Abstract. Noncognitivists must account for the menagerie of moral attitudes of which we are psychologically capable. This paper offers a systematic explanation of these attitudes by means of a recipe semantics. Unlike extant noncognitivist theories, this proposal does not aim to justify the behavior of moral attitudes in terms of any underlying cognitive function. The recipe semantics allows us to characterize them in terms of the distinctive functions that their constituent components play in other contexts, while admitting that they may have no particular function in many. In order to make this more palatable, the paper concludes by offering an account of the evolution of normative attitudes that would make their occasional non-functionality unsurprising.
Target Presentation by Derek Shiller
Normative noncognitivism is a negative theory: it holds that normative judgments are not beliefs. Turning this into a satisfying positive theory requires saying much more about what moral attitudes actually are. To do so, noncognitivists have traditionally looked to the motivational capacity of moral judgments. They have focused on providing an alternative characterization of straightforwardly predicative moral judgments such as insurance fraud is wrong, charitable donations are morally exemplary, ann selfishness is sometimes permissible. Such judgments are said not to be beliefs, despite their surface presentation, because they do not represent the world as being any particular way. Instead, they reflect something about how we are moved to act.
However, not all moral attitudes are straightforwardly predicative moral judgments. Questions over the nature of logically complex moral judgments have been much discussed as part of the Frege-Geach problem, but logical complexity is just one kind of complication found in our moral attitudes. An adequate positive theory would include a characterization of a range of other attitudes.
In this paper, I will sketch a strategy for characterizing a variety of moral attitudes. My approach will differ from extant noncognitivist views in that I will not aim to rationally vindicate the way that we employ moral attitudes. Instead, I aim only to provide an account of what the attitudes are and why we might have them. In short, I think that the diversity of moral attitudes results from an underlying syntactic commonality with ordinary propositional attitudes. Moral concepts enjoy the same syntactic freedoms as other concepts, even if they don’t have coherent semantic functions in all of the contexts in which they can occur.
I will start by cataloging some of the attitudinal complexities that must be dealt with. Second, I will present an account of our propositional attitudes that will play an important role in shaping how I propose we think about moral attitudes. Third, I will offer what I call a “recipe semantics” for characterizing the other moral attitudes. This approach categorizes attitudes in terms of the distinctive functions that their constituent components play in other contexts. I will conclude with an account of the cognitive evolution of normative judgments that makes a recipe semantics plausible.
1. A Menagerie of Moral Attitudes
Cognitivists hold that moral judgments, such as the sort expressed by (A), can be characterized as beliefs with a certain content.
A) It is wrong to collect trophies of endangered species.
The judgment expressed by (A) takes a proposition about the distribution of a certain property (wrongness) as its object. Cognitivists disagree about the nature of this property, but they agree that there are definite conditions that must be met for (A) to be true.
Noncognitivists, in contrast, hold that such judgments have no special propositional content. There is no property that we ascribe to acts in judging them wrong. Rather, in judging an act to be wrong, noncognitivists typically think that we adopt a certain conative attitude towards it. What makes these attitudes moral judgments is not their representational content, but their motivational force.
An account of straightforwardly predicative moral judgments won’t tell us much about the nature of moral judgments that involve logical complexity. Consider (B), which employs ‘wrong’ under the scope of a negation.
B) It is not wrong to collect trophies of extinct species.
(B) doesn’t express an attitude that is motivational in the same way as (A), so an account of predicative moral judgments will not carry directly over to negational judgments. We are not moved to act solely because we believe that something is not wrong.
This problem with negation is only the most basic form of logical complication. The same problem also arises for conjunction and disjunction, as in the attitude expressed by (C), and quantification as in the attitudes expressed by (D) and (E).
C) Human beings either have a special moral status or we should not be testing cosmetics on animals
D) Everything you did today was worse than the thing you did before it.
E) Most killings are deeply morally wrong.
Quantification hasn’t received the same attention as conjunction, disjunction, and negation, but it is equally problematic for noncognitivists. We can make quantificational moral judgments about an issue without making any straightforward predicative judgments about it.
In addition to ordinary quantification, moral judgments can also take plural and generic forms, such as are expressed by (F) and (G).
F) The wrongs that the colonists did to the natives were more varied and numerous than the wrongs that the natives did to the colonists.
G) Wrongs are a stain upon the moral character of decent people.
Geach was especially interested in conditionals that play an important role in moral arguments:
H) If lying is wrong, then getting your little brother to lie is wrong.
These conditionals can be given a truth-functional interpretation, however, this interpretation is widely thought to be inadequate. The indicative conditionals found in expressions of moral judgments like (H) are subject to the same considerations that motivate non-truth-conditional interpretations of indicative conditionals in other contexts.
In addition to purely logical complications, moral judgments can take the form of subjunctive conditionals, as expressed by (I), and can involve tense, as expressed by (J).
I) If it had been wrong to bring children into this world, we would not have done it.
J) Abortion is as wrong today as it was fifty years ago.
Noncognitivists must also account for the existence of mixed normative judgments. Mixed normative judgments, as expressed by (K) and (L), combine different flavors of normativity into a single attitude
K) You ought (epistemically) to know that what you did was wrong.
L) You should (rationally) keep track of what you shouldn’t (morally) do.
Finally, I (Forthcoming) have argued that noncognitivists owe an explanation of moral attitudes other than judgment. Not only can we judge that actions are wrong, but we can desire that they be (or not be) wrong. We can regret that we acted immorally, and hope that we chose the right thing to do. We can wonder whether utilitarianism or deontology is correct. We can find some moral principles intuitive and others unintuitive. We can imagine fictional scenarios where different moral principles hold. We can suppose novel or bizarre moral principles for the sake of arguments.
We must make sense of these other attitudes just as we must make sense of straightforwardly predicative moral judgments. An account of what it is to judge that an action is wrong will not tell us what it is to hope that or wonder whether we did the right thing. It makes no sense to take ordinary conative attitudes for granted for the purposes of arguments (though we can take it for granted that we have these conative attitudes).
The challenge facing noncognitivists is to provide a suitable account of all of these other moral attitudes. Ideally, noncognitivists would offer a systematic characterization of all of these different attitudes. After laying out some assumptions about the cognitive structure of attitudes, I will return to give a sketch of one systematic account.
2. Context and Meaning Determination
Since moral attitudes have so much in common with other propositional attitudes, any inquiry into the former ought to start with the latter. In this section, I will present one view about cognitive structure of ordinary propositional attitudes.
On the view that I will present, our propositional attitudes involve relations to cognitive representations whose representational contents are determined by the representational properties of their parts. Those parts can be individuated by means of non-semantic properties, and they receive their representational content from their designed function in certain contexts. I will describe each of these components in turn.
The Representational Theory of Mind
According to the Representational Theory of Mind (RTM), ordinary propositional attitudes involve relations to mental representations whose contents are propositions. What sort of attitude it is (belief, desire, etc.) will generally depend on the relation taken to the mental representation. To believe a proposition might be to be disposed to token representations of that proposition in a certain way and use those representations in deciding what to do with the goal of getting what one wants.
In order for this theory to go beyond merely unpacking our concepts of propositional attitudes, “mental representation” must be understood in a robust and realistic sense. We may discover that our minds do not employ anything that qualifies as a mental representation in this robust sense, even we can use propositions to individuate them (such as through rationalizing their behavioral consquences). In that case, RTM would be false.
Those who accept RTM owe an account of how mental structures can be associated with propositions. For simplicity, I will assume that a teleological theory is correct. Roughly put, mental representations get their representational contents by virtue of being designed, in some sense or other, to have the function of representing propositions. This theory shouldn’t differ from other plausible theories of cognitive representationality on points that are relevant to my argument, and so readers may substitute their own preferred psychosemantic theory wherever appropriate.
Representations can be either structured or unstructured. Structured representations are composed of constituents that themselves have representational contents. Natural language sentences are structured, since many of the individual phrases and the words that constitute them represent properties, entities, or propositions. Maps are also structured, insofar as they have sub-regions that themselves have representational content (Blumson 2012). I will assume that mental representations are compositional, and refer to the ultimate constituents of individual mental representations as “concept tokens”. The term “concept” will be reserved for types of concept tokens as individuated by their meaning-granting properties.
Structured representations are typically compositional, meaning that the representational contents of the parts help to determine the representational contents of the whole. The representational contents of concepts tokens can’t themselves determine how it is that they, as neurophysiological entities, can and cannot be combined into a whole, even if they will have a say in whether the combinations they enter into are meaningful. Semantic properties simply aren’t the right kind of thing to determine that, any more than the representational content of LEGO head explains why it fits snuggly onto a LEGO torso and not a LEGO horse.
Non-semantic properties must enable some concepts and not others to be strung together in the mind. These properties, whatever they are, are ‘syntactic’ properties. As a result, there may be some syntactically viable combinations of concept tokens that are semantically incoherent – viable combinations that can be assembled without contributing coherently to an overarching representational content for the whole.
