Semantic compositionality and Berkeley's divine language argument

Abstract

Critics of Berkeley's divine language argument usually dismiss it for one of two main reasons: (1) it appears to be a mere variation on Descartes's argument for the existence of other minds, or (2) there is too little similarity between human languages and the ‘discourse of nature’. I will first show that the compositional features of language on which Berkeley partially bases his argument include systematicity and productivity – not merely the generativity on which Descartes's is based. I will then show that the analogy between human languages and the discourse of nature is stronger than typically appreciated, even given contemporary understandings of language.

Introduction

In Hume's Dialogues Concerning Natural Religion, Cleanthes (Hume's advocate for theism) uses a thought experiment in which an articulate voice emanates from the clouds. This voice speaks to everyone in their native tongues, conveying religious and moral information. Philo (Hume's sceptic) concedes that such a voice would give us good reason to believe in God, even though ‘this extraordinary voice, by its loudness, extent, and flexibility to all languages, bears so little analogy to any human voice’ (Hume (1993), §3.3). What Hume appears here to concede, rightly, is that not all dissimilarities between two effects are of a kind that undermines an argument from analogy about their causes. It is deeply unintuitive that a mind could become so intelligent that it ceased to be a mind. In much the same way, it is deeply unintuitive that a language could become so universal or articulate that it ceased to be a language.

Philo remains unconvinced, of course, because neither he nor Cleanthes seriously entertains the thought that any such voice exists in the actual world. Instead, they immediately move on to discuss a ‘vegetating library’ in which books reproduce like biological organisms. After dismissing this second thought experiment on the grounds that no books in our universe actually reproduce in this way, they move on to discuss biological organisms themselves – and this because, supposedly, ‘the anatomy of an animal affords many stronger instances of design than the perusal of Livy or Tacitus’ (Hume (1993), §3.6). From a Berkeleyan perspective, however, Cleanthes has made a critical mistake in allowing this shift from linguistics to biology – for if he had mounted a genuinely linguistic design argument, Hume's various counterarguments¹ that follow would be largely beside the point.²

In the fourth dialogue of Alciphron Berkeley attempts to mount just such an argument. He contends that nature (or the world of sense experience) is a language in and through which the deity communicates with human beings just as directly as they do with each other. According to Berkeley, ‘we [see] God with our fleshly eyes as plain as we see any human person whatsoever, and he daily speaks to our senses in a manifest and clear dialect’ (ALC 4.14)³ – and there is no better evidence for the existence of another mind than to receive such linguistically ordered communication from it.

Critics have usually dismissed Berkeley's divine language argument (DLA) for one of two main reasons: (1) it appears to be a mere variation on Descartes's argument for the existence of other minds, and (2) the natural order appears to bear too little analogy to human languages, particularly given contemporary understandings of language. My goal in this article is to show that casual dismissal of the DLA for either reason is unwarranted, and consequently that the DLA can be useful in contemporary natural theology (especially if re-expressed in more modern terms).

My primary contribution pertains to the role of semantic compositionality in Berkeley's argument. A better understanding of this role shows both that Berkeley's insights into language exceed Descartes's and that the DLA enjoys greater initial plausibility than commentators have generally appreciated. In a key passage, Berkeley explains that it is ‘the articulation, combination, copiousness, extensive and general use, and easy application of signs’ (ALC 4.12) found in languages which distinguishes them from mere sign systems. Various scholars have seen in this passage an appeal to semantic compositionality, but to date the focus has been almost exclusively on generativity, or the recombination of component signs into novel compound signs. This is probably due to the parallels between Berkeley's divine language argument and Descartes's argument for the existence of other minds, since the generative use of signs is, for Descartes, the principal hallmark of intelligence. Such generativity, however, is only one of several distinctively compositional features of language. So, I argue, Berkeley also takes the productivity and systematicity of language into account. It is not only the ability of nature to compose appropriate stimuli out of basic signs that cries out for explanation – it is also our ability easily to understand these compositions, particularly those which we have never encountered before.

