Replies to Critics

From my perspective, replying to Santorio and Szabo was both enjoyable and challenging. There are relatively few points of clear disagreement to focus on. But regarding certain theoretical options, we're inclined to make different bets about which ones will prove most fruitful. So I'll emphasize this, if only to keep things interesting, and pass over agreement largely in silence. In the précis above, and in the book, I focused on a certain tradition in the field of semantics. Santorio correctly notes that in many neighborhoods, this tradition has expanded to include various “information-theoretic” approaches that I do not explicitly discuss. This is partly due to my limited education. But I've always found it hard to understand the operative notions of information. They sometimes seem reminiscent of Grice's (1957) notion of natural meaning, or Quine's (1960) notion of stimulus meaning, despite suggestions of both Shannon's (1948) technical notion and murkier though perhaps intuitive notions of epistemic possibility.1 In CM, I focused on Davidson, Lewis, and Montague because of their historical importance and their tendentious claims. I tried to make some contact with the later developments that Santorio mentions, in the context of comparing Kripke and Lewis on unicorns, by saying why I think it's a mistake to characterize meanings in terms of possibilities—metaphysical or epistemic—and why I think it's better to describe meanings in terms of instructions for how to access and assemble concepts. But I would welcome explicit comparison of the proposals in CM with alternatives of the sort Santorio has in mind. That would help me, and perhaps others. Though one needs to avoid smuggling in (e.g., via the claim that Slangs are for communication) any stipulation to the effect that sentence meanings are the intentional contents of corresponding assertions, characterized as sets of possibilities. Santorio says that the views he has in mind “appeal to notions of possibilities and information that are more fine-grained than standard intensions and extensions,” and that while they are “more complex than standard intensional semantics, they are generalizations of it, rather than alternatives.” Here, let me set aside my discomfort with talk of fine-grained information. The more important point is that abandoning the extensionalism that was originally interwoven with truth-theoretic conceptions of meaning does not obviously yield a “generalization,” as opposed to a class of alternatives according to which Slang expressions have meanings that are mentalistic in some way that is not made explicit. Practitioners can choose to say that their proposals are part of the “broad” tradition of truth-conditional semantics, which has often mixed descriptive claims about Slangs with proposed regimentations of ordinary talk. But why not just talk about “epistemic-possibility semantics,” let the talk of truth go, and then (especially if one grants that Slang expressions are polysemous) say how the views in question differ from the idea that expressions are used to access and assemble concepts? I prefer my mentalism straight up. But in any case, when discussing a “semantics for normative language” like the one offered by Gibbard (2003), it's important to be clear about whether one is discussing—whatever the author's intentions—some non-normative hypothesis about Slang expressions themselves, a partly normative claim about how such expressions ought to be used for certain communicative purposes, or something else.2 This chapter extends a line of argument, against truth-theoretic conceptions of meaning, that began with the discussion of ‘unicorn’ and other nouns we can use to express concepts that apply to nothing. In later discussions of words like ‘sky’ and ‘blue’—along with the conceptual equivocality of ‘book’, ‘fish’, and even ‘water’—I argue that there is nothing special about the meaning of ‘unicorn’. Words are not true of things, and meanings don't determine extensions. Initially tempting responses to Foster make Liar Sentences seem even more problematic, and vice versa. In chapter five, I argue that verbs like ‘chase’, as used in sentences that provide good evidence for event analyses, also turn out to be like ‘unicorn’ and ‘true’ in an important sense: you get into trouble if you assume that the word meanings determine extensions. No single argument in this family is decisive. And in a debate, I wouldn't begin by talking about chases; the point was that even here, things don't look good for the hypothesis that Slang expressions have extensions. But taken together, I find the arguments persuasive, in part because they highlight an important question: are there Slang expressions that provide confirming instances for truth-conditional theories of meanings, as opposed to ever more apparent counterexamples and opportunities for future research? Davidson (1967a) offered his bold conjecture more than fifty years ago, and in my view, investigation keeps making the conjecture seem less plausible. That said, it's certainly fair