By supposition, the representational content of a mental representation depends on its designed function. The designed function of a representation may result from the designed function of its constituent concepts. A representation may be designed for a particular representational function by virtue of the fact that its constituent concepts were designed to be used to build representations. This is analogous to the situation in natural language semantics, where the meaning of a sentence is determined by the meanings of its words, and the meanings of words are in turn determined by the kinds of contributions they make to the meanings of sentences.
By supposition, concept tokens get meanings from their designed functions. Designed functions depend upon behavior in some range of contexts, and so it is necessary that concept tokens have identities across contexts to ground their representational content. Consider the challenge of assigning contents to concepts without the means to recognize the same concepts in different contexts. Quine’s radical interpreters had the phonological properties of ‘gavagai’ to re-identify the natives’ word at different times and with different speakers. If the word had sounded arbitrarily different each use, their task would have been impossible.
The same goes for cognitive semantics. It would be difficult to assign representational contents to parts of an attitude without being able to re-identify them in different attitudes. The difficulty is not merely epistemic; a concept token must have a non-semantic identity in order to support a robust enough pattern of usage to count as having a meaning under any reasonable psychosemantics.
Since it is used to assign representational contents, the property whereby a concept token may be identified from context to context must be non-representational. Given the analogy with written language, I will refer to such a property, whatever it is, as an ‘orthographic’ property. Orthographic properties might be closely associated with syntactic properties, but they need not be the same.
Our concepts may or may not have a designed function in every context in which their syntactic properties allow them to occur. In biology, phenotypical traits often outstrip designed function. Natural selection exerts imprecise forces; adaptations that are helpful in one context may have effects elsewhere. We have a variety of vestigial and byproduct physiological traits. We should expect this to be as true for cognition as it is physiology. I will refer to contexts in which a concept type is tokened in which it has no designed function as a ‘spandrel’ context.
Suppose that that we identify the concept horse as a concept whose designed function is to help track the properties of horses by figuring into representations of horses. It isn’t part of the designed function of the concept horse to help distract us when we are bored, but its tokens can be employed for this purpose. This use need not count against the concept’s designed function, as the concept may not have been designed not to be used as a distraction. Instead, its function may be given by what it is supposed to contribute in the context of certain kinds of attitudes (beliefs and desires), and it may have no functions in others, such as daydreams. If this is the case, then these other contexts are spandrel contexts for the concept. The syntactic properties of concept tokens allow them to appear in contexts which play no role in their meaning determination.
3. Recipe Semantics
In the last section, I put forward a view on which propositional attitudes involve mental representations. I suggested that individual mental representations can be identified across contexts by virtue of their orthographic properties. My proposal for handling the other moral attitudes will depend on extending this approach to moral attitudes. I will start by generalizing the notion of mental representations into a notion of a class of mental entities that may lack representational properties. Then I will provide a recipe semantics for daydreams to serve as an analogue for the view I hold about moral attitudes. Finally, I will combine the two ideas to provide a recipe semantics of non-straightforwardly predicational moral attitudes.
The orthographic properties of non-mental representations can be divided up into resemblance classes. Written words, for instance, are individuated by their shapes. They look distinct from each other, but they share more in common with each other than they do with spoken words. Written English sentences also share commonalities, across different typefaces and handwriting styles, that they lack with Arabic or Greek sentences. An ‘orthographic class’ is a set of entities whose orthographic properties resemble each other, and that includes a system of representations as a subclass. There is an orthographic class for sentences, English sentences, and English sentences in 12pt Times New Roman. There are other classes for maps, tables, and graphs.
Members of an orthographic class are members of that class solely by virtue of the similarities among the non-semantic properties that they share with other members. There is no reason to think that each member of an orthographic class must be representational. There are sequences of letters, sounds, and collections of squiggles on a two dimensional field that have no representational significance. Take for instance, the final stanza of Lewis Caroll’s Jabberwocky:
’Twas brillig, and the slithy toves
Did gyre and gimble in the wabe:
All mimsy were the borogoves,
And the mome raths outgrabe.
This belongs to orthographic classes that also contain natural language sentences in English, but since many of the words have no established meaning, they do not represent anything.
A ‘presentation’ is a member of an orthographic class. Representations are presentations, but not all presentations are representations. The stanza from Jabberwocky contains presentations which not representations. Since there are kinds of meaning that are nonrepresentational, there may be some meaningful natural language presentations that are not representational and have no representational significance.
If the mind utilizes representations to have propositional attitudes, it may also utilize non-representational presentations in other attitudes. Non-representational mental presentations are conceptually coherent. They are simply presentations that fail to meet a contingently-satisfied condition for representationality. According to a teleological psychosemantics, any presentation that was not designed to function as a representation will count as non-representational.
The Presentational Theory of Mind
According to RTM, propositional attitudes are relations to mental representations. If beliefs involve a relation to mental representations, then it is not far-fetched that moral judgments should involve relations to mental presentations. Beliefs and moral judgments are similar in a number of ways (Horgan and Timmons 2006). It would be more surprising if they should operate fundamentally differently at the cognitive level.
If beliefs and moral judgments involve relations to presentations, then we can locate the difference between beliefs and moral judgments either in the kinds of presentations they employ, the relations they involve to those presentations, or both. I will take it for granted that the primary difference between ordinary beliefs and ordinary desires is the former. To believe something and to desire it involve taking different relations to the same (or related) representations.
Now to my proposal: straightforwardly predicative moral judgments are attitudes that involve relations to complex structured presentations that contain non-representational concepts whose proper function is not to represent but instead to influence action. These concepts are moral concepts, and they are applied in just the same way as their representational analogues. Since the designed function of neither the whole presentation nor its constituent moral concepts is to represent, neither is representational. But since the designed function conveyed to presentations by the presence of the moral concept is to direct action, they are also not without a designed function altogether.
We can proceed to characterize the designed function of moral concepts in a variety of different ways, so this proposal does not tell us exactly what sorts of attitudes straightforward moral predicates are. We haven’t said anything about the important question of how to distinguish moral judgments from other normative judgments, or indeed from other motivational attitudes. However, extant proposals can be carried over rather straightforwardly to the present context.
A Recipe Theory of Daydreams
Philosophers of mind have traditionally focused their analytical projects on attitudes, such as belief and desire, that play an important role in influencing our behavior. Not all of our attitudes are like this. Daydreams, for instance, are not. This means that many strategies applied to characterizing propositional attitudes cannot be fully generalized. Contra functionalism, we cannot make sense of what it is to daydream about a certain situation in terms of what that daydream does in our mental life. In order to understand these states, we may need to approach them as byproducts of other cognitive faculties.
I propose that daydreams, and in particular their representational contents, ought to be understood in terms of their parts, and that those parts in turn are to be characterized in terms of how they are used elsewhere. We can’t recognize a concept token as representing a horse, for instance, solely in terms of its deployment in daydreams. Instead, we must look to how the same concept token is used in other cognitive contexts, such as perceiving horses, having beliefs about horses, and organizing intentions regarding our behavior toward horses.
A dream gets its content by employing concept tokens that play an important role in other contexts. These concept tokens are recognized by their orthographic properties. A concept token of a horse, as it occurs in a dream, represents a horse only insofar as it was designed to have a certain representational content in other contexts. The fact that we are dreaming of a horse, and not a teapot or a fermion, depends on the fact that the same orthographic representation has the function of keeping track of the properties of horses on other occasions.
A Recipe Semantics for Moral Attitudes
In the previous sections, I suggested that straightforwardly predicative moral judgments involve a special kind of mental presentation. In the present, I will explain how to extend this characterization to handle more complex moral attitudes.
Moral concepts are characterized by a non-representational proper function that operates in the context of straightforwardly predicative moral judgments. In order for a concept token to be an instance of a moral concept, it must be designed to be used in a certain way in certain contexts. Given that a concept token can be recognized in a variety of contexts based on its orthographic properties, and given that it need not have a designed function in all of the contexts in which it occurs, it is easy to systematically extend this account to handle each member of the menagerie of other attitudes.
The general strategy is to characterize attitudes in terms of a recipe for constructing them (a “recipe semantics”). The recipe will describe the ingredients that must be combined to produce the attitude. The ingredients may be characterized by their designed function in other contexts. There need not be anything specially distinctive about these attitudes beyond the parts from which they are made. They may lack any distinctive function or any important role in cognition.
By following this approach, we avoid the need to describe the function of these other attitudes. This makes this approach very different from traditional approaches to the problem. Unlike Simon Blackburn and Allan Gibbard, I do not think we need to vindicate the forms our attitudes take. We do not need to explain why it is that an attitude behaves as it does in terms of some essential function. Some of our attitudes may not play any important roles in our psychologies and so not every aspect of human psychology needs to have an intelligible rationale. The moral attitudes that make up the menagerie in the first section may be spandrel contexts for moral concepts.
