Forthcoming O’Rourke and Washington (ed.) Situating Semantics: Essays on the Philosophy of John Perry. MIT Press
Misplaced Modification and the Illusion of Opacity
Kenneth A. Taylor
In a number of ground-breaking publications, John Perry has stressed the importance of avoiding a class of fallacies that he and Barwise (1983) formerly called Fallacies of Misplaced Information and that Perry (2001) now prefers to call subject matter fallacies. One commits a subject matter fallacy when one supposes that:
…the content of a statement or belief is wholly constituted by the conditions its truth puts on the subject matter of the statement or belief; that is, the conditions it puts on the objects the words designate or the ideas are of. (Perry 2001)
Avoiding the bewitching influence of subject matter fallacies, Perry has argued, is one key to seeing that two alleged failures of referentialism -- viz., its apparent inability to solve both what Perry calls the co-reference problem and what he calls the no-reference problem -- are really only apparent. The co-reference problem is the problem of explaining how possibly, consistent with referentialism, co-referring expressions may differ in cognitive significance. The no reference problem is the problem of explaining how possibly, consistent with referentialism, names entirely lacking a referent may be cognitively significant at all.
Naïve forms of referentialism maintain that all there is to the meaning, semantic content, or cognitive significance of a name is its property of standing for a certain object. Referentialism of this sort has a prima facie problem on each of these fronts. The naïve referentialist would seem to be committed to saying that sentences differing only by co-referring names must have the same cognitive significance and that sentence containing names with no reference must be entirely devoid of cognitive significance. By distinguishing what he calls referential content from what he calls reflexive content, Perry insists that the referentialist can offer satisfying solutions to both the co-reference and no reference problems. The reflexive content of an utterance is, roughly, a proposition about that very utterance and the conditions under which it is true. The referential content of an utterance, on the other hand, is typically a proposition about some object -- the object that is the subject matter of the relevant utterance. Armed with this distinction, the referentialist can claim that even if there is no difference in referential or subject matter content between two sentences that differ only by the presence of co-referring names, there can still be differences in the reflexive contents of such utterances. And the crucial further claim is that although what is said by an utterance is a matter of the referential or subject matter content of the utterance, the cognitive significance of an utterance is a matter of its reflexive content.
Perry makes a compelling enough case for the claim that the cognitive significance of an utterance should be explained by appeal to its reflexive content rather than by appeal to referential or subject matter content.1 In fact, the case he makes is so compelling that one might have expected him to say something similar about apparent failures of substitutivity within propositional attitude contexts. After all, once one conjoins the thesis of referentialism with the notion of a subject matter fallacy, the pieces for such an approach would seem to be in place already. So, for example, one might begin by conceding that an utterance of:
(1) Jones believes that Hesperus rises in the evening
does not convey the same information, at least not in toto, as an utterance of:
(2) Jones believes that Phosphorus rises in the evening.
However, it is open to the referentialist to maintain that the difference in conveyed information has nothing to do with any difference in subject matter or referential content between the two reports. Instead, the referentialist can say that (1) and (2) attribute to Jones exactly the same relation to exactly the same singular proposition. Consequently, (1) and (2), as uttered at a given moment, should either both be true or both be false. To be sure, denying that (1) and (2) may differ in truth value involves a bit of bullet biting. But the strategy Perry applies to the co-reference problem in general is available here as well. He need only insist that despite their shared referential content, (1) and (2) need not convey the same information in toto. Some of the information conveyed by an utterance of (1) as opposed to (2) might be thought to be a matter of the reflexive content of (1) as opposed to (2). And because of the potential differences in reflexive contents of (1) and (2), there can be conversational contexts in which it is appropriate to utter (1), but inappropriate to utter (2), even though (1) is true when and only when (2) is true. The take home message of such an approach would be to deny that our evident reluctance to utter reports like (1) and (2) interchangeably has anything to do with Fregean reference shifts, Quinean opacity, or any other such subject matter level semantic peculiarities of embedded clauses. Indeed, one might have expected Perry to say that the mistake of many previously extant approaches to attitude statements is to assume that embedding somehow effects, for good or for ill, the subject matter of the relevant sentence. And one might have expected an avowed referentialist, like Perry, to deny this assumption. To think otherwise, one might have expected Perry to say, is to commit a subject matter fallacy.
