Not all dispositions are the same. Many classifications of dispositions have been proposed, based on a variety of dimensions. In this paper I focus on two dimensions:

Using two dimensions naturally creates a four-way distinction: universal dispositions whose bearer is an agent/causer, existential dispositions whose bearer is an agent/causer, universal dispositions whose bearer is not an agent/causer, and existential dispositions whose bearer is not an agent/causer. I will call this classification The Square of Disposition.

In this paper, I argue that the Square is expressed linguistically. Specifically, each of the four classes of disposition represented in the Square corresponds to a linguistic construction.

I should clarify that the goal of the paper is different from standard linguistic papers in that it does not contribute a new empirical discovery but rather aims to provide a conceptual groundwork and a novel perspective on previously unrelated work.

The distinctions plotted on the Square of Disposition are notoriously elusive to define precisely. It proved very hard to come up with clear, widely accepted criteria for being active, for example.

In this paper I want to suggest an alternative. While metaphysical distinctions such as active/passive are hard to make, in linguistics the corresponding distinctions are often much clearer, with useful tests. If we could map the metaphysical Square of Disposition onto a corresponding linguistic Square, we would get much clearer criteria. Of course, linguistic judgments are not always clear and uncontroversial, but often they are; and even when they are not, the judgments can, on many occasions, be corroborated using experimental methods or corpus studies.

This paper is concerned with the construction of such a linguistic Square; I will argue that each of the types of disposition in the Square corresponds to a specific linguistic construction.

But first, a natural question arises. Even if the project of this paper is successful, what is the significance of this fact? How could a linguistic distinction be relevant to a metaphysical one?

This very question is addressed by Bach (1986: 573). He writes: “Metaphysics… deals with questions like:… What kinds of things are there and how are they related?” But then he notes that “anyone who deals with the semantics of natural language is driven to ask questions that mimic those just given:… What kinds of things and relations among them does one need in order to exhibit the structure of meanings that natural languages seem to have?”

Of course, the metaphysical and linguistic questions, although parallel, are not the same. But is an answer to the linguistic question relevant to the metaphysical question? Bach (1986: 593) asks: “Do the fundamental distinctions that are reflected in the overt and covert categories of natural language correspond in any way to the structure of the world?” and he answers in the affirmative, for the following reason: “One of our main resources for coming to understand the world is, after all, language, a sort of tool box for doing whatever it is we want to do.”

In other words, while the distinctions drawn by natural language may not reflect the way things really are, they do correspond to some extent (though perhaps not perfectly) to our perception of them. And to that extent, the investigation of linguistic distinctions between different constructions is relevant for the investigation of the dispositions expressed by these constructions. In the rest of this paper we will attempt to discover this mapping between dispositions—as expressed by the Square—and linguistic constructions.

Let us begin with the newly coined term of passivilities: passive dispositions with an existential modal force.

A natural candidate for the linguistic expression of passivilities is the class of adjectives with the suffix -able or -ible in English: irritable, conceivable, washable, accessible, etc.

Clearly, the argument of these adjectives is not a causer, let alone an agent, hence they are appropriate as expressions of passive, rather than active, dispositions.8

What about their modal force? First, we must demonstrate that -able adjectives are modal to begin with. Consider the following example. Let us suppose there are many databases on some computer server. Some are accessible, but others are protected by a password; some are searchable, whereas others lack a search engine. Let us further suppose that, in point of fact, ever since the server had been placed online, every database that was accessed, was also searched, and vice versa. So, in this scenario, the property of being accessed and the property of being searched have the same extension. However, it still doesn’t follow that a database is accessible iff it is searchable. There might be a database that is accessible, although it has never, in fact, been accessed; but if it had been accessed, the user would have found that it is not searchable. Hence, although the properties of being accessed and being searched have the same extension, the properties of being accessible and being searchable do not. Hence, -able adjectives are modal, and are therefore appropriate as expressions of dispositions.

The modal force of -able adjectives is usually existential; they can, but do not have to, be realized: if something is accessible, it means that it can be accessed; if something is searchable, it means that it can be searched; if something is conceivable, it means that it can be conceived, etc. Indeed, regarding the corresponding suffixes in German, Kratzer (1981: 40) points out: “In general, the suffixes -lich and -bar express possibility.”

It has long been known, however, that there are exceptions to this generalization. Kratzer herself notes a clear exception—payable:

Kratzer notes: “Here, it is not that the twenty guilders can be paid, they definitely have to be paid on the first of January” (original emphasis). So, in this case it appears that -able can mean (deontic) necessity rather than possibility.

