David Kaplan’s (1989b) semantic theory for indexicals yields a distinctive logic for indexical languages, generating a set of logical truths that are entirely absent from non-indexical languages. These logical truths are notable in that they invalidate the rule of necessitation (\(\models \phi \rightarrow\ \models \Box \phi\)), because there are sentences that Kaplan thinks cannot be uttered without being true, despite the fact that they express non-necessary states of affairs. Consequently, they are often cited as examples of contingent a priori truths. Recent discussion of indexicality in the philosophical literature has challenged Kaplan’s proposal to grant this privileged status to certain indexical constructions, however, by drawing attention to numerous apparent counter-examples in natural language. These challenges almost unanimously agree that Kaplan is too restrictive in his analysis of the sorts of contexts in which an indexical sentence can be employed.1 All of the proposed counter-examples appear to show that under certain conditions uses of indexical sentences can align an indexical sentence with a context that is not recognized by Kaplan’s theory and, therefore, that Kaplan’s apparent cases of contingent a priori truths do not reflect genuine semantic features of English, but only reflect Kaplan’s mistaken intuitions about the admissible range of contexts in which indexical sentences may be uttered.
In this paper I will defend Kaplan’s semantic theory against this challenge. I will proceed by first arguing that the proposed counter-examples in question are not just the result of aligning an indexical sentence with an unusual context, they are the result of applying a context-shifting operator on the character of an indexical sentence. Kaplan calls an operator on character a “monster” and argues that monsters are entirely absent from the semantics of English. There are, however, metalinguistic devices such as quotation that do behave like monsters, as acknowledged by Kaplan. I will argue that the proposed counter-examples to Kaplan’s account all share important similarities with these metalinguistic operators and are thus best understood as resulting from pragmatically introduced metalinguistic operators on constructions that are perfectly acceptable on Kaplan’s analysis. I thus conclude that Kaplan’s semantic theory does not stand in need of revision to accommodate these examples, and we have every reason to retain the view that indexicals can generate the sorts of contingent a priori truths predicted by Kaplan. I will begin by elucidating Kaplan’s theory, then discussing the challenges to that theory. I will next introduce the notion of a monster and present the argument that all of the proposed challenges depend on what I will call monstrous operations. I will then defend the view that these particular monstrous operations are best explained pragmatically as resulting from metalinguistic operations, rather than semantic ones. Finally, I will consider some objections and replies.
1 The Logic of Indexicals
Pure indexicals are expressions whose literal meaning both requires a context for saturation and specifies precisely what role context must play in the saturation (in English, examples include “I,” “now,” “today,” some uses of “here,” etc.). Demonstratives differ from pure indexicals insofar as they require an additional demonstration such as a gesture or other such directing intention (in English, examples include “that,” “this,” “she,” some other uses of “here,” etc.). To account for this distinctive class of meanings, Kaplan proposes a two-level semantic theory, coupled with a double-indexing of indexical sentences with formal representations of contextual situations. Firstly, an indexical expression is associated with both a character and a content. The character of the expression is a function from context to content. Less formally expressed, this means that the character can be thought of as a rule governing the contribution required by context in order to fix the semantic value or reference of the term. For example, the first-person English pronoun “I” has a character of the form “the agent of this utterance.” This description specifies a function that will yield a different value depending on who is speaking. Sentences have the same two-level semantic profile and each level is compositionally derived from the individual expressions contained in the sentence. For example, the sentence “I am walking” has a character that maps a context \(c\) on to the proposition that \(a_c\) is walking, where \(a_c\) is the agent of the context \(c\). Different contexts will be mapped to different contents by this function. If character is a function from contexts to contents, then individual agents, objects, times, places and so on will be the contents of individual indexical and demonstrative expressions when they are used in context. The content of an indexical is thus its reference.2 The content of a sentence is a proposition. It follows that indexical sentences cannot contribute truth-evaluable propositions to a semantic theory without the assistance of context. Hence, within the semantic theory, it is ordered pairs of sentences and contexts that are the bearers of truth-values. Contexts themselves are precisely specified as sequences of parameters corresponding to the indexical elements in the sentence. So, for example, a sentence containing “I,” “here,” and “now” will demand a context with parameters for an agent, a spatial location, and a temporal location. We also add a world parameter to every context. Thus the context in this case will be of the form \(c\) = \(\langle a_c, l_c, t_c, w_c \rangle\). The sentence-context pair \(\langle s,c \rangle\) models the proposition expressed by the sentence \(s\) in the context \(c\). Utterances of sentences are thus indexed to contexts. Similarly, evaluation of propositions requires what Kaplan calls “circumstances of evaluation,” which are (at a minimum) pairs of world and times. As we will now see, this double-indexing reveals that some sentences will be true with respect to any context they are paired with, despite expressing propositions which are not true at every circumstance of evaluation (thus, are not necessary).
With the above system outlined, we can make sense of Kaplan’s claim to have discovered examples of the contingent a priori in English. Consider the following four sentences:
(a) I am me.
(b) This is of the same chemical kind as that.
(c) I am here now.
(d) I am not alive.
Any utterance of (a) will be true because any context \(c\) will be such that \(a_c\) = \(a_c\). Thus we know a priori that any utterance of (a) will be true. And, as this truth holds in all possible worlds, (a) will be true at every circumstance. Thus (a) is a necessary truth. An utterance of (b) in a context \(c\)1 in which the demonstratum of “this” is a sample of liquid water, and the demonstratum of “that” is a sample of water-ice, will be true. But an utterance of the same sentence with different demonstrata could well be false. Hence it is certainly not true a priori. However, if it is uttered in \(c\)1, then the truth it expresses will hold at every circumstance of evaluation. Thus it is an example of a necessary a posteriori truth. Kaplan also holds that any utterance of (c) must be true, on the grounds that there is no context in which an agent can fail to be at the location of that context at the time of that context when uttering something in that context. All the same, their being located at whatever part of space-time they are at when they make that utterance is obviously just a contingent fact—they could have been elsewhere. Hence the proposition expressed does not hold true at every circumstance. Accordingly (c) is assumed by Kaplan to be a contingent a priori truth. Likewise for (d), no-one can utter this truly (as the saying goes, dead men tell no tales), but there is nothing necessary about one’s being alive. Thus an utterance of (d) is known to be false on a priori grounds, but is not necessarily false.
What makes Kaplan’s defence of the contingent a priori so compelling is that Kaplan’s proposed cases require no investment in any kind of metaphysical speculation. They are just immediate consequences of the correct semantic analysis of indexicals. Or so it seemed. Many have reacted to Kaplan’s logic of indexicals, however, by pointing out that Kaplan’s analysis does not seem to be proceeding on purely semantic grounds but in fact makes significant assumptions about the conditions under which utterances and other forms of communication can be made that are empirically questionable. Consider again the example (c) above. This apparent logical truth is routinely negated as an answerphone recording: “I am not here now. Please leave a message after the tone…”. Similarly, one may record a message to be replayed at the reading of one’s will that contains (d): “If you are hearing this recording, then I am not alive. I have left you this message to communicate my wishes to you after my death…”. In such circumstances it appears that truths, not logical falsehoods as Kaplan’s analysis seems to predict, are being communicated.
These counter-examples to Kaplan’s analysis highlight, and challenge, an assumption in Kaplan’s theory about the interplay between sentence and context. Kaplan’s assumption is that contexts of utterance, inscription, or other sorts of linguistic performance, always conform to a minimum structural norm such that agents of utterances are always located in the time and place of that utterance. Indeed Kaplan explicitly acknowledges this, arguing that we must restrict contexts of utterance to these “proper” contexts: “[I]mproper indices are like impossible worlds; no such contexts could exist and thus there is no interest in evaluating the extensions of expressions with respect to them” (1989b, 509). Prima facie, these counter-examples seem to show that Kaplan is wrong. Furthermore, the counter-examples are ubiquitous. Here is another, from Predelli (2005, 43). Jones writes the following note just before leaving his house at 8am, which he then leaves at home for his wife, who is not due to return until 5pm:
I am not at home now. If you hurry, you’ll catch the evening flight to Los Cabos. Meet me in six hours at the Hotel Cabo Real.
In this example, the note is obviously not intended to be, nor will it be, interpreted as indexed to the time at which it was inscribed but, rather, is intended to be indexed to the time at which it will be read. Another example, offered by Corazza, Fish, and Gorvett (2002), challenges Kaplan’s assumption that the agent, utterer, and referent, of “I” in a given context must always coincide. They invite us to consider the case of an academic who leaves a note on his door saying “I am not here today” to signal his absence when it is read. This, already, is an example equivalent to Predelli’s above, but they continue the story by imagining that the academic returns to work several days later and then removes the note and reattaches it to a colleague’s door to signal their absence. It now seems that the referent of “I” (along with other indexicals) has shifted while its inscriber and possibly even agent have not. What are we to say of these challenges to Kaplan’s theory? In the next section I will argue that all of these counter-examples display “monstrous” properties.
