Some natural language expressions seem to express the same meaning in all ordinary contexts.1 For example, the predicate “is a mammal” is ordinarily associated with a meaning that yields the collection of all warm-blooded animals that have mammary glands. Yet many expressions seem to express different meanings relative to different contexts. Each of the following includes such an expression:

For instance, “bank” is associated with one meaning that yields the collection of financial institutions and another meaning that yields the collection of areas of ground bordering rivers and other bodies of water. To know whether (1a) is true relative to a given circumstance, we need to know whether the meaning of the occurrence requires Yemi to have gone to a building or a riverbank.

To distinguish the forms of meaning variation exhibited by (1a)–(3b), some minimal assumptions about the nature of meanings are required. A standard approach replaces talk of “meanings” with a distinction between characters, contents and extensions (see Kaplan 1989). A character is a “standing meaning” associated with an expression on the basis of linguistic conventions, while a content is what is expressed by a particular occurrence of the expression relative to a context of use. Then, an extension is the non-linguistic item—such as a collection of individuals in the case of a one-place predicate and a truth value in the case of a sentence—that results from evaluating a content at a circumstance. A circumstance of evaluation is a specification of at least a world, possibly in addition to other features (see Kaplan 1989: 502). More formally, characters are treated as functions from contexts to contents, and contents are treated as functions from circumstances of evaluation to extensions. This approach makes it possible to distinguish several ways in which sentences may express different meanings.

First, (1a)–(2b) include lexically ambiguous words, which are associated with multiple characters.2 “Bank” is associated with one character that yields a content that maps each circumstance to the collection of financial institutions, and a second character that yields a content that maps each circumstance to the collection of areas of ground bordering bodies of water. Similarly, one character associated with “dog” yields a content that maps each circumstance to the collection of members of canis familiaris, and another yields a content that maps each circumstance to the collection of male members of canis familiaris.

There is a further subdivision between types of ambiguous expressions: (1a)–(1b) include homonymous expressions that are associated with multiple unrelated characters, whereas (2a)–(2b) include polysemous expressions with multiple related characters. Determining whether an expression is homonymous or polysemous can be difficult. Etymological evidence and speakers’ linguistic intuitions are generally used to support the existence of some relation, although these methods can be unreliable (see Lyons 1977: 550–552). A further complication is that some have taken relatedness of characters to come in degrees, predicting that expressions will fall on a spectrum that moves gradually from regular polysemy to homonymy (Apresjan 1974; Vicente and Falkum 2017).

Next, (3a)–(3b) include context-sensitive words. A character counts as context sensitive if and only if it maps different contexts to different contents; and an expression is context sensitive if it is associated with at least one context-sensitive character.3 Hence the word “I” counts as context sensitive because it is associated with a character that maps each context to a content that yields the speaker of that context for any circumstance. Similarly, “is tall” is typically assigned a character that yields different contents depending on a contextually determined standard: the content expressed at one context might map any circumstance to the collection of individuals that are taller than the average woman, and the content expressed at another context might map any circumstance to the individuals that are taller than the average professional basketball player.

An ambiguous word need not be context sensitive: for although the discourse context helps hearers to decide which character is relevant for the interpretation of a particular occurrence, each character might map every context to the same content. Nevertheless, a word may be both ambiguous and context sensitive. For example, “light” is homonymous (it is associated with one character pertaining to colours and another pertaining to weights) and context sensitive (the first character yields different contents depending on contextually determined standards of what counts as a light colour, as does the second character with respect to what counts as a light weight).

Linguistic alethic pluralism states that the word “true”—when it applies to expressions that denote truth-bearers—expresses different contents in different contexts; that is, it is either homonymous, polysemous or context sensitive.4 A number of theorists have advanced this thesis. Kölbel (2008; 2013) aims to show that “true” is lexically ambiguous or context sensitive, although he favours the former view.5 The view that “true” is ambiguous is also briefly advocated by Tarski (1944), and defended more extensively by Yu (2016; 2021) and Égré (2021). An analysis of “true” as context sensitive has been endorsed by Burge (1979; 1982), Simmons (1993; 2018), Yu (2016; 2021), Henderson (2019; 2021), and Égré (2021).6

It is clear that linguistic alethic pluralism has become increasingly popular. But why would anyone endorse it? Consider sentences like the following:

Suppose you think that these sentences (or truth-bearers associated with them) are true. Are they true in the same sense? Many theorists think that they are not. Some of these theorists have detected a contrast in the sense in which truth-bearers associated with different domains of discourse or subject matters can be true. This contrast emerges for the “factual” truth-bearer presented by (4a) and the “non-factual” one presented by (4b).7 This perceived contrast has led some theorists to endorse metaphysical alethic pluralism: the thesis that there are multiple truth properties or concepts, which are associated with different domains of discourse (see Cotnoir 2013a; b; Lynch 2001; 2006; 2009; 2013; Wright 1992; 2001; 2013). Other theorists have detected a contrast between truth-bearers associated with non-paradox-generating sentences like (4a) and (4b), and paradox-generating sentences like (4c). This perceived contrast has resulted in a number of theories where occurrences of paradox-generating sentences can be true in some special sense (see Glanzberg 2001; 2004; 2006; Parsons 1974; Schlenker 2010).

