The Logical Possibility of Moral Dilemmas in Expressivist Semantics: A Case Study

1.1. Ethical expressivism and the logical possibility of moral dilemmas

A moral dilemma is defined as a situation where incompatible courses of action A and B are both morally obligatory for an agent. The reality of such dilemmas in human life seems indubitable. One can easily imagine, and often find oneself in, situations where general moral precepts such as “Do not lie”, “Do not steal”, “Help your family and friends”, may come to conflict with one another. One can see the reality of moral dilemmas most vividly in situations where one and the same moral precept seems to generate conflicting but equally strong demands. Consider the often-cited case of Sophie’s choice (Styron 1979): Sophie and her two children are at a Nazi concentration camp, and a guard tells Sophie that only one of her children will be allowed to live but the other will be killed. For each child, Sophie has an obligation to save him/her, but she cannot save both. In this case, it is implausible to think that Sophie can resolve the conflict by thinking that one of her obligations overrides the other, because there is no obvious reason why either of them should be stronger than the other. This and similar examples suggest that genuine, that is, irresolvable dilemmas are possible and often real.1

As McConnell (1978, 2022) points out, however, the possibility of irresolvable moral dilemmas apparently conflicts with some of the prima facie plausible axioms of the standard deontic logic. (The presentation of the problem below follows McConnell (1978)). Crucially, the problem concerns the logical possibility of moral dilemmas—adopting the relevant axioms leads to the result that moral dilemmas are impossible as a matter of logic and the meanings of the relevant normative expressions alone. For the purpose of this paper, I will specifically focus on the two axioms, which are meant to capture the following prima facie plausible theses: (1) permission can be defined in terms of obligation, and (2) obligation implies permission.2 Let OA stand for “A is obligatory” and PA stand for “A is permissible”. First, a moral dilemma is a situation where incompatible courses of action are both obligatory. To capture its troublesome nature, one can characterize a moral dilemma as a situation where the following holds:

A moral dilemma is a situation where A is obligatory, but not doing A is obligatory as well because it is necessary for doing B, another obligatory action—e.g., saving one child’s life is obligatory for Sophie, but not doing so is obligatory as well because it is a necessary means for saving the life of the other child, which is another obligatory act for her. Second, (1) states that doing A is permissible if and only if it is not the case that not doing A is obligatory:

Lastly, the symbolic representation of (2), obligation implies permission, is the following:

(1) and (2) entail OA ⇒ ~O~A. Assuming the material conditional, OA ⇒ ~O~A is equivalent to ~(OA ∧ O~A). This, however, directly contradicts (MD). Hence, (1) and (2) are jointly inconsistent with (MD). Adopting the apparently intuitive theses (1) and (2) as axioms in one’s theory thus rules out the logical possibility of moral dilemmas.

Since McConnell (1978) provided a formal presentation of the puzzle, many answers have been proposed in the literature.3 Those who think moral dilemmas should be at least logically possible seek ways to justify abandoning (at least) one of the axioms that give rise to the inconsistency, whereas opponents argue that any putative solution to the puzzle is bound to be ad hoc.4 (At this point, it may need to be noted that there are also other combinations of axioms that are known to be inconsistent with the logical possibility of moral dilemmas. I will come back to this point below.)

In this paper, for the sake of discussion, I tentatively assume that expressivists should generally side with the pro-dilemma view.5 A prima facie justification for this assumption stems from the fact that ethical expressivism is usually construed as a semantic view, that is, a view about the meanings of normative sentences in a natural language, such as English. (In the next section, I will elaborate on this point in more detail.) As such, proponents of ethical expressivism should be committed to formulating a theory that correctly reflects ordinary speakers’ use of, and linguistic intuitions about, the target normative expressions—they constitute the data that should guide one’s theory construction. Crucially, ordinary speakers’ language use and linguistic intuitions seem to suggest that they take moral dilemmas to be at least conceptually possible. Ordinary speakers do often find themselves in dilemmatic situations and describe them as such, and, accordingly, they do not seem to take “A is obligatory for S, but not-A is also obligatory for S” as an utterly confused, non-sensical statement. To accommodate the data, one’s expressivist semantic account should not entail that moral dilemmas are impossible, as a matter of logic and meaning alone. When conjoined to the puzzle concerning the logical possibility of moral dilemmas presented above, this means that expressivists should construct their semantic theories so that obligation implies permission will not turn out to be formally valid (assuming that they do not want to reject the interdefinability of obligation and permission).6

