We deploy logical notions in moral thought as we do in non-moral thought. But how can a moral expressivist understand the meaning and role of logical constants in moral thinking? That is the question. Here I supply the framework for an answer. What is called ‘expressivism’ typically comprises a negative and a positive claim. The negative claim is that moral thought does not represent the existence of moral facts or states of affairs. This implies that moral judgments (which might be occurrent thoughts or settled dispositions) should not be understood as beliefs about moral facts or states of affairs, and true moral judgments are not true in virtue of moral facts or states of affairs. The positive claim is that moral judgments should be explained in terms of ‘sentiments’, such as pleasures, emotions or desires, which are intentional mental states with non-moral contents, and these are ‘expressed’ in moral judgements.1 We may assume a distinction between beliefs that have moral contents, and sentiments, such as pleasures, emotions and desires that have non-moral contents.2 Thus guilt and anger are not the kinds of sentiments on which expressivists should build, for such states have moral content—that I did something wrong or that someone has wronged me. Given this, we can describe what is called ‘the Frege-Geach problem’ like this: if moral judgments are interpreted in terms of beliefs that represent moral facts, then we have a good understanding of the meaning and role of logical constants in moral thinking, since they bind moral propositions that represent moral facts. That means that logically complex moral propositions represent logically complex moral facts in a way that is a function of their logical and non-logical constituents. Furthermore, on this account, moral propositions mean the same when embedded in complex moral propositions, such as disjunctive, conditional or belief contexts, as they do in non-complex or atomic contexts. In both cases they represent moral facts. There need to be such identities of meaning if the usual entailments (modus ponens, disjunctive syllogism, truth elimination etc.) are to be valid. It is this requirement that generates the most serious problems for expressivism. For, on that account, moral judgements are expressive of sentiments of pleasure, emotion or desire that have non-moral content. But, then, what would be the meaning of moral propositions when embedded in logically complex propositions? When the proposition [kicking dogs is wrong] occurs in a disjunction, in a conditional or in a belief context, it does not express a negative sentiment towards kicking dogs. But in the atomic case, it does. At very least we are short of an account of its meaning in the complex cases. Furthermore, the point seems to be stronger than that since it seems that whatever conceivable expressivist account is given of the meaning of moral propositions, it cannot respect the constraint that the meaning is the same in and out of complex contexts. If so, there cannot be the usual entailments involving moral propositions, which are part of ordinary moral thinking. It is difficult to see how there can be unasserted moral propositions that mean the same as when asserted. This is the main thrust of what is known as ‘the Frege-Geach problem’, after Peter Geach who put the point forcefully and who invoked Frege in doing so (Geach 1965). This objection to expressivism has seemed to many philosophers, not just to be damaging, but to be fatal. Expressivists need to answer this challenge. This enterprise of answering this challenge lies broadly within what Simon Blackburn has called ‘quasi-realism’, which is the general project of showing how expressivism can explain and justify the surface conceptual features of our moral thought and talk that might tempt one to realism (Blackburn 1984, 1988, 1998; see also Gibbard 2003, 2012). Since logic is pervasive in moral thinking, an important part of the quasi-realist project is the attempt to give an account of the role of logical constants in moral thinking.3 However, I shall deploy resources other than those that Blackburn offers; and I shall not frame the issue quite as Blackburn frames it. Nevertheless, I think that he would or at least should be sympathetic with the way I describe the issue. The initial goal in this paper is to understand the Frege-Geach problem properly, and to consider exactly what is at issue, rather than to decide the issue one way or the other. We need a neutral view of what would count as success or failure for expressivism in explaining and justifying the role of logic in moral thinking. We need to do this before rushing to try to solve the problem. However, I shall not remain neutral. For once we have a proper conception of what is at stake, it will become clear that quite a few objections to expressivism arise from an improper understanding of the issue. Furthermore, a proper understanding of the issue should lead to a greater sympathy for the expressivist project, or at least an openness to seeing how far it can get with specific logical constants. I shall not pursue this in this paper, since we have enough to do to understand what such proposals would amount to. Any proposals need an interpretation, and it is the job of this paper to provide an acceptable interpretation.4 There is a move that need to be made, a matter of fundamental reorientation, without which the problem is insoluble. On this basis, and only on this basis, we can offer some concrete proposals. Because of the role of this fundamental reorientation, I shall give an expressivist account of embedding, without in the first instance embedding the account much in the literature on the Frege-Geach issue, since that literature does not take off from this reorientation. Later on, once we have something substantive under our belts, we will address some writers who differ in their approach, where we can learn from these differences. An understandable