Volitional modality in the double-modal construction in Southern US English

Volitional modality in the double-modal construction in Southern US English Cynthia Kilpatrick and Chris Barker, UCSD 1. Introduction: double modals Dialects of English in the American South are notorious for sometimes allowing two modal auxiliaries: (1) We might should meet at 9:30 so we can be done by noon. (2) If I had known, I might would have bought it myself. (3) You might could set it out in the sun and give it lots of water. In (1), the idea is put forth that if a class wanted to, meeting early would ensure that they get out at a specified time. In (2), speaking of a piece of furniture that she did not realize was for sale, a Texan uses might would to say that if she had known the furniture was for sale, she might have made the decision to purchase it. In (3), a suggestion is made to a friend that, if she chose to, she could try to revive a dead plant by putting it in the sun and watering it. Despite frequent mention in the literature, as well as in the verbal lore of the field, the semantics of this construction has never been adequately described, as far as we know. This paper will propose an explicit analysis in terms of Kratzer s (1981, 1991) analysis of modal expressions as supplying a modal base and an ordering source. In brief, we will argue that the second (lower) modal typically establishes an epstemic or circumstantial modal base, and the first (upper) modal contributes a bouletic ordering source that reflects the desires and intentions of some agent involved in the described event. As a result, the meaning expressed by (1) is more subtle and more interesting than the mere conjunction of its two constituent modals: (4) a. We might meet at 9:30 so we can be done by noon. b. We should meet at 9:30 so we can be done by noon. Though these sentences are consistent with the proposition expressed by (1), (1) suggests in addition that the participants may or may not want to meet then, and the choice to do so is up to them. In section 2, we outline the data, showing how double modals differ systematically from single modals. In section 3, we apply Kratzer s (1981, 1991) standard analysis of single modals to the double-modal construction. Section 4 develops a number of predictions of this analysis, illustrated 1

with relevant data and heretofore unnoted restrictions on the use of double modals. Section 5 discusses a conversational implicature typical of double modals, and explains how this implicature can shed light on previously noted but mysterious syntactic idiosyncrasies. 2. Modality 2 2.1. Preliminaries: crucial distinctions among types of modal meanings In order to understand the meaning distinctions in the double-modal construction in southern English, it is necessary to first understand what single modals do. Modals, in general, deal with the attitude of a speaker toward a proposition and are typically characterized as either epistemic or root, with root being a catchall term for anything not epistemic. Epistemic modals deal with a speaker s judgment of possibility, probability, or necessity based on what is known. The examples in (5) illustrate an epistemic use of modals: (5) a. The students must be finished by now. (They ve been working for hours.) b. The students might be tired. (They worked late finishing their papers.) In (5a) the modal must reflects epistemic necessity: the only logical conclusion one can come to, based on what is known at that time, is that the students are finished. Epistemic possibility is expressed in (5b), where might indicates that it is possible but not definite that the students are tired. In each of these, the speaker is making a claim as to what he believes may or may not be the case given his current knowledge of the the world. While epistemic modals deal with a speaker s judgment of possibilities based on evidence, root modals encompass a variety of different situations, the most common of which are deontic and dynamic. Deontic modals typically deal with permission or obligation, as shown in (6), while dynamic modals deal with ability or disposition, as in (7). These modals state necessities or possibilities implied by certain facts: (6) a. The students must finish by 3:30 or their papers won t be accepted. b. The students may register on Friday if their last name begins with A M. Otherwise, they have to wait until next week. In (6a), must is used to express obligation: if the students want their papers to be accepted, they have no choice other than to turn them in by 3:30. (6b), on the other hand, deals with options, where may does not state an epistemic possibility, but instead gives permission (possibility with respect to what is allowed). As a comparison of (5a) and (6a) reveal, in one context a modal

may express an epistemic thought, and in other contexts express a deontic one. (7) a. You must be in bed by 10. (von Fintel 2005) b. Maybe I will take a bath. (Copley 2004) The examples in (7) illustrate bouletic modality, which will be crucial to our explanation below. Bouletic modality is a type of non-root modality, somewhat similar to deontic modality, but specifically concerning the desires and wishes of some relevant agent: (7a) describes what it would take to satisfy the desires of a stern father, and (7b) offers one possible action consistent with the desires of the speaker. 2.2 Previous assumptions regarding double modals Double modals have been characterized in the literature as being epistemic. We will show that this assumption is not adequate. The bulk of the literature on double modals has concentrated on the syntax of the construction, but has assumed an epistemic meaning with little explanation. Shay (1981) differentiates between epistemic (knowing and believing) and volitional (desiring and willing) modals, but concludes that the double modal construction is epistemic. De la Cruz (1995) claims that the meaning of the first modal in an ordinary two-modal construction is normally to be taken as having an epistemic sense. Neither offers additional explanation or evidence as to the meaning of these constructions. King (1994) claims that the subjunctive use of could has been lost in Southern English, and the use of a double modal construction provides the subjunctive sense to an otherwise indicative expression. Regardless of whether or not his characterization is true, it is simply another way of saying what the others have said: double modal constructions add a greater degree of emphasis to the uncertainty that already exists from the use of the lower modal. While this may describe some general idea behind what these constructions mean, the focus of the above-mentioned work has been syntactic rather than semantic, and the meaning of the construction has been assumed to be somewhat transparent. Only a few investigators have even address the question of what these constructions mean. Bernstein (2003) briefly addresses the issue of semantics, pointing out that the dynamic modality of double modals expresses ability or volition, and implies that the double modal expression is a combination of willingness, ability, and uncertainty. However, she goes no further than the mention of semantics, leaving the details still to be dealt with. DiPaolo (1986, 1989) assumes a ordinary epistemic sense for might, but claims that the meaning of the double modal is entirely determined by the 3

