Actions, reasons, and becauses

Abstract

How are sentences that express reason explanations related to sentences that express rationalizing psychological explanations? How are sentences like ‘Jane is going to the pub because John is there’ related to sentences like ‘Jane is going to the pub because she knows that John is there’? Are the former merely elliptical, in some sense, for the latter? Are the former used to express nothing more and nothing less than the latter are used to express? If so, then what explains this? Are the sentences syntactically ambiguous, context-dependent, lexically ambiguous, or is there some merely pragmatic explanation available? This paper examines various linguistic hypotheses which would support the “ellipsis hypothesis” and argues that they all fail to support the hypothesis. It then argues that this has deeper implications for first-order action theory than one might think, as the failure of these hypotheses lends support to an alternative first-order action theory I call analytic anti-psychologism.

1 Introduction

In this paper, I take up several issues concerning our use of the term ‘because’ and related expressions in the explanation of action, issues that, while often noted by action theorists, are rarely taken up in any detail. The central issue I will be concerned with is related to the use of the term ‘because’ in the following sentences:

• (1) Jane is going to the pub because John is there

• (2) Jane is going to the pub because she believes that John is there

• (3) Jane is going to the pub because she knows that John is there

The issue is related to whether (1) can be taken, in some sense, to be ‘merely elliptical’ for (2) or (3). Some theorists claim that in using a sentence like (1) we express little more than what we would express if we used a sentence like (2) or (3) instead.¹ As the view is often put: explanations like (1) are simply elliptical—in some sense—for explanations like (2) or (3). Of course, we express more by using (1) than we do by using (2) because (1) entails or presupposes that John is at the pub, whereas (2) does not. Indeed, (1) seems to entail or presuppose that Jane knows or at least justifiably believes that John is at the pub.² Something closer to an equivalence can be achieved if we focus instead on the relation between (1) and (3). The claim, then, is that (1) can be used to express nothing more and nothing less than what a sentence like (3) can be used to express. To make such a claim about (1) and (3) is to make a claim about the syntax, semantics, and pragmatics of these English sentences, a claim that would need to be defended by appealing to syntactic, semantic and pragmatic hypotheses. How is it that (1) can be used to express a relation between the fact that Jane is going to the pub and the fact that Jane knows that John is at the pub, when, on the face of it, the ‘because’ in (1) is used to express a relation between the fact that Jane is going to the pub and the fact or consideration that John is at the pub?

Why should the answers to these questions matter for first-order action theory, which attempts to provide an account of what it is to act for a reason (among other things), and need not particularly concern itself with how we express facts about acting for a reason? Well, in the first instance, there is an intimate relation between explanations like (1) and the notion of acting for a reason. If (1) is true on the relevant reading of ‘because’, then Jane is acting for a reason, and Jane’s reason for acting is the fact or consideration that John is at the pub.³ If all it takes for (1) to be true is just for (3) to be true, then whatever it takes for (3) to be true will suffice for its being the case that Jane is acting for a reason and for Jane’s reason for acting being that John is at the pub. Why this should matter is something I will get to in a moment.

Another reason why we should be interested in these linguistic issues is that the ellipsis hypothesis provides a quick linguistic argument for the view which has come to be known as psychologism.⁴ Psychologism, as I will understand it, holds that reason explanations like those expressed by (1) either just are rationalizing psychological explanations, or are wholly grounded in rationalizing psychological explanations like those expressed by (3). That is, psychologism holds that the explanation expressed by ‘Jane is going to the pub because John is there’ either just is the same explanation as that expressed by ‘Jane is going to the pub because she knows that John is at the pub’ or is wholly grounded in the explanation expressed by ‘Jane is going to the pub because she knows that John is at the pub’. The quick linguistic argument provides an argument for the first disjunct involved in this disjunctive understanding of psychologism. Of course, psychologism can be defended by other means than an argument to the effect that sentences like (1) are elliptical for sentences like (3). But an argument of this kind certainly provides a quick argument for psychologism. And such an argument is at least implicit in appeals to the ellipsis hypothesis.⁵

It may be worth relating my understanding of psychologism to other understandings of psychologism in the literature. Much of the debate over psychologism in the literature has been cast as a debate over whether “motivating reasons” are the kind of thing that appear in deliberation—considerations, propositions, facts, etc.—or consist of certain psychological states of an agent or facts about these. Within such a debate, psychologism is taken to be the view that motivating reasons consist of certain psychological states of an agent or facts about these.⁶ Since “motivating reason” is largely a term of art, it is hard to see what is at issue in the debate when it cast merely in these terms.⁷ It may be better to understand the debate as one over the relation between explanations like those expressed by sentences like (1) that appear to cite the kind of thing that might have appeared in an agent’s deliberation as that which explains what they are doing, and explanations like those expressed by sentences like (3) that appear to cite facts about an agent’s psychological states as that which explains what they are doing. Opponents of psychologism, like Dancy, are no doubt impressed by the fact that in explanations like (1) it is the very same kind of thing that might have appeared in the agent’s deliberation that seems to explain what they are doing. They may feel this kind of explanation ought at least to be recognised as a distinct kind of action explanation. They may even feel that it ought to be recognised as having priority over explanations in terms of the psychological states of an agent. I am understanding psychologism here as the denial of these thoughts: either what appear to be distinct explanations are not distinct (the first disjunct), or, if the two kinds of explanation are distinct, reason explanations are nonetheless wholly grounded in rationalizing psychological explanations (the second disjunct), and so are not (conceptually or metaphysically) prior to explanations of actions in terms of psychological states.⁸ Although I will primarily be concerned with linguistic and conceptual arguments for and against psychologism in what follows, I nonetheless take psychologism to be a metaphysical thesis, not a linguistic or conceptual thesis.

Returning to our theme, the deeper relevance of the quick linguistic argument for psychologism can be appreciated by reflecting on a related pair of sentences:

• (4) Jane is going to the pub because she believes that John is there.

• (5) Jane is going to the pub because she has a drinking problem.

Here, (5) is naturally taken to express an ordinary causal explanation. The fact that Jane has a drinking problem merely causally explains the fact that she is going to the pub. How are the uses of ‘because’ related in these sentences? Is ‘because’ ambiguous between (4) and (5)? Is ‘because’ used with different senses in (4) and (5)? It is plausible to think that ‘because’ is used with different senses in (4) and (5). Moreover, it is plausible to think that the sense of ‘because’ used in (4) is a hyponym, or a restriction of the sense, of the ‘because’ used in (5), so that the former expresses a species of causal explanation.⁹ If so, we have a quick linguistic argument for a causal theory of rationalizing psychological explanation, according to which rationalizing psychological explanations—explanations like (4)—are just a species of causal explanation.¹⁰

If both the quick argument for psychologism and the quick argument for a causal psychological theory of rationalizing explanation can be made out, then we have a quick linguistic master argument for the causal-psychological theory of acting for a reason, according to which reason explanations are a species of causal explanation: if an explanation like (1) is simply elliptical for (and so identical to) an explanation like (4), and (4) is a kind of causal explanation, then (1) is a kind of causal explanation.¹¹

My aim in this paper is to show that the quick linguistic argument for psychologism is harder to make out than many theorists assume. My interest in this argument goes deeper than merely showing that it fails to support what it is usually taken to support, however. Its failure points towards an alternative to both psychologism and causal theories of rationalizing psychological explanation. It will turn out that rejecting the argument has deeper consequences for the causal-psychological theory of acting for a reason than expected. The causal-psychological theory is, after all, a metaphysical thesis, just as psychologism is a metaphysical thesis, and it need not be defended on linguistic or conceptual grounds, as the quick linguistic argument attempts to do.¹² So it might not be expected that rejecting a linguistic or conceptual argument for the theory would have significant consequences for the theory itself. It does, however, since in rejecting the linguistic or conceptual argument, a conceptual argument for an alternative to the causal-psychological theory suggests itself. I call the alternative analytic anti-psychologism.¹³ According to analytic anti-psychologism, the concept of rationalizing psychological explanation has an analysis in terms of the concept of reason explanation and the concept of causal explanation. If this analysis is correct, then it precludes the relation of reason explanation having a conceptual or metaphysical definition in terms of the relation of rationalizing psychological explanation, as required by psychologism (on the second disjunct). While it might be true that reason explanations can be identical to or wholly grounded in rationalizing psychological explanations without the concepts of reason explanations or rationalizing psychological explanations being identical or without the former having an analysis in terms of the latter, if the concept of rationalizing explanation has a conceptual analysis partly in terms of the concept of reason explanations, it follows, on plausible assumptions, that reason explanations themselves are prior to rationalizing psychological explanations (both conceptually and metaphysically). If analytic anti-psychologism is true, then all forms of psychologism are false, and so too is the causal-psychological theory of acting for a reason.