although misguided first thought would be that we should begin with some general principles concerning the nature of logic before turning to consider the expressivist treatment of logical constants in moral thinking. The thought would be that we should proceed this way because we need to know what has to be attained in order to account for the role of logical constants wherever they appear. Thus, we should lay down a general conception of logic and then draw on that conception when we turn to consider the role of logic in moral thought, for then we are dealing with an application of general principles to a particular case. This, as we shall see, is exactly the wrong approach, and this paper proceeds by dropping this priority assumption. Indeed, it deliberately flouts it. Consider the doctrine called ‘inferentialism’ about logic, according to which understanding logical constants is either being bound by certain norms for propositional attitude revision or perhaps being disposed to conform to those norms. Thus, to understand and or or is either to be subject to norms such that we should infer A from [A and B], and [A or B] from A, or perhaps it is to be disposed to infer in accordance with those norms of inference. In particular, each logical constant is supposed to have distinctive ‘introduction’ and ‘elimination’ rules that define what it is to be that logical constant. This is a popular view (Wittgenstein 1932-33, Gentzen 1935, Strawson 1952, Kneale 1956, Hacking 1979, Brandom 1994). However, any such view is subject to fundamental difficulties. One central objection to inferentialism is an objection from explanatory direction: such norms for propositional attitude revision hold in virtue of something; and in particular, they hold in part in virtue of the logical constant contents of the propositional attitudes in question. That is, we should infer in those ways because our propositional attitudes involve those logical constants: the logical constant contents explain the norms, not vice versa. Hence, inferentialism as a general doctrine about logical constants seems implausible (see further Zangwill 2015, 2021). If so, it seems that expressivists should not seek to explain the logical constants that figure in moral propositions in terms of inferential norms. For if there are norms bearing on complex moral thoughts, thoughts with logical contents, they derive from the logical components of complex moral propositions. It is at this point that expressivists need to make a decisive move, one that opens up a distinctive approach to understanding how logical constants figure in expressive moral thought and discourse. Expressivists need to claim that inferentialism is right about moral thinking; there, the usual order of explanation is reversed. There are inferential norms in both non-moral and in moral thought and talk, but in moral thought and talk the direction of explanation that holds between logical constants and inference differs from the usual case of non-moral thinking. It is fundamental to the expressivist project to stand anti-inferentialism on its head. This move is present in an ironic way in Blackburn’s development of expressivism because Blackburn mistakenly embraces a general inferentialism about logic (Blackburn 1998). But by making a mistake about the general case, Blackburn hit on the right strategy in the moral case. Blackburn made a good mistake! Given this, we arrive at the right way to approach the Frege-Geach problem: the aim is to give an account of the norms of sentiment revision that simulate or mimic what they would be if moral judgements were beliefs about moral facts. This is what it is to generate moral logic and to have the basis from which to solve the Frege-Geach problem for expressivism.5 In moral thought, the logical particles are defined or explained by norms of propositional attitude revision unlike in the non-moral case, where it is the other way round. In the non-moral (that is, the non-normative) case, norms of belief revision are explained by the logical constants that figure in the belief contents; the norms hold partly in virtue of those logical constants. By contrast, for an expressivist, to operate with moral contents with logical constant constituents is to be bound by norms of inference—those corresponding to the ‘introduction’ and ‘elimination’ inferential rules for the logical constants in question. Those rules do not hold in virtue of the logical constants, as in non-moral cases. In morality, expressivism should flagrantly flout the direction-of-explanation point. The intuitive direction of explanation and justification is reversed. The idea pursued in this paper is to address the Frege-Geach problem by appealing to norms bearing on the sentiments that expressivists identify as central to making moral judgments, and then to work from those norms to an account of the logical constants that figure in the content of moral thoughts and discourse. Moreover, this is surely an essential part of expressivism. For, if we have not stood anti-inferentialism on its head in this way, then we bound to wonder how expressivism can possibly succeed given the category difference between propositional contents and norms bearing on propositional attitudes to these contents (Harman 1986, chapter 1). How can we work from one to the other? The Frege-Geach problem would be intractable. But the whole point is to give a different understanding of the grasp of logical constants such that the norms for inference mirror those in play in non-moral thought and talk. This simulation is what it is to construct logical constants, and also the moral propositional contents that they bind, for expressivist forms of thought and talk. What will be developed here should be distinguished from two kinds of approaches to the Frege-Geach problem that go under a similar name. First, a number of philosophers have declared themselves to be ‘moral inferentialists’ and claimed that this