2 Monsters and Monstrous Operations
Kaplan gives the name “monster” to any would-be operator on the character of an indexical. He maintains that no such operators exist in English. Take for example, the indexical “I.” A monster operating on this expression would effect a context shift such that the reference of the expression shifted on to someone other than the agent of the context of utterance. But no such operation seems possible within the confines of ordinary English. If I say “in some contexts, I am not me,” this is simply false (if interpreted literally—there may of course be figurative uses of this sentence which are understood to communicate a truth). I am always identical to myself. Similarly, if I embed the same indexical within a propositional attitude verb, the attitude verb has no impact on the character of the indexical, which immediately takes wide scope with respect to it: an utterance of “John believes that I am happy” communicates the speaker’s report that John believes the speaker to be happy. The character of “I” picks out the speaker, regardless of any operators contained in the utterance.
Kaplan does point out, however, that monstrous operations can be created by metalinguistic devices. The most common is quotation. Compare the following:
(e) John said that I am happy.
(f) John said “I am happy.”
By naming the indexical sentence “I am happy” we can shift the character of that sentence in (f), whereas it is impossible in (e). Kaplan’s position, then, is that the only operators resembling monsters that can be applied to English expressions are metalinguistic operators. To keep this distinction between linguistic and metalinguistic operations intact in what follows, I will use the expression “monster” to denote the sort of lexicalized linguistic operator that Kaplan maintains is absent from English,3 and the expression “monstrous operator” (hereafter “MO”) to denote any operation, including the metalinguistic operators such as quotation, on character. Thus, according to my usage, every monster is an MO but not every MO is a monster.
3 Displaced Communications
The counter-examples to Kaplan’s proposed truths of the logic of indexicals have been responded to in varying ways, but most of the responses conform to one general strategy. The counter-examples are usually understood as somehow involving a displacement from its point of origin of the information that is communicated. To put it another way, a distinction is drawn between the point at which the communication is encoded and the point at which it is decoded. The alleged flaw in Kaplan’s reasoning has thus been almost unanimously identified as the mistaken assumption that communications occur at the point when (and where) they are encoded (either recorded in the cases like answerphone messages, or inscribed in the case of written notes and messages). By contrast, these counter-examples all seem to be intuitively understood as communicating information at the time when they are decoded. Sidelle (1991, 535) describes the production of an answerphone message as a process of “arranging to make an utterance at a later time, or, if one likes, deferring an utterance.” This interpretation has gone largely unchallenged despite the differences in opinion as to the semantic or pragmatic mechanism by which this proposed procedure is thought to be realized.4 In what follows I will use the term “Displaced Communication” (hereafter “DC”) to denote this proposed act of encoding content for later decoding.5
It is rarely, if indeed ever, noted that any such act of encoding content for later communication is itself an MO in the sense that all such cases involve the use of an indexical to communicate a content which is shifted away from the context in which the communication is encoded. For example, the use of “now” to refer to a time other than the time at which it is being used to encode the communication, the use of “I” to refer to someone other than the agent who is encoding the communication, and so on. Furthermore, it is clear that the operation of deferring the communication must involve this monstrous operation. If it did not, then we would not have our counter-examples to Kaplan’s proposed analysis. DCs are not monsters in the sense of being discrete lexically encoded operators within the language that introduce MOs, but they are certainly monstrous in this wider sense that includes pragmatic processes and operations (which is clearly encompassed by the definition of an MO provided above).
Two objections may be raised against the above description of displaced communications as being (or being the result of) MOs, which should be addressed before continuing. Firstly, one may point to the fact that displaced communications appear to perform (or result from the performance of) an operation directly on the context with which the communicated sentence is paired, rather than the character of the sentence. Secondly, one may point to the fact that monsters, as traditionally understood, play a recognisably semantic role by operating directly on lexical items through binding operations, imposing scope relations on them, and so forth. I will reply to each objection in turn.
On Predelli’s (2005) interpretation of DCs, the correct context that a sentence must be paired with for accurate semantic evaluation is determined by the intention of the speaker. That I can intend my utterance at time \(t\)1 of the sentence “I am not here now” to be evaluated at a later time \(t\)2 demonstrates, on this view, Kaplan’s error in defining proper contexts too narrowly. On this view, the counter-examples to Kaplan’s theory arise because sentences have been paired with contexts that he did not recognize. As such, it does not automatically appear that any MO has been applied. The character, Predelli (2005, 44) insists, does not change at all. Rather, careful inspection of the “preparatory operations” (2005, 58)—a phrase Predelli borrows from Quine to describe the decisions we make about how to regiment particular utterances or inscriptions to make them ready for semantic evaluation—simply reveals that sentences can in fact be paired with a wider selection of contexts than Kaplan anticipated. All of the work that separates Predelli’s position from Kaplan’s occurs at this pre-semantic stage: “[O]nce the appropriate clause-index pair has been identified, the indexicals proceed with their customary characters, and results of truth-value are obtained on the basis of the usual mechanisms of compositional analysis” (2005, 58). Nonetheless, further reflection makes it clear enough that this position entails the existence of MOs. An MO is an operation which shifts the context of an indexical element or sentence by operating on its character. We can easily modify the examples of displaced communication to make this effect more explicit. For example, one can download an audio file of Arnold Schwarzenegger’s famous utterance of “I’ll be back” from the Terminator movie and set this file to play as an answerphone message. This performs an MO on the original utterance (made by Schwarzenegger), shifting the context to one in which the agent is no longer Arnold’s cyborg character from the movie, but a real person who has set up an answerphone message to denote their temporary absence or unavailability to a caller.
A further worry may arise by comparing these sorts of MOs with the operators that Kaplan defines as monsters. A monster is an operator on the character of an expression. One way to identify the presence of a monster would be to look for the observable effects that the monster has on the scope of the expression operated on. For example, if we try to shift the scope of the indexical “now” by using a phrase like “in some contexts,” we get the following:
(g) In some contexts, yesterday is now.
The attempt to shift the context fails because “now” and “yesterday” resist embedding under the scope of the operator; i.e. they take wide scope over the operator. On the surface, this seems to be a quite different operation to any present in cases of DCs. The answerphone message “I am not here now,” for example, does not display any distinctive impact on the scope of the indexicals contained in it. While this is undeniable, I don’t think it counts against the view that an MO is at work in cases of DCs.
For one thing, there are many examples of DCs which are quite naturally understood as imposing a binding operation on an indexical element. Take for example the logo used on merchandise by the Rock Climbing equipment manufacturer DMM. They produce clothing with the following phrase emblazoned on it:
(h) Climb now, work later.
This phrase can be naturally used in conversation in a way that pairs it with a proper Kaplanian context:
Speaker A: I don’t know whether I should go climbing now, or
after I have finished my essay. What do you think?
Speaker B: Climb now, work later.
When it occurs as a logo on the aforementioned clothing, however, it cannot be interpreted this way. There is not just one time which fixes the referent of “now” and “later” in this case. Rather it expresses something along the lines of “always go climbing before doing your work.” In fact, it seems to have much the same logical structure as a puzzling case noted (but not addressed) by Kaplan:
(i) Never put off until tomorrow, what you can do today.
The difference, of course, is that (i) contains a lexical item “never” which quantifies over temporal values, making the binding of “tomorrow” and “today” explicit. But this difference is trivial—it is obvious enough that the quantifier “always” is implicitly present as an unarticulated constituent of (h). Thus we have a case of a DC in which the indexicals are forced to take narrow scope with respect to an operator. This is a clear example of an MO.
4 Metalinguistic Monstrosity and Varieties of Quotation
The above sections give us reason to accept that DCs are MOs. I will now argue that, although they are MOs, they are not monsters. This leads naturally to the conclusion that their context-shifting powers are not the result of any semantic operation but is best thought of as the result of a pragmatic process. This will lead me to conclude that DCs pose no serious challenge to Kaplan’s theory of indexicals and, in particular, to his restriction of the range of admissible contexts to proper contexts. Consequently, I shall argue that Kaplan is correct to classify sentences like “I am here now” as encoding contingent a priori truths.
That DCs are not monsters is fairly self-evident. Monsters are linguistic operators. No lexicalised operator can be discerned in the DCs discussed in the literature. It is not the addition of a new constituent to the sentence “I am here now” which performs the role of an MO when this sentence occurs within a DC. It is the particular use the sentence is put to. One could perhaps pursue the line that an unarticulated constituent is responsible for the MO and is thus a monster but there seems little evidence or motivation for such a view.6
If DCs are MOs but are not monsters, then they are most naturally understood as behaving exactly like the paradigmatic metalinguistic MOs recognized by Kaplan which were discussed earlier. Quotation, for example, is an MO because it can take an indexical sentence and block the indexicals within it from taking their customary wide scope positions. It does so by mentioning, rather than using, the sentence. I suggest that DCs are the result of MOs which do exactly the same thing. A DC is created by taking an indexical sentence and recording it ready to be mentioned in a new setting at a later date. Construed in this way, DCs are not constructed by a semantic operation. Thus the criticism levelled at Kaplan which draws on DCs as apparent counter-examples to his semantic theory for indexicals is misplaced. A distinctive feature of DCs is that they require a rich contextual setting. This contextual setting is not the minimalistic sequence of parameters required for saturating indexical expressions, but a far wider notion of “context,” incorporating various complex conventions surrounding human interaction. These conventions are essential to the performance of a DC. But such features are not semantic features. Thus it is natural to understand the MOs involved in generating DCs as pragmatically licensed, rather than semantic. I will now explain in detail the pragmatic process that I suggest is at work in these cases.