To see the intuitive plausibility of this assumption, it may also be helpful to point out the fact that those who deny the possibility of moral dilemmas are usually motivated by substantive moral-theoretic concerns that do not necessarily coincide with ordinary speakers’ language use and linguistic intuitions (McConnell 2022, Sec. 3 and 4). For example, a Kantian theorist might argue that moral dilemmas should be impossible, because one’s (deontological) moral theory should be uniquely action-guiding in the sense that it will never prescribe incompatible actions for an agent in a given situation. To meet this requirement, one might propose that there should be some way of hierarchically structuring moral precepts so that irresolvable conflicts will never arise (for an influential critique of this idea, see e.g., Ross 1930, Ch. 2). Whether this line of reasoning is plausible or not, it clearly concerns a requirement that is to be imposed on one’s substantive theory of morality, not one’s semantic account of normative expressions in a natural language. Semantics must respect the fact that ordinary speakers often talk about moral dilemmas meaningfully and they do seem to take dilemmas to be at least conceptually possible. The assumption of this paper is that this constitutes a prima facie reason for expressivists to try to secure the logical possibility of moral dilemmas in their semantic accounts. Whether the possibility of moral dilemmas should be ultimately excluded in one’s substantive moral theory is a separate question, which I do not intend to address in the current paper.

In the rest of the discussion in this paper, as a case study, I will examine how this general problem concerning the logical possibility of moral dilemmas arises taking a specific shape in the expressivist semantic framework developed by Mark Schroeder (2008a) and explore how one might respond to it. There are two reasons why I specifically focus on Schroeder’s expressivist semantics to explore this issue. The first is that, although ethical expressivism is one of the most discussed views in the literature on the semantics of normative language, apparently no proponents of expressivism have done a better job than Schroeder in actually constructing a semantic theory that seems to provide a promising response to the so-called Frege-Geach problem (Geach 1960, 1965). The Frege-Geach problem, in short, is a demand for expressivists to develop a compositional semantic theory that correctly captures logical-semantical relations between non-atomic, logically complex normative sentences (e.g., negations, disjunctions, conditionals, sentences with quantifications, etc.). In most expressivist proposals before Schroeder, the details of expressivist semantic theories were not fully spelled out, and it was even unclear if they could adequately explain such basic semantic phenomena as the logical inconsistency of “A is wrong” and “A is not wrong”—as I will explain in the next section, this is what Unwin (2001) calls the “negation problem” for expressivists, which is an instance of the Frege-Geach problem as applied to negation. An important contribution of Schroeder’s work is that it identifies a structural requirement that any expressivist semantic account should meet to deal with the Frege-Geach problem.7 Exploring the issue of the logical possibility of moral dilemmas in his framework will provide a useful case study, because my initial goal is to show that an expressivist semantic theory that provides an adequate solution to the Frege-Geach problem does not necessarily succeed in securing the logical possibility of moral dilemmas as well. The second reason is purely pragmatic: at one place in his book, Schroeder (2008a, Ch. 5, Sec. 4) himself discusses the relevant thesis, obligation implies permission, in connection to the logical possibility of moral dilemmas. I will develop my own discussion by building on Schroeder’s.8

Relatedly, as I noted earlier, there are also other combinations of intuitively plausible axioms that are known to be inconsistent with the logical possibility of moral dilemmas. For example, obligation implies possibility (i.e., ought implies can) and the principle of agglomeration, which states that if A is obligatory and B is obligatory, then A and B is obligatory (Williams 1966), are jointly inconsistent with the possibility of moral dilemmas (I will not provide a proof here—interested readers should consult surveys such as McConnell 2022, Sec. 4 and McNamara and Van De Putte 2023, Sec. 6.4.) If expressivists need to side with the pro-dilemma view, they will ultimately have to find ways to avoid such combinations as well. Due to the limit of space, I will not explore other combinations and focus only on obligation implies permission. In Section 4, I will briefly discuss how one might interpret the results of this paper in light of this broader point.