solves the Frege-Geach problem. (Examples are Chrisman 2015, Warren 2015, 2018, Woods 2017, Frapolli 2019.) This moral ‘inferentialism’ is supposed to contrast both with realism and expressivism. These philosophers often have a general sympathy to general inferentialist accounts of meaning, which typically claim that possessing some concept just is being bound by rules that that define it. Unfortunately, such general inferentialist accounts face serious problems as well as being intuitively implausible. One central objection—to put it briefly—is that the rules that are held to be definitive of entertaining a certain content are norms for inference. But then there is the question: in virtue of what do those norms hold? The answer to that question, which must lie beyond inferentialism, will give substance to the inferentialist claim, which was not there before. Without answering the in-virtue-of-what? question, the theory is not yet properly formed, leaving too much open. But once it is answered, the contrast with realism and expressivism dissolves. For realism and expressivism are naturally construed precisely as giving different accounts of why the rules of inference hold. There are distinctively realist and expressivist explanations. Furthermore, if the norms for inference were held to be irreducible then we would have no theory, since either norms are not grounded in anything, or else they are explained in terms of other norms, and we are back where we started. This general ‘moral inferentialism’ is not a happy option. This is to say nothing of doubts that are bound to surface concerning specific proposals as to what these rules might be. Second, Luca Incurvati and Julian Schloder have pursued a program that they also advertise as a kind of expressivist ‘inferentialism (Incurvati and Schloder forthcoming. But they pursue inferentialist expressivism about the logical constants, quite generally, not just as they figure in normative thinking. So, their project is also quite different from what is envisaged here under a similar label. Thus, the first kind of theory pursues inferentialism about all moral content, not just about logical constants as they figure in moral thinking, while the second pursues expressivism of an inferential sort about all logical constants, again, not just about the logical constants as they figure in moral thinking. The project pursued here is more delimited; it is an inferentialist account of logical constants only as they figure in moral (or other normative) contexts. This paper is a prolegomenon to the theoretical reconstruction of norms for specific logical constants and specific implication relations. Such specific proposals I pursue in another paper in progress. This paper aims to describe what such proposals would be assuming, and what they would be attempting to achieve. In both moral and non-moral thought with logical constant constituents we can ask: in virtue of what do norms of propositional attitude revision hold? In the non-moral case, the norms of belief revision hold in virtue of the logical constant contents of the beliefs in question plus the nature of the propositional attitude of believing; that is, they hold partly in virtue of being thoughts with certain logical constant contents and partly in virtue of being beliefs. But for expressivism, our understanding of logical constants depends on norms of sentiment revision (inckuding stability). Furthermore, expressivists will have views about why norms of sentiment revision hold. They will say that they hold in virtue of the purposes of moral thought—the purposes being such that we need our system of sentiments to be complex and to be structured in certain ways, pre-eminently with certain consistency requirements, if it is to serve its purposes, socially and individually. These purposes, which are extrinsic to the sentiments themselves, explain the norms of revision, adherence to which just is our grasp of logical constants in moral thought. To appreciate the power of the inferentialist move, consider the weakness of certain objections to expressivism. In the 1980s, Bob Hale and Crispin Wright complained that Blackburn’s account captures only the moral necessity of drawing inferences, not the logical necessity of doing so (Hale 1986, Wright 1988). But given the way we have set up the issue, we can see that this objection is completely ineffective, because the whole point of the inferentialist development of expressivism is to simulate the inferential norms arising from logic that hold among non-moral beliefs (Zangwill 1992). When the similarity is close enough, that is, when there is enough of a structural isomorphism of inferential norms to inferential norms, that just is success for the expressivist. Asserting a distinction between moral and logical necessity thus begs the question against appealing to constraints on sentiment combination in order to mimic the norms at work in the non-moral case that derive from the logical constituents of propositions. That is, the expressivist project is to try to explain the existence of norms of sentiment revision, in terms of consistency and inconsistency between sentiments, so that they are similar to, or run parallel to, or simulate, the norms of belief revision binding non-moral beliefs with logical constant contents. Perhaps there are difficulties of detail in the way of the account. But the entire project cannot be ruled out of court by saying that even a perfect simulation would not suffice for real and true and genuine logical necessity. At very least, the onus lies with such opponents of expressivism to show that ordinary moral thought contains more than the simulation. Even to admit that it is hard to tell whether or not ordinary thought contains real and true and genuine logical constants or ‘merely’ simulations constructed from norms binding sentiments is in effect to concede victory to expressivism. A much stronger and more interesting objection