1.Representing nothing: The contradictory features of negative existentials

It is well-known of certain syntactically singular NPs in natural language that they lack singular semantic values. Paradigm examples of these NPs are names of literary characters like ‘Sherlock Holmes’, names of mythical figures and legends such as ‘Zeus’ or ‘Santa Claus’, and definite descriptions that purport to denote failed scientific posits like ‘the ether’. According to standard static semantic theories, expressions that lack semantic values are not able to participate in the process of propositional composition. It is not too surprising, therefore, that many semanticists and philosophers of language think that utterances of simple declarative sentences containing empty singular NPs in argument position do not express structurally complete propositions or thoughts.11.Moderate views, such as Braun’s (2005)Braun, David 2005 “Empty Names, Fictional Names, Mythical Names.” Nous 39: 596–631. , contend that simple sentences containing empty NPs express gappy propositions/thoughts which can be represented schematically as <___, F>, where F stands for a property predicated of the (empty) denotation of the NP. A somewhat narrower but conceptually similar view on structurally challenged propositions is suggested by Salmon (1998)Salmon, Nathan 1998 “Nonexistence.” Nous 32: 277–319. . More radical accounts, such as Walton’s (1990)Walton, Kendall L. 1990 Mimesis as Make-Believe. Cambridge: Harvard University Press., hold that, taken literally, sentences like ‘Sherlock Holmes is a detective’ or ‘Santa Claus is coming tonight’ do not express any proposition/thought at all. Structural incompleteness does not pose insurmountable obstacles to online sentence processing. Speakers endowed with the requisite background knowledge and metalinguistic abilities understand utterances of ‘Sherlock Holmes is a detective’ or ‘Zeus is a God’ quite easily. And they are surely able to grasp the acceptability differences between ‘Sherlock Holmes lives at 221B Baker Street’ and ‘Zeus lives at 221B Baker Street’ with minimal effort. Processing data like these are explanatorily independent from the theoretical issue of truth-evaluability, however. In spite of their understandability and acceptability, uttered sentences with empty singular NPs are usually taken to be non-truth-evaluable within standard static semantic frameworks. In these frameworks, only complete propositions or thoughts can be evaluated with respect to truth and falsity. Propositional or thought structures which contain empty constituents are therefore judged to be neither true nor false.22.Of course, those who agree with this claim hold that propositions have a constituent structure. It is worth noting here that static semantic frameworks may be combined with an alternative view of propositions. On this alternative view, propositions are simple, unstructured entities without a constituent structure. Note also that although empty singular NPs do not pose a compositional problem for such a view, their truth-conditional contribution remains highly controversial. The crucial question is that of the truth-value of unstructured propositions that are about nonexistent things. For a detailed discussion of this difficulty, see Merricks (2015)Merricks, Trenton 2015 Propositions. Oxford: Oxford University Press. .

This conception may yield adequate explanations in many theoretical contexts. If we are prepared to grant that empty singular NPs lack semantic values because they lack objective worldly correlates, we can explain why utterances of simple declaratives containing tokens of ‘Sherlock Holmes’, ‘Zeus’, and their likes have no truth value. There is nothing in reality – neither in the realm of concrete objects, nor in the realm of abstracta – to which these tokens might stand in a semantic relation. The intended objects of representation, i.e. the values of token NPs, are clearly absent in these types of utterances. So it is hard to imagine exactly how we could assign truth values to utterances in which tokens of empty singular NPs occur.

One obvious difficulty with the above sketched truth-value analysis is that it cannot be extended to negative existentials. Consider the following sentence schema where the left subscript indicates lexical/semantic emptiness:

Simple declaratives with _eNPs differ from the instances of (1) in at least three respects. First, taken at face value, simple declaratives with _eNPs cannot be interpreted as representations of real states of affairs. An utterance of ‘Holmes is wise’, for example, does not represent Holmes’s being wise since this state of affairs does not obtain in the actual world. The same is obviously true of all other predicative adjective phrases which at first sight may appear to be able to represent the distinctive properties of the character in question. Therefore, declaratives of the form ‘Holmes is AP’ cannot be used to represent states of affairs that actually obtain.33.One might object that ‘Holmes is fictional’ provides a clear counterexample to this generalization. It can be argued that Holmes exists only as a character in a fictional story, and this state of affairs obtains in the actual world. On my view, the objection would be successful only if being fictional were conceived as one of the distinctive properties of Holmes. However, if we are willing to connect Holmes’s existence to the “world” of the story in which he figures, then we cannot say that he is fictional. In contrast, negative existentials with _eNPs seem to have the function of representing how things stand in the actual world. Perhaps the most natural interpretation of ‘Holmes does not exist’ is that it represents the whole world as it actually is; that Holmes does not belong to the overall inventory of what exists. Second, since simple declaratives with _eNPs fail to express complete propositions or thoughts, they cannot be used for contributing new information to the discourse. In contrast, some utterances of negative existentials with _eNPs seem to be suitable to add new information to an existing representation of the world. For example, ‘Santa Claus does not exist’ can obviously express disappointing new information for children believing in the real existence of the mythical figure. And third, most importantly, while simple declaratives with _eNPs are taken to be truth-valueless, negative existentials with _eNPs are thought to be able to express true propositions or thoughts. The relevant observation is that ‘Santa Claus does not exist’ is not merely apparently, but literally true. The assumption that (1) has true instances is especially disturbing for the proponents of standard static semantics because it generates a sharp contradiction in their argumentative framework. On the one hand, _eNPs are classified as valueless expressions which are not appropriate for propositional composition. On the other hand, these very same expressions are seen as constituents of utterances that are literally true. The distinctive features of negative existentials can now be summed up in the following three statements:

On my view, anyone striving for a comprehensive account of the behaviour of _eNPs should give an explanation of the intuitive data encompassed in RELATIONALITY, INFORMATIVITY and TRUTHFULNESS. All these data show us that _eNPs behave in negative existentials as if they were ordinary filled expressions; in other words, it is shown that despite their lexical/semantical emptiness, _eNPs are able to perform all the usual compositional and propositional functions when placed in an appropriate sentential environment. This anomalous behaviour must be explained. The question is: how can this be done in a principled way?

The most promising explanatory strategy is to try to find a value-providing procedure that somehow compensates the anomalous effects arising from the lexical/semantical emptiness of these noun phrases. The proposed strategies should satisfy the following two basic conditions: (a) their reliability must be demonstrated on independent theoretical grounds, and (b) they should be coherent and general enough to accommodate the intuitive data mentioned above. In recent years, two such strategies have been proposed and developed in the literature.

The first is conservative in the sense that it leaves the framework of standard static semantics intact. Followers of this strategy claim that the required values for empty names and empty definite descriptions are supplied by some kind of pragmatic procedure. In general, while _eNPs are acknowledged to be lexically/semantically empty, instances of (1) are nevertheless seen as conveying complete propositions or thoughts. The background idea is that, within a given conversational network, every _eNP is associated with a more or less stable set of pieces of individuating information. The link between an _eNP and an information set is established usually through associative learning or through other forms of conditioning. When it comes to utterances of negative existentials, participants of the conversation make an attempt to mobilize certain common elements of this set. Thus, in a successful conversation, an _eNP gets to be associated with some piece of individuating information which is assumed to be currently available for each party. The procedure is supposed to work in the following way:

Whichever element is selected from the available set of information in (2′), the pragmatically enriched sentence will be interpreted as conveying something true. Adams and Dietrich (2004)Adams, Fred, and Laura L. Dietrich 2004 “What’s in an (Empty) Name?” Pacific Philosophical Quarterly 85: 125–148. think that the transition from (2) to (2′) is implemented by a Gricean conversational implicature. This is intended to mean that in order to identify the proposition or thought expressed by (2′), conversational participants need to perform a pragmatic inference. In contrast, Wyatt (2007)Wyatt, Nicole 2007 “The Pragmatics of Empty Names.” Dialogue 46: 663–81. thinks the transition is implemented by a kind of descriptive enrichment. That is, according to Wyatt’s hypothesis, (2′) is actually expressed and not merely implicated by an utterance of (2). In spite of the difference in their approaches, Adams and Dietrich and Wyatt are in agreement that negative existentials can convey truths by virtue of the fact that the _eNPs in them are substituted pragmatically with some associated piece of information.44.These and similar views are collectively called Pragmatic Millianism in the recent literature. Millianism makes essentially the same predictions about empty expressions as any other standard static semantic theory. I don’t want to delve too deeply into the details of this strategy because it rests on a shaky assumption: namely, that merely associating an _eNP with some additional piece of information is sufficient to generate a truth-evaluable proposition or thought. Mousavian (2014)Mousavian, Seyed N. 2014 “Empty Names and Pragmatic Millianism.” Thought 3: 49–58. rightly observes that there are too many counterexamples to this assumption. Consider again the transition from (2) to (2′). Let us suppose that in a particular conversational network the empty name ‘Sherlock Holmes’ has standardly become associated with the individuating description ‘Dr. Watson’s best friend’. Pace Adams and Dietrich and Wyatt, tokenings of the pragmatically enriched sentence ‘Dr. Watson’s best friend does not exist’ should then be interpreted as literally true utterances. This cannot be right, however. Given that ‘Dr. Watson’ lacks an objective worldly correlate, it is an empty name, too. And if it is an empty name, it is valueless (according to the background semantics), so it cannot be used to generate a true utterance. The problem in general is that, in principle, certain _eNPs might be exclusively associated with sets of information the elements of which are themselves genuinely empty expressions. In such cases, it is impossible to find a non-circular way to accommodate TRUTHFULNESS.55.One might try to remedy the situation by limiting information sets only to those containing non-empty elements. Negative existentials containing _eNPs that are associated with empty expressions would then be treated as truth-valueless at the level of pragmatics. The success of this move is doubtful since even if we substitute ‘Sherlock Holmes’ with ‘Dr. Watson’s best friend’, (2) still seems to express something true. For these reasons, the first strategy fails to satisfy condition (b).

The second explanatory strategy is innovative in the sense that it rejects the framework of standard static semantics and adopts instead a dynamic approach to meaning. This strategy has been developed in two variants: one based on the tradition of dynamic semantics, and another one applying a dynamic pragmatic approach. Both variants are suggestive and interesting in their own right, so I shall discuss each of them in separate sections. I shall argue, however, that the results of this second explanatory strategy, no matter how well formulated or implemented, also fail to satisfy condition (b).