lower modal, rather than by might, and that the construction most often carries an ability sense or a deontic (permissive) sense. Even at that, the semantics of the construction is glossed over rather than investigated fully. Certainly modal-doublers readily acknowledge that there is a clear meaning distinction between double modals and their single modal counterparts. And they do view these constructions just as Bernstein (2003) does: as expressing a combination of willingness, ability, and uncertainty, used in specific contexts in which politeness or negotiation is at stake. However, we will show that the meaning of a double-modal construction cannot be equated with any interpretation of the lower modal alone, nor can it be paraphrased by the meaning of any single modal. Rather, each modal makes its own characterisitic and irreducible contribution to the meaning of the construction as a whole. 2.3 Problems with the assumptions about double modals Though the literature assumes that the double modal construction expresses epistemic possibility, with simply an extra layer of possibility added, native speakers of dialects that allow double modals make a clear distinction between these constructions and simple epistemic constructions. If the double modal were simply an epistemic construction with an extra layer of possibility added, then these constructions should be both synonymous with epistemic constructions, and acceptable in the same instances that simple epistemic constructions are. Consider first the question of meaning. 4 (8) a. Maybe I should write my squib now. b. I might should write my squib now. Modal doublers, however, do not find these two sentences to be synonymous. If the use of a double modal is a simple case of an additional level of possibility, creating an epistemic sense, this result is unexpected. The basic difference between (8a) and (8b) above is the degree to which the speaker is actually willing to engage in the activity. A paraphrase of (8a) might be It s possible that I should write my squib now, which might be used to question whether or not it is really necessary to write the squib now. However, the same paraphrase is not appropriate for (8b). Instead of questioning the necessity of writing the squib now, this construction accepts that the squib should be written now, but questions the speaker s willingness to actually do it. An appropriate paraphrase might then be I should write my squib now, but I m not sure I want to (and I m certainly not willing to commit one way or the other just now).

In addition to not being synonymous with simple epistemic constructions, double modals also cannot appear in all the instances that single epistemic modals do, nor do they appear in the same instances as root modals. For instance, consider the examples given in (9), in which the epistemic modal in (a) is perfectly acceptable but the double modal in (b) infelicitous: (9) a. It might be possible to do it later. b. #It might could be possible to do it later. If this construction were simply epistemic, (9b) should be acceptable, but modal doublers do not consider this an acceptable construction. (As we will show, the problem here is the lack of a volitional participant whose bouletic desires are at stake.) The double modal, while it may have some epistemic sense to it, is not the same as other epistemic constructions. Not only is the double modal not simply epistemic, it is also not simply deontic or dynamic. This can be seen in instances where a single deontic or dynamic modal is acceptable, but a double modal is not. For instance, dynamic modals such as could and can are easily embeddable under verbs like hope and wish, but infelicity results when double modals are embedded under these predicates: (10) a. I wish I could do that. b. #I wish I might could do that. (11) a. I hope I can go. b. #I hope I might can go. Again, in instances where single modals are okay, double modals are not. (The problem here is that the speaker is trying to express a desire for the speaker herself to be undecided about going, which is a deviant desire.) Simply put, while these constructions may share some characteristics with single modals, there are some clear differences between both the meaning and the uses of single and double modals. Native intuitions also point to a clear meaning distinction between single and double modals. As noted, southerners who double modals distinguish between single and double modals. Native southerners are quick to recognize when a non-native is attempting to use the construction, but uses it in a situation in which a native doubler would not. Bernstein (2000) points out instances where novelist Richard North Patterson misrepresents the way that Southerners use double modals, citing the following examples: (12) a. Might could be rape. [Patterson 1981:8] b. Might could be easy for you to check into a motel. [Patterson 1981:144] Finding both of these examples to be inauthentic ones, a judgment supported by native Alabamians at Auburn University, Bernstein suggests that 5