On my proposal, complex moral attitudes are like daydreams. Daydreams get their particular contents because they involve representations whose characteristic function is restricted to other contexts. Complex moral judgments will often involve concepts whose function is specific to other contexts.
Negational moral judgments are the sorts of judgments that we typically express with negations of moral predicates. They have the appearance of logical inconsistency with straightforwardly predicative moral judgments. The judgment that eugenics is not wrong is an example of a negational moral judgment. It appears to be inconsistent with the judgment that eugenics is wrong.
It has been common to characterize negational attitudes by their functional roles; they are states that are inconsistent, in some way or other, with straightforwardly predicative moral judgments (Blackburn 1988). And, according to the traditional story, the incompatibility arises from the attitude’s functional role. For instance, the functional role of both straightforwardly predicative and negational moral judgments might be to direct us to realize a certain state of affairs. If incompatible attitudes are directed to realizing incompatible states of affairs, then they cannot be jointly satisfied, and might therefore be deemed inconsistent.
I am not opposed to ascribing straightforward predicative moral judgments and their negations inconsistent functional roles, but I believe that noncognitivists should not feel obliged to do so. Instead, noncognitivists can simply state that negational moral attitudes involve presentations with a special sort of negation concept.
In order to make sense of the appearance of logical inconsistency among presentations, it is helpful to attribute them some structure (Baker and Woods 2015). This can be done through the presence of a distinctive concept in the cognitive presentation. A negation concept is a concept that has a certain function in representations. That doesn’t mean it can’t occur elsewhere.
The characteristic function of a negation concept may be transformational: a representation governed by a negation concept represents a proposition with the opposite truth value of the proposition represented by its negated content. This designed function of the concept is specific to representations. The concept cannot have this function in all contexts. In the context of a moral judgment, the negation concept may have no more function than in: “twas not brillig”. It simply adds another component to the presentation. In those other contexts, the existence of the negation concept may influence the behavior of the attitude in our cognitive lives. It could produce the appearance of inconsistency and lead us to treat some attitudes as disagreeing with others. But this behavioral influence isn’t a matter of that concept’s function in that context.
A negational moral judgment involves a presentation that includes a negation concept as a constituent. Though the negation concept is designed to play a certain function in the case of representations, it need not play that specific function in the case of other presentations and so it may have no designed function whatsoever in the context of moral presentations. What identifies the moral judgment as a negational moral judgment is simply the presence of a constituent which has a certain function in other contexts.
Generics pose a difficult problem for linguists. On the one hand, generics seem to behave like ordinary quantifiers such as ‘all’ or ‘most’, on the other hand, they defy straightforward analyses in terms of quantities.
One of the most promising approaches to understanding generics treats them as a human cognitive default form of generalization. The proper applications of generics are guided by purposes of this default generalization, which serves to keep track of important properties of a kind without requiring any specific frequency among that kind. Where it is advantageous to associate the kind with that property, the generic is properly applied.
We may assume that generics are used to express mental states with the default mode of generalization, and that such modes are orthographically distinctive. Perhaps we employ particular generic quantificational concept tokens, but for something as basic as generic generalizations, the orthographic property may come from the structure of the presentation rather than from a proper constituent of it. Either way, there is something non-semantic that is distinctive about the form that generic judgments take. This distinctive property can be identified with generic quantification by virtue of designed function in certain representations.
The existence of moral generics can be explained by the fact that the same distinctive orthographic traits that underlie generic judgments can be found among presentations in combination with moral concepts. Generic concepts or generic structures that serve a function when combined with representational concepts might also be syntactically applicable to non-representational concepts. In the latter cases, they need not have any distinctive designed function. They can instead be categorized by their distinctive function in making default generalizations in other contexts.
4. Belief, Exaptation, and Moral Judgments
The value of this recipe semantics depends on the plausibility of concepts appearing in spandrel contexts. It is biologically and conceptually coherent for concepts that have a function in one context to also appear in contexts where they lack a function, but it may still be a defect of a theory if it is forced to say this.
The reason that we are able to combine concept tokens in the way that we do depends ultimately on their syntactic properties. The semantic properties, which depend partly on the historical or counterfactual behavior of the state, cannot themselves explain why it is that we are able to form particular representations.
Cognitivist and noncognitivists both must explain our capacity to have the attitudes we’re capable of having. Cognitivists face the challenge of explaining how we can form diverse beliefs. Jerry Fodor has acerbically argued for a solution invoking the structure rather than the content of our attitudes (Fodor and Pylyshyn 1988, Fodor 1998, 2008). Fodor explained how we have the attitudes that we do in terms of how they can be built out of recombinable units. There is little reason to think the challenge of accounting for our diverse attitudes (or its solution) will look much different noncognitivists than it does for cognitivists. The mechanisms by which cognitive units are combinable need have little to do with the properties that give these units their semantic values.
Nevertheless, it might be doubted that our moral concepts have syntactic properties that allow them to be combined into attitudes which have no designed function. Why is it that we’re able to hope that what we did was right, if the concept right doesn’t make any specific contribution to the attitude?
The evolutionary history of moral attitudes can make these syntactic properties unsurprising. What follows is one story of how this history may have gone. It isn’t the only viable story, but it goes some way toward warding off serious worries that arise from syntactic properties of moral concepts.
Coordinating Social Expectations
Our tendency to moral behavior is surely the product of evolutionary forces. Full-fledged moral attitudes play an important role in regulating our behavior, and the behaviors they produce contribute to our fitness. It is fitter for us to reciprocate acts of kindness and cruelty, and to act altruistically toward our kin. However, full-fledged moral attitudes are not necessary to produce this sort of behavior. Moral sentiments suffice. We could have evolved to feel grateful, vindictive, protective, and magnanimous without having anything that looked quite like moral judgments.
Moral judgments are complex and sophisticated attitudes. They appear to rely on the same mechanisms that underly our capacity to formulate complex representations, even if moral concepts are not representational. If moral sentiments are capable of driving moral behavior by themselves, why have we also evolved the capacity to judge actions to be right or wrong?
One explanation is that moral judgments evolved to enable us to coordinate social expectations. They not only play a role in guiding moral behavior, they introduce a greater amount of social flexibility. They give us the cognitive resources necessary to formulate, assess, and communicate norms that can then be used to regulate our sentiments. We can give voice to the moral rules we agree to obey, and learn the rules others favor following. Furthermore, if we are flexible and open to influence by others, we can coordinate our expectations so that the rules that govern a community are a compromise of the interests of its members. This makes it easier to accommodate changing social structures and power dynamics.
In the rough and rapidly changing world of the upper paleolithic and neolithic, moral flexibility would surely have been advantageous. Our species spread out very quickly to occupy environments as different as ice age Europe and Polynesia. In doing so, they had to adjust their societies to their new homes. Further, as they moved from small egalitarian communities to large hierarchical societies over the last ten thousand years (Boehm 2010), humans have had to learn to live in radically different social structures. Moral flexibility and the ability to coordinate moral expectations would have been an advantage to both individuals living within societies in changing environments, and also to individuals who moved between societies that might have different rules.
A system of norms for coordinating attitudes would not have sprang up overnight. Chimpanzees and bonobos, our closest relatives, do seem to have some rough analogues to moral sentiments, but have nothing clearly recognizable as judgments about morality. They recognize some social norms and are able communicate with each other about their observation in very basic ways. They are not capable, as far as we know, of formulating, evaluating, and communicating them (Rudolph von Rohr et al. 2011). Our nearest cousins might have some sense of fairness, reciprocity, permission, and obligation, but they do not reason, discuss, or negotiate them in the way that we do.
Our ancestors almost certainly would have started making cognitively sophisticated moral judgments only after our split with our nearest cousins six million years ago, and the greatest cognitive changes most likely came in the last fifty to hundred thousand years. In that period of time, cultures sprang up, art and religion came into being, tool use expanded far beyond the making of simple hand-axes, long distance trade commenced, and sophisticated forms of language probably evolved.
It makes sense that moral attitudes (as opposed to our moral sentiments) were largely a product of this time period, and if so, they must have evolved extremely quickly. Our pre-moral ancestors would have much to gain by coordinating social expectations about behavior. This would have produced an evolutionary pressure on the human mind to allow for such flexibility. This pressure would have set to work on cognitive system with a stock of non-moral attitudes including beliefs and desires. Moral judgments might have emerged sui generis from the cognitive void, but it seems far more likely that they would have been spun off from existing attitudes. It is easier to co-opt existing structures than to build them from scratch. Beliefs and desires seem like the two best candidates for the ontogeny of moral attitudes, and there are some reasons to favor the former over the latter.
Our pre-moral ancestors might have started with ordinary beliefs about social rules and regulations: they judged actions according to their social appropriateness within their community and expressed these judgments with very basic language. Generation by generation, they became increasingly disposed to form attitudes that functioned not to simply represent social standards, but to motivate them to live up to them. And gradually, the set of standards they were motivated to live up to might have come to differ slightly from those that were accepted by society. The divergence between what was accepted by individuals and what was accepted by other members of society could grow to a point where the attitudes fell outside of the range of representationality, especially if they came to have a use in influencing those standards, and not just reflecting them.