2 Preliminaries

My primary focus in the first part of this paper will be on the relation between the ‘because’ which occurs in reason explanations like (1) and the ‘because’ which occurs in rationalizing psychological explanations like (2) and (3). In order to discuss theses about this relation, I am going to make an assumption about the relation between the ‘because’ which occurs in rationalizing psychological explanations like (2) and (3) and the ‘because’ which occurs in ordinary causal explanations like (5). The assumption is simply that ‘because’ is lexically ambiguous between rationalizing psychological explanations and causal explanations—it is used with different senses in each kind of explanation. Using ‘because_P’ to represent the ‘because’ of rationalizing psychological explanations, and ‘because_C’ to represent the ‘because’ of ordinary causal explanations, we can disambiguate the relevant sentences as follows:

• (6) Jane is going to the pub because_P she believes that John is there.

• (7) Jane is going to the pub because_C she has a drinking problem.

The main point of making this assumption at the outset is to allow us to disambiguate between pairs like the following:

• (8) Jane is going to the pub because_P she believes that John is there.

• (9) Jane is going to the pub because_C she believes that John is there.

The latter expresses an ordinary causal explanation with the fact that Jane believes that John is at the pub as its explanans. The former expresses a rationalizing psychological explanation with the fact that Jane believes that John is at the pub as its explanans. We can now more clearly state the question we will take up in the first part of this paper: what is the relation between the ‘because’ that occurs in reason explanations like (1) and ‘because_P’? Many of the proposals we examine in the first part of the paper attempt to understand the ‘because’ that occurs in reason explanations like (1) at least partly in terms of the meaning of ‘because_P’.

3 Reason explanations and the ellipsis hypothesis

Consider the following pair of sentences again:

• (10) Jane is going to the pub because John is there

• (11) Jane is going to the pub because she knows that John is there.

As I noted above, many theorists hold that when we use (10) to give an explanation of an action, this is just an elliptical way of expressing exactly the same thing we would express if we were instead to use (11). Call this the ellipsis hypothesis. Theorists are often not very precise about what they mean by saying that one explanation is elliptical for another.¹⁴ I take it the claim is meant to encompass the many different ways we could use a sentence like (10) to express the same thing we might express by using a sentence like (11). In this section I will examine various hypotheses about how we might use (10) to express the same thing that we use sentences like (11) to express. That is, I will explore various ways of vindicating the ellipsis hypothesis. We will see that all of the hypotheses in this section fail to do so. I devote the entire following section to a more promising alternative. I will begin with a hypothesis about the syntax of (10).

3.1 The syntactic ellipsis hypothesis

According to the syntactic ellipsis hypothesis, the ellipsis hypothesis can be explained in terms of the underlying syntactic equivalence of the sentences used to express the explanations in question.¹⁵ The model underlying this hypothesis can be illustrated by considering the following pair of sentences:

• (12) Jane believes that John is at the pub and Mark that John is home.

• (13) Jane believes that John is at the pub and Mark believes that John is home.

It is pretty obvious that (12) can be used to express exactly what (13) can be used to express. This is because (12) is syntactically elliptical for (13). What this means is that, syntactically speaking, (12) and (13) are one and the same sentence. (12) just has an unpronounced syntactic constituent ‘believes’.¹⁶ We can bring this out by presenting the sentence with the unpronounced syntactic constituent struck out:

• (14) Jane believes that John is at the pub and Mark believes that John is home.

The syntactic ellipsis hypothesis holds that sentences expressing reason explanations and sentences expressing rationalizing explanations are related in a similar manner. According to this model, we can capture the relation between (10) and (11) as follows:

• (15) Jane is going to the pub because she knows that John is there.

• (16) Jane is going to the pub because she knows that John is there.

If the sentences we use to express reason explanations are syntactically identical to the sentences we use to express the corresponding rationalizing psychological explanations, then it would not be surprising to find that the former can be used to express exactly what the latter can be used to express.

How plausible is the syntactic ellipsis hypothesis? Not very, unfortunately. The kind of syntactic ellipsis involved in (15) does not fit any known model of syntactic ellipsis, and such models have been extensively studied by linguists.¹⁷ The closest model is the one given above, where a repeated verb can be left unpronounced. There are two further, more general, problems, however. The first is that syntactic ellipsis typically requires the presence of the elided constituent in the (antecedent) syntactic context.¹⁸ This condition is met in (14) and it is met in examples like the following: ‘Jane is going to the pub, and John is too’ (where the elided material is the verb phrase ‘going to the pub’). This condition is not met in the case of (15) since the verb ‘knows’ does not appear in the (antecedent) syntactic context. The second problem concerns the way that the hypothesised elision in (15) crosses over constituent boundaries. Neither ‘she knows’ nor ‘she knows that’ is a single syntactic constituent in (15) in the way that ‘believes’ is in (14) and ‘going to the pub’ is in ‘John is going to the pub’; ‘She knows that’ is a noun-phrase and part of a verb phrase and is not a syntactic constituent of ‘Jane is going to the pub because she knows that John is there’. Syntactic ellipsis typically targets entire syntactic constituents and does not cross constituent boundaries like this.¹⁹ In my view, these are insurmountable difficulties for the syntactic ellipsis hypothesis. It is a good thing, then, that theorists who adopt the ellipsis hypothesis probably do not intend it as a syntactic hypothesis. Perhaps they only intended it as a pragmatic hypothesis. Let’s examine such a hypothesis now.

3.2 The Pragmatic ellipsis hypothesis

According to the pragmatic ellipsis hypothesis, while the sentences we use to express reason explanations and rationalizing explanations are not syntactically equivalent, and so may not semantically express the same thing relative to a context (as a consequence of this equivalence), the former can nonetheless be used to express exactly the same thing that the latter are used to express.²⁰

I will not attempt to provide a model for pragmatic ellipsis here, as it is hard to find uncontroversial models of pragmatic ellipsis.²¹ Operating on a relatively standard model of the syntax/semantics/pragmatics interface, there is a very tight relationship between what a sentence semantically expresses in a context of use, and what the sentence can be used to express.²² The former is tightly constrained by syntax and semantics. It is often held that what a sentence semantically expresses is just a matter of what the semantic values of the constituents of the sentence are relative to the context of use, and a matter of how these are combined as a matter of syntactic structure. What a sentence can be used to express is then closely related to what a sentence semantically expresses in a context. On such a view, there is no room for pragmatic ellipsis of the kind that would be needed to defend the pragmatic ellipsis hypothesis. Nonetheless, many theorists reject the tight link between meaning-in-context and what a sentence can be used to express. So I propose to examine the hypothesis on its own terms in the context of understanding action explanations. Consider the hypothesis, then, that someone could use a sentence like (17) to express exactly the same thing they could use (18) to express, such that (19) is an accurate representation of the ‘pragmatically provided’ material:

• (17) Jane is going to the pub because John is there

• (18) Jane is going to the pub because she knows that John is there.

• (19) Jane is going to the pub because she knows that John is there.

There’s a question to be answered about how it is that (17) can be used to express the same thing as (18). The answer would be a contribution to pragmatic theory. As such, it is likely to appeal to things like the communicative intention with which someone utters a sentence like (17). The idea might be something like this: when someone utters the sentence (17) with the intention of conveying a rationalizing explanation, they are understood to have the intention of communicating something like what they would communicate if they had instead uttered (18). (17) is thus “pragmatically enriched” in the way indicated by (19).

The problem with this theory, however, is that it is too unconstrained. Or, rather, it predicts that we should be able to use (17) to express things that we cannot. Suppose I intend to communicate to you that Jane is going to the pub because she hopes that John is there, or that she is going because she wants it to be the case that John is there. I should be able to use a sentence like (17) to express such explanations according to this hypothesis. But I cannot. This casts some suspicion on the pragmatic ellipsis hypothesis in this context. These considerations do not tell decisively against the pragmatic ellipsis hypothesis, but they do make it look decidedly ad hoc as an explanation of how (17) might be used to express the same thing as one would express with (18). In any case, as we will see in a moment, pragmatic theories are too weak. It is plausible that reason explanations at the very least semantically entail corresponding rationalizing psychological explanations.

A less ad hoc explanation would appeal to a weaker relation between uses of (17) and (18). Perhaps, at best, (17) can be used to convey an explanation of the kind that (18) can be used to directly express. This explanation would appeal to familiar forms of conversational implicature where I might say and express one thing but convey another thing.²³ Now a serious question for this hypothesis is this: what do uses of (17) express when they are used to convey or imply something else? It is implausible to claim that what they express are ordinary causal explanations, for then in many cases, what we say or express using sentences like (17) would be literally false. It is simply false that the fact that John is at the pub causally explains the fact that Jane is going to the pub. (If you need another example, consider ‘Jane is getting her telescope out because there will be an eclipse tonight’. Surely the fact that there will be an eclipse tonight doesn’t causally explain Jane’s getting her telescope out).²⁴ But when we use sentences like (17) we do not have a sense that what is said is literally false. The model is not like the case of using an obviously false statement in order to communicate something true; it is not like saying ‘Jane is one of the most unreliable people I know’ to convey that Jane is, in fact, reliable. We would immediately recognise the falsity of ‘Jane is going to the pub because John is there’ if it were understood as expressing a causal explanation. Yet we don’t.

