Clouds and conceptual analysis

Or why we should think of the meanings of many natural language expressions as clouds or bundles of closely related but distinct intensions

I’m going to use my first real post to lay out a few ideas about definitions that I’ve had running around in my head for a while. It will be a relief to finally get the ideas down on paper, and it may even turn out to be interesting and useful for other philosophers.

Kenny Easwaran recently wrote a post titled “Pragmatic Conceptual Analysis.” The basic idea of that post is that concepts often admit of more than one analysis, depending on the inquiry we’re engaged in. He starts off pointing to the various different definitions we find of mathematical concepts such as “dimension” and “number,” but then he moves on to more ordinary concepts, such as “tall,” and philosophical concepts such as “knowledge,” “justification,” “meaning,” and “intention.” His key point with respect to all of them is that there are many different ways of analyzing those concepts, all equally good for different purposes, and no one of them has the claim to being the right definition, at least simpliciter. So the whole exhausting and typically fruitless game of finding counterexamples to definitions is based on a mistake. If there are many different legitimate definitions of a concept, some of which include something in the extension and some which exclude it, it’s no use to use that thing as a counterexample to the various definitions, because such counterexamples presuppose a standard of correctness for conceptual analysis that simply doesn’t exist.

I agree with this completely, although I think there’s more to say about the picture of meaning and definition underlying Kenny’s view, and the point of this post is to say it. But first let me tell you a story about how I came to agree with the view. Recently I’ve been doing work in the philosophy of language—in particular, in pragmatics, on issues related to manipulative speech, lying, misleading, implicature, dogwhistling, etc. In one paper I aim to give a definition of manipulative speech; I aim to single out a category that includes many of the forms of speech that we intuitively categorize as manipulative, and makes a number of plausible predictions concerning novel forms of manipulative speech. The reaction to that paper, in various talks and in the review process, was often “THAT’S not manipulation! The right definition of manipulation is [insert objector’s preferred definition of “manipulation” here]!!!” The objectors would point out that my definition of manipulation makes manipulation, for instance, intentional, or deceptive, or that it excludes paternalistic manipulation (it actually doesn’t, but nevermind that). They would then point to various actions that they consider manipulation that violate one of these conditions, and so conclude that I am mistaken. Indeed, some audience members were so dismayed at my definition that they called it “obviously wrong,” and “ridiculous.”

But I wasn’t moved. On the contrary, I was convinced that the definition I had offered captures a central “strand” of our usage: roughly, I defined manipulation as intentional, covertly non-cooperative influence. Manipulators are, so to speak, false friends. They pose as cooperative to achieve their ends at the expense of our own. I think this definition is extremely useful, because it connects manipulative speech directly to the Gricean framework that so many philosophers of language know and love. It also draws on a number of definitions already present in the literature—I didn’t invent it out of thin air. Nonetheless, it got a huge amount of counterexample-style pushback.

What do you do when several people object to your definition in completely different ways? One thing to do is to play the counterexample game with them by complicating your definition. In this case I could have tried to give necessary and sufficient conditions for the application of the term “manipulative” in English, broadening and narrowing my definition in the requisite ways. But then again, there are already maybe 20 different definitions of manipulation in the literature, and no matter how I changed the definition, it was sure to get counterexampled by someone. You can’t please everyone, after all.

Instead of doing that, I was stubborn. Or, to put a nicer spin on it, I used these responses as an opportunity to think more carefully about what is going on in these kinds of debates, and ended up drawing the following conclusion: that there are several closely related concepts or definitions of manipulation¹ that agree on a wide range of ordinary cases, but diverge in certain edge cases that are of interest to philosophers, and these different concepts all “do business” and go undistinguished under the heading of “manipulation.” My concept was among those initially going undistinguished under that heading, as were those of my interlocutors. All of our definitions were “strands” that are typically woven together, but in fact can be teased apart.

To illustrate, let’s contrast two definitions of manipulation: one that does not require manipulation to be intentional, and one that does. The former we can call the “inclusive” definition, and the latter the “exclusive” definition. Clearly, the inclusive case includes all of the instances that the exclusive case does—so the two definitions agree on cases of intentional manipulation, which, we can assume, are the majority of ordinary cases. It’s only when we consider cases of unintentional manipulation, which are plausibly less common, that the more and less restrictive definitions disagree. Much the same is true with the issues pertaining to deceptiveness, paternalistic manipulation, etc.

