Appeared in Nous 34 (2000), pp. 203-236.

Russellianism and Psychological Generalizations

David Braun

University of Rochester

I can perhaps best introduce Russellianism by describing its most notorious consequence. Russellianism (also known as `neo-Russellianism' and `the naive theory') entails that any two attitude ascriptions, that differ only in containing distinct co-referring names, express the same proposition and so have the same truth value. For example, since `Twain' and `Clemens' refer to the same person, Russellianism entails that (1) and (2) express the same proposition and have the same truth value.

(1) Harry believes that Twain is a writer.

(2) Harry believes that Clemens is a writer.

I say that this is Russellianism's most notorious consequence because it is so often used to argue against the view: many philosophers think that it is obvious that (1) and (2) can differ in truth value, and so they conclude that Russellianism is false. Let's call this the Substitution Objectionto Russellianism.

Russellians have responded at length to the Substitution Objection. The two best-known advocates of the view, Nathan Salmon and Scott Soames, have said that (1) and (2) really do have the same semantic content and truth value, and that our intuitions to the contrary are due to a failue to distinguish correctly between their semantic contents and their pragmatic implications. I have elsewhere presented an alternative response to the Substitution Objection, one that explains our anti-Russellian judgments without relying on claims about the pragmatics of attitude ascriptions (see Braun, 1998).

In this paper, however, I wish to consider an objection to Russellianism that has been relatively neglected by Russellians. This objection hinges on attitude ascriptions inpsychological generalizations, such as (3).

(3) If a person wants Twain to autograph her book, and she believes that (if she waves, then Twain will autograph her book), then (other things being equal) she will wave.

(3) seems to be true (or at least approximately true). Yet some philosophers, such as Mark Crimmins, Mark Richard, and Michael Devitt, claim that if Russellianism were true, then (3) would be false. Here is one (rough) argument for their claim. ⁽¹⁾ Suppose that an agent wantsTwain to autograph her book, and suppose that she believes that (if she waves then Clemens will autograph her book). Then according to Russellianism, she also believes that (if she waves thenTwain will autograph her book), even if she would vigorously dissent from `If I wave then Twain will autograph my book'. Such a person would not wave, and yet, according to Russellianism, she satisfies the antecedent of (3). So, if Russellianism were true, (3) would be false. So Russellianism is false. This is one version of what I shall call the Generalization Objection to Russellianism.

There are at least two reasons why the Generalization Objection seems worthy of Russellians' attention. First, it is intially plausible. Second, it can be used to support further initially plausible objections to Russellianism. For example, it might be claimed that if psychological generalizations like (3) were false, then attitude ascriptions could not be used toexplain behavior. But clearly attitude ascriptions can be so used. So a critic might (again) conclude that Russellianism is false. ⁽²⁾

For these reasons, I think that anyone who is (tenatively) attracted to Russellianism (as I am) should be interested in whether Russellians can provide a plausible reply to the Generalization Objection. I try to provide such a reply in this paper. To be precise, I present twoquite different Russellian replies. In my first reply, I make a few observations about the semantics of generalizations like (3), and then use those observations to argue that the generalizations would be true under Russellianism. ⁽³⁾ I believe that this first reply is successful, but I suspect (for various reasons) that it will not convince everyone. Therefore I also offer a second, quite different, reply. In it, I concede (for the sake of argument) that if Russellianism is true, then generalizations like (3) are false. But in this second reply, I argue (contrary to the objection) that the generalizations really are false, and offer a Russellian explanation for why they appear to be true to ordinary speakers. This second reply to the Generalization Objection resembles the standard Russellian reply to the Substitution Objection, in that it tries to explain away ordinary speakers' mistaken intuitions about truth values. The upshot of the two replies is that Russellians can plausibly argue that one or the other of the major premises of the Generalization Objection is false.

In addition to responding to the Generalization Objection, I respond to two other closely related objections, one of which is due to Mark Richard, the other of which is a variant on the Substitution Objection.

Before turning to those tasks, a few words about what I do not do below. First, some readers who are familiar with the views of Salmon and Soames might expect me to rely heavily on their distinction between the semantic contents and pragmatic implications of generalizations like (3). They might expect me to claim, for instance, that utterances of generalizations like (3) pragmatically convey true propositions that are not semantically expressed by the generalizations. ⁽⁴⁾ But I am skeptical of the pragmatic theories of Salmon and Soames, for reasons that I give elsewhere (Braun, 1998). So in this paper, I respond to the Generalization Objection in such a way that Russellians are not forced to rely on Salmon and Soames's claims about the pragmatics of attitude ascriptions.

Second, I shall not try to defend Russellianism against all of the objections that might be motivated by the Generalization Objection, or that might be (in some sense) "closely related" to it. For example, I shall not respond to the earlier objection concerning explanation and psychological generalizations; neither shall I respond to claims that ascriptions of Russellian content (or, more generally, "broad content") cannot explain behavior because Russellian content (or broad content) is causally irrelevant to behavior. These objections raise various complex issues about the nature of explanation, causation, and causally relevant properties that would take far more space to discuss than I have here. ⁽⁵⁾ In this paper, I restrict myself to discussing the Generalization Objection, and whether the psychological generalizations would be true under Russellianism. ⁽⁶⁾

1. Russellianism

According to Russellianism, propositions are the objects of attitudes such as belief, desire, and assertion. ⁽⁷⁾ They are also the semantic contents (or simply `contents') of sentences,with respect to (or relative to, or simply in), contexts; they are what sentences semantically express with respect to contexts. Propositions on this view are Russellian: complex structured entities that have individuals, properties, and relations as constituents. The proposition expressed by a sentence, in a context, has as its constituents the contents of the sentence's parts, in that context. The truth value of a sentence, in a context, is the truth value of the proposition it expresses, in that context. The content of a proper name or indexical, in a context, is its referent, in that context. The content of a predicate, in a context, is a property or relation. Thus the proposition expressed by `Twain is an author', in a context, has as its constituents the referent of `Twain' and the content of `is an author', in that context. We can represent the proposition it expresses, in a context, with the ordered pair consisting of Twain and the property of being-an-author, as follows. ⁽⁸⁾

<Twain, being an author>

According to Russellianism, the content of the predicate `believes', in any context, is a certain binary relation that can hold between an agent and a proposition. A standard belief sentence of the form A believes that S expresses a proposition, in a context, whose constituents are the referent of A, the believing relation, and the proposition expressed by S, in that context. More precisely: the `that'-clause that S refers, in a context, to the proposition expressed by S, in that context. The content of the `that'-clause is its referent. So the proposition expressed by (1)

(1) Harry believes that Twain is a writer

has as its constituents Harry, the believing relation, and the proposition that Twain is an author, and can be represented by

<Harry, <Twain, being an author>, believing>.

I shall assume that Russellianism also includes a parallel account of attitude sentences whose complements are infinitival clauses with explicit subjects, for example, `Harry wants Twain to smile'. An infinitival sentence such as `Twain to smile' expresses, in a context, the same proposition as either `Twain smiles' or `Twain will smile', in that same context. (I am ignoring tense here; see note 8.) If S is an infinitival clause with an explicit subject, then a sentence of the form A wants S expresses, in a context, a proposition whose constituents are the referent of A, the binary wanting relation, and the proposition expressed by S, in that context. ⁽⁹⁾

Russellianism has a number of virtues as a theory of attitude ascriptions. It is appealingly simple. It is naturally suggested by the arguments against description theories of names and indexicals presented by Saul Kripke, Keith Donnellan, David Kaplan, and others. ⁽¹⁰⁾ It deals well with our uses of belief sentences containing indexicals in their `that'-clauses. And it gives the most straightforward account possible of apparent quantification into `that'-clauses. Unfortunately, I do not have space to describe further the considerable virtues of Russellianism. For a short summary, see Braun (1998). For more detailed descriptions, see Salmon (1986, 1989) and Soames (1987, 1995).

2. Ways of Taking Propositions

Before turning to the Generalization Objection, it will be useful to have before us the standard Russellian view about the metaphysics of belief. ⁽¹¹⁾ Russellians think that believing and desiring are binary relations that hold between people and propositions. But most Russellians hold that this relation is mediated: one believes or desires a proposition in virtue of standing in some psychologically significant relation to a third thing that determines the proposition that one believes or desires. ⁽¹²⁾ This third thing is called a `way of taking', or a `way of grasping', or a `mode of presentation for', or a `guise for', a proposition. When the relevant attitude is believing or desiring, I shall often say that the relevant way of taking a proposition is a `way of believing' or a `way of desiring' that proposition. Different Russellians have different views about the nature of this intermediary entity. It may be said to be a linguistic meaning, a Kaplanian character, a natural language sentence, a mental state, or a mental representation. One believes a proposition by accepting (in a technical sense of `accept') either a meaning, a character, or a sentence; or by being in a mental state; or by having the mental representation in one's head in the right way. Furthermore, one can believe the same proposition in two distinct ways; and one can believe a proposition in one way without believing it in other ways. Finally, one can rationally believe a proposition, taking it in one way, while also believing the negation of that proposition, taking it in another, suitably different way. Analogous points hold for desiring. ⁽¹³⁾

To illustrate, let's consider (4) and (5).

(4) Twain is a writer.

(5) Clemens is a writer.

According to Russellianism, (4) and (5) express the same proposition. But a person can believe that proposition in at least two distinct ways. Believing the proposition in one way will incline one to assent to (4) and believe that it is true; believing it another way will incline one to assent to (5) and believe that it is true; and believing the proposition in the first way, but not the second, will incline one to think that (4) is true, but not incline one to believe that (5) is. ⁽¹⁴⁾ Moreover, one can rationally believe the proposition expressed by (4) in the first way, while believing the negation of that proposition in a way "corresponding" to the negation of (5). Such a person would be inclined to think that (4) is true and (5) is false.

So much for the standard Russellian metaphysics of belief. On encountering it, one might want to know more about ways of taking propositions, and how a person can believe a proposition in one way, while failing to believe it in another way. I shall describe how this might occur by presenting one theory of ways of taking propositions, which I call the mental-sentence theory. I think it is plausible, but other Russellians who accept ways of taking propositions may not.

On the mental-sentence theory, people believe propositions by having sentence-like representations in their heads. These representations express propositions; the state of having such a representation in one's head may have the right causal role for a belief, or a belief state. These mental sentences are ways of taking propositions. Let's say that a mental sentence that has the right causal role for a belief is in one's belief box. Similarly, one desires a proposition by having a sentence in one's head that expresses the proposition and that has the right causal role. Let's say that a mental sentence that has the right causal role for a desire is in one's desire box.

To illustrate, let's assume (for convenience) that the sentences that one has in one's belief and desire boxes are sentences of one's natural language. So if Harry speaks English, then he has English sentences in his belief and desire boxes. If he has both (4) and (5) in his belief box, then he believes the proposition that Twain is an author in two distinct ways. He will also think that (4) and (5) are true. But Harry could have (4), but not (5), in his belief box. Then he would believe the proposition that Twain/Clemens is a writer in one way, but not in another. He could also have (4) and the negation of (5) in his belief box. By having both (4) and the negation of (5) in his belief box, he would believe a proposition and its negation. But he could rationally do so, given the differences in the ways in which he believes them.

The same kind of phenomenon can occur with respect to attitude ascriptions. Paula could have both (1) and (2) in her belief box.

(1) Harry believes that Twain is a writer.

(2) Harry believes that Clemens is a writer.