2.The dynamic semantic approach to negative existentials

Static semantic theories are standardly committed to a few central theses. Within this tradition, it is commonly held that the semantic values of complex expressions are compositionally determined by the values of their constitutive parts and the syntactic relations between these parts. There is also a consensus that the values of declarative sentences are propositions or thoughts that represent things as being a certain way. It is widely accepted, too, that declaratives have, or determine, truth-conditions relative to contexts to which they belong.

Fundamental theses like the ones just mentioned have obviously far-reaching theoretical consequences. One might assume, on this basis, that static semantic theories involve a closed hierarchy of theoretical claims, all of which are in strong opposition to the claims of their dynamic rivals. This is not so. As Yalcin (2012)Yalcin, Seth 2012 “Dynamic Semantics.” In The Routledge Companion to Philosophy of Language. ed. by G. Russell, and D. Graff Fara, 253–279. London: Routledge., Lewis (2017)Lewis, Karen 2017 “Dynamic Semantics.” Oxford Handbooks Online. and others have pointed out, these two frameworks are not wholly exclusive. One point of agreement between static and dynamic semantic theories concerns the issue of meaning constitution: both parties take the principle of compositionality for granted. Dynamic semanticists need not necessarily deny that declarative sentences in context have truth-conditions, which may be identified as a further common theoretical feature. Moreover, by assigning a key role to the notion of possible world in their metatheories most dynamic views acknowledge the representational character of language. This strongly suggests that there is indeed no sharp dividing line between static and dynamic theories of meaning. On the other hand, dynamic views are much more sensitive to the phenomena of conversational changes than their static predecessors. Sentences are thought of as having a potential to induce changes in the mental states of the participants of the conversation. If dynamic semanticists are on the right track and conversational contexts are indeed sets of mutually accepted beliefs, then one should conceive sentence meanings as rules or instructions to change contexts in a certain manner. Seen from a static perspective, such phenomena would largely lie outside of the domain of semantics proper. Groenendijk and Stokhof (2000)Groenendijk, Jeroen, and Martin Stokhof 2000 “Meaning in Motion.” In Reference and Anaphoric Relations, ed. by K. von Heusinger, and U. Egli, 47–76. Dordrecht: Kluwer. characterize the current research situation in a helpful way:

[Dynamic semantics] links meaning in a principled way to interpretation viewed as a process, one that relates to larger units than just isolated sentences. The bite lies not in propagating a new slogan (A: ‘Meaning is truthconditional content!’; B: ‘No, meaning is context change potential!’), but rather in developing a conceptual and logical apparatus which is suitable to implement this idea, and to apply it to concrete phenomena.(Groenendijk and Stokhof 2000Groenendijk, Jeroen, and Martin Stokhof 2000 “Meaning in Motion.” In Reference and Anaphoric Relations, ed. by K. von Heusinger, and U. Egli, 47–76. Dordrecht: Kluwer. , 68)

The question for us is whether the conceptual and logical apparatus mentioned by Groenendijk and Stokhof is capable to handle effectively the problem posed by negative existentials containing _eNPs. Clapp (2009)Clapp, Lenny 2009 “The Problem of Negative Existentials does not Exist: A Case for Dynamic Semantics.” Journal of Pragmatics 41: 1422–1434. answers this question affirmatively. Clapp’s main claim is not only that dynamic semantics can explain the intuitive data concerning the instances of (1), but that the contradictory features of negative existentials disappear completely in a dynamic framework.

Now, let us examine, in some detail, how Clapp arrives at this claim. The argument begins with a clarification of the required conceptual background. The first important thing to note is that Clapp applies a dynamic conception of language use according to which utterances are guided by sets of beliefs mutually shared by the participants involved in the conversation. Each of these sets contains a particular representation of the world that speakers and hearers are committed to and thus they record which possibilities are excluded and which are still open at a given stage of the conversation. Given this common ground view of context, utterances can be treated as if they were instructions to induce changes in the belief sets of the participants.66.The notion of common ground was introduced into the semantic/pragmatic literature by Robert Stalnaker in 1970. For a more recent interpretation of this notion, see Stalnaker (2014)Stalnaker, Robert 2014 Context. Oxford: Oxford University Press. . Accordingly, the semantic value or meaning of a particular utterance can be described in terms of how it affects the common ground. Under normal circumstances, utterances of simple declarative sentences are successful in this respect, that is, they are able to perform an update operation on the common ground. When an update is successful, certain possibilities become eliminated from the prior context, namely those that are incompatible with the sentence that performed the update. As a result, the belief set of the posterior context remains consistent. Unsuccessful updates exclude exactly those open possibilities from the prior context on which the performative effect of their own update operation would depend, so the posterior contexts they generate are inconsistent.

