«Abstract Answering machines and other types of recording devices present prima facie problems for traditional theories of the meaning of indexicals. ...»
Indexicality and The Answering Machine
Eliot Michaelson†and Jonathan Cohen‡
Answering machines and other types of recording devices present prima
facie problems for traditional theories of the meaning of indexicals. The
present essay explores a range of semantic and pragmatic responses to
these issues. Careful attention to the difﬁculties posed by recordings
promises to help enlighten the boundaries between semantics and pragmatics more broadly.
The invention is of great importance for telephonic purposes, as by providing a suitable apparatus in combination with a telephone communications can be received by the apparatus when the subscriber is absent, whereas upon his return he can cause the communications to be repeated by the apparatus.
US Patent 661,619, for the Telegraphone – 13 November 1900 Indexicals (paradigmatically ‘I’, ‘here’, and ‘now’) have long served as a focal point for philosophical debates about context-sensitivity. Famously, Kaplan (1989) noted that these expressions exhibit more regularity than many other context-sensitive terms (e.g., demonstratives), and proposed a straightforward set of rules (‘characters’) for associating each with a semantic value, relative to a context. Thereby, he provided an intention-free formal semantics to account for indexicals.
Unfortunately, this simple and powerful picture is threatened by examples involving inscriptions and audio recordings. The best-known is the ‘answering machine paradox’: since there is no speaker when an answering machine is triggered, Kaplan’s theory predicts that answering-machine occurrences of ‘I’ fail to refer. Yet answering machines regularly and successfully communicate information about speciﬁc individuals.
∗ Thispaper is entirely collaborative; authors are listed in anti-alphabetical order.
† Department of Philosophy, University of California, Los Angeles, Box 951451, Dodd Hall 321, Los Angeles, CA 90095, email@example.com ‡ Department of Philosophy, University of California, San Diego, 9500 Gilman Drive, La Jolla, CA 92093-0119, firstname.lastname@example.org This essay explores a range of semantic and pragmatic solutions to this paradox. These proposals merit attention not only for their intrinsic interest, but also because of the larger issues they raise about the appropriate range of data for semantics and pragmatics, and about the division of explanatory labor between these components of our total theory of language.
1 Answering Machines and the Classical View As Kaplan (1989, 491, note 12) observes, the apparent fact that answeringmachine produced tokens of (1) communicate something true and informative about particular individuals conﬂicts with the now-canonical semantics for indexicals developed there.1 (1) I am not here now.
For Kaplan, the semantics of indexicals turns on the interaction of two aspects of their meaning — character and content — with the context of use (and, ultimately, with circumstances of evaluation).2 An expression’s content is a mapping from circumstances of evaluation to extensions (501). (For sentences, we can think of contents as “the what-is-said in a given context” (494) — something that maps from circumstances of evaluation to truth values, which are the extensions for sentences). Kaplan’s second kind of meaning, character, is described as “a meaning rule,” that which is “set by linguistic conventions and, in turn, determines the content of the expression in every context,” and a “function from possible contexts to contents” (505). Kaplan captures the context-sensitivity of indexicals by saying that indexicals have contextually-invariant characters that map to different contents as a function of the context: ‘I’ interpreted relative to a context whose agent is Barack Obama has one content, while ‘I’ interpreted relative to a context whose agent is David Kaplan has another.
Suppose, with Kaplan, a context c can be represented by a sequence of particulars including an agent cA, a location cP, a time cT, and a world cW.
1 The version of “Demonstratives” ﬁnally published in 1989 circulated widely from around
1977. Thus, despite claims to this effect by, e.g., Perry (2003, 376), it is false that Kaplan had, in constructing his theory, failed to anticipate the technological possibility of answering machines, or the threats they raise to his semantics. In any case, though the behavior of answering machines makes the clash vivid, the problem arises for journal entries, inscriptions left on blackboards, notes sent on postcards, or cases (such as one Kaplan (1989, note 12) attributes to Donnellan) where sound moves very slowly; and it is hard to believe (even if it were true that he had failed to consider answering machines speciﬁcally) that Kaplan had avoided encounters with the other pieces of technology that raise the problem — e.g., the pencil. Intuitively, what unites such cases and makes them problematic for Kaplan’s (otherwise successful) semantics is that the speaker who linguistically encodes the message needn’t be located at the place and time at which the message token is decoded by a recipient.