If she does, then she believes, in two distinct ways, the proposition that Harry believes that Twain is a writer. But she could, instead, rationally have (1), but not (2), in her belief box. In fact, she could have (1) and the negation of (2) in her belief box. Then she would believe a proposition and its negation; she would also believe that (1) is true while believing that (2) is false. But she could do all of this rationally, because she believes these propositions in suitably different ways.

Thus (1) and (2) express the same proposition, but a rational person could understand both, and think the first is true and the second is false. This is the gist of my response to the Substitution Objection (for details, see Braun, 1998). Notice that this reply does not appeal to the pragmatics of attitude ascriptions, or rely on pragmatics to explain ordinary speakers' intuitions about their truth values.

I wish to emphasize that the replies to the Generalization Objection that follow below do not rely on the mental-sentence theory. What I do assume in my replies is the following: (a) there are distinct ways of believing the same proposition; (b) a person can believe a proposition in one way without believing it in other ways; and (c) a person can rationally believe a proposition in one way while believing its negation in another, suitably different, way. Similarly, with appropriate changes, for desiring.

3. The Generalization Objection

Let's now consider the Generalization Objection in a bit more detail than before. (6a)-(6c) are psychological generalizations.

(6a) If a person wants Twain to nod, and she believes that (if she waves then Twain will nod), then she will tend to wave.

(6b) If a person wants Twain to nod, and she believes that (if she waves then Twain will nod), then she will be disposed to wave.

(6c) If a person wants Twain to nod, and she believes that (if she waves then Twain will nod), then, other things being equal, she will wave.

Richard, Crimmins, and Devitt claim that ordinary psychological generalizations like (6a)-(6c) are true. ⁽¹⁵⁾ But they claim that if Russellianism were true, then the generalizations would be false. So they conclude that Russellianism is false. ⁽¹⁶⁾

The argument's crucial premise is the claim that the generalizations would be false if Russellianism were true. Here's one argument for that claim. Suppose that Sue sincerely assents to `I want Twain to nod', but sincerely dissents from `I want Clemens to nod'. Suppose that she sincerely assents to `If I wave then Clemens will nod', but sincerely dissents from `If I wave thenTwain will nod'. (Finally, suppose she dissents from `Twain is Clemens', and that she understands all of these sentences, and is reflective, attentive, and rational.) Now in this situation, Sue would not wave, and so she fails to satisfy the consequents of (6a)-(6c). But according to Russellianism, she satisfies their antecedents. For if she sincerely assents to `If I wave, then Clemens will nod', then she believes that (if she waves then Clemens will nod). Therefore, according to Russellianism, she also believes that (if she waves then Twain will nod). Furthermore, if she sincerely assents to `I want Twain to nod', then she wants Twains to nod. So Sue satisfies the antecedents of (6a)-(6c). And yet she does not wave, and so fails to satisfy the consequents of (6a)-(6c). Therefore, if Russellianism is true, then Sue falsifies (6a)-(6c), and so these generalizations are false, if Russellianism is true. ⁽¹⁷⁾

The criticism alleges simply that the psychological generalizations are false under Russellianism. It does not allege that Russellians cannot provide some explanation of why Sue does not wave. Indeed, Russellians can claim (and many non-Russellians would agree) that Sue does not wave because there is a kind of mismatch in the ways in which she believes and desires the relevant propositions. Let me explain this notion of mismatching ways by first explaining the notion of matching ways.

Suppose that Ethel assents to both `If I wave then Clemens will nod' and `I want Clemens to nod'. So she believes the proposition expressed by the former sentence, when she takes it in a certain way. (On the mental-sentence view, she has the sentence `If I wave then Clemens will nod' in her belief box.) When she takes that proposition in that way, there is also a certain way in which she takes the consequent of that proposition. (`Clemens will nod' appears in her belief box as the consequent of the sentence `If I wave then Clemens will nod'.) Ethel also desires the proposition expressed by `Clemens will nod', when she takes it in the way that she takes the consequent of `If I wave then Clemens will nod'. (She has `Clemens will nod' in her desire box.) Let's say that her ways of believing and desiring the two propositions match and that she believes and desires those propositions in matching ways. (On the mental-sentence view, the sentence in her desire box is identical with the consequent of the sentence in her belief box.) As a result, she waves.

Now according to Russellians, Sue believes and desires the very same propositions that Ethel does, but not in the same ways that Ethel does. There is no match between the ways in which Sue believes and desires those propositions. Sue believes and desires the propositions inmismatching ways. So Sue's belief and desire do not cause her to wave. So Russellians can explain (in some sense) why Sue does not wave. The problem for Russellianism, the critics claim, is that the generalizations turn out to be false under the view.

Non-Russellians avoid this problem by designing their theories of attitude ascriptions so as to allow Sue to fail to satisfy the antecedents of (6a)-(6c). So on these non-Russellian views, Sue is not even a prima facie counterexample to the generalizations. Some of these views (for example, those of Crimmins, Perry, and Richard) agree with Russellianism on the metaphysics of belief, and the existence of different ways of believing and desiring a single proposition. But they allow utterances of belief and desire ascriptions to express propositions that are partly about the ways in which the agent believes and desires the relevant propositions. On their views, an agent will not satisfy the antecedents of typical utterances of (6a)-(6c) unless she believes and desires the relevant propositions in matching ways. So an agent who satisfies the antecedents will wave.

Thus the Generalization Objection turns on the following facts. First, according to Russellianism, the propositions expressed by belief and desire ascriptions say only that the agent believes and desires certain Russellian propositions; the propositions do not contain information about the ways in which the agents believe and desire the propositions. Yet, second, a person's behavior often depends not only on the propositions she believes and desires, but also on the ways in which she believes and desires them.

I shall not dispute these latter claims. But I shall nevertheless argue in my first reply that the generalizations are true under Russellianism.

4. Exceptions and the Generalization Objection

Let's first note that the argument given above, for thinking that the generalizations would be false under Russellianism, is flawed. The objection points out (in effect) that, if Russellianism were true, then there would be certain exceptions to (6a)-(6c). It then concludes that if Russellianism were true, then the generalizations would be false. So the objection seems to assume that if there were any exceptions to the generalizations, then the generalizations would be false. But this last assumption is dubious, for generalizations of this sort seem to "tolerate" exceptions. Moreover, anyone who thinks that the generalizations are true should reject the assumption, for there are exceptions to the generalizations that hold even if Russellianism is false; that is, there are exceptions in which the agent believes and desires the relevant propositions in matching ways. Suppose that Wilma assents to both `If I wave then Twain will nod' and `I want Twain to nod'; so the ways in which she believes and desires the relevant propositions match. Nevertheless, she could fail to wave if, for instance, she believes that there is a more effective way to get Twain to nod than waving, or she has a strong countervailing desire not to wave, or she suddenly becomes paralyzed. So if the generalizations are true, they must be true despite certain sorts of exceptions; and the critics of Russellianism, who say the generalizations are true, must admit this. ⁽¹⁸⁾

Call exceptions that do not falsify the generalizations tolerable exceptions to the generalizations. To be a bit more precise: let tolerable exceptions be cases that satisfy the antecedent, and fail to satisfy the consequent, but nevertheless do not falsify the generalization. ⁽¹⁹⁾ To draw the conclusion that the generalizations would be false under Russellianism, the critic must argue that these mismatching exceptions are not tolerable exceptions to the generalizations, but rather are genuine counterexamples to the generalizations. Yet so far we have not seen such arguments. We have seen no reason to think that the exceptions involving mismatches falsify the generalizations, whereas the exceptions involving matches don't. (Call these mismatching exceptions and matching exceptions, respectively.)

Thus the objection as it stands is weak. But perhaps it can be modified to include some recognition of the difference between tolerable exceptions and genuine counterexamples. In the next section, I describe the differences between these two sorts of exceptions. Later I use the distinction to present an improved version of the Objection, and to provide a reply to that version of the Objection.

A bit of terminology first. Generalizations that tolerate exceptions are often called `ceteris paribus generalizations'. Some of these generalizations contain phrases like `other things being equal' and `tends to', as do (6a)-(6c) Others contain phrases like `usually', `generally', or `under suitable conditions'. Sometimes ordinary, unadorned, generalized conditionals like (6d) seem to function (semantically) like ceteris paribus generalizations.

(6d) If a person wants Twain to nod, and she believes that (if she waves then Twain will nod), then she will wave.

Instances of these generalizations, such as (7), I shall call `ceteris paribus conditionals'. ⁽²⁰⁾

(7) If Sue wants Twain to nod, and she believes that (if she waves then Twain will nod), then she will tend to wave.

5. Ceteris Paribus Generalizations and Suitable Conditions

To understand the difference between tolerable exceptions to generalizations and genuine couterexamples to them, it will be helpful to begin with ceteris paribus generalizations that do not contain attitude ascriptions. ⁽²¹⁾ Consider (8); variants are indicated in parentheses.

(8) If a car's ignition key is turned, then (generally/other things being equal/under suitable conditions) its engine will (tend to/be disposed to) start.

Utterances of (8) would usually strike us as being true. ⁽²²⁾ But obviously there are actual and possible exceptions to (8), in which, for example, the car's battery is dead, or its starter motor is broken, or its engine explodes before it starts (as in gangster movies). These exceptions seem to occur under conditions in which other things are not equal, or in conditions that are abnormal or atypical or unsuitable in some way. Let's say that suitable conditions hold when other things areequal and nothing relevant is abnormal or atypical. The truth conditions of an utterance of aceteris paribus generalization are determined, in part, by the suitable conditions that are, somehow, associated with the utterance of the generalization. A ceteris paribus conditional is true iff (roughly): if the antecedent were true and suitable conditions held, then the consequent would also be true. A ceteris paribus generalization is true iff (roughly): every individual is such that if it were the case that the individual satisfied the antecedent and suitable conditions held, then the individual would satisfy the consequent. The preceding examples are tolerable exceptions to (8) because suitable conditions do not prevail in them. A genuine counterexample to (8) would be an example in which (very roughly) the antecedent is true, and suitable conditions hold, but the consequent is false.

So to determine whether an exception to a generalization is a tolerable exception or a genuine counterexample, we need to determine whether the generalization's suitable conditions hold (that is, whether "other things are equal"). But when doing this, we need to keep in mind that the suitable conditions associated with a generalization vary from context to context. Let me explain.

Ceteris paribus generalizations and conditionals are context-sensitive: their truth conditions and contents vary from context to context, or from utterance to utterance. Consider (9).

(9) If a car with an automatic transmission is put in reverse, it will move backwards.

In most contexts, an utterance of (9) would strike us as true. But consider a context in which a driving instructor is emphasizing to her students that a car with an automatic transmission moves only if it is in gear and the driver's foot is off the brake pedal. In such a context, (9) might strike us as false; in fact, it seems that it is false in such a context. (10), however, would strike us as true in this same driving-school context.

(10) If a car with an automatic transmission is put in reverse, and the driver's foot is off the brake pedal, then it will move backwards.

So generalizations like (9), and (10), can be true in one context and false in another. They vary in their truth conditions and contents from context to context.