Clapp maintains, following Heim (1982)Heim, Irene 1982 The Semantics of Definite and Indefinite Noun Phrases. Ph.D. Thesis, Amherst: University of Massachusetts., that updates of the common ground are constrained by presuppositions that are inherent in the lexically/semantically encoded meaning of certain expressions. For example, proper names and definite descriptions standing in argument position seem to carry both a familiarity presupposition and an existential presupposition that the common ground must satisfy. The sentence ‘Freddie comes to the party tonight’ can be appropriately uttered only if the proper name ‘Freddie’ is already familiar to each participant prior to the utterance and the existence of Freddie is presumed to be common ground. And similarly, an utterance of ‘The winner of the race comes to the party tonight’ is appropriate only if the participants of the conversation already possess information about a conversationally relevant race and they believe that the unique winner of this race really exists.77.It is worth noting that this view is not universally accepted. For example, Coppock and Beaver (2015)Beaver, David, and Elizabeth Coppock 2015 “Definiteness and Determinacy.” Linguistics and Philosophy 38: 377–435. suggest, in contrast to Heim, that definite descriptions presuppose only a definiteness property that they call weak uniqueness. And they argue, persuasively to my mind, that weak uniquness is independent of existence. For the purposes of the present paper, however, I assume that Heim’s view is correct. Heim’s view predicts, and Clapp seems to agree, that when a particular presupposition is not satisfied by the common ground, then the utterance that carries that presupposition will be felt to be conversationally inappropriate.

There are some cases, however, in which an uttered sentence is perceived to be appropriate in spite of the fact that the common ground does not satisfy its presupposition(s). If someone says that ‘My dog has been lost since the storm’, the utterance may be appropriate even when it was not common belief before the utterance that the utterer owns a dog. In this case, the presupposition concerning the existence of the dog will be accommodated after the utterance has taken place. This type of accommodation may be called global since it has consequences for the common ground as a whole. From then on, one can no longer suppose that the utterer is not a dog owner, which indicates that the common ground has been successfully updated. Presumably, presupposition accommodation may also happen in non-global, local contexts. Consider an utterance of the following pair of sentences: ‘I don’t know whether every nation has a king. But every nation that loves its king is peaceful’.88.The example is taken from Pearson (2010)Pearson, Hazel 2010 “Presupposition in Local Contexts: Why Global Accommodation is not Enough.” Language at the University of Essex (LangUE) 2009 Proceedings, 70–80.. Is the second sentence an inappropriate continuation of the first one? Surely not. But in this case there is no reason to assume that the presupposition of the second sentence should be accommodated globally. It would be misleading to think that it updates the common ground with the belief that every nation has a king. The first sentence has already expressed ignorance in this regard. Thus, if there is any accommodation at all, it ought to occur at a local context: the quantifier phrase ‘every nation’ is accommodated within the boundaries of its clause without entailing any global consequence.

Now let us move on to Clapp’s argument which makes use only of this limited conceptual apparatus. Clapp applies a two-step analytical procedure to demonstrate that negative existentials containing _eNPs do not pose any problem within a dynamic semantic framework. The following short conversation serves to illustrate the main point:

In order to make this conversation suitable for dynamic analysis, let us assume two things: first, assume that the fact that B is a dog owner is part of the common ground; and second, assume that there is a disagreement between A and B concerning the existence of Sherlock Holmes. A believes (mistakenly) that Holmes exists, but B believes (correctly) that there is no such person. Therefore, when it comes to the interpretation of A’s utterance, B recognizes that the common ground fails to satisfy the existential presupposition carried by the proper name ‘Sherlock Holmes’. Then, says Clapp, B may attempt to repair the presupposition failure in two consecutive steps. The first step consists of the accommodation of A’s utterance: B simply adds to the common ground the belief that the name ‘Sherlock Holmes’ represents an existing person. But, after accommodating globally the missing presupposition, B discerns that the resulting common ground is incompatible with her broader belief set. She then makes explicit this incompatibility by uttering the negative existential ‘Sherlock Holmes does not exist’. This is the second step. Note that B’s utterance is entirely appropriate, since the belief that concerns the existence of Sherlock Holmes is already part of the common ground. But B’s final intent is to provide for A an updated common ground in which the existential presupposition of the name is not satisfied. This can happen only if it is possible for the utterance to deny its own presupposition. On Clapp’s view, negative existentials can be expected to fulfil exactly this requirement. In his own words:99.Note that instead of ‘existential presupposition’, Clapp originally uses the term ‘referential presupposition’ in his formulation, but he emphasizes repeatedly that an NP triggers a referential presupposition when it requires that the common ground include the information that the mental file to which the NP is linked pertains to an existing entity. See, for example, Clapp (2009Clapp, Lenny 2009 “The Problem of Negative Existentials does not Exist: A Case for Dynamic Semantics.” Journal of Pragmatics 41: 1422–1434. , 1430–1431).