Even supposing that the proponent of this kind of explanation could tell some story about what sentences like (17) are used to express, they’d still face serious difficulties in arguing that sentences like (17) can be used to conversationally implicate rationalizing explanations like those expressed by (18). The most serious problem comes from the need to explain away the fact that the supposed conversational implicature cannot be cancelled, and seems to be redundant when explicitly reinforced.²⁵ Consider:

• (20) Jane is going to the pub because John is there, but she isn’t going because she knows/believes that he is there.

• (21) Jane is going to the pub because John is there. She’s going because she knows/believes that he is there.

In the case of (20), when the first sentence is understood as a reason-explanation, it appears to be in strict conflict with what is asserted in the second conjunct (following ‘but’) and the implication cannot be cancelled. Compare ‘Jane is one of the most unreliable people I know. She never shows up on time’. Any implication that Jane is in fact reliable is cancelled by the follow-up sentence. This suggests that the relation between the two is something stronger than a conversational implicature.²⁶ Similarly, the second sentence in (21) sounds redundant, again suggesting that the relation between the two is stronger than a conversational implicature.

A further consideration against this hypothesis comes from considering conditionals like the following:

• (22) If Jane is going to the pub because John is there, then Jane is going to the pub because she knows that John is there.

If the sentence ‘Jane is going to the pub because John is there’ merely implied that Jane is going to the pub because she believes/knows that John is there, then this conditional should not appear to be obviously/trivially true. But (22) appears to express an obvious truth (if the antecedent is understood as a reason explanation). If so, then the pragmatic hypotheses we have considered in this section are simply not strong enough to explain what needs to be explained. Because the sentence ‘Jane is going to the pub because John is there’ appears in the antecedent of this conditional, it is not itself asserted, and pragmatic explanations do not come into the story. We need a stronger story in terms of the syntax/semantics of ‘Jane is going to the pub because John is there’ in order to explain the appearance of the obvious/trivial truth of this conditional. It is worth noting, at this point, perhaps, that the syntactic ellipsis hypothesis could explain this, since it holds that the conditional has the form ‘if A, then A’. However, as we have seen, the syntactic ellipsis hypothesis is otherwise implausible. If we are going to explain why (22) appears to be obviously/trivially true we had better look elsewhere. The next two hypotheses I consider (in the remainder of this section and in the entirety of the next section) would, if true, explain why (22) appears to be obviously/trivially true.

3.3 The context-dependence hypothesis

The first semantic hypothesis I will consider is the context-dependence hypothesis. According to the context-dependence hypothesis, ‘because’ is context dependent. When it is used to express a reason explanation, it expresses one relation, and when it is used to express a rationalizing explanation or a merely causal explanation, it expresses a different relation. Which relation it expresses is a matter of the context in which it is used. Examples of context-dependence are familiar. The indexical ‘I’ refers to a different individual depending on the context in which it is used. The adjective ‘tall’ picks out a different property or contrast class depending on the context in which it is used. Some claim that the verb ‘know’ may express a different relation of knowing depending on the context in which it is used. Or, to appeal to a different kind of example, sentences involving words like ‘imported’ and ‘local’ are context-dependent. They express different relations depending on the context in which they are used.²⁷ So why not claim that ‘because’ is context-dependent and appeal to this fact to explain the ellipsis hypothesis?

The claim that ‘because’ is context-dependent is independently plausible, since ‘because’-sentences already exhibit at least two kinds of sensitivity to context. Whether ‘The bridge collapsed because the bomb exploded’ is true may depend on whether what is at issue in the context is the bridge collapsing rather than the tower, or whether what is at issue is the bridge collapsing rather than staying in place. In a context where the bomb’s exploding caused the bridge but not the tower to collapse, and what is wanted is an explanation of why the bridge (but not the tower) collapsed, the sentence ‘The bridge collapsed because the bomb exploded’ might be taken to be false, since the bomb’s exploding does not provide the relevant contrast. It is plausible that ‘because’ sentences are content-dependent in at least this sense and that they are sensitive to a contextually salient contrast class.²⁸ Moreover, ‘because’-sentences, like other bits of explanatory language, seem to be sensitive in other ways to the interests of conversational participants. Since ‘because’-sentences are used to provide answers to ‘why’-questions, they seem to exhibit the same kind of sensitivity to interests that ‘why’-questions do.²⁹ ‘Why’-questions are often request for salient or relevant aspects of complex explanatory clusters. A ‘because’-sentence that fails to provide a salient or relevant aspect of an explanatory cluster in one context may be judged to be false, while being judged to be true in another, without there being a change in the explanatory facts.³⁰ Might we appeal to one of these forms of context-dependence to vindicate the ellipsis hypothesis? Might ‘because’ exhibit further context dependence that can be used to vindicate it?

In order to properly discuss the context-dependence hypothesis and many of the remaining hypotheses considered in this paper we are going to have a more sophisticated understanding of how the sentences we are interested in work. In the discussion so far, I have assumed without argument that ‘because’ expresses a relation between two facts. We can express the idea a little bit more formally by using an expression like the following to refer to the relation at issue: ‘p because q’. That is, we simply treat ‘because’ as expressing a relation between two facts or true propositions. This may be an over-simplification of the syntax/semantics of ‘because’. But it will serve our purposes well enough.³¹

Now there are various ways to develop the context-dependence hypothesis. The most plausible of these would attempt to appeal to some kind of contextual parameter with a range of variation. This is how context-dependence often works. But it isn’t clear how we might understand both reason explanations and rationalizing explanations in such a way that there is some parameter of contextual variation whereby in some contexts a sentence like ‘Jane is going to the pub because John is there’ would express the same thing as ‘Jane is going to the pub because she knows that John is there.’ One natural contextual variation here would be over the agent whose attitudes are relevant to explaining the action or event in question. But variation in an agent wouldn’t be enough to explain the difference between, say, an ordinary causal explanation of the form ‘Jane is going to the pub because she has a drinking problem’ and a reason explanation of the form ‘Jane is going to the pub because she has a drinking problem’. The difference between these two sentences cannot be accounted for in terms of variation in the agent argument of some relation. We seem to need a variation in the relation itself. Let’s consider such a view.

Suppose that in a context where Jane is the salient agent, the following explanatory relation is salient: that relation which holds between an action and some psychological state just when the psychological state is a state of Jane and it rationally explains that action. More formally, the relation at issue when Jane is the salient agent in a context is this: ‘p because_P Jane knows q’. This is a rather strange relation. It is a strange concoction out of the relation of rationalizing explanation (the relation expressed by ‘because_P’) and the partially applied relation of knowing: it is partially applied in the agent place of the relation. But let the strangeness of the relation pass. Suppose, in addition, that in contexts where there is no salient agent, the relevant explanatory relation picked up by ‘because’ is just the relation of causal explanation, that is: ‘p because_C q’. (This explains the merely causal reading of ‘Jane is going to the pub because she has a drinking problem’. Jane isn’t a salient agent in this context.).

We are now in a position to explain how ‘Jane is going to the pub because John is there’ could be used to express exactly what is expressed by ‘Jane is going to the pub because she knows that John is there’. Suppose we are in a context in which ‘because’ expresses the relation ‘p because_P Jane knows q’, and someone asserts ‘Jane is going to the pub because John is there’. Plugging the relevant relata (the fact that Jane is going to the pub, and the fact that John is at the pub) into this relation expressed by ‘because’ in this context, we get ‘Jane is going to the pub because_P Jane knows that John is at the pub’. This is, trivially, semantically equivalent to what is expressed by ‘Jane is going to the pub because_P she knows that John is at the pub’. In this way, we see how the context-dependence hypothesis might be appealed to in order to vindicate the ellipsis hypothesis. Note that it does so in a way that also explains why our conditional (22) appears to be trivially/obviously true. It thus improves on the pragmatic hypotheses.

It may be needless to say, however, but this is an implausible hypothesis. It is incredibly ad hoc. Sure, ‘because’ could be context-dependent in this way, but we have been given no independent reason for thinking that it is. Moreover, the relevant contextual parameter, relations of rationalizing explanation specialised to salient agents, is not a particularly plausible candidate for being a contextually available parameter. Compare: the speaker in a context (for ‘I’), a demonstrated object (for demonstratives), a contrast class (for causal explanations), a comparison class (for adjectives), a location being thought about (for contextuals like ‘imported’). On this view, the relation ‘p because_P Jane knows q’, would have to be salient in some contexts, ‘p because_P Jill knows q’, in other context, and so on.

In any case, there are interesting independent reasons against the context-dependence hypothesis. Like the pragmatic hypothesis it turns out not to be as constrained as it needs to be. Consider the following pair of sentences:

• (23) Jane is being examined by John because her test results were troubling.

• (24) John is examining Jane because her test results were troubling.

The thing to notice about these sentences is this: (23) has two readings: one on which the fact that Jane’s test results were troubling is John’s reason for examining Jane, and one on which the fact that Jane’s test results were troubling is Jane’s reason for being examined by John; (24) has only one reading. What this example shows is that ‘because’-adverbials pattern like other passive-sensitive adverbials, that is, whereas they allow for only one reading in active sentences, they allow for two in corresponding passive sentences. Consider, for instance, the following pair:

• (25) Jane is being examined by John intentionally.

