Abstract
Contextualism about ‘might’ says that the property that ‘might’ expresses varies from context to context. I argue against contextualism. I focus on problems that contextualism apparently has with attitude ascriptions in which ‘might’ appears in an embedded ‘that’-clause. I argue that contextualists can deal rather easily with many of these problems, but I also argue that serious difficulties remain with collective and quantified says-that ascriptions. Herman Cappelen and John Hawthorne atempt to deal with these remaining problems, but I argue that their attempt fails.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
A speaker who utters (2) may well be trying to convey some stronger propositions, for instance, the proposition that Holmes’s being in Paris is more likely than many other propositions regarding his location that are consistent with the speaker’s knowledge. I will assume here that these stronger propositions are at most pragmatically conveyed, and are not the semantic contents of (2) with respect to any contents. This view is commonly held by contextualists.
More colloquially, ‘Holmes couldn’t be in Paris’. For some reason that I do not understand, it is easy to hear this latter sentence with negation wide-scope, whereas it is nearly impossible to hear ‘Holmes might not be in Paris’ with negation wide-scope.
Not all semanticists accept the claim that \( \left\lceil {{\text{that}}\,S} \right\rceil \) refers, in a context C, to the semantic content of S in C. Some, for instance, hold that ‘that’-clauses refer to sentences, and other adhere to Davidsonian theories. But the view that ‘that’-clauses refer to semantic contents is natural and widespread, and published contextualists accounts of ‘might’ either endorse it or fail to explicitly reject it. See note 35.
Dowell (2010a, b) develops a Kaplanian theory of ‘might’ in a somewhat different way than I do. We agree that such contextualist theories are more flexible and resilient than often supposed. ‘Might’ may be able to express types of modality other than those I list in the text; perhaps the only restriction is that the content of ‘might’ in a context is not a deontic modal property.
Utterances are events. Possible Kaplanian contexts are properties that it is possible for some utterance to have, in roughly the same way that possible worlds are properties that it is possible for the concrete universe to have.
The realization relation must be relativized to a world because a single utterance can have (somewhat) different locations and times at different worlds.
For instance, if U is an utterance of ‘that’ in W by a speaker, then the speaker’s demonstrative intentions (if any) determine whether her utterance U realizes in W a context whose demonstrated object is D, or a context whose demonstrated object is something else, or a context that has no demonstrated object (Perry 2009).
It is easy to hear an utterance of ‘Holmes might have been in Paris’ as an assertion of either epistemic or alethic modality. It is easy to hear an utterance of (2) as an assertion of epistemic modality, but difficult to hear such an utterance as an assertion of alethic modality. This contrast surely has something to do with the difference in tense and aspect between the two sentences, but I cannot offer a deeper explanation. See Portner (2009), chapter 5, for a discussion of interactions between tense, aspect, and modality.
Some semanticists, following Kratzer (1977) on ‘must’, think that ‘might’ determines, with respect to a context, both (i) a quantifier content that quantifies over worlds and (ii) an ordering of worlds (that is, a ternary relation among worlds). This is similar to the familiar view that counterfactual conditionals express propositions that, in part, concern an ordering of worlds by similarity. It is consistent with the view of the last section that M C determines such an ordering relation, but I ignore cases in which such an ordering seems relevant to the truth-values, in contexts, of sentences containing ‘might’.
Teller (1972), DeRose (1991), Egan et al. (2005), von Fintel and Gillies (2008, 2011), and MacFarlane (2011) discuss such uses of ‘might’. The above contextualist view can allow that in some contexts, the semantic content of ‘might’ is some other sort of property concerning group knowledge, such as the property of being known by every member of G or the property of being compatible with the union of all propositions that are known by some member of G. (For discussion, see von Fintel and Gillies 2011; Portner 2009, pp. 158–167; Dowell 2010a, b; MacFarlane 2011) The theory is also consistent with saying that, in some contexts, ‘might’ expresses (roughly) the property of being compatible with everything that is known by someone who is F, where being F is a property that (contingently) all and only members of G have (I am indebted for this last point to Egan 2010).