The rest of the discussion in this paper will proceed as follows. In Section 2, I will briefly review several basic features of ethical expressivism construed as a semantic project and introduce Schroeder’s expressivist semantic theory against that general background. As noted, Schroeder’s theory is mainly motivated as a response to the negation problem, which is a special instance of the Frege-Geach problem. The discussion may get technical in places, and to avoid unnecessary complications, in Section 2, I will not discuss how the technical tools developed by Schroeder connect with the issue of the logical possibility of moral dilemmas. I will turn to this issue in Section 3: I will explore how one might respond to the problem of securing the logical possibility of moral dilemmas within Schroeder’s framework and make some general observations from the case study. In Section 4, I will conclude by articulating the moral of the case study in the most general terms.

The last caveat before the main discussion: in presenting Schroeder’s expressivist semantics, I will, as Schroeder himself does, focus on the predicate “is wrong” as the main target of semantic analysis, instead of predicates such as “is obligatory” or “is permissible” that I used in introducing the issue of the logical possibility of moral dilemmas. As I will explain in Section 3, however, there is an easy and relatively uncontroversial way to translate claims that contain the predicate “is wrong” to those that only contain “is obligatory” and “is permissible” (for example, “A is obligatory” could be rephrased as “Not A is wrong”). Although this invites some complication, it does not pose any serious problem for the main discussion of this paper. Schroeder himself also notes that his discussion is applicable to expressivist views that take different normative predicates (e.g., “is rational”, “is the thing to do”, “ought”, and so on) as basic (Schroeder 2008a, 7, see, also, 39). I will come back to this point in Section 3.2.

3.1. The logical possibility of moral dilemmas in Schroeder’s expressivist semantics

So far, following Schroeder, I have been focusing on the predicate “is wrong”. To address the question whether one can make obligation implies permission formally invalid in Schroeder’s framework, it is necessary to translate all the sentences that contain “is wrong” to the sentences that contain “is obligatory/permissible”. The required translation is shown in the following table.

Table 1

The proof of the table only requires two uncontroversial assumptions: (a) “Not φ-ing is obligatory” is a translation of “φ-ing is wrong”, and (b) obligation and permission are interdefinable (i.e., “φ-ing is permissible” can be defined as “Not φ-ing is not obligatory” and vice versa). These directly yield the result that “φ-ing is not permissible” is a translation of “φ-ing is wrong” (represented in the first row), and one can similarly prove the rest of the table.18

I believe that these are natural assumptions in extending Schroeder’s analysis of “is wrong” to “is obligatory/permissible”. In Schroeder’s framework, (a) amounts to the idea that “φ-ing is wrong” and “Not φ-ing is obligatory” express the same being for attitude, namely, FOR(blaming for φ-ing). (b) amounts to the idea that “φ-ing is permissible” and “Not φ-ing is not obligatory” express the same attitude, namely, FOR(not blaming for φ-ing). In fact, once one accepts (a), it is unclear how one can avoid (b) in Schroeder’s framework: to reject (b), one would have to maintain that the pairs of the sentences in each of the rows (such as “φ-ing is permissible” and “Not φ-ing is not obligatory”) express non-equivalent being for attitudes. It is questionable whether one can come up with any reasonably simple assignment of being for attitudes that meets this condition. (And, as I noted in footnote 6, Schroeder himself takes the interdefinability of obligation and permission to be uncontroversial anyway.)