is this: does not the account, or its main presuppositions, imply that the meaning of logical constant concepts and words differs between moral and non-moral cases? An initial reply to this objection is to say that the difference in what the logical constants are, or what logical constant concepts or terms mean, need not be available to those who engage in the conceptual practice. It is a standard expressivist move to insist that what they are aiming at is not conceptual analysis but an explanatory enterprise; and that those who engage in moral thinking and apply moral predicates need not know the expressivist explanation of their thinking (for instance, Blackburn 1984: 189 and 1993: 152-53). Expressivism is, as it were, the ‘deep structure’ that explains the superficial conceptual nature of the form of thought that ordinary thinkers are aware of. Those who wield the logical constants in their thought know the rules of inference—the introduction and elimination rules that govern their thought. But they need not be aware of the deep explanation of their acceptance of those rules, and of the deep source of their right to deploy them. Thus, it seems that there can be a topic-neutral conception of logical constants on the expressivist account. What about mixed propositions? Since there is an overlap of meaning between moral and non-moral conceptions of logical concepts, it is difficult to see a problem here. Grasping the rules of inference for the logical constants should enable a thinker to pass freely and seamlessly between moral and non-moral propositions, and hybrid combinations of these, without any sense of changing gear and swapping to logical constants with new meanings. So, it seems that conjunctions or disjunctions of moral and non-moral clauses do not present a difficulty. While it is true that the deep explanation and justification of the norms defining good practice with the logical constant varies between moral and non-moral thought, the superficial meaning, that ordinary thinkers grasp, and that conceptual analysis might aspire to reveal, is blind to this explanatory difference. Ordinary thinkers are blind to the deep differences that do not surface in their superficial conceptual practice: there the norms suffice. However, a follow-up objection, at this point, might be: even if the meaning of logical constants concepts does not differ between moral and non-moral contexts, did we not say that the direction of explanation from logical constants to norms is intuitive in the ordinary non-moral case? Surely, then, it should be intuitive in the moral case too. But we said that in the moral case, the logical constants are just bundles of norms, which means that the intuitive direction of explanation is lacking in the moral case. I think that this objection can only be met by conceding that, in the moral case, the intuitive direction of explanation is illusory. The expressivist should concede that this is a cost, but not a large one. People might be right to deploy logical notions and they might be right about the norms of modification in moral thought that they generate. It is just that on those occasions when people are reflective, they are wrong in their self-understanding of why those norms hold. However, most of the time, in ordinary non-moral cases, it is enough to grasp the concepts and embrace the norms, and further, more rarified matters of explanatory priority, are not pressing. Such matters are important to us philosophers, but ordinary thinkers just need the concepts and norms in order to get on with moral thinking, not its explanation and justification. Inferentialism, then, is the basis from which an expressivist can launch explanations of logical constants that figure in moral thought. Without inferentialism, they are not explainable. By itself, the inferentialist approach does not remove technical difficulties with formulizing the precise inferential norms for the conditional, conjunction, disjunction, negation, or whatever other logical constants there are. (For discussion see Unwin 1999, Hale 2002.) However, it is clear that the feasibility of any technical proposals depends on the inferentialist view, and they assume some explanatory account of the source of the norms of inference in moral thinking—one that does not lie in the logical constants themselves. This gives us an understanding of what is, or ought to be, at stake when expressivists and anti-expressivists debate the success of specific proposals for accounting for logical constants as they figure in moral thinking. It is not that there are no technical difficulties to be addressed—just that the debate over the correct understanding of particular logical constants in moral thought and discourse, expressivistically interpreted, depends on the prior inferentialist view, and without it, the technical proposals are beside the point. Indeed, those inferential norms are what are codified in the technical proposals. Hence those who think of the issue as primarily a technical problem are misguided.6 A technical proposal without an interpretation—a proper psychological and explanatory basis—is, as I once heard Bernard Williams say, “…like a torch without a battery”. (See also Kripke 1976: 419, and Zangwill 2022.)7 This simulation goal can be put into relief if we compare it with the way Gibbard describes the issue (at Gibbard 2012: 273-74), in terms of an explanation of moral content in terms of ‘states of mind’. Gibbard wants to generalize the expressivist program to all contents. He wants to pursue a global program. And he thinks that the expressivist explanation of logical constants is not worse off than the usual ‘Fregean’ explanation (see especially Gibbard 2012, appendix 2). But this project is problematic since the expressivist construction of moral content takes off from certain sentiments with (non-normative) contents and on that basis aims to construct a richer (normative) content. Therefore, some content