Scope

Suppose that we had a theory only of the meaning of “legal obligation” in sentences like (1). An obvious concern would be that the scope of this theory is too narrow. Semantic theories should aspire to account for a broader range of related linguistic phenomena.

This consideration is particularly germane when we recall that the language of law includes plenty of normative terms besides “obligation.” Recall the epigraph from Gardner: “The place to begin, nobody doubts, is with the language that law-applying officials use. In explaining the law, they cannot but use the language of obligations, rights, permissions, powers, liabilities and so on.” Law-applying officials cannot help but use all of this language because the content of law includes legal obligations, rights, permissions, powers, liabilities, and so on.Footnote ⁸ Positivists need to explain all of this. And if there is a problem with positivists deriving an “obligation” from an “is,” there is a similar problem for positivists deriving a “right” from an “is.”Footnote ⁹

Unfortunately, many philosophers of law offer semantic theories of “legal obligation” without explaining how the theory is meant to generalize beyond this narrow scope. I will focus on this project on generalizing such theories to encompass sentences like:

1. (5) People are legally permitted to commit adultery.

This is for three reasons. First, we have a good starting point for how to understand (3) and (4). The deontic modals “must” and “can” express “obligated” and “permitted,” respectively;Footnote ¹⁰ (5) could be restated as, e.g., “Legally speaking, people can commit adultery.” So, in Section 2, we will use Kratzer’s semantics for “must” and “can” to clarify the meaning of “obligated” and “permitted,” respectively, in sentences like (1) and (5). Second, there is plausibly a close semantic relation between “legal obligation” and “legal permission,” akin to the semantic relation between “must” and “can.” If “A must φ” is true, then “A can ~φ” is false, and vice versa. To illustrate: if it is true that “You are legally obligated to pay taxes,” then it is false that “You are legally permitted to not pay taxes”; and if it is true that “People are legally permitted to commit adultery,” then it is false that “People are legally obligated not to commit adultery.”Footnote ¹¹ Third, many think we can explain “rights” and “powers” and so on in terms of “obligation” and “permission.” To have a power, for example, is to be permitted to change someone’s obligations.

Univocality

Is “obligation” univocal in moral and legal sentences? Recall (1) and (2): “You have a legal obligation to pay taxes” and “You have a moral obligation to pay taxes.” The question here amounts to asking: Does “obligation” have the same meaning in these sentences? I think so. This is supported by applying the standard conjunction reduction test for univocality. I’ll introduce it with a different example. Consider the sentences: “Pale colors are light”; “Feathers are light”; “Pale colors and feathers are light.” The first two sentences contain a purportedly ambiguous term (“light”); the third sentence conjoins them using that term only once in a context that encourages both meanings. The third sentence should elicit a linguistic intuition of oddness, as “light” is being used with two different meanings simultaneously. That kind of oddness has a name: zeugma. Because the third sentence is zeugmatic, we should think that “light” is ambiguous.

We can now easily apply the same test to “obligation” in sentences like (1) and (2):

1. (6) You have a moral and legal obligation to pay taxes.

We simply conjoined and reduced the two uses of “obligation.” But there’s no zeugma. I elsewhere argued that this test and others like it provide strong evidence for the univocality of “obligation.”Footnote ¹² This provides a constraint on a theory of the semantics of “legal obligation.” If such a theory says “obligation” is ambiguous in sentences like (1) and (2), then it predicts that “obligation” is zeugmatic in (6), which is false. We can also construct similar examples using other terms to modify “obligation” and cognate terms (e.g., “There is both a grammatical and moral requirement to φ”) that are not zeugmatic. So this constraint on a semantic theory should be construed quite broadly.

We can also provide a similar test for terms like “permission.” Consider an example:

1. (7) People are legally, but not morally, permitted to commit adultery.

If “permitted” were ambiguous in relation to “legally” and “morally,” then (7) should be zeugmatic. But it isn’t. In both (6) and (7), we vary legal and moral uses of the same normative term to generate evidence that those normative terms are not ambiguous in these contexts. But there is also a parallel issue worth considering: Similar sentences with legal uses of normative or non-normative terms. Consider the two conjuncts of (8):

1. (8) Positivism is a theory of legal institutions and legal obligations.

Is “legal” univocal in each of the two conjuncts? Let’s check!

1. (9) Positivism is a theory of legal institutions and obligations.

If “legal” is ambiguous when it modifies “obligation” and “institution,” then its single use in (9) should be zeugmatic. But it isn’t. This provides another consideration that bears on theories of the semantics of “legal obligation.” If such theories say “legal” is ambiguous in its two uses in (8), they falsely predict that “legal” is zeugmatic in (9).Footnote ¹³

How costly is it to deny that these terms are univocal? To some degree, it depends on what is affirmed instead. One view would be that “obligation” et al. are homonyms. Another view would be that “obligation” et al. are polysemes, which have distinct but related meanings. “Bank” is homonymous rather than polysemous because its distinct meanings (riverside; financial institution) are unrelated. Some may contend that the tests above provide evidence that settles that “obligation” et al. are not homonymous, but does not settle whether these terms are univocal rather than polysemous. As such, the reader may wonder at various points below whether a view about the semantics of (1) that suggests that “obligation” et al. are not univocal is in such dire straits.

It’s a fair question, but I won’t consider specific versions of it in the sections below. Instead, I’ll make three general points about this response now. The first is that if one makes this response on behalf of any view discussed below, one owes an account of why the relevant distinct meanings of “obligation” (etc.) are sufficiently related for the term to be polysemous. But typically, those who have denied that “obligation” (etc.) is univocal have held it to have multiple unrelated meanings.Footnote ¹⁴ Second, other considerations also count in favor of taking “obligation” to be univocal. We need a general explanation of the close semantic relation between “must” and “can,” and between “obligation” and “permission,” that was described above. Moreover, we use terms like “must” and “obligation” not just to “refer to duties,” but to refer “to duties of different kinds,” duties of “different people,” and duties “at different times” (Kratzer 1977: 339). If such terms are ambiguous when used to refer to different kinds of duties, etc., then we must somehow “keep billions of different ‘must’s” (1977: 441). We are finite beings. It is implausible that we store so many meanings for “must.”Footnote ¹⁵ Such concerns about our finite cognitive resources for storing and processing language are widely understood to apply to views that posit homonymy and polysemy.Footnote ¹⁶ The third is that those who have most forcefully challenged the orthodox view that modals are univocal have, given the concerns above, affirmed that deontic modals are univocal.Footnote ¹⁷

Embedding

Philosophers of language have long emphasized that a semantic theory of the meaning of “obligation” needs to account for simple sentences, like (1), and semantically complex sentences. By complex sentences, I mean sentences where a bare claim like (1) is used with ordinary logic operators (and, or, not, if … then). I’ve already mentioned several examples involving conjunctions above, and cases of disjunction are similar.Footnote ¹⁸ So I will mainly focus on cases in which claims like (1) are embedded.

One simple way to embed a claim like (1) is through negation. Here’s an example:

1. (10) There is no legal obligation to be a Good Samaritan.

A theory of the meaning of sentences like (1) should also explain what (1) means when it is negated, so it must explain the meaning of sentences like (10). Moreover, plenty of legal discourse includes the utterance of sentences like (10). So philosophers of law explaining the meaning of “legal obligation” need to explain the meaning of (10).

Adam Perry (Reference Perry2023: 718) also provides a more complex example:

1. (11) All and only persons over 18 have a legal right to vote.

1. (12) A has a legal right to vote.

1. (13) So, A is over 18.