In point of fact, Perry once did endorse such views. For example, in Barwise and Perry (1983), we find the following:
But, a Lockean or Fregean might ask, “Does an innocent theory have any right to even notice what name is used in an attitude report? It is not the name but the individual referred to which gets into the interpretation of the report, so how can the name be in any way relevant, even to the appropriateness of the report? (emphasis added)
This objection contains an instance of the fallacy of misplaced information. [emphasis added] The change from TULLY to CICERO makes an enormous difference to the information made available by the report, and an innocent theory need not overlook this if it is combined with a relation theory of meaning. Part of the information you can get is the information that someone is called ‘Cicero’, and of course you do not get this information if ‘Tully’ is used instead. This is so even though the interpretation of the report stays the same.
For reasons he has never spelled out in complete detail, however, Perry rather quickly abandoned this approach to attitude statements and became convinced that reports like (1) and (2) can and do differ not just in total information conveyed but in subject matter.
Even in the face of Perry’s altered convictions about what attitude reports state, he remained steadfast in both his commitment to referentialism and his commitment to Davidsonian semantic innocence. In collaboration with Mark Crimmins, he sought to reconcile the combination of innocence and referentialism with the essentially Fregean intuition about the attitude statements that he had earlier eschewed -- viz., that propositional attitude statements that differ only by co-referring proper names may, nonetheless, differ in truth value. The key to such a reconciliation is the supposed insight that there is more to an attitude statement than meets either the eye or the ear. The propositions expressed by an attitude report typically contains one or more unarticulated constituents, as Perry calls them, where a constituent of a proposition is unarticulated if it does not correspond to any constituent of the sentence (or utterance thereof ) that expresses that proposition. The central thought is that although a device of “direct reference” may retain, when embedded, the semantic role of standing for an object -- thus preserving innocence -- the embedding of such an expression somehow triggers the introduction into the expressed proposition of an unarticulated constituent. Moreover, at least in some contexts, distinct, but co-referring names may trigger the introduction of distinct unarticulated constituents. That is why, despite referentialism and semantic innocence, substitution of co-referring names is not guaranteed to preserve truth value.
In this essay, I take issue with the Perry-Crimmins approach to propositional attitude reports and urge upon Perry a return to the wisdom of his earlier days.2 In particular, I shall argue that the proposition strictly, literally expressed by a propositional attitude statement contains no unarticulated constituents. I doubt, in fact, that any proposition literally expressed by any sentence or utterance contains unarticulated constituents. However, although I will dwell at some length on some reasons to be skeptical about unarticulated constituents in the general case, my main aim in this essay is to show that even the most intuitively compelling motivations for positing unarticulated constituents in the general case have no force when it comes to propositional attitude statements. Propositional attitude statements simply lack the sort of felt semantic incompleteness to which Perry generally appeals in attempting to motivate the positing of unarticulated constituents. Though I reject the mechanism of unarticulated constituents, I shall argue that there is still a way to put notions and ideas at semantic issue in an attitude report. Moreover, just as Perry desires, that way of doing so is consistent with the conjunction of referentialism and semantic innocence. One merely has to deploy what I have elsewhere called a fulsomely de re ascription. (Taylor 2002, 2003). And I will show that the device of fulsomely de re ascriptions actually provides a less problematic way to capture many of the quite correct insights about beliefs and their ascriptions that lie at the core of Perry’s approach. In the end, then, my arguments may constitute less of a refutation of Perry’s views than a friendly, if rather far-reaching, amendment to those views.
II. What is an Unarticulated Constituent?
A constituent of a proposition is unarticulated when it is not the semantic value of any syntactic constituent of the sentence or utterance that expresses the relevant proposition. It is important to distinguish Perry’s claim from the claim that unarticulated constituents are semantic values of “hidden” or “suppressed” constituents of some sort or other. On such approaches, although some propositional constituents would not be associated with any constituent at the level of surface syntax, they would be associated with syntactic constituents that show up in logical form or in what I have elsewhere called the subsyntactic basement of the lexicon. (Taylor 2003) Such approaches amount to the view that there really are no unarticulated constituents, that a principle of full articulation holds at some level or other. Perry quite explicitly rejects any such view. As he puts it:
…we do not need to first find an expression, hidden in the “deep structure” or somewhere else and then do the semantics of the statement augmented by the hidden expressions. Things are intelligible just as they appear on the surface, and the explanation we might ordinarily give in non-philosophical moments, that we simply understand what the statement is about, is essentially correct. (Perry 1986, pp. 176.)