Quotation is surprisingly varied. Whereas it was once assumed that quotation is a simple device for self-nominalisation, enclosing a string of expressions within quotation marks to generate a name of that string, it is now widely noted that quotation is not restricted to this simple operation. Consider a case where it does seem to behave in such a way. For example:
(j) “Schnee” is a German expression which stands for snow.
In (j) the name “Schnee” names the expression “Schnee” but that expression is not used in any meaningful way in (j). We are simply exploiting the convention whereby quotation marks name the expressions enclosed within them. We could just as easily have exploited a different naming convention or indeed stipulated one. For example, I can stipulate a convention for naming an expression as follows:
(k) Let whichever German word stands for snow be called “Angelika.” Angelika is often uttered by German speakers when in the presence of snow.
The convention employed in (k) is perfectly clear and comprehensible. It does not require any grasp of the German expression for snow or even an ability to recognize that expression. We simply report facts about the expression by utilising a (descriptively introduced) name of it. These sorts of examples demonstrate that quotation behaves as a purely “mentioning” device in such contexts. The semantic content of the expression that is being quoted is wholly inert in these contexts (hence its unproblematic absence in (k)).
Other uses of quotation, however, are less simple. Newspaper headlines, for example, commonly employ quotation not only to report speech but also to convey information encoded by that speech. Here are a few examples taken at random from the BBC News website on one visit:
Woman “killed dad and buried him”
Army ads “won’t appeal to new soldiers”
Financial services “pivotal to Brexit deal”
“Chronic” nurse shortage and Meghan “mania”
In all of the above cases, the quoted material is a speech report. However, there is more going on here than just a speech report. Compare it to the following, more straightforward speech report, taken from Rap Artist Chuck D’s (1997, 193) autobiography:
Ice-T was in the video because I saw him while he was in Arizona and asked him if he wanted to be in the video. He said, ‘Cool.’
This is simply a speech report—reporting the words used to accept an offer to feature in a promotional video. In the previous examples, however, the direct quotation is not simply a report of the words used by whoever uttered them—it also draws attention to the speech act they were used for and, in doing so, draws attention to (and uses them to express) their content. In this usage, which is very common in newspaper headlines, the quotation does not just name the expressions used but also establishes that they were used to allege something. In such cases where an allegation is reported, the reporter does not implicate herself as one making the allegation, she simply reports the allegation and reports that such an allegation has been made. But we can also find cases where quotation is employed not just to report a claim, but also to endorse that claim:
(l) Kaplan’s example of a kidnapped heiress, locked in the trunk of a car, who has lost all track of time and of her location, yet who can still think to herself “it is quiet here now,” demonstrates clearly that, “ignorance of the referent does not defeat the directly referential character of indexicals.”
In (l) the quoted material (from Kaplan 1989b, 536) at the end of the passage both reports Kaplan’s view and, at the same time, endorses it. The quoted material is not simply named; it’s content is asserted. Quotation of this sort, labelled “mixed quotation” by Cappelen and LePore (1997) because of its dual role as reported and asserted, is quite common.7 Récanati (2010) helpfully distinguishes between “closed” and “open” forms of quotation to make sense of the distinctions at play in these cases. Following Davidson (1979), Récanati interprets quotation marks as performing a demonstrative role whereby the linguistic material (L) quoted is demonstrated as the referent of the quoted expression “L” in reports of the form “S said that ‘L’.” The difference between open and closed quotation is that closed quotation recruits the demonstrated material to play the syntactic role of a singular term, whereas open quotation is any form of quotation that cannot be so construed:
The contrast between open and closed quotation is illustrated by the following pair of sentences:
(7) Stop that John! ‘Nobody likes me’, ‘I am miserable’ … Don’t you think you exaggerate a bit?
(8) John keeps crying and saying ‘Nobody likes me’.
In (7) a token of ‘Nobody likes me’ and ‘I am miserable’ is displayed for demonstrative purposes, but is not used as a singular term, in contrast to what happens in (8), where the quotation serves as a singular term to complete the sentence ’John keeps crying and saying ___’. Sentence (7), therefore, is an instance of open quotation, while (8) is an instance of closed quotation. (Récanati 2010, 231)
If we acknowledge this distinction, we ought to agree with Récanati that there is a fundamental difference in linguistic role between the demonstrated linguistic material in closed quotation and that in open quotation. Whereas closed quotation recruits (a token of) the linguistic material as a singular term which is naturally understood as referring to itself (as a type in most instances), open quotation does not feature any singular term which naturally presents itself as requiring a referential interpretation. Récanati’s proposal is that the sense in which the quoted material acts as a demonstration in open quotation is wider than the customary sense in which demonstratives are taken to have their reference fixed by an accompanying demonstration. In open quotation, quoted material demonstrates in the sense of providing a performance or picture that represents through a form of mimicry. Thus, in Récanati’s example (7) above, John’s speech is quoted as a means of mimicking his self-indulgent utterances.
Understood as mimicry, open quotation has monstrous potential. This is unsurprising, of course: we have already noted that quotation is a metalinguistic monster. Mimicry is clearly an attempt to represent a content which is, in some sense, displayed from a perspective other than that of the speaker, namely the perspective of the one that the speaker mimics. Consider my report of my teenage daughter’s recent request for a new pair of trainers:
(m) Amy has asked me to get her a “sick” new pair of trainers.
I do not, ordinarily, use the word “sick” with the sense it has been assigned in (m). If I am honest, I confess that I am not entirely sure what the extension of the term “sick” is when my daughter and her friends use it in their conversations. But I know enough about it’s meaning to know that it is a desirable property of footwear in my daughter’s opinion (hence this use of the expression is not synonymous with other, more common, uses of “sick” in English) and this is readily communicated (to others who have at least the same level of acquaintance with this term as I do) by my utterance of (m). But I do not communicate my judgement as to the possession of this property by any footwear when I use this term in (m); I communicate my daughter’s judgement. Thus we have a form of context-shifting operator present in (m). The quoted expression “sick” shifts the context to one in which Amy judges things to have the property that she takes that expression to encode. As Récanati (2010, 260) notes this may not amount to a full-blown MO as it is not clear that shifting from my idiolect to Amy’s is best represented by a shift in the sequence of parameters we would normally associate with a linguistic (as opposed to metalinguistic) context. Nonetheless, it illustrates nicely the potential that open quotation has for shifting the perspective away from that of the speaker in a way that is commonplace in ordinary discourse. Indeed we can extend the usual notion of a linguistic context to accommodate such metalinguistic operations easily enough, by incorporating an “idiolect” parameter for the context (see Récanati 2010, 260), according to which my utterance of (m) will be interpreted as employing a context-shifting operator (quotation marks) to recruit the content of the expression “sick” as assigned in Amy’s idiolect to act as a constituent of a proposition whose content is otherwise assigned in accordance with my idiolect.
Open quotation, understood thus, has a number of advantages, most notable among them being the fact that we can simultaneously maintain that quotation is (i) metalinguistic, (ii) an MO, and (iii) used (rather than merely mentioned) as a means of communicating information. Furthermore, as I will now illustrate, it provides a perfect explanatory model for the MOs discussed in this paper. My proposal is that DCs are best analysed as akin to instances of open quotation. It follows that DCs are MOs but this poses no threat to the Kaplanian claims that there are no linguistic monsters in English, only metalinguistic MOs, and that DCs do not provide counter-examples to the proposal that there are special logical truths of indexicals languages or to any of Kaplan’s proposed logical truths of indexical languages.
To see how MOs can be interpreted on the same lines as instances of open quotation, it will be helpful to consider a range of similar phenomena involving intuitive context-shifting operations.
Expressives are expressions which encode a speaker-attitude alongside, but independently of, their truth-conditional content (if they have any).8 If I utter the expression “yummy,” upon encountering a delicious foodstuff, I express my positive attitude towards its flavour. But now consider the phenomenon of child-directed speech. In the years before my daughter was old enough to desire “sick” new trainers, I used to feed her baby food: pureed vegetables, rusks, and other assorted foodstuffs which I personally do not find even remotely appetising. Yet, it was common for me to feed her and to accompany the process with enthusiastic utterances of “yummy.” Intuitively, I was not expressing my positive attitude towards the taste of the food; rather I was expressing (or perhaps encouraging) my daughter’s positive attitude. A simple explanation of what is happening here is that the expressive “yummy” encodes the attitude of the speaker. But in cases like this, we have an implicit open quotation operator, which effects a context shift from speaker-attitude to the attitude of the quoted speaker. This is realised, as in the cases considered above, by an act of mimicry. By mimicking the reaction (or, perhaps, desired reaction in this case) of my daughter, I implicitly apply a form of open quotation to her utterances.