Our concept tokens employed in tracking rules and social regulations combine normally with negation, conjunction, quantification, generics, conditionals, tense, and other attitudes. If our moral concept tokens exapted from these representational concepts, then they could easily continue to share some of their syntactic properties.
This is supported by the rapidity of the development of moral judgments. It is quite possible that we only started internalizing social regulations as moral judgments in the last twenty or thirty thousand years. Keeping the antiquated uses of syntactic structures does not do us much harm, and selection has not yet weeded these uses from our cognitive repertoire.
This evolutionary story is far from the only story to be told about the origins of our moral attitudes, and it may well be false in many of its details. However, its plausibility undermines concerns about the syntactic properties that allow moral concepts to appear in the contexts they do. The fact that moral judgments do display the same syntactic properties as beliefs without having a clear function is not a major problem for the view.
In this paper, I have argued that noncognitivists can provide a recipe semantics for a variety of moral attitudes. Moral judgments are special insofar as they involve moral concepts. Moral concepts are characterized by the fact that they play a certain motivational role in the context of moral judgments. Moral concepts can be identified in spandrel contexts by their orthographic properties. The other attitudes that make up the menagerie of moral attitudes, including both complex moral judgments and non-judgment attitudes, are characterizable in terms of their orthographic similarities to other attitudes.
The lack of particular explanatory meaning-determined rationales for our use of such attitudes may be disconcerting, but they can be accounted for by the rapid evolutionary development of ordinary moral judgments.
Anderson, Michael L. (2010). Neural Reuse: A Fundamental Organizational Principle of the Brain. Behavioral and Brain Sciences 33 (4):245.
Ayer, Alfred J. (1936). Language, Truth and Logic. London, V. Gollancz, Ltd.
Baker, Derek & Woods, Jack (2015). How Expressivists Can and Should Explain Inconsistency. Ethics 125 (2):391-424.
Björnsson, Gunnar & McPherson, Tristram (2014). Moral Attitudes for Non-Cognitivists: Solving the Specification Problem. Mind 123 (489):1-38.
Blackburn, Simon (1984). Spreading the Word. Oxford University Press.
Blackburn, Simon (1988). Attitudes and Contents. Ethics 98 (3):501-517.
Blumson, Ben (2012). Mental Maps. Philosophy and Phenomenological Research 85 (2):413-434.
Boehm, Christopher (2010). Moral Origins: The Evolution of Virtue, Altruism, and Shame. Basic Books.
Dretske, Fred (1995). Naturalizing the Mind. MIT Press.
Field, Hartry (1978). Mental Representation. Erkenntnis 13:9-61.
Fodor, Jerry A. (1998). Concepts: Where Cognitive Science Went Wrong. Oxford University Press.
Fodor, Jerry A. (2008). Lot 2: The Language of Thought Revisited. Oxford University Press.
Fodor, Jerry A. & Pylyshyn, Zenon W. (1988). Connectionism and Cognitive Architecture. Cognition 28 (1-2):3-71.
Geach, Peter T. (1965). Assertion. Philosophical Review 74 (4):449-465.
Gibbard, Allan (2003). Thinking How to Live. Harvard University Press.
Horgan, Terry & Timmons, Mark (2006). Cognitivist Expressivism. In Terry Horgan & Mark Timmons (eds.), Metaethics After Moore. Oxford University Press: 255–298.
Joyce, Richard (2007). The Evolution of Morality. MIT press.
Leslie, Sarah-Jane (2007). Generics and the Structure of the Mind. Philosophical Perspectives 21 (1):375–403.
Mackie, John L. (1977). Ethics: Inventing Right and Wrong. Penguin.
Merli, David (2008). Expressivism and the Limits of Moral Disagreement. Journal of Ethics 12 (1):25 – 55.
Millikan, Ruth G. (1984). Language, Thought and Other Biological Categories. MIT Press.
Papineau, David (1984). Representation and Explanation. Philosophy of Science 51:550-72.
Pulvermüller, Friedemann. (2002). The Neuroscience of Language: On Brain Circuits of Words and Serial Order. Cambridge University Press.
Rudolf von Rohr, Claudia; Burkart, Judith & van Schaik, Carel (2011). Evolutionary Precursors of Social Norms in Chimpanzees: A New Approach. Biology and Philosophy 26 (1):1-30.
Schroeder, Mark A. (2008). Being For: Evaluating the Semantic Program of Expressivism. Oxford University Press.
Schwitzgebel, Eric (2002). A Phenomenal, Dispositional Account of Belief. Noûs 36 (2):249-275.
Shiller, Derek (2016). A Primitive Solution to the Negation Problem. Ethical Theory and Moral Practice 19 (3):725-740.
Shiller, Derek (Forthcoming). The Problem of Other Attitudes. American Philosophical Quarterly.
Searle, John R. (1962). Meaning and Speech Acts. Philosophical Review 71 (4):423-432.
Stevenson, Charles L. (1937). The Emotive Meaning of Ethical Terms. Mind 46 (181):14-31.
Stich, Stephen P. (1983). From Folk Psychology to Cognitive Science. MIT Press.
Invited Comments from Derek Baker
Commentary on Shiller’s The Unity of Moral Attitudes
Derek Shiller’s “The Unity of Moral Attitudes: Recipe Semantics and Credal Exaptation” is an ambitious paper with a radical proposal to solve the Frege-Geach problem. Shiller’s proposal, if I’m not misunderstanding it, is that it is sufficient for the non-cognitivist to solve the Frege-Geach problem that she can offer a plausible explanation of why the concept wrong (and other moral concepts) would have the same syntactic properties as a concept representing a property, even though this concept is not representational. Solving the Frege-Geach problem does not require giving an account of the meanings of complex moral sentences, however. This is because the mere fact that a syntactic combination of concepts is possible does not tell us that the resulting combination has any semantic value. Thus, the sentence in question may not mean anything (assuming I haven’t misunderstood Shiller on this point).
First, a bit of background. The problem for the non-cognitivist, remember, is that she says that “Abortion is wrong” serves not to describe a state of affairs, but to express some attitude (of condemnation, say) towards abortion. This sounds plausible enough for the sentence just offered, but there are sentences like “If abortion is wrong, then stem cell research is wrong.” This does not express an attitude of condemnation towards anything. The speaker could easily have no opinion on abortion; she may even be pro choice, and trying to convince her interlocutor that abortion is not wrong.
So what does the more complex sentence mean?
With a few recent exceptions, the standard strategy of non-cognitivists has been to say that the more complex sentence expresses some new more complex attitude. What is this new more complex attitude? There’s some debate about this, but roughly it needs to be an attitude that combines with other attitudes of condemnation in the right way. If the person who has this attitude also has an attitude of condemnation of abortion, the two in combination need to cause her to condemn stem cell research as well. If she has this attitude, and also has whatever attitude is expressed by “Stem cell research is not wrong,” these two in combination need to cause her to form the attitude expressed by “Abortion is not wrong.”
Since this is not Shiller’s solution, the only thing I want you to notice here is that making this solution work looks like it will be very very hard. We have to identify an attitude that stands in the right sort of functional roles to other attitudes of condemnation—and what attitude is that? We also have to identify an attitude expressed by “not wrong,” and this attitude will also have to have the right sort of functional connections with other attitudes. Finally, we need to have some sort of system that allows us to identify an attitude to correspond to any logically complex sentence you might want to express (e.g., “If abortion is not wrong then stem cell research is not wrong unless it has bad consequences or the stem cells are acquired unethically and…”).
Shiller suggests a much simpler route for the non-cognitivist to take. The problem of accounting for logically complex sentences is a special case of the problem of accounting for compositionality in general. After all, it’s not just that moral sentences can be negated or embedded within conditionals, they can also figure in claims like, “Abortion used to be wrong,” “Bob wonders whether abortion is wrong,” “If only abortion were wrong!” and so on. So really, what we need to explain is the fact that moral judgments can show the same kind of compositional complexity as non-moral judgments.
Understanding the Frege-Geach problem as ultimately a problem of explaining compositionality is fairly standard at this point. There’s a reason, however, for the focus on the logical connectives as opposed to other forms of compositionality, one that I think Shiller glosses over, which I’ll get to in a minute. In any case Shiller’s explanation has an attractive simplicity to it, and it promises to massively simplify the Frege-Geach problem.
His solution, as I understand it, is that compositionality is explained by the syntactic properties of a representation (or presentation) rather than the semantic properties. So what the non-cognitivist needs to do is explain why we should expect that moral concepts would have the sorts of syntactic properties that underwrite the requisite compositionality. Once we do that, we can account for which attitudes complex moral sentences express. However, just because certain combinations of concepts are syntactically possible, it doesn’t mean that the combination has any semantic value. So the non-cognitivist, in solving the Frege-Geach problem, should not feel obligated to offer a semantic value for these complex sentences (the attitude expressed may not represent anything or have any functional role in the speaker’s psychology). The resulting position is summed up in Shiller’s introduction: “Moral concepts enjoy the same syntactic freedoms as other concepts, even if they don’t have coherent semantic functions in all of the contexts in which they can occur.” (This quote is one of the reasons why I take Shiller to be saying that complex moral sentences might be meaningless, for all we know. If this is not what he means, I’d appreciate some clarification.)
