Philosophy 1850

Grice has two main goals in this paper. The first is to distinguish 'natural' from 'non-natural' meaning. Ultimately, Grice's interest is in non-natural meaning, which is supposed to be the kind of meaning that linguistic utterances have (and so the kind of meaning that language has). The second goal is to offer an account of 'non-natural' meaning, or at least of one important variety of it.

So one main reason we are reading this paper is to try to get a little clearer about what notion of meaning concerns the philosophy of language. However, when Grice talks about 'meaning' here, even 'non-natural' meaning, he does not have in mind 'meaning' in the sense in which words have meaning. See below for more on this.

There have been attempts to reduce the notion of non-natural meaning to that of natural meaning. The motivation for this is that natural meaning seems, in some cases, to be a natural phenomenon. Consider, e.g., the case of tree rings: that the tree has 36 rings means that it is 36 years old. So some philosophers (such as Fred Dretske and Robert Stalnaker) have wanted to start with this notion of 'information' and use it to try to explain non-natural meaning, which is more puzzling. We will not discuss such issues, however, as they are more in philosophy of mind.

On p. 387, Grice mentions five 'tests' that can be used to separate natural from non-natural meaning. What exactly are the five tests? How good are they? Are some better than others? (One of the tests just seems wrong to me.)

<--! 1. Factivity 2. 'what was meant' (similar to next one) 3. someone meant, or not 4. inverted commas 5. "The fact that..." is what means in the natural case -->

One of Grice's tests is that, where natural meaning is concerned, "x means that p" implies that p is true. A verb with this property is said to be 'factive'. The usual example is "know": If Alex knows that p, then p must be true, so "know" is factive. Grice is thus claiming that "mean" in the natural sense is factive, whereas "mean_NN" is not factive.

One good way to check your understanding of Grice's distinction would be to give some of your own examples of the two kinds of meaning. Do Grice's 'tests' seem to characterize them correctly?

The use of 'mean' that Grice notes at the bottom of p. 378—where "A means to do X" means something like: A intends to do X—is arguably not a use of the same word. Other languages have different words for these two 'senses' of mean, which suggests that this is a genuine lexical ambiguity.

Grice very much seems in this paper to be trying to 'analyze' (roughly, define) certain ordinary uses of the verb "mean". That, presumably, is what justifies the close concern that some of his tests display for the niceties of ordinary usage. But is that really what Grice's focus is (or what it should be)? What else might one read him as trying to do?

Grice's second goal is to offer an account of non-natural meaning. He first considers and rejects a 'causal' account due to Charles Leslie Stevenson. Our main attention will be elsewhere, but you should nonetheless read this part of the paper, as it helps to motivate Grice's own view, which in some ways builds on Stevenson's account.

Note that Grice's target here is not what words or sentences mean. It is, rather, what someone might mean by an utterance of certain words, which might diverge from what those words mean in their own right. Irony and metaphor would be obvious examples. (We'll look more deeply into this matter when we read Grice's paper "Logic and Conversation", and then further when we get to the unit on Contextualism.) Note also that Grice means his analysis to apply to gestures and other non-linguistic forms of communication, as well as to the use of language. So Grice's goal is to explain what it is for a particular 'utterance' (in a very general sense) to mean something, in the non-natural sense. He initially restricts himself, for convenience, to 'informative' utterances (as opposed, e.g, to questions).

What exactly does Grice have in mind when he speaks of "the difference between 'deliberately and openly letting someone know' and 'telling'" (p. 382)?

Grice's account may be formulated as follows: S meant by the 'utterance' U, made to an audience A, that p iff S made this utterance with the intention that A should come to believe that p as a result of A's recognizing that S has that very intention. There are three main aspects to this account:

1. It incorporates Stevenson's idea that meaning something is, in some sense, connected with trying to get someone to believe something. But, as Grice notes, there are different ways in which that can be done, so there has to be more to it than that. (Here again, a good way to check yourself is to try to come up with your own example that serves the same purpose as Grice's.)
2. It incorportates an 'overtness' condition: that one's intention to get someone to believe something should be an intention that one intends the audience to recognize. But that is not enough, once again, to rule out certain cases that Grice think should not be included. (Try to formulate your own example again.)
3. It incorporates a condition that (1) and (2) should be connected: The audience's recognition of the speaker's intention—that they should come to believe that p—is supposed to play a role in getting the audience to believe that p. Grice motivates this condition with a single example involving a photograph. Is that really enough?

As Grice notes, the 'self-referential intention' with which he is supposing people speak might seem puzzling. But this is not usually taken to be a problem. Self-reference by itself does not lead to paradox. (Consider "All complete English sentences have a subject and a verb". Among the English sentences in question is that very sentence.)

Can you think of any cases that pose a problem for Grice's account? (There are plenty!)

How might one try to extend Grice's account to questions? What about other sorts of 'speech acts'? (Grice mentions imperatives on p. 384, but does not offer a detailed account of them, as he has for indicatives.)

Grice briefly mentions on p. 386 that he wants to restrict the relevant intention to the 'primary' intention with which one speaks. But he only describes in general terms why and does not give an example. Can you? (Note how much easier it would be to understand this paragraph if Grice had given an example. Let that be a lesson.)

Grice finishes the paper by making some remarks about the notion of intention that is deployed in his account. What is the purpose of those remarks?

Does Grice's account of the non-natural meaning of 'utterances' presuppose some prior account of the 'meaning' of mental states, such as beliefs and intentions? Does his account presume that we're able to have beliefs and intentions with particular contents before we understand language with that same content? If so, are these problems for his view?

Strawson refers in this discussion to a number of 'logical laws' that he had listed earlier in the book. These are:

(6) ~(p . ~p)

(7) p ∨ ~ p

(9) ~~p ≡ p

(11) p . q ≡ q . p

(19) ~p ⊃ (p ⊃ q)

(20) ~p ⊃ (p ⊃ ~q)

(21) q ⊃ (p ⊃ q)

(22) q ⊃ (~p ⊃ q)

(23) ~p ≡ [(p ⊃ q) . (p ⊃ ~q)]

(19)-(23) are all forms of the so-called 'paradoxes of material implication' (and there are some redundancies here, too). Note that Strawson uses "." for conjunction and "⊃" for the material conditional.

The main question Strawson discusses here is: What is the relationship between the truth-functional connectives studied in logic and the corresponding words of natural language? With the possible exception of negation, Strawson argues that none of the usual identifications are really correct. For example, "and", according to Strawson, does not mean truth-functional conjunction.

As said, Strawson is prepared to accept the claim that "not" means truth-functional negation, though of course there are complications involving how negation is expressed in natural langauge.

The remarks at the end of §7, about truth and falsity, result from Strawson's view that some sentences, though perfectly meaningful, do not always have truth-values. An example would be "Alex has stopped smoking crack", if Alex has never smoked crack. In this case, the sentence 'presupposes' that Alex used to smoke crack. For Strawson's reasons for this view, see his "On Referring", Mind 59 (1950), pp. 320-44 (PhilPapers) See the SEP article listed as optional for a survey of work on presupposition.

Strawson raises two sorts of objections to the identification of truth-functional conjunction and "and". The first is that "and", in English, has other uses than as a sentential connective. The most interesting of these involves cases such as "Alex and Drew got married", where that may not be equivalent (in any sense) to "Alex got married and Drew got married". But, as Strawson notes, the identification can reasonably be restricted to the case of sentential "and". In that case, however, Strawson observes that

(a) They got married and they had a child.

is "by no means logically equivalent" to

(b) They had a child and they got married.

This seems to show that "and", in English, can have some sort of temporal meaning.

What is Strawson's argument that (a) and (b) are not logically equivalent? What assumptions does that argument make?

Can you think of similar examples where the 'additional content' is not temporal but some other kind of relation between the clauses?

Strawson raises lots of objections to the identification of "⊃" with "if...then...". Unfortunately, however, he plays fast and loose with the distinction between indicative and subjunctive conditionals. A standard example of the contrast involves:

(e) If Oswald did not shoot Kennedy, then someone else did.

(f) If Oswald had not shot Kennedy, then someone else would have.

Note that (e) seems clearly true—someone shot Kennedy—but (f) is much less clear. Conditionals like (f) are known as "counterfactuals", because the antecedent is contrary to (known) fact. No one thinks, or has ever thought (so far as I am aware), that the conditional in (f) was material.

See the SEP article on counterfactuals, listed as optional, for a survey of work on them.

If we set aside Strawson's objections concerning subjunctive conditionals, then there seem to be two sorts of worries left. First, Strawson claims (p. 83) that we 'use' a conditional when (i) we do not know whether the antecedent or consequent is true but (ii) believe "that a step in reasoning from [the antecedent] to [the consequent] would be a sound or reasonable step...". (See also the restatement of this point on p. 88.) This implies that it would be a misuse to assert a conditional whose antecedent one knew to be false or when there was no connection between the antecedent and consequent.

Do Strawson's two conditions seem accurately to characterize the use of the English conditional? If so, does it follow that the English conditional is not material?

Strawson's second objection is that the "joint assertion in the same context" of "If it rains, the match will be canceled" and "If it rains, the match will not be canceled" is "self-contradictory" (p. 85), whereas "p⊃q" and "p⊃~q" are consistent (and imply ~p). Is that right? If so, does it follow that the English conditional is not material?

Suppose that ~p ∨ q. I aim to show that, if p, then q. So suppose p. But of course p and ~p ∨ q together imply q. So q. Hence, the inference from p to q is reasonable. That is: If p, then q. Since Strawson concedes that "if p, then q" implies "~p ∨ q", it would seem to follow that "if p, then q" is equivalent to "~p ∨ q". (This is sometimes known as the 'direct argument' for the conclusion that the English indicative conditional is material.) What should Strawson say about this argument? See p. 92 for some ideas.

Perhaps most surprisingly, Strawson rejects the identification between "∨" and "or". He gives two very different sorts of reasons, both of which involve the rejection of the inference from p to "p or q". (There are logics, known as 'relevance' logics, in which this inference is not valid. See the Stanford Encyclopedia article listed as optional for a survey.) What are Strawson's two reasons? How do they differ? What argument does Strawson give (or what argument might he give) for the claim that "p ∨ q" can be true even if "p or q" is not true?

Grice's central goal in this paper is to distinguish what is "strictly and literally said" in an utterance from other things that might be "meant" or "communicated" in the making of that utterance and, in particular, to articulate a notion of what is implicated by an utterance that is not part of what is said but is part of what is 'meant'. (The term "implicate" is a technical one, but it is meant to echo one informal use of 'implies'.)

As is made clear at the outset, Grice's target here is, in part, Strawson. But he does not, in this lecture, actually formulate his reply to Strawson. His target is really more general. There was a common sort of claim at this time that Grice also wants to oppose. For example, people would often say such things as that it is only appropriate to say "X knows that p" if there is some doubt whether p, so "X knows that p" is only true if there is some such doubt (hence, the usual sorts of analyses of knowledge must be wrong); it is correct to say "X performed action A voluntarily" only if there is some question whether X's action was coerced, so the sentence is only true if there is some such question (which was alleged to have various consequences for discussions of free will); and so forth. In a larger sense, then, Grice is interested in the question when facts about 'correct usage' tell us something about the literal meaning of a word and when they do not.

The basic claim Grice wants to make is this: It is common, even normal, for utterances to suggest or convey or communicate things beyond, in addition to, or even at odds with the literal meaning of the words used. Hence, the fact that "They got married, and they had a baby" will very often suggest or convey a temporal order of these events cannot by itself show that the literal meaning of the sentence is not just: these two things both happened.

Grice begins by distinguishing between what is "said" from what is, in Grice's technical sense of the term, 'implicated'. He introduces this distinction with an example (always a good idea). Its chief advantage is that the 'implicature' (e.g., that C's colleagues are treacherous) is clearly no part of what B's words mean. Grice then makes some very general remarks about the nature of this distinction. What is said is supposed to depend upon the "conventional meaning of the words...uttered" plus certain other factors, such as the resolution of ambiguity and 'context-dependence' (e.g., who was referred to by "he"). What is implicated depends upon a good deal else.

Grice distinguishes two types of implicatures, which he calls "conventional" and "conversational". Conventional implicature will not be our focus, but it is worth making sure you understand the difference between them. Make sure you can give a couple examples of your own, of each type.

It is commonly said that the difference between "and" and "but" is a conventional implicature. So, on this view, "A and B" and "A but B" will always have the same truth-value, though they will often differ in what they 'implicate'. So, "Joan is a woman, and she is good at math" and "Joan is a woman, but she is good at math" seem very different, but the latter is (on this view) still true. What might be the advantages and disadvantages of such a view? (Other cases to consider: It's gorgeous, but it's expensive; she's a great soccer player, but not such a great violinist; I have visited Paris but have not visited Rome.)

There is some controversy whether there is any such thing as conventional implicature. See the optional papers by Bach and Potts.

Grice proceeds to sketch an account of how conversational implicature works. It has two parts: The first is a very general principle that Grice dubs the "Cooperative Principle":

Our talk exchanges do not normally consist of a succession of disconnected remarks, and would not be rational if they did. They are characteristically, to some degree at least, cooperative efforts; and each participant recognizes in them, to some extent, a common purpose or set of purposes, or at least a mutually accepted direction. This purpose or direction may be fixed from the start..., or it may evolve during the exchange; it may be fairly definite, or it may be so indefinite as to leave very considerable latitude to the participants.... But at each stage, some possible conversational moves would be excluded as conversationally unsuitable. We might then formulate a rough general principle which participants will be expected (ceteris paribus) to observe, namely: Make your conversational contribution such as is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange in which you are engaged. (p. 26)

Grice then describes (on pp. 26-7) four "categories" that are, in effect, special cases of the Cooperative Principle, with each category containing various "maxims". These tend to apply most obviously to conversations in which information is being exchanged, though some of them apply to other sorts of conversations, as well. It is important to see that the maxims are not meant to exhaust the possible bases for implicatures. That is, Grice is open to the idea that there are implicatures that do not involve any of these maxims.

NOTE: People nowadays tend to speak of, e.g., the "maxim of Quality". Grice, by contrast, tends to speak of various 'maxims' that fall under the category of Quality.

Grice spends a bit of time on pp. 29-30 exploring the question what might be the basis of the Cooperative Principle: Is it merely an empirical fact about how we happen to talk? Or is it somehow inherent in the nature of conversation? Can you sketch an argument for one of the maxims of the form: If one wants to exchange information in an efficient way, then one should obey that maxim (i.e., that it is rational to do so, and would be irrational not to do so)? It would also be worth checking your understanding by giving your own examples of implicatures that "flout" each of the four (categories of) maxims.

There is, in a way, an even deeper underlying principle, namely, that speaking is a form of acting, so it is something that is (typically) done for reasons.("[O]ne of my avowed aims is to see talking as a special case or variety of purposive, indeed rational, behavior..." (p.28).) Grice's idea is that people's reasons for saying what they do can sometimes be revealing of other of their attitudes and that, when this is so, it can lead to their meaning things they do not (strictly and literally) say.

Implicatures are supposed to be generated when speakers "flout" conversational maxims: that is, when they appear not to obey a maxim, but do so openly, but without simply refusing to cooperate. The question then arises: "How can his saying what he did say be reconciled with the supposition that he is observing the overall Cooperative Principle" (p. 30)? Answering this question will reveal the implicature.

Grice will proceed to give a number of examples of this phenomenon. First, though, he gives a general definition:

A man who, by...saying...that p has implicated that q, may be said to have conversationally implicated that q, provided that (1) he is to be presumed to be observing the conversational maxims, or at least the Cooperative Principle; (2) the supposition that he is aware that, or thinks that, q is required in order to make his saying or making as if to say p (or doing so in those terms) consistent with this presumption; and (3) the speaker thinks (and would expect the hearer to think that the speaker thinks) that it is within the competence of the hearer to work out, or grasp intuitively, that the supposition mentioned in (2) is required. (pp. 30-1)

The key point here is that "a conversational implicature must be capable of being worked out" by the audience (p. 31, emphasis added), where that amounts to formulating a certain kind of argument, which Grice goes on to illustrate. In working out the implicature, the audience will draw upon what they know about "the conventional meaning of the words used", which means that literal meaning has a kind of primacy.

It's notable that Grice's definition already assumes that q is "implicated" and only purports to say when it has been conversationally implicated. Why might that be? Is this condition actually required?

Grice's account of implicature refers to what the speaker believes, but what is implicated seems to be the proposition itself. E.g., in the example at the top of p. 32, what's required to make the statement accord with the Cooperative Principle is that the speaker believe the gas station is open, but what's implicated is just that the gas station is open. By contrast, in the example at the bottom of that page, the implicature is just that the speaker doesn't know exactly where C lives. Is there a problem here? What exactly is implicated in the example involving the letter of recommendation? that the writer believes that X is no good at philosophy, or that X is no good at philosophy?

Toward the end of the paper, Grice introduces the notion of generalized conversational implicatures. These are similar to conventional implicatures, because they are supposed to be, in a way, the normal case. The examples Grice gives here are not terribly convincing, it seems to me. A better one involves the word "most". (This is a variety of what are known as 'scalar' implicatures.) If someone says, "Most of the students passed the final", then one would normally suppose that not all of the students passed. This is a conversational implicature. (Which maxim is involved?) One of the signs that it is an implicature is that it can be explicitly "canceled" (p. 39): It would be fine to follow up by saying, "In fact, all of them did". If so, "most" cannot mean most but not all. Otherwise, the follow-up would contradict the original statement.

Consider the example mentioned above, involving "most": Is the 'generalized conversational implicature' that not all something that is always, normally, or typically meant_NN by the speaker? What might be the significance of this point for Grice's account?

As said, Grice does not formulate his reply to Strawson here, but it is clear what its outlines will be: What Strawson has noticed are certain implicatures. Can you fill in some of the details? How might Grice use the machinery developed here to argue, say, that "or" in English really is just truth-functional disjunction?

Conversational implicatures, Grice suggests, have a number of typical features (pp. 39-40). Conversational implicatures are:

1. Cancelable: One can say, e.g., "There is a gas station around the corner. But I'm not sure if it's open." Note the difference with conventional implicatures, which are not cancelable: "Joan is a woman, but she's good at math. But I don't mean to suggest that women aren't generally good at math!" Sorry, you just did suggest that.
2. Non-detachable: Since implicatures depend upon what has been said, they will usually be preserved by changes of wording. For example, the exact words used in the letter of recommendation do not matter. There are, though, special cases where the implicature does depend upon the specific words used. (Generally, these are manner implicatures.)
3. Not part of the literal meanings of the associated words but, rather, depend upon the literal meaning.
4. Not relevant to the truth or falsity of what is strictly and literally 'said'.
5. Often somewhat non-specific (despite Grice's often speaking as if something quite specific is implicated).

It's a reasonable thought that conversational implicature is a type of non-natural meaning, in Grice's own sense. Can we argue for this claim? I.e., can we show that what someone conversationally implicates, according to the definition Grice gives on pp. 30-31, is also meant_NN by them? One interesting question here is what guarantees that the overtness condition for meaning will be satisfied.

Our main goal in reading this material from Aspects is to get clear about some basic distinctions that structure linguistic theory, even today.

The first of these is the distinction between competence and performance. The basic idea is that performance, "the actual use of language in concrete situations" (p. 4), is the result of a speaker's applying some general ability that they have: their competence. They might make mistakes of various kinds, or for some other reason be unable to make use of the ability they have. Performance is not a direct reflection of competence.

Chomsky offers a number of examples to illustrate this point. I'll mention some others, for the sake of variety. Consider:

(1) The horse raced past the barn fell.

This might seem like gibberish, but careful consideration shows that (1) is in fact a grammatical sentence of English with a perfectly good meaning. It means the same as:

(1') The horse which was raced past the barn fell.

(1) is a so-called 'garden path' sentence. What makes it hard to parse is that you tend to hear "raced" as the main verb of the sentence, which then seems to end with "barn", so what is "fell" doing? Or consider:

(2) No eye injury is too trivial to ignore.

This sentence was once found on a sign in a hospital, and most people, on first encountering it, think it means: No eye injury is so trivial that you should ignore it. But careful consideration shows that it means the opposite: No eye injury is so trivial that you shouldn't ignore it; i.e., no matter how trivial an eye injury may be, you should ignore it. It's (in part) because that makes little sense that we tend to hear the sentence differently.

Other examples, which Chomsky discusses later, concern possible ambiguities. To modify one of his examples (in a way I owe to Jim Higginbotham), consider:

(3) I almost had my wallet stolen.

As Chomsky notes, there are (at least) three things this sentence can mean, though one of them is almost impossible to hear, because it makes no sense. (It is, as it is sometimes said, 'pragmatically excluded'.)

What are the three possible meanings of (3)? Which one is so hard to hear, and why? Can you explain in detail what point Chomsky trying to make with this example?

Linguistics, Chomsky insists, is interested in this underlying ability, that is, in competence. He also insists that this ability consists in certain knowledge (or, maybe better, information) that speakers have about their language. The question then becomes: What knowledge do speakers possess about their language that allows them to speak it? In thinking of competence in this way, Chomsky was at the forefront of the 'cognitive revolution' in psychology (though he was by no means the only one). In particular, Chomsky is here rejecting the behavioristic psychology that was still popular, indeed dominant, in the 1960s. (For detailed criticism of behavioristic accounts of language, see Chomsky's review of Skinner's Verbal Behavior, listed as optional. For an even earlier anti-behavioristic work, one of the very first papers in what we now call 'cognitive science', see the optional paper by Tolman.)

Chomsky also insists that the grammar should be 'fully explicit'. What he means is clearest from the examples he gives: traditional grammars tended to focus on exceptions (e.g., 'irregular' constructions) and to say little about the rules that govern the normal cases. By contrast, Chomsky is insisting that the rules that govern the normal cases are precisely what should be of interest to linguistics.

Even more importantly, Chomsky notes that the range of possible sentences a competent speaker might understand (or reject as ungrammatical) is unbounded. Speakers' knowledge cannot, therefore, be 'list-like': We don't just have a list of grammatical sentences, or know one-by-one what the meanings of sentences are. So one has to state some general principles that speakers are using to interpret utterances, and that they can use to interpret sentences they have never previously encountered. Chomsky took as inspiration work in logic and the theory of computation: What he is seeking is something like a set of rules for 'generating' the set of grammatical sentences, or for assigning a meaning (or range of possible meanings) to a sentence.