Other exceptions to the generalization that -able indicates possibility are not hard to find. If something is edible, it does not mean simply that it can be eaten; rather it means that it can be eaten safely; and if something is chewable, it does not mean merely that it can be chewed, but indicates that it is tasty (Fortin 2013).

But are these really exceptions? Fortin (2013) argues that the answer is no. He proposes that the basic sense of -able is possibility; all the additional senses are cases of conversational implicature.

One piece of evidence for his proposal is cancelability—the strengthened sense can be canceled:

Sentenc (7a) means that benefits may be paid, not that they have to be paid; in (7b), B affirms only the possibility sense of edible; and (7c) clearly canceles the tasty sense of chewable, though not its existential modal force.

Another argument involves reinforceability: implicature, but not entailment, can be reinforced without the sentence being odd. For example, (8a) is fine, whereas (8b) seems redundant and odd:

Now, note that the strengthened meaning of -able is reinforceable wthout oddness:

Of course, conversational implicature must be calculable: it must be shown to follow from Grice’s maxims. Fortin analyzes the implicature of x is payable by date d as follows. Its literal meaning is simply “x can be paid by date d.” By scalar implicature, we can conclude that “x can’t be paid by date d+1”, “x can’t be paid by date d+2”, etc. We therefore get the strengthened meaning, namely “x must be paid by date d.”

The literal meaning of edible is simply “can be eaten.” But Fortin points out that almost anything can be eaten. Hence, saying of something that it can be eaten is really giving very little information. This is a flouting of Quantity; hence, the strengthened meaning is conversationally implicated. A similar explanation is proposed for chewable.

An alternative explanation is provided by Maier (2016). He agrees that the basic meaning of -able adjectives is existential, and that their meaning is often strengthened, but he argues that strengthening is semantic rather than pragmatic. Specifically, it has to do with the ordering source according to which the expression is evaluated, which restricts the existential quantifier. Thus, for example, to say of something that it is edible means that there exist worlds where it is eaten, but these worlds have to be “normal”, in the sense that they are closest to the ideal: “Thus moldy bread, though there may be an accessible world where it is eaten, may not be not edible, for the worlds at which it is eaten depart from an ideal according to which all eatings are good in the sense of being, for example, tasty and nutritious” (Maier 2016: 450).

Both approaches can be formalized, but I choose Maier’s approach, where the strengthened meaning is part of the logical form.9 If g is the relevant ordering source, we can say that x is edible iff:

We can conclude, then, that -able adjectives are modal, their argument is not a causer, and they have an existential modal force. They are, therefore, the linguistic expressions that express passivilities.

Recall that capabilities are active dispositions with an existential modal force. I propose that they are expressed by dynamic modals of ability, such as the dynamic can:

Note first that dynamic can is modal. Suppose everybody who climbed the Everest learned Klingon in two months, and everybody who learned Klingon in two months climbed the Everest. So the properties of climbing the Everest and learning Klingon in two months have the same extension. Yet, clearly, they have different intensions: in different worlds, the set of individuals who climbed the Everest is not equal to the set of individuals who learned Klingon in two weeks. Indeed, dynamic can is sensitive to this difference in intensions: (12a) need not have the truth value as (12b).

Having satisfied ourselves that dynamic can is a modal, let us consider whether it behaves as a disposition. Recall that we have said that dispositions do not have to be manifested; indeed, dynamic can seems to fit the bill:

However, let us not be too hasty. Sometimes, dynamic can does require the occurrence of actual events, as the oddness of (14) indicates.

Is this a counterexample to the proposal that dynamic can expresses dispositions?

The answer is no. It has, in fact, been argued that some dispositions do require manifestation of actual instances demonstrating them:

In fact, dynamic can and its synonyms can even have an actual reading, describing a specific manifestation:

Thalberg (1972: 121) evaluates (15) in the following situation: “Before he hit the three bull’s-eyes, he fired 600 rounds, without coming close to the bull’s-eye; and his subsequent tries were equally wild… Therefore he does not have this sort of ability at target shooting.” He concludes: “‘Was able’ sometimes means ‘had the ability’ and sometimes means ‘did’.”11

However, it should be pointed out that was able to in (15) doesn’t mean merely ‘did’; the sentence implies that hitting three bull’s-eyes in a row is remarkable, difficult, or otherwise unexpected. From this we can conclude that even actual readings of ability modals have a modal flavor: (15) says that the actual world, where Brown hit three bull’s-eyes in a row, deviates from what one would expect.