Von Fintel and Gillies (2011) are contextualists, but their theory is somewhat unorthodox. They react to cases like Moriarty and his henchmen by holding, roughly speaking, that there are many contexts that are appropriate for interpreting a given utterance. More precisely, they say that many contexts are admissible, with respect to an utterance. The admissibility of a context has something to do with the speaker’s (and audience’s) thoughts, intentions, and presuppositions. The sentence that Moriarty utters has at most one propositional semantic content with respect to each admissible context. Moriarty’s utterance puts into play all such propositions. Moriarty can appropriately utter his sentence because he is in a position to assert at least one of the propositions that his utterance puts into play. I assume that more orthodox contextualists would prefer to say that Moriarty’s utterance realizes only one context (in my technical sense of realization), and Moriarty asserts the semantic content of his sentence in this context, but may also assert, or otherwise convey, a number of other propositions that are not the semantic content of his sentence in any context realized by his utterance. The contextualist view in the text hinges on the (orthodox) idealization that there are some distinguished thoughts and intentions of Moriarty’s that determine that his utterance realizes exactly one context. This assumption may turn out to be too severe. Some differences between orthodox contextualist theories and von Fintel and Gillies’s theory may have to do with the division of labor between semantics and pragmatics.
Some theorists may think that that the semantic content of ‘might’, in every context in which the agent uses it epistemically, must concern the knowledge of (roughly speaking) a group of people that includes the speaker. Such a theorist would say that the previous two exogenous utterances are non-literal. But I suspect most contextualists would prefer the view I present in the main text, for it allows the semantic contents of such sentences to be true in contexts where the speaker has exogenous intentions. Hacking (1967) and Teller (1972) discuss uses of ‘might’ in which the propositions asserted concern (roughly) compatibility with sets of true propositions that are as yet known by no one, but which can become known by the speaker, or some group the speaker has in mind, by using known methods. I count such uses as non-epistemic. The above contextualist theory can account for them in much the same way that she can account for alethic uses.
Kripke (1979) and others often use the term ‘disquotation’ for cases in which the reporter utters a belief or knowledge ascription. By contrast, I shall restrict the term to cases in which the reporter utters a says-that ascription.
As I mentioned in note 11, von Fintel and Gillies (2011) have an unorthodox contextualist theory. The unorthodox features of their theory make it unclear (to me) how it should be extended to says-that ascriptions. Moriarty’s utterance of ‘Holmes might be in Paris’ puts into play every proposition that the sentence expresses in a context that is admissible with respect to his utterance. But von Fintel and Gillies’s theory does not tell us which, if any, of these propositions Moriarty says when he utters the sentence. Similar questions arise for says-that ascriptions on their view. If we assume that \( \left\lceil {{\text{that}}\,S} \right\rceil \) refers in C to the semantic content of S in C, then their view entails that ‘Moriarty said that Holmes might be in Paris’ has at most one semantic content in every context. But it is unclear whether we should assume that when Watson utters that ascription, his thoughts and intentions determine only one admissible context or many. Their theory also does not say which propositions Watson says, asserts, or puts into play when he utters his ascription. We could consider various possible ways to extend their theory, but I do not have the space to do so here.
Two points: (a) I assume here that asserting is sufficient for saying. So if Moriarty asserts proposition P, then a sentence of the form \( \left\lceil {{\text{Moriarty}}\,{\text{said}}\,{\text{that}}\,S} \right\rceil \) is true in a context in which the semantic content of S is P. (b) There are tricky cases in which Moriarty intends to speak of G MH but believes that he is not a member of that group. Such uses are exogenous, and a contextualist can deal with them in the same way he does with reports of other exogenous uses—see below in the main text. These cases may also raise Frege-type puzzles, but such phenomena are beyond the scope of this paper.