It is clear what unarticulated constituents are not, on Perry’s view. It is a little less clear exactly what they are and why he supposes that there are any of them. On this score, it may help to compare and contrast Perry’s views with other views in the neighborhood. On the one hand, there is impliciture approach defended by Kent Bach (1994, 2001a, 2001b). Bach endorses what he has called the Syntactic Correlation Constraint on what is said by (an utterance of) a sentence. According to that constraint, what is strictly, literally said (by an utterance) must correspond to “the elements of [the sentence] their order, and their syntactic character.” At first glance, that constraint may seem quite alien to the spirit of Perry’s approach, since Bach’s constraint fairly directly entails that there can be no unarticulated constituents in what is strictly literally expressed by (the utterance of) a sentence. However, the disagreement between Bach and Perry may be less deep than it first appears. Bach’s approach allows that certain sorts of “enriched” propositions -- that is, propositions containing constituents not correlated with any either explicit or suppressed syntactic constituent of the relevant sentence -- may be “pragmatically imparted,” but not strictly expressed by (the uttering of) a sentence. Such propositions are one sort of “impliciture,” as Bach calls them. Moreover, since Bach allows that a syntactically complete sentence may, nonetheless, be semantically incomplete, the “first” or even only fully truth-evaluable thing pragmatically imparted by an utterance of such a sentence may be an impliciture. To be sure, Bach insists that an utterance of a syntactically complete, but semantically incomplete sentence will not strictly literally express the relevant impliciture. Implicitures, he maintains, are extra-semantic pragmatic conveyances that go beyond what is strictly literally said in the uttering of a sentence.
The common ground between Bach and Perry is substantial. Both agree, for example, that a syntactically complete sentence, even one with no hidden or suppressed constituents, may be semantically incomplete in the sense that no complete proposition is determined merely by the interaction of its meaning, syntax, and the contextual provision of values for its explicit indexicals and the like. Moreover, both agree that a speaker, in uttering a semantically incomplete sentence, may nonetheless pragmatically impart a complete proposition. The one issue over which Bach and Perry appear to part company concerns what counts as what is strictly literally said by an utterance of a sentence. On Bach’s view, the pragmatically imparted implicitures generated by utterances of semantically incomplete sentences are “extra-semantic” and, as such, form no part of what is strictly literally said by a speaker in making the relevant utterance. Perry, on the other hand, holds that the proposition pragmatically imparted by the utterance of a syntactically complete, but semantically incomplete sentence may, nonetheless, be what the speaker strictly literally says in making the relevant utterance. On Perry’s view, but not on Bach’s, pragmatics can play a role in constituting what is strictly literally said by an utterance, even when the relevant sentence is indexical-free. Bach, on the other hand, seems to want to deny any such role to context and pragmatics in constituting what is said by a speaker in uttering such a sentence.
Since Bach and Perry both agree that it will often be the case that nothing propositionally complete is yielded by the interactions of syntax, lexical meaning, semantic composition rules, and the contextual provision of values for explicit indexicals and demonstratives, it is not unreasonable to think that the real dispute between them, to the extent that there is one, boils down to what may amount to a merely verbal dispute over the proper ownership of the phrase ‘what is said.’ In particular, the question arises whether what is said by an utterance has, in all cases, to be something fully propositional. Bach evidently thinks that what is said need not be fully propositional, while Perry seems think that what is said must always be something propositional. Because Bach allows that what is said by an utterance may fall short of a proposition, he has no particular motive to identify the “first” fully propositional content generated in a certain communicative circumstance with what is strictly literally said by the speaker in making the relevant utterance. By contrast, Perry seems at least tacitly to hold that a speaker hasn’t yet said anything fully determinate until a determinate propositional content has been generated. Moreover, he seems willing to count whatever goes into generating an, as it were, “initial” proposition a determinant of what is said.