It's plausible to suppose that (9) changes in truth conditions and content, from context to context, because the suitable conditions associated with it change from context to context. In contexts of the first (ordinary) sort, the driver's foot must be off the brake pedal in order for conditions to count as suitable (or normal or typical). But in the driving-school context, an event in which a car is put into reverse and the driver's foot is on the brake pedal, counts as occurring in suitable conditions (as being normal or typical), according to the standards for suitability in this context; so the provision that the driver's foot be off the brake must be explicitly added to the antecedent to obtain a true ceteris paribus generalization. In view of this, let's say that a context (or utterance) determines the suitable conditions associated with that sentence in that context (or with that utterance). ⁽²³⁾

In the above example, the change in the suitable conditions associated with (9) in the different contexts seems to be due to differences in the thoughts, assumptions, and intentions of the speakers and hearers in the two contexts. In some contexts, speakers (and hearers) tend to think of certain sorts of cases that satisfy the antecedent as being "normal" or "typical" or as occurring in "suitable conditions". In other contexts, their judgments about this change. In the driving-school context, the possibility that the driver's foot is on the brake is salient to the instructor and students. In such a context, events in which the driver shifts into reverse and has his foot on the brake pedal seem to come into focus and seem "typical". The possibility that the driver has his foot on the brake needs to be ruled out explicitly to obtain a true generalization. But in the first context, this possible interference to normal operations is no more salient than any other (for example, the engine's being off or the tires' being flat), and so the condition that the driver's foot be off the brake seems to be "automatically included" in the context's suitable conditions. ⁽²⁴⁾

Although speakers' thoughts, assumptions, and intentions partially determine suitable conditions, speakers who utter a ceteris paribus generalization (or conditional) may be ignorant of important aspects of the suitable conditions determined by their utterances. For example, many speakers who would be inclined to utter (9), and to think that their utterances are true, do not know that cars have driveshafts and differentials. Nevertheless, one important aspect of the suitable conditions determined by their utterances of (9) is that the driveshaft of the relevant car be correctly connected to its differential. (If it isn't, then the car will not move.) That is, the suitable conditions determined in typical contexts hold only if the driveshaft is connected to the differential. Thus unknown aspects of the world help determine the suitable conditions associated with a generalization in a context. ⁽²⁵⁾

6. A Sketch of a Semantics for Ceteris Paribus Conditionals

I have now presented the observations about ceteris paribus generalizations on which I shall rely in formulating an improved version of the Generalization Objection, and in providing my first reply. But these observations might leave some readers with some concerns. Some readers might wonder whether the observations I have made could be integrated into a systematic, precise semantics for ceteris paribus generalizations; this worry might lead them to question the earlier observations; some might even go so far as to wonder whether such generalizations are ever true. Other readers might worry that they do not yet understand enough about the semantics of ceteris paribus generalizations to evaluate either the (forthcoming) revised Objection or my reply to it. Readers of both sorts might be reassured if I could give them some reason to think that it is possible to formulate a formal semantics for ceteris paribusgeneralizations that is consistent with my claims from the preceding section. (Some of these readers might also find such a semantics interesting in and of itself.)

For this reason, I shall sketch a semantics for such generalizations below. I want to emphasize that this semantics is tentative, and, in any case, my replies to the Generalization Objection do not hinge on the details of it. To simplify matters, I concentrate below on ceteris paribus conditionals, and ignore the quantification involved in generalizations. ⁽²⁶⁾

I earlier used the counterfactual or subjunctive conditional to state rough truth conditions for utterances of ceteris paribus conditionals and generalizations. I did this intentionally, forceteris paribus conditionals are similar to counterfactual conditionals in important respects. On the well-known Stalnaker-Lewis view, the truth conditions of counterfactuals are determined by relations of similarity between worlds. (See Lewis 1986a and 1973, and Stalnaker 1968.) According to Lewis's (1986a) version, the truth conditions of a counterfactual can be given as follows:

If A were the case, then C would be the case is true at world w iff: either (i) A is false at all worlds, or (ii) some world at which both A and C are true is more similar to w than any world at which A is true and C is false.

A counterfactual can be true even though there are some worlds in which A is true and C is false, for some such worlds are too gratuitously dissimilar from the actual world to falsify the counterfactual. (They are less similar to the actual world than other worlds in which A and C are both true.) Analogously, a ceteris paribus conditional of the form If A then ceteris paribus Ccan be true though there are worlds in which A is true and C is false, because these worlds are either too dissimilar from the actual world, or too "gratuitously unsuitable", to falsify the conditional.

The most important differences between the two sorts of conditional arise in cases in which the antecedent is true and the consequent false. A counterfactual conditional is false at a world if its antecedent is true and its consequent false at that world. ⁽²⁷⁾ But a ceteris paribusconditional may be true at a world at which its antecedent is true and its consequent is false, with respect to a context that determines some associated suitable conditions, as long as the suitable conditions associated with the conditional, in that context, fail to hold at that world. So we can intuitively think of a ceteris paribus conditional of the form If A then ceteris paribus C as being like a counterfactual conditional that includes in its antecedent an "implicit statement" S of suitable conditions: If (A and S) were the case, then C would be the case. (But we should not rely too heavily on this intuitive guide, because the suitable conditions might not be finitely expressible in English.) We can modify the Lewis-style truth conditions of counterfactual conditionals to obtain the truth conditions of ceteris paribus conditionals as follows. ⁽²⁸⁾

If A then ceteris paribus C is true at world w, with respect to context c, iff: either

(i) there is no world v such that: (a) A is true at v, with respect to c, and (b) the suitable conditions S determined by c hold at v; or

(ii) there is a world v such that A is true at v, with respect to c, and the suitable conditions S determined by c hold at v, and C is true at v, with respect to c; and vis more similar to w than any world at which A is true, with respect to c, and Sholds and C is false, with respect to c.

The above view concerning the truth conditions of ceteris paribus conditions is consistent with various plausible Russellian views about their propositional contents, in contexts. Here is one such view. The ceteris paribus connective expresses, in a context, a conditional relation between propositions, different ones in different contexts. Let COND_c be the conditional relation expressed by the connective in context c. Thus (7) expresses, in context c, a proposition whose constituents are the proposition expressed by the antecedent in c (call it `P'), the proposition expressed by the consequent in c (call it `Q'), and COND_c. We can represent this proposition with <P, Q, COND_c>. The suitable conditions determined by c for (7) are "built into" the conditional relation COND_c, in the following sense. COND_c determines a property of worlds, being-suitable_c, and so also determines a set of suitable_c worlds. COND_c also determines a ternary relation that ranks worlds for their similarity to any given world w. ⁽²⁹⁾ A proposition of the above form is true at w iff: either there are no P worlds among the suitable_c worlds, or there is a P-and-Q world v in the set of suitable_c worlds such that v is more similar to w than any P-and-not-Q world within the set of suitable_c worlds. ⁽³⁰⁾

7. Modifying the Generalization Objection

With these observations concerning ceteris paribus generalizations in mind, let's return to the Generalization Objection. Recall that the earlier version of the Objection failed to argue that the mismatching exceptions are genuine counterexamples to the generalizations, rather than tolerable exceptions. The modified version of the Objection that we are seeking should provide an argument for this claim. But now that we have seen that ceteris paribus generalizations are context-sensitive, we can see that the earlier version of the Objection has a further problem. The earlier version says that generalizations (6a)-(6d) are true. But the generalizations are context-sensitive, so they cannot be true simpliciter; they can only be true with respect to contexts. So a modified Objection should claim that the generalizations are true in some relevant set of contexts, and then argue that the generalizations would be false in all (or some) of those contexts if Russellianism were true.

But what are the relevant contexts? In which contexts should the modified Objection claim that the generalizations are true? It is not immediately clear, but perhaps the most obvious claim would be this: generalizations (6a)-(6d) are true in all contexts. If the revised Objection made this claim, then it could argue that, if Russellianism were true, then (6a)-(6d) would be false in all, or at least some, contexts, and so conclude that Russellianism is false.

But this proposed revision is unsuccessful, for the claim that (6a)-(6d) are true in all contexts is incorrect: there are many contexts in which (6a)-(6d) are false, whether or not Russellianism is correct. For instance, there are contexts in which the speakers are discussing an agent who has the attitudes described by the generalizations and who tries to wave, but fails to do so because she is suddenly paralyzed. In such contexts, the generalizations will seem false to the speakers; this is a good indication that the generalizations are false in such contexts. To be true in such contexts, the consequents of the generalizations need to be modified to read `she tries to wave', or `she intends to wave', or something similar. In other contexts, speakers might discuss an agent who has the attitudes described by the generalization, but who also has an "overriding" desire not to wave. The generalizations will strike speakers in such contexts as false; the generalizations' antecedents need to be supplemented to rule out such "overriding desires" in order to be true in such contexts. In yet other contexts, the speakers may be concerned with an agent's weighing of options, decision procedures, planning, and other such matters. Again, (6a)-(6d) might seem false to the speakers in such contexts; to become true, their antecedents need to be supplemented with provisions concerning decision procedures and such matters.

Note that (6a)-(6d) might also be false in contexts in which sophisticated philosophers are discussing the alleged defects of Russellianism, or different ways of taking the same proposition. The antecedents of the generalizations might need to be modified, to mention matching ways of taking propositions, in order to be true in such extraordinary contexts. Indeed, I believe that Russellians should concede that the generalizations are false in such contexts. But Russellians need not worry about making this concession, for several reasons. First, the generalizations are already false in some contexts, whether or not Russellianism is correct. Second, contexts in which philosophers discuss Russellianism are truly extraordinary, and so there is no particular reason to think that the generalizations should be true in such contexts. Third, the concession is consistent with the claim that the generalizations are true in other, more ordinary contexts, under Russellianism. Finally, the falsity of the generalizations in such extraordinary contexts could help "explain away" the initial appeal of the Generalization Objection. For suppose that the generalizations are false in these sophisticated "philosophical" contexts, and suppose also that the generalizations appear to be false to speakers in such contexts. Then if the speakers in those contexts did not realize that the generalizations are context-sensitive, they might be misled into thinking that the generalizations themselves are simply false under Russellianism.

In any case, this first attempt at modifying the Generalization Objection does not succeed. But the critic of Russellianism need not give up yet. There clearly are contexts in which ordinary speakers use or discuss generalizations like (6a)-(6d), and in which paralyzed agents, overriding desires, decision procedures, alternative ways of taking propositions, etc., are not under discussion. In many such contexts, ordinary well-informed speakers think that the generalizations are true. Call these the seems-true contexts for the generalizations. Now the critics of Russellianism can argue as follows. Consider a representative seems-true context, and suppose that Russellianism were true. Then an agent could satisfy the antecedents of generalizations like (6a)-(6d), and the suitable conditions determined by this seems-true context could hold, and yet the agent could believe and desire the relevant propositions in mismatching ways. But then the agent would fail to wave. Such a case would be an exception in which "other things are equal" by the standards of the context. Therefore such a case would be a counterexample to the generalizations, with respect to the seems-true context. So if Russellianism were correct, generalizations like (6a)-(6d) would be false in seems-true contexts. But the generalizations are true in seems-true contexts. Therefore, Russellianism is false. ⁽³¹⁾

We now have a substantive objection to which Russellians need a substantive reply. Russellians need to refute the argument for the claim that the generalizations are false in seems-true contexts; or, if they admit that they are false in seems-true contexts under Russellianism, then they need to explain why the generalizations seem to be true in such contexts.