We encounter the same thing when we consider expressives with a truth-conditional component. The expression “eurocrat” is a mild, and slightly comical,9 pejorative expression used by anti-EU British politicians (and those who support them in this regard) to denote the politicians and their fellow officials who form the European Parliament and administer the bureaucracy of the European Union. An utterance of the sentence, “I think it is hilarious that Farage has to spend his time hanging out with all those eurocrats,” made by someone who obviously does not share Farage’s attitude of contempt towards the bureaucrats in question, is naturally read as an open quotation which would most perspicuously be represented as such:
(n) I think it is hilarious that Farage has to spend his time hanging out with all those “eurocrats.”
Interpreted thus, it is obvious that the pejorative force of the expression “eurocrat” has shifted away from the speaker-attitude to the attitude of Farage. The use of the term is again interpreted as mimicry.
Predicates of personal taste have also been noted as displaying similar behaviour. If Mary says “Rollercoasters are fun” and John says “Rollercoasters are not fun,” Mary and John are disagreeing faultlessly. That is to say that although it appears that one is asserting a proposition while one asserts the negation of that proposition, there is a sense in which both are speaking the truth, without either speaker misrepresenting the facts. Semantic Relativists like Lasersohn (2017) explain this by taking the truth of their utterances to be sensitive to a contextual parameter included in the circumstance of evaluation which ensures that the standard for truthful attribution of fun may differ between the two utterances. But now consider the case where Mary asks John, immediately following his rollercoaster ride, “Was that fun?”. Intuitively, the relevant standard here is not Mary’s but John’s. She is asking if he found it fun. Again, we can make sense of this scenario by understanding the evaluation of the attribution of the property encoded by “fun” to be relative to a parameter which is usually set to the speaker of the expression but in cases like this is shifted to the addressee. Again this can be understood as resulting from an implicit open-quotation device to mimic the addressee of the question. Note that mimicry here does not have to be a convincing performance, it simply needs to present the attitude or perspective of the target agent to whom the attitude is being attributed. I can do the same thing when I feed my pet guinea pigs some dried pellets of food and ask them, “Is that tasty?”. I do not need to be providing a convincing impression of a guinea pig to make it clear that the relevant standard of, and perspective on, tastiness here is that of my guinea pigs (or, at least, that which I attribute to them), not mine.
The above examples demonstrate that context-shifting is familiar for a range of expressions.10 What then of the content of indexicals? Can we provide examples where the same context-shifting operation shifts the reference of an indexical in the way that we expect MOs to do? In fact such examples are easy enough to find. First consider another example of child-directed speech. A nursery teacher, talking to a very young child who had her birthday the day before says: “Did mummy and daddy take you somewhere nice for your birthday?” Not only is the contextual standard for “nice” shifted to that of the addressee, but also the content of the terms “mummy” and “daddy” have shifted. These expressions behave very much like indexicals in that when uttered without qualification, they refer to the parents of the speaker. But here, the only qualification arises as a consequence of the nature of the context. That context generates a construction best understood as an open quotation: “Did ‘mummy’ and ‘daddy’ take you somewhere ‘nice’ for your birthday?” where the open quotation operation shifts the context away from that of the speaker to the addressee in order to fix the content of the quoted expressions.
Of course, it might be replied that this example can equally be explained by appeal to ellipsis. It might be thought that the indexical-like features of “mummy” and “daddy” are best explained by appeal to an elided possessive marker “\(\alpha\)’s mummy,” which may be an obviously indexical possessive pronoun “my mummy,” “your mummy,” “her mummy,” etc. Be that as it may, there are other cases which make it perfectly clear that indexicals can be shifted by open quotation. Indeed we saw one above from Récanati, which I repeat here:
(o) Stop that John! “Nobody likes me,” “I am miserable” … Don’t you think you exaggerate a bit?
It is clear that the referent of the “me” and “I” in (o) is not the speaker of (o) but the person they are mimicking, namely John. Examples like (o) are not uncommon and are a clear example of the use of open quotation as an MO that shifts the context that the indexicals contained within it are indexed away from the parametric settings of the overall sentence to another context for those quoted segments of the utterance. Using subscripts to display the relevant indexes, the situation is something like this:
[Stop that John! “[Nobody likes me]c\(_2\),” “[I am miserable]c\(_2\)” … Don’t you think you exaggerate a bit?]c\(_1\)
John is the addressee of c1, and not the agent; he is the agent of c2, not the addressee.
Notice that open quotation is being considered as an explicit (albeit metalinguistic) operation in the above examples. The operator (quotation marks) is ambiguous between open and closed quotation producing functions but it is explicit in the syntax of the written language. But, of course, utterances are not always inscribed. Except in rare cases where quotation marks are “signed” by a gesture which conventionally signals that the words uttered contemporaneously with that gesture are being quoted, it is up to hearers to identify quotation from features of the context. Pragmatic aspects of utterance interpretation come to the fore in such situations. Consider the following pairs of utterances:
(p) That guitarist, John, is performing tonight.
(q) That guitarist, John, who can’t actually play the guitar to save his life, is performing tonight.*
The awkwardness of (q) (indicated by the *) follows from the apparent contradiction which results from simultaneously describing John as bearing a property and then denying that he bears that very property. Were one to hear an utterance of (rather than read an inscription of) (q), however, one would most likely apply a principle of charity and interpret the utterance in a way which resolved this potential infelicity, such as (q1):
(q1) That “guitarist,” John, who can’t actually play the guitar to save his life, is performing tonight.
In other words, the term “guitarist” is interpreted as being subject to an open-quotation operator, shifting its usual extension to one that includes John (who is exempt from the extension of the standard English term). It may be read as synonymous with “so-called ‘guitarist’,” hence behaving much like the term “sick” discussed above: the target of a metalinguistic operator that shifts the idiolect (or other metalinguistic feature) according to which it is interpreted.
This process of pragmatically guided utterance interpretation need not apply only to subsentential elements but can equally be applied to whole sentences. A few years ago, there was something of a craze for purchasing audio recordings to be played as an answerphone message. A popular recording, used as an example above, was the snippet of Arnold Schwarzenegger’s character from the Terminator movies uttering the line “I’ll be back.” If I call my friend and hear this message, I do not interpret it as expressing the proposition it was originally used to encode. I interpret it as saying that the person who I have called is temporarily absent and soon to return. It is, in fact, interpreted as if the person being called were able to respond to my call from their current location and say “As Arnold Schwarzenegger says: ‘I’ll be back!’.” In other words, I understand the utterance as displaying the utterance made by Schwarzenegger and recruiting it to communicate information. I interpret it in precisely the same way as an instance of open quotation. The message is interpreted in just the same way as I would interpret the utterance of my friend who explicitly mimicked Schwarzenegger’s character, monotone pronunciation (perhaps even accompanied by distinctive bodily movements) and all, when in my presence.
Of course, mimicking an iconic actor or fictional character, by uttering an iconic line from an iconic movie is one thing, but what about ordinary answerphone messages, written notes, etc.? Who, or what, is being mimicked in these cases? Mimicry in these cases is more mundane but best understood as mimicry nonetheless. All that happens in these cases is that the speaker mimics themselves saying what they would say, were they able to inhabit the impossible (that is, improper) contexts they would need to be in to otherwise communicate this information. That is to say, when one needs to communicate information from a context unavailable to them, one must find an alternative method of relaying the information. By preparing in advance a message to be retrieved by ones intended audience in this context one is able to overcome this obstacle. But this is achieved not, as is often assumed, by somehow making an utterance “from a distance” but by recording in one form or another an instance of oneself performing the speech act one would want to make at that context if able to, ready to be displayed there. In doing so, one does not encode the proposition that would be obtained by pairing the uttered sentence with the improper context in question, but simply prepares a string of linguistic material that mimics the intended performance and then exploits the various media which permit this mimicry to be planted in advance ready to be deciphered when encountered. What is deciphered is not an utterance by an absent agent, it is a previous utterance deliberately placed in a situation where it will pragmatically trigger a process of interpretation precisely akin to that by which we read utterances of open quotation as discussed above. There is no fundamental difference between my employing open quotation to use an instance of the indexical “I” to pretend to be someone I am not, and my employing open quotation to use an instance of the indexical “here” to pretend to be speaking at a location where I am not. DCs are not utterances made at a distance, they are recorded performances pragmatically recruited to mimic intended and otherwise impossible utterances. This distinction is far from trivial: it demonstrates that DCs are generated by pragmatic features of communication and are thus not data to be accommodated by semantic theory.
The only difference between the uses of open quotation that I have discussed above and the full-blown DCs is that the syntactic role of quotation to mark off the shift will clearly be uncalled for in the latter case. Accordingly, we should not expect quotation to be readily recoverable in a case where a DC consists entirely of a mimicked performance, whereas it is essential for embedded occurrences like we see in “Amy has asked me to get her a ‘sick’ new pair of trainers.” Only when DCs are embedded would explicit quotation marks be felicitous, as we saw with “As Arnold Schwarzenegger would say: ‘I’ll be back’.” Other DCs would need to be placed in similarly embedded constructions to achieve the same result. For example: “If I could speak at the context where you will hear this message I would report that, ‘I am not here now’,” and so on.