On p. 8, Chomsky writes: "Obviously, every speaker of a language has mastered and internalized a generative grammar that expresses his knowledge of his language". Is that so obvious? Why does it seem so to Chomsky? What sorts of assumptions might he be making? What does he mean when he adds: "Any interesting generative grammar will be dealing... with mental processes that are far beyond the level of actual or even potential consciousness"?

You can skim or skip §2. This is mostly a critique of approaches that attempted to account for linguistic performance without reference to an underlying theory of the structure of linguistic competence. Behavioristic psychology is Chomsky's main target here. The notion of 'acceptability' is what behavioristic theories of language used in place of grammaticality.

You can also skim or skip §3. Here, Chomsky outlines the overall structure of a linguistic theory, dividing it into syntactic (concerned with what expressions are grammatical), semantic (concerned with meaning), and phonological (concerned with sound) components. Chomsky also distinguishes "deep structure" from "surface structure", but this will not be particularly important to us. (Although something like this distinction is still made, it is drawn in a very different way.)

In §4, Chomsky addresses some methodological questions about how the study of competence should proceed. The basic idea is fairly simple: The data for the theory are actual linguistic performances (including 'intuitions' of native speakers about whether a given sentence is grammatical and, if so, what it can or must mean). The goal is then to formulate a set of rules that will explain this data—that is, to specify the 'tacit' (roughly: unconscious) knowledge that gives rise to the observed performance.

Chomsky's repeated dismissal of 'objective tests' and 'operational procedures' is again bound up with his rejection of behaviorism.

This brings us to the second crucial distinction that Chomsky draws in this chapter: between descriptive adequacy and explanatory adequacy. In the case of syntax, a grammar is descriptively adequate if it correctly categorizes each sentence as grammatical or ungrammatical. But explanatory adequacy demands more: that the theory be one that a child might 'acquire' on the basis of the limited exposure to 'primary linguistic data' that they actually enjoy.

The term 'explanatory adequacy' suggests, rather, that what we are looking for is a theory that does not just correctly categorize sentences, but also explains why those sentences are grammatical, or not. Or perhaps: a theory that explains why speakers so characterize them. How might this be related to the way Chomsky explains 'explanatory adequacy'?

In §5, Chomsky introduces a third important idea (though one that will not be our central focus): that the child brings with it, to the setting of language acquisition, innate knowledge of certain 'linguistic universals' (that is, principles that apply to all possible human first languages). This is what is known nowadays as universal grammar, or UG.

In §6, Chomsky sketches a general picture of what an explanatorily adequate linguistic theory might look like. This amounts to sketching a picture of what the process of language acquisition might itself look like, and how a child's innate linguistic knowledge might be put to us in developing a grammar for the language to which they are being exposed.

Part of what Chomsky is elaborating here is what have been called 'poverty of the stimulus' arguments. The thought is that the actual linguistic data to which actual children are exposed are extremely limited: too limited to determine any set of principles that would suffice to characterize arbitrary strings as grammatical or not, or to assign the grammatical ones determinate meanings. If so, then the child must bring something to the learning environment which, when added to the data, does suffice to select a particular 'grammar' for its language. The way Chomsky imagines this happening, in Aspects, is that the set of possible grammars is itself severely restricted: If the problem were to settle upon an arbitrary set of principles that accounted for the 'primary linguistic data', the problem would be unsolvable; but if the problem is to select one of a limited range of possible grammars on the basis of that same data, then the problem becomes much easier. Indeed, the smaller the range of possible gramamars, the easier the problem.

The papers we will be reading next all concern, though they do not speak in quite these terms, what a generative semantic theory might look like.

What is distinctive of language, as opposed to other forms of communication, is that words have "conventional meaning" as Grice put it, and that this meaning is partly determinative of one core part of the meaning of an utterance: what is said. But note that we think of both individual words and sentences as having meaning: Somehow, it seems, the meanings of the words in the sentence "Snow is white" 'combine' to give one the meaning of the sentence. The central question of the next couple papers we will read is: What does that really mean? What is sentence meaning? What is word meaning? And how do the meanings of words "combine"?

In this paper, Davidson is concerned to motivate what has come to be called the 'principle of compositionality', which asserts that the meaning of a sentence is a function of the meaning of the words of which it is composed, and of how that sentence is composed from those words. Additionally, he argues that compositionality imposes genuine constraints on theories of meaning, i.e., that some theories of how some constructions work can be ruled out on the ground that they conflict with compositionality.

Davidson begins by considering a dispute between Strawson and Quine over whether there could be a natural language that contained no singular terms (roughly, names of objects). You need not worry too much about the details of this discussion. Davidson's main point is just that claims about what is required for a language to be learned are empirical and cannot be established a priori.

Note the need for care in reading here. Davidson outlines a 'theory' of language learning on pp. 3-4, and you might well get the impression, from the way he presents it, that he means to endorse it. But then he says that it "is now discredited in most details".

At the beginning of section II, Davidson outlines, very quickly, a set of considerations that are supposed to motivate compositionality. He first introduces an analogy with syntax: It must be possible, he says, for us to "define a predicate of expressions...that picks out the class of meaningful expressions" (p. 8), that is, roughly, the grammatical ones. He does not explain why, but the thought, presumably, is that this is something that ordinary speakers know (e.g., which expressions are grammatical sentences of their language), and so that there must be some answer to the question how they are able to tell whether an expression is a (grammatical) sentence; if so, then (in principle) we should be able to tell some story about what features of an expression make it a sentence (or not).

Similarly, then, for semantics (the study of meaning): Ordinary speakers are able to determine, even of many sentences they have never previously encountered, what those sentences mean; there must be some way they do this; there must be some story to be told about how; the only reasonable hypothesis seems to be that they know what the words comprising some "novel sentence" mean, and they are able to put those word-meanings together to determine the meaning of the sentence.

Do not worry too much yet about the reference to Tarski's theory of truth. Davidson will have more to say about that in the next paper we read.

Consider now a language that (i) contained, as natural languages do, infinitely many sentences and (ii) did not satisfy the principle of compositionality, so that there was no story to be told about how one could generate, from some finite basis (of word-meanings), the meanings of all the sentences of that language. Then, Davidson claims, the language would be unlearnable in principle. What is his argument for this claim? The argument is on pp. 8-9.

Davidson goes on to argue that English, e.g., must contain finitely many "semantic primitives" from which the meanings of all other expressions can be generated. To what extent does the argument depend upon there really being infinitely many sentences of English? Are there really infinitely many sentences of English? In what sense? Is each of us actually capable of understanding infinitely many sentences? In what sense?

Davidson proceeds to give four examples of theses philosophers have held about language that would imply that English contains infinitely many semantic primitives. We will work through Davidson's discussion of quotation in class, in some detail. The key claim here is that a quotation name (such as: "snow") cannot be thought of as having its meaning determined as some function of the meaning of the word contained within the quotation marks. (This is what Davidson means when he says that quotation marks are not a functional expression.) The contained word is not used at all, and its meaning is irrelevant. This is particularly clear in the case of quotation names of non-words, e.g., in the sentence: The expression "snaw" is not a word of English.

Davdison alludes on p. 9 to paradoxes involving quotation. But it is not so much quotation as semantic notions like denotation and truth that threaten to give rise to paradox. For an example of one involving denotation, consider Richard's paradox. Only finitely many numbers can be uniquely picked out by expressions of less than twenty-two syllables. This is because there are finitely many such expressions. So some number is not so named and, by the least number principle, there is a least such number. But then "The least number not named by any expression of less than twenty-two syllables" picks out that number, and does so in twenty-one syllables! Contradiction.

Here are some questions to think about ahead of class: Why might all of that have led Church to say that quotation is "misleading"? What is misleading about it? Why does Davidson insist that the claim that a quotation names its interior (i.e., what is within the quotes) does "not provide even the kernel of a theory" (p. 9)? Quine's proposal, which Davidson mentions but does not explain, was to replace quotation by a form of spelling out: Thus, instead of "'snow'" we would have: ess + en + oh + double-u. How would that help solve the problem? Does it solve the problem?

Exactly why can we not think of quotation marks as a kind of functional expression? If we write "q(·)" for quotation of, then what is wrong with: q(snow)? Note that this is meant to be analagous to thinking of, say, "the color of" as a function: so color-of(snow) = white, color-of(grass) = green, etc. Why not something similar for quotation?

Davidson considers three more examples which we may not have time to discuss in class. You can skip the last one. But you should read the second and third examples, and think them through, as they will be mentioned in later readings.

In "Theories of Meaning and Learnable Languages", Davidson argued that, if a language is learnable, we should be able to give a compositional theory of meaning for it: a theory that shows how the meaning of a sentence depends upon, and is determined by, the meanings of its parts. In this paper, Davidson argues that a theory of truth can play this role and, moreover, that nothing else can. This amounts (or is intended to amount) to an argument that the meaning of a sentence is the condition under which it is true, i.e., for what is known as a truth-conditional theory of meaning. This sort of view is well represented in the history of analytic philosophy, having been held by Gottlob Frege, Ludwig Wittgenstein (in the Tractatus), Rudolf Carnap, and many others.

The central question of the paper is thus: What is required to give a genuine explanation of how sentence-meaning is determined by word-meaning? The paper largely proceeds by dismissing various proposals until only one is left.

One desideratum on a proper solution is to avoid a certain regress that Davidson mentions in the second paragraph. If you just take each word to denote some entity as its meaning, then you have the problem how to combine them into the meaning of the sentence. What you cannot do is identify some further thing (instantiation, e.g.) as what does that. For then you just have yet another thing that needs to be combined with the rest. (How do the meaning of "Theaetetus", the meaning of "files", and instantiation combine to give the meaning of "Theaetetus flies"?)

One of the papers by May and me listed as optional, "The Function Is Unsaturated", attempts to unpack and defend the view of Frege's that Davidson derides.

Davidson then considers the problem of explaining the meaning of all terms of the form "the father of...the father of Annette". He notes that the following two axioms will allow us to do so:

• "Annette" refers to Annette
• "the father of t" refers to the father of whoever the term t refers to

From these two principles, one can prove, e.g., that "The father of the father of Annette" refers to the father of the father of Annette. And, Davidson says, we have done that without assigning any "entity" as meaning to "the father of". Implicit here is the idea that an "explanation" of sentence-meaning in terms of word-meaning might amount to a theory with axioms specifying what the words mean, which will then allow us to derive theorems saying what the sentences mean.

The discussion in the full paragraph on p. 305 seems to me confused. Don't the axioms 'give the meanings' of the atomic parts? Probably what Davidson has in mind is that these axioms only state the reference of "Annette" and 'what contribution' the phrase "the father of" makes to determining the reference of larger expressions. But it seems odd to treat such a phrase as syncategorematic.

Davidson then argues that, although we can try to extend this treatment to sentences, that will not work. His argument for this claim is known as the Slingshot, and it is a large topic in its own right. Suffice it to say that Frege's own approach takes sentences to refer to their truth-values. (The Slingshot—which is not an argument Frege himself gives—purports to show that this choice is forced: that if sentences refer to anything, it must be to their truth-values.) But then it turns out that all true sentences refer to the same thing, as do all false ones, which seems a bit odd if we're thinking of the truth-value as the meaning of the sentence.

The notation x̂(...x...) means: the class (or set) of x such that ...x.... See the paper by May and me, "Truth In Frege", for discussion of Frege's views about truth.

As Davidson notes, however, Frege himself insisted (in part for this kind of reason) that we distinguish meaning from reference. (What Davidson is calling "meaning" here, Frege called "sense".) E.g., Frege thought that the names "Mark Twain" and "Samuel Clemens", though they refer to the same person, have different meanings (i.e., senses), because "Mark Twain is Mark Twain" seems to be a logical truth, whereas "Mark Twain is Samuel Clemens" is not (but looks to be something one could discover). So the next set of proposals attempts to work with the notion of meaning rather than that of reference. This is where the real work of the paper begins.

Davidson considers this approach on pp. 306-7. The first problem is that we have been given no substantive account of what the meaning of "flies" is or how it "yields" the meaning of "Theaetetus flies" when given the meaning of "Theaetetus" as argument: The meaning of "flies", on this account, just is whatever function maps the meaning of "Theaetetus" to the meaning of "Theaetetus flies", and so forth. It is also explanatorily useless: One can't specify the function without already knowing what all sentences containing "flies" mean.

The second problem, which is the really serious one, and which is shared with the suggestion Davidson considers on pp. 307-8, is that such a theory makes no real use of the structure of a sentence, that is, of how its words are combined into a sentence.

The paper by May and me, "The Composition of Thoughts", attempts to explain how Frege thought this was supposed to work. The paper of mine, "Is Compositionality a Trivial Principle?" develops these arguments of Davidson's in detail.

We'll discuss this point in class, but one important instance with which you should already be familiar involves quantification: In effect, what Davidson is pointing out is that, while quantifiers appear in essentially the same places that names do, they work semantically in a completely different way: To understand how the meaning of "Someone smokes" is determined by the meanings of "Someone" and "smokes", we need to understand this difference. Can you develop that point a bit?

Davidson traces the underlying problem to a lack of clarity about what 'meaning' is and a corresponding lack of clarity about what a 'theory of meaning' is supposed to do: We want the theory to show us how the meaning of a sentence is determined by the meanings of its parts, but we do not even seem to know what kind of thing the meaning of a word or sentence is! (I.e., we do not know how properly to specify these things in our theory.) Davidson concludes that the shift from reference to meaning has not accomplished what we had hoped it would.

On p. 309, Davidson introduces his positive proposal. It seemed, Davidson suggests, that what we needed was a theory that would somehow deliver, from axioms that specified the meanings of the words, theorems of the form:

• S means that p

where "S" is replaced by a name of a sentence (that is close enough, for our purposes, to what Davidson means by a 'structural description') and "p" is replaced by that very sentence (or a translation of it), e.g.:

• "Alex runs" means that Alex runs
• "Drew smokes" means that Drew smokes

The problem is that it is very unclear how to do this. Davidson's remarks about the non-extensionality of "means that" amount to an observation that we do not even know what the logic of such a theory might be like.

What Davidson speaks of the 'meta-language', he means the language in which we are stating our theory of meaning. By contrast, the 'object language' is the language for which we are giving the theory.

But, Davidson suggests, maybe "the success of our venture depends not on the filling but on what it fills" (p. 309), that is, not on the words "means that" but on the relation between "S" and "p" that the theory secures. So we might instead take the target to be theorems of the form:

• S is true if and only if p

where, again, "S" is replaced by a name of a sentence and "p" is replaced by that very sentence. (This is Tarski's "Convention T" which says, more or less, that such sentences will always be true, and that any predicate for which they are all true will be co-extensive with "true" itself.) What we want, that is to say, is for our theory to deliver such theorems as:

• "Alex runs" is true iff Alex runs
• "Drew smokes" is true iff Drew smokes

on the basis, again, of axioms specifying the meanings of the words. Note that, on this approach, saying what a sentence means amounts to saying what is required if it is to be true, i.e.: the meaning of a sentence is the condition for it to be true, its 'truth-condition'.

The great advantage to this proposal, for Davidson, is that we have some idea from logic how to do this, at least for a reasonable range of sentences, thus:

• "Alex" refers to Alex
• The extension of "runs" is the set of things that run
• A sentence of the form "name one-place-predicate" is true iff the thing to which the name refers is in the extension of the predicate

Those three principles imply that "Alex runs" is true iff Alex runs. And we can do similar things for relational predicates, like "loves", and even for quantifiers (although the formal details get messy in that case).

Question: Does Davidson really have an argument for this proposal? If so, what is it?

Davidson suggests, on p. 311, that there is no trouble telling when we have a correct theory. At least in the case where the meta-language contains the object language (i.e., we are giving a theory of meaning for English in English, "anyone can tell whether it is right". Is this consistent with his claim, earlier in the same paragraph, that the theory is "empirical"?

On pp. 311-2, Davidson considers an important objection, one that hearkens back to the Slingshot. Both the following sentences are true:

• "Snow is white" is true iff snow is white
• "Snow is white" is true iff grass is green

(That is because "iff" here is the material conditional. If it were not, we'd be back in the intensional soup.) For all that has been said so far, then, a theory that yielded the latter would be as good as one that yielded the former. But surely that has to be wrong? Surely the latter gets the meaning of "Snow is white" wrong?

What is Davidson's response? How good is it? (We will consider this issue in more detail in a week or so.)

The end of the paper is a catalog of open problems that such an approach to semantics suggests. Why, in particular, does "Bardot is a good actress" pose a problem? How is that similar to or different from the problem posed by "Katelyn is a tall gymnast"? (Hint: She might be short for a 21-year old woman.)

Strawson is concerned in this paper to mediate a dispute between two approaches to questions about meaning. As you will see, he is not exactly a neutral party.

The central question, as Strawson sees it, concerns meta-semantics: what it is (that's a metaphysical question) for sentences to have meaning (that is, to have semantic properties). More precisely: "What is it for a particular sentence to have the meaning or meanings it does?" As Strawson sees it, there are two broad approaches to the problem, represented here by Grice and Davidson.

• The 'theorists of communication intention' hold that a sentence's meaning what it does consists, somehow, in speakers' having certain communicative intentions: their using that sentence with the intention to communicate some particular proposition. Of course, that will not always be true (due to implicature, irony, and the like), but the hope is that those issues can be finessed somehow. Strawson sketches some of the ideas about how to implement this approach on pp. 173-6, but you need not worry too much about them. The crucial issue does not lie there (at least according to Strawson).
• The 'theorists of formal semantics', on the other hand, hold that there are 'rules' that determine the meaning of sentences but that these rules are not, in any sense, rules for communicating. Rather, the 'rules' fix truth-conditions. As Strawson sees it, Davidson's claim is that someone who knows the truth-condition of a sentence knows what it means, without necessarily having any idea how that sentence might be used to communicate with someone.

Most of the beginning of the paper is spent outlining these two views.

On p. 179, Strawson arrives at what he takes to be the crucial question. Strawson is prepared to agree that the meaning of a (declarative) sentence can be identified with its truth-condition. But we "cannot be satisfied that we have an adequate general understanding of the notion of meaning unless we are satisfied that we have an adequate general understanding of the notion of truth" (p. 180). So the question is: How is the notion of truth-condition to be explained, without any appeal to the notion of a communicative intention? Or, more precisely, how is the notion of truth itself to be explained?

In the middle of p. 180, Strawson considers the response that truth itself is explained by Convention T, or by a definition of truth that satisfies it. He alleges that this is circular. Why?

One might worry that Strawson's formulation of the "crucial question" is imprecise. What is really at issue here is not so much the notion of truth-condition but what it is for a given utterance to have a certain truth-condition. To answer that question, we do need to know, among other things, what it is for an utterance to be true. Can Strawson just accept this clarification? Or does it affect his argument?

What Strawson initially suggests is that truth should be regarded, primarily, as a property of 'sayings', characterized roughly as follows: "One who makes a statement or assertion makes a true statement if and only if things are as, in making that statement, he states them to be" (p. 180). If so, Strawson continues, then the very notion of truth (as applied to sayings) is bound up with the notion of what is stated by someone who makes a given utterance. And that, one might think, can only be explained in terms of the general notion of 'making a statement', that is, in terms of that sort of 'speech act', which is, in the normal case, performed with certain communicative intentions. Indeed, Strawson suggests, what it is to 'make a statement' is, at least in the normal case, to speak with such intentions.

You will note that Strawson speaks here as if the communicative intention characteristic of assertion is that one's audience should come to believe not, say, that there are pigs in the garden, but that the speaker believes themselves that there are pigs in the garden. This is because of counterexamples to the former view that led to the latter view. Is the view he's articulating more or less plausible depending upon which of these two views one holds?

On this conception, then, the 'rules' that determine the meaning of a sentence determine what that sentence can be used to say, in Grice's sense, and so what that sentence might (in otherwise unexceptional circumstances) be used to mean or communicate (see p. 181).

The discussion on pp. 182-3 pushes the issue from a slightly different direction, noting that merely 'associating' the sentence "Snow is white" (say) with the 'state of affairs' of snow's being white cannot suffice.

The discussion recalls, for me, a question asked by Sir Michael Dummett. Suppose that we take Davidson's theory but substitute some nonsense word for "true", say, "alby". So we have such theorems as: "Snow is white" is alby iff snow is white. Clearly, this does not carry any information about the meaning of "Snow is white". But why not? What was it about the use of the word "true" that was so important? Why should truth-conditions seem to have any essential connection with meaning? Does Davidson have an answer to this question?

Seeing no possibility of progress in that direction, Strawson offers the formal semanticists another option: to connect the notion of 'what is stated' not with the notion of belief-communication but rather with the notion of belief-expression, where expressing a belief need not involve any intention to communicate it to someone else. So, on this account, "...the meaning-determining rules for a sentence of the language are the rules which determine what belief is conventionally articulated by one who...utters the sentence (p. 184). (Chomsky, to whom Strawson also refers, has certainly made this kind of suggestion.) As Strawson notes, this is very reminiscent of his own view: If one looks again at p. 181, one can see Strawson himself suggesting that the truth-condition of a sentence determines what belief someone who makes a literal utterance of that sentence would thereby be expressing (that is, what the 'content' of that belief is).

The main objection Strawson pushes here (on pp. 185-6) is that it less clear than it might seem what 'expressing a belief' really is. Dialectically, the key point is that this cannot—from the point of view of the formal semanticists—be something one does with the intention of getting other people to think one has that belief. That would be a communicative intention, which is what we're supposed to be trying to avoid. But if we don't think of it that way, then what is it?

Strawson goes on to suggest that, even waiving that point, such a view would make it a mystery why people who live in the same area tend to speak the same language, rather than all having their own way of expressing their beliefs. The thought, I take it, is that, if it's just incidental whether one's expressions of belief are understood as such by other people, then why can't one just have one's own special way of expressing one's beliefs? But Strawson does not really have much to say about this matter, other than that it seems to him to lead to a view that is "too perverse and arbitrary to satisfy the requirements of an acceptable theory" (p. 188). But this issue, though interesting in its own right, is very complex and it's a reasonable guess that the theorists of formal semantics will have wanted to get off this train long before it reached this station.

Here's another way one might try to raise the question in which Strawson is interested. Davidson's idea seems to be that someone who knows the truth-conditions of all the sentences of some language will therefore be able to speak it. But why exactly? How is one to put this 'knowledge of truth-conditions' to work? What precisely does knowledge of truth-conditions enable one to do, and how?

What difference might it make to this discussion if we make room for implicature? It's not, in general, true that someone who utters a sentence wants their audience to believe its content. The sentence might be obviously false, and the fact that it's obviously false is being used to allow the speaker to implicate something else. Will one of the two views have an easier time explaining this phenomenon? Or is it somehow unfair even to raise this issue?