Indeed, such readings pass the test for modality. Suppose everybody who climbed the Everest wore purple socks and a yellow hat, and everybody who wore purple socks and a yellow hat climbed the Everest. So the properties of wearing purple socks and a yellow hat and having climbed the Everest have the same extension, though of course different intensions. Now, suppose (16a) is true. This does not make (16b) true, since although climbing the Everest is quite a remarkable accomplishment, wearing purple socks and a yellow hat is not.

An important property of dynamic can and its synonyms is that they are subject-oriented. Brennan (1993) proposes that dynamic can combines with the VP, resulting in a modal property predicated of the subject. For example:

Sentence (17) does not mean merely that it is possible that Mary will swim; rather, it says that the subject (Mary) has the property of being capable of swimming. One way to see this is to assume that Mary doesn’t know how to swim. It is still possible that, purely by chance, she will perform the movements appropriate for swimming; yet (17) is nonetheless false. But if Mary does know how to swim, intends to swim, the conditions are suitable for swimming, and there is no hindrance, the truth of (17) implies that Mary will, in fact, swim.

Let me briefly discuss two pieces of evidence presented by Brennan as part of her argument that these modals are subject-oriented. One is the in virtue of test. This expression can only attach to dynamic modals, as in (18a) but not, say, epistemic modals, as indicated by the unacceptability of (18b).

Now, note that in virtue of is only fine when it describes properties of the subject, as in (18a), but not the object, as in (19).

Note that the problem is not that in virtue of cannot describe a property of the object, because, in general, it can:

Hence, the in virtue of test indicates that dynamic can is, indeed, subject-oriented.

The second piece of evidence produced by Brennan involves expletives. Brennan notes that, when a sentence contains an epistemic modal, the subject can be replaced by a semantically empty expletive:

The same holds for deontic modals:

However, this is impossible in the case of dynamic can:

It must be admitted, however, that there appear to be counterexamples—sentences where it looks like can does not predicate a capability of the subject. For example (24a) predicates of the subject (John) the capability to lift that big stone; but no capability is predicated of the subject in (24b).

Why, then, is (24b) acceptable?

Tao (2011: 68) proposes: “While [(24a)] expresses John’s ability [(24b)] asserts some objective possibility of John’s lifting that stone.” In other words, can is reinterpreted as a circumstantial modal.

The following examples probably makes the point more clearly:

Suppose John knows nothing about cracking safes; he merely tries codes at random. Of course, there is a possibility that he will guess the right code by chance. In this case, (25a) is false but (25b) is true. The explanation is that whereas the former ascribes to John the capability to open this safe, the latter merely expresses a circumstantial possibility that John will open the safe.

However, treating can as subject-oriented is not quite accurate; rather, it applies to the external argument. The external argument is usually the subject, but not always. Unaccusatives are a case in point: their subjects are not external arguments. And, indeed, capability can is incompatible with unaccusatives:

Sentence (26) can only receive a non-capability interpretation, perhaps a deontic reading (“Terry is permitted to arrive late”).

It should be noted that the external argument is not merely a formal syntactic notion, but is associated with semantic notions of agentivity and causality (Marantz 1984). Therefore, dynamic can is a good candidate for representing capabilities: existential dispositions that require an agent/causer.

If can indicates capabilities, it cannot indicate dispositions that do not relate to actions. We would therefore expect it to be bad, in general, with stative verbs. Indeed, can appears to be bad with the stative know, compared with the non-stative speak:

It should be noted that Maier (2014) argues that can can be used to indicate powers that are not abilities, i.e. that do not have to relate to actions. As an example, he presents (28), and argues that “the power to understand French will be a power, but not an ability.”

Sentence (28) is, indeed, perfectly acceptable; and understand is, indeed, normally a stative verb. However, this sentence does not constitute a counterexample, since understand can also be used non-statively—for example, it can be used in the progressive (Comrie 1976: 36):

Sentence (28), then, is not a counterexample, but is just another case where understand is used non-statively.13

In formalizing dynamic can, Brennan makes the ordering source dependent on the properties of an individual, which she takes to be the subject, but we can take it more generally to be the external argument. Formally, putting aside irrelevant details, if g(m) is the ordering source dependent on the properties of Mary (17) would be true iff

First, note that middles are modal. Suppose every book that was read easily was also translated easily, and vice versa. So the properties of being read easily and being translated easilty would have the same extension, though different intensions. Now, suppose a new book has just been published, and has not yet been read (or translated). Then (31a) would not entail (31b).