So we can state this … conclusion more precisely: to solve the problem of negative existentials we must countenance the fact that negative existentials deny the existential presupposition carried by their definite NP, if such a presupposition is carried.(Clapp 2009Clapp, Lenny 2009 “The Problem of Negative Existentials does not Exist: A Case for Dynamic Semantics.” Journal of Pragmatics 41: 1422–1434. , 1429)

An alternative way to state this conclusion would be to say that utterances of negative existentials can be regarded as examples of presupposition cancellation (or presupposition suspension). What consequences does this have? If their primary function in conversation is to cancel certain presuppositions, instances of (1) lose their contradictory features. Obviously, to deny a presupposition of an utterance is not the same as to produce a contradiction. Add to this that negative existentials with _eNPs have typically existential presuppositions. This kind of presupposition pertains to beliefs that concern the worldly correlates of _eNPs. Thus, it can be understood how negative existentials can be used to express, at least indirectly, something about the world. Moreover, it can be demonstrated that utterances of negative existentials are informative in the sense that they can update the common ground. Recall the example above. B’s utterance in (4) is clearly informative for A, because it changes the conversation in a way which is incompatible with her prior beliefs (i.e. her beliefs according to (3)). In interpreting B’s utterance, A acquires a new belief. And finally, because the existential presuppositions of _eNPs are cancelled, and because _eNPs have in fact no worldly correlates, predicating nonexistence of _eNPs may yield true utterances. Truth should be considered here from a dynamic perspective: dynamic truth signals how the belief sets on which the context change potentials of utterances operate correspond to something in the world.1010.It should be kept in mind, however, that in dynamic semantics only conversational units or whole conversations are said to have truth-values, not individual utterances. Taken as a coherent unit, our Example (3)–(4) may thus be regarded as expressing something true. In short, Clapp’s view seems to be in agreement with RELATIONALITY, INFORMATIVITY and TRUTHFULNESS. If correct, this would be a significant result.

Unfortunately, the argument does not succeed. According to dynamic semantic theories, one should globally accommodate a presupposition only if the resulting common ground is consistent.1111.On the consistency requirement, see, for example, Roberts (1995Roberts, Craige 1995 “Domain Restriction in Dynamic Semantics.” In Quantification in Natural Languages, ed. by E. Bach et al., 661–700. Dordrecht: Springer., 683). In connection with this, Simons (2003Simons, Mandy 2003 “Presupposition and Accommodation: Understanding the Stalnakerian Picture.” Philosophical Studies 112: 251–278. , 272) argues that in dynamic semantic theories “sentence presuppositions impose “hard” constraints: if a context fails to satisfy the presuppositional requirements of an incoming sentence, the sentence is undefined relative to the context, and update cannot proceed.” Gauker (2008)Gauker, Christopher 2008 “Against Accommodation: Heim, Van Der Sandt, and the Presupposition Projection Problem.” Philosophical Perspectives 22: 171–205. is more radical in that he denies that accommodation can function as a repair mechanism when presuppositions are not entailed by the common ground. After hearing an utterance of ‘My dog has been lost since the storm’, the participants of the conversation may believe that the utterer has a dog even if this information was not part of the common ground before the utterance. Normally, a global case of accommodation like this wouldn’t impose even a minimal risk of inconsistency. The same does not hold for sentences containing _eNPs. Utterances of these sentences are puzzling since the presuppositions they carry seem to resist accommodation altogether: a consistent common ground cannot be updated with an utterance that presupposes the existence of a nonexistent thing. If this is so, Clapp is wrong when he claims that after hearing A’s utterance, B can accommodate the presupposed existence of Sherlock Holmes, in spite of the fact that the accommodation yields an inconsistent common ground. The belief concerning the existence of Sherlock Holmes may not so simply be added to their commonly accepted set of beliefs. But then, again contrary to what Clapp holds, the sentence ‘Sherlock Holmes does not exist’ cannot be appropriately uttered by B. I think Clapp’s analysis of the example rests on an implausible assumption (i.e. the assumption that a missing presupposition can be accommodated even if it results in an inconsistent belief set), and therefore provides no support for the conclusion he wants to draw.

One possible response to this critique would be to say that the existential presupposition of ‘Sherlock Holmes’ is only locally accommodated in B’s utterance. This would make the utterance conversationally appropriate without implying the process of common belief update. There are two problems with this response. The first problem is that, in this case, local accommodation would be motivated by pragmatic rather than semantic factors. One might argue, following the insights of Schlenker (2010)Schlenker, Philippe 2010 “Presuppositions and Local Contexts.” Mind 119: 377–391. , that the local grammatical environment of the name (‘Sherlock Holmes does not ___’) entails the existence of the relevant person. Regardless of how the sentence ends, the required presupposition is thus accommodated. That is, at this point in the incremental processing of the utterance, B recognizes that she should accept temporarily ‘Sherlock Holmes’ as a name of a person. Otherwise her utterance would make no sense. But, to repeat, this recognition is based on a pragmatic, not a semantic inference.

Even if one sets this aside, there is a second problem. On Clapp’s view, negative existentials deny or cancel the presuppositions that are carried by their (empty) noun phrases. But one might wonder how B’s utterance can cancel the existential presupposition of ‘Sherlock Holmes’, if that presupposition has already been accommodated within the local environment of the name. The view Clapp proposes makes use of two components of the dynamic framework – accommodating and cancelling the existential presuppositions of _eNPs – which cannot be linked together so that they form a coherent explanation. From these observations, one may conclude that this explanatory strategy also fails to satisfy condition (b).