to ask how examples like (14-19) bolster the more familiar reasons for skepticism about truth-theoretic semantics. There are, I grant, some bad arguments in the vicinity. Santorio and Szabo both note that in my view, event variables are not plausibly construed as ranging over actual events that can be truth makers for sentences like (14-19). But they seem to think that I am motivated, at least in part, by metaphysical scruples that conflict with positing fine-grained events. To the best of my knowledge, I have no such scruples. My concern is the second one Santorio mentions: the events required by event analyses will need to be individuated about as finely as the relevant Slang phrases, and this threatens the attractive accounts of adverbial implications; cp. Davidson's (1969) response to Kim (1966).3 Theorists can distinguish Al's chase of Theo from Theo's chase of Al—on the grounds that Al is the Agent of the first chase but not the second, while Theo is the Agent of the second chase but not the first—even though Theo did a lot of running in Al's chase of him, Al did a lot of running in Theo's chase of him, the running that Theo did in Al's chase of him seems to be the very running that Theo did in his chase of Al, and likewise for all the running Al did. That is, we can use ‘chase’ to talk about things individuated in terms of thematic roles that respect an exhaustion condition that Slangs impose: in Al's chase of Theo, Al is the only Agent. Semanticists can (and in my view, should) use terminology like ‘Agent’ in this Slang-focused way. But if one uses ‘chase’ and ‘Agent’ in these technical ways, one is not entitled to assume that ‘chase’ (so used) is true of the actual chases that occur when one animal chases another. One can make this point about technical vocabulary vivid with examples of chipmunks chasing each other.4 Theorists can reply that chases are not quite what we thought they were. But this reply requires defense. Moreover, Davidsonian accounts of why (15) seems to imply (14)—and likewise for endlessly many similar pairs of sentences—were supposed to be part of the argument that ‘chase’ is true of chases. But if a theorist says that chases are not quite what we thought they were, and that Al's chase of Theo differed from Theo's chase of Al, why does the theorist still get to assume that Al's chase of Theo was the same chase as Al's gleeful chase of Theo? (Or that the occurrence of the latter ensures the occurrence of the former?) One wants to see an independently plausible conception of events that preserves truth-theoretic glosses of event analyses, not merely a conception tailored to a tendentious account of meaning. The metaphysics required by the truth-theoretic gloss seems to be at odds with the explanatory work being done by ‘∃’ and ‘&’. Again, I wouldn't lead with this somewhat complex argument. But I think it helps reveal a broad of pattern of failure, endemic to the idea that linguistic meaning can be characterized in terms of truth. Meanings, which reflect grammar and the concepts that humans use to think about things, are not beholden to what there is (or ways things could be). Truth, which is independent of any parochial ways of thinking or talking about things, is not beholden to grammar or the concepts we assemble by using Slang expressions.5 If there were independent reasons for distinguishing Al's chase from Theo's, I would withdraw this objection to truth-theoretic conceptions of meaning, but then re-evaluate my view that (15) implies (14) in much the way that ‘A brown dog barked’ implies ‘A dog barked.’ Though just as I would be suspicious of a semantic theory that required subtle individuation of brown dogs, leaving room for some that are not identical with the co-located dogs, I am suspicious of semantic theories that require subtle individuation of chases. And I don't think one needs special metaphysical scruples to wonder how hypotheses about the meanings of (14-19) could motivate the requisite ancillary assumptions about chases. Drawing conclusions about chases from premises about ‘chase’ seems, on its face, like a dubious enterprise. This will seem unpersuasive if one assumes that meaning has more to do with ontology and/or disquotation than understanding. But I think this assumption is independently implausible. And to repeat, the dialectic is not that truth-theoretic conceptions of meaning look good apart from some technical issues illustrated with (14-19). It's that such conceptions don't look good even on their home turf. Szabo offers a relevant modal argument regarding the example of Simon playing a song on his tuba. For what it's worth, I cannot find any “clear modal difference” between the event of Simon playing the song and the event of Simon playing his tuba. But I've often been told that my modal-difference-finder is defective. So let me grant the possibility of a tuba-playing-by-Simon that is identical in terms of motor commands and acoustic wave forms produced, while being a case of spontaneous improvisation as opposed to song-playing. I still doubt that any of