My secondary contribution is an analysis of contemporary literature on the DLA. When critics dismiss the DLA for reason (2), it is typically because they consider only a narrow range of similarities between sense perception and natural languages (such as ‘informativeness’ or ‘arbitrary signification’), identify one or more dissimilarities, and conclude – since an argument from analogy is only as strong as the similarity between the things compared – that the argument is weak. The similarities these critics consider, however, are often non-overlapping. Once we have (a) addressed the putative dissimilarities and (b) collated the various similarities already identified in the literature, the argument becomes more compelling.

Finally, I will discuss the relevance of the divine language argument both for those who do not share Berkeley's unique metaphysical views and for contemporary philosophy of religion in general.

Descartes and compositionality

Kenneth Pearce is one of the commentators on Berkeley most attuned to the compositional features of language – but he is interested mainly in how the view of nature as a language informs other aspects of Berkeley's philosophy and does not directly apply these insights to the divine language argument.⁴ He writes that Berkeley's requirement that the signs constituting a language be ‘apposite’ (ALC 4.7) means that they must be ‘organized into a system in such a way that it is easy to construct the complex signs appropriate to any given situation’ (Pearce (2017), 177). This sort of flexibility ‘is to be obtained by articulation and combination – in other words, by the compositionality of language. The basic signs must be put together to compose complex signs so as to be capable of dealing with indefinitely many situations’ (ibid., 177).

So far so good, but regarding the divine language argument Pearce says just that ‘it is the ability to judge which sign is apposite in a given situation that Alciphron takes as evidence of intelligent agency, [and] if this is Alciphron's point, then he is likely following Descartes’ (ibid.). Here I demur: there are more compositional features of language than the generativity that Pearce (and others) discuss, and more to Berkeley's argument than Descartes's argument for the existence of other minds in The Discourse on Method.⁵

Pearce is one of several who think that Berkeley's argument is an adaptation of Descartes's. According to David Kline, Descartes recognizes three features of linguistic behaviour that are the marks of intelligence: (1) the sign system is composed of ‘arbitrary’ signs, (2) there is generativity, or the novel use of signs through recombination of component signs, and (3) the linguistic behaviour exhibits understanding by being appropriate to the background environment. Kline spends most of his time on the third. He contrasts appropriate behaviour with the tropistic behaviour of machines and animals, writing that ‘we must be able to distinguish the object from its environment and then assign goals to the object. If, as the environment changes, the object behaves in accord with its goals we can speak of the behavior being appropriate’ (Kline (1987), 195). Applying these ideas to Berkeley's argument, Kline identifies the behaviour/environment distinction with the distinction between vision and the other senses and identifies the goal of the visual language with informing us about tangibilia. Because visual signs respond to widespread contextual variation in the environment in rule-governed ways, and always consistently with the goal of informing us about tangibilia, these visual signs are appropriate in the relevant sense.⁶

Kline's approach would indeed explain the primacy of visual language that Berkeley suggests when he writes of an ‘optic language’ (ALC 4.14) and that while ‘other senses may indeed furnish signs . . . It is the articulation, combination, copiousness, extensive and general use, and easy application of signs (all which are commonly found in vision) that constitute the true nature of language’ (ALC 4.12). However, I follow Pearce, Printz, Baldwin, and Olscamp, who all contend that Berkeley's considered opinion is that all the phenomena of nature constitute a language.⁷ They cite PHK §66, PHK §108, and Siris §§252–254 to make their case, but there is textual evidence within Alciphron as well – for example, when Berkeley says that God ‘speaks to our senses’ (ALC 4.14). Drawing a firm distinction between vision and the other senses is not mandatory in the face of these passages; it is more important that we do justice to the language/sign system distinction, since this is what undergirds the alleged primacy of vision.⁸

If we identify the divine language with the entire system of nature (as I think we should), then Descartes's behaviour/environment distinction is not available to us. That said, there is another even deeper problem for Kline's notion of ‘appropriateness’: the sort of linguistic behaviour that is best explained by an intelligent mind does not require any contextual variation whatsoever. A copy of the Iliad says exactly the same things no matter where I take it or what I happen to yell at its pages, and even if I begin reciting it aloud in the most inappropriate of contexts this does nothing whatsoever to diminish the conviction that it must have had one or more intelligent authors. What gives rise to this conviction is, among other things, that the linguistic behaviour is apposite in Berkeley's sense, not appropriate in Descartes's sense – the Iliad is composed of signs which, given the semantic and syntactic rules of the sign system, could be recombined to form an infinite variety of other equally intelligible sentences. Such generativity may help enable appropriate use, but such use is not essential to the inference we are trying to make.⁹