Linguistic alethic pluralists tend to be motivated by their inclination towards one of these positions. For it might be natural to think that the different types of truth-bearers end up in distinct extensions for “true”, and that applying “true” to expressions associated with the different types of truth-bearers causes it to express contents that yield these distinct extensions. Linguistic alethic pluralists may thus be divided into two categories, depending on the types of truth-bearers that they take to provide evidence of the variability of “true”:8

Factual alethic pluralism has been advocated by Kölbel (2008; 2013), while advocates of paradoxical alethic pluralism include Burge (1979; 1982), Simmons (1993; 2018), Beall (2013), Yu (2016; 2021), Henderson (2019; 2021), and Égré (2021). Some linguistic alethic pluralists provide further details about the way in which the contents of “true” differ; for example, it might be claimed that factual truth-bearers are true in a “correspondence” sense, whereas truth-bearers that are non-factual or associated with Liar-style sentences are true in a “deflationary” sense (see Kölbel 2008; Yu 2016).

Advocates of both forms of linguistic alethic pluralism may make different predictions about whether two occurrences of “true” will express different contents. For instance, factual alethic pluralists predict that a factual truth-bearer associated with the sentence “Dogs are mammals” is in one possible extension for “true” while a non-factual truth-bearer associated with “Corndogs are tasty” can only be in another extension for “true.” Paradoxical pluralists may allow truth-bearers associated with both sentences to be in the same extension for “true.” Hence advocates of the two views disagree about whether “true” is likely to express non-identical contents that yield different extensions when it occurs in “It’s true that dogs are mammals” and in “It’s true that corndogs are tasty.”

There are also broader differences in empirical predictions. Factual alethic pluralists predict that evidence of linguistic alethic pluralism should emerge from the ordinary usage of “true.” After all, naive speakers of English regularly issue claims with factual and non-factual subject matters. Paradoxical alethic pluralists need not commit themselves to this prediction, since ordinary speakers do not routinely issue sentences that produce semantic paradoxes. This difference will affect how factual alethic pluralists and paradoxical alethic pluralists handle evidence unfavourable to their views (see §3.4).

In sum, versions of linguistic alethic pluralism may be distinguished according to the mechanism that causes “true” to express different contents in different contexts (homonymy, polysemy or context sensitivity). They may also be distinguished according to the type of truth-bearer that can end up in one extension of “true” but not in another extension of “true” (non-factual or paradoxical truth-bearers). Empirical predictions diverge based on both the mechanism and the type of truth-bearer emphasised. This will affect the testing of linguistic alethic pluralism, as discussed in §3. In the remainder of the current section, the implications of endorsing any version of linguistic alethic pluralism will be considered.

Linguistic alethic pluralism has been relatively neglected in the literature; for instance, Pedersen (2006) describes it as “uninteresting” compared with metaphysical alethic pluralism. Whether or not linguistic alethic pluralism is inherently interesting, it holds implications for the nature of truth.

First, there is a connection with deflationism. Deflationists hold that the word “true” serves a purely expressive or logical function—disquotation or denominalisation, and forming generalisations—that is captured by all non-paradoxical instances of the schemas (for a sentence s or proposition p)9:

Inflationists hold that “true” expresses a more substantive property that instances of the schemas do not fully capture, such as correspondence with reality, coherence with relevant beliefs, empirical verifiability, and so on. If linguistic alethic pluralism holds, then there would be good grounds for rejecting deflationism as a general thesis, at least as defined above.10 For it is difficult to see how different occurrences of “true” could express different contents if “true” only ever serves the same function captured by instances of the schemas. At the same time, there might be justification for a deflationary analysis of some uses of “true”: linguistic alethic pluralists often claim that “true” has one construal where it serves a function captured by the schemas, and another involving a more substantive property (e.g., Kölbel 2008; 2013; Yu 2016).