the reason is in the degree of willingness expressed and to whom it is attributed. She compares these to an authentic double modal construction, which conveys uncertainty regarding a person s intentions. Unlike the felicitous double modal construction, the examples given in (12) instead express uncertainty as to the events surrounding a murder (12a) and probability regarding a suspect s actions (12b). Neither of these contexts are ones in which native Southerners would naturally use a double modal. It is clear, then, that the double modal construction is used in only a subset of instances in which a single modal would be acceptable. In addition, it seems to carry with it additional meaning distinctions that are not always evident in the use of a single modal. Just what these distinctions are and how they can be formally modeled within a formal model of modality is the issue we address here. Bernstein s (2003) description provides just the distinction that modal doublers find instinctively to be true: these constructions express a combination of willingness, ability, and uncertainty. We propose that these three aspects of the meaning of the double modal are due to the following semantic restrictions: The willingness aspect is due to a presupposition that there exists a volitional agent whose desires will dictate whether or not an action is carried out; the ability aspect is reflected in the entailment of the truth of the lower modal, which is crucially dependent on circumstances; and the uncertainty is derived from restrictions on when evaluation of context worlds takes place. We will briefly examine the willingness and ability aspects, and show how Kratzer s (1981, 1991) formalism can be adapted to accurately reflect the truth of a double modal construction. We will then turn to the uncertainty aspect and discuss how it is derived from restrictions on when evaluation takes place. 2.4 Willingness: presupposition of a volitional agent The most salient feature of a double modal construction is that it seems to express a degree of willingness, or unwillingness, as the case may be, on the part of some relevant agent. This is most clearly seen in the kinds of verbs that can appear as the main verb in a modal construction, as well as in the way that double modals are assumed to have an agent at hand to effect an action. The idea that double modals presuppose the existence of an agent who can control the action is supported by the restrictions evident on the main verb of a modal clause. Without exception, this verb must express a controllable action. For example, with verbs like love, hate, or realize, over which one typically has no control, a double modal is unacceptable: 6

7 (13) a. *I might could hate that movie. b. *You might could love the view from here. c. *She might could realize that her husband is cheating on her. In each of the examples in (13), the use of a double modal with a verb over which an agent has no control is simply ungrammatical. While the agent may often be the subject in a double modal construction, this is not a necessary requirement. As long as it is possible that the action can be controlled by some agent who makes the decision, the double modal is acceptable. It is not even necessary that the agent be present in the discourse. For instance, in a passive construction, the agent may never even be mentioned, but is in some way accessible or assumed to exist. Consider (14): (14) Mike Fuller and Rick Enos might could be repackaged as a monster tag team. In this example, a blogger puts forth the idea that two wrestlers could be repackaged as a tag team, but indicates that the speaker is unsure if the powers that be would think this a desirable option. Even though these wrestlers managers are never mentioned, or even referred to in any way, the use of the double modal construction presupposes that they exist and hold some decision-making power. In addition, when native Southerners are presented with a double modal construction in which no agent is evident, they still understand it as a construction in which an agent has control over an action. For instance, consider (15): (15) That book might should be blue. One modal doubler, when questioned as to the acceptability of (15), thought about it for a moment before finally responding that she just couldn t imagine why anyone would ever want to change the color of a book. The use of the double modal indicated to her that there was some volitional agent that was considering effecting the change in color. Even as a context-free utterance, the existence of an agent in (15) was presupposed, and without that agent, infelicity would arise. 2.5 Ability: entailed circumstantial modality In addition to presupposing the existence of a volitional agent, the double modal construction also entails the truth of the lower modal; that is, the truth of the sentence if the upper modal were removed. So for instance, the example in (16a) entails that (16b) is true:

(16) a. I might should read that article before going to class. b. I should read that article before going to class. In (17a), the construction does not question whether or not the article should be read, but rather accepts that truth, questioning only if the agent actually wants to read the article. This is a constant across double modal constructions: the proposition corresponding to the lower modal alone is entailed as true. Nevertheless, though its truth is entailed, the lower position is restricted as to the types of modals that can appear in it. This modal always has a root sense, and more specifically, a dynamic sense to it.4 That is, it is dependent on, as Brennan (1993) defines dynamic modality, abilities, susceptibilities, and dispositions. In no case is this modal ever epistemic, nor truly deontic in the sense of permission. Instead, the lower modal is always dependent on facts or circumstances in the world that create necessities or possibilities. Consider, for instance, that an epistemic modal expressing logical necessity is unacceptable as the lower modal of a double modal construction: (17) a. Josh must/should be here by now: he left home an hour ago. b. #Josh might must be here by now: he left home an hour ago. c. #Josh might should be here by now: he left home an hour ago. In (17b) and (c), the lower modal is not expressing a circumstantial modality, but instead an epistemic one: based on the evidence, the only logical conclusion is that Josh is already here. The speaker is making an inference based on what he knows about the world. This results in an infelicitous construction. Notice however, that a similar construction with a non-epistemic modal is fine: (18) Josh might should be here by seven, or we ll be late for dinner. In (18), the use of should as the lower modal cannot be an epistemic judgment of what the world must be like based on what the speaker knows (that there is evidence Josh will be here by seven), but instead deals with an obligation placed on John as a result of known circumstances in combination with some notion of how the future ought best to unfold. Similarly, an attempt to interpret a lower modal as expressing epistemic possibility is infelicitous for analogous reasons: (19) #Seena might could be sleepy; she didn t take a very long nap. These unacceptable readings of epistemic lower modals indicate that the lower modal is limited to a non-epistemic reading. Nor can the lower modal express permissive modality. Consider the following example which may at first seem to express permission, if, for instance, it were spoken by a boss to his secretary: 8