Further, while Descartes does recognize the generative capacities of language, he does not recognize (as Berkeley seems to) the infinite generative capacities of language. He discusses a machine's inability to ‘arrange words differently in order to respond to the sense of all that which will be said in its presence’ contrasted with the ability of humans to ‘invent for themselves various signs by means of which they make themselves understood’ (Descartes (1994), §§5.10–11), but at no point does he suggest that these generative abilities are unbounded or that they involve grammatical rules that can be applied recursively and therefore infinitely – instead saying only that it is ‘morally impossible that there would be enough different organs in a machine to make it act in all the circumstances of life in the same way as our reason makes us to act’ (ibid., §5.10). Berkeley, in contrast, writes that visual signs ‘being infinitely diversified and combined form a language’ (ALC 4.10) and that God communicates by ‘compounding them and disposing them to suggest and exhibit an endless variety of objects’ (ALC 4.12).¹⁰

Finally, while Descartes recognizes the generativity of language, he does not recognize the productivity and systematicity of language. Competent speakers of a language ‘can understand a large – perhaps infinitely large – collection of complex expressions the first time we encounter them’ (productivity), and ‘if we understand some complex expressions we tend to understand others that can be obtained by recombining their constituents’ (systematicity).¹¹ Both phenomena pertain to the radical intelligibility of language; it is one thing for a sign system to allow meaningful complex signs to be composed out of basic signs, and quite another for these meanings to be readily transparent to speakers of the language. This is why Berkeley is careful to add ‘general use and easy application of signs’ (ALC 4.12) to a list of linguistic features that already includes generativity (‘articulation, combination, variety, copiousness’). Since nothing can be easier or harder for an omnipotent spirit to do, and finite spirits are not directly involved in the production/recombination of most sensible signs, ‘easy application of signs’ can only refer to the ease with which finite spirits interpret the signs in the language of nature. Likewise, ‘general use’ suggests that our knowledge of the meanings of signs generalizes across contexts.

Berkeley further remarks that the language of nature ‘is learned with so little pains; it expresses the differences of things so clearly and aptly; it instructs with such facility and dispatch, by one glance of the eye conveying a greater variety of advices, and a more distinct knowledge of things, than could be got by a discourse of several hours’ (ALC 4.15). His argument is thus based (in part) on the fact that human beings not only have an unbounded ability easily and systematically to understand sentences of human language (including those they have never encountered before) but also an unbounded ability easily and systematically to understand sensory experiences (including those they have never encountered before) – and this by one glance of the eye. Just as compositional grammar is commonly cited as the best explanation of the productivity and systematicity of human languages, so is it the best explanation of the productivity and systematicity of our perceptions – and such compositional grammar plausibly belongs only to languages. That perception exhibits the same sort of productivity and systematicity may be a subject of debate, but it is hardly unreasonable in view of everyday experience. We encounter novel visual scenes all the time, and in most cases have no difficulty in drawing the normal range of action-guiding inferences and successfully navigating our environment.

The putative disanalogies

Most commentators freely concede to Berkeley that the natural world resembles human language insofar as our sensations have and convey rich representational content to us ‘not by similitude, nor yet by inference or necessary connexion, but by arbitrary imposition just as words suggest the things signified by them’ (ALC 4.10). Unsurprisingly, it takes more than this for a system of signs to form a language – a fact which ostensibly makes Berkeley vulnerable to critique. I will begin by showing that the putative disanalogies that have appeared in the literature fail to undermine the argument. I will then move into various additional positive analogies (beyond arbitrary signification) that previous commentators have noted.

The language of nature is one-directional

Donald Baldwin, E. G. King, and Désirée Park¹² object that while all the languages we are familiar with involve a community of speakers and two-way communication, the language of nature (if indeed there is one) appears to operate only in one direction: God speaks, whereas we only listen. To avoid a significant disanalogy with ordinary languages, we must also be able to communicate with God (and perhaps each other) in the language of nature.