As one can see from the table, the claim that “φ-ing is obligatory” implies “φ-ing is permissible” translates to the claim that “Not φ-ing is wrong” implies “φ-ing is not wrong”. As I discussed in Section 1, whether obligation implies permission turns out to be formally valid in one’s semantic theory is an important question, because of its connection to the classic puzzle concerning the logical possibility of moral dilemmas. The puzzle was that the logical possibility of genuine dilemmas conflicts with the apparently plausible theses, each of which one might be inclined to treat as an axiom in one’s semantic theory: (1) permission can be defined in terms of obligation, and (2) obligation implies permission. As noted above, (1) is an independently plausible assumption, and, specifically in Schroeder’s framework, it is difficult to find a way to reject it. Hence, one should either maintain that (2) obligation implies permission is not formally valid or admit that moral dilemmas are impossible as a matter of logic and semantics alone. As I will explain below, this also tracks how Schroeder himself pursues this matter.

At one point in the book, Schroeder (Ch. 5, Sec. 4) discusses a possible treatment of obligation implies permission in his framework, in connection with the issue of the logical possibility of moral dilemmas. There, Schroeder simply registers the fact that there are theorists who believe that moral dilemmas should be logically impossible, and he goes on to explore whether his semantics could accommodate such a claim. Unlike me, Schroeder does not make any assumption concerning whether or not expressivists in general should aim to secure the logical possibility of moral dilemmas in their semantics. (In the next subsection, I will explain the ramifications of the divergence in stance here.) His aim is, rather, to show that his semantic framework is compatible with either of the opposing views on this issue: it “can remain neutral on this question [about the logical (im)possibility of moral dilemmas], offering ways for those who like either result to capture their views” (74-5).19 Specifically, Schroeder maintains that there is a way to “supplement our system with an auxiliary assumption that will yield the result that ‘murdering is wrong’ [i.e., ‘not murdering is obligatory’] and ‘not murdering is wrong’ [i.e, ‘murdering is obligatory’] turn out to be inconsistent” (72). So, on Schroeder’s view, one can either adopt or reject the “auxiliary assumption” in question to reflect one’s preferred view on the logical possibility of moral dilemmas. Below, let me introduce Schroeder’s “auxiliary assumption” and explain how it will make moral dilemmas logically impossible in the current framework.20

The auxiliary assumption in question states that “blaming for not murdering entails not blaming for murdering” (73). According to Schroeder, this validates obligation implies permission in the current framework. Later, I will question exactly what the auxiliary assumption is claiming in substance and how one might justify it—for now, let us simply confirm the technical point first. Notice that (given Table 1) “φ-ing is obligatory” and “φ-ing is permissible” express the following attitudes, respectively.

Table 2

The auxiliary assumption in question states that the following holds:

This, Schroeder claims, results in the following entailment relation between being for attitudes that captures obligation implies permission:

Strictly speaking, to move from (AA) to (OP), one would require an assumption that for any pair of being for attitudes, entailment relations that hold between the contents of the attitudes will reflect in the corresponding entailment relations between the attitudes (that is, FOR(α) entails FOR(β) if and only if α entails β). Schroeder does not seem to explicitly discuss this, but let us accept it as a generalization of Schroeder’s basic proposal that being for attitudes are inconsistency transmitting-attitudes—they are, recall, attitudes that are inconsistent with one another if and only if their contents are inconsistent. This is motivated by the general idea that being for attitudes are, like beliefs, attitudes such that their logical relationships (such as inconsistency) are reducible to the logical relationships that hold between the embedded contents. If one can assume this much, then the auxiliary assumption in question does capture obligation implies permission in Schroeder’s framework.

Now, let us confirm how this will rule out the possibility of moral dilemmas. A moral dilemma is, again, a situation where φ-ing and not φ-ing are both obligatory for an agent. Below, I will demonstrate that “φ-ing is obligatory and not φ-ing is obligatoy” and obligation implies permission are jointly inconsistent in Schroeder’s framework. The dilemma’s conjuncts, “φ-ing is obligatory” and “not φ-ing is obligatory”, express the following attitudes, respectively:

“φ-ing is obligatory” expresses FOR(blaming for not φ-ing).