(20) You might could go down to Office Depot and find a stapler there. However, native intuitions do not accept this as an expression of permission, but instead view it as one of suggestion. The boss is simply saying that the circumstances are such that staplers are sold at Office Depot, and if she wanted to, the secretary would be able to get a stapler there. It is not a statement that gives her permission so much as one that allows for the possibility that maybe the secretary would not want to go to Office Depot for some unstated reason. Even statements like that in (21) are not understood as permissive so much as circumstantial: (21) You might can go to the movies if you get all your homework done. Even spoken by a parent to a child, the lower modal is not one of permission. Rather, it expresses that the possibility exists: there is no relevant circumstance that would preclude the possibility of going to the movies if the homework gets done. In general, the lower modal does not express a speaker s judgment regarding epistemic possibility or necessity, but instead makes a statement about some relevant facts and circumstances about the world. Following Kratzer, we will refer to this as circumstantial modality, encompassing both dynamic and deontic necessity modals when the necessity is due to facts and circumstances. The fact that the lower modal in a double modal construction is limited to a circumstantial reading may help explain some of the facts in relation to what modal combinations are possible. As has been widely noted (DiPaolo 1986, 1989, Feagin 1979, Boertien 1986, among many others), the higher member of a modal pair is either might or may, while the lower can be one of several choices, as shown in (22): (22) Higher modal Lower modal may might *must *shall *will *can *could *would *should *may *might *must *shall?will?can could would should This is, for the most part, a case of complementary distribution. The modals may and might, which are the only options for the higher modal position, are never possible as the lower modal. Meanwhile, the modals could, would, 9

should, and, with slightly less acceptability, will and can, appear as the lower modal, but never as the higher one. The modals must and shall simply never participate in doubled constructions. These facts find a reasonable explanation in light of the requirement that the lower modal be circumstantial. May and might, acceptable in the higher position, do not appear as the lower modal. However, it should be noted that these two modals are primarily used with an epistemic sense8 (Coates 1983, Palmer 1988) and Palmer (1988) claims that clear cases of dynamic modality do not use may even in single modal constructions, while dynamic cases of could refer primarily to abilities in the past tense. So might and may, while acceptable in the higher position, are unacceptable in the lower position, where a circumstantial modal is required. This requirement that the lower modal be circumstantial seems to have a hand in controlling the ability of modals to occur in the lower position as well. For instance, must is used most often as either an epistemic modal indicating logical necessity, or as a deontic modal, indicating an obligation or necessity. Without the possibility of a circumstantial reading, must is not permitted as the lower modal of a double modal construction. The same is true of shall, which, lacking a circumstantial reading, does not appear in double modals. Other modals, such as could and can, for which circumstantial or dynamic readings are possible, appear commonly as the lower modal in these constructions. Consider, though, the case of will and would, both of which can and are often used in future senses, and which may seem to present a problem for the claim of the lower modal being circumstantial. However, both of these have a dynamic reading in which they denote a person s disposition, i.e., they can have circumstantial interpretations. In double modal constructions, rather than making a statement about the future, they simply state that a person is disposed to do something, or at least consider it. Modal doublers are clear that a might would construction is not one that expresses unmistakable intention, but rather one that is even more indecisive than the typical double modal, but still related to someone s desires. They create a situation in which, in essence, someone is expressing their willingness to do something, but not their intention to do it. The double modal construction, then, is characterized by a combination of willingness and ability, which seem to correspond to the presupposition of a volitional agent and the entailment of the truth of the lower modal, circumstantial, modal. These two conditions must be accounted for in any semantic analysis of double modals. 10