When someone says “but look, there are things that we call manipulation in English that are not intentional,” there are two options. One is to admit that the exclusive definition is wrong, because they’ve come up with a counterexample. The other is to say that the exclusive definition picks out a different concept, but one that is no less legitimate. After all, it seems clear that there is a perfectly coherent definition of “manipulation” on which it is intentional, and another perfectly coherent definition on which it need not be. Likewise, it seems clear that there are perfectly coherent definitions of manipulation on which it is deceptive, and perfectly coherent definitions on which it need not be. It’s not like we don’t understand these concepts!

The question, then, is whether one of them has a claim to being the unique, correct definition of “manipulation” in English or not. Spoiler: I don’t think there is one unique correct definition—to echo Easwaran, I don’t think that any of them have a claim to being the right definition—because I don’t think there are strong enough metasemantic constraints to guarantee anything like uniqueness or perfect determinacy. Instead, all of these different definitions are left open by the metasemantics of ordinary English, and what is left is a cloud or bundle of candidate concepts, none of which can be ruled out definitively by counterexample. But when we offer a definition, we are choosing one of the definitions left open by the metasemantics of English for a particular philosophical purpose.

As a way of warming you up for my arguments for this view, I want to point out just how common it is for there to be competing definitions of expressions of English on which there is little or no convergence, and how weird this would be if there were a unique standard of correctness. If there were a unique, correct analysis of a concept, we’d expect there to be some convergence on it, rather than massive disagreement—conceptual analyses are supposed to analyses, and analyses are supposed to be a priori, after all. But this is definitely not what we find. Rather, disagreement over definitions is ubiquitous. For whatever reason, it seems to be particularly prominent in philosophy of language and pragmatics. For instance, is implicature intentional? Philosophers disagree. Are dogwhistles intentional? Again, disagreement. Must lies be intended to be deceptive? Here again, major disagreement.

Or consider a more mundane case: the verb “see.” You might think that it is totally determinate what it is for a perceiving subject to see something. But what about in cases of hallucination? If I hallucinate a tiger, do I also see a tiger? I’ve actually done some empirical work on this question, and it turns out that there is just a significant amount of disagreement, although what I found was that speakers lean slightly toward the verb being fully extensional (although see here for empirical work that seems to show the opposite). As with manipulation, I think the right view is that both concepts are available, and in ordinary cases, English does not distinguish between them, because they overlap for the vast majority of ordinary instances of perception. But hallucinations, as fringe cases, reveal that our concept is not as determinate as we might have thought, and that there are in fact more inclusive and more exclusive definitions available for “see,” both of which are consistent with the present metasemantic constraints of English.

At risk of belabouring the point, here is Orwell from “Politics and the English Language”:

The words democracy, socialism, freedom, patriotic, realistic, justice, have each of them several different meanings which cannot be reconciled with one another. In the case of a word like democracy, not only is there no agreed definition, but the attempt to make one is resisted from all sides. It is almost universally felt that when we call a country democratic we are praising it: consequently the defenders of every kind of régime claim that it is a democracy, and fear that they might have to stop using that word if it were tied down to any one meaning.

For all of these expressions, it seems relatively clear that there are many different definitions available, and the terms are often used while running these definitions together, consciously or unconsciously, strategically or innocently. Perhaps context will sometimes help us get to a unique definition, but not always. Many of these terms are used perfectly felicitously without being tied down to any one definition.

My proposal is that, for many, many terms of English, including a wide range of terms that play important roles in philosophical debates, there is a “cloud” or a “bundle” of admissible definitions. What determines whether a definition is admissible as opposed to inadmissible are the conventions of English at a certain point in time (more on this below). When a word with a cloud of admissible definitions is uttered, context may narrow this cloud down, although often not all the way. When a speaker utters a word with many candidate definitions, but context fails to single out a unique definition (concept, intension), what we have is an instance of felicitous underspecification, a phenomenon discussed at length by Jeff King and that I discuss here.

This kind of indeterminacy—i.e., the admissibility of many different definitions, given the metasemantics of English—is fairly close to a number of different familiar philosophical notions, including vagueness, polysemy, and open texture. You might try to assimilate it to vagueness, although as far as I can see, the kind of indeterminacy at issue here isn’t one that gives rise to the standard features of vagueness. The type of indeterminacy that I am here positing for “manipulation” and “implicature” and “see” is not quantitative. These are not gradable adjectives associated with some sort of scale, where indeterminacy is just a matter of not having enough of some underlying quantity. Rather, there are many candidate definitions available, but these candidate definitions differ on whether they require things in their extension to have some particular feature. This is different from ordinary cases of vagueness.