Second, there is a connection with metaphysical alethic pluralism. Recall that this thesis takes there to be multiple truth properties or concepts, which are generally linked to different domains of discourse. Metaphysical pluralism has been more widely discussed and endorsed than linguistic alethic pluralism, and is frequently deemed a more interesting thesis (Pedersen 2006). Metaphysical pluralists are often keen to distinguish their position from a linguistic thesis. For instance, Wright (1996: 924) emphasises that his view is “not that ‘true’ is ambiguous, that it means different things as applied within different regions of discourse […] the concept admits of a uniform characterization wherever it is applied.” Still, the fact that one version of linguistic pluralism is typically endorsed due to an inclination to accept metaphysical pluralism (see §1.2) emphasises the close connection between the views.

Several theorists have argued that at least the concept-based formulation of metaphysical pluralism entails linguistic alethic pluralism. Lynch (2001: 726) states that “a plurality of truth concepts entails that the word ‘true’ is ambiguous,” given the assumption that occurrences of “true” can express contents associated with these different concepts (also see Kölbel 2013: 287; Yu 2016: 22).11 In the other direction, if the content expressed by a predicate can be thought of as (or as determining) a property or concept, then at least some versions of the linguistic thesis that different occurrences of “true” express different contents entail the metaphysical thesis that there are multiple truth properties or concepts.12 Evidence for at least some versions of linguistic alethic pluralism would therefore provide evidence for at least some versions of metaphysical alethic pluralism, given further assumptions about the connections between contents and concepts or properties. Evidence against linguistic alethic pluralism would apply pressure to at least the concept-based formulation of metaphysical alethic pluralism.

As indicated above, there are areas of open debate surrounding the connections between different versions of linguistic alethic pluralism and metaphysical theses about the nature of truth. Still, linguistic alethic pluralism is more amenable to empirical assessment than metaphysical theses about truth are. Evaluating linguistic alethic pluralism is therefore a promising strategy for assessing deflationism about some or all occurrences of the word “true,” in addition to at least some versions of metaphysical alethic pluralism.

One tries to specify a context where a simple sentence that includes the target expression can be both affirmed and denied, without contradiction (Quine 1960: 128–9; Zwicky and Sadock 1975: 7–8).14 If one can specify such a context, then this is evidence that the target expression is ambiguous, and if one cannot, then this is evidence that the expression is unambiguous. Consider:

Contexts for which occurrences of (5a) and (5b) are non-contradictory include ones where the relevant item is a financial institution rather than a riverbank, or an abstract literary work that has been accepted for publication but not yet printed as physical book-copies. However, it is not possible to imagine a context where speakers are using “mammal” in the ordinary way but (5c) is non-contradictory.

Gillon (1990: 408) argues that passing the contradiction test is neither necessary nor sufficient for ambiguity: some sentences that contain ambiguous words might not permit affirmation and denial (e.g., non-declarative sentences, such as “Go to the bank”), and some sentences that lack ambiguous words but contain context-sensitive expressions might pass the test if contextual features are allowed to shift (e.g., “That’s a mammal, but that isn’t a mammal,” where the occurrences of “that” are understood to pick out distinct individuals). A simple way to circumvent these concerns is to apply the contradiction test exclusively to sentences of the form “DP is XP, but pro isn’t XP,” where “DP“ is a phrase that is understood to pick out an item (“that”, “the man,” “she,” etc.), “pro” is a pronoun (“it,” “she,” etc.) that is understood to pick out the same item as “DP,” and “XP” results from combining a potentially ambiguous word with the minimal number of other words required to produce a grammatical phrase (e.g., “a bank,” “the bank,” “light,” etc.). Moreover, it is important to specify a single context for which a non-contradictory occurrence results, where any contextual features that might affect the interpretation of context-sensitive expressions are held fixed.15

One forms a pair of sentences containing the target word that meets two conditions: first, plausible interpretations of the sentences require a distinct construal of the target word in each case (e.g., “ATM machines are found in banks,” “Water voles are found in banks”).16 Second, it must be possible to combine those sentences by means of a construction that “reduces” or “deletes” one of the constituents containing the target word. These constructions include anaphoric verb phrases (as in (6a)) and coordination (as in (6b) and (6c)). If the resulting sentence sounds zeugmatic (odd in a sense that results from inappropriately linking two expressions), then this is evidence that the target expression is ambiguous, and if it does not, then this indicates that the target expression is either unambiguous or ambiguous between closely related characters.

A sense of zeugma is produced by (6a) and by (6b) (originally from Cruse 1986: 64), but not by (6c).