So the Frege-Geach problem gets a lot easier to solve.
I strongly disagree with the resulting position, but before criticizing it, I just want to say that the points Shiller makes in the process of working up to that conclusion are consistently insightful and helpful. The first eight pages worth the price of admission on their own, and despite my disagreements with the resulting theory, it is an interesting position worth engaging with.
So what is wrong with Shiller’s position? Well, I don’t think that the non-cognitivist can just shrug off the demand for a systematic account of the meaning of complex sentences, and the reasons for this are tied with the reasons for focusing on logical compositionality.
Let’s start with “Abortion is not wrong.” The person who says this seems to be disagreeing with the person who says, “Abortion is wrong.” But disagreement seems like a semantic notion. That is, two syntactic entities disagree in virtue of meaning something (and if they don’t mean anything, they can’t disagree). Now Shiller hasn’t denied that “Abortion is not wrong” is meaningful. But he has denied (I think) that non-cognitivist needs to give a systematic explanation of its meaning. This can’t be right, though, because the phenomenon is very robust. “Stem cell research is not wrong” disagrees with “Stem cell research is wrong,” “Giving to charity is not obligatory,” disagrees with “Giving to charity is obligatory,” etc. More generally, “not P” disagrees with “P” for all P. It would be very strange if there weren’t some general explanation of that fact—some explanation of what the word “not” is doing (or the concept not in the case of an attitude) that explains why it always results in disagreement.
A standard explanation is that the truth-value of “not P” (or not P) is a function of the truth-value of P—it is true if P is false and false if P is true. Sentences or attitudes disagree when they cannot be jointly true. So “P” and “not P” will always be in disagreement, given the truth-functional meaning of “not.”
Whatever we think of this explanation, it cannot be the non-cognitivist’s. An alternative explanation among non-cognitivists has been to claim that there is some sort of rational or functional conflict (or both) between the attitude expressed by “P” and the attitude expressed by “not P.” The two claims disagree then, because one cannot sincerely accept both without being in a (somehow) defective state of mind. You can go back above to look at how the characterization of the attitude expressed by the conditional claim “If abortion is wrong, then stem cell research is wrong” is meant to account in a similar way for why modus ponens and modus tollens are valid arguments.
Geach, in his paper “Ascriptivism,” was worried that on a non-cognitivist account moral terms, instances of modus ponens involving those terms turn out to be cases of fallacy of equivocation (p. 223). The meaning of “Abortion is wrong” is explained in terms of the condemnation of murder. But “If abortion is wrong, then …” expresses no condemnation. So the meaning of the embedded “abortion is wrong” is different from the unembedded. So use of modus ponens is fallacious. I like to think of this as the charge that for a non-cognitivist, logical terms work like the predicate “fake.” “Fake gun” does not refer to a subclass of guns (like “large gun” does). It refers to things that aren’t guns at all. We can’t understand the meaning of “fake gun” in terms of applying the property fake to the object gun. Instead, in this context, the semantic value of “gun” is something different, even though the syntactic object is the same. But this seems to be exactly the kind of picture that Shiller is endorsing, or at least leaving open as a live possibility: “Moral concepts enjoy the same syntactic freedoms as other concepts, even if they don’t have coherent semantic functions in all of the contexts in which they can occur.”
Geach’s concern with equivocation can be generalized into a constraint on our compositional accounts. The meanings of complex sentences containing “Abortion is wrong,” or any other moral claim, as an embedding, need to be properly related to the meaning of the unembedded claim. For example, someone tells you, “Bob believes that abortion is wrong.” Think about some of the relations this stands in to the claim that abortion is wrong. Most obviously, if you are willing to sincerely assert that abortion is wrong, it tells you Bob agrees with you; and if you are willing to assert that abortion is not wrong, it tells you Bob disagrees with you. But it’s hard to see how it could impart that information if the meaning of “Bob believes that abortion is wrong” weren’t related to the meaning of “Abortion is wrong”—if the semantic function weren’t preserved when embedded. Similarly, the claim gives you some reason to accept that abortion is wrong if you think of Bob as an expert on these things. Again, this only makes sense on the assumption that the meaning of the complex and simple sentence are related.
Now I think part of the reason for the focus on logical embedding by the non-cognitivist, is simply that in these cases the semantic relations between the simpler and the more complicated sentences are much more clear-cut, obvious, and seem basic to the use of language. Some of the proposed relations between “Bob believes that abortion is wrong” and “abortion is wrong” might be debatable or imprecise. It gets even worse if you consider the relations between “If only abortion were wrong!” and “Abortion is wrong.” But we have a pretty good understanding of the semantic relations of disagreement that are supposed to hold in virtue of logical form, or the relations of consequence that are supposed to hold in virtue of logical form. What’s more, these semantic relations seem absolutely central to language and thought. So they provide an especially good test case for the expressivist, or a nice first step in building a non-cognitivist solution to the general problems of compositionality and embedding. Basically, if the non-cognitivist can’t even account for this much, the case is hopeless.
What’s more, these are semantic relations we are especially confident in. Responding to Geach’s original worry that moral modus ponens is fallacious according to the non-cognitivist through wholesale bullet-biting, as Shiller seems to do, strikes me as on a par with embracing skepticism about the external world. It’s an overwhelmingly radical revision of my, and I think most people’s, picture of the world.
Invited Comments from Tristram McPherson and David Faraci
Noncognitivism and ‘Recipe Semantics’
Ohio State University
In “The Unity of Moral Attitudes,” Derek Shiller offers a novel way of characterizing noncognitivism about a certain class of psychological states. In these comments, we briefly summarize key elements of Shiller’s approach, and raise several concerns about it. We conclude by asking why his broad picture does not in fact make cognitivism more appealing than noncognitivism.
Shiller begins by correctly emphasizing that to be adequate, noncognitivism about moral thought needs to be able to account for a wide range of psychological states that apparently involve moral contents (§1). This includes logically complex thoughts with moral contents (both the familiar cases of negation and conditionals, and less familiar cases such as generics and plural quantification). It also involves explaining the range of attitudes that we can apparently take towards moral contents: we can hope that an action is permissible, or imagine that another is not, etc. One can imagine attempting to develop accounts of these phenomena ‘one at a time,’ but Shiller hopes to cut the Gordian knot, and provide an account that addresses all of these phenomena simultaneously.
Shiller begins by assuming that ordinary propositional attitudes are understood in terms of distinctive relations borne towards representational contents. So, for example, to “believe a proposition might be to be disposed to token representations of that proposition in a certain way and use those representations in deciding what to do with the goal of getting what one wants” (§2). Shiller also assumes a compositional account of representational contents: the content of any particular representation is a systematic function of what Shiller calls the ‘concept tokens’ that it is composed of. These assumptions are controversial, but widely shared, and we grant them here.
The distinctive character of Shiller’s proposal arises from two further assumptions. The first is that the compositional rules utilize certain non-semantic features of concept tokens. That is, (a) particular concept tokens can be individuated by certain non-semantic features, and (b) the compositional function uses these features as inputs. Shiller calls the non-semantic features that individuate concepts orthographic properties. In light of this, Schiller introduces the idea of an orthographic similarity class (§3). These are classes of tokens sorted by some orthographic property. Some orthographic similarity classes will be compositionally significant: their defining orthographic properties will be the non-semantic inputs to the compositional rules. The second assumption is a teleological theory of content determination: the representational content of a concept (or complex representation) is determined by its ‘designed function’, in some sense of this phrase. Together, these assumptions are meant to make room for spandrel contexts within which concepts can be tokened (due to their orthographic features) but for which they were not designed. Shiller proposes that there is an orthographic similarity class – the presentations – which is shared by all contents fit to tokened within belief and other propositional attitudes. On Shiller’s proposal, presentations include, but are not exhausted by, representational concepts.
This provides the context for Shiller’s distinctive noncognitivism: the proper function of moral concepts is to influence action, rather than to represent. However, moral concepts are also presentations: ordinary moral judgments involve bearing the belief-relation (or a similar relation) towards these non-representational moral concepts. This account is supposed to provide the noncognitivist with a way of explaining the range of contexts that moral thoughts can enter into, without having to provide a semantics for them (§3). Consider negation as an example: rather than attempting to provide a constructive ‘logic of attitudes’ to address the negation problem, Shiller proposes that the noncognitivist can simply suggest that the felicity of negated moral judgments, and their apparent inconsistency with their non-negated correlates, is a function of compositional rules that are presumed to operate on presentations.
This is an ambitious and provocative proposal. In what follows, we raise several questions.