Two things about the example make it a good test case for a theory of the meaning of sentences like (1). First, it is a “mixed argument”: “Some steps are legal claims, some steps are not.” Second, the “argument is valid” (Perry Reference Perry2023: 718). So, a theory of the meaning of (1) better be able to explain these features of the argument in (11)–(13).

Linguistic Data

Many subfields of linguistics, including semantics, rely on intuitions in testing theories. Much can be said about how to understand intuitions as linguistic data,Footnote ¹⁹ and how to elicit probative linguistic intuitions, but the following remarks are fairly uncontentious.

In “On the Methodology of Semantic Fieldwork,” Lisa Matthewson noted that in aiming to “establish facts about the meaning of utterances, and parts of utterances,” semantic fieldwork intuitions are helpful data. But the relevant “semantic facts are often subtle” and “are almost never accessible by direct native-speaker intuitions” (Reference Matthewson2004: 370). As such, “a researcher interested in the semantic contribution of the English definite article the … cannot ask a native speaker, ‘What does the mean?’” (Reference Matthewson2004: 371). Instead, the researcher must “construct a range of example sentences” to generate “primary data involving judgments about the felicity and truth of whole utterances,” from which “the semanticist reasons backward to establish the precise contribution of the.”

We want to know the semantic contribution of “obligation.” But we cannot simply ask people the meaning of “obligation.” Instead, our semantic fieldwork requires us to “construct a range of example sentences” that use this target word, and elicit intuitions about the acceptability (felicity and truth) of those sentences. The example sentences should be simple, and contextual information should be supplied where relevant. If we construct sentences where rival semantic theories issue different predictions, this will elicit probative data. A central desideratum for a semantic theory is that it better coheres with such “primary data.” A theory that fails to do so is in dire straits.

Linguistic data can also concern more specific notions than acceptability. We discussed one above: zeugma. Another is intuitions about semantically anomalies. “Colorless green ideas sleep furiously” is a famous example. It is grammatical, but it is intuitively unacceptable because the only eligible meanings would need to be utterly bizarre. Semantically anomalous sentences are often marked with a hashtag. Now consider:

1. (14) # You have a legal moral obligation to pay taxes.

This sentence is, I venture, semantically anomalous. There are legal obligations and moral obligations, but there are no legal moral obligations. A theory of the meaning of “legal obligation” should explain intuitive data about sentences like (14).

Psychological Plausibility

One final consideration concerns the motivational plausibility of a semantic theory. This consideration stems from a central point made by Hart in The Concept of Law.Footnote ²⁰ Hart notes that many theories of law—associated with Oliver Wendell Holmes and John Austin, among others—suggest that people are motivated to comply with the law solely to avoid sanctions. Hart was concerned that this defines “out of existence” what Hart called the “internal point of view”: The point of view of one who accepts the law and uses it as a guide to conduct (Reference Hart2012: 89–91). The “external point of view,” by contrast, is a catch-all term: it includes “a practical attitude toward the law that does not involve acceptance,” such as those who follow the law solely to avoid sanctions, as well as the non-practical attitudes of someone who “simply wishes to describe how members of a group regard and respond to a set of rules and, perhaps, who wishes to make predictions as well” (Shapiro Reference Shapiro2006: 1159–1160). A semantics of “legal obligation” should avoid implausible commitments about the attitudes of those who take the internal or external point of view. I’ll refer to this as the psychological plausibility of such theories.

This consideration is relevant because it is often noted that the internal and external points of view are expressed using the same language. Since Hart, it has been common for the same language—e.g., (1)—to be called an “internal” or “external” legal statement depending on speaker’s corresponding point of view. For example:

Notice that the exact same string of words can be an internal or external legal statement. I can say (1) as an “assertion of law” or as a normative statement made from the point of view of an adherent of the legal system. I can say (1) as a descriptive statement of a legal observer, as an anarchist who hates the law, or as an advisor telling you how to avoid being audited. If a semantics for (1) suggests that one who says (1) must be an adherent of the legal system, it has implausible commitments about speakers’ attitudes; and ditto if it suggests that one who says (1) must, say, only care about the law’s sanctions.

I’m going to focus on these five factors, but they do not exhaust the considerations that bear on how we should evaluate semantic theories. As such, there are significant limitations on the current project. Among these is the fact that it solely concerns English. Clearly a theory of the meaning of “legal obligation” is less plausible if it is not applicable to (at very least) closely related Indo-European languages.Footnote ²¹ But I lack the expertise to do that analysis with any rigor. And cross-linguistic comparisons are just one of the important considerations that I will not discuss at all below.Footnote ²²

My hope, then, is to offer a survey of how philosophers of law approach the meaning of “legal obligation,” and to show how the debate could become more theoretically sophisticated by drawing on tools from philosophy of language and linguistics. The hope is not to offer anything close to the final word on any of the theories we discuss. It is, instead, to catalyze a richer debate between these views and other unexplored alternatives about this critical issue in philosophy of law, and to explore its (sometimes substantial, and sometimes overstated) ramifications for the plausibility of specific jurisprudential theories, including for positivism in general.

Scope

Kratzer offers a general account of the semantics of modals. The difference between “A must φ” and “A can φ” tracks the difference between universal and existential quantification. With “A can φ,” there is a contextually salient modal base that generates a set of possible worlds, and a contextually salient ordering source that ranks those worlds. “A can φ” is true when A φs in some of the best-ranked worlds. The reader can probably see how this view will explain the meaning of (4): “You can file for an extension before the tax deadline.” This sentence is true when in some of the best-ranked worlds, you file for an extension before the tax deadline.

It is easy to see how to generalize this proposal for “can” to talk of permissions. Consider (7): “People are legally, but not morally, permitted to commit adultery.” Here a single use of “permitted” is modified by “legally” and “morally.” We can start by taking the set of worlds that are consistent with the contextually salient modal base (whatever it may be). This sentence is true when in some of those worlds that are best-ranked by the demands of the law, people commit adultery, and in none of the worlds that are best-ranked by the demands of morality, people commit adultery.

Notice that this proposal explains the close semantic relation between “must” and “can,” and between “obligation” and “permission.” If “People are legally permitted to commit adultery” is true, then “People are legally obligated not to commit adultery” is false. Why? Because “legally” marks an ordering source, and the first sentence says that in some of the worlds that are ranked best by that ordering source, people commit adultery. The second says that in none of the worlds that are ranked best by that ordering source, people commit adultery. Those sentences, then, are contradictory.

Univocality

A core feature of Kratzer’s view is that deontic modals like “must” and “can” are univocal. Consider examples like (6): “You have a moral and legal obligation to pay taxes.” A single use of “obligation” that is modified by “moral” and “legal” is not zeugmatic; it seems to have a clear meaning. The proposal above explains this neatly. Indeed, a notable feature of this proposal is that it allows “must,” “can,” and other closely related terms to be univocal when they are modified by any set of demands—be they aesthetic, prudential, rational, grammatical, and so on.

Similarly, the view does not predict that “legal” is ambiguous in a sentence like (9): “Positivism is a theory of legal institutions and obligations.” Kratzer’s view does not offer an account of the meaning of “moral” or “legal.” Instead, it suggests a natural view about how these words modify the meaning of words like “obligation.” Because “obligation” is relative to a contextually salient ordering source, the inclusion of modifiers like “legal” and “moral” makes a particular ordering source salient. This in no way suggests that “legal” or “moral” always marks an ordering source.

Embedding

Kratzer’s view is widely thought to face no problem explaining uses of “must” and “can” in complex sentences involving logical operators. (Kratzer [1977], notably, discussed many legal examples involving “must” and negation.) So we should expect a similar view about “obligation” and “permission” to face no problem explaining the meaning of these words when they are conjoined, disjoined, negated, or embedded in conditional clauses. Indeed, we already walked through an example involving conjunction with the discussion of (7) above. I won’t dwell on this issue further.