8. First Reply to the Generalization Objection

The final version of the Generalization Objection claims that, if Russellianism is true, then it's possible for an agent to satisfy the antecedents of (6a)-(6d), while the suitable conditions determined by a seems-true context hold, even though the agent believes and desires the relevant propositions in mismatched ways. This is the premise of the objection that I wish to challenge in my first reply. I maintain that (even if Russellianism is true) the suitable conditions determined for (6a)-(6d), by seems-true contexts, include the requirement, or condition, that the agent believe and desire the propositions in matching ways. Call this the matching-ways condition. If this is correct, then if an agent believes and desires the relevant propositions in mismatching ways, then the suitable conditions for (6a)-(6d), in seems-true contexts, do not hold. So the mismatching exceptions are cases in which "other things are not equal", with respect to seems-true contexts. Thus the mismatching cases are not counterexamples, but rather tolerable exceptions, to the generalizations, with respect to seems-true contexts, even if Russellianism is correct. Therefore, if the suitable conditions for the generalizations, in seems-true contexts, include the matching-ways condition, then the last version of the Generalization Objection fails. ⁽³²⁾

Let's now consider the evidence for the hypothesis that the suitable conditions include the matching-ways condition, in seems-true contexts. Call this the matching-ways hypothesis. We saw earlier that the suitable conditions associated with ceteris paribus generalizations, in a context, are partially determined by the thoughts, assumptions, and intentions of the speakers in those contexts. So to find evidence for (or against) the matching-ways hypothesis, we can look to the thoughts, assumptions, and intentions of speakers in those contexts. In particular, we can see what sorts of cases they think satisfy the antecedents of the generalizations, and see which of those cases they consider to be "normal" or "typical". We can inspect those latter cases to see whether the agents in them believe and desire the propositions in matching ways. We can also see what speakers in these contexts think of the mismatch cases, and whether they consider them to be counterexamples, or tolerable exceptions, or neither.

Let's begin with ordinary speakers' judgments about which cases are typical. When ordinary speakers consider the generalizations, in seems-true contexts, and think about cases in which the antecedents are satisfied, the cases that first come to their minds are ones in which the agent does, as a matter of fact, believe and desire the relevant propositions in matching ways. So they tend to think of these cases (and only these cases) as "typical" or "normal". But their judgments about typicality in these contexts partly determine the suitable conditions associated with the generalizations in those contexts. So it seems likely that the suitable conditions for the generalizations in seems-true contexts include the matching-ways condition.

Next, consider speakers' judgments about mismatching exceptions to the generalizations (in seems-true contexts). Ordinary speakers treat some mismatching exceptions that involvedemonstrative beliefs as if they were tolerable exceptions to the generalizations. Suppose, for example, that Harry sincerely asserts `If I wave then he will nod', while pointing at Twain; suppose he also sincerely asserts `I want Twain to nod'. Then in seems-true contexts, speakers would tend to take the sentence `Harry believes that if he waves then Twain will nod' to be true, and also consider `Harry wants Twain to nod' to be true. And so they would take Harry to satisfy the antecedents of generalizations (6a)-(6d). But suppose Harry does not assent to `He is Twain' (while pointing at Twain); so Harry's ways of believing and desiring the relevant propositions do not match, and he fails to wave, and so he fails to satisfy the consequents of the generalizations. Therefore Harry is a mismatching exception, according to Russellians. But ordinary speakers, in seems-true contexts, would not take Harry to be a counterexample to the generalizations. In seems-true contexts, they would tend to take his case to be one in which the antecedent is "strictly speaking" satisfied, while the consequent is not; but they would not take it to falsify the generalization. They would consider it to be a "funny" case, one that does not really "count against" the generalization. They would, in effect, judge Harry's case to be a tolerable exception to the generalization (though they would not put it that way). This is further evidence that the matching-ways condition is included in the suitable conditions of the generalizations in seems-true contexts.

Speakers consider other mismatch cases not to be exceptions at all. These judgments are at least consistent with the matching-ways hypothesis. Consider Sue again. She sincerely assents to `I want Twain to nod', but dissents from `I want Clemens to nod'. She sincerely assents to `If I wave then Clemens will nod', but sincerely dissents from `If I wave then Twain will nod'. According to Russellians, she satisfies the antecedents of generalizations (6a)-(6d), but in mismatching ways, and so she does not wave. Thus Russellians say that Sue is a mismatching exception to the generalizations. Now if ordinary speakers, in seems-true contexts, judged Sue to be a counterexample to the generalizations, then their judgments would be strong evidence that the suitable conditions of these contexts do not include the matching-ways condition. But, as a matter of fact, they would not judge Sue's case to be a counterexample. Ordinary, well-informed, non-Russellian speakers in seems-true contexts would think that the belief ascription `Sue believes that if she waves then Twain will nod' is false. So they would deny that Sue satisfies the antecedents of the generalizations. Therefore they would deny that she is a counterexample to the generalization. In fact, they would judge that Sue is not an exception to the generalizations. Therefore the judgments of ordinary speakers about mismatch cases like Sue's, in seems-true contexts, are consistent with the matching-ways hypothesis. (Moreover, if all data that is consistent with a hypothesis is evidence for it, then their judgments provide some evidence in favor of that hypothesis.) ⁽³³⁾

Taken all together, the above pattern of speakers' judgments (in seems-true contexts) supports the matching-ways hypothesis. For speakers take cases to be typical (or normal) when, as a matter of fact, the agents in them satisfy the antecedents, and believe and desire the propositions in matching ways. Moreover, they do not take mismatch cases to be normal or typical. Some of the mismatch cases they (in effect) take to be tolerable exceptions (for example, Harry's case), and so not typical. Others they judge not to satisfy the antecedent (for example, Sue's case), and so not typical cases of the sort covered by the generalization. ⁽³⁴⁾ Thus speakers in seems-true contexts in effect take matching cases to be typical, and fail to take mismatching cases to be typical. So it is likely that their judgments determine that the suitable conditions for the generalizations, in seems-true contexts, include the matching-ways condition.

Thus there is evidence that the matching-ways hypothesis is correct. But if it is correct, then the mismatching exceptions are tolerable exceptions to the generalizations rather than counterexamples, in these contexts, even if Russellianism is correct. And so these mismatching exceptions do not show that the generalizations would be false, in such contexts, if Russellianism were correct. Therefore the final version of the Generalization Objection is unsound.

We can draw an even stronger conclusion. Suppose that the mismatching exceptions are not counterexamples to the generalizations, in seems-true contexts, if Russellianism is correct. Then we can safely conclude that there are no counterexamples to the generalizations, in those contexts, if Russellianism is correct. (The critics of Russellianism would agree. They think the generalizations would be true, in seems-true contexts, if Russellianism were false. So they think that there would be no counterexamples to the generalizations, in the seems-true contexts, if there were no "Russellian" counterexamples to them.) But if there are no counterexamples to the generalizations, even if Russellianism is correct, then we can conclude that they are true, in seems-true contexts, if Russellianism is correct. ⁽³⁵⁾

Let's consider two replies before moving on. A critic might point out that most ordinary speakers lack the concept of a way of taking a proposition. So ordinary speakers do not think about ways of believing and desiring propositions, or about their matching or mismatching. They do not, for example, think to themselves "Those agents are believing and desiring those propositions in matching ways." But if this is so, then (the critic might say) the thoughts and intentions of speakers in seems-true contexts cannot determine that the suitable conditions associated with the generalizations in those contexts include the matching-ways condition.

This objection overlooks the fact that speakers' thoughts in a context can determine suitable conditions that include conditions of which the speakers are ignorant. Recall that the connection of a car's driveshaft with its differential is (usually) an unknown part of the suitable conditions for true utterances of (9). This can occur because the cases that speakers consider to be typical are, whether they know it or not, cases in which the driveshaft is connected to the differential. Similarly, if ordinary speakers tend to think of cases, in which the agent believes and desires the propositions in matching ways, as typical or normal, then the matching of ways of taking propositions could be an unknown aspect of the suitable conditions for the generalizations in seems-true contexts.

Next, a critic might contend that my reply to the Generalization Objection appears to work only because I have been unfairly focussing on generalizations of just one sort, namely generalizations relating attitudes to non-linguistic physical behavior. But consider (11)-(13).

(11) If a person intends to give Twain some money, and she believes that (if she buys a book by Twain, then she will give Twain some money), then she will intend to buy a book by Twain.

(12) If a person believes that Twain is an author, and someone asks her whether Twain is an author, then she will assert (or affirm) that Twain is an author.

(13) If a person believes that Twain is an author, and someone asks her "Is Twain an author?", then she will say "Yes".

A critic might argue, first, that (11)-(13) are true, in suitable contexts, and second, that (11)-(13) would be false in those contexts under Russellianism. (The critic's reasons for the last claim could be similar to those given in section 7.) So a critic might contend that, even if my response to the Generalization Objection works with respect to generalizations (6a)-(6d), Russellianism is still shown to be false by the truth (in suitable contexts) of generalizations like (11)-(13).

But similar replies can be given to anti-Russellian objections that rely on generalizations like (11)-(13). Consider (11). A person can intend a proposition in various ways. The "Russellian" exceptions to (11) occur when the agent's ways of believing and intending the propositions mentioned in the antecedent do not match in a suitable manner. But it's plausible to think that the suitable conditions for (11), in contexts in which it seems true, include the condition that the agent's ways of believing and intending the propositions mentioned in the antecedent match in a suitable manner. So the "Russellian" exceptions to (11) are tolerable exceptions rather than genuine counterexamples, in the contexts in which (11) seems true. Similarly, there are distinct ways to ask a proposition (that is, ask whether it is true); in contexts in which (12) seems true, the suitable conditions associated with it include some suitable matching condition between the way in which the agent believes the relevant proposition and the way in which she is asked that proposition. Finally, regarding (13): the sentence "Is Twain an author?" is a way of asking the proposition that Twain is an author. Among the suitable conditions in seems-true contexts for (13) is the condition that the way in which someone asks the agent the proposition, and the way in which she believes it, must match in some suitable manner. ⁽³⁶⁾

This concludes my first reply to the Generalization Objection. As I mentioned at the beginning of this paper, I have a second, alternative, reply to the objection. But before turning to it, I would like to consider a closely related objection from Mark Richard, and another objection that might be provoked by my first reply.

9. Richard's Singular Objection

Richard (1990) presents an objection to Russellianism that is similar to the Generalization Objection, except that it relies on sentences concerning a single agent, rather than generalizations. Richard asks us to consider sentences similar to (14)-(16).

(14) Sue believes that if she waves, then Twain will nod.

(15) Sue wants Twain to nod.

(16) Sue waves.

He presents his objection to Russellianism as follows. (I have substituted my example for his; I indicate my substitutions in square brackets.)

....what a sentence like [the conjunction of (14) and (15)] says can explain [Sue's waving]...On a Russellian view, this idea is somewhat puzzling. On Russellian terms the idea that what makes a claim like [(14) and (15)] true tends also to make [(16)] true is a mistake. What makes [(14) and (15)] true for a Russellian is just that [Sue] had some way of desiring-true the proposition that [Twain will nod] and some way of believing-true the proposition that [if she waves, then Twain will nod]. Since there needn't be much of a relation between these ways of getting to the propositions, what makes [(14) and (15)] true isn't something that makes, or tends to make, it true that [Sue waves]. (Richard 1990, p. 174)

Richard thinks that what makes (14) and (15) true tends to make (16) true. And he thinks that this would not be so if Russellianism were true, apparently because on that view (14) and (15) could be true even if Sue believed and desired the relevant propositions in mismatching ways. So he concludes that Russellianism is false. Let's call this Richard's Singular Objection. ⁽³⁷⁾

In reply, I first want to note that there is a perfectly reasonable sense of `tends to make true' according to which what makes (14) and (15) true does tend to make (16) true, even if Russellianism is correct. To understand this sense of the phrase `tends to make true', consider the following two sets of worlds, A and B. In each world in set A, Sue wants in some way or otherthe Russellian proposition that Twain will nod, and believes in some way or other the Russellian proposition that if she waves then Twain will nod. Thus (14) and (15) are true in all worlds in setA, if Russellianism is true. By contrast, in each world in set B, Sue completely fails to believe and desire those propositions; there are no ways in which she believes and desires those propositions. Thus (14) and (15) are false in all worlds in set B, if Russellianism is true. Suppose furthermore that, in all other respects, the worlds in sets A and B are very much like the actual world. Now there are worlds in both A and B in which Sue fails to wave. But there is a perfectly intuitive (though admittedly imprecise) sense in which she is "more likely" to wave in the worlds of set A than in the worlds of set B. For in every world in set A, Sue at least has beliefs and desires with the right sorts of propositional contents to cause her to wave (even though in some of those worlds her ways of believing and desiring the propositions are mismatched). But in many (most?) of the worlds of set B, Sue has no beliefs or desires concerning Twain or waving; in many of them, she does not even have beliefs and desires with the right sorts of propositional content to cause her to wave. So there is an intuitive sense in which Sue is "more likely" to wave in a world that is "randomly chosen" from set A than in a world that is "randomly chosen" from set B. So there does seem to be a reasonable sense in which what makes (14) and (15) true tends to make (16) true, even assuming that Russellianism is true.