The weakened inference provided by the absence of zeugma derives from problems that the test encounters when applied to polysemous expressions (see Geeraerts 1993; Lascarides et al. 1996; Viebahn 2018). Viebahn (2018: 754) produces examples where polysemous expressions appear to yield no zeugma, such as (7a); although as he points out, there are also often some sentences for each polysemous expression that do sound zeugmatic, such as (7b):

Viebahn concludes that “the closer the meanings of an expression are related, the less likely it is that forcing them together will lead to zeugmaticity” (Viebahn 2018: 754). It follows that producing a number of sentences in which an expression yields no zeugma cannot suffice to show that the expression is unambiguous: for the expression might be polysemous, and be associated with characters that are so closely related that zeugma rarely or never arises.

Another issue with the test concerns the extent to which assessors are able to distinguish between zeugma and other forms of oddness (see Lyons 1977: 407–408). For instance, (8) exhibits a form of pragmatic oddness that results from conjoining clauses with poor thematic relatedness, which makes it difficult to imagine a natural question under discussion:

This type of pragmatic oddness disappears when the sentence is assessed relative to a context with a suitable question under discussion. For instance, the oddness of (8) is eliminated when it is considered relative to a context where a speaker has asked any of the following questions:

In contrast, seeking suitable contexts for (6a)–(6b) does not mitigate their oddness.

To summarise, a single sentence that yields no zeugma does not show that the target expression is unambiguous (e.g., for “book” in (7a)). It does, however, indicate that the target expression is neither homonymous nor polysemous between distantly related characters. A sentence that yields zeugma does provide evidence that the target expression is ambiguous (e.g., for “book” in (7b)). Finally, the reliability of the evidence depends on establishing that the oddness of a target sentence is indeed zeugma.

Viebahn and Vetter (2016: 4–5) claim that certain typical relations often hold between the characters of polysemous words, which they describe as constitutive (one character yields extensions consisting of objects and another yields the matter of which the objects are constituted), causal (one yields a producer and another yields the product), instantiating (one yields an abstract type and another yields its concrete tokens), metaphorical extension (one character originated as a metaphorical or figurative use of another character) and pragmatic strengthening (one character originated as a pragmatic inference (see fn.5 above) associated with another character). To apply the test, one checks whether such relations hold between the proposed characters of the target word. If any of the relations hold, then this provides evidence that the word is polysemous.

An instantiating relation holds between the character of “book” that yields informational objects and the one that yields physical book-copies. A metaphorical extension relation holds between the character of “long” that is linked to a property of spatial distances and the character linked to a property of temporal “distances.” A pragmatic strengthening relation holds between the character of “since” used to express temporal succession and the one used to express causality: the former, original character often leads to a pragmatic inference of causal connection between events (e.g., “Since Yemi returned from the library, she has been reading”), and this inference was eventually encoded in a separate character that does not involve temporal succession (e.g., “Since Yemi is in the library, Yash assumes she is reading”). On the other hand, none of the relations seem to hold between the different characters associated with “bank”. Moreover, if one tried to assign a distinct character to “I” for each context in which a different speaker utters it, then one would find that none of the typical relations hold between the proposed characters.

Viebahn and Vetter note that it can be difficult to determine whether the typical relations hold, especially in the case of metaphorical extension and pragmatic strengthening. Judgements regarding the latter two often draw on etymological facts, which can be unclear. Moreover, further relations that are less straightforward to characterise can hold between the characters of polysemous expressions, such as ones based on a more general form of resemblance; hence the absence of the typical relations isolated by Viebahn and Vetter does not justify any conclusions about the polysemy of a word. Still, the test may deliver a clear verdict in cases where the presence of one of the typical relations—especially a relation of the first three types—is obvious (e.g., for “book”).

Viebahn and Vetter (2016: 8) describe a test that involves estimating the approximate number of candidate contents for an expression, where these are the different contents that the expression might express relative to different contexts of use. If the expression has relatively many candidate contents, then this provides evidence that it is context sensitive, and if it has relatively few, then this provides evidence that it is context insensitive (although it is potentially ambiguous if it has more than one candidate content). For example, there are as many candidate contents for “I” as there are speakers of English, and as many for “light” as there are standards for counting as a light colour or a light weight. There are comparatively few candidate contents for “book”: there is one content yielding the collection of abstract works, one content yielding the collection of physical book-copies, and potentially several more.

Concerns might arise about the imprecision of the instructions for carrying out this test. It is unclear how to reliably estimate candidate contents, and exactly how many candidate contents are required in order to provide evidence of context sensitivity. The results should therefore be interpreted with care, and employed in combination with the results of other tests.