Like Kenneth Pearce, however, I see no reason to think that the communication is only one directional.¹³ Pearce writes that ‘according to the theory of sense perception as language, our every interaction with the physical is a statement in an ongoing discourse with God himself’ (Pearce (2008), 256). If finite spirits have the power to produce ideas of sense in other spirits, as we seem to when we interact with shared environments or draw maps for each other (for example), there is nothing inconceivable about finite spirits speaking the divine language.

In principle, moreover, finite spirits could speak the divine language even if they turned out to lack this power (that is, if occasionalism were true, as some passages in Berkeley such as PHK §147 suggest). For if occasionalism were true, then neither would English-speaking finite spirits produce ideas in other spirits when they communicate in English. If God must causally mediate our communication in the divine language by taking our volitions as inputs and providing ideas of sense to other finite spirits as outputs, this no more disqualifies us from communicating in the language of nature than the mediation of an email server disqualifies me from communicating with my friend in English. The same goes for our communication with an infinite spirit. Though God may not perceive by sense, God knows the ideas of sense God has produced in response to our volitions; at worst, this is just one additional layer of transcription. We as humans are far from fully competent speakers of the language of nature (God alone is fully competent and/or fluent), but this is no obstacle to humans being speakers simpliciter.¹⁴

The language of nature lacks structure/instructors

James Danaher (2002) and Paul Olscamp (1970) argue that the natural world cannot be a language in the normal sense because perceivers are alone responsible for forming empirical concepts. In Berkeley's system, supposedly, there is no innate structure in the sensible world and it is merely our mental acts of consideration which unite ideas of sense into sensible bodies. If this is so, then God cannot communicate God's own concepts to us – and it is a basic requirement of any true language that it be ‘capable of producing internal states of understanding between communicants’ (Danaher (2002), 369).

One obvious reply would be that the liberty we have as individuals to form concepts is no obstacle to normal communication between human beings. Danaher counters, however, that the cases are not the same. With a language like English, he writes, ‘the liberty we naturally have to form concepts is eliminated by a myriad of instructors who reinforce the culturally acceptable extensions of a specific word or phrase in order that our concepts form in such a way as to reflect the meaning of our culture or language community’ (ibid., 369). In contrast, he claims, there are no such instructors for the divine language.

I reply that there is a perfectly reasonable sense in which we do receive instruction in the divine language. Although the contingent patterns of associations between ideas of sense that God ordains do not necessitate that we adopt any particular conceptual schema, it is certainly the case that some schemas conduce to our survival and well-being better than others. If we did not learn to group certain ideas of sense into individual animals or plants, for example, we would quickly starve. Thus, we are constantly being conditioned to bring our concepts into line with those of the author of nature, just as our linguistic communities constantly condition us to bring our use of words into line with other people. This is why Berkeley writes that by means of the language of nature ‘we are taught and admonished what to shun, and what to pursue; and are directed how to regulate our motions’ (ALC 4.7).

Expressions in the language of nature are not truth-apt

A closely related issue is whether expressions in the language of nature are capable of truth and falsity – for if they are not, this would mark a major point of disanalogy between divine and human languages. As Walter Creery writes, if expressions in the language of nature have no truth-values, then ‘one cannot say that the Author of Nature asserts anything’ (Creery (1972), 219) – and surely one of the main functions of language is to make assertions.

The most obvious solution would be to identify true expressions in the language of nature with veridical perceptions, and false expressions with non-veridical perceptions. Unfortunately, due to Berkeley's analysis of illusion, this is too simple to work. If there really is a bent visual percept of an oar, and if error exists only when someone infers that the oar will still appear bent when removed from the water (as Berkeley asserts at DHP 238), then the concepts of truth and falsity could apply to our sensations only on the level of suggestion. If truth and falsity are contingent upon the success of our inferences, then no expressions in the language of nature could be true or false in themselves (as at least some expressions in our human languages appear to be).