I do not take this to be a conclusive reply to Richard's Singular Objection, because Richard may well have intended to use the phrase `tends to make true' in some sense other than the one on which I relied above. In fact, it seems likely that Richard intended this phrase to be construed in more or less the same sense that it has in psychological generalizations like (6a). For when Richard says that what makes (14) and (15) true tends to make (16) true, he seems to be implying that the ceteris paribus conditional (8) is true.

(8) If Sue wants Twain to nod, and she believes that (if she waves, then Twain will nod), then she will tend to wave.

Let's consider whether (8) is true under Russellianism.

(8) is an instance of (6a). I argued in the last section that (6a) is true in seems-true contexts, under Russellianism, despite exceptions in which an agent has mismatching ways of believing and desiring. A similar point holds for (8): the matching-ways condition is part of the suitable conditions determined for (8) by seems-true contexts. So the (possible) cases in which Sue has mismatching ways of believing and desiring are tolerable exceptions to (8) (in most seems-true contexts), rather than counterexamples. Thus (8) is true, in seems-true contexts, even if Russellianism is correct. So I conclude that, even if Russellianism is true, what makes (14) and (15) true, in seems-true contexts, does tend to make (16) true, in those contexts, in the sense of `tends to make true' that Richard (apparently) intends.

10. A New Substitution Objection

My first reply says that (6a)-(6d) are true (in seems-true contexts), even if Russellianism is true. A critic might, for the sake of argument, concede this, but then present the following new objection: If Russellianism is true, then (6a) and (6e) express the same proposition.

(6a) If a person wants Twain to nod, and she believes that (if she waves then Twainwill nod), then she will tend to wave.

(6e) If a person wants Twain to nod, and she believes that (if she waves then Clemenswill nod), then she will tend to wave.

But (a critic might continue), (6a) is true while (6e) is false. Therefore Russellianism is false.

This objection is a variant on the more popular Substitution Objection that I outlined at the beginning of this paper. The most obvious difference is that the new objection involves substitution of coreferring names in generalizations containing attitude ascriptions, rather than substitution in simple attitude ascriptions like (1) and (2). I shall therefore call it the New Substitution Objection.

The New Substitution Objection involves complications that are not present in the original. Simple attitude sentences are not context-sensitive. ⁽³⁸⁾ But (6a) and (6e) are; they vary in their contents and truth values from context to context. Thus it is, strictly speaking, incorrect to claim that (6a) and (6e) are true or false; instead, they are true or false with respect to contexts. The critic of Russellianism should modify his argument to take this into account. Perhaps the most plausible modified objection would go as follows. If Russellianism is true, then, in every context, (6a) and (6e) express the same proposition. If in every context they express the same proposition, then in every context they have the same truth value. But there are contexts in which (6a) is true and (6e) is false. Therefore, Russellianism is false.

I am confident that Russellians can give a plausible response to this objection. But the exact response Russellians should give depends on the resolution of some rather delicate issues concerning context-sensitivity, issues that I glossed over when I discussed ceteris paribusgeneralizations (sections 5 and 6). In particular, the correct response depends on the resolution of issues concerning the semantic evaluation of sentences in contexts in which they are not uttered.

The difficult issues arise because the content of (6a) in a context of utterance depends on the thoughts and intentions of the relevant speaker of the context. This intention-dependence poses no problems if the speaker intentionally utters (6a) in the context. But consider a context in which (6a) is not uttered. Suppose, further, that the speaker is not uttering, or considering, any other ceteris paribus conditional or generalization. It seems that a semantic theory should not assign a propositional content to (6a) in such a context, for the speaker does not have any thoughts or intentions relevant to determining the content of the ceteris paribus phrase (that is, the ceteris paribus connective) in (6a) in that context. So it seems that (6a) expresses no proposition with respect to such a context.

Now consider a context in which the speaker utters (6a), but does not utter (6e). Then the speaker does have the right sorts of thoughts and intentions to assure that (6a), and the occurrence of the ceteris paribus conditional connective within it, have propositional contents with respect to this context. But what about (6e)? Suppose that the speaker never considers (6e). Does (6e) nevertheless have a propositional content with respect to this context? In particular, does the occurrence of the ceteris paribus conditional connective in (6e) have the same content, in that context, as does the occurrence of the connective in (6a)? If so, then (6e) does have a propositional content with respect to the context---in fact, the very same content as (6a), if Russellianism is correct. But should we assume that, in every context, every occurrence of aceteris paribus connective, in every sentence, has the same content? We might hesitate to say the same about other context-sensitive expressions. For example, suppose that we say that all occurrences of `that' have the same content with respect to any given context. Then our theory would entail that `That is not that' expresses a false proposition in every context (unless, perhaps, we allowed something like shifts in context in mid-sentence).

Many of the problems that (6a) and (6e) raise are similar to those raised by multiple occurrences of the same demonstrative within a single context. For discussion of these issues, see Kaplan (1989), Braun (1996), and Garcia-Carpintero (1998). The issues are difficult and complex, and so I do not wish to assume any one solution to them in my reply to the New Substititon Objection. Therefore I shall present two different Russellian responses to it; they differ in the assumptions they make about the above issues.

The first response accepts the first premise of the New Substitution Objection: for every context, (6a) and (6e) express the same proposition in that context. (This response assumes that for every context, all occurrences of the ceteris paribus connective, even those in different generalizations, have the same content in that context.) ⁽³⁹⁾ But this response denies the second premise of the objection, the premise that (6a) and (6e) have different truth values in some contexts. According to this response, they have the same truth value in every context. The mistaken intuition that (6a) and (6e) could differ in truth value in some contexts is due to the mistaken intuition that, in some contexts, an agent who does not wave can satisfy the antecedent of (6e), and yet fail to satisfy the antecedent of (6a). For example, Paula may think that Sue does not wave, and mistakenly think that (17) is true and (18) is false, and so think that Sue satisfies the antecedent of (6e) and fails to satisfy the antecedent of (6a).

(17) Sue believes that if she waves then Clemens will nod.

(18) Sue believes that if she waves then Twain will nod.

But Russellians have already offered explanations of these mistaken intuitions. Salmon and Soames say that Paula can think that (17) is true and (18) is false, if she confuses the semantic contents of attitude ascriptions with their pragmatic implications. I say that Paula can believe that (17) is true and (18) is false, if she believes the proposition that they express, and the negation of that proposition, while taking those propositions in suitably different ways (see section 2 and Braun, 1998). ⁽⁴⁰⁾

The second response makes different assumptions about context-sensitivity. According to it, the first premise of the objection is false: (6a) and (6e) may express different propositions in the same context, even if Russellianism is true. One might think this because one thinks (roughly) that the suitable conditions determined by a context for a ceteris paribus generalization depend partly on the speaker's intentions when she uses the generalization. A speaker's intentions (in a single context) might vary depending on whether the speaker is using the name `Twain' or `Clemens' in a generalization, so in some contexts the occurrences of the conditional connective in (6a) and (6e) can differ in their associated suitable conditions; perhaps in some contexts the suitable conditions associated with (6a) include the matching-ways condition, whereas those associated with (6e) do not. ⁽⁴¹⁾ Thus the occurrences of the connective would differ in their propositional contents in that same context. So (6a) and (6e) can differ in their propositional contents within the same context, and the first premise of the objection is false. But according to this reply, the contents of the constituent simple attitude ascriptions in (6a) and (6e) are still the same, with respect to any single context. So the hypothesized difference in content between (6a) and (6e), in certain contexts, is consistent with the Russellian view of attitude ascriptions.

11. A Second Reply to the Generalization Objection

I argued in my first reply that the psychological generalizations are true in seems-true contexts, even if Russellianism is true. I am inclined to think that my arguments for this conclusion are sound. But I suspect that some (perfectly reasonable) readers may still harbor some doubts. Some may doubt my claims about what differentiates genuine counterexamples from tolerable exceptions. Others may accept these claims (more or less), but doubt that the matching-ways condition is part of the suitable conditions determined for the generalizations in seems-true contexts, and so doubt that the mismatching exceptions would be tolerable exceptions to the generalizations in seems-true contexts, if Russellianism were true. Because some readers may have such doubts about my first reply, I wish to present a second reply to the Generalization Objection. This second reply concedes that if Russellianism is true, then the generalizations arefalse in seems-true contexts. But it disputes another part of the objection, namely the premise that the generalizations are true, in those contexts. According to this second reply, the generalizations are actually false. I believe that this second reply is more plausible than it may initially appear to be. Let's consider how a Russellian could argue for it.

First, a Russellian could argue that the critics of Russellianism are not justified in assuming that the psychological generalizations are strictly true, in seems-true contexts; all that they are justified in assuming is that the generalizations are approximately true, in those contexts. (In fact, some critics of Russellianism cautiously claim only that the generalizations are approximately true.) But under Russellianism, the generalizations are approximately true in seems-true contexts. For the generalizations need only a slight addition to their antecedents, to the effect that the agent must believe and desire the propositions in matching ways, in order to become strictly true, in those contexts.

Second, a Russellian can explain why the generalizations appear to us to be (strictly) true in the seems-true contexts, even though they are (strictly) false. Here is how he might do it.

Suppose that Russellianism is true; so an agent could satisfy the antecedents of (6a)-(6d), even if he believes and desires the relevant propositions in mismatching ways. In most of those cases, the agent would not wave. Let's suppose that these cases are genuine counterexamples to the generalizations, in the seems-true contexts. Nevertheless, ordinary speakers (in seems-true contexts) would not think that these exceptions are counterexamples to the generalizations, for they would think that the agents in these counterexamples fail to satisfy the antecedents of the generalizations. Consider Sue again. Sue sincerely assents to `I want Twain to nod' but dissents from `I want Clemens to nod'. She assents to `If I wave then Clemens will nod' but dissents from `If I wave then Twain will nod'. So she does not wave. Ordinary speakers (in seems-true contexts) would think that (14) is false.

(14) Sue believes that if she waves then Twain will nod.

And so they would deny that Sue satisfies the antecedents of generalizations (6a)-(6d). So, in the opinions of ordinary speakers (in these contexts), Sue's case is not a counterexample to the generalizations. So ordinary speakers would think that the generalizations are true, even if Russellianism were correct and the mismatching exceptions were genuine counterexamples to the generalizations. ⁽⁴²⁾ And so Russellians can explain why the generalizations appear to be true (to ordinary speakers in seems-true contexts), even though they are false (in such contexts).