• (26) John is examining Jane intentionally.

Here (25) has two readings, one on which Jane is doing something intentionally (namely, being examined by John) and one on which John is doing something intentionally (namely, examining Jane); (26) has only one reading, one on which John is doing something intentionally.

What’s the upshot? Well, if the context-dependence hypothesis were true, then (24) should have both readings. There’s no reason to think that we couldn’t utter (24) in a context where it is Jane’s relation of rationalizing explanation that was salient. But no such reading is available. This pair of sentences bring out nicely a more general problem: why is it that the available relation is always restricted to a relation belonging to some agent mentioned in the sentences involved? What this suggests is that we don’t have context-dependence at work here, but rather have something more tightly controlled by the syntax/semantics of the sentences.³² In active sentences, the agent argument is provided by the subject position, in passive sentences the agent argument is either provided by the “surface” subject or the “deep” subject, options uniquely available for the passive. There is much more to say about the syntax and semantics of passive sensitive adverbials, but we have already seen enough, in broad outline, to conclude that the context dependence hypothesis is not constrained enough.³³ Not only is it ad hoc, it is too unconstrained.

An interesting upshot of the observation that ‘because’-adverbials are passive-sensitive adverbials is that we have some motivation for assuming that there is a “hidden argument” place in sentences expressing reason explanations, a hidden argument associated with “because”. We have motivation for assuming that in such sentences ‘because’ expresses a relation between an agent and two facts or two propositions. So it may well be that the context-dependence hypothesis was half right when it appealed to relations like the following: ‘p because_P Jane knows q’. This relation is, of course, closely related to the following three-place relation: ‘p because_P x knows q’. Now, for technical reasons, the context-dependence hypothesis could not have appealed to this relation and held that the x argument is provided by context. This would have left the hypothesis without an explanation of the merely causal reading of the relevant sentences. Moreover, as we have just seen, appealing to this relation and the idea that the x argument might be filled by context would be too unconstrained. The argument is not provided by context but some kind of interaction between the syntax and semantics of the sentences involved. The hypothesis to be examined in the next section takes all of this on board, and treats the ‘because’ of reason explanations, a ‘because’ which is distinct from the because of rationalizing psychological explanations, as expressing a three-place relation ‘p because_P x knows q’.³⁴

Before examining this hypothesis, I want to briefly mention an alternative way of appealing to the context-dependence of ‘because’ to argue for psychologism.³⁵ One might concede that, in one sense, the explanations expressed by “Jane is going to the pub because John is there” and “Jane is going to the pub because she knows that John is there” are distinct. But one could then hold that, in another sense, there may only be one explanation here, with these sentences picking out different parts of it. Both sentences would be elliptical, in a sense, for a single explanation. This would be to appeal to the second kind of familiar context dependence involved in ‘because’ sentences noted above. It is a familiar point that most explanations are partial explanations. Jane may well be going to the pub because John is there. But she might also be going because her earlier plans fell through. One can think of these explanations as just picking out different parts of a fuller explanation or different parts of a single explanatory cluster.³⁶ Perhaps “Jane is going to the pub because John is there” and “Jane is going to the pub because she knows that John is there” just pick out different parts of the same explanation, then, depending on which part of the explanation is most salient in the explanatory context. Mightn’t an appeal to an idea along these lines be used to defend psychologism?

Now, I doubt that there is a single explanation or a single explanatory cluster of which both explanations are distinct parts. It is true that a single reason explanation is often part of a larger explanatory cluster consisting of other reason explanations. It is true that a single rationalizing psychological explanation is often part of a larger explanatory cluster consisting of other rationalizing explanations. But I don’t think that reason explanations stand to rationalizing explanations in this way, with both picking out distinct parts of a fuller explanation or distinct parts of a fuller explanatory structure.³⁷ Evidence for this comes from the difference between saying, “Jane is going to the pub because John is there and because her earlier plans fell through” and “Jane is going to the pub because John is there and because she knows that he is there”.³⁸ The former does, indeed, give a fuller explanation of why Jane is going to the pub than an explanation that just cited one of Jane’s reasons. The latter involves a kind of redundancy which suggests that the sentences do not pick out different parts of a single explanatory cluster. If anything, they seem to pick out the same part of a single explanatory cluster. In any case, even if this hypothesis were plausible, it would only allow us to say that, in some sense, “Jane is going to the pub because John is there” and “Jane is going to the pub because she knows that John is there” appeal to different parts of the same explanation (or explanatory cluster). This would not help the proponent of psychologism, since they want to claim that reasons explanations are elliptical for rationalizing psychological explanations, not that both explanations are elliptical for a broader kind of explanation involving both reasons explanations and rationalizing psychological explanations. Typically when proponents of the ellipsis hypothesis say that reason explanations are elliptical, they mean that they are elliptical for rationalizing psychological explanations, not that they are elliptical in some more generic sense.

4 Reason explanations and the lexical ambiguity hypothesis

The second semantic hypothesis we will consider is the ambiguity hypothesis. According to the ambiguity hypothesis (10), that is ‘Jane is going to the pub because John is there’, has a reading, or disambiguation, on which it semantically expresses nothing more and nothing less than what (11), ‘Jane is going to the pub because she knows that John is there’, semantically expresses. If a sentence is ambiguous (and is not syntactically or structurally ambiguous) then it must have one or more ambiguous constituents. According to the ambiguity hypothesis, ‘because’ is an ambiguous constituent. On one disambiguation of ‘because’ in (10), it semantically expresses nothing more and nothing less than what (11) semantically expresses.³⁹ The question for the ambiguity hypothesis concerns how the distinct senses of ‘because’ are related to each other in reason explanations and rationalizing psychological explanations.

The ambiguity hypothesis takes its cue from the fact that ambiguous expressions will often semantically express things on one reading or disambiguation which can be stated in ways which are closely related to the relevant meaning of the expression. So, for instance, the sentence ‘John hit the ball with a bat’ can be used to express something roughly equivalent to the sentence ‘John hit the ball with a hitting instrument’. In the case of polysemous expressions, in order to express what some expression is roughly equivalent to, it will often be useful to use the expression with a different sense. So, for instance, the sentence ‘Meet me at the mouth of the river’ can be used to express something roughly equivalent to the sentence ‘Meet me at the part of the river that resembles a mouth’. A somewhat contrived model for the ambiguity hypothesis can be given by expressions which have one sense which is just a narrowing of another sense the expression has. Consider, for instance, the technical term ‘dog’, which refers to male members of the canine species, and the ordinary term ‘dog’ which refers to members of the canine species. Let us use ‘dog₁’ for the former and ‘dog₂’ for the latter. The expression ‘Rex is a dog₁’ expresses nothing more and nothing less than the expression ‘Rex is a male dog₂’. This is plausible because the meaning of the technical term ‘dog₁’ can be given just as ‘male dog₂’. This example provides a model for the ambiguity hypothesis. We distinguish two senses of ‘because’, ‘because_R’ and ‘because_P’, and claim that the expression ‘Jane is going to the pub because_R John is there’ expresses nothing more and nothing less than the expression ‘Jane is going to the pub because_P she knows that John is there’.⁴⁰ This will be plausible if we can make good on the idea that the meaning of ‘because_R’ can be given at least partly in terms of the meaning of ‘because_P’. We will come to the ambiguity hypothesis’s “meaning hypothesis” after assessing the more general hypothesis that ‘because’ is ambiguous/polysemous.

4.1 The ambiguity/polysemy hypothesis

The hypothesis that ‘because’ is ambiguous/polysemous is very plausible. Standard tests for ambiguity offer some support for the hypothesis.⁴¹ Suppose Jane is going to the pub because John has a drinking problem. She is going to see if John is there and to check up on him if he is. Suppose John is going to the pub because he has a drinking problem. Now consider:

• (27) Jane is going to the pub because John has a drinking problem, and John is too.

• (28) Jane and John are going to the pub because John has a drinking problem.

These sentences strike many as odd, or, as linguistics would say, zeugmatic. One explanation of why they are odd is that they require a shift in the sense of ‘because’ in order to interpret them. For comparison, suppose that Jane took two apples and one banana from the bowl. Suppose she took three minutes to eat the apples. Now consider:

• (29) Jane took two apples from the bowl and one banana

• (30) Jane took two apples from the bowl and three minutes to eat them

There’s something odd about (30); it is zeugmatic. Examples like this involve syntactic ellipsis, which imposes the requirement of sameness of sense. That is why (29) sounds fine: we can fill in the ellipsis with a use of ‘took’ with the same sense. (30) sounds odd, however, since we cannot fill in the ellipsis with a use of ‘took’ with the same sense.⁴² So, according to a standard test for ambiguity, ‘because’ is ambiguous between reason explanations and ordinary causal explanations.