Now I doubt that there is a single intuitive notion of “what is said” against which to measure competing claims of this sort.3 In particular, any merely intuitive notion of what is said is likely to be tied up with notions of samesaying and with the direct and indirect ascriptions of content to utterances in delicate ways difficult to untangle. For example, in some contexts, relative to certain communicative purposes, speakers would seem to count as saying the same thing again just in case their utterances express the same proposition. In others contexts, and relative to other communicative purposes, speakers would seem to count as samesayers just in case what one says about object a the other says about object b. So, for example, if I say of myself that I am hungry and you say of yourself that you are hungry, there would seem to be a sense in which we say the same thing and a sense in which we say different things. This suggests that on one way of assessing “what is said” by a speaker in making an utterance it may be only complete propositions that count – just as Perry apparently believes -- while on other ways of assessing what is said something less than a complete proposition may do – just as Bach apparently believes. But if that is right, then there may be no settling the apparent dispute between Perry and Bach, once and for all, independently of particular discourse contexts. Indeed, it would not be surprising if a fuller, more systematic exploration of the various ways in which we determine either what counts as what is said by a speaker or when two speakers count as having said the same thing again revealed that our practices fail to add up to anything fully precise and determinate. There may, in fact, be no isolable thing in our ascriptive practices against which we can directly test competing claims of the sort defended by Bach and Perry. Choosing between them may be a matter of who has the best, most comprehensive, most explanatory overall theory of meaning and communication. On that score, the jury is still out.4
Sill, such disagreements as there are between Bach and Perry over what goes into what is said suggests that it may be more natural to assimilate Perry’s views to those of radical contextualists like Recanati (2001, 2003a, forthcoming), Sperber and Wilson (1995), Carston (2002), and others. Radical contextualists tend to hold that the gap between semantically incomplete sentence meaning and pragmatically determined utterance content is bridged not by the provision of contextually determined values of either explicit or hidden parameters, but by so-called “primary” pragmatic processes such as free enrichment. Primary pragmatic are supposed to play a role in the very constitution of “what is said.” Such processes are supposed to operate antecedently to the determination of a complete propositional content. As such, they are supposed to stand in sharp contrast with so-called secondary pragmatic processes. Secondary pragmatic processes are supposed to operate on an already constituted propositional content to yield something further -- a conversational implicature, an indirect speech act -- as output. One’s first thought may be that primary pragmatic processes function mainly to assign contextually determined values for relatively tractable elements of a sentence such as tense, aspect, indexicals, demonstratives, and quantifiers in some relatively systematic and semantically constrained fashion. But the radical contextualist holds that there are primary pragmatic processes that enrich contents to include propositional constituents not tethered to any particular syntactic constituent of the relevant sentence.
The free enrichment of the radical contextualists bears a striking affinity to what Perry (2001) calls “content-supplemantal” uses of context. We use context in a content-supplemental way, according to Perry, when context provides propositional constituents “after we have all the words and their meanings identified.” That is, context is used in a content-supplemental way when word meaning, sentence meaning and the contextually determined values of explicit indexicals and demonstratives still do not determine a “complete” content, that is, a content for which the question of truth and falsity meaningfully arises. Consider, for example, the supposed unarticulated location constituent introduced in context by one who utters:
(3) It is raining.
Perry says that:
There is a debate about whether in such cases the ‘logical form’ of a sentence [like (3)] contains an argument place for the place, or does not. Francois Recanati and Jason Stanley have commandeered the term ‘unarticulated constituent’ for the purposes of this debate. (Stanley 2000) Stanley claims that there is such an argument place in logical form, hence that the constituent is articulated; Recanati claims that there is no such place, hence it is not. I think it is a bad idea to use the term ‘unarticulated constituent’ for two different questions, and of course I like my use of it better.
On the issue in question, I am inclined to side with Recanati, I think. I conceive of things in the following way. Relations are ways of classifying variations and unknowns across phenomena, against a background of factors that are taken as unchanging or otherwise given. The words for relations will be lexicalized in a way that reflects what is taken as varying and unknown, at least in a typical case, or was at the time the words acquired their grammatical properties. So ‘be simultaneous’ has two argument places for events, and none for inertial frames. ‘Rain’ has tense, but no argument place for places. ‘Be successful’ has tense and an argument place for succeeder, but none for standards of success. We use adverbial and prepositional phrases of various sorts to get at additional relevant factors when we need to. In cases where this happens a lot, it will be easy. There are lots of ways to say where it is raining. In cases in which scientific or philosophical discoveries or insights lead to appreciation of unlexicalized factors, we appeal to phrases like ‘relative to’. So events are simultaneous relative to inertial frames; the 49ers were unsuccessful last year relative to the common standards of success for athletic teams. (Perry 2001)
This passage makes it pretty clear that Perry doesn’t endorse anything like what I earlier called a principle of full articulation. But it also raises, I think, at least as many questions as it answers. To consider just one example, notice that Perry seems to grant that we can often, perhaps always, add explicit, syntactically optional modifiers or adjuncts to “complete” what would otherwise be incomplete. This means that if the speaker wants to explicitly say where it’s raining, she can do so in a quite straight-forward way. She simply adjoins to the verb a modifier that denotes a place. Given that a speaker may optionally adjoin a modifier and thereby explicitly specify where it is raining, it may seem fair to wonder whether it might not be the case that whenever a speaker utters (3) in some context Cwithout adding an explicit place modifier, she conversationally implicates, by the very failure to add a modifier where one is called for, some proposition to the effect that it is raining at m, where m is a place denoted by some conversationally relevant modifier or other. The worry is that if we’ve already got the mechanism of conversational implicature available to explain how someone who utters (3) without a place modifier can, nonetheless, communicate a proposition to the effect that it is raining at a particular place and time, then its not immediately obvious what talk of unarticulated constituents is supposed to add to this essentially Gricean story.