We could, of course, simply disavow Berkeley's analysis of illusion. Many contemporary philosophers of perception, in fact, do think that percepts are capable of truth and falsity. A visual experience of a tree has as part of its content, for example, that there is a tree situated in front of me. As Declan Smithies writes, ‘Perceptual experience is not just a matter of having sensations. Rather, perceptual experience represents its content with assertive force’ (Smithies (2019), 96). He argues, moreover, that ‘perceptual experience justifies believing its contents only because it represents its contents as true’ (ibid., 93, emphasis mine). If he is correct, there is no great mystery about the ability of expressions in the language of nature to take truth-values.

While this is an intriguing option, and while I would prefer that the divine language argument (like much of the rest of Alciphron) not depend on Berkeley's views in his earlier works, I also do not want it to depend on philosophical theses inconsistent with said views. Instead, the solution I favour is to specify what in the natural world corresponds to letters, words, and sentences, and point out that it is only some sentences that need to be able to have truth-values. Additionally, if the natural world contains these distinctions (that is to say, if it has an identifiable alphabet, morphology, lexicon, etc.) then the analogy between divine and human languages becomes that much stronger.

Berkeley occasionally refers to our sensations as ‘natural letters’ (Siris, §252).¹⁵ This is telling, especially once we consider that the letters in a book are what is immediately perceived when we read (DHP 174) and that ideas are grouped into bodies for the same reason letters are combined into words (PHK §65). The implication is that sensible bodies correspond not to sentences but to words or lexemes, and so perceptual experiences of particular objects (like the bent oar) are not the sort of thing that Berkeley ever intended to be truth-apt (any more than the word ‘oar’ is truth-apt).

So much for the letters and words. As for the sentences, my own view is that they are visual or sensory scenes.¹⁶ Unbeknownst to Berkeley, some contemporary vision scientists (chief among them Melissa Le-Hoa Võ) have begun studying so-called ‘scene grammar’ to better understand and explain visual cognition. On Võ's (2018) models, ‘objects in scenes – like words in sentences – are arranged according to a “grammar” which allows us to immediately understand objects and scenes we have never seen before’. Following Võ, an example of a visual scene that violates a semantic rule would be a Penrose Staircase or an Escher diagram (which represent impossible object-relationships). An example that violates a syntactic rule would be one in which a beer bottle is floating unsupported in mid-air (which coheres with the view that laws of nature are the syntactic rules of the language of nature). An example of a visual scene that merely violates a pragmatic norm would be one in which a kitchen sink is anchored to the ceiling.¹⁷ In the absence of a surrounding scene that enables us to make action-guiding inferences, we should no more expect the perception of a single sensible body to be truth-evaluable than we expect most words taken in isolation to be truth-evaluable.¹⁸

Pearce's approach is similar, minus the appeal to scene grammar. Whereas I wish explicitly to identify these sentences with sensory scenes, Pearce prefers to think of the meanings expressed in nature in more holistic terms. On his view, it is our ability to make empirical predictions based on our perceptions that endows the discourse of nature with meaning. Accordingly, the discourse of nature is about finite minds because it enables us to predict their experiences (see Pearce (2017), 202). Regardless of the further details of our respective views, we have the result about truth and falsity we want: ‘the language God speaks is perfectly capable of expressing falsehoods, for God could easily tell me things about other minds that are not so’ (ibid., 203).

The language of nature lacks a referential function

Creery objects that while human languages have a referential function, the language of nature does not – and moreover, cannot, given Berkeley's rejection of material substance. He writes that ‘the words and expressions of the phenomenal language cannot be said to be about anything at all’ (Creery (1972), 219).

Creery's mistake is in thinking that extra-mental objects are the only things to which the phenomenal language could refer. As shown above, sentences in the language of nature can be about finite spirits instead. This includes myriad statements about their volitions, perceptions, and dispositions.

The language of nature is universal and spoken by God

While criticizing the divine language argument, Olscamp writes that ‘among the significant differences between the two languages are the facts that the signs of nature are created by God, and that they are the same in all nations and climes, and for all people’ (Olscamp (1970), 38, emphasis mine). These are certainly disanalogies, but they do not matter.

It would be painfully ad hoc (if not circular) to cite the fact that the language of nature would be spoken by God (and not just humans) as a reason to reject an analogical argument intended to show that nature is a language spoken by God. And yet, Olscamp is not the only one to cite this fact as a ‘significant’ disanalogy. King, for example, writes that ‘the initial difference is that the communication involves God and man, and all other languages involve only men’ and correspondingly that ‘it is because the divine visual language is divine that we need to extend the word “language” in the first place’ (King (1970), 121). He acknowledges that this is not a ‘crushing’ objection, but I see no reason to view it as an objection in the first place.