Another standard test for the ambiguity of a declarative sentence is the contradiction test (Gillon, 1990, pp. 407–410; Quine, 1960; Zwicky & Sadock, 1975, pp. 7–8; Sennet, 2023, 4.3). On one version of the test, we ask whether a particular declarative sentence can be conjoined with its negation and nonetheless has a reading on which it is true relative to some context. If it can, then this suggests that it is ambiguous. If it cannot, then this suggests that it is not. Consider:

• (31) Jane is going to the pub because she has a drinking problem, but she isn’t going to the pub because she has a drinking problem.

If ‘because’ sentences were ambiguous between reason explanations and merely causal explanations, then we should expect there to be a reading of (31) on which it could be true. Such a reading seems to be available. This can be brought out as follows:

• (32) Jane is going to the pub because she has a drinking problem (since that’s what people with drinking problems do), but she isn’t going to the pub because she has a drinking problem (this isn’t her reason for going to the pub, she might not even know that she has a drinking problem).

These two tests together suggest that ‘because’ is, indeed, ambiguous.

In support of the claim that ‘because’ is, specifically, polysemous, we can note that other explanatory language exhibits a similar kind of ambiguity, suggesting that the relation between the different senses of ‘because’ is systematic. Consider the following in relation to the above tests:

• (33) Jane is going to the pub for John is there.

• (34) Jane is going to the pub for the reason that John is there.

• (35) Jane’s reason for going to the pub is that John is there.

Contrary to the claims of some action theorists, even (34) and (35) do not always express reason explanations. This gives us extra reason to think we are dealing with a case of polysemy rather than mere homonymy since the difference in sense is systematic.⁴³

I propose to accept, at this stage in the argument, that ‘because’ has a specific sense when it is used in reasons explanations. I will use ‘because_R’ for this sense. We must now ask how plausible it is to think that ‘because_R’ and ‘because_P’ are related in the way that the ambiguity hypothesis holds that they are in order to vindicate the ellipsis hypothesis.

4.2 The psychological analysis

What must ‘because_R’ mean if ‘Jane is going to the pub because_R John is there’ is going to semantically express nothing more and nothing less than ‘Jane is going to the pub because_P Jane knows that John is at the pub’? The obvious thing for the proponent of the ambiguity hypothesis to do at this stage is model the case on the case of ‘Rex is a dog₁’ and ‘Rex is a male dog₂’, where we construct the meaning of ‘dog₁’ out of the meaning of the constituents of ‘Rex is a male dog₂’, the relevant constituents here being ‘male’ and ‘dog₂’. In the case of ‘Jane is going to the pub because_P she knows that John is at the pub’, the relevant constituents are ‘Jane’, ‘knows’, and ‘because_P’.

It is clear enough what the ambiguity hypothesis’s meaning hypothesis needs to be. We have already suggested, informally, that it will appeal to the following relation: ‘p because_P x knows q’. However, in order to state the meaning hypothesis in a perspicuous way, it will be helpful to introduce some notation. Let us use R for the concept expressed by ‘because_R’, P for the concept expressed by ‘because_P’, and K for the concept of knowledge. We can now present the meaning hypothesis as follows:

Psychological Analysis: R(x,p,q) $:=$ P(p,K(x,q))

This representation makes clear how the meaning of ‘because_R’ is to be understood in terms of the meaning of ‘because_P’ and the meaning of ‘knows’. Indeed, according to the ambiguity hypothesis, Psychological Analysis simply gives the meaning of ‘because_R’. It should be obvious how, on this hypothesis, sentences like ‘Jane is going to the pub because John is there’ can be used to express nothing more and nothing less than what is expressed by ‘Jane is going to the pub because she knows that John is there’. In virtue of the relation between the meaning of ‘because_R’ and the meaning of ‘because_P’, these sentences have exactly the same meaning. Moreover, we can see how this hypothesis explains why our conditional (22) appears to be trivially/obviously true.

We have seen, then, how we might construct the meaning of ‘because_R’ out of the meanings of ‘know’ and ‘because_P’ and how this hypothesis can be made, at least initially, linguistically plausible. And we have seen how the ambiguity hypothesis’s meaning hypothesis can be appealed to in order to explain the ellipsis hypothesis. Have we finally vindicated the ellipsis hypothesis, then? Well, no. While we have a hypothesis about the meaning of ‘because_R’ that supports the ellipsis hypothesis, we do not have an argument for the meaning hypothesis. I turn to arguments for and against the meaning hypothesis now.

4.3 For the psychological analysis

A potential form of consideration for or against the meaning hypothesis would come from linguistic precedent. If the hypothesis is just a particular instance of a more general polysemy phenomenon, this would be a significant reason in its favour. Unfortunately for the meaning hypothesis, however, there appears to be no plausible linguistic precedent for the relation between the meaning of the different senses it appeals to.

There have been many attempts to systematise the different ways that polysemous terms can be related. Many regularities have been discovered. Indeed, as Emanuel Viebahn and Barbara Vetter note, “there are a number of established patterns for spelling out the relatedness of meaning for polysemous expressions” (Vetter & Viebahn, 2016, p. 4). Many senses of polysemous terms are related in terms of metonymy, for instance. But these patterns and regularities are unlikely to tell us much about what to expect in the case of ‘because’ and other explanatory vocabulary, and this is because such vocabulary is part of the functional vocabulary of English and is a largely closed class. If the relation between the senses of ‘because’ does not have a model elsewhere this hardly counts against it, since it is plausible that ‘because’ and other explanatory vocabulary are sui generis.

That said, some general constraints may still be appealed to. Many senses of polysemous terms are generated by pragmatic strengthening. The basic idea behind pragmatic strengthening is something like this: “that an expression with a given initial meaning may, by a gradual process, acquire a further meaning, that is typically pragmatically implied by its initial meaning” (Vetter & Viebahn, 2016, p. 4). A plausible case can be made that this is how the causal and non-causal senses of ‘since’ are related. ‘Since’ may have been commonly used to pragmatically imply a causal relation. Over time, ‘since’ may have picked up a conventionalised meaning expressing a causal relation. Might the same be said for ‘because’? Perhaps. Perhaps sentences like ‘Jane is going to the pub because_P she knows that John is there’ pragmatically implicate that Jane’s reason for going to the pub is that John is there. Perhaps the use of ‘Jane is going to the pub because_R Jane is there’ developed out of this use. This may be so, but then it wouldn’t help the ambiguity hypothesis, for it looks like this use of ‘because_R’ would express something more than that Jane is going to the pub because_P she knows/believes that John is there, that is, something that goes beyond this. It is called pragmatic strengthening after all.

We cannot appeal to linguistic precedent, then, in motivating the meaning the hypothesis. How else might it be motivated? Well, it turns out that there is a familiar form of argument that might be taken to support the hypothesis. The argument appeals to the need to explain certain entailments between certain sentences. It suggests that the best explanation of the entailments is given by some hypothesis about the meaning of the terms involved. The basic idea here is that our conditional from earlier in the discussion, (22), represents an entailment of some sort. The conditional seems obviously or trivially true because the antecedent appears to trivially entail the consequent. More generally, and using ‘$\implies$’ for entailment, the following seems to hold:

Psychological Entailment: ‘X is V-ing because_R P’ $\implies$ ‘X is V-ing because_P X knows P’

In other words, sentences expressing reason explanations appear to entail sentences expressing rationalizing psychological explanations. Now, it is a familiar point that entailments like these provide data that theories need to explain. Here is Jerrold Katz putting the point nicely:

Saying an inference is valid is just saying that its conclusion is true in every circumstance in which its premises are, but saying that an inference is linguistic, or logical, or mathematical, or metaphysical is saying something about why the necessary connection obtains. The why may have to do with the sense structure of language, the laws of truth, the nature of number, or the a priori conditions of being. The explanation of necessary connection is a different matter in each such case. (Katz, 1988, p. 184)

We can initially agree that Psychological Entailment is an entailment, or represents a ‘valid’ inference in the sense identified by Katz: every circumstance in which (an instance of) ‘X is V-ing because_R P’ is true is a circumstance in which (a corresponding instance of) ‘X is V-ing because_P X knows P’ is true. Let’s reserve the unqualified ‘entailment’ for this kind of entailment. It seems that we can say something stronger about the kind of entailment involved here, however. Psychological Entailment additionally seems to be an a priori entailment, in the following sense: whenever we are in a position to know that (an instance of) ‘X is V-ing because_R P’ is true in some circumstance, we are, all other things being equal, in a position to know that (a corresponding instance of) ‘X is V-ing because_P X knows P’ is true in that circumstance on the basis of reasoning alone, and independently of having any further empirical justification for believing that ‘X is V-ing because_P X knows P’ is true. I propose that we accept that the entailment is both an entailment in the ordinary sense, and is an a priori entailment in something like the sense just isolated.