If Lestrade’s audience does not know his or Holmes’s intentions, then they might not grasp what Lestrade asserts. But this is a constraint on pragmatics rather than semantics.
I suppose that a contextualist could hold that Hudson asserts something true when she utters (17), though this true thing is not the semantic content of (17) in her context. However, I assume here that most contextualists would want to hold not only that Hudson asserts a true proposition, but also that the semantic content of (17) is true with respect to Hudson’s context.
(17) and (18a) may have both collective and distributive readings (compare them with ‘Every male in 221B Baker Street lifted Hudson’s piano’). I am interested only in the distributive readings. I claim that Mrs. Hudson says something true if she utters it while intending the distributive meanings. I include (18b) because it does not have a collective reading. Thanks to Michael Glanzberg for discussion.
I have preserved one rather idiosyncratic feature of C&H’s notation, by allowing ‘(to x)’ to appear as a subscript on ‘nearby’, as in ‘nearby(to x)’. I will discuss some issues about how to interpret their notation later. I have changed their notation in another respect that I think is unimportant. They use ‘λx (x said that Naomi went to a beach nearby(to x))’, which changes the surface order of ‘nearby’ and ‘beach’. My lambda-predicate preserves the surface order of ‘nearby’ and ‘beach’.
C&H hold that there may be semantic limits on the sorts of functions that can serve as the referent of ‘f’ in a context. Strictly speaking, C&H (p. 48) say that ‘f’ “picks out” a function. They seem to mean that the referent or extension of ‘f’ in a context is a function. If the extension of ‘f’ in C is a function, then an appropriate semantic content for ‘f’, in C, would be a functional relation, that is, a binary relation R such that, for any x, there is at most one y such that xRy. But C&H do not mention functional relations; indeed, they never specify the semantic content of ‘f’ in a context. Perhaps they ignore semantic content because they “ignore complexities introduced by time and modality” (C&H, p. 48).
Thanks to a referee for pointing this out.
I assume here that the semantic content of ‘mightf(x)’ in a context C (under an assignment to ‘x’) is a structured entity whose constituents are the content of ‘might’ in C and the content of ‘f(x)’ in C (under that assignment). This assumption entails that (32) is true in Hudson’s context only if ‘might’ has the same content in Hudson’s, Lestrade’s, and Mycroft’s contexts. (Thanks to a referee for suggesting that I clarify my reasoning here.) An advocate of C&H’s view might instead hold that the content of ‘mightf(x)’ is simply an (unstructured) property determined by the content of ‘might’ and the content of ‘f(x)’. But even on this view, it would be remarkable if (32) is true in Hudson’s context while ‘might’ has different contents in her, Lestrade’s, and Mycroft’s contexts. (There is an alternative to C&H’s view on which ‘might’ is “doubly context-sensitive,” not because ‘might’ is context-sensitive, but because ‘might’ is associated with two context-sensitive function terms. I discuss this alternative view in note 27 below.)
Kratzer (1977), in fact, makes no claims about ‘might’. The phrase ‘in view of’ in her examples always appears before clauses containing ‘must’ or ‘can’. But many who follow her extend a similar view to ‘might’.
I assume here that attitude verbs, such as ‘believe’, are not modal expressions. But even on (mistaken) theories that say that they are modal, they are not expressions that take the contents of the above prefixes as arguments.
