In the following passage from Wittgenstein and the Vienna Circle dated December 1930, Wittgenstein is reported to have raised a remarkable question:

In this paper I am not interested in relating these early remarks to Wittgenstein’s later views on religious language, but rather in Wittgenstein’s question of whether talking is internally related to religion.

Wittgenstein’s comments are puzzling. At first glance, the remark beginning ‘I can well imagine …’ looks like a non-sequitur. Can we not describe a religious tribe with speech and practices that we would call ‘religious’ but nothing that we might call ‘doctrinal propositions,’ construed as a creed? For example, when we ask the tribe for a systematic expression of their beliefs, they do not understand us: doctrine has no role in this community. Perhaps Wittgenstein means by ‘doctrinal propositions’ not creeds but grammatical remarks such as ‘God loves you just as God loves everyone.’ He is suggesting that if there is religious speech, then there is a grammar and propositions that express this grammar (even if at this stage in his thinking he still considered moral and religious propositions as ‘nonsense’).

Clearly, Wittgenstein wants to do more than point out that religious speech must have a grammar, since he finds it ‘obvious’ that ‘the essence of religion cannot have anything to do with the fact that there is talking.’ Why is it obvious? And why does he write as if this remark is logically connected to the remark beginning ‘when people talk’? In fact, I cannot make sense of Wittgenstein’s early suggestion that there could be a religion without talking. I do not know what it means to speak of a religion, or a religious community, or a religious tradition to which speech is external.

Although I am not sure that Wittgenstein would have said in his later work that speech is external to religion, the view does survive in the work of some Wittgensteinians. I make my case in reference to Peter Winch’s classic paper ‘The Meaning of Religious Language,’ in which we have a more contemporary formulation of the idea that there can be a religion without religious speech. [2] In his illuminating paper, Winch deals with a variety of issues relating to religious language and belief. Particularly excellent are his reflections on the relation between theology and speech, the way in which religious practice is an aspect of religious belief, and the analogies between geometrical and theological propositions. I concern myself solely with that section of his paper in which he speaks of speech as something that can be ‘added to’ or can ‘grow out of’ primitive religious behavior.

Winch states his goal as ‘a characterization of religious language as such’ (193–4). He agrees with Wittgenstein’s comment from ‘Remarks on Frazer’s Golden Bough’ that in questions of religious belief and practice, ‘The practice does not spring from the view, but both of them are there.’ Rather than treating religious beliefs as the foundation of practice, Winch suggests that ‘what makes a belief a “religious belief” can best be understood by investigating the roots in religious practice of the concepts at work in religious beliefs. The possibility of doing this seems to depend on our being able to identify a set of practices as “religious” independently of any beliefs associated with them’ (p. 196).

So, what are the roots of religious concepts? Winch quotes the famous passage in Investigations § 244 in which ‘words are connected with the primitive, the natural, expressions of the sensation and used in their place.’ In this way, the adults ‘teach the child new pain-behavior.’ Winch suggests that ‘we might also think of the possibility that in the dawn of language, a ‘language of sensations’ gradually grew up and developed out of the ‘primitive’ (nonlinguistic) expressions of sensations. We could (and indeed can) recognize people (babies—as well as animals) as being in pain even though they have no concept of pain, in the sense that they have no “pain language” and no beliefs about pain’ (196–7).

Writing in response to a paper by Norman Malcolm, Rush Rhees criticizes the view of language as something that ‘grew up’ out of expressions of sensations. [3] He makes two points that I wish to elaborate. First, the concept ‘pre-linguistic’ is logically parasitic on the concept ‘linguistic’ and so cannot be the foundation for it. Speaking from within a community that has a pain-language, we can refer to a baby that cries in pain although it does not yet speak. But it makes sense to say that a baby cannot yet express pain in language only when a pain-language exists in which it could express itself. That is, without the language-game of pain, it does not make sense to refer to aspects of the game that are ‘primitive’ in the sense of either ‘pre-linguistic’ or ‘not based on reflection.’ Similarly, we could not show how a chess board is constitutive of the rules governing the movement of chess pieces if there were no such game as chess. [4] To see the point, consider someone who argues that the rules of chess emerged from the board that is its primitive source. This suggestion is nonsense because the board is not a chessboard unless there is the game of chess: the rules for chess could not, as it were, grow up out of a chessboard. The game of chess and the board on which it is played are internally related. In the case of pain-language and religious-language, the primitive responses are partially constitutive of the language. A board is only a chessboard within the game of chess, just as a human cry is only a cry of pain within a pain-language. [5]