Additional positive analogies

I have addressed the main disanalogies that critics have raised. If this seems like a short list, it is because most have critiqued Berkeley's argument for a paucity of similarities rather than for putative dissimilarities. Olscamp writes, for example, that ‘Berkeley did not hold that nature and artificial languages had all or even most of their properties in common, for clearly this would be silly’ (Olscamp (1970), 37). Is it silly? Let us see.

Several important positive analogies have come up already, such as those incidental to my comparison of Berkeley and Descartes. I now turn to the other positive analogies that commentators have noted.

The language of nature exhibits displacement

Paul Olscamp and Talia Mae Bettcher are the only two commentators specifically to note the role of displacement, as it is called in linguistics, in Berkeley's argument.¹⁹ Displacement is the ability of language to convey information about states of affairs arbitrarily distant in both space and time. Berkeley twice cites this ability as a similarity between human and divine languages (ALC 4.7 and 4.12). It seems uncontroversial that the language of nature exhibits displacement both temporally and spatially; a fossilized footprint can tell us that a dinosaur was present millions of years ago, and a telescope can tell us facts about objects light years away that help us reason about states of affairs in the here and now. Displacement is not a particularly compelling basis for an argument on its own (since mere signs may be able to exhibit it), but it is one more positive analogy to add to our list.

The language of nature can communicate without corresponding ideas

Olscamp helpfully notes that ‘Just as we do not have ideas occurring to us each and every time that we use signs in artificial languages, so the occurrence of images in actual cognition does not happen every time we use the signs of the natural language’ (Olscamp (1970), 37). This similarity fits well with Berkeley's anti-Lockean views of language more generally. Berkeley observes that ‘words may not be insignificant, although they should not every time they are used, excite the ideas they signify in our minds’ (ALC 7.5). Armed with this insight, nature resembles a language insofar as it can (and often does) communicate using non-iconic forms of representation – for example, ‘a threat of danger [is] enough to make us afraid, even if we do not think of any particular evil that is likely to befall us or even form an idea of danger in the abstract’ (PHK §20).

Closely related is a point by Danaher. In human languages, he writes, ‘a sound or letter could be missing in a given communication but we could still detect the meaning of the missing sound or letter because of the surrounding sounds or letters’ (Danaher (2002), 362). The same is true in the language of nature. We might never actually perceive the underside of a table, but we can still interpret our experience of the table appropriately due to the surrounding contextual clues.

The language of nature can express the same meanings through multiple modalities

Berkeley tells us that ‘visible figures represent tangible figures much after the same manner that written words do sounds’ (NTV §143). The relationship between written words and their corresponding sounds is not, however, one of normal reference or signification. Instead, the written word ‘table’ and the spoken word ‘table’ both signify tables. It follows that ‘visual ideas signify tangible stimuli in the way written words signify spoken words, rather than in the way words signify their referents’.²⁰ If we take sensible bodies to be the words in the language of nature, as I earlier discussed, then we have another positive analogy: both human and divine languages allow multiple representative modalities for the same meanings, for the same word can be represented using letters or sounds just as the same table can be (mediately) seen or touched.

The language of nature is interpretable absent a systematization of its semantics

Just as one need not be a linguist or a grammarian to understand a language, one need not be a scientist to navigate the world. As Berkeley observes, ‘a man may understand natural signs well without being able to say by what rule one event is a sign of another’ (PHK §108). Herein lies a further positive analogy: ‘As with human language, we can, in most cases, successfully interpret the perceptual language even in the absence of a rigorous systematization of semantics’.²¹

The divine language argument for non-idealists

An important concern remains: even if Berkeley is right, how do we know that the linguistic structure we observe is intrinsic to nature (i.e. imposed on it by another mind) rather than a product of our own cognitive apparatus (e.g. imposed on it by our own subconscious minds in order to process our experiences more efficiently, as Chomsky might have it)?