Simon's actual performances could have been such an event. And in any case, we should distinguish three questions: could there have been a truth-maker, of the sort Szabo imagines, for some thought of the form ∃e[Agent(e, Simon) & Past(e) & PlayHisTuba(e)]; could any of Simon's actions have differed in the way Szabo imagines; and if the answer to either modal question is affirmative, does that warrant positing distinct satisfiers of ‘play the tuba’ and ‘play the song’ in the situation where Simon does what he does? I can imagine revising my tentative answers—yes, no, and no. But I'm loathe to revise my answers based on modal intuitions that I don't share; see below. And I think that reflection on the full range of relevant cases makes it less plausible that Davidsonians can have their truth conditional semantics while eating their event analysis cake.6 Santorio notes that if names are instances of type , semanticists need a composition rule like Function Application (or old school Fregean Saturation). He thinks the relevant literature favors this familiar conception of names. I don't. Readers can make their own assessments of what the totality of evidence, regarding Slang syntax and semantics, suggests. Here, let me just say why I don't think that Santorio's observations about anaphora—or more generally, initial appearances that English names are grammatically atomic denoters—outweighs the case, reviewed in CM, for treating lexical proper nouns as semantically monadic expressions that can combine with determiners (overt or covert) to form phrases that are also semantically monadic but perhaps used to construct concepts that apply to at most one thing. I don't deny that children have atomic concepts of type that can be lexicalized with proper nouns like ‘Napoleon’. On the contrary, I suspect that such concepts are often used to introduce monadic concepts like identical(_, Napoleon) that apply to at most one person, perhaps Napoleon Bonaparte; cp. Quine (1951). I also suspect that proper nouns, like other words, often become polysemous or conceptually equivocal. For example, ‘Napoleon’ might also get used to access a concept that applies only to some other Napoleon—not Bonaparte—or a more permissive metalinguistic concept like properly-called-with(_, NapoleonPN) whose singular constituent is a concept of the proper noun itself; cp. Katz (1994). I don't think that all uses of a proper noun correspond to a common conceptual content, or that simple uses as in ‘I saw Napoleon’ have contents that must be specified without appeal to concepts of type . Conceptual reduction is no part of my project or the kind of predicativism that I embrace. Initially, this tells against the hypothesis that (22) includes a modifiable clause in which the intransitive verb ‘break’ combines with ‘a vase.’7 But given Baker's (1988) discussion, it is quite plausible that the grammatical process of incorporation blocks this logically possible reading of (22); see Pietroski (2003) for detailed discussion. Again, a corresponding “blocking hypothesis” may not be needed to account for the data that Santorio sites. But I don't think such hypotheses acquire special burdens; similarly, I don't reject type analyses of names simply because they need to be supplemented with ancillary hypotheses to accommodate examples like (21). We're looking for the best overall theory, given the evidence. No proposal gets to be the default option. Unlike Santorio, Szabo seems happy to abandon the idea of characterizing Slang meanings in terms of Fregean types and Function Application, but less happy about some aspects of my internalism. I'll continue to follow the usual practice of suppressing remarks of appreciation—apart from repeating here that I'm grateful for how Santorio and Szabo have captured the spirit of CM and conveyed its main ideas in interestingly different ways. Szabo's opening paragraphs set up and motivate my internalism in a useful way that also highlights his worry that I go too far in “opting for the Chomsky-inspired view that natural languages generate pronounceable meanings.” He suggests another option that comports with his modal intuitions about expressions. So let me say why I still want to say that Slangs generate the meanings (semantic instructions) they connect with pronunciations (phonological instructions). If it matters, we can represent this pronunciation in IPA: wi wɑʧt hɜr dʌk nɪr ə ˈmʌdi bæŋk. More importantly, given any resolution of the lexical homophonies, ‘near a muddy bank’ can modify ‘duck’ or ‘watched her duck.’ So there are at least eight ways of generating a string of words with the pronunciation of (23), each corresponding to a way of understanding it as (the pronunciation of) a generable English expression. And note that on each construal, deleting ‘muddy’ or ‘near a muddy bank’ seems intuitively compelling. So it seems that the pronunciation of (24) is shared by two expressions, each with its own meaning, but the pronunciations of (25) and (26) correspond to one expression/meaning each. So the second option supports the reading indicated with (27b) and not the unattested reading indicated with (27c). Such examples, and there are many, suggest that it's one meaning per expression (or “derivational history”); see Chomsky (1964) and the discussion in chapter one. Sidepoint: such facts let us gesture towards the meanings that Slang expressions connect with pronunciations, without relying on tendentious assumptions about what meanings are. (Compare spotting a few quasars and wondering whether they are stars, nebulae, or galaxies.) Lexical homophones are distinct expressions with distinct meanings; (24) has two meanings, while (25) and (26) have one each; (27) has two but not three; etc. I think these meanings are instructions for how to access and assemble concepts of a special sort. Others can try to argue that English somehow connects the pronunciation of (27) with two but not three truth conditions. I agree with Szabo that we could invent a language that connects the pronunciation of (27) with any interpretation(s) we like.9 In particular, we could invent a language that connects the pronunciation of (27) with structures like (27α) and (27β), stipulating that the second way of generating the relevant word-string goes with the kind of interpretation indicated with (27c). Let ‘saw a spy’ and ‘with a telescope’ correspond to predicates of individuals, and let the adjunction signify predicate conjunction as in ‘∃y[Saw(x, y) & Spy(y)] & ∃y[Had(x, y) Telescope(y)]’ or some analog with an event variable that cannot be modified by ‘with a telescope.’ We could also let the adjunction signify disjunction or conditionalization. But I see no reason for thinking that the actual Slang expressions, with the pronunciations of (27a) and (27b), could have had such meanings. I agree that speakers of English could have expressed the meaning of ‘dog’ with the pronunciation of ‘cat.’ But in my view, it's not similarly “conventional” that adjunction is conjunctive; see the discussion of Lewis in chapter one. I suspect that the nature of the human language faculty (a.k.a. Universal Grammar) determines the mapping from modes of “syntactic” combination to the modes of semantic combination, whatever they turn out be. We can imagine worlds in which certain hominid-counterparts acquire a mental faculty that lets them hear the pronunciation of, ‘This is a cow that we saw yesterday’ as meaning that the demonstrated thing is either a cow or a thing we saw yesterday. But it doesn't follow that our language faculty could have combined a noun with a relative clause to form a phrase with a disjunctive meaning. We can also imagine that the hominid-counterparts live in a world with Putnamian XYZ, instead of H2O. But such flights of imagination tell us little about the nature of water or Slang meanings. I tend to distrust the modal “intuitions” that philosophers report regarding Slangs and their expressions. It's too easy to assume, for no good reason, that Slang expressions are to be individuated like the well-formed formulae of familiar invented languages. I have more (albeit shaky) faith in Kripkean intuitions that suggest the importance of ancestry and natural history. So when thinking about Slangs as languages of a special kind, acquirable by certain primates who enjoy a distinctive mental faculty, I am suspicious of the idea that species-universal features of Slangs are metaphysically contingent. I certainly don't think it's a metaphysically contingent fact that combining English (or Japanese) words doesn't signify Function Application. In my view, the meanings of Slang expressions are not composable in that way. Of course, it's epistemically possible—indeed, likely—that I'm wrong about the actual features of Slang expressions. But pace Szabo, I don't see any good modal argument against the following simple idea: Slang expression meanings correspond one-to-one with ways of generating Slang expressions because Slangs are biologically instantiatable procedures that generate the (biologically instantiatable) meanings/instructions in question and link them with pronunciations in constrained ways; where pronunciations can be viewed as Slang-generable instructions that interface with human “articulatory-perceptual” systems, as suggested by Chomsky and Halle (1968) and Halle (1990). I find it hard to see how else Slang expressions could have the meanings they do, in the sense of being understood in the ways they are understood, as opposed to other logically coherent but never attested ways.10 This doesn't make it absurd to ask why the pronunciation π of some actual expression is paired with the meaning(s) it is paired with, as opposed to other coherent interpretations, or why π is the pronunciation of certain actual expressions as opposed to other logically possible expressions. (I assume that explanation plays out against a background of epistemic possibilities, many of which then get regarded as mere epistemic possibilities.) It's relevant here that on the view urged in CM, speakers can use polysemous words—including ‘language’, ‘meaning’, and ‘water’—to express kind-concepts; see