Lewis here offers a different perspective on the sorts of questions that exercise Strawson in "Meaning and Truth". Lewis himself advertises it as a 'synthesis' of the two sorts of positions that Strawson considers (though he does not mention Strawson, and this paper actually pre-dates Strawson's, though it was only published later). In particular, Lewis argues that (what he calls) languages, which are the concern of "formal semanticists", are central to an account of (what he calls) language, which is the social phenomenon of concern to philosophers like Grice. In particular, Lewis wants to explain what it is for a sentence to have a certain meaning in a given speech community by analyzing the relation: L is the language used by some population P, where that relation is itself explained in terms of a particular convention that prevails in P. In the background is a general account of what a convention is.

Lewis suggests that we think of a language, as far as meaning is concerned, as a function from sentences to 'meanings', which he takes to be truth-conditions. Following many then, and inspiring many now, Lewis suggests that we may think of truth-conditions as sets of "possible worlds" (situations, circumstances, ways that the world might be). So, e.g., the truth-condition of "Dogs meow" is the set of possible worlds in which dogs meow. Fortunately, this aspect of Lewis's view will not matter much for our purposes.

Lewis has much to say elsewhere about exactly what 'possible worlds' are supposed to be. But this metaphysical question will not be our concern. If you're curious about these issues, see the SEP article on possible worlds.

The notion of language as a social phenomenon is harder to characterize, but Lewis characterizes it in terms that should, at this point, be recognizably Gricean. What he emphasizes (even more than Grice does) is the conventional character of the association between "sounds and marks" and certain kinds of communicative intentions.

Lewis then offers a philosophical analysis of the general notion of a convention. Whether it is correct is a controversial issue in its own right, quite independent of questions about language. So we'll not pursue that issue and, for now, let Lewis have this analysis.

Test your understanding of Lewis's account of convention by explaining why, according to it, it really is a convention (in the US) to drive on the right.

On pp. 7-8 (166-7), Lewis turns to the question what connects languages (as formal objects) with language (the social phenomenon). His answer is: certain kinds of conventions. As usual, Lewis focuses on informative uses of language (assertion and the like). The trick is to say what the convention is. To focus on a single sentence (instead of a whole language), Lewis's proposal is that S means that q in the population P iff there prevails amongst the members of P a convention (i) to utter S only if one believes that q (the convention of truthfulness) and (ii) to "tend to respond" to someone else's utterance of S by coming to believe that q (the convention of trust).

Quick question: Why not "iff" in (i)?

Less quick question: Might it be better to modify (ii) so that people tend to believe that the speaker believes that p? Why or why not? (If you look closely at the two papers from Grice that we read, you will see that his view changes between them in exactly this way.)

Paper- or book-length question: How exactly is Lewis's view related to the Grice-Strawson view? (Lewis does address this question, obliquely, later in the paper.)

On pp. 8-9 (167-8), Lewis argues that, if some language L is used by a population P, then there always will be convention of truthfulness and trust in L that prevail in P. The argument consists in checking that the six conditions that characterize conventions are all satisfied.

It would be worth working through these for the case of a single sentence. (In doing so, it is fine to ignore or bracket context-dependence, e.g., tense. Lewis discusses how his view accommodates context-dependence on pp. 13-5 (171-2).) Must all these conditions be satisfied? Are there really conventions of truthfulness and trust?

On pp. 17-23 (175-9), Lewis considers a series of objections based upon the fact that languages, as he describes them, only pair sentences with meanings and do not involve any sort of compositionality. The first of these is clearly a version of Davidson's main idea in "Theories of Meaning and Learnable Languages". In response, Lewis articulates an account of what (perhaps following Chomsky) he calls grammars. These assign meanings to sub-sentential components of sentences (i.e., words, roughly) and explains how these combine to determine the meanings of complete sentences. The details are not crucial, so you can skim pp. 18-19 (175-6), until the objection is restated (at the top of p. 20 (bottom of 176).

The new version of the objection is: We need to be able to say that grammars are used by populations, not just languages, i.e., that words have certain meanings for speakers, not just sentences. In response, Lewis says that he simply does not know how to make sense of that idea or, more precisely, that he does not know what a convention to use a certain grammar would be like. Of course, one might reply that that only reveals a limitation of the notion of convention. But what other option might there be?

This kind of worry was originally raised by Quine (who was Lewis's PhD supervisor). The rough idea is that only sentences ever really get used by speakers, so everything about meaning must ultimately cash out in terms of sentences. But if two grammars generate the same language, then they make all the same predictions about sentences.

The second objection seems to echo Chomsky: If we want to explain how it is that speakers can have an 'infinite capacity' to understand novel sentences, then we need to think of them as 'knowing' grammars, i.e., knowing what their words mean and how those combine to determine sentence meanings.

Lewis's response is that this is an empirical hypothesis. As such, it can have no role in an analysis of what it is for a language to be used by a population. What does this say about how Lewis understands his project? How might Chomsky respond?

Oddly enough, Davidson ends up holding a similar view in "Reality Without Reference", but I have never understood how to reconcile that view with what he argues in "Theories of Meaning and Learnable Languages".

On pp. 22-3, Lewis considers the suggestion that the convention ought just to be to 'bestow' certain meanings on certain sentences. His objection is that this is not a convention at all, because conventions are regularities in action and belief, and 'bestowing meaning' is neither.

So consider the following view: S means that q in the population P iff everyone in P believes that S is true iff q (relativizing to worlds, if one so likes), and if the fact that everyone so believes is common knowledge in p. Why is this not a convention? Does it matter if it is convention? When Lewis says, on p. 7 (166), that "It is a platitude—something only a philosopher would dream of denying—that there are conventions of language...", is that meant to rule out this kind of view?

Foster's paper is important for a certain sort of objection that it brings against what Foster calls "Davidson's Initial Thesis". As we shall see, the objection threatens rather more than just that.

In the first section of the paper, Foster attempts to motivate a certain sort of approach to philosophical questions about meaning that puts "theories of meaning" (in the sense in which Davidson uses that term) at the center. To some extent, this just reiterates the way Davidson himself motivates the principle of compositionality in "Theories of Meaning and Learnable Languages". As Foster initially motivates the proposal, the goal is to give an articulate statement of what competent speakers know about their languages, specifically, what they know about the meanings of words, and the significance of various modes of combination, which allows them to understand sentences they have not previously encountered.

However, Foster worries that the explicit statement of these principles—think, perhaps, of the semantic principles governing words like "every" and "some"—are not ones many speakers would even be able to understand. So Foster proposes to recast the original project: What we seek is a theory explicit knowledge of which would suffice for competence. Foster argues that approach implicitly requires that the theory be 'scrutable'. He illustrates how this condition might be violated by a theory that, nonetheless, did assign the right meanings to the right sentences.

Would it be reasonable to interpret Foster as requiring that the meaning of a sentence should be effectively calculable from a 'structural description' of that sentence? Should we require something even stronger?

We shall soon turn ourselves to questions about the nature of semantic knowledge. But it is worth asking now what difference it might make whether we think of a semantic theory as, in some sense, known by competent speakers. Why might it be important to say that it is?

Foster then raises the question what philosophical significance theories of meaning might have. His answer is that "if we put the right constraints on what giving-the-meanings involves, then characterizing the general method by which such theories can be constructed and verified reveals what meaning-in-language really amounts to" (p. 4). That is: If we know (i) what a theory would need to tell us to tell us what "Snow is white" means and (ii) what would make such a theory correct (or, at least, 'confirmed'), then we will have the answers to the philosophically interesting questions about the nature of (linguistic) meaning.

The argument Foster gives on pp. 4-6 is a version of one that has been given by many philosophers, including Davidson and Lewis. It concerns the question whether a translation manual could serve as a theory of meaning. What is Foster's argument that it cannot? How good is that argument?

It is in the second section that Foster criticizes Davidson's Initial Thesis. That is the thesis that a theory of meaning may take the form of a theory of truth: that it is enough to have a theory that, for every sentence S, yields a theorem of the form:

S is true iff p

where p is a translation of S, and so can serve to 'give the meaning' of S. This is, of course, the view put forward in "Truth and Meaning".

One way to think about the issue here is to focus on the following inference, which I'll call the 'T-M inference':

1. S is true iff p
2. So, S means that p

In general, such an inference is invalid: Davidson already notes in "Truth and Meaning" that "Snow is white" is true iff grass is green, but we should not want to conclude that "Snow is white" means that grass is green.

Foster emphasizes and elaborates this point on pp. 10-1. Davidson remarks in "Truth and Meaning" that his theory "works by giving necessary and sufficient conditions for the truth of every sentence, and to give truth conditions is a way of giving the meaning of a sentence" (p. 310). Foster argues that Davidson's theory does no such thing. The truth-condition of a sentence, Foster says, is what would have to be true for the sentence to be true. But the "iff" occuring in T-sentences is the material biconditional, so

(SW) "Snow is white" is true iff snow is white

only tells us that "Snow is white" is true iff snow is white. It does not tell us that "Snow is white" would have been false had snow not been white. In fact, (SW) has nothing whatsoever to say about the truth-value of "Snow is white" in any circumstances other than the actual circumstances.

But one might think that if we add some additional premise, then the T-M inference would be valid. And now the question becomes: What should that additional premise be? Davidson's original suggestion, on pp. 311-2, was that it is enough if we expand our view from the T-sentences themselves to the theory as a whole: If (1) issues from a theory which states correct T-sentences for all sentences of the language, then the inference goes through. I.e., we have something like:

1. S is true iff p
2. (1) issues from a truth-theory that implies correct T-sentences for all sentences of the language.
3. So, S means that p

Foster's counter-example to this revised T-M inference (slightly simplified) is as follows. Waiving a lot of complications, let us imagine that the semantic clause for "is white" is:

(W1) "is white" is true of x iff x is white

Suppose this clause, which is of course true (at least in this simplified form), is part of a theory that issues in correct T-sentences for all sentences of the language. If we replace (W1) with some other clause, then that will change what T-sentences are implied by the theory. But so long as we replace (W1) with some other true clause, the implied T-sentences will still be true. And that, Foster notes, is easy to do:

(W2) "is white" is true of x iff (x is white and the earth moves)

This is still true, since the earth does indeed move. Since true axioms imply true theorems, the T-sentences we get if we replace (W1) with (W2) will still be true. And they will still be generated in a compositional way. But the T-sentence for "Snow is white" will now be:

(SW2) "Snow is white" is true iff snow is white and the earth moves

And so the revised T-M inference would lead to both of these conclusions:

• "Snow is white" means that snow is white
• "Snow is white" means that snow is white and the earth moves

And those two things cannot both be true.

We can pull a similar trick even for sentential connectives. What might a Foster-style clause for "and" look like? What is the lesson of that clause for theories of meaning? (Foster does discuss this, very briefly, on p. 16.)

Recall that, in "Truth and Meaning", Davidson observed that both of these "T-sentences" are correct:

1. "Snow is white" is true iff snow is white
2. "Snow is white" is true iff grass is green

But, whereas the former seems in some sense to 'give the meaning' of the sentence "Snow is white", the latter does not. So it cannot, in general, be true that a correct T-sentence is also meaning-giving. Nor, as Foster pointed out, is it enough simply to require that these T-sentences be compositionally generated, since

3. "Snow is white" is true iff snow is white and grass is green

can be compositionally derived. (Of course, we're ignoring all the actual complexities here.) As it happens, Davidson had discovered the "Foster problem" for himself, and "Radical Interpretation" is (among other things) his attempt to solve it. (He also addresses it in "Reply To Foster", which is listed as optional.)

At the beginning of the paper, Davidson elaborates an approach to questions about meaning. He takes the central issue to be: What could we know that would allow us to interpret (understand) the utterances of other people? How could we come to know that? As he emphasizes, these are hypothetical rather than empirical questions.

It's obvious why the empirical question how we do manage to understand 'novel' sentences is worth asking. What's the interest of Davidson's 'doubly hypothetical' question? Could the answer to it explain our ability to understand novel sentences? In what sense? Another important question, which we'll spend significant time discussing shortly, is: What notion of knowledge does Davidson have in mind here when he denies that ordinary speakers must know such things? What notion should one have in mind?

What does Davidson mean when he writes: "All understanding of the speech of another involves radical interpretation"? How does that relate to his claim that the question in which he's interested is 'doubly hypothetical'?

Davidson proceeds to eliminate a number of possible approaches and to elaborate other aspects of the problem, such as the need to account for the unbounded capacity to interpret novel utterances (as discussed in the other papers of Davidson's we have read). He emphasizes that we need not think of a theory of interpretation as defining a function from utterances (or, in the simplest case, sentences) to 'meanings' or 'interpretations', contra Lewis, but instead simply as enabling someone who knew the theory to 'interpret' the language in question, that is, to understand it (and, presumably, to speak it, though Davidson seems most focused on the case of language comprehension).

On p. 316 (128, in the reprint), Davidson claims that "a theory of interpretation" must be able to "be supported or verified by evidence plausibly available to an interpreter". But he gives no explicit reason for this claim. What might he have in mind? How does this relate to the 'doubly hypothetical' character of his investigation? (Davidson will come back to the other claims he makes here.)

On pp. 316-7 (128-30), Davidson considers Quine's suggestion that such a theory may take the form of a translation between the 'new' language and one's own. What is Davidson's objection to this approach? How good is it?

Unsurprisingly, given Davidson's earlier writings, the alternative he suggests is a theory of truth that issues in T-sentences like:

4. "Es regnet" is true at time t iff it is raining at t.

I.e., his claim is that, if you knew a theory of truth for (say) German, then you could use that theory to interpret (speak and understand) German.

On p. 319 (131), Davidson lists three questions about this approach. The first, whether a theory of truth can actually be given for a natural language, will not concern us (though it does much concern people who work on natural language semantics). The second and third questions are the crucial ones philosophically.

The second question is: Can a theory of truth be verified by evidence plausibly available to a radical interpreter? Note that the question here is simply whether we can decide, on the basis of such evidence, whether the T-sentences the theory implies are true—so we are not yet trying to distinguish (1) from (2). The question is especially pressing because Davidson has a very austere conception of what evidence is available to the radical interpreter. It amounts to facts about what sentences are 'held true' under what sorts of circumstances. So, roughly, if Kurt tends to 'hold true' the sentence "Es regnet" when it is raining, that is some evidence for (4).

Contra Grice and Strawson, Davidson denies the radical interpreter access to "finely discriminated" facts about beliefs and intentions. (He seems to have in mind any mental states more finely discriminated than that a given sentence is true: see p. 322 (135).) He argues for this restriction first on p. 315 (127), and then states a general restriction on the possible evidence on p. 316 (128): "[I]t must be evidence that can be stated without essential use of such linguistic concepts as meaning, interpretation, synonymy and the like". What is Davidson's argument for this restriction? How good is it?

How much of an argument does Davidson offer for an affirmative answer to his second question? It might help to imagine that we have to deal only with a very simple language, say one with just ten proper names, a few one-place predicates, a few two-place predicates, and (perhaps) the propositional connectives. (As we'll see shortly, Gareth Evans discusses this sort of simple language.)

The third question is: Could such a theory, if known to be justified by such evidence, be used to interpret the target language? It is here, on pp. 325-6 (138-9) that the Foster problem rears its head. It might seem plausible that a truth-theory will allow you to interpret German if the theory proves (4), but not so much if it proves

5. "Es regnet" is true at time t iff it is raining at t and there are everywhere continuous real-valued functions that are nowhere differentiable.

even though such a T-sentence is still true. Hence, not just any correct theory of truth will suffice for interpretation. A theory of truth that is 'adequate for interpretation' must thus meet some additional condition. The additional condition is supposed to be, precisely, that the theory is verifiable on the basis of the sort of evidence that is available to a radical interpreter. What Davidson is proposing is thus that the inference from T-sentences to meanings might be understood as follows:

1. "Schnee ist weiß" is true iff snow is white.
2. The T-sentence (i) is a theorem of a truth-theory that is verifiable on the basis of the sort of evidence that is available to a radical interpreter.
3. So, "Schnee ist weiß" means that snow is white.

Another way to think of this is that the inference from (i) to (iii) goes through so long as one's justification for (i) is of the kind described by (ii). I.e., we treat (ii) more as a background, 'enabling' condition for the inference than as an additional premise.

What Davidson means by a 'canonical proof' is one that derives the T-sentence in a particularly direct way. (We can define precisely what that means but will not try to do so here.) Some such restriction is necessary because any theory of truth will, in fact, prove lots of T-sentences for any given sentence. For suppose that the theory proves "S is true iff p". Then if q is any other statement the theory proves, it will also prove "S is true iff (p and q)". (Convince yourself of that.) So, really, we are interested only in the 'canon ically proven' T-sentences. Those are the ones that are supposed to be 'interpretive'. Still, Foster's objection shows that this cannot, by itself, resolve the problem posed by (4) and (5): If there is a correct theory of truth that canonically proves (4)—that is, one whose canonically proven T-sentences are all true—then there is also a correct theory that canonically proves (5).

Davidson's argument that the inference described is valid (or usually valid, or something in that vicinity) is contained in the final two paragraphs of the paper. What is the argument? One very important question, which Davidson does not really address, is: What is it to 'interpret' a sentence anyway?

Our main interest here is in Soames's re-articulation of the Foster problem, on pp. 19-25, and his criticism of Davidson's response to it, on pp. 25-9. In between, on p. 25 itself, are two paragraphs (and two very long footnotes) about James Higginbotham's view, which we shall read next. (The two papers were delivered as part of symposium.) Since we have not yet read Higginbotham's paper, you need not worry about that material now, but you should come back to it after you do read Higginbotham's paper.

Soames begins with a very nice formulation of the motivation for Davidson's approach: "[I]f a theory tells us everything we need to know in order to understand a language, then it must be counted as specifying all essential facts about meaning, even if it does not issue in theorems of the form 'S' means in L that p, which state the meanings of individual sentences one by one" (pp. 17-8).

Soames claims that there are basically two ways to interpret the claim that a theory of truth may "serve as" a theory of meaning.

1. Knowledge of what an interpretive (or, as Soames puts it, "translational") truth-theory states is sufficient for understanding.
2. Knowledge of what a truth-theory states is necessary for understanding.

The latter view comes in two forms, as well: that knowledge of everything the theory states is necessary for understanding, or that the theory states everything knowledge of which is necessary for understanding.

Soames argues on pp. 19-21 that all forms of the sufficiency view fall to a form of the Foster problem. The crucial argument is the one that surrounds (3) on p. 21. Can you summarize this argument in your own words? How might Davidson respond?

Note that Soames's example does not just threaten Davidson's view but also views that would strengthen T-sentences in the way suggested by Foster:

"Snow is white" is true in world w iff snow is white in w.

Or, alternatively, by requiring the "iff" to be not material but 'strict', so that the equivalence is necessary. The point is that we can make the same move Foster makes but make the extra conjunct itself be some necessary truth. That is why Soames claims that even knowledge of truth-conditions is insufficient for knowledge of meaning.

Soames is a bit sloppy here, using "arithmetic is incomplete" as his stock example of an English sentence that is necessarily true. Arithmetic itself is not 'incomplete'. Rather, formal theories of arithmetic, satisfying certain sorts of conditions, are always incomplete. But, of course, the point does not matter here.

The argument against the first form of the necessity view (on pp. 22-3) has two parts:

1. Knowledge of the compositional axioms, etc, does not seem necessary for understanding, since ordinary people do not have such knowledge. (What Soames has in mind here are e.g. the axioms that govern quantifiers, or attributive adjectives, or whatever.)
2. Even a totally trivial theory would state things necessary for understanding.

The argument against the second form of the necessity view is that the Foster problem applies to it, as well. The central claim here is that no truth theory can ever state everything knowledge of which is necessary for understanding, since knowledge of the truth-theory is compatible with false beliefs about what sentences mean.

Here again, see if you can sketch Soames's argument in your own words.

All that might make one wonder whether the second form of the necessity view is really very different from the sufficiency view. What is the difference between saying that knowledge of what some theory T states is sufficient for understanding and saying that T states everything knowledge of which is necessary for understanding? If you know everything necessary for understanding, won’t you then understand? If there's a gap there, is it one that will matter for this argument?

On p. 25, Soames begins consideration of Davidson's response to the Foster problem. He interprets Davidson as holding that, if one adds to one's knowledge of a truth-theory the knowledge that the theory is "translational"—i.e., that the sentences that appear on the right-hand sides of the T-sentences it generates translate the sentences mentioned on the left-hand sides—then that will be enough information to allow one to draw inferences about what the various sentences mean. He formulates this, on pp. 27-8, as a six-step argument. Soames finds almost every step of the argument to be problematic.

The sort of objection mentioned at (iii) above is what I call a `fiddly techincal objection'. Such objections can sometimes reveal serious problems with a view, so they cannot just be dismissed. But then tend to have `fiddly technical answers'—which is to say that they often do not go very deep.

Here's a different sort of view we might consider. One way to think of the Foster problem is as raising the question what additional premise might permit the inference from "S is true iff p" to "S means that p". As Soames interprets Davidson, and as Davidson states his view in "Radical Interpretation", the idea is that the additional premise is that the T-sentence is a theorem of a truth-theory that is verifiable on the basis of the sort of evidence that is available to a radical interpreter. So what's needed is knowledge about how the theory can be justified. The problem is that, if we think of it that way, then we have to explain how one can get from (i) knowledge of a T-sentence and (ii) knowledge of how one might justify a theory that proves that T-sentence to (iii) a claim about what the sentence means. And that looks complicated. This complication is what is driving Soames's objection to the six-step argument.

But if we're thinking of the theory as one that someone is actually using to interpret other speakers, then we do not necessarily have to suppose that they have knowledge about how the theory is justified. Rather, we could say that they will be able to make the truth-to-meaning inference so long as their knowledge of the T-sentence is in fact justified in a certain way, namely, on the basis of the sort of evidence that is available to them as a radical interpreter. The thought is that the mode of justification for the T-sentence is working as a kind of background, 'enabling' condition for the inference.

One might compare this to so-called 'rules of proof' in certain sorts of logical theories, such as modal logic. In such systems, there is an inference rule of the following form:

A.
So, it is a necessary truth that A.

Obviously, such an inference is invalid, in general. But, in such systems, the inference is permitted only when A is itself a theorem, i.e., if it has been justified in a certain way (namely, by means of a proof with no undischarged assumptions).

How much help (or how little) might all that be to Davidson?