Here Berkeley's metaphysics can vindicate him. Plausibly, given immaterialism and the essential passivity of ideas, there could be no such thing as subconscious mental processes (much less mental activity taking place in a physical brain). The transparency of the mental precludes the possibility that the appearance of linguistic structure has an ‘internal’ source of which we are not consciously aware. If we can know by introspection that we are not the source of our ideas of sense (as Berkeley claims at PHK §146), then presumably we can also know by introspection that we are not the source of the rich and complex structure we observe in nature.²²

I believe this answer would satisfy Berkeley, but it is unlikely to satisfy those who are not already sympathetic to his metaphysical views. The lesson, perhaps, is that Alciphron is not as independent from Berkeley's previous works as he seems to have wished. Regardless, in the name of metaphysical ecumenism, I will briefly outline an alternative answer.

The alternative answer is that we do not know – and perhaps we cannot know, especially given how little we know about the extent to which our subconscious cognition conditions our perception. This matters little, however, for the same problem can be raised about our knowledge of other finite spirits. We have a tu quoque at our disposal: how do we know that the behaviour we observe in other human beings is intrinsically linguistic rather than a product of our own cognitive apparatus? We do not – but that does not stop our belief in other minds from being justified, and so it should not stop our belief in God from being justified.

Alvin Plantinga (1990) famously contends that the best argument for other minds shares the same defect as the teleological argument. Rather than accept scepticism, however, he concludes that the obvious rationality of belief in other minds entails the rationality of belief in God. Though their respective arguments share very little in substance, we can see Berkeley as engaged in the same kind of endeavour. As he states, the divine language argument gives us ‘as much reason to think God speaks to your eyes, as you can have for thinking any particular person speaks to your ears’ (ALC 4.12, emphasis mine).

Admittedly, Berkeley does then go on to claim that the operations of nature are ‘utterly inexplicable and unaccountable’ (ALC 4.14) by any scientific principles, which suggests he thinks the argument provides far better evidence for God's existence than we have for other finite minds. He is not under any obligation to prove this claim, however. All he needs to do is show that there is at least as much reason to attribute the language of nature to the operations of an external mind as there is to attribute the linguistic behaviour of another human being to the operations of an external mind. Such an inference could go through even if the operations of finite and infinite minds alike turned out to be entirely explicable by scientific laws. As King points out,

we so associate speaking with persons that we conclude there was a person nearby if we can rule out an animal or hallucination and so forth even if we cannot detect the person or even evidence that he was there. The move from speaking to a person is often a conceptual move. (King (1970), 122)

We are free to leave the causal relationship between languages and minds unanalysed so long as we are only trying to prove that knowledge of God is on an equal footing with knowledge of other minds.

Berkeley's argument need not be fully compelling on its own in order to be of interest and value to contemporary philosophers of religion, idealist or otherwise. This is because the argument rests on evidence largely incommensurate with the sort of evidence typically cited by traditional nomological, teleological, or cosmic fine-tuning design arguments. We could easily imagine, for example, a world with cosmological constants fine-tuned for life but which lacked the linguistic structure Berkeley claims to observe (or vice versa). However much or little Berkeley's argument ultimately adds to the credence we ought to give to theism, it should be cumulative with other a posteriori arguments.

Conclusion

While acknowledging that Berkeley has shown that nature exhibits some characteristics of language, King asks ‘would he be able to satisfy a description of language used in empirical linguistics?’ (ibid., 117). This is not a fair question, because a satisfactory definition of language in empirical linguistics has not been forthcoming. Perhaps that will change, but the best that we can do for now is to consider as wide a variety of generally accepted aspects of language as possible. In light of those that I have highlighted, we can read the divine language argument as a challenge: Berkeley defies us to find a plausible definition of language which is empirically adequate for all human languages while at the same time excluding nature itself.

Creery writes, ‘No doubt the forcefulness of the argument is increased if the set of characteristics identified for the concept of “language” is as large as possible’ (Creery (1972), 214). Accordingly, I have tried to show that this set is much larger than has been hitherto appreciated. Even so, I do not claim that Berkeley's argument is conclusive – there may be many other important analogies and disanalogies to consider²³ – but it is hardly ‘silly’, as Olscamp would have it.