Cappelen and Lepore (2003; 2004; 2005) develop a test that applies to unambiguous target expressions.17 One tries to identify some sentence s containing the target expression such that a speaker can truthfully utter a version of the following schema at some context: “There can be false utterances of ‘s’ even though s.”18 If there is some such s, then this provides evidence that the target expression is context sensitive, and if there is no such s, then this provides evidence that the target expression is context insensitive. For instance, a true occurrence of the schema is given in (10a) (relative to a context where the speaker is a philosopher), which provides evidence for the context sensitivity of “I”. Since it does not seem that a speaker could truthfully utter (10b) relative to any context, (10b) provides evidence for the context insensitivity of “mammals.”

One concern about this test is that assessors often disagree about the verdict, or are unable to reach a clear verdict. For instance, Cappelen and Lepore (2003: 43–45) argue that expressions like “is tall” are classified as context insensitive by their tests, but it is far from obvious that this is the case.19 Indeed, others have argued that expressions classified as context insensitive by Cappelen and Lepore count as context sensitive according to their inter-contextual disquotation test, including proper names, quantifier expressions, “is tall,” “ready,” and “red” (Bezuidenhout 2006: 8; Szabó 2006: 36; Leslie 2007: 141–142). These observations might raise doubts about whether the application of the test relies exclusively on pre-theoretic, linguistic intuitions available to any speaker of the language.

With five tests now in place, we are in a position to see how they have been used to evaluate linguistic alethic pluralism.

Both factual and paradoxical alethic pluralists have claimed that the contradiction test supports the ambiguity of “true,” although others have disputed this claim for factual alethic pluralism. First, Kölbel (2008; 2013) argues that a version of the contradiction test provides evidence that “true” is either lexically ambiguous or context sensitive.20 Kölbel (2008: 370) states that “[m]any users would accept, or make, utterances like the following two,” and that occurrences of such sentences “do not seem to be incompatible”:21

He takes these results to provide evidence that some occurrences of “true”—such as the one in (11a)—express a “deflationary” notion of truth, while others—such as the one in (11b)—express a “substantial” one. He additionally claims that the collection of truth-bearers that count as true in the substantial sense is a proper subclass of the collection that count as true in the deflationary sense, consisting of all and only the factual truth-bearers.

The reason that Kölbel’s version of the test could not provide evidence for lexical ambiguity in particular is that the context might shift between occurrences of (11a) and (11b).22 This would allow two occurrences of a context-sensitive target expression to express distinct contents. Yet §2.1 advised applying the test by seeking a single context for which a sentence of the form “DP is XP, but pro isn’t XP” is non-contradictory, where ‘pro’ is a pronoun understood to pick out an item fixed by “DP.” Applying the test in this manner reduces the likelihood of a context shift: assessors are instructed to envisage a single context, and any context shift would have to occur in the middle of a simple sentence. The best strategy for evaluating factual pluralism would be to apply the contradiction test as advised, along with other tests that identify ambiguity. Tests that identify context sensitivity can be separately applied later.

Boscolo and Pravato (2016: 48) apply the contradiction test in the correct manner in order to assess Kölbel’s view. They conclude that “[i]t does not seem that an ordinary English speaker can utter [(12)] without contradiction”:

On the face of it, a factual alethic pluralist who takes “true” to be lexically ambiguous predicts that (12) should pass the contradiction test as smoothly as sentences like “That’s a bank, but it’s not a bank”.

Yu (2016: 225–226)—a paradoxical alethic pluralist—claims that “true” passes the contradiction test when applied to λγ, which names the contextual Liar sentence that reads “λγ is not true in any context”:

Yu claims that “true” is polysemous between a “correspondence” and “disquotational” character (2016: 172), such that the collection of truth-bearers that count as true in the correspondence sense is a proper subclass of those that count as true in the disquotational sense.23 He thinks that, relative to a context in which the Liar reasoning is considered, (13) is non-contradictory: the contextual Liar sentence does not say anything true in the correspondence sense, but does say something true in the disquotational sense.

A difficulty with Yu’s argument is that sentences such as (13) are intuitively deemed contradictory when they occur as the conclusion of Liar-style arguments. Indeed, theorists go to great lengths to either block the derivation of the apparent contradiction, show that there is no genuine contradiction, or argue that the derivation of a contradiction is unproblematic. Were it to be obvious that a sentence like (13) could be understood as non-contradictory, it would be unclear why Liar-style paradoxes would be considered paradoxical in the first place.

In sum, the contradiction test provides no clear evidence that “true” is lexically ambiguous. Indeed, those who are unable to imagine any context in which (12) and (13) can be uttered without contradiction will judge the test to provide some evidence that “true” is unambiguous.