This is an extremely difficult paper—maybe the hardest paper we will read. Part of the difficulty is that the dialectical structure of the paper is very complicated. I'll provide guidance by outlining the paper and asking some questions along the way. Note that by ¶1, I mean the first full paragraph on a page; ¶0 is the paragraph continuing from the previous page, if any.

In section 1, Higginbotham introduces his own account of the relationship between truth and meaning. It is perhaps worth emphasizing that Higginbotham is looking for an account of what meaning is that is "in harmony with semantic theory as it is actually practiced" (p. 3).

On pp. 3-4¶2, Higginbotham quickly sketches an account of why reference is essential to the theory of meaning. This essentially summarizes the arguments of "Truth and Meaning". (I.e., reference is the key to compositionality.) The rest of p. 4 quickly introduces the Foster problem, arguing that no appeal to structure (as Davidson had suggested in "Truth and Meaning") can solve it. The basic problem is that, so long as different words can have the same reference, there will be different sentences with the same content. But speakers can, in principle, 'respond' to those sentences differently.

In the rest of section 1 (pp. 5-6), Higginbotham articulates a positive conception of what role a theory of truth might play in an account of a speaker's linguistic competence, in particular, of their knowledge of meaning. He proposes to regard knowledge meaning as knowledge of facts about reference. E.g., someone's understanding the sentence "Alex runs" would be regarded as consisting in their knowledge that "Alex" refers to Alex, that "runs" is true of things that run, and so that "Alex runs" is true iff Alex runs. But that is not quite all. The thought is that these are facts one not only knows but also expects other speakers to know, somewhat the way that Lewis regards conventions as commonly known. Thus:

From this point of view, meaning does not reduce to reference, but knowledge of meaning reduces to the norms of knowledge of reference [that is, to facts about what one is supposed to know about reference]. Such norms are iterated, because knowledge of meaning requires knowledge of what others know, including what they know about one's own knowledge. To a first approximation, the meaning of an expression is what you are expected, simply as a speaker, to know about its reference. (p. 5)

Higginbotham does not here provide much motivation for this view. As he notes in footnote 6, however, there is an obvious link to notions like overtness that figure prominently in Grice and Lewis.

"[K]nowledge of meaning reduces to the norms of knowledge of reference." What precisely does that mean? What norms does Higginbotham have in mind?

How might one's own knowledge of T-sentences, and ability to rely upon others' knowing them, figure in an account of interpretation and communication? Take a simple case: Sam says to Toni, "Alex runs", in an effort to communicate that Alex runs. How, on Higginbotham's account, might this lead Toni to believe that Alex runs? How does it compare to the way that Lewis uses common knowledge in his account of conventions of truthfulness and trust and their role in communication?

What is most distinctive of this view, compared with others are have seen, is that it is unapologetically psychological: It is a view about what competent speakers actually know that allows them to speak and understand their language. Note, however, that some such knowledge is, as Higginbotham remarks at the bottom of p. 5, "tacit": These are not things one knows explicitly, but only sub-consciously. We'll spend a good deal of time, shortly, talking about this notion (which plays a major role in linguistic theory and in other parts of cognitive science).

In section 2, Higginbotham argues against certain other types of solutions to the Foster problem, and then argues that his own view does solve it. The first page or so rehearses Soames's way of formulating the problem. The question is: Why isn't knowledge of the truth-conditions of "Firenze è una bella città" compatible with false beliefs about what it means?

At p. 7¶1, Higginbotham distinguishes two sorts of responses to the Foster problem, very briefly articulating the "immanent" response, but then turning his attention to the "transcendent" response. He'll return to the 'immanent' response later. (This is one of the places the structure of the paper is most confusing.)

The transcendent response is discussed from p. 7¶2, through p. 8¶1. The idea is to deny that there are any facts about meaning beyond what would be apparent to a radical interpreter. That is: If there really is a difference of meaning of the sort on which the Foster problem rests (e.g., between "Snow is white" and "Snow is white and arithmetic is incomplete"), then that difference will have to be one that "disrupt[s] communication". If it does not, then the theories do not really differ in any way that matters.

The criticism is that, insofar as that seems plausible, it is because we are appealing to facts about the contents of speech acts, in which case such an account "swallows meaning whole" (p. 8). One might see this line of thought as continuous with Strawson's: What I need to know, on this account, to decide what theory of truth is correct for Gianni, is what his communicative intentions are when he speaks to me. (See esp. p. 8, ¶1: "...I guessed at Gianni's meaning based upon my beliefs about what he would be likely to be interested in telling me".) But if we have access to facts about communicative intentions, then we already know quite a lot about meaning. (And, of course, Davidson explicitly denies that he wants to appeal to such facts.)

How would Davidson respond to this argument? One option would be to insist that there are facts that would be apparent to a radical interpreter that would distinguish the two theories in question. What might those facts be? (Remember, again, Davidson's insistence that the radical interpreter does not have access to 'fine grained' facts about beliefs and intentions.)

The other option would, indeed, be to deny that there is any significant difference between the two theories. That makes the transcendent response essentially a version of Quine's thesis of the indeterminacy of translation, for which he argues in Word and Object. Davidson endorses a version of this claim at the very end of "Radical Intepretation". But one might wonder just how much indeterminacy it is possible to tolerate. Can we simply accept that there is no difference of meaning of the sort Soames and Foster are claiming?

Higginbotham criticizes the immanent response from p. 8¶2 through p. 9¶0. This part of the argument is a bit easier to understand. What is the argument?

Lepore and Loewer would definitely reject the claim that their response to the Foster problem "amounts to the thesis that translation into one's own speech...is sufficient for understanding..." (pp. 8-9). See their paper "Translational Semantics", listed as optional. But Higginbotham is not claiming that this is explicitly their view, only that their arguments commit them to it.

At p. 9¶2, Higginbotham explains own response to the Foster problem: The account outlined in section 1 is immune to it. On his account, we are actually to think of Gianni as knowing that "Firenze è una bella città" is true iff Florence is a beautiful city, and that this is what others also know and expect one another to know. This is of course an empirical fact (assuming it is a fact). But, Higginbotham is insisting, knowing that is different from 'knowing' that "Firenze è una bella città" is true iff Florence is a beautiful city and arithmetic is incomplete, and that this is what others also know and expect one another to know.

The crucial question here is parallel to one asked above: How would 'knowing' one of these things rather than the other affect one's use of language to communicate beliefs to other people, and to acquire new beliefs from them? That is: How might such knowledge be put to use in communication? The key to answering this question is implicit in Higginbotham's remark that "Statements of truth conditions that go beyond these bounds are irrelevant to understanding, resting as it does on common knowledge, and so irrelevant to meaning as well" (p. 10).

In the remainder of the section, Higginbotham briefly explores the role that knowledge, and common knowledge, play in his account. He remarks that "...the theory of truth...is not something that one starts with, augmenting it with conditions or constraints so as to make it acceptable as a theory of meaning. Rather, truth comes in as something [a competent speaker] knows about, and the deliverances of the theory are of interest only insofar as knowledge of them is part of [their] linguistic competence" (p. 10). This echoes Higginbotham's earlier remark that his account "makes use of the information that a person tacitly possesses about the truth conditions of her own utterances" (p. 5).

There is an implicit criticism of Davidson here, and an attempt to re-orient the focus of the theory of meaning (that is, of semantics). What is the criticism, and what is the new focus supposed to be?

You will recall that Strawson claimed that 'formal semanticists', such as Davidson, could give no reasonable account of the notion of expressing a belief, without appealing to the notion of a communicative intention. Rumfitt here means to answer Strawson, though he is going to argue that, to explain that notion (or one in the same vicinity), we need both the sorts of resources typically deployed by Griceans and the notion of truth-conditions. Rumfitt's focus, for our purposes, is on the notion of expressing a thought. This is something one does not only when one makes an assertion, but even when one makes a conjecture or a supposition.

The first two sections mostly summarize material we have already discussed. But there are a couple of new points. In §II, Rumfitt argues, on Strawson's behalf, that the notion of truth, as it appears in the T-sentences that are supposed to 'give meaning', cannot be one that is simply defined in a certain way. This kind of point has a long history in the literature on Tarski. (See e.g. the paper by Hartry Field listed as an optional reading.) As Rumfitt says, Davidson does not (by 1990, anyway) disagree, so we need not worry too much about this point, a version of which we have already seen in Strawson himself.

The short version of this point is that if truth is defined by a Tarski-style recursive definition, then T-sentences like:
(*) "Snow is white" is true iff snow is white
turn out to be things we can prove mathematically. But the ordinary version of (*) is simply not a theorem of mathematics: You cannot find out what "Snow is white" means by doing math.

It is also important to note that it is simply not clear what kind of 'account' of truth is needed here, if any. As Rumfitt notes, Frege famously insisted that truth is not definable at all, and Davidson once wrote a paper titled "The Folly of Trying to Define Truth" (PhilPapers). Perhaps we can simply treat truth as 'primitive'. Or perhaps the most we can hope is to explain how truth is related to certain other equally basic concepts: meaning, belief, and the like.

In §III, Rumfitt considers whether Strawson is right that the notion of assertion, and that of the content of an assertion, can be explained in terms of communicative intentions. The question that Rumfitt presses is what, precisely, my communicative intention is supposed to be when I assert, say, that (my cat) Lily is asleep. We've discussed this question before, and Rumfitt recounts some of the back and forth that led to the idea that my intention is to get my audience to believe that I believe that Lily is asleep, not to believe that Lily is asleep. But Rumfitt argues that there are problems with that view, too, and suggests at the beginning to §IV that there is not going to be a satisfying resolution to this problem: There just isn't any intention that one must always have when asserting that Lily is asleep.

Formulate, as clearly and precisely as you can, Rumfitt's argument against Gricean analyses of assertion. Explain how that argument would equally threaten views that take communicative intentions to involve knowledge.

Why not just say that the intention constitutive of asserting that Lily is asleep is: that one intends to express the thought that Lily is asleep?

For additional discussion of accounts of assertion, see the SEP article thereon.

What Rumfitt suggests is that we should abandon the attempt to explain in such terms what assertion is and attempt instead to explain a more general notion: that of expressing a thought, e.g., the thought that Lily is asleep. The key idea is to appeal to the structure of a speaker's reasons for speaking as they do, and to try to extract from those reasons an understanding of what it is to express a thought. The main point in this section is that, whatever the speaker's reasons are, they will always have to involve (and, in a certain sense, terminate in) the speaker's reasons for uttering a particular sentence. This is what is known as a 'basic action': It is something one can 'just do'; one does not have to do it by doing anything else. (By contrast, Rumfitt is claiming, expressing a thought is not something I can 'just do'. I have to do it by uttering a sentence and, indeed, by choosing a particular sentence which will express that thought in the context in which I find myself.)

There are some complex issues about the nature of action in the background here. To utter a sentence, I have to move my lips and tongue, etc, in a certain way, and contract and relax my vocal chords. But these are not things I can do as such: They happen when I speak, but they are not directly under my intentional control. (Compare: To type these words, the muscles in my fingers have to move in certain ways; but the very specific ways in which they move is not something I directly control.) See the SEP article on action for further information.

Formulate, as clearly and precisely as you can, Rumfitt's argument that our reason for speaking must always involve reasons to utter a particular sentence. What would speaking be like if that were not so?

As Rumfitt remarks, there is no assumption here that people consciously go through the kind of reasoning he discusses. The point of his reconstructions are to allow us to get a sense for what people's reasons actually are, and how our beliefs and desires lead us to perform the linguistic actions we do. This is a common and relatively uncontroversial way to think about reasons for action. Suppose I walk across the room and pick up the eraser. One might ask why I did so. The answer might be that I wanted to erase something on the board, and I wanted to use the eraser to do so (rather than my hand, or my shirt). I do not have consciously to have thought, "Oh, I should use the eraser rather than my shirt" for that to be a correct answer to the question why I did what I did—even in a really strict sense according to which my 'reasons' must also be causes of my action.

Rumfitt's analysis is developed in §V. The key thought here is that the speaker's reasons for uttering a given sentence will, in typical cases, include something that establishes a link between that sentence and a certain proposition that they mean to express. (Maybe one just wants to hear that particular sound, but that is not the typical case.) T-sentences are then supposed to make this connection between a sentence and the proposition one expresses.

We can pose the central question this way. Consider the (A) syllogism on p. 838. Why will the students learn that the Battle of Waterloo was fought in 1815 if the instructor utters the sentence "The Battle of Waterloo was fought in 1815" (that sentence is the one dubbed S₀)? The connection between the sentence and what the students learn is left out of this account of the instructor's reasons for speaking as they do. Rumfitt's idea is that we can fill in the missing details as follows:

1. The students know that an utterance of "The Battle of Waterloo was fought in 1815" is true iff the Battle of Waterloo was fought in 1815. (Why? Because they understand English.)
2. The students will regard my utterance as true. (Why? Because I'm their teacher, and they trust me to speak the truth about such matters.)

If I utter "The Battle of Waterloo was fought in 1815", then, I can expect the students to use (i) and (ii) to reach the conclusion that the Battle of Waterloo was fought in 1815. So it is rational for me to utter that sentence, if my goal is to get them to believe that the Battle of Waterloo was fought in 1815.

Something similar applies in the case of the (B) syllogism. Fill in the details.

So the idea is: What thought someone expresses when uttering a given sentence is determined by the T-sentence that, the speaker expects their audience to deploy in interpreting and responding to their utterance.

A form of the Foster problem arises on p. 841, as an objection to the analysis stated at the top of that page. The observation is that various T-sentences might appear among my reasons for speaking. (Rumfitt gives a very complicated example to show this.) The idea behind his response is that, nonetheless, among these, there will be one that is most basic. Here is a simpler case. Suppose I want you to come to believe that Alex has left, but that I already know that you know that Drew has left if, and only if, Alex has left. (They always leave together.) Then I could utter "Drew has left" and leave the inference to you. So one of my reasons for uttering "Drew has left" is that

3. You will know that my utterance of "Drew has left" is true iff Alex has left.

But my utterance does not express the thought that Alex has left, but rather the thought that Drew has left. Rumfitt's idea is that (iii) is not my most basic reason for uttering "Drew has left" but is itself grounded upon other things, namely:

4. You will know that my utterance of "Drew has left" is true iff Drew has left.
5. You know, independently, that Drew has left iff Alex has left.

So it is (iv), the more basic T-sentence, that tells us what thought was expressed.

By the way, if you replace "Drew has left", in the foregoing, with an arbitrary sentence and "Alex has left" with some arbitrary other sentence, then this kind of example can be used to prove that the utterance of any sentence can implicate any proposition, given an appropriate context. For that purpose, one only needs the left-to-right part of (v), not the biconditional. (Rumfitt's argument needs the biconditional in (v), because otherwise (iii) would not follow from (iv) and (v).)

To explain this notion of a 'most basic' T-sentence, Rumfitt introduces the notion of an "I-source" of knowledge of a biconditional. It is certainly worth considering whether that notion, as Rumfitt explains it, is adequate to the task for which it is introduced. However, it's also worth remembering that the underlying idea is simply that one of these T-sentences is 'most basic'. There are other ways one might try to explain that. (I pursue a somewhat different strategy in the paper of mine listed as optional.)

Say, as clearly and precisely as you are able, in what sense (iv) is meant to be 'most basic'. (As just indicated, this is far from trivial. It could make for a good final paper topic.)

Restricted version of the previous question: Is the notion of an I-source of knowledge really needed in the analysis with which Rumfitt finally settles on p. 842? That is: Could the T-sentence that figures in the 'lowest practical pro-syllogism' be one that did not issue from an I-source of knowledge? Why or why not? If not, can we explain what is special about those cases? (This again could make for a good final paper.)

There is arguably a close relationship between Rumfitt's proposal and Higginbotham's answer to the Foster problem. In particular, both of them talk about "expectations" about what one's conversational partners will know. What more can be said about the relation between these two views? (Exploring that question would also make for a good final paper topic.)

Part II, which you do not have to read, but are invited at least to skim, explores the same sort of issue, but in connection with "taking in" or "apprehending" a thought expressed by someone else. The analysis Rumfitt eventually gives, at the bottom of p. 850, is parallel to the analysis of expressing a thought, and many of the arguments Rumfitt gives in favor of it are parallel to those that are given in Part I.

The main reason we are reading this material is to get an idea of the way that (most) theoretical linguists think about knowledge of language, and their reasons for endorsing a broadly 'cognitive' approach to linguistic competence. We have seen some of these ideas in our earlier readings, but the views presented here are more developed than those in Aspects. Probably the most important distinction that Chomsky makes, for our purposes, is between E-language and I-language. We will need to appreciate both what the distinction is and why Chomsky thinks that I-language should be the focus of linguistic inquiry.

Chomsky sometimes uses the term "grammar" to mean I-language and "language" to mean E-language. This is an older terminology that he had used in Aspects of the Theory of Syntax. It is potentially confusing, since, if we speak this way, then Chomsky's view is that 'language' is irrelevant to linguistics!

Chomsky begins by motivating the study of language from a psychological point of view. He urges that there are three basic questions to be asked, which I will rephrase as:

1. What does a competent speaker of a language know that allows them to be able to speak their language?
2. How is that knowledge acquired?
3. How is that knowledge deployed and employed in the actual day-to-day practice of speech?

Chomsky himself has suggested that, if we're uncomfortable talking about speakers' 'knowing' linguistic principles, we can instead talk of speakers 'cognizing' them. Another option would be to reformulate (1) as: What information does a competent speaker have...?

Question (1) can be asked in more or less specific ways. One might be interested in what some particular speaker knows; or what speakers of some language like English know; or what sort of knowledge is common to all human language-users (that being 'universal grammar', or UG). All these questions are empirical—including the question whether there is any knowledge common to all human language-users. But Chomsky thinks there are good reasons to pursue these questions.

What kinds of reasons does Chomsky give to be think that these are the right questions to ask? (See especially pp. 7-13.) Which do you find most (or least) impressive?

In §2.1, Chomsky discusses a number of idealizations that are (he suggests) commonly made in theoretical linguistics. He defends these idealizations on pp. 17-8. It's a reasonable question, though, how the study of an idealized situation can give us information about more realistic, usually very un-ideal situations. How might Chomsky answer that question?

Regarding the distinction between E- and I-language: When Chomsky introduces this terminology in §§2.2-2.3, he remarks that E-language is external language, and I-language is internal language, and many of his remarks accord with that usage. But there is another distinction that is at least as important, and that is the one he really has in mind: between language thought of extensionally and language thought of intensionally.

If we restrict ourselves to semantics for the moment, then, on the extensional conception, a language would be thought of as a function from sentences to meanings (or something like that); this is how Lewis thinks of 'languages'. On the intensional conception, language would be thought of as a system of rules for assigning meanings to sentences. Note that, since there can obviously be many sets of rules that yield the same function from sentences to meanings, languages in the intensional sense are finer-grained than languages in the extensional sense: Many I-languages will generate the same E-language.

A similar distinction is often made in mathematics: between functions thought of extensionally and functions thought of intensionally. Extensionally, a function is just a pairing of inputs (arguments) and outputs (values). Intensionally, a function is a rule for calculating the output from the input. So, extensionally, f(x) = x²-1 and g(x) = (x+1)(x-1) are the same function; intensionally, they are different functions.

A similar distinction is also made in logic (from which Chomsky is borrowing the terminology). There are two different ways to think of a 'theory': as a set of theorems, or as a set of axioms that 'generate' a set of theorems (namely, the theorems that can be proved from the axioms). The former is the 'extensional' way of thinking the theories; the latter, the 'intensional' way. Both have their uses, but the intensional notion is arguably the more fundamental one. (In fact, there is an even finer-grained 'intensional' notion of a theory, which concerns not just what the axioms are but how they are 'presented'. This is especially important when there are infinitely many axioms, so that one cannot just present them as a list.)

Think just of the case of syntax, where we are concerned not with meaning but just with grammaticality. How then would we think of languages on the extensional and intensional conceptions?

The focus on I-language is arguably connected to the conscious inattention to 'normative-teleological' aspects of the ordinary notion of language. How so?

Note that I-languages, in Chomsky's sense, are every bit as abstract as 'languages' in Lewis's sense (see pp. 22-3). But they are not just mappings from sentences to meanings. They are collections of rules by means of which a meaning can be determined for a given sentence. So whereas Lewis asked: What E-language L does a given population P use? And what is it for L to be the language of P? Chomsky wants us to ask: What I-language does a given speaker 'know'? What it is for them to know it? How are I-languages acquired and shaped by experience? Universal Grammar then becomes the theory of the 'initial state' from which language acquisition begins and so, derivatively, a theory of what I-languages it is possible for human beings to acquire.

What is the relation between these two ways of interpreting Chomsky's distinction? There are four possible categories: extensional external; extensional internal; intensional external; and intensional internal. Which does Chomsky have in mind? Might some of the others also be interesting and worth studying? Do some of them seem hard to make sense of?

The distinction between E- and I-languages might also seem to be related to the distinction between public or common languages, such as English or German, and 'idiolects': language as understood by a particular speaker at a particular time. Are these related? If so, how? (One might want to consider here Chomsky's remarks in Ch. 1.)

Recall Lewis's remark that he "know[s] of no promising way to make objective sense of the assertion that a grammar Γ is used by a population P whereas another grammar Γ', which generates the same language as Γ, is not", which Chomsky partially quotes. How would Chomsky respond? (See esp. pp. 23, 30-1, 37-8, and 39.)

Chomsky ends up, in §2.4.1, being very dismissive of the notion of E-language, as well as of 'common' languages. Is there something he's missing? Are there, at least, questions he should be taking more seriously? Just how dismissive is he of 'common' languages?

Note: When Chomsky speaks of the 'language of arithmetic' here, you can think instead of the language of basic logic.

Much of Chomsky's discussion revolves around syntax and, in one striking passage (pp. 41ff), phonology. To emphasize a point that may seem obscure: Despite how it will seem to you, there is not actually an "n" sound in the word "bent" (as it is normally pronounced by English speakers). You hear the "n", but it is not really there in the sounds themselves. What you hear is determined not just by the actual sounds but by the phonological representation that your mind constructs of that word. Similarly in the other cases: What is in column (II) represents what you hear; what is in column (III) represents the sounds that are actually present.

It would be natural—despite what Chomsky says on pp. 44-5—to want to apply these same sorts of considerations to the case of semantics. That is what we saw Higginbotham explicitly propose to do: "I am applying to semantics a research program that goes forward in syntax and phonology, asking, 'What do you know when you know a language, and how do you come to know it?'" ("Truth and Understanding", p. 13).