Rhees argues that Wittgenstein never speaks of development from something prior to a language-game. It is not primitive reactions that determine the language-game of ‘pain’; rather, the language-game of ‘pain’ shows what is meant by ‘primitive pain reactions’. Rhees writes: ‘The primitive reactions may become important within a Sprachspiel, they show in the character of certain moves in the Sprachspiel’ (4). [6] The reactions that we distinguish as primitive are useful for bringing out features of language that we might otherwise not notice. For example, when giving a philosophical account of causality, we can indicate ‘looking in the direction of the cause’ as a primitive reaction that is partially constitutive of practices involving causal concepts. When giving a philosophical account of pain, we can appeal to facial expressions, sounds, and movements that in certain circumstances are partially constitutive of practices involving pain concepts. Pointing out unreflective responses within language-games is often philosophically useful, but it is altogether unclear what it would mean for language to ‘emerge’ or ‘dawn’ from these responses. Rhees remarks that if we see someone hurt, we can call him, ask him, get help, discuss matters, even walk away and be criticized for it. He asks pointedly: ‘How would this behavior “emerge” from the instinctive behavior in an animal? (2)’

Second, it is misleading to say that both human infants and animals share the same condition of ‘having no pain language or beliefs about pain’. If we conflate the cases of human beings and animals, we may be tempted to argue as follows: ‘Foxes have no pain-language, yet they manifest primitive pain responses, e.g., crying out in pain. Human babies have no pain-language, yet they manifest primitive pain responses, e.g., crying out in pain. Hence, the same primitive pain responses can exist whether or not a pain-language exists.’ The argument fails because although it makes sense to say that a baby cries in pain but cannot yet speak, it makes no sense to say that a fox cries in pain but cannot yet speak; this is one important difference in the criteria of identity for the concepts of human and animal pain. With a human community there is language, and within this community are individual human beings such as babies who do not speak the language; but in an association of foxes there is no language that an individual fox could either speak or fail to speak. [7] Thus, the contrast between ‘primitive’ and ‘non-primitive’ is different when applied to human beings and animals. With a human being, ‘primitive’ can mean ‘pre-linguistic’ or ‘non-reflective,’ but with an animal it cannot mean this. For example, we can use ‘primitive’ to contrast a baby fox’s primitive techniques for food gathering to an adult fox’s sophisticated strategy of ‘charming’ its prey, or the primitive climbing skills of the red fox to the well-developed skills of the gray fox. But how do such distinctions applied to animals shed light on the communication that characterizes human life?

Perhaps we argue as follows: ‘So, the expression “primitive” means something different when applied to foxes than it does when applied to human beings. Nevertheless, human beings and foxes are both mammals and both exhibit pain-responses. Why could pain-language not have developed among certain types of mammals but not others?’ It is not as if we can argue, however, that both human beings and foxes can be fearful or angry, with the difference between them being that human beings at some point began expressing their feelings in words. What feelings amount to in a human society and in an association of animals is different, which is closely connected with the point that language is internal to ‘society of human beings’ but not ‘association of animals’. For example, a fox could not expect something awful to happen next month, nor pretend to be fearful, nor express its anger in a hypocritical way. It is not that human beings perform additional behavior called ‘speaking’, but that a human way of living, which necessarily involves speaking and trying to understand one another, has a different kind of intelligibility from an animal way of living. On this theme Rhees writes:

We can object that animals, like human beings, have satisfactions and also difficulties to overcome. For example, beavers face problems when building a dam. Is it not simply that our difficulties are more numerous and complicated? Rhees writes:

The point is that the pain involved in religious, moral, or political conflicts is bound up with speaking and understanding one another in discussion, and discussion has no place in animal associations. This is the main reason that it does not help to appeal to animal behavior to try to give sense to the view that language emerges from primitive reactions.