Of course, to complete the above explanation, Russellians need to explain why ordinary speakers (in seems-true contexts) think that (14) is false and that Sue does not satisfy the antecedents of (6a)-(6d). But to do this, a Russellian need only appeal to his explanation of anti-Russellian intuitions regarding simple attitude ascriptions. He could appeal to a confusion between semantics and pragmatics, as do Salmon and Soames. Or (better) he could appeal to the fact that a person can believe the propositions expressed by the antecedents of the generalizations in different ways, as I do. (See the end of section 2.) Thus as long as a Russellian can plausibly explain how speakers have mistaken intuitions about the truth values of simple attitude sentences, he can also explain why ordinary speakers have mistaken intuitions about the truth values of psychological generalizations.

A similar reply can be given to Richard's Singular Objection. Let's assume that if generalizations (6a)-(6d) are false under Russellianism, then (8) is false also.

(8) If Sue wants Twain to nod, and she believes that (if she waves, then Twain will nod), then she will tend to wave.

Let's suppose that if (8) is false under Russellianism, then what makes (14) and (15) true does nottend to make (16) true, in Richard's intended ceteris paribus sense of `tends to make true'. ⁽⁴³⁾ A Russellian could admit that this is a consequence of his view, but insist that it is correct. A Russellian could also insist that he can account for any ordinary intuitions to the contrary. The possible cases that constitute counterexamples to (8) are those in which Sue has mismatching ways of believing and desiring the relevant propositions. Ordinary speakers would think that the antecedent of (14) is false in such cases; and so they would not take them to be counterexamples to (14). That's why speakers (wrongly) think that (14) is true, despite counterexamples, and why they (wrongly) think (insofar as they do) that what makes (14) and (15) true also tends to make (16) true, (in Richard's ceteris paribus sense).

12. Conclusion

I have argued that psychological generalizations like (6a)-(6d) are true, in seems-true contexts, even if Russellianism is true; for in those contexts, the suitable conditions associated with the generalizations require that the agent believe and desire the relevant propositions in matching ways. But even supposing (contrary to my expectations) that those generalizations are false under Russellianism (in those contexts), they would still appear to be true to ordinary speakers (in those contexts), and so Russellians could account for ordinary judgments that they are true. Either way, Russellians have available to them a reasonable reply to the Generalization Objection.

One point emerged several times in my discussion of the Generalization Objection: if a Russellian has a plausible explanation of speakers' anti-Russellian intuitions regarding simple attitude ascriptions, such as (1) and (2), then he can rely on that explanation to give a reasonable reply to the Generalization Objection. Perhaps this goes to show some may have suspected all along: the most important question concerning Russellianism is whether Russellians can give a plausible reply to the Substitution Objection. ⁽⁴⁴⁾

References

Braun, David. (1995) "Causally Relevant Properties." Philosophical Perspectives 9, pp. 447-475.

Braun, David. (1996) "Demonstratives and Their Linguistic Meanings." Nous 30, pp. 145-173.

Braun, David. (1998) "Understanding Belief Reports." Philosophical Review 107.

Chierchia, Gennaro and McConnell-Ginet, Sally. (1993) Meaning and Grammar: An Introduction to Semantics. Cambridge, MA: MIT Press.

Crimmins, Mark. (1992) Talk About Beliefs. Cambridge, MA: MIT Press.

Devitt, Michael. (1996) Coming to Our Senses: A Naturalistic Program for Semantic Localism. Cambridge: Cambridge University Press.

Devitt, Michael. (1997) "Meanings and Psychology: A Response to Mark Richard." Nous 31, pp. 115-131.

Donnellan, Keith. (1972) "Proper Names and Identifying Descriptions." In Donald Davidson and Gilbert Harman (Eds.) Semantics of Natural Language, pp. 356-379. Dordrecht: Reidel.

Dowty, David. (1985) "On Recent Analyses of the Semantics of Control." Linguistics and Philosophy 8, pp. 219-331.

von Fintel, Kai. (1998) "Quantifiers and `If'-clauses." Philosophical Quarterly 48, pp. 209-214.

Fodor, Jerry. (1991) "You Can Fool Some of the People All of the Time, Everything Else Being Equal; Hedged Laws and Psychological Explanation." Mind 100, pp. 19-33.

Fodor, Jerry. (1994) The Elm and the Expert: Mentalese and Its Semantics. Cambridge, MA: MIT Press.

Garcia-Carpintero, Manuel. (1998) "Indexicals as Token-Reflexives." Mind 107, pp. 529-563.

Haegeman, Liliane. (1991) Introduction to Government and Binding Theory. Oxford: Blackwell.

Kaplan, David. (1989) "Demonstratives" and "Afterthoughts". In Joseph Almog, John Perry, and Howard Wettstein (Eds.), Themes from Kaplan, pp. 481-614. Oxford: Oxford University Press.

Kripke, Saul. (1980) Naming and Necessity. Cambridge, MA: Harvard University Press.

Lewis, David. (1973) Counterfactuals. Oxford: Blackwell.

Lewis, David. (1975) "Adverbs of Quantification." In Edward Keenan (Ed.), Formal Semantics of Natural Language, pp. 3-15. Cambridge: Cambridge University Press.

Lewis, David. (1986a) "Counterfactuals and Comparative Possibility." In Lewis 1986b, pp. 3-31. Originally published in Journal of Philosophical Logic 2 (1973), pp. 418-446.

Lewis, David. (1986b) Philosophical Papers Volume II. Oxford: Oxford University Press.

Marcus, Ruth Barcan. (1961) "Modalities and Intensional Languages." Synthese 13, pp. 303-322.

Morreau, Michael. (1997) "Fainthearted Conditionals." Journal of Philosophy 94, pp. 187-211.

Perry, John. (1977) "Frege on Demonstratives." Philosophical Review 86, pp. 474-497.

Perry, John. (1979) "The Problem of the Essential Indexical." Nous 13, pp. 3-21.

Richard, Mark. (1987) "Attitude Ascriptions, Semantic Theory, and Pragmatic Evidence." Proceeedings of the Aristotelian Society 87, pp. 243-62.

Richard, Mark. (1990) Propositional Attitudes: An Essay on Thoughts and How We Ascribe Them. Cambridge: Cambridge University Press.

Richard, Mark. (1997a) "Propositional Attitudes." In Bob Hale and Crispin Wright (Eds.), A Companion to the Philosophy of Language, pp. 197-226. Oxford: Blackwell.

Richard, Mark. (1997b) "What Does Commonsense Psychology Tell Us About Meaning?" Nous 31, pp. 87-114.

Riemsdijk, Henk van and Williams, Edwin. (1986) Introduction to the Theory of Grammar. Cambridge, MA: MIT Press.

Salmon, Nathan. (1986) Frege's Puzzle. Cambridge, MA: MIT Press.

Salmon, Nathan. (1989) "Illogical Belief." Philosophical Perspectives 3, pp. 243-285.

Schiffer, Stephen. (1991) "Ceteris Paribus Laws." Mind 100, pp. 1-17.

Silverberg, Arnold. (1996) "Psychological Laws and Non-Monotonic Logic." Erkenntnis 44, pp. 199-224.

Soames, Scott. (1988) "Direct Reference, Propositional Attitudes, and Semantic Content." In Nathan Salmon and Scott Soames (Eds.), Propositions and Attitudes, pp. 197-239. Originally published in Philosophical Topics 15 (1987), pp. 47-87.

Soames, Scott. (1990) "Belief and Mental Representation." In Philip P. Hanson (Ed.),Information, Language, and Cognition, pp. 217-246. Vancouver: University of British Columbia Press.

Soames, Scott. (1995) "Beyond Singular Propositions?" Canadian Journal of Philosophy 25, pp. 515-550.

Stalnaker, Robert. (1968) "A Theory of Conditionals." In Nicholas Rescher (Ed.), Studies in Logical Theory, pp. 98-112. Oxford: Blackwell.

Notes

1. See sections 3 and 7 for stronger, more carefully stated versions of the objection. See note 16 for comments concerning my attribution of the objection to Richard and Devitt.

2. Devitt (1996, pp. 151-3, 174-5, 184, 230, 243) seems to argue along roughly these lines.

3. As we will see, this rough statement of my first reply needs to be modified. See section 8. I am not the first to claim (roughly) that psychological generalizations like (3) are true, even if Russellianism is true. Soames (1990) and Fodor (1994) have made similar claims. But their views differ from mine in various ways, and I argue for my view in different ways than they do. I can best describe the differences at the end of section 8. See note 35.

4. Crimmins (1992, pp. 33-34) argues that generalizations like (3) are false under Russellianism, and assumes that if a Russellian is to defend his view, then he must maintain that utterances of (3) pragmatically convey a true proposition. I think that most critics of Russellianism, and even most Russellians, would agree with Crimmins that Russellians need to appeal to pragmatics. For example, Richard (1987) takes a pro-Russellian stance, and assumes that Russellians should make heavy use of pragmatics to defend their view from objections concerning generalizations and explanation. (Richard no longer endorses the Russellianism presented in that paper.)

5. I briefly discuss the causal relevance of broad content in Braun 1995. Here is a hint at what I think about the objection concerning explanation. First, I believe that the generalizations would be true if Russellianism were correct. Second, I think that, even if they were false under Russellianism, attitude ascriptions could still successfully explain behavior. (I hope to go into this further in a future paper.)

6. There is one related topic that I will occasionally touch on in this paper: whether psychological generalizations like (3) would be nomologically necessary under Russellianism. See notes 15, 22, and 28 below. But I shall not try to respond to the objection that (3) would fail to be a law under Russellianism. Discussing this would require a discussion of the nature of laws and whether ceteris paribus generlizations of any sort can be laws.

7. I follow Braun (1998) quite closely in sections 1 and 2 below. For more detailed descriptions of Russellianism than the one I give below, see Salmon 1986, Soames 1995, and Richard 1990, chap. 3. Russellians should perhaps say that propositions are among the objects of attitudes such as believing and desiring, in view of the discussion in note 9.

8. I assume that the content of a proper name is constant from context to context. I also ignore matters of time and tense throughout this paper. A Russellian who wished to take tense and time into account might hold that `is an author' expresses different "time-indexed" properties in different contexts. Similarly, `believes' might be held to express distinct "time-indexed" binary relations in different contexts. For a detailed treatment, see Salmon 1986.

9. As far as I know, Russellians have not explicitly discussed attitude ascriptions with infinitival complements. Attitude ascriptions with infinitival complements that lack explicit subjects may need a different semantic treatment from those that do have explicit subjects. Consider, for instance, `John wants to run'. According to some syntactic theories, the syntactic complement of `wants' in this sentence is (roughly) a full sentential clause. The subject noun phrase of this clause is an unpronounced constituent, PRO; `to run' constitutes the verb phrase of the clause. This syntactic analysis suggests (though it does not entail) that the infinitival phrase is, semantically speaking, like a `that'-clause, and that a Russellian semantics could treat this desire ascription in (roughly) the same manner as belief ascriptions. Other syntactic theories say that there is no unpronounced constituent like PRO; the syntactic complement of `wants' is simply some sort of verb phrase. Theories of this sort suggest that the sentence relates John to a property and that the constituents of the proposition expressed by the sentence are John, the relation of wanting, and the property of running. It is also possible to have a "mixed" theory; for example, one could hold that the infinitival phrase is a sentential clause containing an unpronounced PRO, but that the content of the clause is a property. For discussion of the syntax of such sentences, see Haegeman 1991, chapter 5, and Riemsdijk and Williams 1986, chapter 8. For a discussion of their semantics (and syntax), see Chierchia and McConnell-Ginet 1993, pp. 247-252, and Dowty 1985. Notice that the "property analysis" would not preclude a Russellian from maintaining that John also stands in the wanting relation to the proposition that John runs. A Russellian could say that it's metaphysically necessary that if John wants the property, then he also wants the proposition. But the sentence itself would express a proposition that has the property of running as a constituent, and not the proposition that John runs.