How might the sort of program in semantics that Higginbotham is proposing (and pursuing) be motivated, in a broadly Chomskyan way? One way to answer this question would be to consider the three "basic questions" that Chomsky asks on p. 3: How do they arise in the case of semantics? How are those questions similar to or different from the ones that arise in the case of syntax?

For those with relevant background: Chomsky clearly thinks that the psychological perspective he favors demands an internalist conception of mind and language. Whether that is so is controversial. My own view, and that of many others, is that it does not. But we shall not pursue the matter, since it will not significantly affect the issues we will be discussing.

This paper is about some general issues in philosophy of science, but with important applications to philosophy of mind (and so to philosophy of linguistics). These issues have become especially pressing recently, with developments in artificial intelligence.

One way to think about the central question of the paper is: Why are there sciences other than physics? Physics is meant to be 'general' in the sense that everything that happens in the physical world admits of physical explanation. That is as true of monetary transactions and animal development as it is of billiard balls bumping into each other. In some sense, that is to say, the various events (e.g., hand movements, computer transactions, etc) that constitute my paying for groceries can all be explained, in principle, in physical terms; the same goes for, say, the transformation of a caterpillar into a butterfly. Really, it's all just atoms moving around (or various sorts of fields transforming in certain ways).

Why, then, are there such sciences as economics or biology? or linguistics and (other parts of) cognitive science? The epistemological answer would be that the physical explanations are simply too complicated for us to understand. The sorts of explanations given in economics and biology are, so to speak, high-level summaries of physical explanations. But ultimately, and ideally, they would be replaced by the really fundamental explanations, which would be given entirely at the physical level. We have examples of this kind of 'reduction' in the molecular theory of heat (heat is just molecules moving around) and of chemical valence to atomic structure (valence has to do with the structure of electron rings). Hence, as Fodor says, on this view, "the more the special sciences succeed, the more they ought to disappear" (p. 97), because the explanations given in the special sciences will ultimately be replaced by ones at more fundamental levels.

In the case of cognitive science, this idea takes a very particular form. Setting aside Cartesian dualism, mental states are brain states; cognitive development involves certain sorts of changes in the brain; mental events and processes are neurological events and processes; etc. So, one might think, every mental phenomenon must ultimately be explicable in purely neurological terms, without reference to such notions as belief, perception, information, and the like. We do cognitive science and psychology only because neurology is too complicated; once we know enough about the brain, we won't have to do cognitive science any more.

The position just described is sometimes known as 'eliminativism' and is much identified with the work of Paul and Patricia Churchland. See the SEP article on eliminative materialism for more information.

This is not just an abstract philosophical position but one that concretely affects the practice of science. For example, it affects who gets grants, what kinds of departments receive university support, and so forth. The excitement caused by advances in neuroscience and artificial intelligence is today leading (otherwise sensible) people to suggest that linguistics and cognitive science are soon to be out-dated, because we will soon be able to give explanations of linguistic and cognitive phenomena in more basic terms.

Fodor is out to argue that this view is fundamentally mistaken. Fodor considers a very simplified form of 'reduction', one that is focused entirely on 'laws'. The central question is what is required to 'reduce' a scientific law at one level to laws at a more basic level. (I think Fodor would claim that the simplification can only hurt his case: The more kinds of 'reductions' we have to carry out, the harder it will be. So the idealization helps his opponent.)

Fodor mentions Gresham's Law, from economics, as an example. For an example of a psychological law, consider Weber's Law, which states that the change in a stimulus needed to produce a change appreciable by an organism is proportional to the original stimulus. For example, if you have an array of dots, then how many dots you need to add to get it to look noticeably different to an organism is proportional to the original number of dots. (Adding two dots to two dots will do it; adding two dots to a hundred dots will not.) The same sort of thing is true for the number of clicks in an auditory stimulus, or the number of bumps in a tactile stimulus.

Fodor frequently mentions that laws must 'support counterfactuals'. What this means is that laws do not just concern what actually happens but also what would have happened had things been different. Not only is it true that my pen just fell to the floor when I let go of it. It is also true that, if I had let go of it a few moments before, it would have fallen to the floor then.

If we take the special science law to be "S₁ events cause S₂ events", then what's required to 'reduce' this law to physics are:

1. Bridge laws that say what physical events constitute the special science events: S₁s are really just P₁s; S₂s are really just P₂s.
2. A corresponding physical law: P₁ events cause P₂ events.

I'll speak here in terms of 'events', since Fodor does so. But we could also speak in terms of 'states'.

Fodor argues on p. 99 that the bridge laws have to be formulated as identities: Psychological states just are brain states; no weaker claim will do. What is the argument?

Those who are familiar with Saul Kripke's work may balk at Fodor's claim that bridge laws express "contingent event identities". There are things to be said about this, which I'd be happy to discuss outside of class. For now, let me simply assert that nothing Fodor is claiming is actually inconsistent with Kripke's thesis that identities are always necessary. (But Kripke himself is guilty, in my view, of overlooking one of Fodor's main points here in his argument for dualism in the third chapter of Naming and Necessity.)

Fodor then argues for two crucial claims.

1. In order for the reduction to work, kinds of special science events must 'reduce' to kinds of physical events. This is because physical laws relate physical kinds. There can't be a law of the sort mentioned at (ii) unless P₁ and P₂ are physical kinds. Fodor suggests that, in so far as this is not obvious, it is because we forget that not every physical description determines a physical kind: for example, "has at some time been less than three miles from the Eiffel Tower" does not determine a physical kind; there are no laws about the things less than three miles from the Eiffel Tower. Of course, other physical laws will apply to those objects, but not in virtue of their satisfying that description.
2. This sort of 'kind identity' is not guaranteed by what Fodor calls "token physicalism": the claim that every individual event that falls within the domain of any special science is (identical to) a physical event. The reason is that token physicalism (as opposed to "type" physicalism) allows events that are of the same kind from the point of view of the special science to be (identical to) events that are of different kinds from the point of view of physics.

And this is how Fodor thinks things are: Events that are of the same kind, from the point of view of a special science, are typically 'multiply realizable'; even though each S₁ event is identical to some physical event, they are not all identical to the same kind of physical event. From the point of view of physics, that is to say, S₁ events might be 'wildly disjunctive' (that is, of arbitrarily many different physical kinds that have little or nothing in common, physically speaking). And yet, there might be "interesting generalizations" (i.e., laws) to be stated about S₁ events (p. 103). The special sciences, as Fodor sees it, are in the business of making just such generalizations.

To see what kind of thing Fodor has in mind, note just how unlikely it is that all utterances of the word "extraordinary" have anything physically in common. Not only can the word be spoken, signed, and written (with ink, pixels, or even condensed water vapor, in the sky), even spoken forms differ significantly: As Linda Wetzel once pointed out, "extraordinary" can be pronounced with anywhere from two to six syllables. (The two syllable version is the British: strord-nreh.) But there are interesting things to say about utterances of that word (for example, how they are perceived), in all these different forms.

It is important to see that the argument here is not at all special to the case of the human sciences. Fodor argues elsewhere that similar remarks can be made about geology and aerodynamics: Aerofoils, for example, can be made out of just about anything. In particular, the explanation of why planes can fly is pretty much the same as the explanation of how it is possible to sail into the wind, which is pretty much the same as the explanation of how hydrofoils work (both on dolphins and on boats), even though sails and wings and flippers tend to be made of different things and to have quite different shapes, and even though hydrofoils move through water, whereas wings and sails move through air. Aerodynamics is in the business of telling us what all these things, physically different as they may be, have in common.

Fodor’s claim might therefore be understood as follows: There can be no general reason to think that psychology must one day give way to neurology that would not equally show that aerodynamics, geology, and biology must also give way one day to more basic sciences. But, simply as a matter of sociological fact, there does not seem to be anything like the same clamor for the latter as for the former.

Is there some significant difference between the human sciences and non-basic physical sciences that Fodor might have overlooked?

In §III, Fodor sketches an alternative account of how special sciences relate to more basic sciences. The discussion proceeds in terms of yet another simplifying assumption: that there is some disjunction of physical event types to which a special science kind will correspond. One might think that this assumption is less innocent: that it matters just how 'wild' the disjunction is. For example, if there were just two physical kinds that 'realized' a special science kind, that would frustrate reduction strictly speaking, but perhaps not in spirit. In practice, however, special science kinds are, indeed, wildly disjunctive. (See the example of aerodynamics above.)

The basic idea here is that the more basic sciences can help us understand how the special science laws are 'implemented': for each of the possible realizations of S₁, we can see how it gives rise to a corresponding realization of S₂. But why it is worth grouping the possible realizations together in this way is opaque from the point of view of the more basic science.

Another important point Fodor makes here is that the laws of special sciences very often have exceptions, and that it is hard to see how that could be if reductionism was true. Special science laws only hold, as it is put, ceteris paribus, that is, other things being equal. (A simple if boring way to see this is to note that there's always the possibility, however small, of some bizarre quantum mechanical phenomenon that leads to a violation even of the laws of chemistry, let alone those of economics or psychology.) Fodor's idea is that this is because some realizations of S₁ will not be lawfully related to realizations of S₂, even though almost all of them are. A different idea would be that the reduced laws—that P₁s cause P₂s—might not hold in full generality, but only 'other things equal'. (This need not imply that real physical laws have exceptions.) For our purposes, this point will not be particularly important, but see the optional readings for more, if you're interested.

The moral of the story is that special sciences do not exist simply because it would be so difficult to understand a purely physical explanation of, say, the relation between inflation and unemployment. Rather:

[T]here are special sciences not because of the nature of our epistemic relation to the world, but because of the way the world is put together: not all natural kinds (not all the classes of things and events about which there are important, counterfactual supporting generalizations to make) are, or correspond to, physical natural kinds. (p. 113).

Explain exactly what Fodor means in the passage just quoted and why it implies that there are, and will continue to be, special sciences. How does it ensure the 'autonomy' of psychology and so 'protect' it from eventual elimination by neurology?

Evans's main goal in this paper is to try to explain what motivates the requirement that theories of meaning should be compositional, and to explain as well what might allow us to distinguish "extensionally equivalent" theories (that is, theories that prove all the same T-sentences, but use different sets of axioms to do so).

As was the case in Lewis's discussion, Evans is assuming, at least for the purposes of this paper, that speaker's linguistic behavior ultimately concerns sentences, and that facts about the meanings of words (e.g.) must somehow supervene on facts about the meanings of sentences. One could question this assumption.

Evans begins by insisting that, if compositionality is to be motivated at all, then it must be motivated in terms of speakers' actual abilities to understand 'novel' sentences. The question is what the relation is between a compositional (or, as Evans often says, "structure reflecting") theory of truth and actual speakers of the language. There are two options. On the weaker, one 'knows' such a theory if one acts as if one knew it. But then there is no distinguishing between extensionally equivalent theories. (Recall Lewis's skepticism about the possibility of making good sense that one 'grammar' rather than another is correct.) The stronger interpretation is that speakers somehow unconsciously deploy the information contained in the theory to figure out the meanings of novel sentences. The rest of the paper tries to give some substantial content to this view.

What would be the analogous distinction in the case of syntax? A much larger question is: How might Evans's arguments be adapted to that case?

In section II, Evans discusses a toy language with ten unary predicates and ten names, and so 100 sentences, and considers two theories of truth, T₁ and T₂, that (respectively) do and do not ascribe any sort of structure to the sentences of this language. Evans suggests that, considered as accounts of what a speaker of the language knows, these two theories can be distinguished empirically. The core idea is that the axioms of the theories can be associated with certain abilities (or dispositions) that the speaker has.

Evans specifies these dispositions using what are known as "substitutional" quantifiers, but we can (for our purposes, at least) rephrase the disposition that constitutes knowledge that 'a' refers to John as:

(*) For any predicate P and any property φ, if U (tacitly) knows that P is true of the things that have the property φ, and if U hears an utterance of "Pa", then U will judge that the utterance is true iff John has the property φ.

This is supposed to capture the idea that U (whoever that is) take sentences that contain 'a' to be 'about' John, to the effect that he has whatever property is associated with the predicate P.

Test your understanding by re-writing Evans's statement of the disposition associated with the predicate 'F' in a similar fashion.

Explain in detail how these two dispositions give rise to a third disposition: Whenever U hears an utterance of 'Fa', they will judge that the utterance is true iff John is bald. (Do you need to appeal as well to what, on the bottom of p. 327, Evans calls the 'compositional axiom'? Why or why not? Note that Evans never tells us what disposition is supposed to correspond to it.)

Evans insists that these dispositions must "be understood in a full-blooded sense". Evans is alluding to a distinction generally made about dispositional properties. For something to be fragile, for example, is for it to be disposed to break easily. But (we ordinarily suppose) that is not all there is to the story: Something that is fragile has some other property that explains why it breaks easily. It is in its possession of this other property that the fragility of the thing consists. (The underlying property is sometimes known as the 'categorical basis' for the disposition.)

A famous example often used to illustrate this distinction is 'dormativity': the disposition something might have to make one sleepy. If we ask, "Why does this pill make one sleepy?" the answer can't just be: Because the pill is dormative. That just means that the pill tends to make people sleepy. It would be viciously circular to say that it makes people sleepy because it tends to make people sleepy. There has to be some other (presumably chemical, in this case) property the pill has that explains why it tends to make people sleepy. So "there is a common explanation to all those episodes of" the pill's making people sleepy.

So, in this case, Evans wants us to think of (*) not just as saying something about what beliefs U will form under what circumstances but as saying that there is some underlying (presumably cognitive) state corresponding to this disposition that explains why U tends to form the beliefs they do.

Explain as clearly as you can why Evans thinks that the dispositions in which (he is arguing ) tacit knowledge consists must be 'full-blooded'. (The argument is on pp. 329-30, and the basic point is that, otherwise, we can't distinguish extensionally equivalent theories.)

Evans goes on to argue (pp. 331-4) that there are empirical differences between the two theories, i.e., that there could be differences in the way a speaker behaved that would justify our saying that they knew one or the other of T₁ and T₂. What are these differences? Can you think of any other such differences we might expect to find?

It's perhaps worth noting that Evans's 'psychological model' for T₁ is compatible with, and seems even to embody, the assumption that the subject's thoughts are structured. What's at issue is simply whether the subject's language has a similar structure.

The paper by Ladefoged and Broadbent to which Evans refers in note 5 was listed as optional for the excerpts from Aspects. It's really quite a fascinating experiment, so the paper is well worth reading.

In section III, Evans outlines some reasons for thinking that 'tacit knowledge' is quite different from ordinary knowledge (or belief). His main point is that it has what one might call restricted application: One's tacit knowledge of the meanings of words and the principles by which they combine is used only in one's comprehension of language. It is not available, in particular, for one to report verbally. Otherwise, semantics would be easy: If you wanted to know how adverbs work, say, you could just ask someone; after all, we all know how they work, right? But we only know such facts tacitly, and our knowledge is (as it is sometimes said) "modular" and unavailable to us outside certain specific applications.

The idea that the mind consists of a number of distinct systems that are 'modular' in this way is largely due to Jerry Fodor. The view is often combined with the idea that there is also an 'executive' system that is not modular, but which co-ordinates the activity of the various modules. See the SEP article on modularity for more information.

In section IV, Evans discusses the question whether attribution of tacit knowledge of a structured theory of meaning to a speaker can explain their capacity to understand novel sentences. Evans first concedes that, by itself, it cannot. It is, honestly, not entirely clear to me why Evans makes this concession. I think the reason is this. Suppose that the speaker regards some utterance of the sentence "Gb", for example, as being true iff Harry is happy. And suppose that we were to regard that judgement as the exercise of the dispositions that constitute their tacit knoweldge of the theory T₂. Then the judgement cannot also be regarded as explained by her having those same dispositions: that would be like explaining a sleeping pill's effectiveness in terms of its dormativity.

But the usual response to this kind of objection is that it assumes that the disposition is "thin" rather than "full-blooded", as Evans's were meant to be. So why doesn't Evans just say that, if the dispositions are full-blooded, then there is an explanation to be had, in terms of whatever the categorical basis of those dispositions is? Perhaps this is the point Evans is making when he says that "to say that a group of phenomena have a common explanation is obviously not yet to say what the explanation is": To give a satisfying explanation, we would need to know what the states underlying the dispositions actually were.

I suggest that Evans may be guilty here of roughly the same sort of confusion Fodor identifies in "Special Sciences". How so?

Evans goes on to suggest that we can get an explanation of "creativity" if we embed the attribution of tacit knowledge in a larger story about language acquisition. How is that supposed to go? How does it relate to the "full-bloodedness" of the dispositions? (Developing a proper answer to the various questions I've been asking here would make a nice final paper.)

Davies is interested in the question whether there is a way of making sense of the idea that a language (in roughly Lewis's sense) has a certain structure without having to claim that speakers of the language must in any sense, including tacitly, be aware of that structure. So, in particular, in the case of Evans's 100-sentence language, Davies wants to claim that, even if speakers fail to have the sorts of dispositions Evans specifies, it can still be true that the language they speak has a certain sort of semantic structure. Davies is thus arguing that semantics can be neutral on the question whether there is a cognitive basis for semantic competence.

Davies talks both about truth-theories and semantic theories that pair sentences with meanings, or as issuing in statements like: S means that p. This difference does not much matter for the present discussion. I'll speak about 'theories of meaning' below, meaning to include 'interpretational' theories of truth.

Davies argues by example, on p. 137, that the best theory of meaning for a language need not be one to which speakers of that language bear any interesting relation. Is the example convincing? Are the imagined speakers of L₀ "obliged" (p. 139) to revise their opinions about what, say, either "Fa" or "Gb" means if they have revised their opinion about what "Ga" means? Why or why not?

Recall that Evans suggested a theory of meaning for a language should be 'structure reflecting' if there were certain patterns of acquisition and loss among speakers of that language. As I'll remark below, Davies is implicitly adding a third such constraint, concerning change of meaning.

Davies suggests that semantic theories should instead meet what he calls the structural constraint:

If, but only if, there could be speakers of L who, having been taught to use and know the meanings of sentences (of L) s₁, ...,s_n..., could by rational inductive means go on to use and know the meaning of the sentence s..., then a theory of truth for L should employ in the canonical derivations of truth condition specifying biconditionals for s₁, ...,s_n resources already sufficient for the canonical derivation of a biconditional for s. (p. 138)

The kind of case Davies had in mind would be one in which someone had learned the meanings of "Fa" and "Gb" and was, on that basis, able to figure out the meanings of "Fb" and "Ga".

Note (as Davies emphasizes) that this involves what a hypothetical speaker could do, much along the lines of what Davidson suggests in "Radical Interpretation". The basic point here is supposed to be, again, that semantics need not have anything particular to do with facts about how speakers actually do understand their language. It is enough if someone could 'project' meanings in a certain way. (You need no worry too much about the discussion of the alternative Davies calls S*. It is still hypothetical.)

In part I of the paper, Davies considers four objections to the structural constraint. You can skip the discussion of the second, in §6,and the last paragraph of §7, on pp. 145-6.

The first objection is that there might be cases in which one can 'work out' the meaning of some sentence, given what one knows about the meanings of other sentences, but where one would not want to say that there was 'semantic' structure that made such projection possible. (The third objection is in the same ballpark.) Are there better examples where meaning is systematic in some way but where there is no syntactic structure to which it corresponds? If so, how much of a problem is that for Davies?

The most important of the objections is the fourth. Davies' initial presentation of the objection is somewhat complicated (because of the insistence upon generality), but he gives a concrete example on p. 146 (to which you can probably just skip). The core of the objection is that certain actual speakers of a language might lack the ability to "project" the meanings of certain sentences they have not encountered (e.g., "Bug moving"), whereas other speakers might have that ability. If so, then it seems odd to say that the language of the speakers who cannot "project" has the same structure as that of the speakers who can project, simply because someone else could project meaning in that way. Davies's response is to deny that, in such a case, the common sentences of the two languages (e.g., "Bug") actually do have the same meaning.

How exactly does this save the structural constraint? What does it say about how the meanings of sentences are being specified when languages are identified as Davies identifies them at the very beginning of the paper? Can we really say what the meanings of the sentences are without knowing whether they are semantically structured? That is: Is this response compatible with Davies's insistence that, even if speakers understand Evans's 100-sentence language in a non-compositional way, still the right semantics for that language is the compositional one? (Consider carefully Davies's remark that "Members of the S group either do not have, or else do not employ here, the concept of a bug..." (pp. 146-7).) We'll return to this below.

In part II, Davies discusses what is involved in crediting speakers with "implicit" knowledge of a theory of meaning for their language. Davies first step is to introduce a notion of "full understanding" of a language: Someone fully understands a language L if they are in a "differential state" with respect to the various words (semantic primitives) of the language.

The terminology here may be confusing. But, roughly, you are in the "<σ, L> state" if you know what the meanings of all the L-sentences containing the 'word' σ are. You are in the "σ differential state" if your knowledge of the meanings of those sentences is derived from your knowledge of what σ means. What Davies is trying to do here is to say how these two 'states' are to be distinguished.

Exercise: Come up with better terminology. (This is often an important part of writing.)

Ultimately, Davies's view is that someone 'fully understands' a language just in case they have the sorts of dispositions concerning acquisition and loss that Evans discusses, as well as dispositions concerning change of meaning that Evans does not mention. (This allows for answers to certain sorts of objections to Evans.) That is: If you change what you think "Fa" means, then you must either change what you think "Fb", "Fc", etc, mean, or you must change what you think "Ga", "Gb", etc, mean. The idea is that, if you understand "Fa" as being structured in a certain way, and derive your knowledge of its meaning from your knowledge of what its parts mean, then you must have changed your mind either about what "F" means or about what "a" means. (Note again how much easier it would be to understand Davies's paper had he used an example, as I just have, rather than formulating the point in complete generality, as he does on p. 149.)

This leads to a different objection to what Davies argued in part I. Suppose that the members of some group G speak the Evans language but do not have the kinds of dispositions just mentioned. Does the language they speak have 'semantic structure' or does it not? It must at least be a possibility that it does, if the structural constraint is to be different from the mirror constraint. But suppose further that what "Fa" means to them changes, without the meanings of any other sentences changing. (This is possible, since they lack the mentioned dispositions.) Does this new language have semantic structure? If not, does that show that their old language didn't? (If so, with what consequences for Davies's argument?) Or should we say that their new language does not have any semantic structure but that the old one did?

Formulate and discuss a version of the objection just given that involves acquisition of new expressions.