More accurately, this would be so on C&H’s view, as I explicated it above, and on many other contextualist views of ‘might’. On my explication of C&H’s view, the value of ‘f(x)’, relative to a context and assignment, fully determines the content of ‘mightf(x)’ in a context. If the content of ‘f(x)’ in a context can be fixed by a prefix like one of those above (or if ‘f(x)’ can be simply replaced by such a prefix), then a sentence that contains a prefix and ‘might’ semantically expresses the same modality in all contexts. Kratzer’s original (1977) theory also entails that ‘might’ sentences with such prefixes express the same modality in all contexts. So does the version of Kratzer’s theory presented by Portner (2009, pp. 50–56). But some contextualist theories, such as von Fintel and Gillies’s (2011), do not have this consequence. On their view (roughly speaking), ‘might’ expresses a quantifier over possible worlds, and the domain of worlds over which this quantifier ranges may vary from context to context. Prefixes like those above further restrict the domain of worlds over which ‘might’ quantifies in a context, but do not prevent it from varying in its range from context to context. Thus a sentence containing ‘might’ and a prefix can express different modalities in different contexts, and the criticism that follows below in the main text does not apply to their view. However, such a view is seemingly vulnerable to the same sorts disquotation objections that afflict other contextualist views about ‘might’; and it also seems vulnerable to parallel disquotation objections concerning ascriptions that contain prefixed ‘might’ sentences in their that-clauses. (There are complications. See note 24.) A theorist who is attracted to von Fintel and Gillies’s theory, or who is attracted to “double context-sensitivity”, might propose a C&H-like revision of F&G’s theory. Such a theory would say that ‘might’ interacts with two context-sensitive function terms, represented with ‘f’ and ‘g’ in ‘mightf(x)g(x)’. On this view, Hudson’s ascription (31), on one reading, expresses the same proposition as ‘Lestrade and Mycroft said that λx (Holmes mightf(x)g(x) be in Paris)’. (Thanks to a referee for mentioning this alternative view.) An advocate of such a view should say that ‘might’ itself is context-insensitive, for otherwise there would be no guarantee that Hudson’s ascription is true in her context, for reasons like those I gave in Sect. 4 and note 24. But this alternative view has problems similar to, and perhaps worse than, C&H’s. First, it seems committed to saying that ‘might’ is a ternary predicate. (Von Fintel and Gillies themselves are not committed to this.) Second, it has the same problems with validity as C&H’s theory (to be described in the main text below). A more detailed examination of this “two context-sensitive functors” view, and F&G’s theory, will have to await another occasion.
There is a second way of understanding C&H’s lambda-predicate proposal, on which ‘might’ is not a binary predicate. I think it does not capture their intent, but is worth considering briefly. On this unary interpretation of C&H, ‘might’ is a unary predicate that semantically expresses, in all contexts, the same unary modal property of propositions, namely possibility. The function term ‘f(x)’ is syntactically a modifier of ‘might’, much like an adverb or auxiliary verb. Its content, in a context (under an assignment), is a (higher-order) property that can modify the property of being possible, just as the contents of ‘nomologically’ and ‘metaphysically’ modify the content of ‘might’, in the contents of ‘nomologically might’ and ‘metaphysically might’. The following notation better indicates how ‘might’ functions as an argument of ‘f(x)’, on the unary interpretation of C&H: Lestrade and Mycroft λx (x said that f(x)(might) Holmes is in Paris). On the unary interpretation, ‘Holmes might be in Paris’ is semantically complete (since ‘might’ is unary on this proposal), in much the same way that ‘Holmes runs’ is complete on standard views of that latter sentence. So the unary interpretation of C&H’s proposal for Hudson’s says-that ascription is analogous to the view that ‘Lestrade and Mycroft said that Holmes runs’ has a reading (a semantic disambiguation) on which it is synonymous with ‘Lestrade and Mycroft λx (x said that Holmes f(x)(run)), where the content of ‘f(x)’ in a context is a property of running, such as being quickly (done) or slowly (done). I assume the latter proposal is implausible. The unary version of C&H’s analysis is equally implausible.
The conclusion of (46) could alternatively be represented by ‘(Some y:y is a thing) (Lestrade and Mycroft λx [x said y])’. But C&H seemingly assume that conjunctive-term distribution and lambda-conversion are valid, and I shall too. If distribution and lambda-conversion are valid, and the alternative symbolization is valid, then (47) is also valid. See also the next paragraph in the main text.