At times, Perry seems to have in mind a view like Recanati’s. Recanati holds that conversational implicatures are generated by so-called secondary, rather than by primary pragmatic processes. Clear evidence that Perry has something like this distinction in mind can be found in his rejection of Nathan Salmon’s view that apparent failures of substitutivity in propositional attitude contexts are pragmatically generated illusions. By contrast, Perry and Crimmin (1989) claim to assign to pragmatics what they call “a more honorable role.” Pragmatic features do not, they claim, “create an illusion, but help to identify the reality the report is about.” And they go on to suggest that there are at least two different jobs that the pragmatic factors can perform: the job of determining the truth conditional content of an utterance of a semantically incomplete sentence and the job of generating a conversational implicature.5 As they put it:
Last, the move to unarticulated constituents emphasizes the importance of pragmatic facts about language to the study of what seem like purely semantic issues. In order to express claims, we exploit the tremendous variety of facts, conventions, and circumstances, of which the meanings and referents of our terms form just a part. So it is a mistake to relegate pragmatics to matters of felicity and implicature. In the case of belief reports, it is central to the understanding of truth and content. (Crimmins and Perry pp. 232)
But it is fair to wonder whether Perry has some principled basis for deciding just when pragmatics is performing one rather than the other of these two distinct jobs. About all that Perry ever says by way of defending the view that unarticulated constituents are not pragmatic externalities, as I call them, but ingredients of literal truth-conditional content is that absent those very ingredients the relevant utterance would not say anything strictly truth-evaluable at all. In the case of (3), for example, Perry typically says things like the following:
In this case, I say that the place is an unarticulated constituent of the proposition expressed by the utterance. It is a constituent because, since rain occurs at a time in place, there is no truth evaluable proposition unless a place is supplied. (emphasis added) It is unarticulated, because there is no morpheme that designates that place. (Perry 2001)
But Perry’s reasoning here is not altogether compelling. From the fact that rain occurs at a time in a place, it simply doesn’t follow that there is no truth evaluable proposition expressed by an utterance of ‘it is raining’ unless a place is supplied. To see why consider other indispensable properties of rainings. Whenever it rains, it rains a certain amount and for a certain duration. But we can express a fully determinate proposition by an utterance of ‘it rained’ without having to specify how much rain fell or over what span of time the rain fell. It is unclear what explanation there is supposed to be, on Perry’s view, of the fact that we must specify where it is raining in the case of the present tense ‘it is raining’ if we are to express a complete proposition, but we need not specify how much it rained or for how long it rained in the case of the past tense ‘rained’ in order to express a complete proposition.
One might be tempted to appeal here to Perry’s distinction between argument roles of relations and argument places of predicates. He puts the distinction this way:
On the way I like to look at things, relations have argument roles or parameters. These are to be distinguished from the argument places or variables that predicates that express the relations may have. My picture of unarticulated constituents is that there are argument roles that are not represented by explicit argument places. We fill the argument role which is filled from context. (Perry 2001)
But our current worry is that not every argument role demands contextual filling in, on pain of incompleteness. Perry might – but this is really just a guess -- try to make something like the adjunct/argument distinction, not at the level of predicates but at the level of relations themselves. Armed with that distinction, he could perhaps argue that it’s the arguments and not the mere adjuncts of a relation that must be supplied if the relation is to obtain at all.
This move seems unpromising. For any relation in n arguments with a claim to be the unmodified raining relation, there will be other relations in m arguments, for m distinct from n, that appear to have no lesser claim to being the or at least a raining relation. Consider the following two examples. There is a relation that holds between a place, a time, and a velocity just in case it is raining at the time, at that place, with that velocity. There is another, less “articulated” relation that holds between a time and a place just in case it is raining at that time at that place. Does one or the other of these relations have more of a claim to being the raining relation? Is the former relation merely a modification of the latter? If these questions are supposed to be purely metaphysical questions about relations rather semantic questions about the lexical meaning of verbs, then I confess to not having the foggiest clue how to answer them. I suspect that Perry may not either. The problem is that if Perry is to explain, merely on the basis of the distinction between the argument roles of relations and the argument places of predicates, why this rather than that aspect of a raining requires contextual filling in on pain of semantic incompleteness of an utterance, he owes us answers to such questions.