11 3. A Kratzerian analysis Having seen the variety of modal combinations and established that the double modal differs from a simple epistemic or deontic modal, I now turn to the question of how modality is treated within Kratzer s formal model. The standard treatment of modality is in terms of possible worlds under an accessibility relation, established by Kratzer (1981, 1991) as what she terms a doubly relative account of modality. She introduces the notion of graded modality, which is achieved through the use of a modal base and an ordering source, both of which are relevant conversational backgrounds. The modal base determines the set of accessible worlds, while the ordering source imposes an ordering on all the accessible worlds, determining which worlds are closest to some relevant ideal. I first examine how this model operates, and then extend it to an account of double modals. 3.1 Graded modality According to Kratzer s (1991) model, modal words operate on the interaction of a modal base and an ordering source. The modal base, characterized by either what we know and believe (epistemic) or by what the relevant facts and circumstances are (circumstantial), supplies the set of worlds accessible from any given evaluation world. Then the ordering source imposes an order as to which worlds are the closest to the ideal world, based on such considerations as the law, a person s beliefs, or a person s wishes. Modal interepretations differ in their selection of a modal base, of an ordering source, and in the requirements they place on the optimal worlds picked out by the ordering source (i.e., whether the proposition in question is true in some optimal world, all optimal worlds, and so on). Kratzer builds on Stalnaker s conception of a proposition as a set of possible worlds. Update of the common ground with the information contributed by a sentence will eliminate from further consideration those worlds in which the proposition expressed by the sentence fails to hole. The modal base is a function from a world to a set of propositions. An epistemic modal base, for instance, maps each world onto the set of propositions that are known in that world. Since propositions are themselves sets of worlds, that means a modal base is a function from a world to a set of sets of worlds. From the modal base we can calculate the set of worlds that are accessible from that world: they will be those worlds that satisfy all of the propositions associated with the world in question. So given a modal base f, the worlds accessible from a world w will be all those worlds v such that v f(w), that is, the intersection of the propositions in f(w). Following Kratzer, we

will only consider here modal bases that are consistent and realistic: since epistemic modal bases, and likewise circumstantial modal bases, deal in what is known, and since only true propositions can be known, each world will always be accessible from itself. That is, each world will necessarily be compatible with what is known in that world. The other conversational background is the ordering source. The ordering source imposes a partial ordering as follows (Kratzer 1991: 644): (23) Let the conversational background f map a world w to a set of propositions A = f(w). Then A induces an ordering over the set of worlds as follows: for all worlds x and y, x A y iff for all p A, p(y) p(x). Here, A is the ideal, represented by a set of propositions that ought to be true (according to some flavor of ought ). Those worlds in which more of the propositions in A hold are worlds closer to the ideal. So for any A, x is at least as close to the ideal as y just in case x satisfies at least as many A-propositions as y does. Kratzer uses the ordering induced by a conversational background to define a notion she calls human necessity: (24) A proposition p is a necessity in a world w with respect to a modal base f and an ordering source g iff For all u f(w) there is a v f(w) such that v g(w) u; and For all z f(w): if z g(w) v, then z p. (Kratzer 1991: 644) In words, a proposition p is a necessity just in case no matter which world w you start with, moving closer to the ideal always leads you to worlds in p. [poss dual of necc] Consider how this modal base and ordering source operates given a statement such as (25): (25) John should buy some eggs. First consider an epistemic modal base, assuming that among the established facts are the following: (26) a. John has no eggs at home b. Eggs are necessary in order to make quiche Then for each possible world w in the evaluation context, f(w) (the set of worlds that f makes accessible from w) will contain only worlds in which eggs are needed to make quiche, and John has no eggs. Let us now consider a bouletic ordering source, so that for each world w, g returns the set of propositions that characterize John s desires in w. What 12

John desires may differ somewhat from world to world; let s imagine that in each world under consideration in the evaluation context, John has a desire to make quiche, though perhaps he wants to finish at slightly different times in each world. Consider a specific world w in the evaluation context. Then f(w), the worlds that are epistemically accessible from w, is a set of worlds in which John has no eggs, but eggs are needed to make quiche. Then we can divide this set of worlds into four quadrants: (27) J buys eggs J doesn t buy eggs J doesn t make quiche I: w 1, w 2 II: w 3, w 4 J does make quiche III: w 5, w 6 IV: There are no worlds in quadrant IV, because there are no epistemically accessible worlds in which it is possible to make quiche without eggs. It is easy to see that the worlds in quadrant III are all more ideal than any of the worlds in quadrants I or II, since in the worlds in quadrant III, John fulfills his desire to make quiche. Then should(john buy eggs) will be true at w just in case it is a necessity at w that John buy eggs (given f and g). This is the case, since the most ideal worlds (the ones in quadrant III) are all worlds in which John buys eggs. More technically, for every choice in {w 1,..., w 6 }, we can find a world (say, w 5 ) that is at least as ideal, and that no world exceeds in closeness to the ideal. While Kratzer (1981, 1991) deals only with the use of a modal base and an ordering source for a single modal, we propose that these same tools can be used to express the meaning of the double modal construction currently in use in Southern US English. However, this will require not a single modal base and ordering source, but a modal base and ordering source for each of the modals. Crucially, the modal base of the lower modal will be circumstantial, while the ordering source of the upper modal will be bouletic. [need to prove that not equiv to a single circumstantial/bouletic modal.] 3.2 Graded modality and double modals In general, a model for double modals can work in much the same way that Kratzer s graded modality does, with a relevant modal base and ordering source for each modal. We propose that each of the modals in the double modal construction act in just the same way that a single modal does. However, there are two crucial restrictions evident: the lower modal always has a circumstantial base, and the upper modal always has a bouletic ordering source. 13