cannot be expressivistically constructed. It seems that expressivism cannot be generalized across the board as Gibbard envisages, and it can only succeed as a local program. The trouble is that according to what we might call ‘Fregean’ accounts, logical constants figure in propositional contents; and that explains implication relations among those propositional contents (Harman 1986, chapter 2). Crucially, none of that draws on any kind of propositional attitudes with those contents. The propositional attitudes could be wishes, hopes, imaginative acts, wonderings, beliefs or sentiments. It makes no difference given that implications concern the contents of such propositional attitudes. In the non-moral case, the logical contents that generate implications have nothing to do with propositional attitudes to those contents; they concern contents—contents with logical constant constituents. Expressivist logic is a special case, where a system of sentiments mimics the inferential norms for beliefs that derive in part from the propositional contents of beliefs, and in particular from their logical constant contents together with the kind of propositional attitudes that they are. To put it another way: there are implication relations among propositional contents, which hold in virtue of the logical content constituents of those contents; and there are inferential norms for beliefs, which derive in part from the implication relations among belief contents, and in part from the nature of belief. By contrast, there are the normative relations among sentiments, which mimic the inferential norms among beliefs. The ‘Fregean’ basis is assumed by expressivism and cannot be expressivistically explained. Some logic must be non-expressivist. Before we move on to address notions of consistency and inconsistency among sentiments, it is worth remarking here that the simulation approach makes it obvious that quite a few common ways of characterizing the overall issue are flawed. In particular, it is common to say that the opposition needed to characterize expressivism, and frame the Frege-Geach issue, is between kinds of propositional attitudes: beliefs by contrast with desires or pleasures—kinds of states with different so-called ‘states of fit’ (Zangwill 1998). But this cannot be right. For if there are beliefs with moral representational content, there are very likely also to be moral desires and emotions with moral propositional content (see Zangwill 2003). In fact, the kind of propositional attitude is not the issue, as is assumed when the issue is set up solely in terms of an opposition between cognitive vs. non-cognitive mental states. (The issue was set up in section 1 of this paper in terms of propositional attitude kinds and contents.) Instead, we should say that the expressivist is someone who explains moral thought and feeling in terms of propositional attitudes without moral propositional contents. (I have appealed to “sentiments”.) The kinds of propositional attitudes that expressivists appeal to, whatever they are, have non-moral propositional contents, and they get expressed in moral judgments or moral utterances. Thus, the standard label “non-cognitivist” makes a mess—since moral realists also invoke moral non-cognitive states, such as desires and emotions with realistic moral propositional contents. What is important is that if some beliefs have moral propositional contents, then what explains the norms of propositional attitude modification, when we reason logically, is the propositional contents, with their logical constituents, plus the nature of belief. By contrast, what explains the norms of propositional attitude modification, for expressivism, are the norms binding various sentiments with non-moral contents, together with the overall function of the practice of making moral judgements. It remains true that for one theory, moral judgments are beliefs (about moral facts), whereas for the other, they are certain kinds of sentiments. But all sides should admit that there is more to moral thought than moral judgments. In order to develop this line of thought, and to start to put some flesh on the abstract bones, I shall now explore the idea that the consistencies and inconsistencies that we seek on behalf of moral expressivism are to be understood as rational consistencies and inconsistencies. (The development will be somewhat unHumean, a point to which I shall return.) Two rational norms are these: being pleased at x and also being pleased at not-x are rationally inconsistent; so are being displeased by being pleased at x and also being pleased at x. Why? To explain this, a first step for the expressivist is to follow Gibbard’s development of expressivism in which sentiments are internally connected not just with motivations but also with intentions and plans (Gibbard 2003). It then appears that favorable and unfavorable sentiments to the same thing are rationally inconsistent because it is impossible to do both x and not do x at the same time. So, it is irrational to strive to do both. Hence positive and negative sentiment to x are rationally inconsistent given that it is also rational to intend to do x, and to strive to do x, if we have a positive sentiment to x, at least in those cases where doing x is in our control (Stampe 1987). Thus, one kind of rational consistency and inconsistency between sentiments lies in the possibility or impossibility of joint realizability of the contents of sentiments. Favorable and unfavorable sentiments to the same thing are ‘inconsistent’ sentiments in this sense, given an internal relation between sentiments and desires. Such a pair of sentiments 'clash'. Compare belief. Suppose I believe that elephants have big ears. Suppose someone then asks me what I think about the view that elephants do not have big ears, and I say, “Oh I agree with that too”. Or suppose I am completely neutral about it. That’s irrational. More than that, it’s stupid. Sentimen