Perry gets himself tied up in knots, I think, just because he denies that it’s a fact about the verb ‘to rain’ that it demands contextual provision of a place, but not contextual provision of an intensity, amount or duration. That denial is, I think, motivated in part by a mistaken conception of where in syntax the argument places of a predicate may sit. He seems to suppose that where there is no sentence level syntactic constituent and no morpheme, there can be no argument place for a predicate. But there are, I think, reasons to believe that not every to-be-contextually-evaluated argument place is explicitly expressed in the syntax of the sentence. Sometimes, I claim, a to-be-contextually-evaluated argument place hides in what I call the subsyntactic basement of suppressed verbal argument structure. (Taylor 2003) Take Perry’s favorite verb as an example. On the view that I favor, the verb ‘to rain’ has a lexically specified argument place that is thetamarked THEME that takes places as values. My claim is that in the lexicon, rainings are explicitly marked as a kind of change that places undergo. But from the point of view of sentence level syntax, such lexically specified parameters are what I call subconstituents rather than constituents. No constituent of the sentence (3) need serve as an argument place for the verb ‘rain’. Yet, despite the fact that this lexically specified argument place need not be expressed as a sentence level constituent, it makes its presence felt by “demanding,” on pain of semantic incompleteness, to be assigned a contextually supplied value. Thus though Perry is right to say that (3) is missing no syntactically mandatory sentential constituent, it is, nonetheless, semantically incomplete. The semantic incompleteness is manifest to us as a felt inability to evaluate the truth value of an utterance of (3) in the absence of a contextually provided location (or range of locations). This felt need for a contextually provided location has its source in our tacit cognition of the syntactically unexpressed argument place of the verb ‘to rain’.
To be sure, there are many changes that places plausibly undergo that are not explicitly marked as such in the sub-syntactic basement of the lexicon. Consider ‘dancing’. It is certainly true that there can’t be a dancing that doesn’t happen somewhere or other. So one might plausibly conclude that a place undergoes a dancing when a dancer dances in that place. But suppose that without saying where Laura danced, a speaker utters:
(4) Laura danced the tango until she could dance no more.
Has the speaker left something out, something required for the semantic completeness of her utterance? The answer seems clearly to be no. One can say something fully determinate, something fully truth evaluable, by uttering (4) even if context provides no place as the place where the dancing took place. Why does (4) differ from (3) in this regard?
The answer, I suggest, depends not on language independent facts about relations but on language dependent facts about verbs. ‘To dance’ and ‘to rain’ relate differently to the places where rainings and dancings happen. Unlike ‘to rain’, ‘to dance’ does not stakes out no proprietary claim on the place where a dance happens as the theme or undergoer of the dance. The theme or undergoer of a dancing is the dancer herself. The place where a dancing “takes place” is, from the lexical perspective, derivatively and indirectly associated with the dancing as the place where the dancer dances. When Laura is dancing in a place, the place does not undergo the dancing, only Laura does. Of course, she undergoes the dancing in a place. But that does not make that very place to be the theme of the dancing. Again, I take this to be a fact about the verb ‘to dance’, not a fact about the dancing as such. It is this fact about the verb, I submit, that explains why, despite the fact that one cannot dance without dancing somewhere or other, a sentence containing ‘to dance’ can be semantically complete, even if the place where dancing happens is not contextually provided. That a dancing must take place somewhere or other is a (mutually known) metaphysical fact about the universe -- a fact that supervenes on the nature of dancing and the structure of space-time. But that metaphysical fact is not explicitly and directly reflected in the lexically specified thematic structure of the verb. To say this is, of course, not to say that the place where a dancer dances is never of conversational relevance to us. It is merely to say that such conversational relevance as the location of a dancing enjoys is not a direct consequence of lexically generated requirements on thematic/semantic completeness. That is why we use optional adjuncts rather than fill a mandatory argument place to specify where a dancing takes place.
Things are otherwise with the verb ‘to rain’. The verb itself – in particular, its lexically-specified thematic structure – is the source of the felt need for the contextual provision of a place or range of places where a raining happens. Facts about the lexically specified thematic structure of the verb directly entail that nothing fully propositionally determinate has been expressed by an utterance of a sentence like (3) unless a place is contextually provided. Notice, however, that although ‘to rain’ does demand, as a consequence of its lexically specified thematic structure, that a place be provided, it permits silence about the duration, intensity or amount of the relevant raining.