Winch argues for an analogy between primitive pain behavior and primitive religious behavior. As we can dig down to the root pain behavior of pain-language, so we can dig down to the root religious behavior of religious language. He writes:

In this example, if someone asks us ‘what the tribe believes’, we would not know how to answer, although we would say that the tribe has religious rituals. Winch tells us that the ritualistic behavior is ‘in some way set apart from behavior associated with everyday practical concerns’ and is associated with a sense of wonder and awe at the tribe’s environment (p. 198). At the same time, Winch emphasizes that there are ‘connections between the ritualistic performances and other aspects of the tribe’s life’ (p. 198). He further suggests that we can expect different degrees of devoutness among the practitioners of the rituals. As long as Winch is arguing only that we can identify practices as religious even though we do not understand the tribe’s language, or even though we understand the language and can recognize nothing like an expression of belief, then the example is intelligible.

The trouble begins when Winch suggests that the tribe’s behavior may profitably be called primitive religious practice. By using the term ‘primitive’, he makes ‘an analogy between the relation of such cases respectively to one in which there is a developed pain language or a developed religious language’ (p. 197). In other words, if we want to understand how religious language can develop from primitive religious rituals, then we should reflect on how pain language can develop from primitive pain behavior. The relationship between instinctive pain responses and verbal expressions of pain is analogous to the relationship between religious rituals and religious language. Winch is quick to point out, however, that important conceptual differences exist between primitive pain behavior and primitive religious practice. First, with the religious, ‘we are already dealing with an established social practice’ rather than with individuals considered separately; second, pain responses are ‘pretty well universally the same’ whereas religious responses are not (p. 197). The crucial point is that Winch refers to religious language rather than religious belief, so he is not merely arguing that we can recognize what people do as religious even if we cannot describe ‘what they believe’. He seems to suggest that there could be religious practices without a religious language.

Winch argues that the primitiveness of the religious responses is shown in the ‘relation’ of the religion and pain cases to one another, but we can raise two logical objections. First, if it does not make sense for pain-language to develop from primitive pain behavior, then how is it analogous for religious language to develop from primitive religious behavior? It is confusing when Winch qualifies his analogy by telling us that the primitive pain behavior, unlike the primitive religious behavior, is not an ‘established social practice’. Certainly, an action in a ritual can be performed correctly or incorrectly, whereas the scream of a baby cannot. But the deeper logical point is, as we have argued, that the pain behavior of the infant can only be distinguished as primitive within the context of a community in which a pain-language is spoken. That is, the pain responses and the pain language must both be there. But in this case, language-games involving pain must already exist in the community, and so primitive pain behavior depends on an ‘established social practice’ in this sense no less than primitive religious behavior. The word ‘primitive’ in the case of pain cannot mean that it makes sense to distinguish human pain responses independently of pain-language, although it can mean both that babies exhibit pain behavior before they can speak, and that adult speakers exhibit pain behavior that is ‘primitive’ in the sense of ‘not based on prior reflection’. The logical problem with the pain-language example leaves it unclear what ‘primitive’ is supposed to mean when applied to religious behavior.

Second, the acquisition of language by an individual cannot show what is meant by the emergence of a new form of speech in a community. In the infant case, a community with a pain-language exists. Within this community, a child exhibits instinctive pain behavior but cannot yet speak—that is what makes the behavior primitive. Over time, the child can mature into an individual who exhibits instinctive pain behavior but who can verbally express pain. In the religion case, we are told that the community has no religious language. Within this community, people allegedly exhibit instinctive religious behavior, but because there is no religious language, the responses are called ‘primitive’. The community as a whole then may or may not develop a religious language. In the pain case, an individual comes to speak the existing language, but in the religion case, a new language emerges in the community. I do not see how the cases are analogous. A person who screams in pain but cannot yet speak the language of the community cannot be used to show what it means for religious practioners to be members of a community in which there is no religious speech at all.