A circumstantial modal base is realistic (i.e., contains only propositions consistent with what is known), but is more permissive than a strictly empistemic modal base. If we consider everything that is known, the answer to a question like Can hydrangeas grow here must be no, since it is known that they don t grow here. But if we consider instead a circumstantial modal base, which includes a only subset of what is known, they we consider only the nature of the soil and the climate, and we may conclude that hydrangeas can grow here, even though they don t happen to currently. The ordering source for the lower modal will vary in the normal way. Thus if the lower modal is should, as in John might should leave, the lower modal should contributes a circumstancial modal base and a deontic ordering source. [Maybe say what you had in mind when you said this: We ll assume that the ordering source for the lower modal is an inertial one (Dowty 1979), which orders worlds along the lines of how things normally proceed.] The higher modal builds upon the truth of the lower modal, taking as its set of evaluation worlds only those worlds at which the lower modal is evaluated as true. So, what then is the modal base of the higher modal may/might? I propose that the modal base of the higher modal is epistemic, and returns for each evaluation world the set of worlds epistemically accessible to the evaluation world, in other words, those worlds which are consistent with the available evidence on which we draw conclusions. These worlds are then ordered by an ordering source, which in this case is crucially bouletic. Thus the words accessible from the evaluation world are ordered according to the desires of some person. It is here that the presupposition of a volitional agent resides. If the ordering source relies on someone s desires, then it must follow that there is some sentient being that has those desires. Let s see how that could work with the double modal construction in (20): (28) I might should buy some eggs. The construction in (28) is first evaluated as to the truth of the construction with only the lower modal, as in (29): (29) I should buy some eggs. So the truth of (29) is evaluated with respect to each evaluation world, after which the context set is updated to only include those worlds at which [I should buy some eggs] is true. After context update takes place, the upper modal construction is evaluated, as in (30): (30) might (should (I buy some eggs)) 14

Then (30) is evaluated with respect to each world in the context set, based on the epistemic modal base and bouletic ordering source of the higher modal. Crucially, since the modal base of the lower modal is realistic, then the facts that held for evaluation in relation to the lower modal must still hold: in each world, we have no eggs, and eggs are necessary to make quiche. The epistemic modal base of should takes as its argument some world under evaluation and returns a set of worlds consistent the speaker s knowledge. The bouletic ordering source supplies the ideal imposed by the desires of some relevant individual. This results in a set of worlds closest to the ideal of what the speaker desires. Evaluation takes place just as it does for a single modal construction with a possibility modal: for every evaluation world, if there exists some world closest to the ideal in which I buy eggs, then [might (should (I buy some eggs))] is evaluated as true. 4. Consequences of the formal analysis As argued in the previous sections, the double modal construction presupposes the existence of a volitional agent. This presupposition of a volitional agent is correlated with a requirement for a bouletic ordering source on the higher modal. If in fact the bouletic ordering source on the upper modal triggers the presupposition of a volitional agent, then this analysis predicts that double modals will 1) be infelicitous in situations and constructions where there can be no volitional agent, and 2) give rise to a volitional reading even in instances where no volitional agent is overtly mentioned. These predictions, as it turns out, are correct. In this section, I illustrate that in fact, in context-free situations in which double modals may seem infelicitous, they do become acceptable when put into a context where someone can control the action. 4.1 Presupposition of a volitional agent The most salient feature distinguishing the meaning and usage of a double modal from that of a single modal is that the double modal presupposes the existence of a volitional agent that has the power to bring about some action. This presupposition is a result of the bouletic ordering source required on the upper modal. If, in fact, the worlds are ordered according to the desires of some individual, then there must exist an individual who has those supposed desires. This volitional agent is evident in the way that these constructions are understood, and in the situations in which they are felicitous or infelicitous. I will first examine the construction in light of standard tests for control predicates, and show that though the construction is not a control predicate 15