In §12, Davies worries that mere dispositions are not enough—compare Evans's insistence on 'full-blooded' dispositions—and so invokes causal and explanatory relations as well in giving a full account. In the case of Evans's language, for example, the claim would be that a 'full understander' would be in certain states S_a, S_b, S_F, and S_G that are "causally operative in the production of" their beliefs about what "Fa", etc, mean; moreover, their believing that "Fa" means that John is bald is explained by their being in the states S_a and S_F.

One might think of Fricker's goal in this paper as to try to answer Lewis's claim that there is "no promising way to make objective sense of the assertion that a grammar Γ is used by a population P whereas another grammar Γ', which generates the same language as Γ, is not" (L&L, p. 20)—and to do so on Lewis's own terms. (Fricker does not mention Lewis in this connection, though.) That is, Fricker accepts that semantic facts about a language must supervene on how it is used—on the linguistic abilities of its speakers—and that these abilities relate only to what whole sentences mean. (These are principles (α) and (A), on pp. 52 and 56, more or less). And yet, she wants to claim, each language has a semantic structure that is essential to it and to which speakers of the language bear some non-trivial relation. (These are principles (β) and (γ) on p. 52.) This is a bold, possibly even 'heroic', view. The strategy, to a large extent, is to effect a synthesis of Evans and Davies.

There's a lot of talk in this paper about 'a priori principles governing interpretation'. This is not language with which I find myself entirely comfortable. But one might simply think of Fricker as claiming that there are certain methodological principles to which we ought to adhere when constructing theories of meaning—never mind if they are a priori or empirical—and that among these is, e.g., a principle of compositionality.

Note the strategy here: If there's some aspect of a view with which you find it difficult to agree, see if you can re-interpret it. It may be that the bit with which you disagree isn't really that central.

Fricker begins, in section I, by rehearsing what is, very roughly, the metaphysical project elaborated by Davidson in "Radical Interpretation": explaining how semantic facts are determined by (grounded in) facts about the use of language. Section II introduces the problem with which she is concerned and argues that both Davies and Evans fail properly to vindicate the idea of semantic structure. In particular, neither of them establishes both that semantic structure is essential to a language and that its having that semantic structure supervenes on how it is used.

Section III formulates an argument that the assumptions mentioned above lead to Lewis's conclusion: that a theory of meaning for a language need not uncover structure. One aspect of Fricker's presentation that is somewhat different from what we have seen before is her characterization of what is distinctive of speakers of a language: that they have "a recognitional capacity for sentence meanings", the ability, when presented with a utterance in their language, to recognize what it means. Note that this is a psychological, cognitive matter. Fricker suggests that, if we take this to be all that is distinctive of speakers (as far as semantics is concerned), then we are liable to be led to Lewis's conclusion.

The main argument of the paper is in sections IV and V. Fricker's way out begins by noting that "...a [theoretical representation of speakers’ abilities] for a language L...will be neutral with respect to L's structure only if the correctness of its ascriptions of meanings to sentences of L is independent of facts about its structure" (p. 58). If sentences that have different semantic structures therefore have different meanings, then it will not, after all, possible for the sentences of two languages to have the same meanings but for one of them to be significantly structured and the other one not. Fricker gives three sorts of reasons in favor of this claim:

1. Unless there were "some form of Structural Postulate requiring that sentences be seen to be composed out of a stock of semantically primitive elements which make the same characteristic contribution to sentence-meanings in all their occurrences" (p. 59), interpretation would be too indeterminate. In particular, the interpretation of one sentence needs to be constrained by how other sentences are interpreted if we're to make any progress at all.
2. There are many sentences of, e.g., English that will never be uttered and yet that have determinate meanings. But if facts about meaning supervene on facts about use, then facts about the meanings of un-uttered sentences must supervene on facts about uttered sentences, and that can only be so if the language is compositional.
3. There is a connection between the meanings we assign to sentences and the contents we assign to speakers' beliefs. And the latter, typically, will themselves be structured, involving certain concepts we take the speakers to possess. This "will tend to ensure that OL sentences are not interpreted by ML sentences with a greater degree of structural complexity" than the speakers are capable of understanding (p. 60).

Fricker concludes that facts about sentence-meanings are not independent of facts about their structure" (p. 60).

Regarding (i): Perhaps one way to think of this is that, if we're to regard the speakers of the language as rational, then we need to be able to see them as reasoning. But, at least in the case of elementary logical reasoning, we need to be able to see the premises and conclusions of inferences as related in various ways, and that will require articulating structure. Can you develop this idea? The really crucial ingredient is that the structure must reflect not just syntactic recurrence but what we might call semantic recurrence.

Regarding (ii): it's important to remember that Fricker is trying to argue here not just for compositionality, but that the structural facts about the sentences of a language are essential to it. We've of course seen considerations about creativity and productivity before, and not everyone has wanted to draw the same conclusion Fricker does. What is it about how she's understanding (ii) that leads her to the stronger conclusion?

Regarding (iii), Fricker's conclusion is that these sorts of considerations will tend to restrict the degree of structural complexity ascribed to sentences. One might think this cannot help her argument. Why?

The arguments just mentioned are elaborated and interwoven in section V. Here, Fricker adapts Evans's proposal to generate a "transcendental" definition of the notion of a semantic primitive. The idea is that the semantic structure of a language must mirror the causal structure of its speakers' linguistic abilities: If, but only if, speakers can project meaning in a certain way should we regard their sentences as being structured. This leads directly to the question whether different speakers' abilities might have different structures—perhaps some speakers of English can project meaning in a certain way, while others cannot— which is what leads to Lewis's pessimistic conclusion. (See pp. 63-4.)

The "immanent/transcendent" language is used in this connection by Evans, who borrows it from Quine. One might think of an 'immanent' definition of F-ness as saying what it is for a theory to treat something as F; a 'transcendent' definition says, rather, what it is to be F, which will then have implications for how a theory ought to treat certain things (namely, as F, if they in fact are F). In our case: It's clear enough what it is for a theory to treat a sentence as structured (that's the immanent bit); what's wanted is an account of what it is for a sentence to be structured (the transcendental bit); if a sentence is structured, then of course a theory should treat it as such. In some cases, one might think, only an immanent definition is possible. For example, we can say what it is for a mathematical theory to treat something as an axiom. But it’s not at all clear that it’s even coherent to ask what it is for something to be an axiom, independently.

Fricker's response is on p. 64: She wants to insist that, if speakers understand a sentence to have different structures, then they cannot understand it in the same way. This is because it will be implausible to hold that the beliefs these speakers associate with the sentence deploy the same concepts. So Fricker concludes that "...the initially plausible thought", which we find in Davies, that "there could be structure in a language, though its speakers were blind to it, is wrong" (p. 65).

I take Fricker's point here to be a version of one Davies makes in response to an objection. Consider an unstructured sentence "Bugmoving" and its structured counterpart "Bug moving". It might be that the two sentences have the same truth-condition, in the sense that the conditions in which the former is true are exactly the conditions in which the latter is true. But one might nonetheless want to say that the meaning of the latter sentence involves predicating movement of a particular bug, whereas the meaning of the former does not. The structure thus turns out to be present in the meaning of the sentence itself and to reflect the conceptual resources that speakers of the language deploy when they use this sentence. The upshot is that, even though you only need to know what a sentence means to understand it, you can't know what the sentence means unless (in some sense) you appreciate its structure (p. 61).

Can such a tight link between semantic structure and the conceptual capacities of speakers be maintained? This would make a good topic for a final paper, though one could not really hope to answer this question in one paper. But one could try to make some progress. One might even make a bit of progress in a Canvas post....

Antony's main goal in this paper is to argue that semantics should be understood as inextricably linked to psychology: to questions about what speakers actually come to know about their languages when they learn to speak (and how they come to know it). She distinguishes three sorts of alternatives to her position: Platonism, represented by Michael Devitt, Jerrold Katz, and Scott Soames; Instrumentalism, represented by John Foster and Donald Davidson; and Reconstructive Rationalism (my term), represented by Sir Michael Dummett and Crispin Wright. (The relevant paper of Wright's is listed as optional.)

The discussion of Instrumentalism and Reconstructive Rationalism, which dominates section I, mostly consists in playing them off against one another. The latter view insists that semantic theory should be concerned with speakers' knowledge, but only with a "systematized" (Dummett) or "idealized" (Wright) version of such knowledge, not with the knowledge actual speakers possess. The question Antony presses is how this is supposed to be different from claiming that semantics aims to articulate knowledge that would suffice for competence (as on the Instrumentalist view).

Antony carefully reconstructs Wright's reasons for thinking we do need to attribute semantic knowledge to speakers. There are two. The first is compositionality. Wright takes the philosophical issue to be to secure the rationality of our 'projections' of meaning to novel sentences. To do that, we need to explain how knowledge of the meaning of a novel sentence could be arrived in a rational way. The second is familiar from Fricker: Meaning must, Wright suggests, be grounded in the way language is used by actual speakers, that is, by how they understand their language. This criticism applies most obviously to Platonism: Surely the facts about (say) English simply cannot be completely independent of facts about how English is understood by actual speakers (see pp. 184-5).

To put this slightly differently: Wright proposes to take seriously the idea that we know what novel sentences mean (when we hear them). The question to which he wants an answer is: How could that amount to knowledge (as opposed to merely true belief, or a lucky guess)?

Antony then turns to critique. She notes (on p. 185) that Wright keeps insisting that there are certain explanatory purposes for which we need to attribute semantic knowledge to actual speakers. If that were not so, then the attribution would be merely heuristic (and Reconstructive Rationalism would be no different from Instrumentalism). But, Antony then suggests, "...the rational reconstruction of linguistic meaning cannot explain the rationality of human language use if the posited linguistic structure is not available to speakers" (p. 188). That is: Just saying what an idealized speaker might know that would make their use of language rational cannot show that ordinary speakers' use of language is rational. But that was what we were supposed to be explaining.

Suppose Wright were to respond that the project was not to show why our actual projections of meaning are rational, but only to show that they could be rational. How should Antony reply?

Is there some other explanatory goal that Wright's "idealized epistemology of understanding" might serve? (You can assume that Antony's presentation of Wright is accurate.)

Antony goes on to argue that the idealizations inherent in Reconstructive Rationalism are difficult to reconcile with the professed goals of its proponents. There is, she says, a "tension between, on the one hand, appealing to human capacities in order to justify features of meaning-theoretic projects, and on the other, ignoring the actual nature of those capacities" (p. 188). The original sin here is the way Davidson motivates compositionality combined with his own refusal to take semantic knowledge seriously. The problem this observation poses for Reconstructive Rationalism is supposed to be that it is hard to see why, if we are not going to idealize away from the finitude of the language-learner—if we did, then compositionality would be unmotivated—we should think it justified to idealize away from any of the other circumstances under which language is, in fact, acquired by human beings. That is, the question is: Which idealizations are legitimate and which are not? And why?

I take Antony's point to be something like this. What motivates the principle of compositionality is the fact that actual speakers are actually able to interpret novel sentences. Presumably, there is some answer to the question how they do it. Either we think they do so in a way that makes their appreciation of the meaning of the new sentence rationally justified or we do not. If we don't, then it's unclear why we should think that there is, as Wright supposes, any rational basis for such knowledge, even in principle. And even if there were, it's unclear why we should care. But if we do think that actual speakers' appreciation of the meanings of novel sentences is rationally justified, then why not just ask how people actually come by knowledge of the meanings of novel sentences? Of what interest is the question how some idealized agent might do so? (As Antony notes, this criticism is very much of a piece with one that Quine makes of 'rational reconstruction' more generally.)

You might recall that Chomsky sometimes speaks of certain idealizations under which linguistic theory operates, e.g., speech communities are completely homogeneous, speakers don't make mistakes in interpreting sentences. Many of these idealizations are designed to focus attention on competence rather than performance. Could Wright appeal to the competence–performance distinction to justify his idealizations? Why or why not?

This point then morphs into a criticism of Instrumentalism. The discussion is directed at Quine's restriction on the data on which radical translation must be based, namely: behavior, construed in behavioristic terms. Even though radical translation is not thought of as a theory of language acquisition, but as an epistemologically-flavored thought-experiment actually concerned with the metaphysics of meaning, Quine's restrictions on the available evidence are motivated by claims about what sort of evidence is available, in principle, to a child learning language. It's because the child does not have access to, e.g., facts about the brain that these facts are excluded from the evidence that is available to a radical translator (see p. 192).

Although Antony's criticisms are directed at Quine, one might wonder whether they could equally be directed at Davidson. He too imposes various restrictions on the evidence that is available to a radical interpreter (what's held true by whom under what external circumstances). To what extent are his restrictions also motivated by assumptions about what 'evidence' is available to a language-learner (or to someone trying to verify that someone else speaks their language)?

Antony's conclusion is that the evidence that is allowed to a radical translator (or interpreter) is both wider and narrower than what is available to actual language-learners. Explain this point in detail, and with examples of your own.

Antony's mention of "cats" and "undetached cat parts" echoes a famous example of Quine's, in which he asks how we can come to know that some word in an alien language, "Gavagai", means rabbit rather than undetached rabbit part. The point is that undetached rabbit parts are found in the same places as rabbits. Whenever you are pointing at one, you are also pointing at the other. Etc.

Antony summarizes her argument this way:

Considered in the context of Quine's metaphysical goals, the idealization involved in permitting the linguist an unlimited amount of behavioural evidence appears concessive to the meaning realist; in fact, it is a slick piece of bait-and-switch. The cooperative tone distracts us from the fact that Quine has already begged the crucial question, by assuming that whatever physical facts metaphysically determine meaning must be identical with the physical facts that constitute the evidence children have available to them during language acquisition. (pp. 192-3)

What does Antony mean here? What might be examples of physical facts on which semantic facts supervene that are not among the facts available as evidence to a child learning language? How, if there are such facts, might they, in some other (non-evidential) sense, be "available" to ordinary speakers interpreting one another? Here's a thought experiment that might help you get at a possible answer: Suppose that, as a matter of empirical fact, there were exactly 1000 concepts humans could possess, and that they were all innate, so that we were all born already possessing all of them and could never come to possess any others. How would that affect language acquisition and interpretation?

There is thus a common criticism of both Instrumentalism and Reconstructive Rationalism: "[T]he epistemic strategies of 'ideal' learners are of no theoretical value to the task of understanding human cognitive competencies if the idealizations abstract away from epistemic constraints that are in fact constitutive of the learning task confronting humans" (p. 193).

The key word in the preceding quote is "constitutive". As mentioned earlier, linguistic theory (like all sciences) idealizes in various ways. So what exactly does Antony mean when she talks about 'epistemic constraints that are constitutive of the learning task'?

Section II of the paper turns to questions about tacit knowledge. Antony begins (on pp. 195-8) by rehearsing some of the reasons people have found tacit knowledge puzzling. She then argues (on pp. 198-200, and largely following Wright) that the strategy pursued by Evans and Davies has a fatal flaw: It purports to justify the attribution of tacit knowledge entirely on the basis of an isomorphism between the structure of a semantic theory and the structure of a speaker's abilities: But "...isomorphisms are cheap: the mere fact that the formal structure of a particular theory can be projected onto some structure of causally related states is not enough to make it true that the set of states actually embodies the theory", that is, that those states carry the sort of information that the theory does (p. 200).

By contrast, Antony insists, what we want is for the causal processes that underlie the understanding of novel sentences to be sensitive to information about the meanings of sub-sentential constituents. And she proposes that we can have that only if we take seriously the idea that there are states in the speaker that encode the very information articulated by a semantic theory, and that these states interact causally in ways that are sensitive to that information, giving rise to new states whose informational content depends upon those of the states that generate them. Accepting those assumptions means accepting a strongly 'realist' conception of the mind and of mental processes. But, or so Antony argues, only such a conception can give us a plausible account of tacit knowledge.

Here again, then, is the big point Antony is trying to make. Almost everyone we have read seems willing to accept something like the following reasoning, which Fricker makes explicit: Whatever the truth about what expressions of natural language mean, that truth must somehow be knowable on the basis of the sort of evidence that's available to a child learning language, or to an ordinary speaker who is just trying to make sure that their conversational partners really do mean what they seem to mean. So, in some sense, meaning must be 'manifest' in linguistic behavior (as Dummett famously put it). Different people seem to have different ideas what 'linguistic behavior' is (that is, what 'use' is), and then they argue about how determinate meaning really is, whether we can really think of words (and not just sentences) as having meaning, and so forth. But Antony thinks this whole set-up is a con game. Why?

More specifically, Antony writes: "The problem with Quine's constraints is not that they are severe, but that they are arbitrary" (p. 202). What does she mean?

When I was in graduate school, David Marr's book Vision was only a few years old. It was one of the main topics of conversation among those working in philosophy of mind. Most of the book describes the state of the art, at that time, in the science of vision, which had been transformed by the computational paradigm. But Marr also spends some time discussing the nature of computational explanation. Much in the spirit of Fodor's "Special Sciences", he articulates three levels of computational explanation, of increasing detail and specificity.

I had originally written "Most of the book reports the state of the art...", only later to realize that this could be a garden-path sentence, "book reports" being read as a noun phrase. This is (a) the kind of thing you need to watch out for when writing and (b) an interesting example of how we are not always sensitive to ambiguities in our own words.

At Level 1, one specifies which function some mental process calculates; at Level 2, one specifies the algorithm used to calculate that function; at Level 3, one describes how that algorithm is implemented in the brain. Note that multiple Level 2 explanations can correspond to any Level 1 explanation (there are lots of ways to calculate any given function); similarly, lots of (potential) Level 3 explanations correspond to any Level 2 explanation. So we have the typical phenomenon of multiple realizability.

Note that this distinction among levels applies to every explanation of a cognitive phenomenon that we might want to give in broadly computational terms. No matter what the phenomenon is, there is a function being calculated; there is a way that function is being calculated; and there is a way that calculation is carried out in the brain.

It's unfortunate that Marr labeled the levels as he did. One tends to speak of 'higher' and 'lower' levels of explanation. These are inverse to the way that Marr numbers the levels. So one has to be careful with such language here.

Recall Chomsky's distinction between E-language and I-language. E-languages (thought of as languages 'in extension') pretty clearly correspond to Marr's Level 1. Where would I-language (thought of as languages 'in intension') sit?

In an earlier paper (listed as optional), Peacocke had identified a level of explanation between Marr's first and second levels, which he dubbed 'Level 1.5'. At this level, one specifies what information is utilized in calculating the function. Peacocke emphasizes (on p. 112) that Level 1.5 has the same sort of many-one relation to Levels 1 and 2 that they have to each other. Level 1.5 is thus distinct from Levels 1 and 2. Peacocke also argues (on pp. 112-3) that there are phenomena that are naturally explained at Level 1.5, and that explanations pitched at that level can be genuine causal explanations (meaning: it can be legitimate to say that such-and-such happens because the mechanism that produces that behavior draws upon certain information).

In the earlier paper, Peacocke argued that many explanations in cognitive psychology can be understood as pitched at Level 1.5. (He mostly gives examples from vision science.) So think about the sorts of explanations you have encountered in linguistics or cognitive science courses. At which of these levels might those explanations be found? Are some of them at other 'levels' not so far identified? (Peacocke suggests that there may well be other levels between Marr's first two.)

At the bottom of p. 113, Peacocke considers the objection that information-based explanations are simply 'reading back' patterns in the output. Much of what we have read recently has been concerned to answer the same objection, in the form: How is tacit knowledge of a theory of meaning (or truth) different from simply behaving as if one knew the theory? So Peacocke is aiming to answer that question, too, but in a way different from Evans.

The rest of the paper will articulate an answer in detail. But Peacocke suggests that we can already see that the objection is wrong-headed. Why?

Peacocke then proposes to explain what it is for a grammar to be 'psychologically real' (tacitly known) in terms of Level 1.5. The detailed analysis is presented at the top of p. 115. But the basic idea is simpler. Suppose that some linguistic theory explains some linguistic phenomenon in terms of some set of grammatical principles. Then those principles are 'psychologically real' (for a particular individual) if the mechanism that is responsible for that phenomenon (in that individual) draws on the information specified by the grammatical principles.

Peacocke gives a few simple examples of how this might work in practice. Can you give other examples from linguistic theory? How does Peacocke's account work in the case of semantics? What would be required for some truth-theoretic explanation of why "Snow is white" is true (for Fred) iff snow is white to be 'psychologically real' (for Fred)?

We can ask the same question another way. Peacocke considers a 'finite' language, not unlike the one Evans considers, on p. 117, though Peacocke's example concerns syntax. How would he explain what it is for a compositional theory to be psychologically real in the case of Evans's finite language? Note that Peacocke does not talk about 'patterns of acquisition and loss' the way Evans does. How should he think of the relevance of such phenomena?

Peacocke emphasizes that this sort of explanation does not specify the algorithm that draws on the information in question. In particular, he emphasizes that there is no requirement that the grammatical rules somehow be 'written out' in Mentalese (p. 118). The informational content of those rules has to be represented somehow—otherwise, it could not be 'drawn upon'—but explanations at Level 1.5 are neutral on how the information is encoded. This will become important (very important) later.

Peacocke notes that the contents of rules of inference cannot be 'written out' explicitly in this way, lest one run into a regress described by Lewis Carroll (yes, that Lewis Carroll) in the paper listed as optional. Since every system of computation must have some rules of inference, it follows that everyone needs to make sense, somehow, of information's being encoded other than by being 'written out'.

Quick summary of Carroll's argument. After losing the race to Achilles (thereby refuting Zeno's paradox), the Tortoise looks to him for instruction in geometry. Achilles explains that p, and also that, if p, then q, inviting the Tortoise to infer that q. But the Tortoise says that he does not see why q follows. Achilles proposes, as a general principle, that, if A, and if also, if A, then B, then B. The Tortoise agrees, and Achilles invites him to infer that q. But the Tortoise says that, while he agrees that (i) p, and also that (ii) if p, then q, and also that (iii) if both p and if p, then q, then q, he doesn't see how those three claims lead to q. At this point, the regress has begun. It will not help to add a fourth claim. At some point, one just has to make the inference. So there is a difference between claims (axioms) and inferences; the former cannot do the work of the latter.

On pp. 118-9, Peacocke discusses the question whether there must be a single state encoding the relevant information that is implicated in every explanation in which a given grammatical principle is involved. (Note the similarity to Evans's remarks about dispositions in the 'full-blooded' sense.) The argument is complex, but the basic idea is this: If there were two different states, whose contents were not somehow co-ordinated, then the content of one could change without the content of the other changing. That is: A different (changed) grammatical principle would be applied in some cases (e.g., perception of speech), but not in others (e.g., production of speech). Peacocke suggests that this is not really a coherent possibility.