The second possibility is that this is just an instance of polysemy. Maybe that’s right, but it depends on how polysemy itself is understood. If you take certain standard tests for polysemy to be definitive—tests such as the zeugma test—these terms typically do not get categorized as polysemous. I don’t think that linguists would agree that “manipulation” is polysemous in 20 different ways, even though philosophers have offered at least 20 different definitions, each of which seems to get at some important principle of categorization underlying the expression. While I don’t want to argue over terminology, my feeling is that the phenomenon I’m describing is somewhat more subtle and less widely recognized than polysemy.

I think the phenomenon I’m pointing to is closest to what Friedrich Waismann termed “open texture.” On Waismann’s view, no matter how precise of a definition of an expression we think we’ve offered, we’ll always end up encountering weird cases that we don’t know how to categorize. In my examples above, these weird cases are things like *non-intentional manipulation*, *bald-faced lies* and *hallucinatory visual experiences*. These weird cases present us with choice points at which we have to pick a definition and so make a decision concerning how to categorize some novel case. Or, we may just concede that there are different concepts around, and that definitive categorization is out of reach—it just depends on what we mean.

Waismann originally used the expression “open texture” to pick out the feature of legal language that makes it the case that we often need to extend our patterns or principles of legal categorization to novel cases through acts of judicial interpretation. I think that this phenomenon arises for basically *all* language—no matter how firm a grasp of a concept we feel we have, we are always going to encounter weird cases that we won’t know quite what to do with. No matter how sure we are that we’ve carved modal space in two, we should always expect the unexpected: we should recognize that eventually, we will encounter novel cases we haven’t anticipated that break our scheme of categorization, and force us to further specify the definition.²

Here’s what this looks like in the case of our example. Suppose we offer a definition of manipulation as “covertly non-cooperative influence.” We might then encounter an instance of what looks like manipulation, but isn’t intentional (or covert). Does this count as manipulation or not? For Waismann, we then have to legislate—we have to make a decision about whether non-intentional manipulation counts as manipulation. But what we realize very quickly is that there are many such choice points for many words other than “manipulation.” The conventions of our language leave open a range of definitions, all legitimate, but when we encounter novel or controversial cases we can prune or narrow the set of definitions between which our original usage was neutral.³

I don’t want to take a stance on exactly how we should handle this kind of indeterminacy. Formally speaking, I suspect it can be handled in one of the same ways that we handle other forms of indeterminacy. My goal is just to offer a new understanding of what’s going on in the lexicon that makes better sense of a range of puzzling and frustrating phenomenon we encounter while doing philosophy. This view makes sense, first of all, of the frustration so many of us feel about the interminable definition-counterexample dialectic, and the nagging feeling that there’s something slightly wrong about counterexampling a definition. It also makes sense of why there are such interminable debates over definitions, and why there are so many conflicting definitions of the same concepts in philosophy, with little to no convergence or resolution.

Okay, now for some arguments. What about someone who comes along and says: “Justin, you’re totally wrong. There is a single standard of correctness for definitions—disagreement doesn’t show otherwise, it just shows that philosophy is hard!” I’m sympathetic to this response when it comes to many philosophical problems, and in general I don’t think we can draw any radical epistemic conclusions from the presence of pervasive disagreement. However, I am not sympathetic to this view when it comes to definitions. I think that disagreement over definitions is often not substantive, and isn’t on a par with ordinary philosophical disagreement. It seems like it’s better to think of definitions not as true or false, but rather as available or unavailable given our current categorization scheme, and better or worse given our interests.

But aren’t there arguments from metasemantic premises to the conclusion that there is a single, unique standard of correctness at which analysis aims? One such argument is that we are competent speakers of English, and competent speakers of English know and agree on the extensions of the terms they use. In offering a definition, we are trying to make explicit their implicit semantic knowledge. But this view can’t be right, for reasons we’ve already seen: speakers often disagree about such definitions. If it were part of our semantic competence or implicit semantic knowledge that “manipulation” has a particular definition, why would so many definitions proliferate? Why would there be so much argument?

Another reason to think that this response is doomed is the incontrovertible fact that conventions differ across different communities of English speakers—there is significant semantic variation in what we call “English.” If different English speakers and subcommunities of English speakers often use the same word differently—maybe one subset uses an inclusive definition and the other uses an exclusive definition—how can we really claim that English has a single convention determinate enough to fix one intension as an expressions meaning? If we are going to talk about “the” conventions of English at all, as opposed to its various dialects or idiolects, we’ll have to admit that conventions admit some degree of indeterminacy to accommodate such manifest differences in patterns of use. And just to shore up this point: this is roughly what I found in doing empirical work on whether “see” has an intensional reading as well as an extensional one: there’s just significant disagreement within the relatively large samples I was working with (>300 English speakers) in each.