syntactically, the semantic requirement for an agent is clear in felicitous contexts. I will then offer additional evidence for the presupposition of a volitional agent, based on the unambiguous reading of predicates that in other contexts may have both volitional and non-volitional readings. 4.1.1 Standard tests for control predicates Standard tests for control indicate that this construction acts in many ways like a control predicate. For instance, we shall see that idiom integrity is not maintained, and pleonastic subjects are unacceptable. However, it is possible to create a context in which these constructions suddenly become acceptable. Crucially, one context in which native speakers of double modal dialects seem to always accept double modals is when there is some type of puppet master or someone who directs and controls the action. It is not necessary that this puppet master be in the subject position of a sentence, which indicates that in fact these constructions are not syntactic control predicates. The simple idea or accessibility of someone controlling the action is enough to render the construction acceptable. 4.1.1.1 Idiom integrity The presupposition of an agent can be seen with many idioms, where double modals seem to be infelicitous: (31) a. The cat might be out of the bag. b. #The cat might should be out of the bag. The construction in (31b), whether literal or figurative, is infelicitous in most contexts. However, in a very specific context namely, one in which someone is planning to let the cat out of the bag the sentence is fine, both literally and figuratively. For instance, this might be used as a suggestion that someone let a secret become common knowledge. The idea of someone willfully controlling the action makes the use of a double modal acceptable even when it seems initially infelicitous. Notice that (31b) also shows that the presence of an animate subject is not enough to ensure felicity of these constructions. Instead, the bouletic ordering source requires that there be someone who is sentient and able to bring the action to pass: a volitional agent. Contrast the situation in (31) with that of (32), where an idiom is used to describe an accidental happening. In this case, it is impossible to maintain idiom integrity with the use of a double modal: (32) a. He could kick the bucket any day now. b. #*He might could kick the bucket any day now. 16

Here, the idiomatic expression kick the bucket loses its figurative meaning when embedded below a double modal. Idiom integrity is not maintained, due to the nature of the figurative expression: kick the bucket means to die, but not be murdered, or commit suicide, or become dead by premeditated means. It implies an accidental or even expected death, but not one brought about by the actions of an agent. When embedded below the double modal, which requires an action that is controllable, only the literal meaning is retained, crucially, a meaning which allows an agent.11 4.1.1.2 Pleonastic subjects Just as the presence of an agent can make an idiomatic expression with a double modal acceptable, this same agent idea also allows the presence of a pleonastic subject, even though pleonastic subjects are typically infelicitous with double modals. For instance, with a pleonastic there, if an agent could possibly bring about the situation given, then the construction becomes acceptable. For instance, consider the examples in (33) and (34): (33) a. #There might could be 100 s of students absent tomorrow. b. There could be 100 s of students absent tomorrow. c. There might be 100 s of students absent tomorrow. (34) a. #It might could be raining. b. It might be raining. c. It could be raining. Crucially, the construction in (33a) is only acceptable if there is a plan being concocted in which 100 s of students will be absent. While both (33b) and (33c) can logically refer to an accidental situation in which students are getting sick and may be unable to come to school, (33a) cannot refer to an event that happens by chance, but only one that is brought about as the result of a conscious choice. This is also true in (34), where both (b) and (c) question whether or not it is actually raining, but (34a) makes a suggestion that someone might want to use rain in order to accomplish some goal. Obviously whether or not it is raining is not a situation which is typically controlled by an agent, but if, for instance, the director of a play wanted to bring about some specific effect, (34a) could be used as a suggestion as to how that effect could be achieved. In essence, it would mean something like you could make it rain if you wanted to. Without a context in which rain can be controlled, the use of (34a) is infelicitous. Standard tests for control predicates indicate, then, that when the context is such that an action can be controlled, the double modal is felicitous, even when infelicity results in context-free situations. This follows naturally as the result of the requirement of a bouletic ordering source, which 17

by definition is dependent on the desires of some individual. The bouletic ordering source results in felicity in contexts where that volitional agent can somehow control an action. 4.1.2 Loss of ambiguity In addition to predicting that felicity will occur when a volitional agent is possible in some context, the analysis also predicts that in constructions that are ambiguous, and allow both a volitional reading and a non-volitional reading, only the volitional reading will be possible with a double modal. This is just what happens with verbs like sink. Consider the example in (35a), which has both a volitional and accidental reading: (35) a. Blaine might sink the ship. b. Blaine might could sink the ship. While a non-volitional reading is possible for the single modal construction, in which Blaine sinks the ship, but not on purpose, the double modal construction is unambiguous: Blaine has the power to sink the ship, and if he does, it will be the result of a conscious decision, not an accident where he causes the sinking. [A refinement: if Blaine is a character in a play, then the director might have the power to rearrange the script, in which case the director or her writing staff might felicitously utter (35b) in response to a question such as How can we give Blaine s partner motive for murder? ] While verbs like sink are ambiguous in single modal constructions like (35a), they lose their ambiguity in constructions like (36a): (36) a. The ship might sink. b. The ship might could sink. In (36b), however, the construction is both felicitous and volitional, relying on the presupposition that there is some agent that is going to make a choice as to whether or not the ship sinks. The examples in (35) also reveal an additional distinction between double modals and their single counterparts. When the single modal construction given in (a) has a volitional reading, it is clearly Blaine who makes the decision to sink the ship. However, this is not the case with the double modal in (36b). With the double modal, the ambiguity is not between volitional/nonvolitional, but between who the agent is: was it Blaine, or was it someone who somehow is able to control Blaine s actions, like his father, his boss, or the director of a play? If it is the requirement of a bouletic ordering source on the higher modal that correlates with the presence of a volitional agent, then this result is expected. The double modal does not require that the subject be the one whose desires define the ideal, it only requires that there be someone against whose desires the worlds are ordered. 12 This is also why 18