3.An alternative proposal: Static semantics with dynamic pragmatics

As we have seen in Section 1, one option for explaining how instances of (1) are able to convey complete propositions or thoughts is to invoke a pragmatic mechanism. The argument consists of two parts. First, _eNPs are said to be associated with conventionally fixed sets of information. Second, these sets of information are thought of as supplying values to _eNPs at the level of their pragmatics. Proponents of this option apply a standard static semantic framework accompained by a static pragmatics. In agreement with Mousavian’s (2014)Mousavian, Seyed N. 2014 “Empty Names and Pragmatic Millianism.” Thought 3: 49–58. opinion, I have argued that explanations of this kind are inadequate. In principle, an _eNP may be associated with a certain piece of information which contains other _eNPs. If this is the case, the pragmatic explanation becomes circular.

However, since Stalnaker’s work on assertion (Stalnaker 1978Stalnaker, Robert 1978 “Assertion.” Syntax and Semantics 9: 315–332.), there have also been attempts to apply a dynamic pragmatic view to the problem of _eNPs and their role in negative existentials. The demand for a dynamic pragmatics was originally motivated by the recognition that static semantics in itself is unable to provide an account of a wide range of truth-conditionally relevant processes. These are typically processes in which a propositional constituent changes its value, but the change is not lexically/semantically mandated. In the recent literature, such processes are categorized under the label ‘modulation’.1212.Modulation, as a genre, includes the cases of meaning transfer, loosening and free enrichment. For a systematic treatment of the topic, see Recanati (2004)Recanati, Francois 2004 Literal Meaning. Cambridge: Cambridge Universtity Press.. Propositional constituents which undergo the process of modulation change their semantic values for general pragmatic reasons. These include the rationality of the conversation, cooperative norms and other non-semantic factors. If pragmatic processes can indeed induce “dynamic” changes in the behaviour of certain propositional constituents, then it is not unreasonable to assume that negative existentials contain expressions whose propositional contribution is also modulated in one way or another.1313.The main explanatory difference between the two dynamic approaches to meaning is rather easy to state (at least in the present context): while the dynamic semantic view locates dynamicness at the level of entire conversations, the dynamic pragmatic view locates dynamic effects at the sentential/compositional level.

The most elaborated version of this idea is to be found in Spewak (2016)Spewak, David C. 2016 “A Modulation Account of Negative Existentials.” Philosophia 44: 227–245. . Although Spewak’s primary goal is to give an account of negative existentials that contain names of literary characters, his view can be extended to other kinds of negative existentials, too.

Let us suppose that utterances of negative existentials with _eNPs undergo a pragmatic process of modulation. Then the first question that needs to be answered concerns what precisely the factors are that generate this process. One such factor, perhaps the most important one, is the topic of the conversation. According to Spewak, the topic of a particular conversation determines how the meaning or sense of a propositional constituent is modulated. However, once this first answer has been given, a second question immediately arises: which of the propositional constituents is or should be involved in the process of modulation? Since negative existentials are uttered in conversations that concern issues of existence, the obvious candidate for this role is the constituent which corresponds to the predicate ‘exists’.1414.In a similar spirit, Hanks (2015)Hanks, Peter 2015 Propositional Content. Oxford: Oxford University Press. contends that the key observation for solving the problem of negative existentials is that the predicate ‘exists’ creates a shifted context. According to this approach, when someone utters that ‘Zeus does not exist’, then the context of predication shifts to the predicative act itself. Instead of talking about Zeus, the utterer says that there is no Zeus-type act of predication which has true instances. An anonymous reviewer suggests that modulation might also be conceived as generating a metalinguistic context-shift. If this is possible, then the modulation of the existential sentence ‘Zeus does not exist’ would yield the metalinguistic sentence ‘The name Zeus does not refer’. This is quite close to the proposal put forward by Walton (2015Walton, Kendall L. 2015 In Other Shoes. Oxford: Oxford University Press., 102), but I think the metalinguistic account of modulation would be incompatible with Spewak’s view, as he presented it in his 2016 paper.

In its most general use, the predicate ‘exists’ may be considered as equivalent with ‘to be’ or ‘have being’. This is the lexical sense of ‘exists’ which can be predicated of all things, be they concrete or abstract. This variant of the predicate is well suited for inquiries into the logical or metaphysical foundations of reality. But in everyday conversations, ‘exists’ is used typically in a different manner. The following passage is an attempt to clarify the difference:

For children, what matters is whether there is something in their closets that is physically the Boogie man, or a real monster. When parents tell their children that the Boogie man does not exist, they are trying to convince them that no one is going to eat them in the middle of the night. For the child and the parents, there is no thought given to whether some abstract artifact is the Boogie man. What matters in most ordinary cases is what physically exists. When we think about existence and what exists, we tend to focus on physical existence.(Spewak 2016Spewak, David C. 2016 “A Modulation Account of Negative Existentials.” Philosophia 44: 227–245. , 233, emphases in the original)

The general idea is, then, that when our ordinary, medium-sized world is the topic of a conversation, the lexical sense of ‘exists’ is strenghtened so that it applies only to a subset of all things, namely those that are part of the physical world. For purposes of illustration, let us introduce a simple technical notation. Consider once again our earlier example (now numbered (5)):

In (5), the downward arrow indicates that the semantic value of the predicate undergoes a process of modulation, while the subscript ‘ct’ denotes a conversational topic with respect to which the utterance is interpreted. If ‘ct’ denotes a topic as described above, then ‘exists’ acquires a sense which is stronger than the sense of its lexical variant. Because of this, ‘↓exist_ct’ in (5) can be said to be equivalent with ‘to be physical’ or ‘to exist as a physical thing’. In the end, Spewak’s reasoning allows us a reading of (5) which is apparently free from contradiction:

Given that the literary character ‘Sherlock Holmes’ is the product of Arthur Conan Doyle’s artistic imagination, there is no physical thing with which it is identical. Therefore, (6) seems to express a true proposition or thought.