First, Shiller suggests that moral judgment differs from belief in terms of what concepts it tokens, rather than what kind of relation is borne towards those concepts. Thus, the relation we bear towards moral concepts in ordinary moral judgment is assumed to be belief, or a very belief-like relation. However, it seems dubious that we are able to believe something merely in virtue of its orthographic features, rather than its semantic features. To bring out our doubts, consider one of Shiller’s own leading examples. Shiller initially motivates the idea of a presentational class by discussing a famous bit of nonsense from Lewis Carroll’s Jabberwocky: “’Twas brillig, and the slithy toves…” This bears certain orthographic similarities to an English sentence, despite presumably being meaningless. But notice that plausibly, one cannot believe that ’Twas brillig, and the slithy toves… What would it be to believe that? The same goes for other propositional attitudes: this also doesn’t seem like the sort of thing we can imagine or hypothesize. It is thus quite unclear whether orthographic similarity to a representation is enough to make something a candidate object of thought. But we can believe or imagine It is not wrong to pet kittens, so the noncognitivist owes us an account of what that thought consists in. And that looks like it is (a mental state version of) the negation problem all over again.
Shiller says that complex moral attitudes are like daydreams (§3). His leading example of a spandrel context is daydreaming about horses. He proposes that the concept horse evolved to represent horses in the context of certain attitudinal relations that are important for our mental lives, such as the belief-relation and the desire-relation. It did not, he suggests, evolve to represent horses in the context of the unimportant daydreaming-relation. Nevertheless, because of its orthographic features, horse is capable of being tokened in daydreaming as well as belief. By analogy, it may be that due to the orthographic features moral concepts share with representational ones, they can be tokened alongside concepts used to create more complex moral judgments, such as those involving negation or generalization.
We are not convinced that daydreams don’t play an important role in our mental lives, as Shiller suggests. It seems to us that daydreaming might, for instance, allow us to imagine various possibilities during ‘down time’, which would facilitate later deliberation involving those possibilities. Of course, the specific example isn’t central to Shiller’s argument, but it would be nice to have a more convincing example of the kind of ‘spandrel context’ Shiller asks to do so much work in his account. Otherwise, there is a natural worry that such cases are rare or idiosyncratic.
A more substantive concern mirrors one already raised: it’s not clear that it is horse’s orthographic features, rather than its semantic ones, that allow it to be tokened in daydreams. To return to his other example, it is no more intelligible to daydream ’twas brillig, and the slithy toves than to believe or imagine it. So, once again, we’d like to hear more from Shiller about why we should think that orthographic similarity is all that’s required here.
Our next question is about the analogy with moral judgment. In the daydreaming case, there is a relation—the daydreaming-relation—that is borne towards something with representational content, namely the concept horse. Since horse didn’t evolve to represent anything in the context of the daydreaming relation, that relation is a spandrel context for horse. Now consider the complex moral judgment that something is not wrong. Our question is: What exactly is the spandrel context for what here?
One possibility is that being tokened in the moral-judgment-relation is a spandrel context for negation. Another is that being tokened alongside a moral concept like wrong is a spandrel context for negation (or vice-versa). The former is unattractive, given that the moral-judgment-relation is supposed to be the same as or similar to the belief-relation, which is presumably a context in which negation did evolve its function. A version of the latter is more promising, but it invokes an entirely different kind of spandrel context from the daydreaming example: one about combinations of concepts, rather than the relations they are tokened in. One idea is that perhaps negation has evolved to function in representational contexts, and thoughts containing wrong are not representational.
Even if this works, it leaves a more pressing concern: does a concept play its ordinary functional role in spandrel contexts? The daydreaming example might seem to suggest that there is no effect: horse still represents horses in daydreams. But we strongly suspect that this is a result of the fact that daydreaming is itself a representational relation. For Shiller, aptness to occur in a spandrel context is secured by merely orthographic features. But if being tokened alongside wrong is possible for negation solely because of the orthographic features of each, it is hard to see why we should expect negation to be doing any functional negating here. And if negation doesn’t functionally negate in thoughts like it is not wrong to pet kittens, then it is unclear whether we can explain the apparent rational tension between positive and negative moral judgments. This seems like a very bad result.
Similarly, suppose we accept that the belief-relation is a spandrel context for wrong. One advantage of noncognitivism is supposed to be that it accounts for the action-guiding nature of moral judgment. Shiller might seem to have retained this advantage by positing an action-guiding function for moral concepts. But this is too quick: we need to understand how that action-guiding function is maintained and deployed through tokening within the belief-relation. Given that noncognitivism is supposed to have this advantage precisely because bearing the belief-relation to something is purportedly unable to issue in motivation, it is far from clear that this will be possible.
Of course, Shiller is welcome to set these concerns aside, offering his proposal merely as an explanation of how noncognitivism can account for moral discourse’s syntactic features. But Shiller himself claims that he is open to further development of his view along lines that would vindicate aspects of moral discourse, such as the apparent tension between positive and negative moral judgments. We worry that the existing structure of his account will make it quite difficult to develop it into a vindication.
Perhaps more importantly, we worry that such a merely syntactic solution to noncognitivism’s problems removes any theoretical reason to prefer it to cognitivism. Consider Shiller’s evolutionary just so story, the heart of which is this:
Our pre-moral ancestors might have started with ordinary beliefs about social rules and regulations: they judged actions according to their social appropriateness within their community and expressed these judgments with very basic language. Generation by generation, they became increasingly disposed to form attitudes that functioned not to simply represent social standards, but to motivate them to live up to them. And gradually, the set of standards they were motivated to live up to might have come to differ slightly from those that were accepted by society. The divergence between what was accepted by individuals and what was accepted by other members of society could grow to a point where the attitudes fell outside of the range of representationality, especially if they came to have a use in influencing those standards, and not just reflecting them. (§4)
We are not sure what it is for an attitude to ‘fall outside of the range of representationality,’ but we take everything else in this story to be compatible with a simple subjectivist view, for example, or, (with minor tweaking) an error theory.
Now consider standard objections to simple subjectivism, such as its failure to make sense of disagreement, or of embedded claims like “If I didn’t disapprove of killing, it wouldn’t be wrong.” As suggested above, if negated and conditional contexts are simply spandrels, it is not clear that the noncognitivist can have anything satisfactory to say about those contexts either.
Or, again, consider the noncognitivism’s standard advantage in accounting for the motivating force of moral judgments. As mentioned above, the standard Humean worry that beliefs can’t guide action threatens the idea that bearing the belief-relation (or anything similar) towards wrong could motivate us. But suppose Shiller can successfully argue that wrong’s alleged action-guiding function can be played within the context of the belief-relation. Unless that argument is in some way dependent on the non-representational nature of wrong—and we see no reason to think this would be the case—the cognitivist could deploy this solution just as easily.
The point of all these worries is simple. Shiller sees it as a boon for his view that it avoids the details typically associated with vindicating certain familiar features of moral thought and discourse. But without tackling those details, it is unclear how Shiller’s view manages to capture any of what motivated noncognitivism in the first place. And we worry that given the merely syntactic approach he takes to things, once those details are fleshed out, it will turn out that you could have everything he offers simply by being a deflationary sort of cognitivist.
 Here we are inspired by James Dreier’s ‘accostivism,’ the lesson of which, we take it, is that securing grammaticality doesn’t suffice to help us to understand the broadly semantic properties of fragments of language. Dreier, James. “Negation for Expressivists,” Oxford Studies in Metaethics) Vol. 1 (Ed. Russ Shafer-Landau), 1996.
 Cf. Mark Schroeder Noncognitivism in Ethics, 2010.
5 thoughts on “The Unity of Moral Attitudes: Recipe Semantics and Credal Exaptation”
** Overview **
Thanks to the organizers for providing the venue and to the commenters for their thoughtful comments and criticisms. This has been a great opportunity for me to think about some sides of the issue that I had so far neglected. This introductory note concerns some themes shared between my responses to each set comments. My specific responses to Faraci/McPherson and Baker will follow.
After reflecting on the comments, I think that most of the key differences in my approach to noncognitivism stem from the thought that the semantic properties of words and concepts should not bear too much explanatory weight. I start with the assumption that semantics is a result of patterns of usage. Semantic properties may provide systematizing explanations of usage, but they do not contribute to an underlying causal mechanism behind usage. The fact that I apply a concept ‘cat’ to a cat can’t be because of the concept’s semantic properties — the semantic properties instead reflect the pattern of usage of which the application of the concept is one instance.
Causal mechanisms do underly the phenomena that semantic properties are traditionally used to explain, and these mechanisms must invoke non-semantic properties. My account takes advantage of the possibility for noncognitivists to make use of the same non-semantic properties to explain the corresponding features of moral attitudes.
In my paper, I use this strategy as follows: something must explain why we are able to take the ‘judgment’ relation toward representational contents, and whatever it is, I propose that the same thing allows us to take that relation to moral contents.