While the only existing application of the zeugma test to factual pluralism takes the results to oppose the ambiguity of “true”, an existing paradoxical pluralist reaches the opposite conclusion. First, Boscolo and Pravato (2016: 47) claim that the zeugma test indicates that “true” is not homonymous between a construal that applies to factual truth-bearers and one that applies to non-factual ones concerning what is funny, due to the absence of zeugma elicited by (14):

However, they acknowledge that the zeugma test faces difficulties in identifying polysemy for sentences where characters are very closely related, or one character is more general than another (see fn.16 above). For instance, Kölbel’s analysis of “true” predicts that (14) may express content that holds at the actual world provided “true” is disambiguated as the more general, deflationary construal.

Next, Yu (2016: 227–228) claims that (15) is clearly zeugmatic, suggesting that “true” is ambiguous between a construal that applies to non-paradoxical sentences and one that applies to the contextual Liar sentence λγ:

Even if Yu is correct that (15) seems odd when presented out of context, it is unclear whether it is zeugmatic. An alternative view would be that it displays a form of pragmatic oddness attributable to the lack of obvious thematic relatedness between the two conjuncts and the resulting difficulty in accommodating a natural question under discussion (see §2.2).

There is some evidence in favour of this alternative view. First, providing a natural question under discussion—such as “What are two sentences that are true?”—appears to reduce the oddness, despite the fact that zeugma typically persists when a suitable context is specified. Second, consider a variant involving the sentence “δ is not true in any context” where δ names a non-paradoxical sentence (say, “I am not here now”). Even when presented out of context, (16) does not sound particularly odd:

A natural explanation for the reduced oddness of (16), compared with (15), would be the increased thematic similarity (i.e., the topic of sentences that are arguably not true as used in any context) between the two conjuncts. Nevertheless, a paradoxical alethic pluralist who accepts that (15) is pragmatically odd rather than zeugmatic might reiterate that polysemous words with closely related characters can fail to produce zeugma in some settings.

In sum, if zeugma is absent for some target sentences—such as (14) and (16)—then this would provide evidence against the view that “true” is either homonymous or polysemous between distantly related characters. This absence of zeugma remains compatible with the view that “true” is polysemous between closely related characters. The test only provides evidence that “true” is polysemous if zeugma is elicited by some target sentences of the type predicted by factual alethic pluralists and by paradoxical alethic pluralists. No candidates for the former type of sentence have been found, although Yu presents (15) as a candidate for the latter. Yet a plausible alternative explanation attributes any oddness elicited by (15) to pragmatic factors rather than zeugma. Hence the polysemy of “true” cannot be ruled out by the zeugma test, but receives no clear support from it either.

The related characters test has not been used to evaluate factual alethic pluralism, although an existing paradoxical pluralist takes it to support the polysemy of “true.” Given that the test is insensitive to the sorts of settings in which “true” expresses different contents, it may be used to evaluate factual alethic pluralism and paradoxical alethic pluralism in tandem. At the same time, the fact that the test requires a comparison of hypothesised characters means that its outcome will depend on each linguistic alethic pluralist’s specific views about the way in which the characters differ.

Yu (2016: 233) claims that a metaphorical extension relation holds between the two hypothesised characters, because “an initial meaning corresponding to correspondence truth gives rise, by a combination of widening and metaphorical extension, to an extended meaning corresponding to disquotational truth.” Yu does not elaborate on this idea, and it is difficult to see how a truth-bearer that is true in a correspondence or other substantive sense might be deemed metaphorically true in a deflationary sense (or vice versa).

Perhaps a linguistic alethic pluralist could try to identify another typical relation that holds between a deflationary character and a character that yields a more substantive truth property, or between some other hypothesised pair of characters. It seems unlikely that any of the first three types of relations—constitutive, causal or instantiating—could hold between the different characters: any character associated with “true” will yield extensions consisting of truth-bearers, and truth-bearers are presumably not the kind of physical objects that stand in these type of relations. A pragmatic strengthening relation would mean that one character often leads to a pragmatic inference that the relevant truth-bearer is true in the other sense, and this inference was eventually encoded in a separate character. But uses of “true” do not obviously contribute to these sorts of pragmatic inferences (see fn.5), and no theorist has argued otherwise.

Hence the related characters test provides no evidence in support of the polysemy of “true.” Part of the difficulty is that the relations that could most plausibly hold between its characters involve metaphorical extension and pragmatic strengthening, which are particularly challenging to identify. An inability to detect such relations might be attributed to this difficulty, or to the absence of any typical relations holding for “true.” Of course, a linguistic alethic pluralist might try to argue that some alternative relation holds between multiple characters for “true”; for instance, perhaps a more general form of resemblance holds between truth-bearers that are true in a deflationary sense and ones that are true in some substantive sense.24 Yet the absence of any typical relations would also be compatible with the non-polysemy of “true.”