Thanks to Chris Kennedy and Nat Hansen for replies along roughly this line, and thanks to Ishani Maitra for a related reply. My presentation of the reply runs roughshod over use and mention. In particular, it ignores the distinction between two claims: (a) for some model M and context C in M, ‘Lestrade said that Holmes mightf(Lestrade) be in Paris’ is true in C and M, and (b) Lestrade said that Holmes mightf(Lestrade) be in Paris in the world of C (and M). But (a) could be true while (b) is false, for there are unintended models in which the members of C (and W) are not really contexts (or worlds), but rather arbitrary objects, such as numbers, and in which the extension of ‘say’ at C and M is not a set of ordered pairs in which the first object says the second, but rather a set of arbitrary ordered pairs. These unintended models are (nevertheless) relevant to logical validity: a sentence is logically valid only if it is true at all such (unintended) C and M (as well as contexts in the intended model). Some of the initial plausibility of the reply may rest on a failure to notice this.
A further point: Logical truth requires truth at all contexts in all models, including unintended models. In some contexts of some unintended models, the extensions of ‘that’-clauses are not propositions, but rather sequences of numbers, and the extension of ‘say’ is an (arbitrary) set of ordered pairs of (i) numbers that are members of the total domain of the modal and (ii) sequences of numbers which are the extensions of ‘that’-clauses in that context and model. In some contexts of some of these unintended models, (54c) is true and (54d) is false. As I just indicated in the main text, I believe that there are also contexts in the intended model in which (54c) is true and (54d) is false.
A logic of this sort would require that all models (and contexts and worlds in those models) satisfy certain general principles (or meaning postulates) regarding ‘say’. This would rule out unintended models of the sort I described in the preceding note.
Thanks to Michael McGlone for suggesting a reply along the following lines.
Thanks to a referee for suggesting that I consider an objection much like the following one.
The contextualist theories that I have considered assume that a ‘that’-clause refers, in a context C, to the semantic content of its embedded sentence, in C. The objections to contextualism from collective and quantified says-that ascriptions, and my criticisms of C&H’s analysis, also rely on that assumption. Perhaps an advocate of contextualism would wish to deny it. (Thanks to a referee for pointing this out.) But no published theory that I know of does so. I believe that contextualist theories of ‘might’ that say that ‘that’-clauses refer to sentences are also vulnerable to objections from collective and quantified ascriptions. So, I think, are Davidsonian theories that hold that attitude ascriptions refer (in contexts) to utterances. Other alternative contextualist theories might claim that ‘that’-clauses containing ‘might’ refer to Kaplanian characters, or other similar non-propositional entities (such as functions of various sorts). I suspect that theories of this last sort would have problems independent of those I mention here. In any case, replying to such alternative theories must await another occasion.
References
Bach, K. (2011). Perspectives on possibilities: Contextualism, relativism, or what? In: A. Egan & B. Weatherson (Eds.), Epistemic modality (pp. 19–59). Oxford: Oxford University Press.
Cappelen, H., & Hawthorne, J. (2009). Relativism and monadic truth. New York: Oxford University Press.
Cappelen, H., & Lepore, E. (2005). Insensitive semantics: A defense of semantic minimalism and speech act pluralism. Oxford: Blackwell Publishing.
Cappelen, H., & Lepore, E. (2006). Replies. Philosophy and Phenomenological Research, 73, 469–492.
DeRose, K. (1991). Epistemic possibilities. Philosophical Review, 100, 581–605.
Dowell, J. (2010a). Flexible contextualism about ‘Ought’ and attitude-attributions. Talk delivered at the annual meeting of the Pacific Division of the American Philosophical Association.
Dowell, J. (2010b). A flexibly contextualist account of epistemic modals. Unpublished paper.