The responses that I give to the questions and criticisms of Faraci, McPherson, and Baker will extend this basic strategy for other phenomena.
Response to Faraci and McPherson
My solution to the problem of the unity of attitudes requires that syntactic properties determine which conceptual structures are viable objects of our attitudes. If this is wrong, then the (absence of) representational properties of a presentation may prevent it from being the subject of a judgment. Faraci and McPherson note that we cannot bring ourselves to judge that ‘twas brillig, and the slithy toves…’. But if the syntactic properties of a content determines its viability as a subject of an attitude, then this too should be a viable attitude.
One of the main thoughts behind my approach is that the semantics of a concept is largely epiphenomenal to its behavior. Concepts mean what they do because of how we (collectively) deploy them, and therefore their meanings cannot guide their deployment. Semantic values may systematize a concept’s behavior, but semantic properties do not actually play a causal role in directing our concepts. For this reason, I don’t think semantic properties can ultimately explain viability. There must be an unknown causal mechanism that determines the viability of a content, and this mechanism should be equally amenable for noncognitivists.
There is no attitude expressed by Lewis Carrol’s sentence ’Twas brillig…’. The reason for this is that the sentence has no coherent analogue in the language of thought. This is not because the sentence is non-representational, but because there are no constraints at all on its content. There is nothing that it is to be a concept of brilligness, and so there is no kind of attitude, not even one that could be characterized by a recipe semantics, that it expresses.
If there were a syntactically viable sentence in the language of thought that the sentence expressed, then it could be the object of a judgment. Even so, we might never be in a position to actually judge it to be true. “Colorless green ideas sleep furiously” may express a syntactically viable judgment composed of concepts that we do possess, but it is one that we never actually believe.
The paper provides a rationale for why moral concepts are often the subject of judgments, and are not just syntactically viable attitudinal possibilia like ‘colorless green ideas…’. If straightforwardly predicative moral judgments were advantageous to our ancestors, then it makes sense that we should be disposed to have them, and the same exaptive explanations for the syntactic viability of the other attitudes may help explain our disposition to actually form them.
Representationality and Daydreams
Next, Faraci and McPherson wonder whether moral concepts maintain their action-bearing function in presentations in the way that representational concepts maintain their representationality in the context of daydreams. This issue relates to the question of whether moral concepts mean the same things in the different contexts in which they occur. Representational contents may be taken to be the same content by virtue of representing the same thing. Nonrepresentational contents cannot.
On my view, our concepts would continue to have their motivating-when-predicated function in cases where they are not predicated something like the way that a spring maintains its tension even when compressed and left in a drawer. Outside of the context of moral judgments, however, the presence of moral concepts won’t actually lead to motivation. So while they may continue to have a function, the function is not manifest.
This might be taken to mean that moral concepts are a bit different from proper representational concepts, insofar as they don’t manifest their function in spandrel contexts. Representational concepts can still represent in the context of attitudes, even if the attitude isn’t set up for this purpose. This might be sufficient for a shared content.
However, I’m skeptical that representational concepts are really all that different. For many contexts, the best analysis of what it means for a concept in that context to represent may be that the concept shares a (orthographically-secured) semantic identity with concepts that play a certain representational functional role in other contexts.
This is what I hoped to get out of the daydream proposal: even if daydreams do serve a purpose for us, which I agree they might, it is at least conceivable that there could be representational attitudes that are not designed to represent and whose contents would borrow their representationality entirely from orthographic relations with concepts in other contexts.
Representation isn’t some special thing that some concepts innately do. It emerges from their behavior. If concepts that are tokened in daydreams do not have a sufficiently robust behavior within the daydream to be representational by virtue of that behavior alone, then the fact that they represent depends upon their resemblance to concepts in other representational contexts.
If this sort of purely parasitic representationality-via-similarity is conceivable for daydreams, then the same thing should be possible for other noncognitive contents. Instead of borrowing their representationality from other contexts, however, noncognitive contents would borrow their bearingness-on-action from other contexts.
Finally, Faraci and McPherson claim that my proposal might undermine some the appeal of noncognitivism. There are many different reasons to be attracted to noncognitivism, and my proposal might not satisfy all noncognitivists. One traditional attraction of noncognitivism comes from its compatibility with an a priori psychology that precludes beliefs from having any motivational capacity except in combination with desires.
They are right that my proposal would undermine that sort of rationale for noncognitivism, but I don’t think that belief states ought to be conceptually barred from being solely responsible for non-cognitive effects. Cognitive function is messy; representational states can have diverse non-representational effects. Observing exercise has been linked to increased rates of ventilation.1 If beliefs are motivationally inert in the way that perception is physiologically inert, then beliefs may in fact make motivate us. If Lebrecht et al.2 are right that concepts routinely bind affective valences, then motivational influences may be the rule rather than the exception with concepts.
Instead, I take the primary appeal of noncognitivism to be that it can easily make sense of the commonality between moral judgments. If there is no single plausible representational content shared by all those who make the same moral judgments, and if their attitudes all share a similar motivational profile (at least in a given context), then it is more plausible that the motivational profile characterizes the attitude.
The notion of a content ‘falling outside the range of representationality’ is critical to this idea. Concepts don’t get representational contents on the cheap. In order for a concept to get a representational content, it must behave in a way that warrants its content. There are many many ways we could arbitrarily deploy concepts that would prevent them from having representational contents. In order to acquire representational meaning, concepts need to both have a coherent behavior and have that behavior be broadly representational. There may be coherent behaviors for concepts that are not representational. I think this is the case with moral concepts.
I have said nothing in the paper about whether noncognitivism has advantages over its representational cousins: fictionalism, subjectivism, relativism, and error theory. If it does, I think that the advantages result from subtle metasemantic considerations. There are certainly cognitivist views that share the metaphysical commitments of noncognitivism. The debate between noncognitivists and these cognitivists should be settled by other factors, and I have not hoped to advance noncognitivism in these debates except by illustrating how it might avoid a problem that the other problems are not faced with.
Paccalin, C., and M. Jeannerod. “Changes in breathing during observation of effortful actions.” Brain Research 862.1 (2000): 194-200. ↩
Lebrecht, Sophie, et al. “Micro-valences: perceiving affective valence in everyday objects.” Frontiers in psychology 3 (2012). ↩
Response to Baker
In the paper, I suggest that we might best understand moral attitudes in terms of their component parts and characterize those parts in terms of their roles in other cognitive contexts. Baker points out that this invites the interpretation that these other moral attitudes are meaningless: perhaps moral concepts are meaningful in the context of straightforwardly predicative moral judgments, but not in other attitudes.
Let me say in response that I am not entirely sure what it takes for a sentence or an attitudinal content to be meaningful. On a strict interpretation, being meaningful might require having an abstract object (i.e. a reference, truth-condition, inferential role, etc.) that is its meaning. On such an interpretation, I do agree that my view implies that the contents of many moral attitudes are meaningless. This would not, however, preclude them from having a coherent and systematic use or from forming important and personally valuable parts of our cognitive lives.
On the other hand, on a liberal interpretation of meaningfulness, being meaningful might only require being a viable content for a judgment. Where ‘judgment’ is flexible to enough cover noncognitivist uses, this liberal interpretation makes my account consistent with the meaningfulness of a range of contents.
There is plenty of middle ground between these two notions of meaningfulness. I don’t have a well-worked out theory of meaningfulness, so I’ve tried not to confidently affirm or deny that the contents of other moral attitudes are meaningful. If they are meaningful, then I don’t think that their meaning is what binds them together as a single kind of content. Instead, their shared meaning results from the fact that they are so bound.
Even if it is granted that the contents of other moral attitudes are meaningful, it needn’t follow that attitudes share meanings in the way necessary to prevent equivocation. Baker notes that this was something that concerned Geach in his early presentations of the Frege-Geach problem. Geach may have been primarily concerned about the inverted commas meaning of moral language that noncognitivists were tempted to use for tricky cases, according to which: “If lying is wrong, then getting your little brother to lie is wrong” means “If most people would predicate ‘wrong’ of lying, then most people would predicate ‘wrong’ of getting your little brother to lie”. This is a fairly different approach from my own. Nevertheless, Baker is right to worry that I am running afoul of the constraints Geach set forth for an adequate solution.
I’m inclined to think that there is some kernel of shared-meaning that is metaphysically prior to the meanings of concepts themselves. If a term has to have a consistent pattern of usage before it can have a meaning, then something must associate its different instances other than their meanings. A term/concept means what it does in part because all its instances are instances of the same term/concept.
I think something similar can be said about moral concepts. A certain level of semantic unity is guaranteed by sub-semantic features that identify different instances of the concept as entities that collectively bear a single semantic content (if any content at all). So on my view, moral concepts may acquire their content collectively as an individual semantic unit, even if their content is characterized as a role in a particular context. Noncognitivists who resorted to an inverted commas interpretation in special contexts were running afoul of Geach’s constraint because they were assigning two different meanings to terms. I do not do this.