If my hypothesis is correct, it will be an especially interesting and pressing matter to determine just why ‘to rain’ allows its mandatory theme to go unexpressed by any sentential constituent, even though the verb itself demands the contextual specification of a theme. For many, many verbs their lexically specified argument structured is realized by a suitable array of sentence level constituents. This must surely be the unmarked case. Why should some verbs, like ‘to rain,’ behave any differently? I admit that a fuller defense of my current hypothesis requires a principled answer to this question or a principled way of separating out the marked cases from the unmarked cases. Indeed, it may be thought that the very fact that we can say things like “it rained in Seattle,” but not things like “Seattle rained” shows that the place where it rains is not, after all, lexically marked as the theme of the raining. But this argument turns on a suppressed principle, viz., Thematicity Requires Constituency (TRC). TRC is at least tendentious. I cannot stop to argue it here, however. I will only say that TRC begs the question against the very possibility of lexically specified, but syntactically suppressed thematic structure. To accept the existence of lexically specified but syntactically suppressed thematic structure is to allow that the lexicon may directly prohibit the possibility of “bare rainings.” It does so by directly specifying both that a raining can’t happen without happening somewhere and, more particularly, by specifying that the place where the raining happens undergoes the raining. Moreover, the lexicon may make such direct thematic specifications without thereby requiring that the relevant place be the value of an overt and syntactically explicit argument place. Indeed, I suspect, but again will not argue here, that something stronger is true. Not only does ‘to rain’ not require its theme to be expressed as the value of an explicit argument. It doesn’t even permit its theme to be expressed by an explicit argument. It does, however, permit a place to be originally expressed by an optional adjunct. I suspect that the prohibition together with the permission explains why ‘it is raining in Seattle ’is okay, but ‘Seattle rains’ is not okay. It is as if the theme is “locked” in sub-syntactic basement and can’t be “raised” to the level of a surface theme-argument. Nonetheless, adjuncts can always be freely added. In particular, we can adjoin a phrase that species a place. The crucial further open question for this hypothesis is whether the allowable adjunct can be used to encode as such the thematic information that when it rains in a place the place undergoes the relevant raining.6 I suspect that the answer is no, but admit that much further argument is needed to establish the point.
Though I admit that to not having a knock-down argument in favor of the series of hypotheses just outlined, I think they are, nonetheless, plausible enough to suggest that Perry is probably half-right and half-wrong about the relationship between the argument roles of relations and the argument places of predicates. He is certainly correct to highlight the fact that a predicate may represent a more or less fulsome lexicalization of a relation. Likewise, he is certainly correct to highlight the fact that the indefinite modifiability of predicates via adjuncts gives us one way to accommodate the mismatch between the adicity of a relation and the adicity of the lexicalization of that relation. Perry’s mistake, I think, is to suppose that the bare mismatch between predicate and relation itself somehow explains why some argument roles must be contextually supplied, on pain of semantic incompleteness, while others need not be. My alternative hypothesis is that it is typically the lexical structure of the verb itself, and not language independent facts about the relation, nor even facts about the mismatch between predicate and relation, that determine what must be supplied and what need not be supplied by context. The verb itself directs its own semantic completion by, as it were, demanding that occupants of certain argument roles, but not others, be supplied in context, sometimes, perhaps, as the value of sentence level constituents, but not always. Indeed, in the kind of cases that motivate Perry’s introduction of unarticulated constituents, we have lexically generated demands on thematic completeness not tied to explicit sentence level constituents.7
Just why a given relation should be lexicalized via a predicate with certain argument places rather than others, whether they are expressed as sentential constituents or locked in the sub-syntactic basement, will often be shrouded in the pre-history of our ancient conceptions of the way things are. As Perry himself points out, before the advent of relativity theory, our linguistic progenitors took simultaneity to be absolute rather than relative to an inertial frame. The discovery that our prior lexicalization of a given relation is insufficiently fulsome may sometimes lead to a more fulsome re-lexicalization of the given relation. But such discoveries need not lead to such re-lexicalization. The indefinite modifiability of predicts via adjuncts provides us the wherewithal to compensate for the under-lexicalization of a given predicate in a highly flexible way that need not require more fulsome lexicalization of argument roles.
One final class of cases bears mentioning before we turn to propositional attitude statements. Sometimes a relation or property so rich in known complexity may be lexicalized by a predicate so thin in thematic structure that sentences containing that predicate may seem to us to express nothing determinate enough to count as complete proposition. Consider, for example, the predicate ‘..is red.’ It would seem to be one thing for a table to be red, another thing for dirt to be red, still another thing for a book to be red. Red dirt is dirt that is through and though red. On the other hand, the redness of a table would seem to depend on the redness of its upward facing surface and not to depend at all on whether its legs are red or its downward facing surface is red. Indeed, even if the downward-facing surface were entirely red that would not, on its own, suffice for the redness of a table. By contrast, the outside surfaces of the two covers of a book have an equal say in determining when a book is red. Neither the inside surfaces of the covers nor the pages themselves have any say in determining the color of the book. A book is a red book just in case both covers of the book are sufficiently red. If just one cover of the book is red, then it is a book with one red cover and one non-red cover, but it is not, it would seem, a red book. Moreover, just how red the relevant surface or surfaces must be in order that the object itself count as a red is a contextually variable matter. I am currently looking at a box of Sun-Maid raisins. No more than half of the total outer surface of the box is red. A large chunk of the outer surface consists of yellow, white and black lettering announcing the product itself and a picture of the sun maid, in all her rural glory, in front of a bright yellow sun, carrying a basket of green grapes. Though she wears a red bonnet, she is mostly dressed in white. The multi-colored surface contains nearly as much red as non-red on it, maybe even slightly more non-red than red. Nonetheless, there are clearly contexts in which it would be true to say that the box is red. For example, suppose there were another box of raisins that was purple where the Sun-Maid box is red. It would surely be correct to call the Sun-Maid box the red box (of the two) and the other box the purple box (of the two). But there are also contexts in which the multi-colored Sun-Maid box would be insufficiently red to count as a red box. Imagine, for example, an art project, the first step of which involves painting a box red. It seems pretty intuitively clear that if you painted the box so as to match the distribution of colors on a Sun-Maid box, you would have failed to follow the instruction given.