the discussion of Putnam (1975). And if linguistic meanings are instructions of a certain kind, generated by languages of a certain kind, then modal intuitions should be diagnosed accordingly. So not unreasonably, Szabo worries that appealing to plural concepts of (restricted) ordered pairs is needlessly artificial or “farfetched.”12 If this is the best way to treat quantificational determiners as instances of my dyadic type , then—especially given that I don't characterize meanings in terms of Fregean types and lambda abstraction—why not drop the demand that quantificational determiners be instances of some semantic type, and just appeal to a syncategorematic rule for each (lexical) quantificational determiner? One reason is rhetorical. Many semanticists abhor syncategoremata, even if they grudgingly grant the need for a few. So I didn't want to make it seem that my account of composition requires giving up the idea that quantificational determiners—the holy of holies for some intellectual descendants of Montague—have meanings that compose with others. Moreover, Pietroski (2005) had grudgingly appealed to pairings of entities with truth values; and I wanted to show that the work done by invoking truth values could be done without them. But I have no brief against syncategoremata per se, and I do worry about using formal tricks, even if my proposal is less artificial (overall) than familiar alternatives. On the other hand, at any given time, we may be stuck with theories that use some formal tricks until we can do better. And I worry that we won't be pressed to do better if we let ourselves simply encode the meanings of quantificational determiners with syncategorematic rules. As suggested above, the real point of encoding the meanings in terms of ordered pairs was to encode them in terms of restricted ordered pairs, thereby describing—and perhaps starting to explain—why determiners like ‘every’ and ‘most’ get understood as restrictable quantifiers, as opposed to other “generalized” quantifiers.13 We can say that an instance of ‘every F is G’ is an instruction for how to build a thought according to which everything covered by the ‘F’-concept is covered by the relevant ‘G’-concept. But then one wants some explanation for the absence of determiners governed by parallel rules like the following: ‘yreve F is G’ … everything covered by the ‘G’-concept is covered by the ‘F’-concept; ‘ident F is G’ … the things covered by the ‘F’-concept are the things covered by the ‘G’-concept; ‘equi F is G’ … the things covered by the ‘F’-concept correspond one-to-one with the things covered by the ‘G’-concept; etc. (One also wonders about the space of polyadic generalized quantifiers.) I don't deny that theorists can try to provide a suitably restricted but still descriptively adequate framework for stating syncategorematic rules. On the contrary, I would welcome such a framework; cp. Chomsky (1957, 1959). Perhaps this perspective can also be extended to verbs, nouns, and their modifiers in a way that avoids the need for any semantic typology. That might provide a “humanly limited” version of Davidson's (1967a) attempt to eschew Fregean apparatus in favor of Tarski's (1944) notion of satisfaction; cp. Davidson (1990). In CM, I discuss the possibility of supplementing my very spare inventory of types, along with my suspicion that overgeneration is the more pressing concern. Invoking even less typology would be attractive in some respects. But part of the project was to explore the idea that a posited semantic typology should earn its keep by limiting the range of admissible interpretations for lexical items, as opposed to allowing for abstractions of the sort that yield Fregean types like <, <, t>> and endlessly many still higher types; see also Pietroski (2017, 2020). The typology of CM was the sparest one I could think of that had a prayer of being descriptively adequate—so long as we drop the idea that meanings determine extensions—while also imposing substantive but empirically plausible constraints on lexical meanings. The fundamental problem of communication is that of reproducing at one point either exactly or approximately a message selected at another point. Frequently the messages have meaning; that is they refer to or are correlated according to some system with certain physical or conceptual entities. These semantic aspects of communication are irrelevant to the engineering problem. The significant aspect is that the actual message is one selected from a set of possible messages. The system must be designed to operate for each possible selection, not just the one which will actually be chosen since this is unknown at the time of design. I don't deny that there are chases, and that in this sense, chases exist even if skies don't. But the existence of chases doesn't show that ‘chase’ is true of them. I also agree that there is a sense in which there are blue skies, but no blue unicorns. But it doesn't follow that ‘sky’ is true of some things, at least not in the sense of ‘true’ that matters for a theory of truth (p.68).