How might we explain the fact that Psychological Entailment is an a priori entailment? Well, one explanation is that it is an analytic entailment. It is widely agreed that analytic entailments are a priori entailments, even if not all a priori entailments are analytic entailments. Now, it must be noted that attempts to explain a priori entailments by claiming that they are analytic entailments have proven to be very unpopular in the recent philosophical literature. “ ‘X knows that P’ $\implies$ ‘X believes that P’ ” is certainly an a priori entailment if anything is, but many theorists reject the view that it is an analytic entailment. At the same time “ ‘X is a bachelor’ $\implies$ ‘X is unmarried’ ” is an a priori entailment if anything is, and most theorists would accept that it is an analytic entailment if anything is. Most theorists would accept it, of course, because they think that the meaning of ‘bachelor’ can be given, at least partly, in terms of the meaning of ‘unmarried’, whereas this might not be the case for the meaning of ‘know’ and ‘believes’. How plausible is it to think that Psychological Entailment is an analytic entailment, and that its being an analytic entailment in virtue of the Psychological Analysis is the best explanation of the fact that it is an a priori entailment?

Since the notions of analytic truth and analytic entailment are going to play a significant role in the discussion going forward, let me be clear about how I understand these notions. I take it that there is widespread agreement on uncontroversial paradigm instances of analytic truths and analytic entailments. ‘Every brother is male’ is an analytic truth, and ‘Jesse is Mark’s brother’ analytically entails ‘Jesse is male’. Other examples are familiar: ‘All bachelors are unmarried’, ‘All vixens are foxes’, ‘All sisters are female’ and so on. These are what Locke called trifling propositions. They are the kind of truth Kant famously attempted to characterise in terms of the notion of containment. One suspects that what characterises this class is that, at some level, these truths and entailments are going to be explained by appealing to the analysis of the meaning, say, of ‘brother’ in terms of the meaning of ‘male’, an analysis of the meaning of ‘bachelor’ in terms of the meaning of ‘unmarried’, and so on. Analytic truths, and analytic entailments, on this understanding, are truths and entailments which bear a particularly important relation to analyses. Analytic truths and analytic entailments “follow from” analyses of some constituent term.⁴⁴ If Psychological Entailment is an analytic entailment, then there must be some analysis of the meaning of the constituents such that one side of the entailment can be derived from a logical truth by means of substituting the term to be analysed with its analysis. This is just what an analytic truth is. My understanding hews close to the etymological connection between analyticity and analysis. As Fodor and Lepore write:

[A]nalyticity is after all supposed to be about analysis. The classic understanding of the notion required that concepts have constituent structure and that analytic sentences are true in virtue of the relation between complex concepts and their less complex parts. (Fodor & Ernie, 2006, p. 121)

As Fodor and Lepore go on to point out, on this conception it is easy to see why ‘Whatever is red is coloured’ is not analytic. Following Fodor and Lepore we can call this understanding of analyticity the “classical understanding”.⁴⁵ How does this bear on the issue of the ambiguity hypothesis’s meaning hypothesis? Well, as Fodor and Lepore note, the claim that some truth is plausibly analytic has traditionally been taken as motivation for particular hypotheses about the meanings of expressions (Fodor & Ernie, 2006, p. 118). Since we have seen that Psychological Entailment is plausibly an a priori entailment, we can, perhaps, argue for the Psychological Analysis on the basis of an argument to the best explanation of this fact. Such an argument would have to show both that the hypothesis that the entailment is an analytic entailment is more plausible than the hypothesis that it is an instance of the synthetic a priori. Moreover, it would have to hold that the Psychological Analysis is the best explanation of it being an analytic entailment.

Do we have independent reason for thinking that the entailment is analytic in this context, especially given the general scepticism about analytic entailment in the contemporary philosophical literature? It turns out that we do. As we have seen, it is plausible that the distinct senses of polysemous terms are related in systematic ways. We, in fact, have every reason to think that we will find analytic connections between the different senses of ‘because’ here. So although we may remain sceptical about analytic entailments elsewhere in philosophy, we have good reason for expecting one here.

The main challenge for the Psychological Analysis comes from establishing that it provides the best explanation of the fact that the entailment is a priori or analytic. What are the alternative explanations? Consider, for instance, the purported analytic connection between ‘knows’ and ‘believes’. As Timothy Williamson has argued, it does not simply follow from the fact that there is a priori entailment or a conceptual entailment, if there is one here, that the meaning of ‘knows’ can be analysed conjunctively, partly in terms of the meaning of ‘believes’. Indeed, perhaps the connection can be explained by analysing ‘believes’ in terms of ‘knows’ (Williamson, 2000, pp. 32, 43). Now, one thing a proponent of the meaning hypothesis can say at this point, however, is that in many of the philosophically interesting cases, the problem with attempted analyses has been with finding sufficient conditions. As the case of knowing and believing demonstrates, one central reason for doubting that there can be definition of knowledge in terms of believing is that nobody has ever produced non-open-ended sufficient conditions for knowing in terms of believing. The other famous example where there is significant reason for doubt is close to home. There is a necessary connection between rationalizing psychological explanation and ordinary causal explanation. One reason for doubting that rationalizing psychological explanations can be defined in terms of the causation of action by psychological states is that nobody has ever produced non-open-ended sufficient conditions for rationalizing psychological explanations in terms of the causation of action by psychological states. These reasons for doubt are absent in this case. At this stage, then, the proponent of the meaning hypothesis is in a position to claim that they have a good explanation of these purported analytic entailments. We have every reason to expect the entailments to be analytic. And there just doesn’t seem to be an alternative explanation of their analyticity available.

The general considerations in favour of the claim that ‘because’ is polysemous between reason explanations and rationalizing explanations, taken together with the argument above, provide good reasons for accepting the ellipsis hypothesis. Indeed, they provide good reasons for accepting analytic psychologism for, on this view, psychologism turns out not only to be true but analytically true. In what remains, I cast significant doubt over the plausibility of analytic psychologism. The main thrust of my argument is that we have better reasons for accepting an analysis of the meaning of ‘because_P’ partly in terms of the meaning of ‘because_R’ than for accepting an analysis of the meaning of ‘because_R’ partly in terms of the meaning of ‘because_P’. If so, then not only is psychologism not analytically true, it is analytically false.⁴⁶

4.4 Against the psychological analysis

As we have just seen, many of the standard reasons against claiming that Psychological Entailment is not analytic do not apply in our context. However, there are good reasons for thinking that Psychological Analysis is false and for thinking that Psychological Entailment is not analytic after all. There are two good reasons. The first is that Psychological Entailment fails a kind of linguistic test for analyticity. The second is that there are good reasons for thinking that the meaning of ‘because_R’ is basic, is not analysable in other terms.

Linguistic tests for analyticity are controversial and need to be handled with care. Nonetheless, they provide defeasible evidence for and against the hypothesis that some entailment is analytic. If some sentence S analytically entails some other sentence S’ there should be some temptation for thinking that someone cannot believe that S without believing that S’.⁴⁷ But it seems to be quite obvious that someone can believe that someone is doing something because_R so and so, without believing that that person is doing something because_P they know that so and so. Consider, for instance, the following pair of belief attributions:

• (36) Mark believes that Jane is going to the pub because_R John is there.

• (37) Mark believes that Jane is going to the pub because_P she knows that John is there.

Does (36) entail (37)? Is it possible for (36) to be true and (37) false? Arguably it is possible for (36) to be true and (37) false (even if we keep in mind that “Jane is going to the pub because_R John is there” a priori entails “Jane is going to the pub because_P she knows that John is there”). This can be brought out more clearly by considering examples in the first person:

• (38) I am going to the pub because_R John is there.

• (39) I am going to the pub because_P I know that John is there.

Is seems to be clearly possible for me to believe (38) and not to believe (39). What is the relevance of this claim? Well, if Psychological Analysis and Psychological Entailment are correct, then (36) should follow from (37) and I should not be able to believe (38) without believing (39). If Psychological Analysis is correct, then (36) and (37) should pattern like the following pair:

• (40) Mark believes that John is a bachelor.

• (41) Mark believes that John is a man.

Now, it is a familiar point that the substitution of analytically equivalent terms in belief contexts may fail.⁴⁸ I am not relying on the assumption that if there is an analytic connection between ‘because_R’ and ‘because_P’ then (36) should entail (37). I agree that it may be possible for (40) to be true and (41) false. My claim is that it takes a rather special contexts to block the validity of substitution of analytic equivalents in belief contexts. No such special context is needed to think that (36) might be true and (37) false, however. This is the sense in which (36) and (37) do not pattern like (40) and (41).

In fact, the striking thing about (36) and (37) is that they have the feel of a Frege case. They seem to ascribe to Mark belief in the same thing but under different guises. Mark may know that Jane is going to the pub because_R John is there. This is to know an explanation of Jane’s action under one guise. For Mark to know that Jane is going to the pub because_P Jane knows that John is there is to know the same explanation under another guise.

More striking still is that the situation is not entirely symmetrical. Consider the pair of belief attributions in reverse order now:

• (42) Mark believes that Jane is going to the pub because_P she knows that John is there.

• (43) Mark believes that Jane is going to the pub because_R John is there.

