John Rawls (1971) suggested that intuitive judgments about justice are systematic. This proposal provided an initial theoretical motivation for LA, suggesting that humans may possess tacit knowledge of a small set of fundamental moral principles, and that this knowledge is analogous in important respects to the implicit grasp of the fundamental principles of grammar that underwrite our linguistic judgments.1 Rawls’ suggestion has often been misinterpreted as a prescriptive claim (see Mikhail, 2010), but it has also been at the heart of 30 years of debate among philosophers, who have deployed moral thought experiments intended to elicit our intuitive moral judgments. A consistent, but still striking finding is that the principles whose acceptance would explain the patterns that we find in our intuitive judgments are rarely consciously accessible to those making them, and the philosophers who have designed such thought experiments often struggle to explain why some scenarios seem to give rise to the moral judgments that they do. Recent attempts to develop the details of the analogy between moral cognition and linguistic cognition as a guiding hypothesis for the study of moral psychology (Dwyer, 1999, 2006, 2009; Harman, 1999; Hauser, 2006; Jackendoff, 2007; Mikhail, 2007; Stich, 1972) suggest that there are at least three themes, central to generative linguistics, that can profitably be extended to and explored within moral psychology: (a) a distinction between competence and performance; (b) poverty of stimulus considerations; and (c) adopting the computational level as the proper level of empirical analysis.2
2.1. Competence and performance
Linguists have long recognized that linguistic performance includes a variety of complexities that stand in the way of directly modeling the cognitive principles that are responsible for language production and comprehension. Speakers utilize incomplete sentences and punctuate their speech with neutral vowel sounds like “um” and “er.” More importantly, the capacities to speak, understand, and acquire a natural language do not require facility with these complexities. Speakers of natural languages are able to understand and produce a discrete infinity of expressions. Thus, this ability must be grounded in a set of generative rules—a competence—that cannot merely be “read off” its output.
The importance of the competence-performance distinction is also captured and reinforced by contrasting the generative grammar of a language (or I-language) with the folk idea of “a” language (e.g., English, French, etc.). Languages (in the folk sense, or E-languages) are local, historically contingent phenomena that are grounded in economic, social, political, and cultural conventions. While the mind-external features of these languages are clearly grist for the sociolinguist’s mill, they are not the only plausible explananda for cognitive science. After all, descriptions of E-languages cannot explain the generative principles that yield our relatively unbounded capacities for linguistic expression and comprehension. Although such data do provide evidence for the necessary existence of an underlying linguistic competence, such descriptive corpora tell us only what is; they cannot begin to provide explanations of what is not, and why it is not. The set of humanly possible languages is not identical to the set of logically possible E-languages, and this raises an important question: What is the space of humanly possible languages? Answering this question calls for a deep understanding of human linguistic competence, which in turn requires discovering the constraints and boundaries on humanly possible languages. This cannot be achieved by a mere cataloguing of languages; it requires experimentation to assess what is acquirable, what is not, and why. Note, however, that this does not negate the importance of linguistic performance and the E-languages that are generated. Rather, it lays out a methodological strategy that distinguishes between mind-internal processes (both language-specific and more domain-general) and the systems that externalize the outputs of those processes (e.g., articulation and pragmatic mechanisms that are sensitive to contingent features of the speaker’s physical and social environment). Such a strategy thus distinguishes between linguistic knowledge (competence) and how that knowledge is put to use (performance).
Proponents of LA also distinguish moral competence from moral performance. Thus, LA highlights the importance of adopting a parallel distinction between I-morality and E-morality. E-moralities are historically contingent phenomena that are the result of economic, social, political, and cultural conventions (including a hodgepodge of rituals, customs, practices, institutions, and behaviors); I-morality, in contrast, includes the generative principles that are required if one is to acquire a morality at all. Although descriptive accounts of folk-moralities are of great interest from the standpoint of many psychologists and anthropologists, we contend that they will always be explanatorily incomplete. Moral competence requires the capacity to produce and understand an unbounded range of morally significant judgments and actions. We are frequently faced with previously unexamined cases where moral judgments must be made. As the ease of eliciting moral judgments by way of thought experiments suggests, we have little difficulty arriving at nearly immediate moral judgments even in unfamiliar cases. Thus, moral psychology cannot rest content with a mere catalog of extant moral judgments. After all, such extensionally characterized “moralities” cannot explain how we make moral judgments in unfamiliar cases; and this is the central issue to be addressed by cognitive scientists interested in the operation of our moral psychology.3 Moral psychologists, like linguists, owe us an account of why human moral systems take the forms they do, and not others. Again, while cataloguing moral systems is of interest and a critical first step towards delimiting a set of empirically tractable phenomena, it remains only a first step insofar as it fails to engage the possibility that there are other moral systems that could be acquired, and yet others that could never be acquired (Hauser, 2009).