Now note further that middles have a necessity flavor. Sentence (32) implies that, whenever this bread is cut, it is cut easily, not that it is sometimes cut easily, and sometimes with difficulty.14

Indeed, it has been suggested (Condoravdi 1989; Ackema & Schoorlemmer 1994; Lekakou 2005) that middles involve a (quasi) universal modal operator. Middles are therefore appropriate for the expression of universal dispositions.

As an aside, it should be noted that, although middles and -able adjectives are sometimes treated together (e.g., Gràcia 1992), this approach is dubious, given that the former are universal whereas the latter, as we have seen, are existential. Curiously, even studies that perform thorough comparisons between middles and -able adjectives ignore the issue of their respective quantificational forces. Indeed, in her thorough investigation of the issue, Oltra-Massuet (2014: 102) candidly admits: “Whether this modal operator must be understood as existentially quantifying over possible worlds or as universally quantifying over situations in -ble remains an open question and need not concern us here.”15

Lekakou (2005) argues that middles do not ascribe a dispositional property to the external argument of the verb, but rather to the internal one. Furthermore, there are variants of middles where the property is ascribed to an adjunct, or even an expletive pronoun, as exemplified by the following Dutch examples (Ackema & Schoorlemmer 1994; 2006; I am indebted to an anonymous reviewer for drawing my attention to this phenomenon and providing me with these examples):

Hence, middles express passive rather than active universal dispositions.

Just as a demonstration, glossing over important but irrelevant details, and assuming a rather simplistic representation of manner adverbs, we can represent middles as quantification over normal events. For example, we can say that (32) is true in world w iff:

Putting this in words, this means that, in normal worlds, all events of cutting this bread are easy.

-Er nominals are words like smoker, jogger, liar, etc. A necessary condition for them to express dispositions is that they be modal, and indeed they are. Specifically, I will show that they quantify over worlds that maintain essential properties of the actual world, i.e. normal worlds.

To see this, it will be useful to go back to the examples from section 2.2 demonstrating that habituals are not modal, to show that -er nominals are.

Let us go back to Sally, whose birthday is December 25th. Since Sally’s birthday is not an essential property of Sally, there are possible worlds that preserve essential properties of the actual world, in which Sally’s birthday is not on Christmas. We can see that these worlds are considered when an -er nominal is evaluated, i.e. they can affect truth conditions.

For suppose Sally drinks on her birthday (and only then). Then (35a) would be true; however (35b) would not follow—Sally may not celebrate Christmas; she may not even be aware that her birthday falls on Christmas.

For another example, consider (36a), and contrast it with the habitual (36b).

Suppose, as before, that Sally hates cigarettes, but she is constantly harassed by a militant smoker who forces her to smoke at gunpoint. Consequently, she smokes several cigarettes every day. In this situation (36a) is false, although (36b) is true.

Now suppose Sally is very fond of cigarettes, but she is confined to a prison where no smoking is allowed, and never leaves it. So Sally doesn’t actually smoke, and therefore (36b) is false, yet (36a) is true.

To give another example, suppose all animals except dogs suddenly died in the actual world. But they continue to live in other worlds, which can be as normal as you like. Then (37a) and (37b) would have the same truth value, since all the animals are dogs; however, (38a) can still be true without (38b) being true, since one can be certified to be a dog trainer without being certified to be an animal trainer.

Since -er nominals are modal, they satisfy the necessary condition for expressing dispositions.

The disposition expressed by -er nominals sometimes does and sometimes doesn’t have to be manifested. Rappaport-Hovav & Levin (1992) draw a distinction between two types of -er nominals: eventive (e.g., saver of lives) and non-eventive (e.g., lifesaver). In eventive -er nominals, the disposition has to be manifested, as exemplified by the unaccepatbility of (39a); in contrast, in non-eventive -er nominals the disposition does not have to be manifested, as exemplified by the acceptability of (39b).

Both eventive and non-eventive readings of -er nominals are dispositional; but -er nominals also have non-dispositional readings:

Sentence (40a) has a dispositional reading: what John does is murder people. However, it also has a non-dispositional reading, the satisfaction of which requires only the existence in the actual world of one event of a murder committed by John. Sentence (40b) can have a dispositional reading, referring to a person whose profession is to sing, but also a non-dispositional reading, referring to a person who was singing at the event time in the actual world. And since the dispositional reading of (40c) is ruled out pragmatically, it only has a non-dispositional reading: there is in the (immediate) future of the actual world an event of John’s being gone.