2 Crucially, the puzzle of the answering machine arises within Kaplan’s treatment of indexicals, which constitute a subset of the occurrences of the word-forms ‘I’, ‘now’, ‘here’, etc. Kaplan (1989, 489–490) clearly points out that the words that interest him have both indexical and non-indexical uses, and explicitly sets the latter outside the scope of his theory. As such, (pace Vision, 1985; Smith,
1989) non-indexical occurrences are not counterexamples to that theory.
Kaplan’s character axioms stipulate that ‘I’ refers to cA, ‘here’ to cP, and ‘now’ to cT ; and he holds that for any “proper” context c, cA must be located at cP and cT.3 It follows from these stipulations that (2) is a ‘logical truth’ within his semantics: it is true whenever tokened (509).
(2) I am here now.
And, given the usual semantics for negation, it similarly follows that (1) is a ‘logical falsehood’: it is false whenever tokened.
Thus, the received semantic machinery for indexicals apparently has as a trivial consequence the denial of a banal fact about natural language made manifest by ordinary 1970s consumer technology. Presumably, then, some modiﬁcation of that machinery is called for. But what?
2 Pragmatic Accounts Though most published discussions respond to the problem in semantic terms, a natural initial thought would be to offer a pragmatic solution.
Thus, one might supplement Kaplan’s semantics with standard Gricean mechanisms to explain the actual communicative impact of answering machine uses of tokens like (1). On such a view, the answering machine recording would literally say something false, but would communicate something true, something like: the original speaker isn’t at the time/place where this answering machine has been triggered. (Or, simpler still: the speaker isn’t available when you’re calling.) Such extrasemantic communicated content is typically explained in terms of conversational implicature.4 If so, two problems arise.
First, unlike paradigmatic conversational implicatures, answering machine
recordings of (1) cannot be canceled, as indicated by the infelicity of (1 ):
(1 ) # I am not here now, but I might be when you are listening to this.
The pragmatic theorist might attempt to bite this bullet, but this means giving up one of our best criteria for distinguishing semantic from pragmatic content.5 Second, the envisaged pragmatic story risks being self-undermining.
3 Kaplan’s reasons for these stipulations are complicated; for discussion, see Michaelson (2013b).
4 Pragmatic theorists might try to invoke conventional implicature instead; but that sort of Gricean content looks ill-suited to the task at hand. In particular, such content is generally taken to supplement what is said, strictly speaking — viz., to convey some further information on top of that content — rather than supplanting it.
5 Perhaps a pragmatic theorist might offer an independent explanation of the infelicity (1 ) by claiming that its cancellation clause (viz., its second conjunct) is either false or uninformative — that it either states something false about the speaker’s state at the time/place of utterance, or something trivial about the time/place of triggering. However, note that the cancellation clause of (1 ) can be informative in the right circumstances — such as when the caller had previously assumed that the listener is not the sort to screen her calls. We take it that (1 ) is still infelicitous even in these circumstances, which suggests that the proposed explanation of its infelicity is unsuccessful.