On the basis of such considerations, proponents of LA suggest that humans are equipped with a moral faculty (FM) that allows us to acquire a morality. FM shapes our capacity for making moral judgments, and in this sense, the computational principles instantiated in FM constrain the range of moralities that can be acquired and deployed in moral reasoning. Paralleling the case of language, this approach positively demands an experimental approach to moral psychology—one that abstracts away from the moral performance manifest in debating the ethics of abortion or in acting in accordance with one’s moral judgments, to uncover the underlying cognitive mechanisms at play in the evaluation of the moral status of various actions. As we argue more directly below, studying the FM as an analog to the language faculty (FL) opens up a host of theoretical questions central to contemporary research in the cognitive sciences. Specifically, it calls for an account of which processes are unique to the FM and which nonmoral processes are required for making the sorts of moral judgments that we do. LA thus requires us to ask about the extent to which the cognitive processes involved in theory of mind, or action perception and segmentation, interface with FM, as well as the extent to which FM is a domain-specific system, and the extent to which it is informationally encapsulated. Additionally, LA raises new strategies for investigating the ontogenetic development of moral competence.
2.2. Poverty of the stimulus
Questions about the ontogeny of our moral judgments lead directly to an important fact about human moral psychology. Every typically developing child, in every part of the world, acquires the capacity to understand and produce a discrete infinity of expressions in one or more natural languages; yet no matter how atypical they are, chimpanzees do not, and neither do dogs, who are exposed to much of the same input as are children. Both empirical evidence (e.g., Crain & Nakayama, 1987; Crain & Thornton, 2006; Marcus, 1993) and conceptual arguments (at least back to Chomsky, 1959) suggest that the acquisition of a first language cannot be exhaustively explained in terms of domain-general learning mechanisms (Crain & Pietroski, 2001; Laurence & Margolis, 2001). The primary linguistic input that is available to children at various stages of development is impoverished relative to their linguistic capacities; so poverty of the stimulus considerations suggest that children must rely on domain-specific computations that shape the space of humanly acquirable languages.
Proponents of LA contend that our moral capacities are species-typical in the same sense as language is (Dwyer, 1999, 2006, 2009; Hauser, 2006; Mikhail, 2010).4 Again, although dogs adhere to some of the rules that we set for them, and appear to have a sense of fairness that may represent a derived cognitive trait (Range, Horn, Viranyi, & Huber, 2009), they do not appear to have an intuitive sense of moral rightness and wrongness of the sort that every typically developing child eventually acquires, one that includes a sense of “oughtness.” From the standpoint of LA, it becomes clear that any systematic explanation of human moral competence must be grounded in a clear sense of the capacities that children possess at various points in development. This is one respect in which moral psychology lags behind linguistics, which has been able to amass a plethora of evidence concerning what children know about their language and how they come to acquire that knowledge.
This is not, of course, to deny that philosophers, legal theorists, and theologians have—for quite some time—been interested in providing an account of the principles that underlie moral judgment. Indeed, the systematic investigations into the nature of moral rules that we find in Cicero, Aquinas, Grotius, Kant, and Sidgwick were attempts to do just this. Moreover, as Mikhail (2009) has recently been at pains to argue, the more recent attempt to systematize common legal rules in the Model Penal Code and the Restatement of Torts, among others, is a contemporary attempt to offer a systematic investigation of the underlying principles that are responsible for the structure of mature moral judgments. However, whatever else may be said about our knowledge of mature moral cognition, at present, we still do not have a comprehensive descriptive account of the developmental trajectory of moral competence. With this in mind, a few remarks are in order by way of motivating some new questions about the psychology of moral development.