The foregoing considerations will suggest to the various advocates of unarticulated constituents, free-enrichment, implicitures and the like that a statement of the form:
(5) x is red
is semantically incomplete and must somehow be completed in context. But I want to suggest that instances of (5) aren’t so much semantically incomplete but rather what I’ll call modificationally neutral. An utterance of (5) makes the weakest possible positive statement about the redness of x. It is not, for that reason alone, semantically incomplete, however. It just that there is no modifier m’ly such that an utterance of (an instance of scheme) (5) will be strictly literally equivalent to an utterance of an instance of scheme (6):
(6) x is m’lyred.
Strictly, literally when a speaker asserts merely that x is red, she has asserted neither that x is wholly red, nor that x is partly red, nor that x is just a little bit red,nor that x is mostly red. In fact, she has asserted no modification of redness of x at all. She has simply asserted that x is red. However, because a modificationally neutral assertion of redness is the weakest possible assertion of redness possible, such an assertion would, in typical conversational contexts, be insufficiently informative. But the mutual recognition by speaker and hearer of the uninformativeness of the speaker’s modificationally neutral assertion may well generate, by roughly Gricean means of a familiar sort, some more informative, because less modificationally neutral, pragmatic conveyances of one sort or another.
The fact that modificationally neutral assertions are minimally informative, together with the fact that speakers who make such assertions typically convey something more informative and less modificationally neutral, has, I believe, led many to see semantic incompleteness where there is only modificational neutrality. But the inference from neutrality to incompleteness is fallacious. Indeed, in the spirit of Perry himself, it is worth giving a name to the relevant fallacy. I will say that one commits a fallacy of misplaced modification when one infers from the modificational neutrality of a sentence to the semantic incompleteness of that sentence. One who commits a fallacy of misplaced modification is liable to believe that sentences with a perfectly determinate, though modificationally neutral semantic contents stand in need of contextual supplementation, even when there are no explicit or suppressed argument places for context to fix a value for. I suspect that all forms of so-called radical contextualism are founded on fallacies of misplaced modification.
On the other hand, some might claim that apparent semantic incompleteness is always really just unappreciated modificational neutrality.8 The main benefit of this approach is that it enables one to explain contextual variability in what is communicated by utterances of a given sentence, while eschewing not only unarticulated constituents, but also any suppressed parameters of the sort I posit for verbs like ‘rain’. If this were right, then one could almost certainly defend an old-fashioned Gricean way of marking the semantics/pragmatics divide, on the grounds of elegance and parsimony. Though I have great affinity both for clean and parsimonious theories and for a roughly Gricean approach to the pragmatics/semantics divide, I suspect that things are much messier than paleo-Griceans envision. Many putatively incomplete sentences are, in fact, not incomplete, but complete and modificationally neutral. Failure to appreciate that fact can indeed lead one to posit both unarticulated constituents and hidden parameters where there are only Gricean implicatures, or the like. But I do not think that all instances of apparent incompleteness can be handled in this way. Indeed the pure neutrality approach, as we might call it, faces pretty much the same obstacle, in the end, that bedevils Perry’s unarticulated constituents. It cannot explain why some “extra” ingredients seem all but mandatory, while others, though they “enrich” the communicated contents remain optional. If all is implicature, why must information about place be contextually available before an utterance of (3) is conversationally acceptable, while information about duration or intensity need not be? I do not think the pure neutrality approach has much of a chance of answering such questions. But since I am not prepared to argue on this front at length in the current essay, I will for the nonce concede that deciding who has the better of this issue is a delicate matter, requiring detailed, construction-by-construction investigation of the marking of thematic roles.