Notice, in addition, that if the subscript ‘ct’ in (5) were used to denote a different conversational topic, (5) would allow another reading. In a philosophical debate, for example, where the topic is existence in its entirety, ‘exists’ probably retains its original lexical sense. Die-hard proponents of fictional realism would then surely contend that utterances of ‘Sherlock Holmes does not exist’ are false because literary characters belong to the class of abstract things, and the members of this class exist.1515.This stance is characteristic of the artefactual theory of fiction. Proponents of this theory claim that literary characters have a similar ontological status as nations, marriages and other cultural products. All such things are abstract artefacts created intentionally by the human mind. For a detailed overview, see Thomasson (1999Thomasson, Amie L. 1999 Fiction and Metaphysics. Cambridge: Cambridge University Press., 2003Thomasson, Amie L. 2003 “Speaking of Fictional Characters.” Dialectica 57: 207–226. ).

Spewak’s dynamic pragmatic account has some commonalities with the dynamic semantic proposal discussed in the previous section. It is certainly independently motivated: it has been shown through several case studies how modulation affects the values of the constituents of propositions (or thoughts). Recanati (2004)Recanati, Francois 2004 Literal Meaning. Cambridge: Cambridge Universtity Press. and others have argued convincingly that the sense of colour adjectives like ‘green’ or ‘red’ may vary under the influence of the topic of the conversation in which they are uttered. The same has been observed in the behaviour of gradable adjectives like ‘small’ or ‘tall’. Moreover, Spewak’s account is no less general than Clapp’s in the sense that it allows the value of ‘exists’ to be modulated in a variety of ways. Spewak describes a case where social scientists are engaged in a debate about the reality of race. Presumably, many social scientists are of the opinion that race is merely a social construct that lacks a biological basis. Thus, in uttering ‘Race does not exist’, they may want to reject the view according to which race is a purely biological phenomenon. If this is so, ‘↓exist_ct’ should be interpreted here as being equivalent with ‘to be biological’ or ‘to exist as a biological thing’. In such a conversational setting, Spewak claims, ‘Race does not exist’ may be regarded as a true utterance.

At first sight, this explanatory strategy might seem to fit neatly with our pretheoretical intuitions about negative existentials (i.e. it might seem to have the capacity to account for RELATIONALITY, INFORMATIVITY and TRUTHFULNESS). It uses a conceptually conservative vocabulary which is flexible enough to cover a wide range of conversational settings where such sentences may be uttered. I contend, however, that we may have serious doubts about the tenability of Spewak’s account because the strategy it uses is one that cannot be applied to all types of negative existentials.

To see the problem, recall the way modulation is supposed to work. As already mentioned, modulation has to be thought of as a pragmatic process that operates on the lexical senses of certain expressions. It is important to stress that sense modulations must be performed before the procedure of propositional composition has been completed. This holds also for the modulation of the lexical sense of ‘exists’. Conversational topic first affects the sense of the predicate, and then the modulated sense is calculated in the compositional procedure. As we have seen, that is why negative existentials like (5) can express true propositions or thoughts. But in order to express true propositions or thoughts, negative existentials must express structurally complete propositions or thoughts in the first place. This is the point at which Spewak’s explanatory strategy seems to break down. For it is not sufficient merely to say that the sense of ‘exists’ occasionally undergoes a process of modulation. One should also explain how negative existentials containing genuinely empty NPs, rather than apparently empty NPs, can express structurally complete propositions or thoughts. To borrow an example from Kroon (2003)Kroon, Frederick 2003 “Quantified Negative Existentials.” Dialectica 57: 149–167. , consider the case of ‘Max’. In talking about Freddie, a speaker says ‘He went to the max’ intending to emphasize her belief that Freddie definitely gave it his all. A hearer overhears the ‘max’ part of the utterance and interprets it as if it were a token occurrence of the proper name type ‘Max’ and asks: ‘What was the reason for doing that? And anyway, who is Max?’ The speaker then corrects the hearer’s mistake by saying that ‘In fact, Max is not a person, he is the result of your mishearing my utterance. Max does not exist.’ According to Kroon, ‘Max’ has been introduced into the above mini-conversation as a genuinely empty NP: it was used neither as a name of a real person nor as a name of a fictional character. Applied to this case, Spewak’s explanatory strategy delivers the following result:

Here ‘ct’ denotes a conversational topic which is centered on flesh-and-blood persons. Accordingly, the modulated sense of the existence predicate in (7) can be said to be equivalent with ‘to be human’ or ‘to exist as a biological human being’.