In raising these objections, I am not suggesting that the word ‘primitive’ does not have an application both to pain responses and religious behavior; rather, I argue that there cannot be the sort of relation between them that Winch suggests. If we use ‘primitive’ to mean ‘not based on reflection’, for example, then an example of primitive pain behavior might be someone who spontaneously screams when his hand is burned, and an example of primitive religious behavior might be someone who spontaneously bows his head before an awe-inspiring sunset. The important conceptual difference is one that Winch does not note: either a baby or an adult can exhibit the primitive pain behavior of screaming when burned, but it is unclear what it would mean for a baby to exhibit primitive religious behavior. That is, an infant can react in pain before it can speak, but can it react religiously before it can speak? For example, can it spontaneously bow its head in respect before an awe-inspiring sunset? These are important conceptual differences between ‘primitive’ as applied to pain reactions and to religious reactions.

After setting the stage with his imaginary tribe, Winch writes, ‘I am suggesting, then, that such a context of behavior, even in the absence of any recognizably religious talk or belief, may already force the description “religious” on us. Suppose, however, that we do now add such talk to the original picture; that we think of such talk as growing out of the primitive ritualistic observances’ (p. 198). For example, talk of the gods who inhabit the mountains is introduced into the community. Winch’s first suggestion, that the behavior may ‘force the description “religious” on us’, is true enough if it means only that we can observe the practices of a people and describe them as religious even if, as it so happens, we do not understand their language. His second remark is an echo of Wittgenstein’s early remark to the effect that talking is externally related to religion, that it is something that it makes sense to ‘add’ to the picture. If I read Winch correctly, he is not suggesting that religious speech exists in the tribe but happens not to be spoken by the participants in the ritual; rather, he is suggesting that no religious speech exists at all.

Can we make sense of Winch’s suggestion that religious speech could be added to a tribe in which there are religious practices? We can easily imagine religious rituals in which speech plays no role. For example, the offering of incense at the altar during a Soto Zen service is typically conducted in silence, and the silence is an essential aspect of the ritual. Imagine that over time the custom develops of saying something personal of religious importance when offering the incense. For example, a monk may say, ‘The fire burns this incense in the way that awareness burns delusion.’ We may call the original ‘pre-linguistic’ practice more ‘primitive’ than the later one (although we might well say the reverse, since the change could be viewed as a corruption). Is Winch’s case of the ‘gods’ like this one? No, because in both the primitive and non-primitive cases, the Buddhist ritual has its intelligibility within a tradition with a religious language. The concepts involved in what is said during the ritual presuppose their intelligibility in other religious connections. Hence, there is a difference between the perfectly intelligible case of introducing religious speech into a silent ritual, and the Winch case of adding religious language to a society that has religious rituals but no religious language.

Because the silent ritual presupposes religious speech in other connections, let us try an example that avoids this problem. There are dozens – perhaps hundreds – of variants of chess: chess played on round, hexagonal, or three-dimensional boards; chess played between more than two players; chess played on small or large boards. To my knowledge, speech does not enter into the rules of any variant. [9] Assume that we invent a variant of chess in which speech is an important aspect of the game, e.g., each player is allowed twice per game to ask why the opponent made a particular move and must be given a truthful answer. The concepts of ‘explanation’, ‘honesty’, ‘lying’, and so forth enter the game of chess. We can refer to the pre-linguistic forms of chess as more primitive than speaking-chess. Just as chess existed for centuries without speech, and speech was eventually introduced into forms of the game, so religious practices in Winch’s tribe existed for centuries without religious speech, and speech was introduced into some of these practices. Are the cases analogous? No, because although the primitive variants of chess do not involve speech, a concept of ‘chess’ must exist in the language in order for the game to be played, for a person to ask someone for a game, for example. The point is not that an individual could not learn chess without talking, or that she could play the game without ever knowing what it was called (although this would be strange enough), but that without a language in we could distinguish chess from other games, discuss strategies of chess, invent new variants of chess, and so forth, we could not distinguish pre-linguistic chess from talking-chess. Winch’s suggestion seems to be that although religious rituals exist in the tribe, it true both that no religious speech enters into them and that there is no language in which they can be distinguished, described, or discussed.