Linguistic Data

Kratzer’s view about “must” and “can” is well-supported by linguistic data.Footnote ²⁶ I think that we should be similarly confident that the same is true of a corresponding view about “obligation” and “permission.” I’ve also illustrated why the view does not predict zeugma in sentences like (6), (7), or (9). But it is worth walking through why the explanation of the meaning of (6) predicts that (14) will be semantically anomalous. If I say “You have a legal and moral obligation to pay taxes,” the use of “legal and moral” makes two ordering sources salient. By contrast, “You have a legal moral obligation to pay taxes” makes a “legal moral” ordering source salient; only there is no such ordering source! The view makes similar predictions about other analogous cases. Conjoining modifiers (e.g., “grammatical and rational”) in modifying words like “requirement” makes two distinct ordering sources salient; stacking modifiers (“grammatical rational”), by contrast, is semantically anomalous unless there is a single eligible ordering source.Footnote ²⁷

Psychological Plausibility

The Kratzerian view under consideration provides a semantics for “legal obligation” in sentences like (1). On its own, it is compatible with different positions about speaker’s attitudes; it does not define out of existence either the internal or external point of view.

As we will see below, many philosophers of law take the meaning of (1) to differ depending on whether it is an internal or external statement (i.e., whether the speaker takes the internal or external point of view). Further, many such philosophers insist that internal and external statements involve a difference in the literal meaning of (1). The Kratzerian view, by contrast, instead provides a univocal semantics for (1). But it is still compatible with holding that the speaker’s point of view can bear on the non-literal meaning of an utterance of (1). The meaning of an utterance is, after all, not all literal.

Here’s an early illustrative example of such an argumentative move. Richard Holton (Reference Holton1998: 611) held that a judge who took the internal point of view and said “You are legally obligated” could have thereby “pragmatically implicated that she believes there is a moral obligation,” but “that doesn’t mean she thinks she has strictly and literally said that there is a moral obligation, or said anything that entails this.” I doubt Holton’s specific proposal about this implicature and how it is generated. But many have endorsed more plausible views about how uses of “must” and “obligation” generate implicatures of the speaker’s practical attitudes.Footnote ²⁸ As such, Kratzer’s view is at least compatible with holding that speakers can say (1) with a different psychological orientation toward the law, and thereby implicate different kinds of claims.

Taking Stock

Kratzer’s approach is widely described as the “dominant,” “canonical,” “classical,” and “textbook” semantics.Footnote ²⁹ That does not mean that the view—especially the simple version sketched here—must be correct, let alone correct in every detail. But it provides a good starting point for this project. It illustrates how a theory of the meaning of “legal obligation” can be compositional: we start with a well-motivated account of the meaning of “obligation,” which then suggests a natural account of how “legal” modifies its meaning, such that we can explain the meaning of sentences like (1) in terms of the meanings of their parts. And, more importantly, it illustrates how such a theory can meet the desiderata discussed in Section 1: It has a broad scope, accounts for the univocality of the relevant terms, explains the role of logical operators, fits with linguistic data, and is psychologically plausible with both the internal and external points of view.

It would seem, then, that philosophers of law providing accounts of the meaning of “legal obligation” would be well-advised to build upon these foundations, and only depart from Kratzer’s view on the basis of careful, considered arguments. But that has not been how the literature has developed to date, which is, in a way, surprising.

In a related context, Brian Leiter (Reference Leiter2011: 665–666) wrote:

Kratzer’s truth-conditional semantics is certainly mature. It has detractors, but hardly seems to be on its deathbed.Footnote ³⁰ Surprisingly, though, it has not been brought to bear on a theoretical question that is centered in the positivism–anti-positivism debate. It may be fine for philosophers of law to march to the beat of a different drummer. They just need to offer views that are either compatible with or superior to a Kratzerian view about the semantics of “legal obligation.” Unfortunately, they’ve mostly offered neither.

Scope

Moral perspectivalism cannot merely be a theory of the meaning of “legal obligation” in sentences like (1). It must have a broad scope, extending to sentences like (5) and (7): “People are legally permitted to commit adultery”; “People are legally, but not morally, permitted to commit adultery.” At first glance, this does not seem to pose a problem. For moral perspectivalists, the literal meaning of (5) is, roughly: According to the legal system’s moral theory, people have a moral permission to commit adultery. The literal meaning of (7) adds that people are not morally permitted to commit adultery. This is not self-contradictory. To assert (7) is, on this view, akin to saying: “Utilitarianism says you are permitted to kill one person to save five; but you are not permitted to do so.”

But moral perspectivalism faces a subtle challenge in extending to “legal permission.” The reason stems from an inconsistency between two claims. First, proponents of the view say that the legal point of view is “not a complete account of morality” (Shapiro Reference Shapiro2011: 186); it is instead an incomplete moral theory. In other words, it leaves gaps: Some acts are, from the legal point of view, neither morally forbidden nor morally permissible.

Second, at least in many legal systems, “everything not forbidden by law is permitted by law” (Gardner Reference Gardner2012: 34). It follows that at least many legal systems leave no gaps: any action is either legally forbidden or legally permitted. For example, if committing adultery is not legally forbidden, then committing adultery is legally permitted.

The inconsistency between these claims arises once we consider moral perspectivalism as a theory of “legal obligation” and “legally permitted,” in sentences like (1) and (5). Take the sentence: “If an act is not legally forbidden, it is legally permitted.” For moral perspectivalists, this literally means something like “According to the legal system’s moral theory, if an act is not morally forbidden, it is morally permitted.” Ditto for any other way of expressing the view that a given legal system has no gaps. Moral perspectivalists, then, either need to say that legal systems do leave gaps (if committing adultery is not legally forbidden, it does not follow that it is legally permitted) or that the law’s moral theory doesn’t leave gaps—which would make it an absurd theory!

Even if the legal point of view is an incomplete moral theory, it may be absurd for other reasons. It is hard to see how the theory will “satisfy the demands of rationality and coherence” (Coleman Reference Coleman2011: 22). Among the class of actions that rational and coherent moral theories would consider to be highly similar, some such actions will be legally prohibited while others will be legally permitted—due to moral constraints on justified enforcement, institutional indolence, or myriad other reasons. This will make a legal system’s underlying moral theory substantively irrational and incoherent. If that seems too harsh, try to construct a rational, coherent moral theory such that in a given legal system, every act that is legally obligatory is morally obligatory and every act that is legally permissible is morally permissible.

The law’s moral theory can also be structurally irrational and incoherent. For example, even if legal systems shouldn’t generate conflicting legal obligations, they sometimes do so.Footnote ³⁷ The law’s moral theory, then, must countenance conflicting moral obligations. That is a highly controversial commitment for any moral theory. In standard deontic logic, conflicting obligations generate deontic explosion: If there are conflicting obligations, then everything is obligatory. (Likewise, in standard classical logic, contradictions generate explosion: everything follows from p and ~p.) At the very least, the structural features of the law’s underlying moral theory can make it deeply implausible.