5 Semantics, Pragmatics, and Displaced Communications
In this section I want to briefly say a few things in defence of the view presented above and to point out its advantages over competing accounts of DCs. In recent years the most vocal and influential critic of Kaplan’s account of logical truth for indexical languages has been Stefano Predelli. Although couched within a position sympathetic to Kaplan’s semantic project, Predelli takes issue with Kaplan’s decision to limit the possible combinations of sentences with contexts to proper contexts. Drawing on the DCs discussed above, Predelli argues that Kaplan is simply wrong to assume that utterances require their agents to be located at the times and places they occur. Furthermore, he offers an ingenious proposal as to how an extension of Kaplan’s theory to include improper contexts can be motivated and put into practice.
According to Predelli (most comprehensively in 2005) no semantic theory for indexical languages can be complete unless it has the resources to accommodate the role of speaker intentions in fixing the parametric settings of indexicals as uttered. In particular, Predelli maintains that speaker intentions are crucial to determining the context with which an indexical sentence must be paired in order to correctly model the actual utterance. Presented thus, Predelli’s position may not sound particularly distinctive—challenges to the attempt to model meaning by formal semantics alone without recognition of the role played by the speaker intentions behind the utterances whose meaning we are attempting to model are common from those who maintain that a theory of pragmatics is needed to explain linguistic meaning. The novelty of Predelli’s position however rests on his desire to reconcile his approach with a philosophy of language that assigns the core role of explaining meaning to formal semantics. To bring about this reconciliation of speaker intentions and semantically assigned meanings, Predelli draws a distinction between the workings of a formal system which calculates truth-conditions for utterances and a “pre-semantic” arena in which the inputs to this system must be first determined. It is in this latter area that speaker intentions become significant. Before I can employ a formal system to calculate the truth-conditions of a speaker’s utterance, I first must determine which utterance she has made (which proposition, in other words, she has said). To take a trivial example not involving indexicality, I can only know the truth conditions of an utterance of the sentence “John is sitting beside the bank” if I know which lexeme the ambiguous English word “bank” encodes in that sentence. So I must determine which lexeme the speaker intended before I input her utterance into the formal semantic theory which then returns its truth-conditions as output. The same thing happens when one utters an indexical sentence, according to Predelli, but now the pre-semantic task is to determine the correct context that this sentence must be paired with and, as with the case of disambiguating a lexical ambiguity, the only correct answer here will be that which identifies the speaker’s intention.
Certainly there is much to agree with in Predelli’s account, and its subtleties are not always recognised by his critics. For example, the objection often directed at Predelli that he is guilty of Humpty-Dumptyism11 is misguided. Humpty-Dumpty does not resolve ambiguities or select contexts to pair indexicals with, he simply rejects on whim the existing semantic assignments given to the elements of his vocabulary and selects alternative ones again on whim. Humpty’s speech is thus effectively unreadable to any semantic system employed by anyone other than himself. He scrambles the inputs to semantic theory into a code known only to himself. Thus communication breaks down. Applying this analogy to indexical terms, Humpty would be guilty of modifying the characters of indexicals at whim. We would not know which function from contexts to contents was encoded by his use of, e.g., “I” and hence could not calculate its content. But, as Predelli (2005, 58) explicitly states, characters are left untouched by the pre-semantic task of sentence-context pairing. Agents of answerphone messages are not using “I” to pick out anyone other than the agent of the context, they are simply selecting an improper context to pair their use of the term with.12
Predelli will therefore reject my interpretation of DCs as MOs.13 There is no operator on character according to his view, only a selection of a context that we have been wrongly denied in Kaplan’s theory. Dropping Kaplan’s restriction to proper contexts allows us to accommodate DCs as respectable utterances made “at a distance.” What, then, is to be said in favour of my view over Predelli’s? I think that Predelli’s view, for all its ingenuity, suffers a number of drawbacks that my account is not prone to.
Firstly, as argued in detail by Stevens (2009), Predelli’s position stands or falls on the strength of the intuition that DCs really are cases where utterances are made at a distance.14 But this intuition is fragile and sensitive to varying examples. The presentation of the data makes a difference to the intuition. For example, while it is true that when I phone Amy and hear her recorded answerphone message “I am not here now, please leave a message,” I understand that this expressed a prior intention on her part to communicate to anyone who hears the message the fact that she is not present at that later time, there are features one would expect an utterance to have which are lacking in this scenario. For example, it would be very odd of me to accuse her of lying, or even unwittingly telling an untruth, if I knew that she was in fact present at the location of her answerphone when I called her. The appropriate things to say in such a case would be something like “your answerphone is wrong/misleading/in need of updating, etc.” not “you are wrong…”.15 Similar points can be made about all DCs. We intuitively recognize a gap between these devices of communication and ordinary utterances, but this gap goes intrinsically unrecognised on Predelli’s intentionalist account, according to which the DC is a straightforward utterance.
Secondly, Stevens (2009) also points out that the intuition that a DC is a genuine utterance appears hard to reconcile with the equally strong intuition that an utterance is made at the time of encoding of the message. To, as Sidelle (1991, 535) puts it, engage in “arranging to make an utterance at a later time, or, if one likes, deferring an utterance,” is not to engage in uttering something while making those arrangements. This is especially clear in Predelli’s case. If the context of utterance is the intended context of utterance, then I utter nothing at all when recording my answerphone message; I simply get things ready for an utterance to occur later on. Perhaps this is so, but insofar as the position is motivated by our intuitions regarding DCs, this counterintuitive consequence counts against Predelli’s account. On my account, however, it can be explained easily enough. One is simply engaging in an act of pretense when recording the message, mimicking what one would say if located at the time and place of the context at which our intended audience will hear our performance. Unlike Predelli’s account, this entails no claim about utterances being displaced from their proper contexts. The only utterance that takes place is that which is made when recording the message, although it is not uttered with assertoric force; it is simply the product of an act of mimicry, ready to be displayed in a different context. The underlying intuition that motivates Predelli’s intentionalist account, namely the intuition that I am deliberately aiming to communicate things at contexts other than the one in which I am located, is preserved without endorsing the counterintuitive consequences of construing this as a form of utterance at a distance, by accommodating that intuition within a purely pragmatic explanation.
Technology can open up the possibility of previously outlandish uses of language but this is best explained pragmatically, not through a reconstruction of an otherwise perfectly acceptable semantic theory. Consider a recent technological advance which facilitates an unusual application of indexicals: The Rock Group, Dio, recently performed a series of concerts in which their deceased vocalist Ronnie James Dio was replaced on stage by a hologram. The hologram appears to be singing as it mouths along to pre-recorded vocal tracks from Ronnie. This holographic rendition of Ronnie, convincing though it may appear, is not of course really singing. The hologram is not causing any vibrations in the air, picked up by a microphone, etc. It is just a visual representation of a dead person, carefully synchronised with recordings of that person’s voice. But we can exploit this pretense to the full. When performing in London, we can make our holographic Ronnie “say” things (i.e., mouth along to recorded utterances of Real Ronnie’s) like “it is great to be in London tonight!”. Of course, Ronnie himself is not saying anything; Ronnie, unfortunately, is dead. Suppose that on the evening of 20th December 2019, the hologram is made to “say” this sentence: “It is cold tonight in London!”. Has the proposition that London is cold on the evening of 20th December 2019 been expressed? It seems reasonable to agree that it has, although it is equally obvious that Ronnie was not the agent who expressed that proposition (the most likely agent, or agents, would be those responsible for generating and controlling the hologram). But what about a case where Holographic Ronnie “says” something using the first-person pronoun like “I am so happy to be here with you in London tonight!”. Again, I think it is obvious enough that Holographic Ronnie has not said anything (not being an agent, he cannot be the agent of an utterance after all). Nor, for that matter, has Real Ronnie said that he is happy to be in London on the evening of 20th December 2019 (not being alive, he is not able to be an agent and hence not able to be an agent of an utterance). But this seems hard to square with Predelli’s view, according to which the parametric settings that determine the content of an utterance are fixed by intentions. In this case there clearly is an intention, just not an intention on the part of Ronnie (Holographic or Real). But whoever produced the hologram (let us assume it is a single individual for simplicity’s sake) had an intention to combine a recorded utterance of Real Ronnie’s (perhaps reconstituted from several samples taken from previous utterances and hence not identical with any one previous actual utterance) with a visual representation of Real Ronnie to ensure that Holographic Ronnie “said” that he was happy to be in London on the evening of 20th December 2019.