Does (42) entail (43)? Is it possible for (42) to be true and (43) to be false? Arguably is it not possible (if we keep in mind that (42) is meant to be a rationalizing psychological explanation), and the former entails the latter. In terms of guises, it seems that for Mark to know that Jane is going to the pub because_P Jane believes that John is there under the guise associated with rationalising psychological explanations is also to know that Jane is going to the pub because_R under the guise associated with reason explanations.

This asymmetry spells trouble for Psychological Analysis and Psychological Entailment. It suggests that, if anything, we should expect an analysis of the meaning of ‘because_P’ in terms of the meaning of ‘because_R’ rather than the other way around. As I shall argue below, such an analysis is available, and so can be taken to support our judgements about these cases.

A further consideration against Psychological Analysis comes from the fact that the concept of reason explanation is a good candidate for being a basic concept, a concept not defined in terms of any other concept. Just as the concept of causation may have no definition in terms of more basic concepts, the concept of being a reason for action may have no definition in terms of more basic concepts. Such a view can be made plausible by considering just how primitive the explanation of action in terms of an agent’s reasons seems to be. We are able to give and understand explanations of our own actions and the actions of others, in terms of our reasons for acting, it seems, without being able to understand third-person rationalizing psychological explanations. The child explains that they hit the other child because that child hit them. The child can produce this explanation without understanding an explanation of their own action in terms of their psychological states. Indeed, it isn’t obvious that the child needs to understand any other explanatory practice in order to understand reason explanations. This makes Dancy’s claim that the concept of reason explanation may well be basic a plausible one:

[W]e have to allow that we have offered no analysis or philosophical account of the ‘in the light of relation’. I suspect, however, that no such analysis or account is available to be given.... It is what it is, and not another thing; and if it cannot be analysed, so much the worse for the more global pretensions of analysis. (I agree, however, that it would be good to produce an account of the ‘in the light of’ relation—if one could only think of some way of producing one.) (Dancy, 2000, 163)

This point reinforces our previous one. To the extent that we think that the meaning of ‘because_R’ is basic, we shouldn’t expect Psychological Entailment to be analytic in virtue of the meaning of ‘because_R’.

I take these points to show that Psychological Analysis is false. Since the truth of Psychological Analysis was the last hope for making good on the ellipsis hypothesis we can now conclude that the ellipsis hypothesis is (probably) false. It is not the case that sentences used to express reason explanations are used to express nothing more and nothing less than what sentences used to express corresponding rationalizing psychological explanations are used to express. There is no quick linguistic argument for psychologism.

At this stage, a proponent of psychologism might think: so what? If the argument of this section is correct, all we have seen is that there is no quick linguistic argument for psychologism and that analytic psychologism is false. Psychologism is not a linguistic thesis, but is offered as a metaphysical thesis.⁴⁹ Indeed, it can be argued for on the basis of the need to explain why Psychological Entailment is an ordinary entailment even if it is not an analytic entailment.⁵⁰ We need an explanation of that entailment. And psychologism provides one since it tells us that what it is for a reason explanation to obtain just is for a rationalizing psychological explanation to obtain.

The proponent of psychologism cannot rest at this stage, however. If the proponent of psychologism wants their psychologism to be part of a causal-psychological theory of reason explanations, then they need to make good on the idea that rationalizing psychological explanations are a species of causal explanations and they need to do so in a way that avoids appealing to the notion of reason explanation. Why the latter restriction? Because they have already understood reason explanations in terms of rationalizing psychological explanations. They can hardly turn around and understand rationalizing psychological explanations in terms of reason explanations! That would make for a tight circle. As I argue in the next section, however, rationalizing psychological explanations might best be analysed partly in terms of reason explanations. We are partly encouraged in this view by the observation we made in this section that there appears to be an analytic entailment running from rationalizing psychological explanations to reason explanations. If our analysis can explain that entailment, that is a mark in its favour. If our alternative analysis is correct, the proponent of psychologism is in trouble. Not only will it turn out that is psychologism false. It will turn out that it is analytically false.

5 For analytic anti-psychologism

The motivation for analytic anti-psychologism is twofold. Analytic anti-psychologism is motivated, in the first instance, by an old puzzle about rationalizing psychological explanations. It is motivated, moreover, by the need to explain the analytic entailment running from rationalizing psychological explanations to reason explanations. At the heart of analytic anti-psychologism is an analysis of rationalizing psychological explanation partly in terms of reason explanations. The challenge is to offer an analysis which explains both the analytic entailment running from rationalizing explanations to reason explanations and the analytic entailment running from rationalizing explanations to causal explanations. We begin our discussion with the latter entailment and the analysis it is usually taken to motivate, namely, the causal analysis of rationalizing psychological explanations.

Earlier we briefly discussed the following explanations:

• (44) Jane is going to the pub because she believes that John is there.

• (45) Jane is going to the pub because she has a drinking problem.

We assumed that ‘because’ is ambiguous between rationalizing psychological explanations and ordinary causal explanations like these. The intuitive motivation for this claim comes from considering sentences like (44) and noting that they appear to have an interpretation on which ‘because’ just means whatever it means in (45) and also has an interpretation on which what is required for the sentence to be true is more than just that there be some causal connection between the psychological states and the action. Indeed, there are famous examples which show that it is not enough for the state to both rationalize the action and cause it. Something else needs to obtain. In other words, on this interpretation, it is not enough for the truth of (44) that Jane’s belief be among the causes of her going to the pub. As Davidson argues, it is not enough for the climber’s belief that her climbing partner poses a danger to her to be among the causes of her dropping the rope for it to be true that the climber dropped the rope because she believed that her partner was a risk to her (Davidson, 2001, p. 79).

Now, the hypothesis that ‘because’ is ambiguous between (44) and (45) is a linguistic thesis. And to establish it, we would have to consider the full range of linguistic evidence and consider alternative explanations of how these explanations are related. The options are analogous to those we considered in the case of reason explanations (except there is no analogue of the syntactic explanation). One might attempt to argue that sentences like (44) merely pragmatically imply that Jane’s belief caused her action in the right kind of way. One might attempt to argue that ‘because’ is context-dependent. Rather than rehearse the arguments here, I will simply point out that they face similar difficulties as those we have already covered. Moreover, in the case of (44) and (45) there is linguistic precedent for the meaning hypothesis that would accompany the claim that ‘because’ is ambiguous here. There’s a good case for thinking that ‘because_P’ expresses a species of the relation expressed by ‘because_C’. Since nobody tries to draw any interesting conclusions about rationalizing psychological explanations from a linguistic claim about them, I propose that we continue to assume that ‘because’ is ambiguous between (44) and (45).

Granted this, the question remains: how exactly are the two senses of ‘because’ related here? Well, we get a clue in this case by considering the fact that rationalizing psychological explanations entail causal explanations. Consider:

• (46) If Jane is going to the pub because_P she believes that John is there, then Jane is going to the pub because_C she believes that John is there.

The idea here is that the ‘because’ of rationalizing psychological explanation entails the ‘because’ of ordinary causal explanation, the ‘because’ which applies equally in cases of deviant causation, like the case described by Davidson above.⁵¹ We can generalise the point to arrive at the following entailment:

Causal Entailment: ‘X is V-ing because_P X knows that P’ $\implies$ ‘X is V-ing because_C X knows that P’

The linguistic tests we used in the previous section give us some evidence that this entailment is an analytic entailment. If so, it might be thought that it can easily be explained on a (partial) analysis like the following:

Causal Analysis: P(p,K(x,q)) $:=$ C(p,K(x,q)) & $\ldots$

According to Causal Analysis, the meaning of ‘because_P’ has an analysis partly in terms of the meaning of ‘because_C’. The analysis is partial because the meaning of ‘because_P’ clearly goes beyond the meaning of ‘because_C’. If Causal Analysis is correct, then we have a straightforward explanation of why Causal Entailment is an analytic entailment. If it is the best explanation, then we may have good reasons to accept it.

Now, of course, Causal Analysis is closely related to the familiar causal theory of rationalizing explanation, which holds that rationalizing psychological explanations are a species of causal explanation. Indeed, the Causal Analysis entails that the causal theory is correct. Yet few proponents of the causal theory of rationalizing explanation today would hold that the theory is analytically true. This may be partly due to a resistance to claims about analyticity in general. But it is more likely to be due to a familiar problem: the problem of deviant causation.

The problem of deviant causation is, in short, the problem of filling in the dots ‘$\ldots$’ in the Causal Analysis in such a way that one can provide not only necessary conditions for the truth of ‘X is V-ing because_P X knows that P’ but sufficient conditions. There have been many attempts to do so, and more and more informative necessary conditions have been provided. But, it would be fair to say that nobody has succeeded in providing an analysis which provides sufficient conditions for the truth of ‘X is V-ing because_P X knows that P’.⁵² This is sometimes taken to be an issue for the causal theory of rationalizing psychological explanation. However, as proponents of that theory are fond of pointing out, the theory isn’t committed to the relevant sufficient conditions being non-open-ended in the way required for conceptual analysis. It seems to me to be a moot point as to whether the problem of deviant causation is a problem for causal theories or not. What it does seem to be a serious problem for is the Causal Analysis. It casts serious doubt on the idea that we can analyse the meaning of ‘because_P’ in terms of the meaning of ‘because_C’. One cannot claim that Causal Analysis is a partial analysis without giving good reason for thinking that there is a complete analysis that it is partial with respect to. It is doubtful that there is one.