Spewak might be right in that (7) allows a pragmatically modulated reading, (8), that is non-contradictory. It is confusing, however, that the proposition or thought expressed by (8) still contains a compositionally inert constituent (i.e. ‘Max’). Unfortunately, there is no hint given in Spewak’s modulation account of how this remaining structural semantic incompleteness might be removed. It bears emphasis that utterance misinterpretation is not the only source of genuinely _eNPs. Dream-like hallucinations and figments of imaginations may encourage speakers to create similar expressions. Note also that genuinely _eNPs can be introduced into conversations even by deliberate intentional acts.1616.An example for this can be found in Salmon (1998)Salmon, Nathan 1998 “Nonexistence.” Nous 32: 277–319. . Let us say that ‘Nappy’ stands for the actual emperor of France if there is such a person, and it is an empty name otherwise. Since there is no such emperor, ‘Nappy’ is a deliberate genuinely _eNP. We are forced to the conclusion, again, that the proposed explanatory strategy is not general enough, that is, it fails to satisfy condition (b).

4.Is there a way out?

It is not easy to answer this question. A promising line of inquiry would be to revise the core principles of standard static semantics. One might draw some conceptual inspiration from the rather neglected tradition of Goodman (1968)Goodman, Nelson 1968 Languages of Art: An Approach to a Theory of Symbols. Indianapolis: Bobbs-Merrill.. Goodman makes a sharp conceptual distinction between representations of objects and object-representations. According to this distinction, utterances containing lexically/semantically filled NPs may be thought of as representations of existing objects. Utterances of sentences with _eNPs, in contrast, may be interpreted as expressing merely object-representations. An important feature of the latter type of utterance is that although it is not related to the worldly correlate of its NP constituent (since there is no such thing), it is semantically non-defective. Instances of (1), for example, may be taken as involving _eNP-representations. This enables us to interpret ‘Sherlock Holmes does not exist’ as a structurally complete, meaningful utterance. If so, one may say that it has a certain kind of representational content, a content that can enter in logical/inferential relations with other contents. I believe Goodman provides at least a good starting point for an adequate analysis of negative existentials with _eNPs, but I will not argue for this here.1717.In other work, I have argued that empty names occurring in fiction are best conceived as non-relational object-representations. For the details, see Vecsey (2015)Vecsey, Zoltán 2015 “Fiction and Representation.” SATS 16: 202–215. . Instead, I would like to call attention to a methodological issue that has been rarely, if ever, discussed in this particular area of research.

The methodological issue in question concerns the status and treatment of data in semantic theorizing. It is not hard to notice that certain types of data are taken as theoretically reliable in the literature on negative existentials. Perhaps the most striking example is the treatment of intuitive judgements about the truth-conditionally relevant features of such sentences. It is often said that a particular utterance of a negative existential is “intuitively truth-evaluable” or is “intuitively true” with respect to a context. Intuitive judgements like these are considered to be reliable in the sense that they are assumed to provide a secure basis for theorizing. In this paper, we have implicitly taken for granted that this methodological stance is correct.

The main disadvantage of this methodological stance is, however, that it assimilates intuitive judgements to facts. Intuitive data may be thought of, in a certain sense, as given, but they hardly have such a determinate status as facts. Quite the contrary, their basic property is their uncertainty. Intuitive judgements are typically unstable both within and between conversational participants, and, what is worse, they are influenced by several contingent factors that cannot be easily controlled. Hence, we should rethink what we mean when we say that our best theory about negative existentials must have a capacity to accommodate the intuitive data. In this regard, we may follow the advice of Kertész and Rákosi (2012)Kertész, András, and Csilla Rákosi 2012 Data and Evidence in Linguistics: A Plausible Argumentation Model. Cambridge: Cambridge University Press. , and adopt a methodological framework in which intuitive judgements are considered to be only more or less reliable truth-candidates, not unquestionable facts. Kertész and Rákosi’s model interprets the uncertainty of data generally in terms of plausibility. On their view, plausibility comes in degrees and the plausibility values of data can be traced back to the sources from which they originate. Certain data may be highly plausible according to one source and less plausible according to others. Data that are gathered from completely acceptable sources (i.e. data with maximal plausiblility values) may substantiate true theoretical judgements. It is claimed, on this basis, that the primary function of intuitive and other kinds of data is to introduce plausibility values into theories. Given that all theories of negative existentials refer to intuitive judgements,1818.Clapp claims, for example, that “negative existentials are often judged by interpreters to be both felicitous and true” (Clapp 2009Clapp, Lenny 2009 “The Problem of Negative Existentials does not Exist: A Case for Dynamic Semantics.” Journal of Pragmatics 41: 1422–1434. , 1423). Similarly, Spewak argues that the sentence ‘Sherlock holmes does not exist’ “is intuitively true” (Spewak 2016Spewak, David C. 2016 “A Modulation Account of Negative Existentials.” Philosophia 44: 227–245. , 230). it would be helpful if we could determine the plausibility values these judgements introduce into our theories. It would also be interesting to see how and why these plausibility values clash with the plausibility values of our background theoretical assumptions. But this investigation must be left for future work.1919.I thank the editor and two anonymous reviewers for their constructive comments, which improved the quality of the manuscript substantially.