The tests discussed so far do not support the view that “true” is ambiguous. At this point, linguistic alethic pluralists who posit ambiguity might try to seek and apply further tests, additional to the ones discussed in §2.1–2.3.25 The possibility cannot be excluded that further tests could be found that provide support for ambiguity. Yet even if this turned out to be the case, the linguistic alethic pluralist would need to explain why the classic tests used by natural language theorists do not provide this support. On one hand, a lack of evidence for ambiguity does not in itself constitute evidence for non-ambiguity. On the other hand, one might think that in the absence of positive evidence for an expression’s being ambiguous, the default assumption should be that it is unambiguous. Yet I believe that a stronger argument can be mounted in favour of the view that “true” is unambiguous.

The argument begins by observing a disanalogy between the interpretation of “true” and paradigm ambiguous expressions. Kölbel (2008: 373) himself mentions this disanalogy, before attempting to mitigate it: if an expression is ambiguous then “interpretation usually requires a decision in favour of one of the readings. Not so in the case of ‘true,’ it seems.”26 His response is that, in fact, disambiguation is often unnecessary for polysemous expressions like “dog,” such as when “Yash is a dog” is interpreted at a context where it is known that Yash is a male animal. He claims that treating “true” as polysemous rather than homonymous then “explains why the ambiguity of “true” is not easily recognized, and it also explains neutral cases, i.e., cases where we feel no need to disambiguate” (2008: 375). Initially, it seems that the potential to avoid a decision about how to understand “true” might be explained by—or even motivate—the hypothesis that “true” is polysemous between closely related characters.

According to the theory of meaning set out in §1.1, a sentence can only be interpreted if every expression is assigned some unique character and content. Given the standard view that polysemous expressions are associated with multiple characters, it follows that “both speaker and hearer must select a reading (the same reading) if the sentence is to play its part in a normal conversational exchange” (Cruse 1986: 51).27 Hence the standard view of meaning and polysemy rules out cases where polysemous words are used without disambiguation. If Kölbel is right that there are “neutral” cases involving “true”, then “true” could not count as polysemous.

An alternative view has recently been proposed, where at least some polysemous words are associated with a single, underspecified character that covers all of the more specific construals.28 The requirement to assign a character could then be satisfied by assigning an underspecified one. Yet advocates of this position grant that further specification of the character will sometimes be necessary for interpretation, such as when plausibility emerges with one specific character but not another. Kölbel (2008: 374–376) states that users will recognise the polysemy of words like “dog” and “true” when they are confronted with appropriate examples. If factual and paradoxical alethic pluralists are right, then plausible interpretations of (18a)–(19b) require distinct disambiguations of “true,” just as (17a)–(17b) do for “dogs”:

An assessor should recognise the need to disambiguate the target word, and which disambiguation she has selected, for at least one sentence in each pair.29 While this appears to happen for (17a)–(17b), it is far from clear that it happens for (18a)–(19b).

In sum, if the standard view of polysemy is correct, then the existence of any cases where users do not decide between multiple characters for “true”—as in the “neutral” cases raised by Kölbel—would indicate that “true” is non-polysemous. If the alternative view of polysemy is correct, then assessors’ failure to recognise a disambiguation process for “true” in cases where a specific character is required for plausibility—as in (18a)–(19b)—would indicate that “true” is non-polysemous.

We might ask whether parallel considerations raise doubts that “true” is context sensitive. A context-sensitive word has a character that maps different contexts to different contents because some parameter provided by the character takes different values in different contexts. But it is widely thought that the values of at least some of these parameters are fixed by contextual factors that go beyond users’ intentions (see Glanzberg 2007; Viebahn 2020). The view that “true” is associated with this sort of parameter would be compatible with claiming that users need not make a decision about the value, or even be aware of which value the context has assigned. On the other hand, it is generally accepted that it is speakers and hearers—not contexts—that select characters. So the preceding argument can only convincingly target the view that “true” is ambiguous. In light of this argument, and the absence of evidence that “true” is ambiguous, the grounds for suspecting that “true” is unambiguous are stronger than merely the claim that this should be the default hypothesis for any expression. The remainder of §3 considers whether there is evidence that “true” is context sensitive.

At this point, it is worth drawing attention to a hypothesis that paradoxical alethic pluralists are likely to deploy when theorists attempt to apply the current tests. The hypothesis is that the tests cannot be used to elicit reliable linguistic intuitions from speakers when applied to Liar-style sentences and paradox-generating contexts. After all, Liar-style sentences are generally not used in ordinary discourse.30 Yet the tests are designed to probe speakers’ linguistic intuitions about the contents expressed by target words, based on their ordinary usage. If the outcome of a test does not support the prediction of a paradoxical alethic pluralist, then this hypothesis might be deployed.