Univocality

Moral perspectivalism holds that “obligation” is univocal in (1) and (2), and hence predicts that mixed sentences that use “moral and legal obligation” are not zeugmatic in sentences like (6). Indeed, as we saw above with Hershovitz, this view is often partly motivated by a commitment to holding that “obligation” has the same meaning in sentences like (1) and (2)—though it is far from clear why the datum that “obligation” has the same meaning supports the inference that it has a moral meaning in (1). The idea that “obligation” could simply have a generic meaning seems to be ignored. It is worth stressing how odd that is. For instance, taking “obligation” to always have a moral meaning suggests that modifying “obligation” with “moral” is strictly redundant.Footnote ³⁸

What about the univocality of “legal?” Recall (9): “Positivism is a theory of legal institutions and obligations.” It is awkward at best to say “legal” here marks the law’s perspective about institutions. Andrew Currie (Reference Currie2020: 121–122) similarly notes that “I have a legal question” does not literally mean that “The law claims that I have a moral question.” There is also a further problem posed by examples like (9). Proponents of the view in question, like Gardner (Reference Gardner2012), explain the law’s perspective in terms of legal officials and legal institutions. So, when “legal” modifies “institutions,” does it mark the law’s perspective about institutions? Circularity looms. It is, notably, hard to find analogous cases for words like “Kantian” which mark the perspective of a moral theory about something like institutions or questions. Proponents of this view, then, need to do one of two things: either they need an explanation for why “legal” marks the perspective of a moral theory only when it modifies some words like “obligation”; or they need to explain why it similarly marks the perspective of a moral theory when it modifies “question,” “institution,” and a whole range of non-normative vocabulary.

To see why this is a serious challenge, it is worth comparing the issue to how the Kratzerian view (Section 2) explains why “legal” marks an ordering source when it modifies “obligation,” but not when it modifies “institution.” The answer is that terms like “obligation,” unlike “institution,” are always relative to a contextually salient ordering source. The modifier “legal” makes this contribution when modifying “obligation” because of a stable feature of the semantics of “obligation.” By contrast, it is not a stable feature of the meaning of “obligation” that it is always relative to a theoretical perspective. This is why the challenge for moral perspectivalism is much harder to meet. What can they say to explain why “legal” modifies the meaning of “obligation” to mark a theoretical perspective, but does not generally mark a theoretical perspective?

Embedding

Moral perspectivalism also faces a serious challenge with embedding. This is, in fact, the source of a recent objection from Adam Perry: moral “perspectivalism is unable to account for the use of legal claims in logically complex arguments” (Reference Perry2023: 718).

Perry’s argument concerns the example of (11)–(13): “All and only persons over 18 have a legal right to vote”; “A has a legal right to vote”; “So, A is over 18.” The problem, Perry notes, is that this argument is invalid under perspectivalism. The literal meaning of (11) becomes “According to law all and only persons over 18 have a moral right to vote,” and (12) becomes “According to law A has a moral right to vote.” But if this is the literal meaning of sentences (11) and (12), they do not entail (13). As Perry notes, “the invalidity is obvious” when we consider structurally analogous cases where the relevant point of view is false, and the legal point of view is meant to be fallible.Footnote ³⁹

Perry considers a way around this problem by appealing to two commitments. The first is to say that the legal point of view is deductively closed. Perry considers the following closure rule: “If p is entailed by q, r, s, …, and according to law each of q, r, s, …, then according to law p” (Reference Perry2023: 720). But Perry notes that this closure rule “leads to absurd results” for moral perspectivalism. Suppose being a competent adult entails being morally obligated to keep your promises. If, according to the law, you are a competent adult, and what’s true according to the law is closed under the rule just specified, then according to the law, you have a moral obligation to keep your promises. For moral perspectivalists, this means that you now have a legal obligation to keep your promises! That is absurd. But it is not clear whether this problem stems from the specific closure rule that Perry considers. Arguably, the moral perspectivalists can endorse a weaker closure rule; the Kantian moral theory seems to be closed under entailment in some sense, but it is not clear that it is closed under the rule Perry considers above.Footnote ⁴⁰

The second commitment is that (13) is implicitly prefixed (“So, according to law A is over 18”). Perry seems to treat this commitment as innocuous, but it strikes me as ad hoc. Why say the literal meaning of (13) is prefixed when it is preceded by legal premises? Suppose we replace “legal” with “moral” in the premises (11) and (12). Unless we also adopt a perspectivalist semantics for moral language, we do not need to similarly say that the literal meaning of (13) is prefixed for the argument to be valid. Nor should we say that. It seems bizarre to hold that when it is accompanied by (11) and (12), (13) literally means “According to law, A is over 18,” but if we replace “legal” with “moral” in (11) and (12), (13) literally means “According to morality, A is over 18.” The literal meaning of “A is over 18” isn’t prefixed, and it is ad hoc to say that the way it is prefixed as a conclusion depends on whether “legal” or “moral” appears in the premises.

Where does this leave moral perspectivalists? Notice that both moves must work to block Perry’s objection. If moral perspectivalists can endorse a suitable closure rule that does not have absurd implications, that helps them explain why the argument is valid only if the conclusion of the argument is implicitly prefixed. So they face an uphill battle.

Linguistic Data

Moral perspectivalism says that “legal” modifies “obligation” in the same way that “Kantian” modifies “obligation” in sentences like (17). How can we test this?

Wodak (Reference Wodak2018: 801–802) offered a way. Consider how we can stack the modifiers “Kantian” and “moral,” as in: “You have a Kantian moral obligation not to lie.” There is nothing unacceptable about this sentence; stacking “Kantian” and “moral” just clarifies that the Kantian point of view is a perspective about morality. This generates a testable prediction: if “legal” and “Kantian” modify “obligation” in the same way, we should similarly be able to stack “legal” and “moral.” But when we do so, we generate semantically anomalous sentences like (14): “You have a legal moral obligation to pay taxes.” This sentence can only be ascribed a bizarre meaning.

It is worth noting two points about such linguistic data. First, perspectival modifiers can typically be stacked the same way as “Kantian” above. In the following normative and non-normative examples, the addition of the modifier in parentheses does not produce semantically anomalous sentences: “You have a utilitarian (moral) obligation to pay taxes”; and “You have a putative (moral) obligation to pay taxes”; “The duality of waves and particles is an Einsteinian (scientific) fact.” By contrast, stacking modifiers of “obligation” that mark distinct ordering sources typically does produce semantically anomalous sentences: “You have a moral prudential obligation to pay taxes.” This broader diet of examples adds considerable weight to the relevant linguistic data.

Second, moral perspectivalists are under significant pressure to explain such data. Moral perspectivalists could brush off such data if they were engaged in, say, analyzing the illocutionary force of sentences like (2), or offering an account of the pragmatics rather than semantics of sentences like (2).Footnote ⁴¹ But its key proponents, like Raz and Shapiro, advance this theory to support a solution to the challenge posed by Hume’s is/ought gap (see Section 6). As such, moral perspectivalism must be understood as offering exactly what it says on the tin: an account of the semantics of sentences like (1).Footnote ⁴² But a central desideratum for a semantic theory is that it accords with linguistic data.

These two points show that moral perspectivalists need to explain linguistic data about modifier stacking, but have no clear path to do so.Footnote ⁴³ But these points also suggest something methodologically awry about moral perspectivalism. This is first and foremost a semantic theory. So, any proponent of the theory should have been considering what distinctive testable predictions it makes about linguistic data. Proponents of the theory have offered no response to a challenge about such predictions. Worse yet, they never sought to elicit any testable predictions in the first place. This observation is about as damning as the data from sentences like (14).

It is worth emphasizing that the problems with embedding and linguistic data target different components of moralized perspectivalism. The view faces a problem with embedding because it is perspectival (it takes “legal” to semantically mark a point of view); and it faces a problem with modifier stacking because it is moral (it takes this “obligation” to mean “moral obligation”). Each commitment, then, is dubious.