Winch says that the rituals of the tribe are not disconnected from the life of the tribe. What happens in the rituals is related to things that are said and done at other times. But if the rituals are connected in this way, then the tribesmen must be able to talk about the rituals and the aspects of life with which they are associated; and if so, then why would this talk not be religious? In other words, even if there is no talk of the gods, the tribal rituals that we identify as religious are ‘set apart’ by the tribe from other practices, and this setting apart must be reflected in the speech of the tribe. The religious significance of the ritual for the tribe is partially constituted by the fact that it is talked about differently from other practices.

Why does Winch suggest that the tribal religious rituals are in some way external to tribal religious speech? I think the explanation has to do with the idea of religious concepts as ‘used’. [10] This idea is evident when he construes his primary question as ‘how do we identify a use of language as a religious use in the first place? (p. 193)’ Further, he suggests that ‘we first become acquainted with religious uses of language in becoming familiar with a particular religious tradition (or pretty limited range of such traditions)’ and in this way ‘acquire some concepts which we feel able to apply outside the context provided by these particular traditions’ (p. 194). The theme of use and application is continued when he writes that ‘how a term refers has to be understood in the light of its actual application with its surrounding context in the lives of its users’ (p. 200). The idea of concepts as used or applied is famously illustrated in Wittgenstein’s analogy of the toolbox. [11] The idea of a conceptual toolbox suggests a picture of language as a limited collection of concepts that can be applied in various ways in speech. Hence, what ‘existence’ means, for example, is revealed ‘in the way people apply the language they speak’ (p. 200).

If we view speech as a matter of using concepts, then it is tempting to think of the distinction between religious and non-religious language as the distinction between religious and non-religious concepts. Speaking a religious language involves using the concepts that belong to religious language. When looking for the concepts that belong to religious language, we may look for distinctively religious concepts such as ‘gods’. Is this a religious language? Well, which concepts are being used? Is this an auto mechanic’s toolbox? Well, how are the tools in the box being used? If we do not find uses that we would identify as ‘religious’, then we may then conclude that there is no religious language. And I think this is something like what Winch does when imagining the tribe in the pre-gods phase.

In his account of the tribesmen, Winch considers concepts that we identify as distinctively religious. He writes: ‘I am thinking of concepts like worship, reverence, religious awe, devoutness. These are concepts which we apply to human beings in certain aspects of their lives, demeanor, and practice, and which we think of as characteristic of descriptions of the religious dimensions (or lack of such dimensions) of people’s lives’ (pp. 194–5). What does it mean for concepts to be characteristically religious? Clearly, Winch is not thinking of characteristically religious expressions. For example, when he emphasizes that religious rituals have their significance in relation to what lies outside of them, he writes that ‘this [contrast] is connected with the ways in which expressions used in religious contexts differ importantly in their grammar and application from the same or similar expressions used in contexts where religious belief is not being explicitly emphasized’ (p. 220). The same expressions can have a religious or non-religious use.

To illustrate this point, imagine that I compile a list of various expressions used in Winch’s tribe before the concept ‘gods’ is introduced and give the list to someone who is familiar with the tribe. As a sort of game, I recite the items one by one and ask him to pick out the religious expressions. Let us assume that there are no expressions in the tribal language that are used exclusively in connection with religious practices. That is, if we were to give to our test-taker a list of all the expressions involved in or used to describe the religious practices, then he would not identify any expression as characteristically religious. For example, ‘bowing’ is on the list, but the tribesmen bow in all sorts of different circumstances (before washing their hair, perhaps) so ‘bowing’ is not an exclusively religious expression. Also, ‘ritual’ is on the list, but the tribe has rituals for beginning sporting events that have no religious significance. The test-taker may conclude that because there are no characteristically religious expressions, there are no religious concepts in the tribe, and hence no religious speech. If I then say that ‘gods’ is on the list, the test-taker may respond, ‘Well, that is surely a religious expression, so there is religious language in the tribe after all’.