the construction in (36b) is acceptable with a double modal: a single modal will only allow the subject to be the volitional agent, but the double modal allows any accessible volitional agent to control the action.13 The loss of ambiguity and lack of a non-volitional reading is consistent with the idea that the double modal is a volitional construction, which presupposes the existence of an agent who is able to choose, based upon their own desires, whether or not to carry out some action. This is consistent with the bouletic ordering source of the upper modal: the fact that the ordering source relies on someone s desires makes it necessary that there exist someone whose desires will determine the outcome of a given situation. 4.2 Entailment of the truth of the lower modal This defining characteristic of volition is not the only thing that distinguishes double modals from single modals. If in fact, the construction functions with two modal bases and two ordering sources, then other characteristics also must be considered. One point of importance is the role that the lower modal plays in determining the meaning of the construction. Specifically, the truth of the lower modal is entailed by the double modal construction. This entailment is a result of the context-updating that takes place between the evaluation of the lower and higher modals. Because the only worlds that affect the outcome of evaluation for the higher modals are those in which the truth of the lower modal holds, the construction entails the truth of the lower modal. For instance, consider (37), which entails that I can leave at 5am, but makes the assertion that it may not be my choice to do so: (37) I might can leave at 5am and then I ll get to Dallas earlier. This is, crucially, a case of circumstantial modality, in which what matters are the facts known about the world at hand. So for (37), perhaps the facts that I know are that I have a car, an alarm clock, and nothing planned at that time. The circumstances are such that nothing prevents the possibility that I leave at 5am. Crucially, can deals with what the relevant circumstances are. In order for (37) to make sense, it must be true that I can leave at 5am. For instance, consider the sentences in (38): (38) a. I might could babysit for you. b. You might could babysit for her. c. Beth might could babysit for you. d. #Amy might could babysit for you, but she has to work. (38a) expresses that a speaker does not want to commit herself to babysitting; it does not question whether or not she has the ability, but rather whether or not she is willing to. (38b) is a suggestion, but one that the 19

speaker is not sure if the listener will agree with. So the speaker is expressing that she realizes that this proposition may not be in accordance with what the listener wants to do, even though she knows that the possibility of babysitting exists. The same is true in (38c); the speaker indicates that she knows that it is possible that Beth can babysit, but carries with it the idea that the speaker is not sure if that is really what Beth would want to do. In each case, the use of the double modal expresses uncertainty, but not in relation to the possibility of a proposition. Rather, the possibility or necessity is acknowledged, but the desirability of the proposition is questioned. However, consider that in (38d), where the possibility for Amy to babysit does not exist due to her having to work, infelicity results. Because the lower modal is evaluated as false, the construction as a whole is evaluated as false. If in fact the only worlds that affect the outcome for the upper modal are those in which the lower modal is true, then these judgments are expected. The use of a double modal does not question that the possibility or ability exists, but instead questions whether or not the fulfillment of the proposition is what the agent wants to do. 5. Additional implications of double modals 20 5.1 Uncertainty: Ordering, Reasoning, and Implicature In addition to the presupposition of a volitional agent and the entailment of the truth of the lower modal, the double modal also expresses uncertainty as to whether the agent will carry out a proposition. This lack of certainty as to how the relevant circumstances will play out seems to indicate that there is some degree of choice, or at least lack of certainty, that is inherent to the use of the double modal construction. Intuitively, this does seem to be true: using a double modal indicates a person s unwillingness to commit himself to bringing about the state of affairs described by the embedded modal. Mishoe and Montgomery (1994) note that double modals are often used in situations where negotiation is an issue; the use of the double modal indicates that there is a degree of choice possible. So for instance, the use of a double modal indicates an agent s desire not to do what is required, and would be felicitously used in a context where the speaker wants someone to offer another option, or convince them to do something else. These constructions have a hedging quality that lends itself to use in situations where the speaker wants to make it clear that some action may or may not be carried out. So even though the double modal construction accepts the truth of the second modal, it questions the willingness of the agent to actually carry out the proposition.