But note that even these non-dispositional expressions are modal, although not dispositional.16 Suppose there are exactly three people who murdered someone; and suppose they are about to be executed, and, on their way to the gallows, they are singing, and nobody else is singing. Then, at that moment, the (non-dipositional) properties of being murderers and being singers have the same extension. Yet, although (41a) entails (41b), (42a) does not entail (42b).

What kind of disposition do -er nominals express (under their dispositional readings)? Many researchers (Fabb 1984; Keyser & Roeper 1984; Burzio 1986; Rappaport-Hovav & Levin 1992; van Hout & Roeper 1998; Cohen 2016a) note the External Argument Generalization: -er indicates that the argument of the nominal receives the thematic role that the verb assigns to its external argument.

As we have noted above, the subject is often the external argument, but not always, and the relevant case is that of unaccusatives. Indeed, -er nominals normally are not formed from unaccusative verbs:

It must be conceded that sometimes such examples are attested, but they do not constitute counterexamples to this generalization. Indeed, Schäfer (2010) notes:

Hence, we can conclude that -er nominals express active dispositions.

-Er nominals have a universal rather than existential flavor, as demonstrated by the following example. By Jewish custom, every baby boy is given a few drops of wine before his circumcision. So, for every Jewish man, in every world (compatible with the custom), there is an event of this man drinking wine. However, this doesn’t make him a wine drinker.

One step further is taken by von Fintel & Heim (1999), who identify the universally-flavored quantifier associated with -er nominals with the generic quantifier.17 They note that, under one reading,

When von Fintel and Heim refer to the generic quantifier, they assume universal quantification over normal worlds. Although, as we saw in section 2.2, this characterization of habituals and generics is quite problematic, the identification of the quantificational force of -er nominals with a universal quantifier over normal worlds is quite convincing. Hence, we can conclude that -er nominals are modal, they require an agent/causer, and have a quasi-universal flavor. Hence, they express active universal dispositions.

In formalizing the meaning of -er nominals, I will adopt Brennan’s ordering source, which is depndent on properties of the external argument. Assuming the same simplistic view of manner adverbs as before, we can therefore say that (44a) is true iff (44b).

Dynamic can is usually considered together with other dynamic modals, primarily dispositional will. Dispositional will does, indeed, seem appropriate for the expression of dispositions, since it usually does not require the occurrence of actual events, as indicated by (45a). However, note that, as can be seen by the unacceptability of (45b), sometimes it does.

As we have seen, a necessary condition for a linguistic construction to express dispositions is modality. Indeed, dispositional will satisfies this requirement. Suppose everyone who ate cardboard wore purple socks and yellow hats, and everybody who wore purple socks and yellow hats ate cardboard. Then the two properties would have the same extension; yet (45a) does not entail (46).

The modal force expressed by dispositional will is universal: (47) says that under normal conditions, whenever sugar is put into water, it dissolves.

Example (47) is interesting for another reason. Note that the verb dissolve is an unaccusative: its subject is not the external argument. Hence, the argument of dispositional will may be the external argument (as in, e.g., (45a)), but does not have to be. Hence, dispositional will can express both types of universal disposition: active and passive.

We can now fill in the Square of Disposition, for each type of disposition, the linguistic expression that is appropriate for representing it.:

Before concluding, I have a confession to make. In this paper I first constructed the Square of Disposition, and then tried to fit linguistic expressions into it. However, in the research leading to the paper, I actually worked backwards: I looked at linguistic expressions, and formed the Square so as to fit them in.

What is the significance of this fact? There are many ways to classify dispositions—the Square is only one option. The debate over the correct classification of dispositions will no doubt go on. The debate is, and should remain, a philosophical one, following philosophical considerations.

What I have tried to show is that there is one particular classification that corresponds to what is expressible in natural language. Following Bach’s (1986) program of natural language metaphysics, we can go one step further and say that this metaphysical classification is motivated by a linguistic classification. This, of course, doesn’t mean that this particular classification is metaphysically correct: the deciding arguments for or against it will still be philosophical, not linguistic, in nature. But, echoing the words of Vetter (2014: 131): “I am certainly not advocating that philosophy be replaced by linguistics… but I do take these data to provide some motivation for an alternative approach.”