Psychological Plausibility

What about internal and external statements? Internal legal statements, recall, are normative statements made from the point of view of an adherent of a legal system. If we suppose that the legal point of view is a moral theory, then internal legal statements seem to express adherence to that moral theory. That is also how proponents of moral perspectivalism understand the internal point of view. Raz, for example, suggests that “judges believe that legal obligations are morally binding,” and this is “what they actually say when asserting obligations according to the law” (Reference Raz1984: 131).Footnote ⁴⁴ Internal legal statements express the internal point of view; so the internal point of view is moral.

Taking the internal point of view to be moral is independently controversial.Footnote ⁴⁵ But there is also a reason to think it is incompatible with moral perspectivalists’ commitments. This reason again concerns legal permissions, in simple sentences such as (5): “People are legally permitted to commit adultery.” I assume (5) is true. I also assume judges think it is true. A judge may say (5) from the bench if, for example, someone is accused of adultery and the judge wants to point out that they are not thereby accused of a crime. But according to moral perspectivalism, (5) literally means that according to the legal point of view, people are morally permitted to commit adultery. By parity of reasoning, such a judge expresses the belief that adultery is morally permissible, and this is what they actually say when asserting its permissibility according to the law.Footnote ⁴⁶

This implication is hard to swallow. And the bullet is harder to bite when we reflect on two points. First, adultery is just one example. On this view, someone who accepts the law as a system of norms thereby accepts that everything the law does not forbid is morally permissible. And as Matthew Etchemendy notes in a related context, “committing adultery” is “perfectly legal in a wide range of circumstances,” along with plenty of other “nasty or pointlessly destructive things” such as “gratuitously insulting loyal friends” (2018: 19). If we accept moral perspectivalism, the internal point of view looks pathological. Second, a theory that holds that everything not legally prohibited is morally permitted has implausible substantive commitments and structural features. It will not treat morally like cases alike, since the law’s prohibitions are selective.

How, then, should we regard the dominant semantic theory from positivists? Taken on its own, the theory fails to satisfy basic desiderata. But we should also evaluate the theory comparatively. That is, how does the theory fare in comparison to its rivals? The Kratzerian view, by contrast, seems to avoid its pitfalls because it takes “obligation” to have a generic rather than moral meaning, and because it takes “legal” and “moral” to modify “obligation” the same way: “legal” marks a distinct ordering source, like “moral,” rather than a theoretical perspective on morality.

Scope

If expressivism is a theory of the meaning of “legal obligation,” then, as Etchemendy notes, the expressivist also needs a theory of the meaning of “statements like ‘you have a legal right to dispose of your own property,’ and any number of other kinds of internal legal statements” (Reference Etchemendy2016: 19 fn. 56). Toh seems to agree; after all, Toh explicitly takes the expressivist view to have a scope that includes statements like “The Fourteenth Amendment allows states to regulate bakery employees’ work hours.” This example involves language of permission, just like (5): “People are legally permitted to commit adultery.” So the expressivist needs an account of the practical attitude that is expressed by committed internal statements of (5), and how it relates to the practical attitude that is expressed by committed internal statements of (1).

I will again focus on statements about legal permissions rather than (say) legal rights, because it draws out a general problem for the theory. If “Legally, A must φ” expresses plural norm acceptance of A φing, what exactly do sentences like “Legally, A can φ” and “Legally, A cannot φ” express? Expressivists struggle to provide plausible answers that explain which of these sentences are consistent and inconsistent.

The main difficulty is familiar from discussions of how metaethical expressivists should handle negation. Here’s how Mark Schroeder explains that general problem:

Legal expressivists need to explain the semantic relation between “legal obligation” and “legal permission”: if it is true that “You are legally obligated to pay taxes,” then it is false that “You are legally permitted to not pay taxes” (Section 1). And if they are to explain this in terms of the attitudes that committed internal uses of these sentences express, then the attitude expressed by the first sentence needs to be inconsistent with the attitude expressed by the second. As Schroder writes:

Suppose, then, that Toh holds that “You are legally obligated to pay taxes” expresses the attitude of acceptance of your paying taxes. This must be consistent with the attitude that is expressed by “You are legally permitted to pay taxes,” as well as inconsistent with the attitude that is expressed by “You are legally permitted not to pay taxes.” As Schroeder’s discussion above outlines, legal expressivists here face the problem that when we run through the available options, there are not enough spots for the “not.”

There is also a way that the problem posed here for expressivists about legal language is harder than the problem for expressivists about moral language. The legal expressivist needs to explain the inconsistency of the attitudes expressed in “You are legally permitted to φ” and “You are not legally permitted to φ,” while explaining the consistency of the attitudes expressed by “Law X permits φing” and “Law Y does not permit φing.” If these are committed internal statements that express practical attitudes—attitudes like desires, whose contents are actions—it is not immediately clear how the expressivist can generate the result that the first pair of sentences expresses inconsistent attitudes but the similar pair of sentences expresses consistent attitudes.

I do not mean for this to convey that these problems are insoluble. Indeed, there are interesting recent proposals in metaethics that take expressivism in a different direction in an attempt to solve this problem—see, e.g., Incurvati and Schlöder (Reference Incurvati and Schlöder2021). These proposals might provide new resources for legal expressivists. But however, the view is developed, it needs to explain the broad scope of prescriptive legal language.

Univocality

Expressivism faces a further challenge with the univocality of “obligation.” In fact, it faces a harder challenge on this front than any theory we considered so far.

In discussing this issue in Section 1, I noted that the same term, “obligation,” appears in legal language, in moral language, and in mixed language. The examples of this were (1), (2), and (6): “You have a legal obligation to pay taxes,” “You have a moral obligation to pay taxes,” and “You have a moral and legal obligation to pay taxes.” As we saw, sentences like (6) provide evidence that “obligation” is univocal in (1) and (2). That is, “obligation” has the same meaning when it is modified by “legal” and “moral.”

Expressivism is inconsistent with this. It is even inconsistent with something more basic: that “legal obligation” is univocal in (1), and that “moral obligation” is univocal in (2). We saw above that Toh makes this explicit: “expressions containing deontic terms such as ‘ought’, ‘may’, ‘obligatory’, etc. are often ambiguous.” Here Toh cites with approval Eugenio Bulygin’s ‘Norms, Normative Propositions, and Legal Statements’, which endorses the view that “deontic sentences”—that is, “sentences in which such deontic terms as ‘ought’, ‘may’, ‘obligatory’, ‘forbidden’, ‘permitted’, and so on occur”—“are typically ambiguous” (Reference Bulygin, Bernal, Huerta and Mazzarese2015 [1982]: 188). Toh does not just happen to endorse this stance. It falls out of the expressivist semantic program, as Toh understands it. The expressivist offers a non-cognitivist semantics for committed internal statements; external statements are said to differ in meaning. So, for the expressivist, terms like “ought,” “must,” “can,” “obligation,” and “permission” are not univocal in legal language like (1), or in moral language like (2), let alone in mixed uses like (6). Whether or not Hart is an expressivist, Hart also seemed to flatly deny “the identity of meaning of ‘obligation’ in moral and legal contexts,”Footnote ⁵⁵ and many take that to be Hart’s considered view.Footnote ⁵⁶ This would suggest that Hart similarly denies the univocality of “obligation” in (6) and hence predicts that its use in that sentence is zeugmatic.Footnote ⁵⁷

The expressivist’s commitment to ambiguity is a radical departure from a Kratzerian view (Section 2). As I noted, on Kratzer’s semantics, deontic modals are univocal. Indeed, on this view, modals are univocal. That is: “must” has the same meaning not only when it is used in relation to duty, but when it is used in relation to necessity, be it deontic, logical, metaphysical, epistemic, or what have you. When we compare the expressivist view to its Kratzerian rival with respect to univocality, they could not be further apart.