10. See Kripke 1980, Donnellan 1972, and Kaplan 1989. See also Marcus 1961 and Perry 1977.

11. For descriptions of the standard Russellian view of the metaphysics of belief, see Salmon 1986 and 1989, Soames 1995, and Richard, 1990, chapter 3.

12. Strictly speaking, the third entity is usually said to determine the proposition believed or desired only given the circumstances of the agent (for example, the causal and historical relations in which she stands to objects, and to instances of properties and relations).

13. On all of these views, the binary belief relation is (in effect) metaphysically analyzable into a ternary relation between agents, propositions, and ways of taking propositions, which we may call `BEL'. The binary belief relation is (in effect) an existential generalization of BEL in the following sense: necessarily, x believes p iff there is some way w of taking proposition p such that BEL[x, p, w]. Similarly, the binary wanting or desiring relation is analyzable into a ternary relation DES such that: necessarily x wants (desires) p iff there is some way w of taking p such that DES[x, p, w].

14. In this paper, I often use a sentence's numerical index as an abbreviation for its quotation-name, especially in `that'-clauses. For example, I use (i) as an abbreviation for (ii).

(i) The speaker believes that (4) is true.

(ii) The speaker believes that `Twain is a writer' is true.

15. Devitt might claim that they were laws or nomological truths. See note 28.

16. I should confess that only Crimmins explicitly presents the Generalization Objection in print; see Crimmins 1992, pp. 32-4. My attribution of the argument to Richard and Devitt is based on interpretive inferences. Richard explicitly says that the generalizations are true; see Richard 1987; 1990, p. 44, n. 16; 1997b, p. 96. Devitt also says that they are true, under one "reading"; see Devitt 1996, pp. 151-4, 184, 230-231, 242-243; 1997, p. 122. They also claim that attitude ascriptions would not be useful for explaining behavior under Russellianism; see the previous passages from Devitt and the passages from Richard cited in note 37. They both hold that true generalizations are needed for explanation (see the previous passages). And they both hold that generalizations that allude to ways of taking propositions are needed for explanation (see previous passages). I infer from all of this that they hold that the generalizations would be false under Russellianism. In addition, Richard (1997a, p. 218, n. 23) says, "Richard (1990) and Crimmins (1992) argue that the view's [Russellianism's - DB] inability to account for the literal truth of folk psychology is a serious flaw". So in this passage, Richard seems to assume that he had previously argued that the generalizations would be false under Russellianism. In any case, the Generalization Objection is worthy of serious consideration, whether or not Devitt and Richard endorse it.

17. I am making a number of assumptions here that I wish to enumerate. First, I am ignoring tense and time throughout this paper, so I am not not worrying about the differences between `Twain nods', `Twain will nod', and `Twain to nod', as they appear in the embedded clauses of (6a)-(6c). Second, I am assuming that all occurrences of `she' in (6a)-(6c) are bound by quantifiers in some manner similar to the way in which quantifiers bind variables in familiar formal languages. I also assume that `she' is bound by `Sue' in some similar way in sentences like `Sue believes that if she waves then Twain will nod'. I am not confident that this latter assumption is correct, but by making it I can avoid thinking about semantic issues that are irrelevant to my concerns here. Third, I assume that if Sue sincerely assents to `If I wave then Twain will nod', or `I want Twain to nod', or similar sentences containing `Clemens' instead of `Twain', then she believes the propositions that the sentences express, with respect to the contexts in which she assents to them, which (I assume) are contexts in which `I' refers to Sue. Thus in the scenario described in the previous sentence, Sue believes the proposition that if Suewaves then Twain will nod, and she believes the proposition that Sue wants Twain to nod. Fourth, I assume that Sue believes these latter propositions in first-person, or "I-ish", ways. We can imagine that she has the relevant `I'-sentences in her belief box; thus she has beliefs about herself, while taking herself in a first-person way. Fifth, I assume that if Sue believes, in a first-person way, that Sue wants Twain to nod, then Sue does want Twain to nod, in a first-person way.

Let me say why I am making the last two assumptions. If Sue believes or desires one of the relevant propositions in a non-first-person way, then she may fail to wave. (For instance, if Sue were willing to assert "If she waves, then Twain will nod", while pointing at herself in a mirror, but unwilling to assert "If I wave, then Twain will nod".) A critic of Russellianism might appeal to such "Perry-like" cases (see Perry, 1979) to argue that the generalizations would be false under Russellianism. I think that such non-first-person cases raise basically the same issues that the `Twain'-`Clemens' cases do; I think that Russellians can respond to such cases in much the same way that I respond to the `Twain'/`Clemens' cases. See note 32.

18. Some readers might think that Wilma's case does falsify the generalizations. Other readers' intuitions about the case might waver. In section 5 below, I present a semantic account of the generalizations that can account for conflicting and wavering intuitions about such cases.

19. To be even more precise: By `the consequent of the generalization' I mean `she waves'. I do not include phrases like `other things being equal' in the consequent. I assume that these latter phrases constitute part of the conditional connective. By a (possible) case, I mean a nomologically possible event such that some individual involved in the event satisfies the antecedent of the generalization. There exists an exception to a generalization iff: there is a nomologically possible world (a world in which all of the laws of our world hold) such that an individual in that world satisfies the antecedent of the generalization but fails to satisfy the consequent.

20. There may be semantic differences between the various generalizations and conditionals I have mentioned, due to differences in the phrases that appear in them (such as `tends to', `other things being equal', and `usually'). But determining the differences (if any) between them would take more space than I have here. Moreover, the proponents of the Generalization Objection do not attend to their semantic differences (if any). Therefore, I shall assume that they can receive a uniform semantic treatment. I shall also assume that (6a)-(6d) are logically equivalent to universal generalizations of conditionals of some sort; thus in an appropriate formal language we could symbolize them with sentences of the form x(Ax > Cx), where `>' is a ceteris paribusconditional connective. According to an alternative analysis, however, phrases like `usually' and `other things being equal' are adverbs of quantification. On such an analysis, the generalizations would be better symbolized with sentences of the form (Ceteris paribus: Ax)Cx. In the latter formula, the ceteris paribus phrase is an unselective variable-binding (modal) operator of some sort, and the open formula Ax is a restriction on it. This sort of analysis might be more easily extended to generalizations that do not contain an explicit conditional connective, such as `Anyone who wants Twain to nod and believes that (if she waves then Twain will nod) will, other things being equal, wave'. So there may be various reasons to prefer the alternative analysis. I strongly suspect, however, that these details do not make a substantive difference to the issues I am discussing here, and so I shall ignore them. For relevant discussion, see Lewis (1975, 1986a, secs. 2.7 and 6), von Fintel (1998), and the references in the latter.

21. I have relied heavily on Silverberg (1996) and Morreau (1997), especially Morreau, in the following remarks on ceteris paribus generalizations and conditionals. I have not followed either of them exactly. In particular, I have simplified Morreau's account considerably, sometimes for convenience, but sometimes also for theoretical reasons having to do with the nature of propositions and the proper treatment of context-sensitive expressions.

22. In fact, in many contexts, utterances of (8) would seem to us to be nomologically true; it isnot accidental that cars' engines start when their ignition keys are turned. Perhaps in these contexts of utterance, an utterance of (8) is true iff an utterance of its nomological necessitation is true also. See note 28.

23. In this section and the next, I ignore certain complications raised by multiple occurrences of a single context-sensitive expression. See section 10.

24. In this paragraph, I have emphasized that speakers' thoughts about the typicality of cases that satisfy a generalization's antecedent determine the suitable conditions associated with that generalization in a context. But one might wonder whether speakers' thoughts about cases thatfail to satisfy the antecedent can also help determine suitable conditions. I think there are two reasons to think that their thoughts about the "satisfying" cases are (at the very least) much more important. First, a claim or thought that a case is typical or normal is always relativized to some reference class or type. If one thinks that a case is typical, one thinks that it is a typical case of a certain class or type of case, and not just typical tout court. In the above discussion, the relevant class of cases seems to be the class of cases that satisfy the antecedent. Second (and more imporant), some of the suitable conditions associated with a generalization simply cannot hold in a case unless that case satisfies the antecedent. Consider, for example, the generalization `if a match is struck, it will light'. Among the suitable conditions for this generalization is the condition that the match be struck with sufficient force and speed. This aspect of the suitable conditions cannot hold in a case unless the case satisfies the antecedent. So it seems that the thoughts of speakers about cases in which the antecedent is satisfed are necessary to determine the full set of suitable conditions for a generalization. (Thanks to Mark Richard for discussion of these matters.)

25. These properties of, or facts about, the world need not appear as constituents in the propositional content of (9) in these contexts. See the end of section 6. Thanks to John Bennett and Graeme Forbes for discussion of the matters mentioned in the above paragraph.

26. I also ignore the fact that a semantics for ceteris paribus conditionals may have to allow distinct occurrences of the ceteris paribus conditional connective to be associated with distinct suitable conditions within a single context. See section 10 for discussion.

27. I am assuming (as Lewis and Stalnaker do) that every world is at least as similar to itself as any other.

28. I derive these truth conditions from Silverberg (1996) and Morreau (1997), with some modifications. Both present truth conditions for ceteris paribus conditionals that resemble Lewis's (1986a) truth conditions for counterfactual conditionals. If need be, they could be reformulated to accomodate the view that `ceteris paribus' is an adverb of quantification. See note 20.

I assumed in the previous section that the context-sensitivity of ceteris paribusconditionals is due to shifts in the suitable conditions determined by contexts. But on Lewis's view of counterfactuals, the similarity relation between worlds can also shift from context to context. Thus I cannot rule out the possibility that the "shiftiness" of truth conditions for ceteris paribus conditionals is due to shifts in the associated similarity relation rather than shifts in the associated suitable conditions. But I believe that, whichever one shifts, the shifts are due to contextual factors similar to those I describe above. Thus I think the exact source of the shift in truth conditions (whether suitable conditions or similarity relation) does not affect the issues I wish to discuss here.

Among the suitable conditions determined by a context might be the condition that the laws of our world hold. If this is so, then for some contexts c, a sentence of the form if A thenceteris paribus C is true at the world of c iff its nomological necessitation is: it's nomologically necessary that (if A then ceteris paribus C). This might help to account for the seeming nomological force of some ceteris paribus conditionals and generalizations in certain contexts. See notes 6, 15, 22, and 30.

29. The preceding determination relations can be represented using a function: there is a function F from conditional relations to pairs of suitability relations and ternary relations between worlds such that, for all contexts c, F(COND_c) = <being-suitable_c, T_c>, where T_c is the appropriate ternary relation between worlds.

30. Schiffer (1991) and Fodor (1991) discuss the truth conditions of ceteris paribuspsychological laws, among other things. Their concerns are different from mine, but are similar enough that I should (perhaps) comment on them here.