There does not seem to be any semantic difference between what is happening here and what is happening if I attached a note authored by Jones which says “I am not here” to Smith’s door to express Smith’s absence. The agent of the note, on Predelli’s account, is presumably Smith because it is his absence I intend to communicate. Accordingly, Predelli’s account predicts that the agent of Holographic Ronnie’s “utterance” is either Holographic Ronnie or, perhaps, Real Ronnie (depending on which of these two, if distinguished, the producer of the holographic performance intended). This, I think, cannot be the right thing to say in this situation. No amount of intention can make dead people agents of utterances after their death.16 Surely what we have is simply a case where someone is doing an extremely sophisticated job of pretending that Ronnie James Dio is present and performing on stage by displaying recordings of his previous speech in an act of mimicry. This, in my view, is what we find in all cases of DCs. Ordinary utterances with Kaplanian meanings are employed to allow us to pretend to say (or pretend that others are saying) things unavailable to us when the utterances are construed literally. There is no need to modify our semantic theory to accommodate a theory of pretence.17
One final approach to the answerphone problem that seeks to accommodate DCs within a wholly semantic framework is suggested (though not endorsed) by Parsons (2011). The view merits brief consideration here as, again, it shares some similarities with my proposal but the differences are significant. I have argued that apparently true instances of “I am not here now” etc., are not really true. They are false utterances made by speakers who utilise a pragmatic process to facilitate their non-literal interpretation as pretences. Speakers are relying on context to allow them to mimic utterances at different temporal or spatial locations (or even by different agents) because that context will make those shifted contexts salient (most routinely because those shifted contexts are the ones that the hearer will be in when they decode the utterance).
It is helpful to compare my view to a radical form of what we might call “content relativism” (CR). CR is the position whereby the content of an utterance is subject to modification depending on the context in which the utterance is assessed (rather than the context in which it is uttered).18 In Kaplanian terms, the context in which an utterance is assessed is the circumstance of evaluation. Whereas Kaplan takes circumstances to be world-time pairs, CR expands the parameters to include all those parameters standardly recognized as elements of contexts of utterance (agents, times, places, etc.). Whereas circumstances of evaluation are usually appealed to in determining truth-value, CR allows them to determine content. Hence the same utterance can change its content (express a different proposition) if the context in which it is assessed changes.
Parsons (2011) considers the possibility of appealing to CR as a way of providing a semantic theory for answerphone cases. On the surface, the suggestion is promising: the utterance of “I am not here now” on the answerphone strikes us as intuitively true, despite the fact that it cannot be true if that sentence is contradictory. But, of course, CR will abandon the claim that it is contradictory, because that claim relies on the belief that the content of the sentence is tied to a proper context of utterance. CR can agree with Kaplan that all contexts of utterance are proper but take a more relaxed view on contexts of assessment, allowing these to impact the content of the sentence uttered in ways that break the tie with contexts of utterance. Hence we have a neat explanation of how an utterance of “I am not here now” can express a truth: although the sentence cannot be true when uttered, it can change its content depending on the context of its assessment so as to become true.
I think there is something right about CR, but we need to be careful about endorsing it as a semantic theory. The problem is that there is nothing systematic about the behaviour of indexicals which tells us in advance whether they are assessment-sensitive or not. Answerphone messages are assessment-sensitive, ordinary utterances of indexical sentences tend not to be. Or, to be more precise, indexical sentences uttered in certain conventionally recognised scenarios are routinely interpreted in accordance with the predictions made by CR, while most utterances do not demand such elaborate mechanisms to interpret them. Of course, we might just maintain that CR applies uniformly to all utterances but that the default interpretation is one where the context of assessment coincides with the context of utterance. Only in certain cases does the context trigger a bifurcation of context of assessment from context of utterance. I see no problem with that view, but it clearly demonstrates that the semantic theory by itself does not do sufficient explanatory work. A pragmatic account of the way in which the interaction of context of utterance and context of assessment is triggered is essential to such a story, and this is what I have attempted to provide in this paper.
One thing that suggests that a CR-based semantics alone is not sufficient to explain displaced communication is that (as we have seen many times in the discussion in this paper) our intuitions are highly unpredictable and subject to the details of the contextual situation. Parsons takes this concern to show that CR cannot explain the answerphone problem. He imagines a case where a time delay on the phone line results in someone hearing the answerphone message after the speaker has in fact returned home. With some reservation, he endorses the view that the message is still true, and is (he claims) able to shift the context of assessment away from the time of decoding to the time of intended decoding. Parsons himself confesses to being unsure of his intuitions in regard to this example. It seems to me to be another case like those I considered previously which just show that we do not have firm intuitions about displaced communications. But without firm intuitions to make concrete predictions about what it meant and what is true or false, the task required of a semantic theory cannot be fulfilled. The situation can be seen quite clearly by reconsidering the holographic Ronnie scenario that I posed as an objection to Predelli’s approach above. It is clear that using a holographic image of Ronnie requires some rich stage-setting to work. It is only because of this stage setting that the intended content (the pretence, as I have argued) is made available. We can recognize a semantic value which is interpreted relative to the context of assessment for the utterance. This will make sense of our intuition that some of Ronnie’s apparent utterances at least sound like they are true (“it is raining in Manchester tonight,” for example), while others don’t sound true (“I am happy to be here tonight” doesn’t sound true when we know that Real Ronnie is both dead and played no conscious role in this utterance). Consider Holographic Ronnie’s production of “I am Ronnie James Dio”—is this true at the context of assessment? According to CR it ought to be possible that the agent really will be Real Ronnie. And Real Ronnie really is Ronnie James Dio. So the utterance should be true. But I don’t have the intuition that this utterance is true—or, rather, I’m not sure that I have any intuition about this sort of case. Intuitions are just not stable in cases like these. And unstable intuitions are not suitable foundations for a semantic theory.
6 Objections and Replies
In this final section, I will consider some objections to the view that I have presented above, and offer some replies, which will hopefully help to clarify my position.
The first objection I want to consider concerns my definition of DCs. A DC is a communication that occurs at a different context to that in which it is encoded. It is tempting to assume (as seems to be the case for each of the examples considered so far) that DCs are always evaluated with respect to the context in which they are decoded (hence, on my view, the monstrosity present in the pragmatic operation facilitating DCs) But what about cases like we see in the following pair (both, imagine, recorded for a posthumously broadcasted will):
(r) Today, I met with my lawyer before recording this will.
(s) Today, you all received a call from my lawyer informing you that you have inherited a large sum.
It seems that both (r) and (s) are clear cases of DCs as commonly discussed in the literature, yet only (s) seems to communicate information that is evaluated with respect to the context in which it is decoded. Far from being monstrous, (r) seems to communicate information about the context of encoding. But is this not a DC?
I do not think that (r) is a DC. While (r) is being used to communicate information at a context subsequent to that in which it is encoded, the information is about the context of encoding. The indexicals “today,” “I,” “my,” and “this” all contribute contents drawn from the context in which the message is recorded. Furthermore, I am sceptical that a construction like (r) could be developed in such a way as to be coherently understood as communicating information about the context of decoding. For example, continuing (r) in the following way, sounds infelicitous to my ear:
(r*) Today, I met with my lawyer before recording this will that you are now listening to.
If we understand this as an attempt to shift the temporal parameter of the utterance from that indicated by “today” to that indicated by “now,” mid-sentence, I think the sentence can only be made sense of if we read an implicit open quotation as present on the “now.” Only in such cases, I suggest, do we have a candidate for a DC. Simply presenting a recording of a message is not sufficient to produce a DC. Only when that message is naturally interpreted as communicating information about the context in which it is decoded, rather than encoded does it count as a DC. If I uncover a forgotten recording from my 10th birthday in which I say “I am 10 today,” I do not stumble on a DC. But if I uncover a recording of my 10-year-old self, saying “when you hear this, you will suddenly remember recording it when you were 10,” I do. It seems to me that (r) is akin to the former, not the latter.
The second objection arises when we consider a very large class of cases of potential DCs that I have said little about above, involving the production of signs containing indexicals. Consider a sign positioned in a hospital waiting room that consists of an inscription “please wait here.” This sign exhibits typical features of a DC as “here” will be naturally interpreted as referring to the location of installation, not of inscription. Tokens of the type of this sign are mass-produced in a factory. Some individual factory worker produced this particular token sign. But, surely, the producer of the sign in this case is not the agent of any instruction. The factory worker is simply a component part in the production of a communication that intuitively seems to occur at the time of decoding. This potentially casts doubt on my claim that DCs are the result of MOs operating on an utterance or inscription evaluated with respect to a proper Kaplanian context.19
I agree that it is implausible to construe the factory worker as the agent of the instruction inscribed on the sign. I am, however, unconvinced that any instruction as such is made in the factory. We should not be misled by the fact that human agents can be involved in the production of an artefact that carries information into inferring that they are the agents of whatever information is thereby transmitted. In this instance, the factory worker is no more the agent of a communicated content, than Stephen Hawkins’s voice-synthesising computer is the agent of his utterances when he relies on it to communicate his thoughts. The factory worker is producing another agent’s message in accordance with their instructions. Who, then, is the agent who desires to communicate the information? The agent here is the hospital (or relevant hospital authority). We can avoid complicated metaphysical questions about how organisations might be agents by assuming an individual consultant, Ms Smith, is the relevant authority. Ms Smith wants to ensure that patients arriving in reception wait in an orderly fashion in the waiting room. One way that she could do this would be to write a sign in her own hand saying “please wait here” ready to be displayed at the waiting room, or utter the sentence “please wait here” into a recording device to be on looped playback in the reception. But, due to the frequent reoccurrence of episodes when consultants need to instruct patients to wait in a specific location, it is of course more practical for signs to be mass-produced rather than produced by flimsy hand-written notes. Hence she orders a batch of ready-made signs designed to meet this common need among consultants. Nonetheless, Ms Smith remains the agent of the instruction. She has simply exploited a labour-saving device that ensures that one factory worker produces signs for the large number of agents who want to issue this instruction. Once she is in possession of the sign, she can exploit the convention that signs routinely signal information about their spatial location to engage in pretence of the sort her hand-written note would exploit.20 The difference in the method of production of her message does not alter the fact that she is the agent of the utterance and its displacement is the result of a metalinguistic pretence, not a deferral of her utterance.