First, the target explanandum in moral psychology is neither the presence of morally good behavior (cf. Dunn, 1999; Eisenberg et al., 1999; Hoffman, 1983) nor the ability to assert moral rules (e.g., “Lying is bad”). Such behaviors and assertions, though of great interest, have wildly different causes and, by themselves, suggest no account of their production. The psychological phenomenon that calls for explanation is how young children have any moral capacities at all. By 39 months, children distinguish between moral and nonmoral transgressions (Helwig & Turiel, 2002; Smetana & Braeges, 1990; Turiel, 1983). At approximately the same age, children recognize the special force of permission rules (expressed as conditionals) and, on the basis of this, seem able to attribute intentions to actors in ways that ground responsibility ascriptions (Harris & Núñez, 1996; Núñez & Harris, 1998). Furthermore, as Leslie, Knobe, and Cohen (2006) report, children manifest the so-called “side-effect effect” (Knobe, 2003): They construe unintended but foreseen side effects of an action as intended when the outcome of the action is bad, but they do not do so when the outcome is good. In addition to these rather subtle normative capacities, there are further important questions about how children acquire the specific moral signature of their cultures and bring their moral cognition to bear on their behavior.
Second, although children do receive some moral instruction, it is not clear how this instruction could allow them to recover moral rules (Dwyer, 1999). The world does not come with the required labels for distinguishing moral violations from violations of local norms of etiquette, and the explicit rules handed down by religion do not map onto the psychological distinctions that are in play in either various strands of moral philosophy or in the intuitive judgments that adults offer in both familiar and unfamiliar moral situations. Moreover, when children are corrected, it is typically by way of post hoc evaluations (e.g., “You should not have hit your brother”), and such remarks are likely to be too specific and too context dependent to provide a foundation for the sophisticated moral rules that we find in children’s judgments about right and wrong. Parents do, of course, offer universal imperatives (e.g., “Always keep your promises!”), but a key question remains regarding how children work out when exceptions to these explicit moral generalizations apply and when they do not. Moreover, it is unclear what can be inferred from these imperatives. What does a child learn when she or he hears “always” or “never” in relation to a norm transgression? These logical operators appear in moral and nonmoral contexts, but how do they interface with morality-specific contents? Furthermore, it seems that nonmoral (i.e., conventional) transgressions are corrected far more frequently than moral transgressions (Nucci, 2001; Smetana, 1989). This leaves open a puzzle: If correction is common for conventional transgressions, but infrequent for moral ones, how is a child to learn moral rules from standard reinforcement? Finally, most children do not commit or witness acts that would warrant correction or instruction of a kind that could account for their broad capacity to make moral judgments, especially when presented with unfamiliar, and often apocryphal cases. Unfortunately, such questions have yet to be directly probed from the standpoint of moral psychology—so we do not have clear answers regarding the precise developmental trajectory of our moral capacities. This is due, in part, to the fact that such questions only arise, and take on their significance, once it is recognized that there is a difference between moral competence and moral performance, and once the generative capacity for moral expression is recognized and explored as an empirical problem.
We acknowledge that, at present, it cannot conclusively be established that the moral input is impoverished relative to the child’s moral capacities. To establish that this is the case, we would need to have a richer account of:
The mature state of moral knowledge (which will benefit greatly from targeted analysis of mature responses to moral dilemmas discussed below, as well as investigations of commonalities and differences in legal codes and explicit moral theories);
The precise state of moral cognition at various stages of moral development;
The processes that underpin the acquisition of the capacity for moral judgment; and,
The precise connection, if any, between correction and patterns of change in moral judgment.