Another possibility is that people are thinking that expressions like “manipulation,” “lie,” etc. are natural kind expressions whose extensions are fixed with respect to some paradigm case, as with “water” or “tiger” or “gold.” A defender of this view might say that an act of manipulation is something that is relevantly similar to that act right there, where the act is some paradigm case of manipulation. If our use of the expression “manipulation” can somehow pick out the essential feature of the paradigm case it is introduced to describe, can’t that essential property provide the standard of correctness for analysis?

Here I think the answer is “no.” First, recall that paradigm cases of manipulation are included by basically all of the definitions we are considering. It is only once we encounter the fringe cases that we even realize that there are different concepts lying around that disagree on those cases. So on what grounds would we say that “manipulation” picks out one rather than another of the properties of the paradigm case? Why, for instance, would we think that “manipulation” picks out the inclusive, rather than the exclusive, concept? I think there simply aren’t any grounds to suppose that it does before we turn to debating the merits of including or excluding edge cases from our definitions. The choice between candidate definitions is typically pragmatic, and is to be judged by the theoretical fruits of each definition.

We might, however, want to distinguish between more and less natural properties of the paradigm case, and say that there is some reason to think that the term picks out the more natural of them—we might think that there is some kind of magnetism that draws our term to the most natural candidate property. But which is more natural, intentional manipulation or manipulation that need not be intentional? Frankly, I have no idea. One thought is inclusive definitions are often more disjunctive, and so less natural. But I don’t think the right conclusion is that inclusive definitions are always less natural than exclusive ones. I think there may be no general way of figuring out which properties our words attach to before going through the process of legislating or deciding on a particular definition for a particular purpose. Naturalness may be an additional factor we can appeal to in such decisions, but it won’t give us reason to think that there is one unique definition.

So what is the upshot of all this? Well one upshot is that a lot of counterexample-style disagreement is actually non-substantive. We’re just using different concepts, and there is an element of stipulation or decision involved when we define our terms. Another upshot is that a lot of agreement is also non-substantive. If you and I agree, but we each are employing slightly different concepts, then our agreement is spurious. Of course, that might not matter for many purposes, because we might interested in a case on which all of the admissible definitions agree. But sometimes it will matter.

Consider how many times people say that we need to “save democracy,” and you basically agree with them. But what are you agreeing to? You may have a guess about what they mean, but it might well be wrong. This is what Steffen Koch calls “Merely verbal agreement.” So the process of actually agreeing or disagreeing on a fully precise proposition is much more difficult than you might think.

Of course, there’s lots more to say here, but I think that the general picture I have sketched is more plausible than a view on which concepts are somehow fixed and determinate, and definitions are just attempts to discover them. And in case you’re concerned with a counterexample to what you otherwise thought was a plausible definition, hopefully this will give you something to say in response.

1. Just to be totally transparent, I’m here going to take “concepts” or “definitions” to just be intensions—functions from possible worlds to extensions, and something that could be specified by a necessitated biconditional. I know definitions are often taken to be finer-grained than this, but I am going to ignore that point for present purposes.

2. This also bears a close relation to what Peter Ludlow calls the “dynamic lexicon,” although I definitely don’t think that, even relative to context, we end up with perfect determinacy. [Edit: apparently in saying this, I mischaracterized Ludlow’s view. Actually, his view is that even relative to a context, words underdetermine their meanings. Given that, and given what Ludlow says in his book, you can see this post as basically endorsing his view, and considering the issue of conceptual analysis in its light.]

3. One way of thinking of definitions, on this model, is as what Carnap calls “explications.” An explication is a definition proposed for a particular philosophical or theoretical purpose, and it is to be judged by its theoretical fruits, rather than against our intuitions about whether it captures all and only what we intuitively take to be instances of the concept. Explications often end up at odds with our pre-theoretical or pre-philosophical understanding, but that is to be expected—explications are definitions made for a particular theoretical purpose.

But an alternative way of thinking of these definitions is as what are sometimes called in the vagueness literature “precisifications,” “delineations,” or “sharpenings.” In this case, the expressions of interest aren’t vague in the ordinary sense, but they can still be precisified or sharpened, but ordinarily, when they are used, they will be used unsharpened.