Another objection responds directly to my analysis of open quotation as an MO. An obvious feature of open quotation is that, even if the quotation operation is not explicit, it should be easily recoverable. Consider this exchange from the movie The Empire Strikes Back. Lando Calrissian has double-crossed Han Solo and his friends, betraying them to Darth Vader and the evil Empire. However, he strikes a deal with Vader to preserve the freedom of Solo’s friends. Informing Solo of the deal, he says “I”ve done all I can. I’m sorry I can’t do more, but I’ve got my own problems.” Solo sarcastically replies: “Yeah. You’re a real hero.” It is obvious enough how we might appeal to an open quotation analysis of this ironic utterance. Solo is not expressing his own admiration for Calrissian (he has in fact just punched Calrissian in the face, unequivocally expressing his real attitude). He does not mean that Calrissian is a hero, rather he is mockingly echoing the use of this term of praise to display his own distance from such a perspective. One thing that obviously stands in favour of the open quotation analysis is that the recovery of the operation as an explicit one is natural. One might very well report Solo’s speech as “Yeah, You’re a real ‘hero’.” Many of the commonly cited cases of DCs in the literature on indexicality, however, do not seem to be so neatly reconfigured with explicit quotation marks. Consider Predelli’s note from section 1:
I am not at home now. If you hurry, you’ll catch the evening flight to Los Cabos. Meet me in six hours at the Hotel Cabo Real.
It would not be natural to add quotation marks to the shifted indexicals in the note (I add the “#” to indicate the marked quality of this):
#I am not at home “now.” If you hurry, you’ll catch the evening flight to Los Cabos. Meet me “in six hours” at the Hotel Cabo Real.
The note, if anything, becomes quite confusing once the quotation marks are made explicit. Why is this, and how can it be the case if a DC is really generated by open quotation, in the same way as Han’s response to Lando?
A key difference is apparent in these two contrasting cases that explains why quotation is not recoverable in the second case. The first case involves the shifting of a sub-sentential element within a context that remains non-shifted. The second case involves the shifting of the entire sentence for its interpretation. But open quotation is linguistically employed for the first kind of case only. Recall that open quotation is appealed to on my account as a way of making explicit a form of mimicry. In a case where a single expression, or string of expressions, contained in a wider linguistic frame are employed in this mimicking role while the wider frame is not, open quotation serves to explicitly indicate this role. When it is an entire sentence or other self-contained linguistic item, this device serves no purpose. Mimicry shifts the context to create a DC. Only when mimicry is embedded within a non-shifted context is explicit quotation required to indicate this. This is why, for example, it makes no sense to add quotation marks to this message:
(t) “I’ll be back”
But they are clearly useful in:
(t*) As Arnold Schwarzenegger would say, “I’ll be back.”
In Predelli’s note, we cannot capture the mimicry that I take it to employ by only quoting “now” and “in six hours,” because they have not been shifted in relation to the remainder of the note. It is the whole note that mimics what the note writer intends to communicate by their pretence. My claim is that the same sort of pretence lies behind both cases, but only when the mimicry is embedded within a wider frame is the quotation device demanded to make this explicit.
The final objection I want to consider is a methodological one that, I think, goes to the heart of the different approaches to this problem taken by Predelli and those who, like myself, have urged a pragmatic explanation of DCs. The intuition that supports the pragmatic explanation of DCs is that distinctive features of the concrete episodes of language use that generate DCs are responsible for them. These features point to ways in which DCs are heavily reliant on a rich background of human behaviour that goes beyond the remit of semantic theory to explain. Like other aspects of communication that are accepted as requiring an explanation from pragmatics such as conversational implicatures, we need to look beyond the literal meanings of the expressions employed to understand what is happening in these cases. But, insists Predelli (2005, 2011), this approach both misunderstands and underestimates the place of semantics. It fails, in his view, to respect Kaplan’s advice that we base our semantic theorizing on “the verities of meanings,” not “the vagaries of actions” (1989a, 585). Indeed, Kaplan’s own insistence on admitting only proper contexts is a failure to follow his own advice, according to Predelli (2005, 60–62). The view that Kaplan’s restriction on contexts strays from is summarised elsewhere by Predelli (2011, 301) like this: “[S]emantics is concerned with the evaluation of sentences with respect to contexts, not with concrete episodes of language use—it is concerned with ‘utterances’ only in Kaplan’s technical sense of the term as sentence-context pairs”. To illustrate the significance of drawing this methodological line, Predelli gives the example of a tautology that is sufficiently long that no mortal human will ever utter it. As a concrete episode of language use, we do not have an utterance. But clearly it is unproblematic to evaluate the sentence as true with respect to any context of “utterance,” in the more careful Kaplanian sense of a sentence-context pair (see 1989b, 522–23). Failing to respect this distinction, and being misled by the peculiarities of how we use language to perform speech acts in particular situations, Predelli maintains, inevitably leads us to the wrong conclusions.
While I agree with Predelli that we ought to recognise the distinction he makes, I do not agree that there is a methodological decision to be taken here that will remain neutral with regard to our intuitions concerning concrete episodes of language use. Keeping the discussion focused purely on the issue at hand, one can of course construct a formal semantic theory that is more generous in the contexts it evaluates sentences with respect to than the proper contexts endorsed by Kaplan. One is limited only by mathematical constraints in this regard. But eventually one has to make a decision about which of those mathematical possibilities correspond to our actual use of natural language expressions, if the mathematical structure in our formal semantics is going to be empirically adequate as a model of the semantic profile of an actual expression or set of expressions in a natural language. Kaplan’s decision to restrict the range of contexts we should be interested in to proper ones is, I take it, based on this desideratum. After noting that an unconstrained range of contexts will provide contexts with respect to which “I am here now” is false, he insists that only the proper ones should be admitted if we are to arrive at an empirically adequate analysis of the indexical expressions contained in this sentence. To repeat the quotation I gave at the beginning of this paper: “[I]mproper indices are like impossible worlds; no such contexts could exist and thus there is no interest in evaluating the extensions of expressions with respect to them” (Kaplan 1989b, 509, emphasis added). As the quotation shows, while Predelli is quite correct to point out that a formal semantic theory should be founded on an abstract pairing of expressions with mathematical objects within a formal structure, we will have to make choices about which pairings are of interest to our concerns as natural language semanticists, and these choices will surely be based on our intuitions about the way the expressions behave in the mouths, pens, and thoughts of ordinary speakers. My own intuition, following Kaplan’s, is that the restriction of contexts to the set of proper contexts best captures the semantic behaviour of indexicals in English, once we recognise the input of pragmatic processes on apparent deviations from this restriction. Predelli’s intuition is to take the deviations to illustrate that Kaplan’s restriction is empirically inadequate. I do not think that either of us is basing our choice about which contexts our semantic theories should recognise on issues independent of intuitions about concrete episodes of language use, nor do I think that we should.
My argument in this paper has been the following. Firstly, I have argued that DCs are best understood as being generated by MOs. I have then argued that, understood as MOs, they are in turn best understood as the result of pragmatically triggered metalinguistic context-shifting operations. I have then given a detailed explanation of this proposed mechanism. If this is correct, then DCs are MOs but are not Monsters, for, while DCs are certainly monstrous, their monstrosity is not generated by any lexicalised semantic operator of English. Furthermore, the argument presented here is also intended to vindicate Kaplan’s insistence that the only proper contexts relevant to the semantic evaluation of English indexicals are those which situate the agent at the time and place of her utterance. DCs are not “utterances at a distance” which result from making utterances in improper contexts; they are ordinary utterances made in the course of a deliberate pretence that they are something more.