Without targeted investigations into these questions in the moral domain, we suggest that it is a live possibility that moral competence develops through a process analogous to language acquisition. The existing data suggest the following view of moral development: The human mind includes a biological mechanism that provides a limited range of possible moral systems, and on this basis, the environment selects among the cultural options on offer to acquire a particular morality. However, if future data reveal that children can and do simply extract moral principles from their environments, relying on nothing more than the pattern of corrections and affirmations that they encounter early in life, then poverty of the (moral) stimulus considerations will have to be abandoned. Similarly, if future studies reveal that there is no stable developmental trajectory for moral cognition, paralleling linguistic growth, or that the moral principles that children rely upon for moral judgments are intimately tied to the cultural environments in which they have been raised, then the claim that moral competence is richer than the available input will also seem less plausible.
Regardless of whether a particular poverty of the moral stimulus argument succeeds, we contend that researchers cannot determine the extent to which experience alone is sufficient to explain specifically moral development without first isolating the components of moral competence and considering how these interact with nonmoral aspects of the human mind. Only with such descriptively adequate understanding in hand can we tackle the problem of moral acquisition. This is a general methodological point that has typically been overlooked but that is made vivid by adopting LA.
2.3. How to study competence
The most successful method for recovering the computational principles of the FL has been driven by the elicitation of acceptability judgments, either from native speakers with no formal training in linguistics or from linguists themselves. Competent speakers are asked (a) to judge whether particular strings are acceptable in their language; (b) to rank strings in order of acceptability; and (c) to identify possible and impossible meanings of particular strings. For example:
1. *She liked pictures of each other
is judged unacceptable by competent English speakers (as indicated by the asterisk); and
2. Which film did you wonder which critic liked?
is ranked as more acceptable than
3. Which critic did you wonder which film liked?
4. The architect called the builder from Boston
String 4 will be judged to mean either that the architect made the call from Boston, or that the architect called a builder who was from Boston. However, no competent speaker will take it to mean that the architect was from Boston.
Linguists use acceptability judgments to test candidate grammatical rules that they believe to be instantiated in the computational architecture of FL. Ordinary speakers’ acceptability judgments are remarkably convergent and confident. However, those same speakers are typically unable to explain the computational principles that are responsible for their judgments of grammaticality.
As we noted at the outset, eliciting folk-moral judgments (as compared to explicitly theoretical moral judgments derived from consciously entertained moral principles) has long been a mainstay of moral philosophy. Philosophers have often constructed thought experiments to evoke moral judgments in support of a variety of moral theories (e.g., consequentialist and nonconsequentialist theories), principles (e.g., the Doctrine of Double Effect),5 and morally significant distinctions (e.g., between actions and omissions) (e.g., Foot, 1967; Quinn, 1989; Thomson, 1985).6 Moral psychologists, in contrast, evoke folk-moral judgments in the service of uncovering the cognitive machinery that is deployed in moral cognition, and to identify the features of actions and agents that are salient to moral cognition. In much the same way that individuals respond to the grammaticality of a sentence, individuals appear to spontaneously and confidently offer moral judgments in response to moral dilemmas. Proponents of LA contend that these intuitive moral judgments do not typically express reflectively held normative principles (Cushman, Young, & Hauser, 2006; Hauser, Cushman, Young, Jin, & Mikhail, 2007; Hauser, Young, & Cushman, 2008; Mikhail, 2007) but instead are driven by reflexive computations that can be uncovered by systematically manipulating the important features of moral scenarios and testing various target hypotheses about the representations over which moral cognition operates.7 Paralleling our comments above, LA would be at least partially defeated, then, if moral intuitions were typically and readily revisable. In other words, sustaining LA depends in part on establishing that the computations responsible for folk-moral judgments are distinct from and, at least relatively, immune to the results of conscious moral reflection on new, potentially morally relevant data.
Keeping these general motivations for pursuing LA in mind, we now turn to a set of experimental results that have been derived from LA, and that provide empirical support for this perspective on moral cognition. These results begin to provide more detailed answers to deep questions about how moral cognition works, thereby demonstrating the value of LA as an approach to moral psychology. Specifically, the work we discuss bears on the following central predictions of LA:
The computational processes responsible for folk-moral judgment operate over structured representations of actions and events, as well as coding for features of agency and outcomes.
Folk-moral judgments are the output of a dedicated FM and are largely immune to the effects of context.
In addition, this work underscores the complexity of questions about how the FM interfaces with other cognitive capacities.