For three reasons, I think this constitutes a significant vice of an expressivist semantic program, and a substantial relative virtue of a univocal semantics, like Kratzer’s. First, as we saw, it is standardly thought that the onus is on those who posit lexical ambiguities to proffer linguistic data that supports their position. Toh does not offer any evidence from any standard test for lexical ambiguity. Nor do other expressivists. As we saw in Section 1, such tests support univocality, not such profligate ambiguity.

Insofar as evidence is offered for the claim that deontic terms are “typically ambiguous,” it seems to be of the wrong kind. In a telling moment, Bulygin moves from a claim about “the ambiguity of deontic expressions” to a claim about “the practical difficulty of determining which speech act or acts have been performed on a given occasion” (Reference Bulygin, Bernal, Huerta and Mazzarese2015 [1982]: 198–199, emphasis mine).Footnote ⁵⁸ This seems to confuse the orthodoxy that a single utterance of (1) can constitute multiple speech acts with the heterodoxy that there is no univocal semantics for a single sentence like (1). Toh (Reference Toh2005: 96–99) similarly holds that the development of J.L. Austin’s speech act theory was a major influence on Hart’s views about the semantics of legal obligation, and Toh appeals to Austin’s speech act theory extensively. But as I noted in Section 1, Austin’s speech act theory is based on the distinction between locutionary, illocutionary, and perlocutionary acts—where only locutionary acts relate to literal meaning. The claim that an utterance of (1) can constitute different kinds of speech acts is not equivalent to the claim that the single sentence (1) has multiple literal meanings depending on the speaker’s attitudes.

Second, a view that holds that deontic terms are univocal will be more parsimonious than a view that holds that they are ambiguous. Again, we are finite creatures. Our brains are limited. As Kratzer notes, we use “must” and “obligation” to refer to many different kinds of duties. If obligation is not univocal in (1), in (2), or in (6), then as we consider more examples of normative language, we soon get the result that our brains must somehow learn, store, and aptly deploy many multitudes of meanings for each deontic term. Toh could say something to soften the force of this blow. Toh writes:

If the meaning of external uses of “legal obligation” is derivative of the meaning of committed internal uses of the same phrase, the proliferation of lexical entries may be less dire: we still have distinct meanings for “legal obligation,” but at least they are related. (In other words, “legal obligation” would be polysemous.) Similarly, we would get distinct but related meanings for “moral obligation,” and for “grammatical requirement,” and so on. But the real problem, I think, is that “obligation” still has a distinct meaning in (6), and ditto for “requirement” in a mixed use like “moral and grammatical requirement,” and so on. At best, this move makes a vice somewhat less vicious; but it does not come close to turning that vice into a virtue.

Third, if “obligation” is not univocal in (1), it becomes harder to explain disagreement. Toh takes it to be an ambition of the expressivist program to “make sense of legal disagreements between people who subscribe to different fundamental norms” (2011: 116). So suppose Mitch says (1) and Kim responds by asserting ~(1)—that is, “You do not have a legal obligation to pay taxes.” They disagree. This remains the case even if Mitch takes the external point of view and Kim takes the internal point of view, or vice versa. As Finlay and Plunkett note, “expressivists like Toh seem committed to a separate account of the semantics of external statements of law, and therefore are hard-pressed to explain how such disagreements are possible” (Reference Finlay and Plunkett2018: 70). Why is that? We saw that Toh takes the external legal statements, as Etchemendy wrote, to be akin to anthropologist claiming “Infanticide was morally permissible in Sparta.” If an anthropologist said that and someone responded, “No, infanticide was morally impermissible in Sparta!”, they would be talking past each other. The confusion could be cleared up by the anthropologist: “I simply meant that Spartans generally thought that infanticide was morally permissible.” The disagreement between Mitch and Kim does not seem similar to this exchange. They do not seem to be talking past each other, and neither could say anything similar to the anthropologists’ clarification to clarify their merely verbal dispute, because their dispute is not merely verbal.Footnote ⁵⁹

Perhaps, then, the expressivist should change course. What if expressivists instead aim to offer a univocal semantics for “obligation” in (1), (2), and (6)? This is no easy task. In fact, they face at least two hurdles in trying to pull it off.

First, let’s consider how the expressivists can explain the univocality of “obligation” in (1). Suppose, that is, one endorsed an expressivist view about internal and external statements of (1). Some have suggested (sloppily, I think) that Hart held this view.Footnote ⁶⁰

The main obstacle to this view is that expressivists aim to explain the meaning of sentences in terms of the attitudes that they express, and hold that internal statements of (1) express a practical attitude, norm-acceptance, but those who take the external point of view lack that attitude. Expressivists would now need to say that any utterance of (1) expresses the attitude of norm-acceptance even when the speaker does not have that attitude. To take this stance is to say that external legal statements are like fake smiles: They express an attitude that the speaker lacks, so they are insincere. But to simply assert this is to posit insincerity without evidence. That’s little better than positing lexical ambiguity without evidence. We need an argument for why (1) has these sincerity conditions, and one that tracks the typical ways of determining sincerity conditions in philosophy of language. The violation of sincerity conditions typically makes a speech act infelicitous. If you promise to read my draft but have no intention of doing so, your promise is insincere and so infelicitous. It is, in the parlance of speech act theory, an abuse. But if someone who takes the external point of view says, “You have a legal obligation to pay taxes,” their speech act does not seem infelicitous.Footnote ⁶¹ So if the expressivist says internal and external legal statements are univocal in expressing the practical attitude of norm acceptance, then the expressivist faces an uphill battle in making a case for the surprising verdict that external legal statements are insincere.

The expressivist also faces a second hurdle when they turn to explaining the univocality of “obligation” in mixed sentences like (6). How can the expressivist capture the datum that “obligation” is univocal when it is modified by “moral” and “legal?” Wodak (Reference Wodak2017) argued that expressivists here face a challenge in explaining univocality without over-generating semantic inconsistency.

There are two starting points for this argument. First, expressivists explain sameness of meaning in terms of sameness of expressed attitudes. If “obligation” has a univocal meaning in (6), then it must express acceptance when it is modified by “moral” and “legal.” If “obligation” expresses acceptance when modified by “legal” and, say, hope when it is modified by “moral,” the single use of “obligation” in (6) should seem zeugmatic. Second, expressivists explain the inconsistency of sentences in terms of the inconsistency of the attitudes those sentences express. Suppose the expressivist adopts what I described as the straightforward option above: “You have a legal obligation to pay taxes” expresses acceptance of your paying taxes. That sentence is inconsistent with “You do not have a legal obligation to pay taxes,” so the negated sentence must express an attitude that is inconsistent with acceptance of your paying taxes. With these two starting points in mind, we can now consider a moral sentence like “You do not have a moral obligation to pay taxes.” Why isn’t the attitude that this sentence expresses also inconsistent with the attitude expressed by “You have a legal obligation to pay taxes?”

You might think that this problem is easy to resolve. The expressivist just needs to say that (1) and (2) express the same type of attitude, but with different contents. For example, Etchemendy finds it plausible that “A has moral obligation not to steal” expresses acceptance of blaming A if A steals,Footnote ⁶² while “A has a legal obligation not to steal” expresses acceptance of punishing A if A steals (Reference Etchemendy2016: 16, 20). Many hold, however, that punishment constitutively involves the expression of blame. If that is true, then “A has legal obligation not to steal” expresses acceptance of punishing and hence blaming A if A steals. So it expresses an attitude that is inconsistent with the acceptance of not blaming A if A steals. More generally, it is hard to see how expressivists can offer a systematic explanation of why moral and legal claims are not inconsistent. Nothing on this view seems to guarantee that sentences like (7)—“People are legally, but not morally, permitted to commit adultery”—are never self-contradictory.