“Not φ-ing is obligatory” expresses FOR(blaming for φ-ing).

Assigning the meaning for the dilemmatic statement, “φ-ing is obligatory and not φ-ing is obligatory”, requires introducing the definition for conjunction that Schroeder provides on page 66:

Accordingly, “φ-ing is obligatory and not φ-ing is obligatory” expresses the following attitude:

Now, it needs to be shown that (MD*) and (OP) are jointly inconsistent. To proceed from here, one only needs to assume that having the attitude of FOR(α∧β) is equivalent to having the attitudes of FOR(α) and FOR(β). If this can be assumed, the proof is obvious. Having the attitude (MD*) amounts to having the following pair of the attitudes, (1) FOR(blaming for not φ-ing) and (2) FOR(blaming for φ-ing). (1) and (OP) immediately yield FOR(not blaming for φ-ing). This is directly inconsistent with (2). Hence, (OP) and (MD*) are jointly inconsistent.

This should suffice to show Schroeder is right to claim that the auxiliary assumption in question rules out the logical possibility of moral dilemmas in his framework. According to Schroeder, one can then either adopt or drop the auxiliary assumption depending on one’s view on the logical possibility of moral dilemmas. Now, the question I want to pursue below is this. Can an expressivist really adopt or drop the auxiliary assumption that flexibly, as Schroeder seems to assume? A potential worry stems from the point that I set aside earlier. In claiming that blaming for not murdering entails not blaming for murdering, one actually seems to be making a substantive claim about blaming. That is, whether it is true or not seems to depend on what blame actually is, or how the notion of blame should be understood. If so, the auxiliary assumption is making a claim whose plausibility may need to be examined independently of one’s view on the logical possibility of moral dilemmas. This might mean that expressivists cannot in fact adopt or drop the auxiliary assumption as they like to deal with the logical possibility of moral dilemmas. (In the current paper, unlike Schroeder, I am assuming that expressivists generally need to side with the pro-dilemma view. So, the question I will focus on in the next subsection is this: can expressivists reject the auxiliary assumption freely, just because they need to invalidate obligation implies permission and make moral dilemmas logically possible?)

Schroeder, in fact, seems to recognize this sort of concern himself. In the same section, Schroeder points out that if one wishes to adopt the auxiliary assumption to make moral dilemmas logically impossible in the proposed semantic framework, one may need to justify it by maintaining that “it is [as a matter of conceptual necessity] impossible to both blame for murdering and blame for not murdering” (73). This, in effect, is to justify the auxiliary assumption by maintaining that it expresses a conceptual truth about blaming (as Schroeder puts it, a truth in the “logic of blaming” (73)) . In passing, however, Schroeder also notes that this may actually seem “a little too strong for plausibility” (74). Although Schroeder does not elaborate on this, certainly we may imagine someone who insists that one can consistently blame someone for not φ-ing and blame the same person for φ-ing. For example, in Sophie’s choice, Sophie is forced to choose only one child from the two, and, whichever child she ends up choosing, she might blame herself for not choosing the other. To take a more mundane example, one can imagine, say, a poor heavy smoker who will be blamed by her family and friends anyway regardless of whether she continues smoking or refrains from doing so.21 The existence of this kind of practice concerning blame can certainly cast doubt on the idea that the auxiliary assumption expresses a conceptual truth about blame.

Generalizing from this point, I think one can arrive at an important observation: the plausibility of the auxiliary assumption in question depends on one’s decision as to which act/attitude should go into the placeholder in the attitude of being for [some unary expression] φ-ing. Schroeder puts this point in this way:

The auxiliary assumption originally states: “blaming for not murdering entails not blaming for murdering” (73). If one decides that avoiding, instead of blaming, should go into the placeholder, the auxiliary assumption would have a different content, correspondingly: avoiding not murdering entails not avoiding murdering. Here, the latter might actually appear more plausible than the former, because it is highly unintuitive to think that an agent can avoid not φ-ing and avoid φ-ing at the same time—this looks similar to the case of intending φ-ing and intending not φ-ing at the same time, which seems impossible or at least deeply irrational. This in turn means that it can be highly controversial to reject this version of the auxiliary assumption (i.e., avoiding not φ-ing entails not avoiding φ-ing), because it appears to capture an independently plausible claim that follows from the “logic” of avoiding. On the other hand, if one instead decides that, say, disliking should be put into the placeholder, one would get disliking not φ-ing entails not disliking φ-ing as the corresponding auxiliary assumption. This may seem rather implausible—we can coherently imagine a universal hater who dislikes pretty much everything, including both φ-ing and not φ-ing. It would of course depend on how one understands the notion of disliking in question, but the point is that rejecting this version of the auxiliary assumption seems less controversial compared to the different version that one gets by putting avoiding in the placeholder.

Let me summarize the current point by connecting it to my exposition of Schroeder’s semantics in the previous section. To adequately deal with the negation problem, one’s expressivist semantic theory needs to meet the structural requirement that the attitude assigned as the meaning of “φ-ing is wrong” should take the following form: being for [some unary expression] φ-ing. Recall that this in itself does not call for any strong material restriction on which specific kind of expression should go into the placeholder. But now, there is at least one important consideration that one should take into account in making one’s decision here: the plausibility of the auxiliary assumption, which validates obligation implies permission, depends on which specific attitude/act gets plugged into the placeholder. I think this is an interesting result that one can extract from Schroeder’s discussion, which is worth pressing further than he actually does. Building on this point, in the next subsection I will explore how expressivists should ultimately deal with this issue and explain how this seemingly technical point actually exposes a more general problem for the expressivist semantic project.

3.2. Basic meaning assignment and its empirical implications

As I mentioned, Schroeder neither endorses nor rejects the auxiliary assumption himself—he merely presents it as an option that one can either adopt or reject, depending on one’s view on the logical possibility of moral dilemmas. Although Schroeder’s neutral stance is justified given the overall aim of his discussion, I think one can actually push this point further than Schroeder himself does to pose a general challenge for proponents of expressivism. One can do so by, for the sake of argument, sharing the assumption of the current paper that it is in fact a requirement for expressivists to secure the logical possibility of moral dilemmas. As I explained, this in turn requires (at least) invalidating obligation implies permission in one’s semantic theory. So, although Schroeder simply allows expressivists to either accept or reject the auxiliary assumption, one can advance the discussion further by assuming that they are actually committed to rejecting it.

This leads to an important question. From the discussion above, we know that the plausibility of the auxiliary assumption depends on which specific attitude/act gets plugged into the placeholder in being for [some unary expression] φ-ing, assigned as the meaning of “φ-ing is wrong (is not permissible)”. Now, can expressivists justifiably decide to put e.g., blaming instead of avoiding into the placeholder, solely on the basis of the fact that this would make it easier for them to reject the auxiliary assumption and thereby invalidate obligation implies permission? The answer I defend below is no. More specifically, I argue that their decision about which specific act/attitude must go into the placeholder should be criticized as problematically ad hoc, if it is motivated only by the need for securing the logical possibility of moral dilemmas.

To see why, notice first that the question of which act/attitude should go into the placeholder is a question that concerns the theory’s basic meaning/attitude assignment for an atomic sentence that contains its target normative expression: what is the non-cognitive attitude that ordinary speakers express by sincerely uttering, “φ-ing is wrong”, after all? Here, recall also that expressivists are committed to the mentalist assumption about the meanings of normative sentences (see Section 2): the non-cognitive attitudes assigned as the meanings of normative sentences are mental states of speakers who express them via their utterances. Therefore, in deciding what to put into the placeholder in being for [some unary expression] φ-ing, expressivists are making a substantive empirical-psychological claim about the mental states that underlie the use of “is wrong” in the actual linguistic practice.22 As such, naturally, their theoretical decisions need to be empirically well-motivated. Therefore, if securing the logical possibility of moral dilemmas in their semantic theory is the only reason for their decision in their basic meaning assignment for “is wrong”, it is problematically ad hoc; this is because it simply ignores other equally important, notably psychological, considerations that expressivists need to take into account in motivating their basic meaning assignment.