Schiffer is concerned with sentences of the form Ms cause Bs, ceteris paribus, where Mis some sort of sentence (or predicate) concerning mental states and B is some sort of sentence (or predicate) concerning behavior. He eventually accepts the following analysis of such sentences (Schiffer 1991, p. 7):

Ms cause Bs, ceteris paribus, iff: for each of "sufficiently many" realizations D of M there is a same-level condition C such that D-&-C is non-superfluously causally sufficient for aB event.

Schiffer thinks that, if this is the correct analysis of ceteris paribus sentences of the type with which he is concerned, then they can express true propositions. But he argues that, if this is the correct analysis, then ceteris paribus sentences do not express psychological laws, for (he claims) they cannot be used as premises in covering-law explanations. I am not sure to what degree sentences of the form Ms cause Bs, ceteris paribus semantically resemble the ceteris paribusgeneralizations and conditionals with which I am concerned here. (Oddly, the sentence that Schiffer displays on p. 2 as an instance of the above form is not an instance: that sentence is an explicit conditional and does not contain the term `cause'.) Insofar as they are similar, I believe that a semantic account of them should be derived from a more general account of ceteris paribus conditionals, as I have done here. I suspect that Schiffer's account suffers because he does not do this. Unlike Schiffer, I am not concerned here with whether ceteris paribusgeneralizations deserve to be called `laws'. But I think that Schiffer's argument against their lawhood is weak, because it does not recognize that ceteris paribus generalizations can provide some degree of (non-deductive) support for explananda.

Fodor (1991) mainly aims to show that sentences of the form ceteris paribus AB may be laws despite the existence of "absolute exceptions". But Schiffer's analysis, and his argument against their lawhood, seem to allow this. The "mismatch" between Schiffer's argument and Fodor's reply may have occurred because Fodor aimed his reply at an earlier draft of Schiffer's paper; see Schiffer 1991, note 1, and Fodor 1991, note 1.

31. For the sake of simplicity, I have presented the above objection using the terminology of "possible cases" and "counterexamples". But, strictly speaking, the above reasoning does not validly support the conclusion, or at least not if the formal semantics sketched in the previous section is correct. To see why, let's restate the crucial premise using explicit talk of possible worlds: in some possible world, an agent satisfies the antecedents and the suitable conditions (of the context) hold, and the agent believes and desires the relevant propositions in mismatching ways. (And in that world, the agent does not wave.) But this is not enough to show that the psychological generalizations are false, if the previous semantic theory for them is correct. To establish that they are false, we need an additional premise: there is no world that is more similarto the actual world in which the agent satisfies the antecedent, and suitable conditions hold, and the agent believes and desires the relevant propositions in matching ways (and so waves). This additional premise is needed, for if there were such a world, then the generalizations still might be true. The reasoning could also be accurately conducted using counterfactuals, as follows. There is an agent such that: if he were to satisfy the antecedents, and the suitable conditions of the context held, then he might believe and desire the relevant propositions in mismatched ways; and if he were to satisfy the antecedents, and the suitable conditions of the context held, and he believed and desired the relevant propositions in mismatched ways, then he would not wave. Therefore, there is an agent such that if he were to satisfy the antecedents, and the suitable conditions of the context held, then he might not wave. But if this is so, then generalizations (6a)-(6d) are false. This reasoning is valid, assuming the earlier semantics for ceteris paribusconditionals, and Lewis's semantics for counterfactuals and his views on the interdefinability of `would' and `might' counterfactuals. See Lewis 1986a, especially p. 8.

32. This reply is also adequate to respond to the "more accurate" version of the Generalization Objection given in the previous note. Using explicit possible worlds talk, the reply says that there is no possible world in which an agent satisfies the antecedents, and suitable conditions hold, and the agent believes and desires the relevant propositions in mismatched ways. Using purely modal talk, the reply would go as follows: It is not possible that an agent satisfy the antecedents, and suitable conditions hold, and the agent believes and desires the relevant propositions in mismatched ways. Therefore, it is not the case that there is an agent such that (if he were to satisfy the antecedents, and suitable conditions held, then he might believe and desire the relevant propositions in mismatched ways).

A similar reply can be given to arguments that appeal to "Perry-like" cases to argue that the generalizations would be false under Russellianism (see note 17): the suitable conditions for the generalizations, in seems-true contexts, include the condition that the agent believe and desire the relevant propositions about himself in first-person ways. My arguments for this claim would be similar to some of those I give below for the matching-ways hypothesis.

33. Mark Richard (in correspondence) argues that if the suitable conditions for (6a), in seems-true contexts, hold in Sue's case, and Russellianism is true, then her case is a genuine counterexample to (6a), even if ordinary speakers do not think that her case satisfies the antecedent of (6a). So to defend Russellianism, I need to argue that suitable conditions do not hold in Sue's case. I agree. I provide such an argument below (in the main text) when I point to the pattern of speakers' judgments about various cases, including those like Sue's. (Richard remains unconvinced.)

34. See also note 24 for relevant discussion.

35. I can now compare my first reply to the Generalization Objection with the responses of Soames (1990) and Fodor (1994) to similar objections.

My response is closest to Soames's (1990, pp. 232-3). Soames says that ordinary psychological generalizations contain ceteris paribus clauses, and that among the conditions specified by such clauses are the condition that the agent apprehend certain propositions in the same way. Soames, however, does not attempt to argue for this claim. He does not discuss the semantics of ceteris paribus generalizations or note their context-sensitivity.

Fodor (1994) is not directly concerned with Russellianism or with any semantic theory of English. He instead wishes to show that there are true intentional psychological generalizations that make use of broad content attributions. He (wrongly) assumes that generalizations that ascribe broad content attitudes to agents must exhibit the same sort of transparency to substitution of co-referring proper names as do ordinary attitude ascriptions in English according to Russellianism. So we can view Fodor as being indirectly concerned to show that ceteris paribus psychological generalizations in English would be true under Russellianism. (But we must be cautious here, for Fodor's use of attitude ascriptions is often inconsistent with Russellianism. See pp. 40-1; but see also p. 46.) Fodor thinks that such generalizations can be true despite Frege cases (as he calls them), because Frege cases do not satisfy the ceteris paribusclauses of such generalizations, and so are tolerable exceptions to the generalizations. In this respect, his response to (something like) the Generalization Objection is similar to mine. But his reasons for thinking that the Frege cases are tolerable exceptions are quite different from mine. Fodor says that agents "can be relied upon to know that a=b if the fact that a=b is germane to their behavior" (p. 41). Presumably what Fodor means is this: if a=b, and an agent believes that a=a in an `a=a' way, then that agent will believe that a=b in an `a=b' way. Fodor seems to think that this fails to be the case in Frege cases, and that this is why the agents do not satisfy theceteris paribus clauses. I believe that Fodor's condition for satisfying the ceteris paribus clause is too strong. On my view, an agent need only believe and desire the propositions in matching ways in order to satisfy the ceteris paribus clauses (in typical contexts); she need not accept the relevant identity sentences to do this. For example, Sue need not believe that Twain is Clemens in a `Twain=Clemens' way in order to believe and desire the propositions mentioned by (6a)-(6d) in matching ways, and so satisfy the ceteris paribus condition. There are other differences between Fodor and me. Fodor fails to note the context-sensitivity of ceteris paribusgeneralizations. Fodor also says that the "falsification" of ceteris paribus generalizations has something to do with the actual frequency and systematicity of exceptions (p. 39). On my view of ceteris paribus generalizations, the actual frequency of exceptions to a generalization (with respect to a context) has nothing to do with its truth or falsehood (in that context). Perhaps it does have something to do with the `systematicity of exceptions', but it's unclear what Fodor means by this.

36. Graeme Forbes (in correspondence) has formulated an objection that appeals to generalizations with a more prescriptive flavor than those I discuss above. He argues that there are true generalizations concerning rationality, such as (i), that would be false if Russellianism were true.

(i) If a person believes that (if P then Q), and believes (or comes to believe) that P, then it is rational for her to infer that Q.

Forbes distinguishes between inferring that Q and coming to believe that Q. It is always rational for an agent who satisfies the antecedent of (i) to infer that Q (roughly, come to entertain Q as a result of reasoning from prior beliefs). But sometimes it is not rational for such an agent to come to believe that Q, for example, when such an agent has strong evidence for not-Q. Forbes claims that (i) is not a ceteris paribus generalization (it is "absolutely true"), and so he claims that I cannot appeal to associated suitable conditions to argue that it would be true under Russellianism. I believe, contrary to Forbes, that (i) is also ceteris paribus, because there are tolerable exceptions to it in which the agent believes the propositions in mismatching ways. But I cannot describe any "non-Russellian" tolerable exceptions to (i); thus Forbes can deny that I have really found any tolerable exceptions to it. Thus it may appear that we have reached a stalemate. But if I am correct in arguing that there are tolerable Russellian exceptions to (6a)-(6d), then (I maintain) it's plausible to conclude that there are also tolerable Russellian exceptions to (i).

37. As is clear from the passage I quoted above, Richard goes on to conclude that (14) and (15) fail to explain (16), if Russellianism is correct. I shall not deal with this part of Richard's argument here (I hope to address it in future work). Richard presents similar objections to Russellianism in a number of places. See Richard 1987, pp. 248, 252, and 257; 1990, pp. 126 and 219; 1997a, pp. 202. For related remarks on attitude ascriptions and explanation see Richard 1990, pp. 44 (note 16); 84; 103; 208; and 260-263.

38. More accurately: the simple attitude sentences are not context-sensitive according to Russellianism. Richard, Crimmins, and Perry say that simple attitude sentences are context-sensitive, simply because they contain attitude verbs. Nevertheless, Russellians and non-Russellians alike can agree that the generalizations involve a kind of context-sensitivity not present in simple attitude sentences, namely that created by the occurrence of the ceteris paribusphrase.

39. Two more details. (i) If two utterances of ceteris paribus generalizations (whether of the same generalization or two different generalizations) are associated with different suitable conditions, then an advocate of the first response is committed to saying that these two utterances occur in different contexts. (ii) An advocate of this response would have to allow that both (6a) and (6e) can fail to express a proposition in contexts where the speakers do not have suitable intentions.

40. There is a second possible source for the mistaken intuition: (6e) is likely to be false in any context in which it is uttered (since the intentions of such a speaker are unlikely to determine that the context's suitable conditions include the matching-ways condition). But according to the above response, (6a) is also false in such contexts.

41. To describe precisely how (6a) and (6e) can differ in content with respect to the same context, we would need a semantic theory that says that their contents are "doubly" relativized, to something like contexts and intentions. For discussion of semantic theories of this sort, see Braun (1996).

42. As I pointed out in section 8, there are other mismatching exceptions that ordinary speakers, in seems-true contexts, would take to be tolerable exceptions (for example, Harry's case). There are no mismatching exceptions that ordinary speakers, in seems-true contexts, would take to be counterexamples to the generalizations.

43. Notice, however, that if `tends to make true' is understood in the sense that I explicated in section 9 using sets A and B, then what makes (14) and (15) true does tend to make (16) true, even if the ceteris paribus generalizations are false under Russellianism.

44. Many thanks to John G. Bennett, Earl Conee, Michael Devitt, Jennifer Saul, Susan Schneider, Theodore Sider, and the members of my seminar in the fall of 1997, for comments and criticisms. Special thanks to Graeme Forbes and Mark Richard for extensive, and very helpful, critical comments. I presented an abbreviated version of this paper at the American Philosophical Association's 1999 Pacific Division meeting. I want to thank my commentator, Simon Evnine, and the participants in the discussion that followed, for their comments.