Interestingly, Toh grants that legal, moral, and even aesthetic judgments can all express the same specific type of practical attitude, plural norm acceptance (Reference Toh, Green and Leiter2011: 130). Toh then says this is not a problem because it “does not mean that all expressions of plural norm‐acceptances, including legal statements, are a species of moral statements,” and writes that there are likely to be “differences among the kinds of norms which are accepted in different kinds of normative judgments.” That is, “A legally ought to φ,” “A morally ought to φ,” and “A aesthetically ought to φ” may all express plural norm acceptance, but will not all express plural norm acceptance of A φing. This doesn’t help solve our problem. Suppose “A legally ought to φ” expresses acceptance of A φing and “A morally ought to φ” expresses acceptance of A ψing. These are consistent attitudes. But now ask: Why are “A legally ought to ψ” and “A morally ought to φ” consistent? Applying the view systematically, we still generate some unwanted inconsistencies.

Finally, there is a related question of whether the expressivist can explain the univocality of “legal” in sentences like (9): “Positivism is a theory of legal institutions and obligations.” It is not clear what to make of these sentences. What does “legal” contribute when a single use of the word simultaneously modifies a term that expresses a non-cognitive attitude (“obligations”) and a term that seems straightforwardly descriptive (“institutions”)? Perhaps the expressivist will respond by also offering a non-cognitivist analysis of “institutions.” But that response will struggle to be systematic. To modify Currie’s earlier example: “I have so many legal questions and legal obligations” can be conjoined and reduced to “I have so many legal questions and obligations.” Will we need to be expressivists about “questions” as well to make expressivism viable? Enoch and Toh “argue that legal is illuminatingly conceived as a thick concept,” such that “declaring an act legal (or illegal) would involve an expression of some evaluative or normative commitment” (Reference Toh2013: 264–65). If legal is a thick concept, and we token that concept once in sentences (9), then it seems to follow, oddly, that declaring an institution or question to be legal similarly involves some normative commitment.

Embedding

The last two problems are serious. But we’re only now coming to the most famous problem for expressivists, which concerns arguments like (11)–(13): “All and only persons over 18 have a legal right to vote”; “A has a legal right to vote,” “So, A is over 18.” The problem that examples like this generate is known as the Frege-Geach problem.Footnote ⁶³ (The negation problem is really a special instance of the broader Frege-Geach problem.)

There are two aspects to Frege-Geach problem. First, if we suppose “A has a legal right to vote” expresses acceptance, and this captures its meaning, what does the sentence express and mean when it is embedded in a conditional? Consider, e.g., “If A has a legal right to vote, that’s not good for your candidate’s odds in November.” It is hard to see how the expressivist account of the meaning of the bare asserted sentence is meant to relate to the meaning of the sentence when it is embedded (or conjoined, disjoined, etc.).

Second, how does the expressivist explain the validity of the inference to “A is over 18?” If “A has a legal right to vote” has a different meaning in the two premises, then the argument commits a fallacy of equivocation. More generally, if we understand validity in terms of truth-preservation—if the premises are true, the conclusion must be true—then how can the argument be valid if its premises are not truth-apt?

Toh writes that “Hart was aware of this second task involved in solving the Frege-Geach problem, and that he thought that it can be carried out to satisfaction,” though Toh notes that Hart does not offer “a fully worked out proposal for a solution to the Frege-Geach problem” (Reference Toh2005: 104–05), and said little about the first task. As Kramer also notes (Reference Kramer2018: 415), the kind of proposed solution that Hart built upon for the second task is at odds with Hart’s own commitments about legal obligations (that they can conflict).

Of course, legal expressivists could seek solutions to this problem that have been developed elsewhere. But despite the enormous volume of work on the Frege-Geach problem over the past decades, not only is there no agreement about whether it has been solved, but “it remains highly controversial whether it can be solved” (Finlay and Plunkett Reference Finlay and Plunkett2018: 70, emphasis mine). It’s not clear that we should be very optimistic about the prospects that legal expressivists will find or develop a solution.

Linguistic Data

Some argue that expressivism also faces problems with linguistic data. Jack Woods (Reference Woods2014) notes that expressivism predicts that some sentences should seem Moore-paradoxical, like “It’s raining, but I don’t believe that it’s raining.” If asserting p expresses the belief that p, the asserting p but I don’t believe that p is marked. Woods notes that a similar pattern exists with, for example, “Go Red Sox! But I don’t support the Red Sox” (Reference Woods2014: 5). But as Woods notes, we do not get a similar pattern with “Murder is wrong, but I don’t disapprove of it.” Expressivism predicts otherwise. It says some sentences like that should seem Moore-paradoxical. Consider an example:

1. (18) You have a legal obligation to pay taxes, but I don’t approve of you paying taxes.

There’s nothing Moore-paradoxical about (18). But if “A has a legal obligation to φ” expresses approval of A φing, there should be! My conjecture is that this problem persists no matter how we fill in the details of the expressivist’s view: (18) will not seem Moore-paradoxical if we change the right-hand side by changing “approve” to “accept” or “intend,” or “your paying taxes” to whatever act-description you like.

Psychological Plausibility

Expressivism explains the meaning of normative sentences in terms of the attitudes that they express, so it might seem that the view will not have any problems with psychological plausibility. If we do not have the relevant attitudes, the expressivist should not be explaining the meaning of our sentences in terms of those attitudes!

But other considerations can push expressivists away from offering a view that is psychologically plausible. As we saw, Toh says internal legal claims like (1) express plural norm acceptance. The reason for this is that Toh thinks this account of the relevant attitudes helps address a common objection to expressivism. But as Toh notes, that solution comes at some cost to the psychological plausibility of the view:

Toh grants that to many this will be a counterexample, but says it is not “damning.”

Regardless of whether Toh is right about that example, it is not the only case where expressivists will face pressure with respect to the psychological plausibility of their view. The expressivist needs their account of our attitudes to explain an awful lot: the semantic relations between “obligation,” “permission,” and negation; univocality and inconsistency; the validity of deductive inferences; why no sentences like “You have a legal obligation to pay taxes but I don’t approve of your paying taxes” are Moore-paradoxical. It is unlikely a simple, familiar attitude can do all of this work. But if the expressivist appeals to some complex, exotic attitude to do this explanatory work, it will become less likely that the attitude is widely held, shared, and expressed in language.

Summing up

Expressivism does have promising features, but it also faces serious pitfalls. Is it worth the cost? That will depend, in part, on whether we can get the goodies another way. And here it is worth comparing expressivism to the view Finlay and Plunkett call quasi-expressivism.Footnote ⁶⁴ As they put it, quasi-expressivism “agrees with expressivism that a central class of (legal or moral) statements is expressive of noncognitive attitudes or prescriptions,” but disagrees with expressivism “because it diagnoses this as a feature of the pragmatics of these statements, rather than of their (purely descriptive) semantics” (Reference Finlay and Plunkett2018: 50). In other words, a speaker’s committed internal use of (1) pragmatically implicates that the speaker has the relevant practical attitude (e.g., norm-acceptance of your paying taxes), but this is not the literally meaning of their use of (1). Its literal meaning is still captured by a familiar truth-conditional semantics.

As a final point, it is worth reflecting on an interesting feature of Toh’s proposal. Expressivists in metaethics have said little in general about flavors of deontic modality (i.e., different ordering sources like morality, prudence, law, and etiquette) and artificial normative systems in particular (like law and etiquette). As Jan Dowell (Reference Dowell, Lepore and Stojnic2024) notes, a sentence like “Members must not bring women to networking events” can be used in a context where it is understood relative to club rules, and “accorded no normative or action-guiding status whatsoever”; its use “seems representational, not motivational.” Dowell notes that such cases seem a poor fit for an expressivist approach. But Dowell’s example is close to the paradigm case for Toh’s view.