Let me demonstrate this point in more concrete terms. Their decision in the basic meaning assignment will, for example, yield predictions concerning what kind of behavioral patterns are generally compatible with one’s sincere utterance of “φ-ing is wrong”. The plausibility of their decision should be then tested by examining whether the predictions it yields fit with ordinary speakers’ actual behaviors as well as their intuitions on this matter. Suppose that one’s expressivist semantic theory tells us that, as Schroeder supposes, a speaker’s sincere utterance of “φ-ing is wrong” expresses the attitude of being for blaming for φ-ing. This proposal has an implication that a speaker who sincerely utters this sentence must be generally disposed to sanction actual instances of φ-ing performed by others. Surely, one who possess the attitude of being for blaming for φ-ing must feel compelled to actually blame others’ performances of φ-ing, at least when circumstances permit. Likewise, people should be likely to find puzzling a situation where a speaker sincerely utters “φ-ing is wrong” but never cares to blame observed instances of φ-ing at all. The question is: does this in fact capture ordinary speakers’ behavior and their intuitions on this matter? If yes, putting blaming into the placeholder is empirically well-motivated—if no, the choice may need to be reconsidered.

Now, contrast this to an alternative proposal, which assigns the attitude of FOR( avoiding φ-ing) as the meaning of “φ-ing is wrong”.23 This account now yields a different prediction concerning how ordinary speakers would react to the kind of situation described above. This is because avoiding φ-ing may be, unlike blaming, a matter of making personal plans for oneself, which may not necessarily concern whether one would also publicly sanction others’ performances of φ-ing. If this is so, my having the attitude of FOR(avoiding meat-eating), for example, might simply mean my being committed to avoid eating meat myself (and, perhaps, vaguely hope others do the same). This attitude, unlike FOR(blaming for meat-eating), need not imply that I am committed to socially sanction those who do not act as I do. Accordingly, even if I am known for overtly asserting “Meat consumption is wrong”, my not taking any corrective actions toward those who continue to consume meat need not appear so puzzling on the alternative proposal. Again, the question is: does this actually fit with ordinary speakers’ behavior and intuitions?

This quick comparison between the two choices above should suffice to illustrate how one’s decision about which attitude/act should be put into the placeholder needs to be motivated by general empirical-psychological considerations concerning the use of “is wrong” in ordinary speakers’ linguistic practice. Whatever decisions they end up making in their basic meaning assignment for “is wrong”, the mentalist assumption of the expressivist semantic project implies that they are also making claims about individual speakers’ psychology. As such, they yield various predictions that need to be tested empirically in light of the data. What I discussed above is just one example, and I suspect that there are also other types of similar considerations that one should take into account (e.g., to what extent do ordinary speakers take a speaker’s sincere utterance of “φ-ing is wrong” to be compatible with her emotional neutral reactions to instances of φ-ing?). It is for this reason that, in deciding which specific act/attitude should go into the placeholder in being for [some unary expression] φ-ing, expressivists cannot simply insist that they are justified to choose whatever act/attitude that would invalidate obligation implies permission and make moral dilemmas logically possible.