I think the argument as Peacocke states it does not work. In particular, his claim that I(pd) is not sensitive to evidence that its content is incorrect could be questioned. What we know, at that point of the argument, is just that it is not sensitive to evidence coming from perception of sentences involving the relevant word (say). For all that's been said so far, it could be sensitive to evidence coming from production (e.g., responses to it, which need not involve hearing that word repeated). Moreover, even if that kind of disconnect seems odd, it's very specific to the example Peacocke considers. What should we say here?

On pp. 120-2, Peacocke first considers a general objection to what he is trying to do in this paper and then considers an alternative criterion of psychological reality: "that a principle of a grammar is psychologically real if it corresponds to a 'determinate complex of neural mechanisms'", a view at least suggested by Chomsky. Peacocke argues that such a view fails because we have not been told what counts as a 'determinate complex': Neural structures could be grouped together in various ways; to group them at the right level, we have to appeal to the information they encode.

On pp. 122-5, Peacocke considers the question whether his proposal is compatible with, or could apply to, mechanisms that use 'parallel distributed processing' (PDP), which was an early ancestor of the technology used in today's large language models. Referencing a simple model of word-recognition, Peacocke argues that such models can encode and deploy the same information that a more conventional sort of algorithm, which computed in some sort of language, would. He concludes "that general arguments in favour of PDP models should not be taken to foreclose the possibility of explanation of linguistic phenomena by an interesting psychologically real grammar" (p. 124).

Peacocke's argument here is best understood as defensive: There is no general obstacle to explaining the behavior of PDP-type models at Level 1.5. The difference between those models and 'classical' computational models is, rather, a difference at Level 2 (which is not, for that reason, uninteresting). It's a different question whether any given device does encode and deploy the sort of information specified by any given grammar. But the mere fact that the principles of that grammar are not 'written out' in the device does not, by itself, show that the informational content of those principles is not somehow encoded in, and used by, the device in question.

One way to understand what Peacocke is arguing in this section is: It has been too easily assumed that the psychological reality of grammars depends upon the employment of certain algorithms. But once we recognize the possibility of explanation at Level 1.5, we can see that linguistics need not be committed to any claims about the algorithms used in determining (say) grammaticality. It need only be committed to claims about what information is used in determining grammaticality. For that same reason, there need be no commitment to the claim that linguistic theories are explicitly encoded, in some language-like way, in the mind. In particular, computational explanations, at least if they are pitched at Level 1.5, are not committed to a 'language of thought'.

Peacocke argues on pp. 124-5 that this need not stop theorists from carrying out something akin to formal deductions from grammatical principles, and even as seeing those formal deductions as having a kind of explanatory status of their own. Can you explain why?

Note that the mere fact that the behavior of some PDP-model fits a particular linguistic theory does not imply that the theory is 'real' for it. Explain why not. What would it take to show that the theory was 'real' for the model? (Careful: As Peacocke repeatedly emphasizes, he is giving a constitutive criterion for psychological reality, not a practical guide for psycholinguists. That is: he's interested in a metaphysical question, not an epistemological one.) Explain further why it would be fallacious to infer from the fact (if it were one) that the theory was not 'real' for the model that it is not 'real' for humans.

Searle is concerned in this paper to argue against a certain conception of the literal meaning of a sentence and in favor of a different conception. He describes his target as follows:

Every unambiguous sentence...has a literal meaning which is absolutely context free and which determines for every context whether or not an utterance of that sentence in that context is literally true or false. (p. 214)

This view obviously allows for certain kinds of context-dependence, such as that exhibited by indexicals like "I" and "yesterday", 'bare' demonstratives like "this" and 'complex' ones like "that cat", definite descriptions like "the cat" (which cat?), and tense. What makes these cases relatively unproblematic is that the context-dependence is "already realized in the semantic elements of the sentence..." (p. 210). The sentence itself, as it were, tells us that there is some work for context to do in determining what is said, and the 'literal' meaning of the sentence includes information about how context affects what is said, that is, about how to 'resolve' the context-dependence. For example, the literal meaning of an utterance of "I am tired" tells us that an utterance of that sentence is true iff the utterer is tired at the time the utterance is made.

Searle's preferred view is:

For a large class of unambiguous sentences...the notion of the literal meaning of the sentence only has application relative to a set of background assumptions. The truth conditions of the sentence will vary with variations in these background assumptions; and given the absence or presence of some background assumptions the sentence does not have determinate truth conditions. These variations have nothing to do with indexicality, change of meaning, ambiguity, conversational implication, vagueness or presupposition as these notions are standardly discussed in the philosophical and linguistic literature. (p. 214)

What makes this view different from the traditional view is Searle's insistence that these "background assumptions" cannot, even in principle, all be "specifi[ed] as part of the semantic content of the sentence, [since] they are not fixed and definite in number" (pp. 214-5). Most importantly, no such specification can ever be complete: No matter how precisely we try to specify the "background assumptions", there will always be other background assumptions in play that can, by clever construction of examples, be brought to our attention and varied, so as to lead to variation in truth-conditions. Note the difference from the case of "I": In that case, we can say quite precisely how the context affects what is said; the word "I" refers to the person who utters it. In the kind of case Searle has in mind, however, we cannot do anything like that. We cannot, that is, specify a mapping from features of the context in which the sentence is uttered to truth-conditions. Such a mapping is how Searle is thinking of the literal meaning of the sentence (as opposed to the meaning of a particular utterance of it).

Formally speaking, the view Searle is criticizing might regard the meaning of a sentence as a function from contexts to truth-values. So, e.g., the literal meaning of the sentence "I am tired" maps a context to True iff the speaker in that context is tired. In this setting, Searle's point takes this form: Even if we assume that we understand how context picks out a definite cat and a definite mat, there is no function from contexts to truth-values we can specify as the literal meaning of "The cat is on the mat". (Searle's paper predates the now-common distinction between contexts of utterance and contexts of evaluation. But that distinction does not seem relevant to his argument.)

One might wonder whether that is so. One might concede that Searle has shown that it is very difficult to specify such a mapping, and that we do not at present know how to do so. But that is different from showing that no such mapping could be specified. What reasons might someone give to think that there must be some such mapping, even if we do not know how to specify it? What reasons does Searle have for thinking that there is no such mapping (and so that our inability to specify one is not due to ignorance)?

In fact, it is not so clear that "I" always refers to the person who utters it. One kind of case involves, for example, ChatGPT, when it responds to an inquiry by saying, "I am a large language model". (An older example to the same point involves a scale with a sign on it that says "I tell your weight".) Another kind of case involves someone reading a statement on someone else's behalf: "I am sorry I cannot be here today", begins the statement. What might we say about such cases? Are they an example of the phenomenon in which Searle is interested? Can you think of examples that throw similar doubt on the most obvious specifications of the meaning of "now" (time of utterance) and "here" (place of utterance)?

What exactly does Searle mean to exclude when he says that "These variations have nothing to do with indexicality, change of meaning, ambiguity, conversational implication, vagueness or presupposition as these notions are standardly discussed in the philosophical and linguistic literature"?

Regarding presupposition: Searle seems to be operating with something like the 'logical' notion of presupposition that originates with Strawson. One might even seem some similarity between Searle's view and the notion of 'pragmatic' presupposition that is due to Stalnaker. In that sort of setting, Searle's question becomes whether we can say, precisely, how the 'common ground' affects what it said in an utterance of "The cat is on the mat".

Searle's terminology is liable to confuse: The 'literal' meaning of an utterance is sometimes taken to be what is 'strictly and literally' said, as Grice puts it. So as to avoid confusing Searle's notion of 'literal meaning' with this one, I shall often use the term 'standing meaning' to refer to what Searle has in mind: the context-independent meaning of the sentence as such. And I shall tend to put Searle's point not in terms not of sentence meaning but in terms of Grice's notion of what is said. Searle's point, so understood, is that what is said, even in the most literal utterance of a very ordinary sentence (meaning: not metaphorical, not ironic, etc) is not always (even often?) completely determined by "stable" features of the sentence that is uttered, but also depends upon background assumptions whose effect on what is said cannot, even in principle, be completely specified.

Yes, to some extent, this does involve enforcing the distinction between contexts of utterance and contexts of evaluation.

The general strategy of argument is to consider a sentence that seems to have a perfectly definite meaning. In practice, we focus on one word and the contribution it is making. (In Searle's example, the problematic word is "on". What exactly does it require for the cat to be on the mat?) We then consider certain unusual contexts and note that it is simply not clear whether to regard a perfectly literal utterance of the sentence as true or false in that context. Indeed, there will be ways of developing the example so that a perfectly literal utterance of the sentence would be true, and there will be other ways of developing it so that such an utterance would be false. Hence, what the truth-condition of the utterance is depends upon exactly which "background assumptions" are in play. And, once again, the exact nature of this dependence—how precisely 'background assumptions' affect literal meaning—cannot be completely specified.

To check your own understanding, it is worth trying to construct examples similar to Searle's for sentences other than the ones he considers. Color words make for one good example. For contrast, one might consider comparative adjectives like "tall". (Recall that such adjectives always invoke something like a comparison class: At 5'11", Jewell Lloyd is a tall woman but not a tall basketball player, since over half of the players in the WNBA are 6 feet or taller.)

Recall Strawson's remarks about "and". Might that be regarded as an example of the phenomenon in which Searle is interested?

On pp. 218-9, Searle argues that the phenomena in which he is interested affect not just the meanings of utterances but intentional mental states, as well. What we believe when we believe that the cat is on the mat is, he suggests, every bit as dependent upon 'background conditions' as the meaning of the sentence is. Moreover, at the very end of the paper, Searle argues that a version of this point applies even to perceptual content: When I see a cat on a mat, my visual state has a 'veridicality condition' much as my belief has a truth-condition; but what is required for my visual state to be veridical (true) depends upon background conditions in a way, Searle claims, that we cannot completely specify.

Searle does not develop these points in detail. Can you fill in some of the details? Once you have done so, does Searle's point seem correct? Does someone who believes that the cat is on the mat in, say, the floating in outer space case believe the same thing as they would in the case described on pp. 213-4? How can we tell? If we do think that they believe the same thing, would it be sufficient for a semantic theory simply to say, in effect, that "The cat is on the mat" expresses that very belief (even though its 'application' will itself be subject to the sorts of variation Searle describes)?

The question we are now starting to discuss is: How do the sorts of phenomena Searle has noticed affect the conception of semantic theory with which we have been operating so far? What should we think of a competent speaker as knowing about such ordinary sentences as "The cat is on the mat" in virtue of being a competent speaker? What might we say about what competent speakers know about the meaning of the word "on", as it occurs in sentences such as "The cat is on the mat"?

Grice writes in "Logic and Conversation":

In the sense in which I am using the word say, I intend what someone has said to be closely related to the conventional meaning of the words (the sentence) he has uttered. (p. 25)

Grice is of course aware that contextual factors may play various sorts of roles in determining what is said, with respect to words like "I" and "you" and "this". But the fixed, stable meanings of the words used are supposed to play an especially important role in determining what is said when someone utters a sentence.

We have more or less been following Grice in this respect, and assuming that we can derive the truth-condition of a sentence from semantic axioms governing its component parts. But Carston, in this paper, challenges this sort of assumption. She is particularly concerned with the requirements of a psychologically adequate account of linguistic competence. And she argues that, if we want such an account, then the difference between what is said—which she calls the explicature associated with an utterance—and what is implicated is much less stark than Grice seems to suppose.

Carston begins with an example similar to one Grice might have given:
 A: How is Jane feeling after her first year at university?
 B: She didn't get enough units and can't continue.
Carston notes that it's clear enough that there is some work context has to do to determine a truth-condition for B's utterance. For example, "she" needs a referent. It's also clear that "Jane isn't feeling so great about it" is an implicature. But what of the connection between her not passing enough classes and not being able to continue? and continue what? Are those part of what is implicated or part of what is 'strictly and literally' said? The first lesson to learn from this paper, then, is that it is not at all clear how we should draw the distinction between what is said and what is meant.

Do you have any initial 'intuitions' about how to answer these questions?

In section 2, Carston reminds us of some of the general principles that govern the notion of implicature. She also mentions some (then) newer ideas, deriving from Dan Sperber and Deirdre Wilson, concerning exactly how implicatures are 'calculated' in practice. One of these is extremely important: that the 'explicature' (what is said) has to be logically independent of the implicature, i.e., that there should be no redundancy between the two.

Carston asserts on p. 158 that this principle of 'functional independence' implies certain answers to the questions raised about the example quoted above. But she doesn't really explain why. Can you?

Can you think of any counter-examples to the principle of functional independence? I.e., cases where something is clearly an implicature and yet implies what is 'strictly and literally' said?

In section 3, Carston argues that an utterance of "A and B" can assert the existence of all sorts of different relations between A and B: temporal, causal, rational, and so forth. Given how many such possibilities there are, the view that "and" is lexically ambiguous is implausible. Rather, Carston suggests, in some ways echoing Grice, "... these meanings are the result of the way our minds organise information into connected scenarios or scripts..." (p. 159). And, indeed, Carston agrees that some kind of 'pragmatic process' is involved here.

However, Carston insists that the fact that some kind of 'pragmatic process' is involved does not imply that the temporal, etc, content is implicated rather than 'explicated' (part of what is said): After all, pragmatic processes enter into the resolution of ambiguity and the reference of demonstratives. Carston then proceeds (on pp. 160-2) to sketch an account of how 'pragmatic processes' might 'enrich' the linguistically specified content so as to include the temporal, etc, aspects. The thing to note is how much has been added in (11): that he used the key to open the door, and that he did so after it was given to him.

One might wonder whether the fact that he used her key to open the door is really part of what is said, even if one does think that his opening the door after she gave him the key is. How does Carston propose such questions should be answered? (Hint: See the last question.) How good is her argument?

More specifically, Carston opposes what she calls the "linguistic direction principle", which claims that any "explicating" process must be triggered by something in the linguistic form that calls for it. In section 4, she suggests that the linguistic direction principle is motivated by two assumptions: (i) that "what is said" must be truth-evaluable and (ii) that the only contribution that context can make to determining what is said is whatever needs to be done to get us something truth-evaluable. Anything beyond that is an implicature. So, e.g., the reference of demonstrative has to be determined, since otherwise one has said nothing truth-evaluable; but one does not need to find any e.g., causal relation for "and" to express, since truth-functional conjunction is already truth-evaluable.

Our main interest will be in the sorts of arguments Carston gives that, e.g, the temporal aspect of certain uses of "and" must be part of what is said. There are four of these:

• Functional: As mentioned already, Carston argues that the explicature and implicature must play different roles; in particular, Carston claims that the implicature cannot logically entail the explicature. For all the work Carston wants this principle to do—she invokes it repeatedly—she does not give much argument for it, and it does not seem overwhelmingly plausible on its face. So feel free to set it aside if you wish, though if anyone has a good idea about how to justify this restriction, please do say so.
• Relevance: Carston argues (in section 5, which you can skim or skip) that relevance plays a central role in communication. In particular, considerations of relevance enter not just into the determination of implicatures but into the determination of what is said, e.g., the resolution of ambiguity, the determination of the reference of demonstratives, and so forth. Claims about relevance thus fuel some of Carston's claims about what is part of the explicature. Here, though, Carston mostly gestures at work by Sperber and Wilson, so these arguments may be difficult for us to evaluate. (I confess that I find the discussion of how relevance and principles governing efficiency constrain the explicated content pretty hand-wavey.)
• Negation: Carston develops this test in section 6. Consider Grice's gas station case. Joe says, "Yo, Bill, I'm out of gas!" and Bill answers, "There's a gas station around the corner". Grice says that Bill implicates that the station is open and has gas to sell. Suppose Fred knows that the station is closed. Then it seems clear that Fred cannot say to Joe, "Bill's wrong", thereby meaning that the station is closed. As it's sometimes said, negation cannot "target" the implicated content.
• Conditionals: There is something similar to be said about conditionals. I'll leave it as an exercise (i.e., feel free to do this in your response) to explain how the "conditional test" is supposed to work.

Can you think of any other tests to which Carston might appeal?

Is it true that one can't 'deny' an implicature in the way Carston has in mind? Can you think of examples where something like that does seem to happen? (Carston mentions a couple cases one might consider in footnote 11. But, in those cases, people have used the negation test to argue that metaphorical and ironical meaning are part of what is said. So one would like an example in which it is relatively uncontroversial that the thing being denied really is an implicature.)

Carston uses the "negation test" and the "conditional test" to argue, in a variety of cases, that the explicature is richer than one might have supposed. There is a whole battery of examples here. Which of these seem to you to be the strongest? which the weakest? and why? What strategies do you think might be available for resisting the conclusion that Carston wants to draw: that pragmatic processes play a surprisingly large role in determining what is said?

It's not always clear that Carston's various tests deliver the same result. She argues based on functional considerations that "The park is some distance from where I live" says something non-trivial. But how does it interact with negation and conditionals?

Stanley's goal in this paper is to defend a form of what Carston calls the 'linguistic direction principle': that the kinds of contextual effects she discusses are 'triggered' by elements in the syntactic structure—the 'logical form', or LF, in the then-current sense of that term—of the sentences in question. This is a theoretical principle, but defense of it necessarily has to be case-by-case, and of course Stanley is not going to discuss every possible case. Rather, his strategy is to explain how the principle can be defended in a wide range of cases.

Some approaches to syntax and semantics do not make use of anything like 'logical form' in this sense. (In particular, the approach pioneered by our own Pauline Jacobson does not.) Stanley suggests in note 13 that, while his arguments would need to be reformulated, if one likes this kind of view, doing so will not affect the central arguments of the paper. He says a bit more about this in the paper "Semantics In Context" that is listed as optional for "Making It Articulated". I'll tend, therefore, to ignore this disagreement in what follows.

Section I lays out some of Stanley's background assumptions. He notes that the terms "semantic" and "pragmatic" did not then (and still do not) have agreed meanings. He therefore clarifies that, as he uses the term 'semantic', semantic interpretation is not limited to the interpretation of the fixed, context-independent meanings of the words used. It also includes the 'resolution' of context-dependent elements present in the sentence itself. Importantly, though, "Extra-linguistic context enters [into semantic interpretation] only when called upon by a linguistic rule governing an element" of the sentence (p. 395).

It is consistent with this claim that the result of semantic interpretation is not a complete proposition—not something that even could be evaluated for truth, not something that has truth-conditions—and that additional forms of pragmatic interpretation are required before we arrive at 'what is said'. The claim for which Stanley is arguing, however, implies that no such additional interpretation is required: "[I]f all effects of extra-linguistic context on the truth-conditions of an assertion are traceable to logical form, then the result of semantic interpretation...will be the truth-conditions of the assertion..." (p. 395).

As Stanley notes later, there is another role that context plays: in disambiguation. In cases of structural ambiguity (e.g., "Someone loves everyone"), this means deciding what the logical form of the sentence is. So there is, as it were, a pre-semantic role for context, as well as a semantic role. But that is not at issue in this debate.

As Stanley emphasizes, on his view, the ways that context can affect the meaning of an utterance are "highly constrained" (p. 396). What sort of worries might one have about a view that allowed context to affect interpretation in less constrained ways? (Hint: The greater the number of possible answers to a question, the harder it will be to find the right one.)

Stanley emphasizes that one cannot just posit extra structure in a sentence in order to save the 'linguistic direction principle'. The claim that a sentence has a particular 'logical form' has to be independently motivated and compatible, at least, with what our best syntactic theories tell us about the structure of the sentence. Stanley's opponents are aware that one could posit additional structure, but they regard doing so as ad hoc and unmotivated. In response, Stanley will be arguing that such additional structure can be argued for on familiar sorts of syntactic grounds.

It will be essential to Stanley's argument that some of the elements of the 'logical form' whose meaning is decided by context are 'covert', that is, not present in the 'surface' form of the sentence. It is not controversial that there are such covert elements in the syntax of many ordinary sentences. These include 'traces', which function like bound variables, and which are left when expressions move out of their original position, and PRO, which typically serves as the unpronounced subject of a non-finite clause. E.g., compare "Alex would like Sam to eat" with "Alex would like to eat". In the latter, the verb "to eat" has no pronounced subject, but (for various reasons) it needs one. Its subject is therefore typically taken to be PRO, which in this case functions like a bound pronoun, referring to Alex. So the sentence means: Alex would like Alex to eat. Stanley will be arguing that there are similar unpronounced elements in the sorts of sentences on which Carston et al. tend to focus.

See the Wikipedia article for more information on the motivation for 'empty categories'.

Section II discusses 'non-sentential assertion', as when one person says to another "Nice dress". That is not a complete sentence, and yet a complete proposition seems to be expressed. So this might look to be a case in which there are aspects of the expressed proposition that correspond to nothing in the uttered words. Stanley argues that, in typical cases, this is an illusion. But you can skip this section: We will not discuss it in class.

In section III, Stanley begins his discussion of examples that might seem to show that context affects truth-conditions in ways not permitted by the 'linguistic direction principle'. These are examples in which there is something present in the truth-condition of an utterance (what is said) that, apparently, corresponds to nothing in the sentence itself. For example, an utterance of "Alex is ready" might mean that Alex is ready to take an exam (rather than to go to a party). But nothing in the sentence itself, one might think, specifies what Alex is ready for.

For the most part, Stanley considers views on which this 'extra' material is required by the meaning of a particular word. Thus, the meaning of the word 'ready' might specify that the thing for which someone is ready has to be provided somehow: if not explicitly, then by context. We'll consider how Stanley's arguments apply to views like Carston's next session.

The argumentative strategy that Stanley uses has come to be known as the 'binding argument'. The basic idea behind the argument is that, if the meaning of some expression E can vary not just with context but with the value of what is, in effect, a variable introduced by some quantifier-like element Q, then there must actually be a covert variable associated with E that is bound by Q. (This is the 'binding assumption' mentioned on p. 412.) I'll explain this below, using the notation of ordinary logic.

You need not worry, for our purposes, about how Stanley's argument applies to what he calls 'narrow indexical analyses'. As mentioned above, you can skip or skim those parts of his discussion. But this extension of the argument is important in other contexts. See Stanley's paper "On the Linguistic Basis of Contextualism", listed as optional.