These are familiar reasons, then, for thinking that the Causal Analysis may not provide the best explanation of the Causal Entailment. This point could be driven home, however, if we had an alternative explanation available. I want to suggest that an alternative analysis may have been sitting under our noses this whole time. I have hinted at the basic idea a few times now: we might attempt to understand the meaning of ‘because_P’ in terms of the meaning of ‘because_R’. This is certainly an option that emerges given the failure of attempts to analyse the meaning of ‘because_P’ in terms of the meaning of ‘because_C’. There is an obvious reason why proponents of the causal theory of acting for a reason cannot help themselves to this option, however, because it would rule out psychologism. They can hardly analyse ‘because_R’ in terms of ‘because_P’ and then turn around and analyse ‘because_P’ in terms of ‘because_R’. That would be a tight circle indeed. Perhaps it is this circle that Davidson had in mind when he despaired of finding non-circular sufficient conditions for rationalizing psychological explanation (Davidson, 2001, p. 73). The main trouble with attempting to analyse the meaning of ‘because_P’ in terms of the meaning of ‘because_R’ is that it just isn’t obvious what such an analysis would look like, and nor is it obvious how it would help to explain Causal Entailment.

The key to understanding alternative analysis is to reflect on the following Euthyphro style question: when someone acts for a reason, and so is caused to act by some mental state of theirs with that reason as its content, is the reason their reason for acting because they were caused to act by that mental state, or were they caused to act by that mental state because the content of that mental state was their reason for acting? Notice that the ‘because’ in this question is the ‘because’ of grounding explanation, the kind of ‘because’ that can be replaced by ‘in virtue of’ talk.⁵³ Psychologism answers that when someone acts for a reason, and so is caused to act by some mental state of theirs with that reason as its content, the reason is their reason for acting partly because they were caused to act by that mental state.⁵⁴ This is why reason explanations entail both rationalizing psychological explanations and ordinary causal explanations according to psychologism. But suppose we consider the alternative answer: when someone acts for a reason, and so is caused to act by some mental state of theirs with that reason as its content, they were caused to act by that mental state because the content of that mental state was their reason for acting. This suggests an alternative account of rationalizing psychological explanations. For a rationalizing psychological explanation to obtain is just for a causal explanation to obtain in virtue of the fact that an appropriate reason explanation obtains. Formally, the analysis looks like this:

Reason Analysis: P(p,K(x,q)) $:=$ G(C(p,K(x,q)),R(x,p,q))

In this analysis, the analysans involves the concept of grounding G. It relates the causal explanation C(p,K(x,q)) and the reason explanation R(x,p,q). The analysans states that for a rationalizing psychological explanation to obtain, the former must hold in virtue of the latter holding.

There are several virtues of Reason Analysis. First, unlike the Causal Analysis, it is complete. It offers an analysis which provides both necessary and sufficient conditions for rationalizing psychological explanations. Second, it can explain why Causal Entailment is analytic, as predicted by the linguistic tests. Since G is factive, G(C(p,K(x,q)),R(x,p,q)) entails C(p,K(x,q)). Finally, Reason Analysis can explain why rationalizing psychological explanations analytically entail reason explanations. The explanation is the same, since G is factive, G(C(p,K(x,q)),R(x,p,q)) entails R(x,p,q). These are significant virtues of Reasons Analysis. Taken together, they offer good reasons for thinking that it is correct.⁵⁵

Reason Analysis is not entirely unprecedented. Dancy comes close to offering a view like this when he writes:

The most direct response to Davidson [is] that the difference between those reasons for which the agent did in fact act and those for which he might have acted but did not is not a difference in causal role at all. It is just the difference between the considerations in light of which he acted and other considerations he took to favour acting as he did but which were not in fact ones in the light of which he decided to do it. (Dancy, 2000, p. 163)

I take Dancy to be suggesting that we might appeal to the concept of reason explanation to answer Davidson’s challenge about what distinguishes cases where an agent’s psychological states cause an action in the right way and cases where they don’t. However, Dancy rejects the idea that there might be a “difference in causal role” here. But we have seen that this isn’t quite right. There is a difference in the causal role of one’s attitudes when one acts for a reason. What we should reject is the idea that there is an independently characterisable difference in causal role, one that can be characterised without appeal to ‘because_R’. According to Reason Analysis the difference in causal role is this: that difference between a causal relation holding because the relation of reason explanation holds, and its holding otherwise. No wonder it has appeared to be impossible to give additional conditions under which one’s psychological states cause one to act in the right way.

Before we can conclude that Reason Analysis is correct, and, therefore, accept analytic anti-psychologism there is one remaining challenge to address. It seems to be a significant disadvantage of Reason Analysis that it does not explain Psychological Entailment. Why do reason explanations entail rationalizing psychological explanations? Indeed, not only does Reasons Analysis not explain the entailment, it precludes what looks like the best explanation. As we saw earlier, we have good reasons for thinking that Psychological Entailment is not analytic in the way analytic psychologism predicts. But it is nonetheless a metaphysical entailment. Psychologism promises to explain this metaphysical entailment by understanding the relation of reason explanation in terms of the relation of rationalizing psychological explanation. If what it is for a reason explanation to hold is just for a rationalizing psychological explanation to hold, then we can explain Psychological Entailment as a metaphysical entailment. It is a mark against an explanation of a particular phenomenon if it fails to explain all the facts that are in need of explanation. Just as the causal-psychological theory of acting for a reason fails to explain why reason explanations entail causal explanations by solving the problem of deviant causation, analytic anti-psychologism fails to explain why reason explanations entail psychological explanations. Of course, it does not follow that analytic anti-psychologism cannot explain this. This may seem to suggest that we have a stand-off, however. But this is not so. Analytic anti-psychologism explains more of what needs explaining than the alternative does. As we have seen, only analytic anti-psychologism explains the relevant facts about entailments which are analytic. We thus have good reason, insofar as this is our best theory, for thinking that we will find some explanation of Psychological Entailment compatible with analytic anti-psychologism. This, however, is a project for another day.⁵⁶

6 Conclusion

This paper set out to explore some under-explored issues related to the language of action explanation. Such issues have often been set aside in philosophical discussions on the assumption that the philosophical discussions are interested in action and the explanation of action itself, and not the linguistic expression of such explanations. Issues of analyticity have largely been set aside for more direct talk about the things and relations themselves, and relations of grounding and identity which hold among these things. We saw how the two quick arguments promised a quick argument for a causal-psychological theory of acting for a reason. We saw that the assumptions behind these arguments are not as straightforward as they seem. This might have been taken to only establish the negative conclusion that we shouldn’t be arguing for theses in the philosophy of action on linguistic grounds. Hence, the proponent of psychologism might have thought that the discussion was, for the most part, irrelevant to their concerns. Perhaps ‘because_R’ and ‘because_P’ cannot be analysed in terms of each other, but that doesn’t stop reason explanations being grounded in rationalizing explanations and ultimately in causal explanations. As we continued to pursue the question, however, we saw that considerations about polysemy led us to take seriously unfashionable issues about analysis and analyticity. This led us to an alternative hypothesis about the meaning of ‘because_R’ and ‘because_P’, a hypothesis which, if true, would establish an alternative to psychologism on the basis of the meaning of the terms alone (and not metaphysical analysis). The irony is that the alternative gives us a kind of causal theory of rationalizing explanations without supporting a causal-psychological theory of reason explanations. On the alternative, rationalizing explanations are a species of causal explanation, namely, that species of causal explanation which obtains in virtue of reasons explanations obtaining. The irony is that it was the thesis that rationalizing explanations are a species of causal explanation which was meant to be a central premise in the master argument for the causal theory of acting for a reason (the other premise being the truth of psychologism).

Let me end with a reflection on the ambition of the causal-psychological theory, and compare it with the alternative I have offered. The causal-psychological theory has a philosophically worthy ambition: to explain one thing in terms of another. Unfortunately, however, it suffers from the vice of being over-ambitious. It attempts to explain both reason explanations and rationalizing psychological explanations in terms of causal explanations and other factors. For too long, like so many over-ambitious theories, it has failed to deliver on its promise. The alternative I have defended in this paper shares the philosophically worthy ambition with the causal-psychological theory. But it is less ambitious. It takes causal explanation, grounding explanation, and reason explanation for granted, and explains rationalizing psychological explanation in terms of them. In virtue of being less ambitious, this alternative is able to deliver on its promise. It delivers a definition of rationalizing psychological explanation. No small achievement. Should it really be so surprising at the end of the day that we are left with reason explanation and causal/grounding explanation as primitives? The former is a primitive of the first-person perspective. The second is a primitive of the third person perspective. Rationalizing psychological explanations exist at the nexus between the first-person and the third-person. Is it really so surprising that, being sandwiched in this way, they can be defined in terms of elements from each perspective? Should we have expected anything else?