Scope

Fictionalism can borrow a Kratzerian semantics, and thereby inherit its theoretical virtues. Recall that for Kratzer (Section 2), “A must φ” means roughly A φs in all of the top-ranked worlds that are consistent with the modal base, while “A can φ” means roughly that A φs in some of the top-ranked worlds that are consistent with the modal base. Generalizing from this account of deontic modals provides a semantics for “obligation” and “permission.” And as we saw, this enables a clear account of the meaning of sentences like (7): “People are legally, but not morally, permitted to commit adultery.” And it explains the close semantic relations between “obligation” and “permission.”

Univocality

For the same reason, fictionalists can offer a univocal semantics for (1) and (2), so they face no problem explaining the univocality of “obligation,” and hence why the term is not zeugmatic in (6). Similarly, the view does not deny the univocality of “legal,” as it does not treat “legal” as if it semantically marks a fictional perspective in (1).

You may still worry about whether there is a similar issue with the assertoric rather than semantic content of mixed sentences like (6). How do we interpret what’s said by the speaker if the speaker is talking about what’s true according to reality and according to the speaker? And does “legal” mark a fictional perspective in the assertoric content of (1) and (6), such that it must also do so when it modifies “institutions” in (9)?

These are good questions. Here are two reasons to think the fictionalist faces no serious problems in answering them. First, consider analogous mixed sentences: “Kim and Atticus Finch are lawyers.” There’s no semantic ambiguity in this sentence, and it also does not seem that we face a serious difficulty in understanding what is being asserted.

Second, recall the above discussion of what moral realists say about the semantics of (2). Claims about “moral obligation” are not always made true because of facts about objective moral reasons. We saw this earlier with our anthropologist’s claim about the moral permissibility of infanticide in Sparta. As Worsnip argues, the speaker’s clear intentions play some role in determining when this is the case. That’s why the anthropologist’s claim about the moral permissibility of infanticide in Sparta turns out to be true: The speaker clearly intends to be talking about the compatibility of infanticide with the positive morality of Spartans, not the compatibility of infanticide with objective moral reasons. Similarly, if we grant the arguments from Perry (Reference Perry2015) and Wodak (Reference Wodak2018) discussed above, the speaker’s attitudes in internal legal statements of (1) play a role in determining why the relevant ordering source for what’s asserted is a legal fiction. The legal fictionalist need not hold (even with respect to what’s asserted) that “legal” always picks out what’s true according to a legal fiction, just as the moral realist does not hold that “moral” always picks out what’s true according to objective morality.

Embedding

Recall Perry’s objection to moral perspectivalism, which concerns arguments like (11)–(13): “All and only persons over 18 have a legal right to vote”; “A has a legal right to vote”; “So, A is over 18.” Since the literal meaning of the premises concerns what’s true according to a (fallible) theory, the argument turns out to be invalid. As Perry noted (Reference Perry2023: 718), this is “analogous” to an objection to “some types of fictionalism” from Vision (Reference Vision1993). So, you may well wonder, is this also a serious problem for fictionalism?

It depends. Which types of fictionalism did Vision have in mind? Those that think sentences like “Hamlet is melancholy” are in some sense true, and so insist upon “prefixing sentences of this ilk with a phrase on the order of ‘In the play Hamlet … ’” (1993: 150). Hence, prefix fictionalism has a problem with embedding. But this is not a problem for fictionalism per se. It is the reason I prefer pretense fictionalism. Many take the view that pretense fictionalism, as Nolan, Restall, and West noted above (Reference Nolan, Restall and West2005: 317), faces no Frege-Geach-like problems with logically complex arguments.

Linguistic Data

Fictionalism avoids the problems with linguistic data that afflict the other views we considered. It does not predict that sentences like (7) are self-contradictory: “People are legally, but not morally, permitted to commit adultery.” It does predict that sentences with stacked modifiers like “legal moral” in (14) are semantically anomalous. It does not predict that any sentences like (18) are Moore-paradoxical. That’s all good news.

There is also some linguistic data that fictionalism is best suited to explain, like:

1. (19) Legally, you have to cross the street at the lights. But you don’t really have to.

1. (20) Legally, you can commit adultery. But you can’t really do that.

I noted how fictionalism explains such data above. Sentences like (19) and (20) fit a similar pattern with “Atticus Finch is a lawyer. But Atticus Finch isn’t really a lawyer.”

Psychological Plausibility

As I noted above, Perry and I argue for fictionalism by arguing that it provides a plausible account of the internal point of view in terms of the attitude of accepting a fiction. The external point of view, for the fictionalist, is the corresponding attitude of rejecting the legal fiction. (I won’t say much about the attitude of rejecting a fiction; it should be familiar for anyone who suffered through Christopher Nolan’s Tenet.)

But some have thought that fictionalism struggles with psychological plausibility in most contexts, and perhaps especially with law. Spaak, for example, said that Kelsen’s fictionalism aspired to “mak[e] explicit what all jurists unconsciously assume in their legal thinking”; Spaak then argued that it is doubtful that all jurists believe that “legal validity (or normativity) as a fiction” (Reference Spaak2022: 184). This tracks a common concern about fictionalist proposals about mathematics, modality, and other domains of inquiry: If when we make claims like (1) we’re engaged in pretense, why don’t we recognize this?

This objection is damning for those like Bentham and Kelsen who assume that when we engage in fiction, we know we are asserting something false. Wodak argued against that assumption in two ways (Reference Wodak2019: 842–844). First, fictions are not always false.Footnote ⁷⁹ Historical fictions are an obvious example: what’s true according to the fiction is in fact often true. And historical fictions are not an exceptional case. If Arthur Conan Doyle writes “London is foggy” and “Holmes lived at 221b Baker Street,” the story contains literal truths and literal falsehoods. Fictions can be false. They are not necessarily false. If they were, why would people read fictions to learn about the real world?Footnote ⁸⁰ So if fictions are not necessarily false, accepting a fiction and engaging in pretense cannot be understood as knowingly asserting falsehoods. This point is especially helpful for the legal fictionalist if we understand the content of the relevant fiction such that it can be true. If a fiction about legal reasons for action is a fiction about reasons which can be real, but need not be real,Footnote ⁸¹ then accepting that fiction will involve accepting propositions which are true according to the fiction and sometimes literally true.

Second, we can get swept up in fictions without recognizing that this is the case. Just as we can believe p without believing that we believe p, we can fictively accept p without believing that we fictively accept p. Opacity of mind is common (see Carruthers Reference Carruthers2011). To know whether our attitudes in a domain of inquiry are fictive requires theoretical reflection (Joyce Reference Joyce2001: 292). Many of us do not ask whether our attitudes in a domain bear the distinct markers of fictive acceptance that were briefly discussed above. And if we did ask, the answer may not be obvious in cases where we take p to be true and true-according-to-a-fiction; it’s harder to tell whether you have the attitude of accepting that London is foggy when you happen to independently believe that London is foggy.

Summing up

Again, a fictionalist approach to the semantics of “legal obligation” is far less developed than its rivals. But given how much the view can borrow from the semantic views offered by others, fictionalism can stick closely to a promising Kratzerian approach (Section 2), while offering a promising account of the ways in which claims like (1) and (2) seem both similar and different. They can receive an identical semantic treatment, so for moral realists the relevant uses of (2) are true when tax-paying is supported by moral reasons, and for legal fictionalists (1) is true when tax-paying is supported by legal reasons. But they are different insofar as those legal reasons need not be real.