 2000: 67], so that both of the ‘many's mean something like ‘surprisingly many’.

¹⁵Ludlow 1989: 520 – 4] and Stanley 2000 attempt to handle examples like (28), (30), and (28b) in another way. Stanley very compactly summarizes their approach at 2003: 272, n. 5]. Ludlow and Stanley assume in their treatments that uses of gradable adjectives have a variable for comparison classes in their logical form, and in Section V, above, we've already seen reason to doubt that. At least insofar as I have a good feel for what's meant by ‘logical form’ in such discussions, the very ‘logical form’ of ‘X is A’ cannot have a variable for comparison classes if the meaning of such claims involves comparison classes only on some resolutions of its context-sensitivity, and there are good uses of it where there simply is no relevant comparison class that is a component of the meaning.

I suspect that if my pluralism is correct about gradable adjectives, the role that ‘logical form’ is often given in explaining various ‘binding’ phenomena will have to be revised. The binding phenomena appealed to by the likes of Ludlow and Stanley are real; I don't question the existence of readings they claim (at least in most cases). But I think we have good reason to suppose that often the variables being bound are not present in the syntax or logical form of the claims in question, but get introduced only as context-sensitive elements in the sentences take on particular values. (Where such a bound reading of the sentence is intended, that makes salient a reading of the context-sensitive element that provides the variable to be bound.) But I also think we have other reasons not dependent on pluralism regarding gradable adjectives for drawing the same conclusion, anyway.

¹⁶If a die-hard defender of IRT were to insist that ‘small’ here is indexed to the smallest type represented in the book—say, butterflies—we should observe that if we were to ask John whether an ant which is not as large as his butterfly though it is quite large for an ant is ‘small’ and a picture of it belongs in his scrapbook of ‘small’ things, he'll say no.

¹⁷Though it's natural to use ‘minimalist’ to describe this proposal, note that this ‘minimalist’ account is very different from the thesis of ‘semantic minimalism’ championed recently by Herman Cappelen and Ernie Lepore 2005. Cappelen and Lepore's ‘minimalism’ explicitly denies the very context-dependence in gradable adjectives that I seek to understand.

¹⁸Thanks to Peter Ludlow and to two referees for this journal for very helpful comments.