Some notable exceptions are discussed below.↩︎
An insistence that all indexicals must be directly referential is misplaced if we take the definition of an indexical simply to be that its character is a non-constant function from contexts to contents. There seems no principled reason to exclude quantifiers, predicates, or unarticulated constituents such as those which are regularly posited to fix the comparison class for gradable adjectives, from this list. King’s (2001) quantificational analysis of complex demonstratives, for example, treats such expressions as indexical quantifiers. Similarly Kaplan’s formal language \(LD\) contains an indexical temporal operator in place of a referring expression as its correlate of “now.” Nonetheless it seems obvious that most paradigm cases of indexicals and demonstratives (“I,” “here,” “this,” etc.) are directly referential.↩︎
Kaplan only says that monsters are absent from English but he is often interpreted as making the wider claim that they are absent from natural languages generally, a claim challenged by Schlenker (2003), who appeals to empirical data concerning Amharic to support the view that monsters are present in some natural languages. For a detailed discussion of Kaplan on monsters, see Predelli (2014).↩︎
It is denied by Stevens (2009). Cohen (2013, 8, fn 8) rightly points out that a major shortcoming of Stevens (2009) is my lack of a positive proposal in place of this interpretation. In particular, no detail is given as to how a pragmatic explanation could explain apparent utterances at a distance, so as to make the idea of a deferred utterance redundant. A proposal like the one I will be offering in this paper is going to be required if we are to reject the deferred utterance analysis. Others have offered pragmatic proposals that share some features with my approach, including Connolly (2017) who also diagnoses the counterexamples as involving some form of pretence, although his analysis of how this is effected takes a very different line to mine. Åkerman (2017) also provides an alternative account of how pragmatic processes can be appealed to in our explanation of apparent cases of context-shifting.↩︎
I prefer the term “displaced communication” to “deferred utterance” as it is not limited to utterances, but can include any communication of information.↩︎
“Pure” expressives like “ouch” and “oops” appear to make no truth-conditional contribution to utterances and to simply encode a speaker attitude, whereas, e.g., pejoratives (including racial or sexual pejoratives) are often thought to encode both a truth-conditional content (an extension, namely those who the speaker intends to denote by the term) and a speaker attitude (of derogation towards the members of that extension).↩︎
Admittedly, its comedic quality has been somewhat diminished since the UK’s recent referendum result.↩︎
Of course, not all will share my analysis of these cases as instances of open quotation. For example, irony of the sort displayed in examples like (m) and (n) may inspire competing analyses from Griceans. I do not have space here to mount a detailed defence of my analysis of irony and related phenomena, but hope to have made it clear that the analysis is a plausible one for a range of phenomena that are importantly similar to the cases we are concerned with. As well as drawing on Récanati’s approach, my analysis has some similarities with the echoic analysis of irony and related phenomena adopted in Relevance Theory (see, e.g., Wilson 2006). The Relevance Theoretic approach is applied by Bianchi (2014) to echoic uses of slurs. An important point, emphasized by her, is that when we echo or imitate the perspectives of others, we do not have to extend the echoing to the whole content of an asserted proposition—we often only express a perspective distinct from our own with regard to a part of that proposition. For example, Bianchi interprets the sentence “As I reached the bank at closing time, the bank clerk helpfully shut the door in my face” as containing an instance of echo or imitation only with regard to the expression “helpfully” (2014, 39). This is the same feature that I am appealing to open quotation to explain in many of the examples above. Bianchi draws on this analysis to explain seemingly non-offensive uses of slurs, such as we see in appropriation (cases where the usual targets of a slur use the expression in a way that removes its derogatory aspect). While I agree that there are echoic uses of slurs such as (n), I would not extend this analysis to appropriation (my own account of appropriation can be found in Scott and Stevens 2019); other examples of echoic uses of slurs and of expressives more generally are discussed in Stevens and Duckett (2019).↩︎
“Humpty-Dumptyism” is the pejorative term for a semantic theory of the bizarre and implausible sort envisaged by Lewis Carroll’s fictional character who insisted that his words mean simply whatever he wants them to mean. Responses to the charge are given in Predelli (2011).↩︎
An alternative source of the humpty-dumpty objection to Predelli that I have heard attacks the account on the grounds that it allows the speaker to pair a sentence with any context that they choose, hence their choices as to that pairing could, in principle, be just as private to them as Humpty’s choices about meaning assignments are to his idiolect. I don’t find this objection compelling—there is no obstacle to Predelli admitting that there are success conditions placed on successful communication that apply to the pairing of sentence and context just as there are for resolving lexical ambiguities. I can successfully encode a number of things by “bank,” but not just anything. My intended meaning will only succeed if it conforms to existing conventions about English usage. Similarly for the intentions I have about the contexts I pair my sentences with.↩︎
Predelli (1996) addresses the relation between monsters and DCs to some extent. Although in this paper Predelli does not consider all alleged cases of monsters, focusing solely on the famous “never put off until tomorrow what you can do today,” he explicitly appeals to his intentionalist framework to explain away the apparent monstrosity of this example by maintaining that it should be understood as encoding multiple intended DCs. Discussion of his account of this particular sentence takes us beyond the scope of this paper. For other interesting discussions of cases that seem to involve one and the same sentence expressing multiple DCs as it is decoded repeatedly, see Egan (2009) and O’Madagain (2014). See Predelli (2014) for further discussion of monsters.↩︎
Predelli (2011) responds to the objections raised by Stevens (2009) by characterizing those objections as founded on a mistaken conception of the proper role of semantic theory. On Predelli’s characterization, Stevens is denying that “the evidence put forth by true instances of ‘I am not here now’ [should] constrain the shape of an empirically adequate semantic account” (2009, 301). I agree that this would be a mistaken view of the role of semantic theory; however, it is a misrepresentation of the objection from Stevens, who clearly rejects the intuition that there are any true instances of “I am not here now” to be accounted for. Of course, Predelli is correct to note that the question of where the line should be drawn between semantic theory and pragmatic theory is a controversial one. I take the considerations in this section and the preceding one to lend compelling support to views like that forwarded in Stevens (2009) and Récanati (2010), according to which that line is decisively drawn in a way that makes pragmatic theory responsible for explaining DCs rather than semantic theory as Predelli maintains.↩︎
As a referee pointed out to me, intuition seems to shift back in the other direction in a case where there is a deliberate deception. Suppose that Amy does not want to speak to me and deliberately leaves her answerphone on so that I will think she is not there. Now the intuition that she is lying has more traction. I take this point. However, I am content to use the example to illustrate that our intuitions are unstable—the intuition that agents make utterances at places and times other than where and when they are situated is malleable in a way that the intuition that agents make utterances in Kaplanian proper contexts is not.↩︎
A similar objection to Predelli is raised by Sherman (2015, 594) who notes that Predelli’s intentionalist account makes apparently correct predictions about the cases where we have some choice over our use of indexicals, but struggles to explain cases where we don’t. The comment is made in passing but I assume he has in mind cases like this: a recently released addition to a range of ice cream has “I am vegan!” written on it. Predelli seems to have a simple explanation of what the “I” means here—whoever wrote this intended that it be paired with a context in which the ice cream is the agent. But now consider a case where I stand next to the freezer in the supermarket shouting, “I am vegan!” to passing shoppers, while intending the sentence I emit to be paired with a context in which the ice cream is the agent. Clearly, my intention will not be fulfilled. This suggests that there is more to the successful case than just the intentions of whoever produced the communication.↩︎
The idea that DCs can be explained as pretend utterances is also defended by Voltolini (2006) and Connolly (2017). Voltolini’s strategy is to situate his explanation within a fictionalist semantics, while Connolly shares my preference for a pragmatic approach. I have a great deal of sympathy with Connolly’s approach which construes DCs as produced by participants knowingly and deliberately entering into a game of “externally-oriented make believe” (2017, 616). However, while I think our approaches are in the same vein, I think the situation he describes must be supplemented by the sort of analysis I propose if it is to explain the monstrous quality of DCs. For example, I have argued that binding of indexicals (Climb now, work later), and embedded context-shifting (Amy has asked me to get her a “sick” new pair of trainers) have important similarities with DCs that require the sort of approach I am urging.↩︎
CR is a more radical theory even than the controversial forms of semantic relativism (or, as it is sometimes called “truth relativism”) developed most notably by Lasersohn (2017) and MacFarlane (2014). Semantic relativism holds that truth is sensitive to context of assessment; CR holds that what is meant by an utterance depends on the context in which that utterance is assessed. Despite the clear logical space for CR to exist within any semantic framework which admits both contexts of utterance and contexts of assessment, few have been persuaded that CR is worth exploring. A rare exception (in addition to Parsons 2011, discussed shortly) is Weatherson (2009).↩︎
Examples such as these motivate both O’Madagain (2014) and Briciu (2018) to distinguish between tokens and proper utterances. This distinction allows for the possibility of utterances at a distance by holding that genuine utterance requires the presence of illocutionary force, whereas the mere production of a token does not. I am inclined to agree that, in the example above, our factory worker is engaged in the production of a token, not an utterance but, as I now argue, I do not think that this means we must recognize the context of decoding as the context of utterance.↩︎
Not all signs are obviously about the location in which they are placed, or object they are attached to, of course. An object may well be emblazoned with the sign “visit [such and such website] to see full product range,” or a pair of running shoes may come in a box marked “consult medical professional before beginning any new program of exercise.” Such signs, while clearly connected to some salient object are not about that object. But there are clearly a multitude of cases where the convention does hold: “twist clockwise” on a food jar lid, “made in England” on a guitar amplifier, “serve chilled” on a beer bottle, “4m high” on a road bridge, etc., all refer to the object they are attached to. “No smoking” in a public building, “Slow Down” on a road sign, “Wear a face covering” outside a shop, etc., all refer to the location in which they are placed.↩︎