Factual alethic pluralists cannot rely on a similar hypothesis: speakers regularly make claims pertaining to factual and non-factual truth-bearers in ordinary discourse. Also, paradoxical alethic pluralists cannot use this hypothesis in order to maintain that “true” is ambiguous: if characters are selected by language users, then theorists who utter or interpret Liar-style sentences should recognise a disambiguation process for “true”, whether or not ordinary users have reliable intuitions about occurrences of those sentences. The described hypothesis is useful only to paradoxical alethic pluralists when they seek to defend the view that “true” is context sensitive.

There is a cost to developing a version of paradoxical pluralism that relies upon this hypothesis; but a discussion of this cost is reserved for §3.7. Now, we return to the tests.

While the candidate contents test has not been used to evaluate factual alethic pluralism, an existing paradoxical alethic pluralist takes it to support the polysemy and context insensitivity of “true”. Yu gives the following argument:31

Yu’s idea seems to be that the revenge paradoxes explicitly formulated at any context involving the actual world never surpass some comparatively low “level”, hence the number of characters (and thus candidate contents) that are actually associated with “true” are similarly limited.

The trouble is that the proposed analysis of the revenge paradoxes provides the only grounds for thinking that “true” has multiple but few candidate contents. A related version of paradoxical pluralism might hold that all of the “concepts in the hierarchy” are “actual concepts of truth”, whereupon the number of candidate contents would turn out to be sufficiently high to provide evidence for context sensitivity. Yet applications of the tests are supposed to be independent of post-theoretic considerations. From a pre-theoretic perspective, however, it is far from obvious that there is more than one candidate content for “true”. This remains the case whether we focus on uses of “true” in ordinary contexts involving factual and non-factual discourse, or additionally consider paradox-generating contexts.

At this point, paradoxical pluralists might deploy the hypothesis described in §3.4. That is, they might explain the absence of a pre-theoretic recognition of multiple candidate contents by claiming that the test cannot be used to elicit reliable linguistic intuitions when applied to paradox-generating contexts. Still, it would then follow that the candidate contents test could provide no useful evidence regarding the status of “true.”

The inter-contextual disquotation test has been used in the literature to evaluate factual pluralism and paradoxical pluralism; in the former case, it was taken to indicate that “true” is context insensitive, and in the latter case, it was taken to indicate the opposite. Boscolo and Pravato (2016: 52–53) present a version where they attempt to describe a scenario in which an individual truthfully asserts that there can be false utterances of “’Chaplin is funny’ is true” despite the fact that “Chaplin is funny” is true:

In their judgement, the scenario is unconvincing, so their application of the test does not support the context sensitivity of “true”. Unless some sentence that yields truthful utterances of the schema can be found, the test provides evidence that “true” is context insensitive.

Simmons (2018: 30) claims that the inter-contextual disquotation test indicates that “true” is context sensitive in the manner predicted by paradoxical alethic pluralists, although he does not explicitly apply it.32 If this claim is correct, then a theorist could truthfully utter (20) in the course of paradoxical reasoning (where L names the Liar sentence that reads “L is not true”).

As far as I can tell, it is difficult to have any pre-theoretic sense of whether or not a theorist could truthfully utter (20).

Hence the inter-contextual disquotation test does not provide support for the context sensitivity of “true”. Paradoxical alethic pluralists might again claim that the test cannot be used to elicit reliable linguistic intuitions when applied to Liar-style sentences. This would entail that the inter-contextual disquotation test can provide no evidence regarding whether “true” is context sensitive in the ways that paradoxical alethic pluralists have maintained that it is. Furthermore, paradoxical and factual alethic pluralists alike might point out that doubts have been raised about the reliability of the test (see §2.5).

The last two tests do not support the predictions of linguistic alethic pluralists. The candidate contents test provides no evidence for either context sensitivity or ambiguity, and the inter-contextual disquotation test provides no evidence for context sensitivity. Either “true” does not express different contents in different contexts, or the standard tests employed by natural language theorists are unable to detect the fact that “true” has this feature.

As explained in §3.4, paradoxical alethic pluralists—but not factual alethic pluralists—could provide a plausible explanation of why the tests fail when applied to “true”. They hold that the different contents expressed by “true” emerge in contexts involving Liar-style sentences. Since such sentences are not used in ordinary discourse, the tests cannot be used to elicit reliable linguistic intuitions. Perhaps paradoxical alethic pluralism may be upheld, then, whatever the verdicts of the tests. Yet this strategy for upholding paradoxical alethic pluralism has a cost: one must endorse a thesis about natural language that cannot be evaluated via methods that are typically used to test theses about natural language.