I think that ‘heart’ is a good analogy for the unity of moral concepts I am postulating. A heart is an organ that pumps blood in the context of a healthy living body. Not just anything that is heart-shaped and inside a person’s chest cavity is a heart. Nor do all hearts pump blood: cadavers have hearts even though their hearts no longer perform their proper function. Although a heart does not play its characteristic role in the context of a cadaver, it is still a heart because it is the same kind of organ as the organs that pumps bloods. Similarly, a moral concept in the context of a moral hope may be a moral concept because it is the same (orthographic) kind of concept as a concept that functions in a certain way in the context of a straightforwardly predicative moral judgment. Its the same meaning-bearing entity that bears a particular motivational meaning manifested in the context of straightforwardly predicative moral judgments.
I don’t think that this answer should fully satisfy Baker, but I hope this is a step in the right direction. What we need, and what I have not yet supplied, is a theory of identity across contexts that does not depend upon meanings, but that can form the basis by which concepts are treated as single semantic kinds. Noncognitivists are not alone in needing such an account, for cognitivists will also need to explain how representational contents that occur in different contexts share representational contents, when their behavior in those specific contexts is not sufficiently robust to secure their representational content. Whatever makes this determination is my key to avoiding equivocation.
This brings me finally to the issue of disagreement. There are two possible things that might require explanation: the fact that some moral attitudes disagree with each other, or the fact that some moral attitudes appear to disagree with each other. Noncognitivists typically aim for the first, and their success is controversial. The second kind of explanation is much easier to find. This kind of explanation is enough, in my opinion, to make a noncognitivist theory workable. Explanations of why we are inclined to think that two attitudes disagree undermine the significance of the unintuitiveness of their failure to disagree.
Noncognitivists might point to the same features to explain the appearance of disagreement as cognitivists. Its highly doubtful that we judge that two attitudes disagree by virtue of analyzing explicit semantic knowledge of the contents of our attitudes. Metaethicists cannot agree on what the meanings of moral attitudes are, and so if moral contents needed to be grasped in order for their disagreement to be perceived, we should expect a lot more uncertainty about the genuineness of moral disagreement than we encounter.
Instead, it seems more plausible that we grasp disagreement through a sensitivity to the mechanisms that produce it. Negational concepts are a good guide to disagreement. The fact that we treat negation as producing disagreement may help to make negational concepts the concepts that they are. It is tempting to project this disagreement-producing appearance to negations in novel sentences whose meanings we do not understand. We get the sense that ’twas brillig’ disagrees with ’twas not brillig’ even if we have no idea what ‘brillig’ means.
On the explanation that I favor, it is tempting to generalize patterns in negational concepts from representational to non-representational contexts. “Not x” appears to disagree with “x” because the meaning of “not” is typically such as to ensure their disagreement. It is hard to discern when “not” isn’t producing disagreement. Seeing that requires developing an adequate semantic theory of the contents involved. The fact that we do not know enough about the meanings of our moral attitudes to easily perceive their difference from representational attitudes, and the fact that they involve patterns that ensure disagreement in ordinary representational attitudes, could explain the appearance of disagreement in moral contexts.
Ultimately, I admit that my view is inconsistent with widespread intuitions. I do not think that it should make it any less appealing than traditional forms of noncognitivism, relativism, or error theory, which all involve some departures form common sense. Philosophy can be surprising, and it is doubtful that ordinary intuitions will make it through a fully developed metaethical theory intact.
Thanks for the response. I really like a lot of what you have to say. But there some points where I feel like I still don’t fully understand the view, and one other point where I strongly disagree. I’ll start with where I disagree.
So my big issue is that you are willing to endorse widespread revisions about our intuitions about disagreement in moral cases. I’m not happy about this, for three reasons. The first one is that my primary reason for finding non-cognitivism compelling is Missionary and Cannibals or Moral Twin Earth type cases. I have a strong intuition that people are in disagreement when standard referential semantics would normally predict that they are not. Non-cognitivists seem to have a nice story about why disagreement using moral terms might be different.
But if it turns out that being a non-cognitivist requires that I think that our pre-theoretic intuitions about disagreement in normative cases are for the most part wrong, then my motivation for being a non-cognitivist is gone. So I think a position like yours requires articulating a new compelling reasons for being a non-cognitivist, or its a bit of Pyrrhic victory.
Second, I understand that addressing the Frege-Geach problem is hard, but you think that it can be addressed. But part of accounting for the logical properties of normative sentences will have to include vindicating the claim that “Murder is wrong” and “Murder is not wrong” are logically inconsistent. If you don’t do that, you have not solved the Frege-Geach problem. But I would think that if two sentences are logically inconsistent, don’t the people who endorse one disagree with those who endorse the other? Isn’t this just a platitude about logical inconsistency and disagreement?
Third, I feel like there’s a tension between the idea that competent speakers can suffer widespread failure to know which of their sentences disagree with each other, and what you wrote in response to Tristram and David: “Concepts mean what they do because of how we (collectively) deploy them…”
Here’s why. Imagine everyone mistakenly believes ‘not wrong’ is in disagreement with ‘wrong’. In that case, everyone will use ‘not wrong’ to disagree with ‘wrong.’ But the predicates express the relevant concepts or combination of concepts. But that means that the use of these predicates counts as a deployment of the relevant concepts. But in that case, the combination of concepts NOT WRONG is deployed to disagree with WRONG. So the believe wasn’t really a mistake.
(You might complain that I’m stipulating that people use ‘wrong’ and ‘not wrong’ to disagree, which is begging the question in the current context. But again, you write, “Semantic values may systematize a concept’s behavior, but semantic properties do not actually play a causal role in directing our concepts.” So what is relevant is whether speakers act like their terms disagree. We’ve stipulated that they do if they are sincere, since we stipulated they believe they disagree. Saying the speakers use the terms to disagree should just be a case of systematizing speaker behavior.)
In short, there seems like a tension between the idea that collective use determines meaning, and the idea that we could be radically mistaken about what our terms mean.
I’m wondering, moreover, if you have to say this. You note that there are probably stricter and looser notions of meaning. Since disagreement is clearly a meaning-related term, shouldn’t there be correspondingly strict and loose senses of disagreement. Perhaps there is a strict representation sense in which WRONG and NOT WRONG do not disagree. But there might be a looser, meaning-is-use sense of disagreement in which they do disagree. (In fact, they’ll probably have to disagree, because meaning-is-use guarantees that widespread false beliefs about meaning will become true beliefs very quickly.)
In writing this paper, I set myself the target of presenting the most extreme version of the view that I thought I might reasonably be able to defend. I am attracted to more moderate versions as well. In particular, I think that a lot of what I have to say is more plausible for other constructions than for negation. One problem facing the noncognitivist is to explain all of the different contexts in which moral concepts can be embedded. It may make sense to give a separate account for moral judgments involving negation, which is a very basic and common context for moral concepts, and reserve my account for rarer contexts, such as in the judgment that murder is as wrong today as it was yesterday or the hope that I did the right thing.
I think you’re right that my project might be better served with a quasi-minimalist form of attitudinal disagreement. We might say that what it is for two attitudes to disagree with each other is for those two attitudes to be part of a cultural practice guided by certain norms. The fact that we treat attitudes as disagreeing is enough to make them genuinely disagree in this minimal sense, because by adopting one attitude I am committing myself (according to the norms of the culture) to not adopting the other and to behaving in a certain way with respect to other people who do adopt that attitude. This approach would still differ greatly from most mainstream noncognitivist accounts because it gives up on any deep attitude-innate rationale for disagreement. Instead, the disagreement is forced upon the attitudes from an external source.
I might note that I have elsewhere also defended a similar minimalist noncognitivist account of disagreement in which regarding two attitudes as disagreeing largely amounts to having a certain set of pro-attitudes toward them. I didn’t specifically target this paper at the Frege-Geach problem, and a full response to that problem invoke this extra machinery.
I also am sympathetic to the semantic importance of disagreement that you raise. In my earlier response, I suggested that the philosophical force of disagreement might be undermined by an explanation of the appearance of disagreement, but I don’t think this is quite right. The appearance of disagreement affects its use, and its use does determine its meaning. If people are sufficiently committed to a cognitivist account of moral disagreement, then that might rule out noncognitivism in the same way that if everyone were to agree that whales were ‘fish’, phylogeny be damned, that might make whales ‘fish’.
I am not really sure what most people think about the substantialness of moral disagreement. I admit I haven’t read much of the literature on the topic. But if the average person were committed to thinking that moral disagreements between attitudes are genuine and substantial in the way my account precludes when they were faced with all the considerations for and against non-cognitivism, that would be a problem for the account. I’m not sure if it is a problem if people actually generally regard moral disagreements as substantial, for they might be swayed by factual mistakes. Use theories of meaning ought to allow for the possibility of widespread misapplications of words and concepts.
In any case, in order to decide between the rival semantic theories, we need to know what the true costs of noncognitivism are. I think we’re still a long way away from that.
Comments are closed.