Winch does not share the view of the test-taker. He does not identify an expression with an application in such a way that ‘religious language’ just means ‘such-and-such expressions’. When Winch speaks of ‘recognizably religious talk’ growing out of the primitive rituals, I take him to mean that recognizably religious uses of language grow out of the rituals—not recognizably religious expressions such as ‘gods’ or ‘holiness’. His view seems to be that the tribesmen are participating in the rituals, but are not yet using language religiously, which suggests a view of religious talk as the refinement of the techniques of a pre-existing religious practice. Does this suggestion make sense? In the hypothetical tribe, we are told that certain practices are set apart from other practices, although these practices are connected to other aspects of the life of the tribe. We are also told that the tribesmen vary in the degree of their devoutness. These are all distinctions that are made in the speech of the tribe. The tribe distinguishes bowing before the mountains when someone has died from bowing that is done at other times. The tribe distinguishes serious bowing from careless bowing. If a ritual exists in the tribe, then there is a distinction in the tribal language between ritual and non-ritual; if this ritual is set apart in a way that we identify as religious, then there is a distinction in the tribal language between what is religious and non-religious. I am arguing that this distinction in the speech of the tribe is partially constitutive of religious speech. Religious speech is certainly not reducible to anything like a set of technical expressions that we pick out from a list, but nor is it a matter of techniques added on to a pre-existing practice; rather, it is shown in what the tribesmen say during and about the practices that we identify as religious. What is religious is revealed in what the tribe says about birth and death, mountains and the sea, childhood and adulthood. If there are religious rituals, then there is talk of the rituals and the things with which they are connected, and so there is religious talk.

So far I have sketched the considerations that lead me to say that it does not make sense to have religious practices without religious speech. Let us formulate an objection. Assume a case in which none of the tribal practices that we identify as religious involve speaking by the participants, nor are they ever talked about by the tribe. Bowing to the mountains, marriages, and adolescent initiations are all conducted in strict silence. The religious practices are such that not only must they be conducted in silence, but must never be mentioned. For example, if a child asks of the people bowing, ‘What are they doing?’, then he is severely punished. Would this be a case of religious rituals without religious speech? No, because in suggesting that no religious language exists in the tribe, we are not saying that religious matters are not permitted to be talked about: we are saying that there is no concept of ‘religious matters’ at all. But if there are religious rituals in the tribe, then criteria of identity must distinguish rituals from non-rituals. It must be possible in the language to distinguish doing the ritual correctly, doing it incorrectly, and failing to do it. It will not do to object that a slap on the hand could take the place of saying ‘You’re doing the ritual wrong’ since the slap would simply be the way that one says ‘You’re doing the ritual wrong’ and so would involve a concept of religious ritual. What belongs to the ritual and what does not is reflected in what is said; the distinction between religious sense and nonsense is found there.

The point can be put a different way. The concept of ‘what should not be talked about’ distinguishes what it is and is not permissible to discuss. This concept has its intelligibility within the language of the tribe, which means that it is shown in what the tribemen say to one another. It is reflected in the circumstances in which the tribesmen speak as well as in what they talk about and fear to talk about. The concept ‘what should not be talked about’ cannot mean ‘what it does not make sense to talk about’ since the former presupposes that it makes sense to talk about it—otherwise what would it mean for it to be forbidden to talk about it? Forbidden to talk about what? When we describe the circumstances in which religious rituals are performed, when we give examples of what the people say in connection with what they do (e.g., ‘They are silent when bowing to the mountains, but chatter when bowing before meals’), then this is part of what it means to give an account of the religious speech of the tribe. In short, if a practice exists that we describe as religious, then there is speech that we describe as religious. The speech is not added to the practice, but both of them are there.