Egan, A. (2007). Epistemic modals, relativism, and assertion. Philosophical Studies, 133, 1–22.
Egan, A. (2010). Comments on Janice Dowell’s ‘Flexible Contextualism about ‘Ought’ and Attitude-attributions’” (Dowell 2010a). Talk delivered at the annual meeting of the Pacific Division of the American Philosophical Association.
Egan, A., Hawthorne, J., & Weatherson, B. (2005). Epistemic modals in context. In G. Preyer & G. Peter (Eds.), Contextualism in philosophy (pp. 131–169). Oxford: Oxford University Press.
Egan, A., & Weatherson, B. (Eds.). (2011). Epistemic modality. New York: Oxford University Press.
Hacking, I. (1967). Possibility. Philosophical Review, 76, 143–168.
Humberstone, L., & Cappelen, H. (2006). Sufficiency and excess. Proceedings of the Aristotelian Society Supplementary, 80, 265–320.
Kaplan, D. (1989). Demonstratives. In J. Almog, J. Perry, & H. Wettstein (Eds.), Themes from Kaplan (pp. 481–563). Oxford: Oxford University Press.
Kratzer, A. (1977). What ‘must’ and ‘can’ must and can mean. Linguistics and Philosophy, 1, 337–355.
Kripke, S. (1979). A puzzle about belief. In A. Margalit (Ed.), Meaning and use (pp. 239–281). Dordrecht: Reidel.
Lasersohn, P. (2005). Context dependence, disagreement, and predicates of personal taste. Linguistics and Philosophy, 28, 643–686.
MacFarlane, J. (2011). Epistemic modals are assessment-sensitive. In A. Egan & B. Weatherson (2011), pp. 144–178.
Montague, R. (1974). English as a formal language. In R. Thomason (Ed.), Formal philosophy. New Haven, CT: Yale University Press.
Nunberg, G. (1993). Indexicality and deixis. Linguistics and Philosophy, 16, 1–43.
Perry, J. (2009). Directing intentions. In J. Almog & P. Leonardi (Eds.), The philosophy of David Kaplan (pp. 187–207). New York: Oxford University Press.
Portner, P. (2009). Modality. New York: Oxford University Press.
Teller, P. (1972). Epistemic possibility. Philosophia, 2, 303–320.
von Fintel, K. (2006). Modality and language. In D. M. Borchert (Ed.), Encyclopedia of philosophy. Detroit: MacMillan Reference USA.
von Fintel, K., & Gillies, A. (2007). An opinionated guide to epistemic modality. In T. S. Gendler & J. Hawthorne (Eds.), Oxford studies in epistemology (Vol. 2, pp. 36–62). Oxford: Oxford University Press.
von Fintel, K., & Gillies, A. (2008). CIA leaks. Philosophical Review, 117, 77–98.
von Fintel, K., & Gillies, A. (2011). ‘Might’ made right. In A. Egan & B. Weatherson (Eds.) pp. 108–130.
Acknowledgments
I presented an early version of this paper at a workshop on contextualis in the philosophy of language at Queens University in Kingston, Ontario, in September 2009. Thanks to Adèle Mercier and Arthur Sullivan for organizing the workshop and inviting me to speak at it. Thanks to Richard Vallée for his comments on my talk. I also thank those who commented on my paper during the discussion period and afterwards, including Kent Bach, Michael Glanzberg, Nat Hansen, Claire Horisk, Chris Kennedy, Barry Lam, Ishani Maitra, Michael McGlone, François Recanati, and Brett Sherman. Thanks to Andy Egan and Janice Dowell for helpful conversations. Thanks to Michael McGlone for helpful written comments on an earlier version. Thanks to an anonymous referee for many helpful suggestions.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Braun, D. Contextualism about ‘might’ and says-that ascriptions. Philos Stud 164, 485–511 (2013). https://doi.org/10.1007/s11098-012-9861-3
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11098-012-9861-3