It might be helpful to elaborate on the current point by connecting it to the fact that when Schroeder chooses to specifically assign FOR(blaming for φ-ing) as the meaning of “φ-ing is wrong”, he purports to be following the proposal by Gibbard (Schroeder 2008a, 58). For Gibbard, “to call something rational is to express one’s acceptance of norms that permit it” (Gibbard 1990, 7); and, accordingly, “φ-ing is irrational” (which should be read as “φ-ing is wrong”, given his overall picture) would express a state of accepting a norm that forbids φ-ing, which looks similar to being for blaming for φ-ing.24 Again, Schroeder is not necessarily committed to this specific choice, and he draws on Gibbard just to “fix examples” (58). However, surely Gibbard himself should have some basic reasons and motivations (including considerations such as above) for analyzing “is wrong” ultimately in terms of blame/forbiddance, instead of avoidance, disapproval, disliking and so on.25 And this means that Gibbard (in his 1990 book) and others who adopt the notion of blame/forbiddance in analyzing the meaning of “is wrong” are prima facie committed to accepting whatever theoretical consequences that follow from “the logic of blame/forbiddance”. If it tells them that blaming for (forbidding) not φ-ing and blaming for (forbidding) φ-ing are inconsistent, they are prima facie committed to accepting its consequence in their semantic theory: obligation implies permission turns out to be formally valid, which in turn makes moral dilemmas logically impossible. My contention is that even if something like this turns out to be the case, they cannot easily switch to a different analysis that invokes e.g., the notion of avoidance instead of blame/forbiddance just for the purpose of blocking this result. Such a response should be criticized as problematically ad hoc. As I argued, securing the possibility of moral dilemmas is only one of the considerations that one should take into account in determining the basic meaning assignment for “is wrong” in one’s theory. If one hastily makes changes in the basic attitude assignment to deal with this particular technical problem, it is likely to produce unintended predictions in other places and even runs the risk of unintentionally abandoning whatever basic insights that motivated one’s theory in the first place.

The current point can be generalized. Different expressivist accounts invoke different basic notions in analyzing the meaning of “is wrong” (or whatever normative predicate or operator that they take to be basic, such as “is irrational”, “ought”, “is obligatory”, etc.). To take a few examples, Blackburn (1984, 1988) analyzes “φ-ing is wrong/impermissible” in terms of booing/disapproving φ-ing; Gibbard (2003) analyzes “φ-ing is the thing to do” in terms of a state of planning to φ, Horgan and Timmons (2006) analyze “One ought to φ” in terms of an ought-commitment that one φ’s. Each of these different proposals should be motivated by some basic theoretical considerations that they take to be important, including observations of ordinary speakers’ behavior and intuitions concerning the use of the target expressions. Depending on which of these proposals one finds plausible and which act/attitude one thinks should be put in the placeholder in being for [some unary expression] φ-ing, different results will follow as to whether moral dilemmas are logically possible. Even if they do not like the result, modifying their basic attitude assignment just for the purpose of blocking it would be problematically ad hoc. I have demonstrated this point in some detail, focusing on an expressivist semantic account that invokes the notion of blaming in its basic meaning assignment. I believe that one can pose, mutatis mutandis, the same point for any kind of expressivist semantic theory.26

Before closing this section, let me address one potential objection. In the discussion above, I assumed that expressivists such as Blackburn and Gibbard have some independent theoretical reasons and motivations to stick with specific notions (such as disapproval, blame/forbiddance and so on) in their analyses of “is wrong”. One might find this assumption dubious and object that their choices are not really based on any substantive, let alone empirical, considerations, because their primary aim is merely to construct structurally adequate expressivist semantic accounts that can deal with the Frege-Geach problem (although, if Schroeder is correct, they do not succeed in achieving this aim either). For example, Blackburn does not provide any lengthy discussion to justify his choice—he merely notes in passing that he is following Ayer (Blackburn 1984, 167). So, one might say, expressivists are free to switch to whatever attitude/act that seems suitable for dealing with technical problems at hand (such as making moral dilemmas logically possible) and there is nothing ad hoc about this move.

My response to the objection would be that expressivists including Blackburn, Gibbard and others should have supported their choices by some non-trivial empirical-psychological considerations, even if they in fact did not do so. As I explained, when taken together with the mentalist assumption of the expressivist semantic project, one’s choice in the basic meaning assignment will yield various predictions that should be empirically tested, whether they like it or not. I demonstrated this point by comparing analyses that invoke different notions (such as blame, avoidance) in their basic meaning assignments. The discussion in this section, if successful, shows that expressivists cannot remain indifferent to this issue and simply maintain that whatever attitude/act will do as long as it allows them to deal with technical problems in their semantic theories. It has to be recognized that, in the expressivist semantic project, one’s decision in the basic meaning assignment always comes with psychological implications.