For if answering machine uses of (1) regularly and systematically convey this extrasemantic content, as per the pragmatic proposal, it is a puzzle why that content wouldn’t become conventionalized, hence semantic. That is, we worry that the current proposal that the association between (1) and the content in question is extrasemantic is potentially unstable — that, just by virtue of its prevalence and regularity, the association threatens to become semantic after all. If so, of course, then the pragmatic proposal is not, advertisements to the contrary notwithstanding, an alternative to the view that the content communicated by (1) is semantically encoded.6 In response, the pragmatic theorist might loosen her attachment to any particular pragmatic theory and claim only that, by some mechanism or other, answering machine machine uses of (1) standardly convey that the speaker isn’t available right now. But it’s hard to see why, if this is the content that answering machine uses of (1) convey, (1) cannot be used to convey the same thing when recorded as an outgoing message for a cellular telephone.7 One natural explanation is that what (1) means doesn’t ﬁt this new situation.8 The semantic accounts we explore below are aimed at making good on this suggestion.
3 Ambiguity One early approach to the puzzle claims that indexicals are (not only contextsensitive, but) ambiguous — that they can refer to elements of either of two distinct contexts. Indeed, in raising the puzzle, Kaplan (1989) himself appears
to endorse this suggestion:9
If the message: “I am not here now” is recorded on a telephone answering device, it is to be assumed that the time referred to by ‘now’ is the time of playback rather than the time of recording.
Donnellan has suggested that if there were typically a signiﬁcant lag between our production of speech and its audition (for example, if sound traveled very slowly), our language might contain two forms of ‘now’: one for the time of production, another for the time of audition... (Kaplan, 1989, 491, note 12).
The idea that emerges (and is developed by Smith (1989)) is that ‘now’ is lexically ambiguous: it may refer either to the time of the context of 6 These considerations parallel a criticism by Devitt (1997) and Reimer (1998) against Kripke (1977), who appeals to Grice to defend the Russellian theory of deﬁnite descriptions. One might object that this line of response overpredicts semantic ambiguity — e.g., that it predicts that the interrogative form ‘can you pass the salt?’ has the regularly associated content do pass the salt as one of its literal semantic contents. For a reply to this objection and defense of this style of criticism, see Reimer (1998).
7 Thanks to Ivano Caponigro for this observation.
8 One piece of support for this explanation is that, to our ears at least, hearing (1) as the outgoing message for an unknown line counts as evidence that one has called a landline.
9 Nonetheless, Kaplan (1989) does not incorporate the idea into his ofﬁcial semantics for indexicality.
production/inscription (ci ) or to that of the context of audition/tokening (ct ).10 Perhaps, likewise, ‘here’ and ‘I’ refer ambiguously to either the place/agent of ci or the place/agent of ct. If so, then since there are three indexical expressions in (1) (ignoring tense), each ambiguous between two readings, there should be a total of 23 = 8 possible readings for (1).11 Presumably the view would be that, while these readings collapse onto the single, standard interpretation (which Kaplan’s semantics guarantees to be false relative to any proper context) when ci = ct, tokens of (1) nonetheless express truths when (because of the use of answering-machines, the slow travel of speech sounds Donnellan suggests, or whatever) ci = ct. How might listeners then disambiguate tokens of (1) to arrive at a unique content when ci = ct ? Perhaps pragmatic constraints — e.g., the semantically guaranteed falsity of the standard interpretation — guide listeners to a speciﬁc alternative disambiguation.
While schematic, the strategy outlined looks initially promising as an answer to the puzzle. Moreover, the proposal is a conservative extension of the classical machinery: it validates the ofﬁcial Kaplanian semantics in the special cases where ci = ct, and incorporates a simple, localized, yet powerful extension where ci = ct.
However, the proposal depends on a postulated lexical ambiguity that is not supported by the evidence.12 Speciﬁcally, application of a standard test involving the possibility of non-contradictory explicit cancellation under crossed readings tells against the presence of an ambiguity. To see this test in action, note that, because ‘bat’ is lexically ambiguous (between an animal and a piece of baseball equipment), there is a non-contradictory reading of 10 Smith denies that his position amounts to an ambiguity view, since it involves associating each indexical word-form not with multiple, distinct meanings, but with a single ‘metarule’ — a function from speakers’ intentions to ordinary characters (Smith, 1989, 168). We ﬁnd Smith’s characterization of his view as something other than an ambiguity theory unpersuasive.