Stanley considers a number of examples, but I'll focus on the case of comparative adjectives (which Stanley discusses on pp. 418-9). It is a common observation that, if someone utters a sentence like:

(1) Jewell Lloyd is tall.

what they mean will depend upon the answer to the question: tall for a what? At five foot eleven, Lloyd is tall for a woman, but not tall for a professional basketball player. (Over half the women in the WNBA are six feet or taller.) To say that someone is tall is, it would seem, always to say that they are tall for an F (or just: a tall F), for some appropriate 'comparison class' F.

The most common analysis today differs, treating such adjectives in terms of 'degrees' (in this case, degrees of height). That difference will not matter for our purposes—though, if you are familiar with such accounts, you might want to think about how Stanley's analysis would be developed in terms of them.

So the question arises: Where does this 'comparison class' come from? Sometimes, it is explicitly provided, as when we say, "Jewell Lloyd is a tall woman". But in (1), there does not seem to be any element of the sentence to which the comparison class corresponds. That is the sort of thought that motivates 'unarticulated constituent' analyses, which go roughly as follows:

(UC) "is tall" is true of x, in some context C, iff x is taller than at least half the people in the contextually salient comparison class in C.

It is not a problem for this analysis that different occurrences of the word "tall" might be associated with different comparision classes, as in "Jewell is tall, and Michael is tall, too". The 'contextually salient comparison class' can be the one for that particular occurrence of "tall", rather than a single comparison class provided once and for all.

So the idea is that the comparison class is 'provided by the context', but not because there is some element of the 'logical form' to which it corresponds. The objection to the analysis given by (UC) is that the 'comparison class' is bindable, as in this example:

(2) Every team has players who are tall.

This sentence could be used to say that every team has players who are tall people, or tall in some other uniform way. But (2) could also be used to say: Every team has players who are tall for the sport that team plays. To account for this reading, Stanley posits a 'covert argument place' for the comparison class, which is filled by a function. So we get something like:

(3) Every team x has players who are tall F_x's.

or, to present it in terms of logical notation:

(3') ∀x{x is a team → ∃y[y is a player & y is on x & y is tall for an F_x]}

Here, F_x is a function from people (more generally, objects) to sets (or perhaps properties) that can function as comparison classes. The reason F_x needs to be there is that, otherwise, there is nothing there for the quantifier ∀x to bind at the relevant location. But the comparison class is, as it were, varying with the team, so there needs to be a variable there to effect the varying. What context does, on this analysis, is choose the function F_x: In this case, it is the function "the sport played by x", but in other cases it would be something else.

To check your understanding: Explain in similar terms one of Stanley's other examples.

Can you come up with a context in which our old friend "Someone loves everyone" could be used to make Stanley's point about quantifier domains? I.e., in which the domain of the second quantifier varies, depending upon which person we are considering? If you have to use a different verb, that will be good enough.

One question Stanley doesn't really address is which variables the function in question takes as arguments. Consider the other reading mentioned: that every team has players who are tall people. Is the logical form still "Every team x has players who are tall F_x's" or is it "Every team x has players who are tall F's"? And what about the variable y? Is it present as an argument to the function? There are thus some open questions about how this analysis ought to be developed.

On pp. 424-6, Stanley considers a response. Perhaps words like "tall" have a variable attached to them when they occur in sentences like (2) but not when they occur in simple sentences like (1). One might just dismiss such a view as ad hoc and unmotivated, but Stanley argues that there is a more serious problem, concerning ellipsis. Imagine that we are at a banquet at which several athletes are getting awards, including Stephen Curry. Suppose someone has been saying that only tall athletes ever win awards. In response, I say:

(4) Stephen Curry is short. In fact, everyone is.

Then this can mean that Curry is short for the sport he plays (basketball, in his case), and that everyone (getting an award!) is short for the sport they play. This would not be possible if there were no bindable variable in "Stephen Curry is short" itself. In that case, "Everyone is" would have to mean "Everyone is short", as compared to whatever group was in question when I said Curry was short. Of course, (4) could also mean that. As Stanley notes, this is the so-called 'strict/sloppy' ambiguity that arises with overt pronouns. His point is that (4) exhibits the same ambiguity, which strongly suggests that a pronoun-like element is present, even if unpronounced.

Vary the example. Suppose Alex had said "Stephen Curry is short", and I had added "Indeed, everyone is". Stanley's argument seems to require that Alex meant (or I understood Alex to mean): Curry is short for the sport he plays. But, one might think, Alex might just have meant: Curry is short for a basketball player. How serious is this problem?

Carston mentions a lot of different examples. To which of these do Stanley's arguments apply? To which do they not apply?

This paper defends the "binding argument" against some criticisms.

As Stanley makes clear in the introduction, his larger goal is to defend a form of what Carston calls the 'linguistic direction principle': "[A]ll the constituents of the propositions hearers would intuitively believe to be expressed by utterances are the result of assigning values to the elements of the sentence uttered, and combining them in accord with its structure" (pp. 150-1). In particular, there is no place for 'enrichment' of the sort to which Carston appeals. Stanley's view, for which he argued in "Context and Logical Form", is that context can affect what proposition is expressed by an utterance (what is said) only by affecting the interpretation of elements of the syntactic structure of the sentence uttered.

The first section summarizes the binding argument and places it in the context of the sorts of arguments often given in linguistic theory for the existence of "hidden" or "unpronounced" elements. If those sorts of arguments are unfamiliar, don't worry about it. The point is that there's nothing unusual about the kind of argument that Stanley is giving: Linguists give such arguments all the time. (Indeed, many of Stanley's arguments have the form: There are other phenomena that are typically explained in precisely the ways Carston et al. think are too implausible to consider.)

To review: What the binding argument is supposed to show is that the 'added' material that is allegedly provided by 'enrichment' actually corresponds to some unpronounced variable (or pronoun) that is present in the 'logical form' of the sentence. The evidence for this is that the variable in question can be bound, i.e., that its value can be made to depend upon some other 'operator' in the sentence. That is, the claim is that, in

(2) Every student answered every question.

the domain of "every question" can depend upon the student. To account for this reading, there needs to be a 'domain variable' associated with "every question" whose value depends upon the student. So, on this view, the universe of discourse is not set once and for all, but can vary for each quantifier, and the universe for one quantifier can depend upon the values of variables introduced by other 'operators' in the sentence.

Consider the sentence "The President (of the United States) will one day be a woman". Explain how the ambiguity of this sentence can be used to argue that there are temporal variables present in the sentence.

The second section is designed to undermine one possible response to this argument: that the 'unarticulated constituents' are provided by a process similar to ellipsis. This suggestion was made by Wilfrid Sellars in the paper listed as optional. There certainly is such a thing as ellipsis, and the example concerning "Seven is" is a standard example of it. But Sellars claims that the same sort of phenomenon is at work in so-called "incomplete descriptions", such as "the table". What makes such descriptions 'incomplete' is the fact that there is obviously not just one table, so one wants to know which table.

Incomplete descriptions pose a serious problem for Bertrand Russell's so-called 'theory of descriptions'. (Descriptions, in this sense, 'definite' descriptions, are expressions of the form "the F", with F singular.)

• On Russell's theory, sentences of the form "The F is G" mean something like: There is one and only one F, and it is G. However, there are obvious problems with this analysis in most ordinary cases. "The table is dirty" cannot, it might seem, mean: There is one and only one table, and it is dirty. Otherwise, it would obviously be false.
• In the paper listed as optional, Strawson had argued, in part for this reason, that we should think of descriptions as being more like names. On Strawson's view, "The F" is used to pick out a particular object, much the way a name would be. So "the table", on his view, is just used to refer to some salient table, much the way "that table" would be (though Strawson recognizes that there are differences between these cases).
• Sellars, by contrast, wants to defend Russell by arguing that incomplete descriptions are not really incomplete. They are, rather, another example of ellipsis: Linguistic material sufficient to determine a unique table has simply been elided. A typical utterance of "The table is dirty" will actually mean something like: The table in this room is dirty, and the extra material has just been elided.

Stanley argues for two points. The first is that we do not reach the 'natural' understanding (in context) of "The table is dirty" by supplying missing words, as we do in standard cases of ellipsis. As Strawson remarks, in standard examples the missing material is present in the surronding linguistic context (and is, on some accounts, literally 'copied' from the surrounding context). Nothing of the sort seems to be true of incomplete descriptions. Stanley then explains his favored account of incomplete descriptions, which treats "the" as a quantifier (meaning: one and only one). If "the" is a quantifer, then one would expect it to exhibit the same kind of domain variation that other quantifiers do.

To give a little more detail, what Stanley is proposing about incomplete descriptions is roughly as follows. According to Russell's account, "The table is dirty" means something like: ∃x[x is a table & ∀y(y is a table → x = y) & x is dirty]. (More modern analyses would treat "The" as being its own quantifier.) The worry, again, is that this will always be obviously false. But if quantifiers have unpronounced domain variables associated with them, then the initial existential quantifier can have a 'restricted' domain; e.g., the domain might be just the things in this room. And it might well be true that there is only one table in this room. The mere fact that utterances of sentences with incomplete descriptions can have sensible meanings thus does not, by itself, threaten Russell's analysis. (It is another question whether Russell's analysis is, in the end, correct.)

Stanley's real point in this section, however, is the one made at the end of the section, concerning what a proper response to the binding argument would have to be like. One cannot simply say that there is some "magical" process through which the right interpretation is generated. One actually has to explain what that process is, when it is operative, what restrictions there are on it, and so forth.

In the third section, Stanley elaborates such a response, drawing upon work by Robyn Carston and Kent Bach. The idea is roughly as follows. Fans of "free enrichment" already accept that, during the process of semantic interpretation, additional material can be added to the (possibly incomplete) proposition that is provided simply by the literal meanings of the words used and whatever compositional rules there might be. For example, if someone utters the sentence "Michael is tall", then this can be 'enriched' to "Michael is tall for a man" or "Michael is tall for a basketball player".

Stanley tends to put this in terms similar to those of Sellars: The sentence is somehow 'enriched'. In fact, Carston and other 'relevance theorists' tend to speak in terms of the enrichment of the mental representations we use to interpret the sentence. But this detail does not affect the argument, for the most part. But see note 16.

Stanley several times makes remarks like the following: "If interpreters tacitly exploit these interpretive processes, then speakers will be tacitly aware that hearers may exploit these sorts of processes" (p. 160). Although this is not central to Stanley's goals in this paper, one might wonder why this is important. Why is it?

Similarly, then, the thought is that the very same process could also provide a pronoun (amongst other material) that can then be bound by a higher operator. E.g., in the course of interpreting "Every senator is disliked by most voters", one might 'enrich' it to: Every senator is disliked by most voters in their state, thus recovering the bound reading.

Stanley himself develops the argument using comparative adjectives, which also, as he notes, figure in bound readings. So, on his view, if a comparison class is not explicitly provided, there is a variable for one that is present anyway. So, according to Stanley, "Michael is tall" really has the form "Michael is tall_F", where "F" can either be given a value expicitly ('for a man') or left to get it from context. What, according to Stanley, shows that the variable is really there is that its value can be made to depend upon other elements of the sentence, as in his example: Every species has members that are small.

In the final section, Stanley argues against this sort of move by arguing that it over-generates. This too is a standard sort of linguistic argument. It is every bit as important that one be able to explain why certain readings of sentences are not available as that one should be able to explain why certain readings are available. For example, we want to know not just why, in "John's brother said Tom loves him", the pronoun can take either "John" or "John's brother" as its antecedent, but also why it cannot take "Tom" as antecedent. So the worry here is that, if "free enrichment" can provide bindable material, then we would expect to get readings of sentences we cannot in fact get.

Stanley claims that:

(15) Everyone has had the privilege of having John greet.

is ungramamtical, but that it could be rescued from ungrammaticality by the addition of the word "them" to the end of the sentence:

(16) Everyone has had the privilege of having John greet them.

(This is itself ambiguous, but we are interested here in the reading where "them" is bound by "everyone".) Since that is the sort of thing that "free enrichment" is supposed to be able to do, it is then a mystery why (15) is not grammatical after all.

But (15) is grammatical (and a similar point applies to Stanley's (13)). Greeting is a task one might have at a church or a meeting, and so (15) can perfectly well mean that everyone has had the privilege of John's performing a certain task for them.

Here's a possible response for Stanley. If 'enrichment' could add the word "them" to the end of (15), then (15) would be ambiguous: It would have both the standard reading just described and the reading with "them". But (15) simply isn't ambiguous in that way. Thus, this account "over-generates", in the sense that it predicts that (15) should have a reading it simply does not have, unless some way can be found to stop (15) from being "enriched" to (16)—while, of course, preserving the other uses of 'enrichment'.

What makes trouble for Stanley's example is the fact that "greet" can be both transitive and intransitive. So one could construct a better example by instead using a verb that only has an intransitive meaning. Then the sentence corresponding to (15) would indeed be ungrammatical. Construct such an example and explain what Stanley's claims about it would be.

It would seem, then, that there are two different sorts of examples that make trouble for 'free enrichment': ones involving ungramamtical sentences and ones involving ambiguity. Is one of these examples better than the other for Stanley's purposes? Or are both needed for different purposes? (See in particular the discussion of the 'final reply' on pp. 163-4.)

It is extremely important here to keep clearly in mind that the issue is supposed to be whether utterances of (15) can express what (16) does; that is, the question is whether what is said when (15) is uttered can sometimes be what would have been said had (16) been uttered. It is not relevant if utterances of (15) can communicate what (16) does, i.e., that (16) could be what is meant (implicated) by someone who utters (15). This is the point made on pp. 165-6.

This paper is one in a volume of essays responding to and commenting upon Herman Cappelen and Ernie Lepore's book Insensitive Semantics, in which they argue for the view now known as semantic minimalism. This is the view that, with the exception of the obvious exceptions ("I", "here", "this", etc), every sentence expresses a unique proposition, so that semantic context-sensitivity is limited to those obvious exceptions. (Of course, they allow pragmatic effects, such as implicatures, to depend upon context.) Borg also defends a form of this view, though an even stronger one. Our interest is in how carefully Borg sets out the different positions and how she distinguishes the sorts of views Carston and Stanley defend. She does not argue for any particular view of her own in this paper. See her book, listed as optional, for a lengthy defense of her own view.

Borg identifies four sorts of arguments against minimalism, that is, in favor of the view that some particular expression is context-sensitive.

1. Context-shifting Arguments: These claim that some particular sentence, e.g., "Jewell is tall", can express a truth in some contexts, but not in others, even if the relevant facts have not changed.
2. Incompleteness: These claim that particular sorts of sentences, such as "Mary is ready", rarely, if ever, express propositions on their own, but require some supplementation.
3. Inappropriateness: These claim that certain sorts of sentences, such as "Every bottle is empty", although there is a proposition they could always express, cannot, or do not usually, express that proposition, because it is too obviously not what speakers mean.
4. Indeterminacy: These arguments claim that even the thought expressed is in certain sorts of cases indeterminate. It's much less clear what an argument of this type would be like, and Borg does not give any examples. Some of Searle's remarks suggest this kind of view, however. For a different kind of example, consider a sentence like "The apple is green", and suppose it clear which apple is in question and that we are talking about color (not ripeness). Is it supposed to be green on the outside or on the inside? Is it OK if it's a red apple painted green? (Note what that use of the word "red" meant in my last question.)

Part of the point Borg wants to make is that some of these arguments, if accepted, lead to more radical departures from minimalism than others.

All of these sorts of arguments should be familiar by this point, with the possible exception of the last. Which of them seem, initially, to be the strongest arguments for 'contextualism', and why?

Following Cappelen and Lepore (C&L), Borg distinguishes two sorts of contextualism: radical and moderate. C&L had characterized the difference in terms of the scope of context-sensitivity, so that radical contextualists regard all expressions as context-sensitive, while more moderate views regard fewer terms as context-sensitive. Borg suggests, however, that this is not a particularly illuminating characterization. As she sees it, the disagreement instead concerns the mechanism of context-sensitivity. One view, which is represented for us by Stanley, holds that, while there are terms outside the "basic set" ("I", "here", "tomorrow", and the like) that are context-sensitive, the way their meaning depends upon context is similar to the way the meanings of the words in the 'basic set' depends upon context. A more radical view, represented for us by Carston and Searle, holds that "there are forms of context-sensitivity that are not capturable on the model of...the Basic Set" (p. 344).

So, again, Borg regards the crucial questions as being:

• What are the mechanisms of context-sensitivity? In particular, what is involved in 'resolving' context-sensitivity?
• Can the context of utterance affect semantic content even when such action is not demanded by the syntax of the sentence?

Moderate contextualists answer the second question "No" and so think that the way context-dependence is resolved in non-Basic cases might be no different from how it is resolved with "you" and "that": It involves assigning values to (possibly covert) elements in the syntactic structure of a sentence. Radical contextualists, by contrast, answer the latter question "Yes" and so think that context-sensitivity is not always 'resolved' in that way. Some other 'mechanism' is involved. Borg's discussion on pp. 344-5 suggests that, in practice, radical contextualism is distinguished by its embrace of 'free enrichment' and 'unarticulated constituents', that is, by its rejection of the 'linguistic direction principle'.

The moderate view is thus much closer in spirit to minimalism, since it regards all context-sensitivity as traceable to syntactic 'triggers'. The disagreement between moderate contextualists and minimalists thus concerns not what kind of phenomenon context-sensitivity is, so to speak, but only how extensive that phenomenon is. (Does it just apply to "I", "here", "that", etc, or also to quantifiers and comparative adjectives?) Radical contextualists, on the other hand, think we need "an entirely different picture of the relationship between semantics and pragmatics" (p. 346), i.e., they think that there is something fundamentally wrong with the model of context-sensitivity that informs miminalism and moderate contextualism.

Using this distinction, Borg then defends moderate contextualism against a charge made by C&L. In their book, they had argued that moderate contextualism "collapses" into radical contextualism: Once one accepts that arguments of the four types listed above can be used to show that there are context-sensitive expressions outside the basic set, then arguments of the same types can be used to show that all expressions are context-dependent. But, if we draw the distinction between radical and moderate contextualism as Borg does—if it is, really, a disagreement about the mechanisms of context-dependence—then it seems clear that simply multiplying examples cannot by itself force a moderate contextualist to accept 'free enrichment' and 'unarticulated constituents'. So there is no immediate danger of 'collapse'.

There seems to be a different sort of argument implicit in Borg's discussion, one that could allow moderate contextualists to resist what we might instead call an 'explosion' of context-depenence beyond the 'basic set', to encompass all expressions. The thought here is that none of the four sorts of arguments that were listed above can, by itself, force a moderate contextualist to regard a given expression as context-dependent. The reason is that a moderate contextualist will regard the expression as context-dependent only if it can also be shown that there is relevant syntactic structure. Can you fill out some of the details of the argument just sketched?

At the beginning of section 3, Borg characterizes the contrast this way:

[T]he fundamental schism in philosophy of language [is] between those who think that sentences (relativized to contexts of utterance) possess genuine, truth-evaluable content in their own right (and who go on to endorse some kind of formal approach to semantics) and those who think that it is only at the level of the utterance that it makes sense to talk about real meaning or to ask for truth-evaluable content (and who thus advocate a speech act approach to semantics). (p. 347)

Here's what she has in mind. For a moderate contextualist, all context-dependence involves assigning values to (possibly covert) elements of the syntactic structure of a sentence. So, in principle, we could make a list of all these 'loci' of context-dependence and consider a 'formal context' simply to be an assignment of appropriate values to all these elements.¹ The semantics, then, could assign truth-conditions to sentences, relative to such 'formal' contexts. There would still be an important question how, in practice, speakers decide what value to assign to a given element (or, if you prefer, what facts make it the case that a given element has a given value in a given speech situation), but that question would not 'infect' the semantic theory itself.

For a radical contextualist, by contrast, context-dependence is too unconstrained for any such approach to work. There is no specifiable list of the 'loci' of context-dependence; rather, various factors will affect how the meaning of the uttered sentence is 'enriched'. There is, presumably, some work for 'semantics' to do—it's not as if the meanings of the words don't matter—but what is said is determined by too many factors for any formal theory to specify truth-conditions 'relative to' the context of a given speech situation.

Note the different ways the term 'context' is being used here. How should we understand them?

Borg spends the remainder of the paper offering a characterization of semantic minimalism. It has four parts.

(1) Sentences (not just utterances of them) typically do express propositions (modulo any obviously context-dependent elements they might contain).

There are different kinds of cases here. There's no particular problem seeing what proposition the sentence "Every bottle is empty" might express. (Which?) It's less clear what proposition the sentences "Jewell is tall" and "Jill is ready" might express (all by themselves, again, not relative to any context). What might a minimalist say about such cases? (There are some hints in §3.2.)

(2) There is something special about the expressions in the 'basic set' and any extension of context-dependence beyond these is momentous.

What reasons does Borg give, in the first paragraph of §3.2, in favor of this claim? How good are those reasons? (The rest of §3.2 discusses a sectarian dispute between Borg and C&L and can be skipped.)

(3) Minimalists draw a strong distinction between semantic content and speech act content, where the latter corresponds to what is said in Grice's sense and is not determined by semantic content.

Most of Borg's own discussion is, again, concerned with a disagreement between her and C&L about what role this 'minimal proposition' is supposed to play. (The problem is that it doesn't really seem to play any role, but to be a useless theoretical appendage.) But there's a deeper worry here for both Borg and C&L. If what's said when someone utters "Jill is ready" isn't determined by the minimal proposition allegedly expressed, then what does determine it? In practice, it would seem to be the sorts of pragmatic processes to which radical contextualists draw attention. Indeed, Borg herself speaks of the "pragmatic strengthening" of expressions. So it seems as if it turns out that, if our interest is in what's said, then minimalism looks to be closer to radical contextualism than one might have thought it should. How might Borg respond to this worry?

Consider the example Borg discusses on p. 354: Alex says, "Jill is tall", but Sam doesn't know what comparison class Alex had in mind. When asked what Alex said, Sam says, "Alex said that Jill is tall". Borg insists that this will be an accurate way of reporting what Alex said. Is that right?

(4) "[T]here is an entirely formal route to semantic content" (p. 355).

Borg distinguishes two different forms of this view. On the weaker, her (a), it amounts to the linguistic direction principle. On the stronger, her (b), it implies that the 'semantic content' of a sentence should be wholly determined by its formal features and so independent of any contextual facts. This is an incredibly strong view. I take the main question about it to be: Why should one think, as Borg seems to think (see e.g., fn. 42), that there needs to be something propositional or truth-evaluable that is wholly determined by the formal properties of a sentence? (It might be that this is really another sectarian dispute between Borg and C&L, i.e., that the argument here is directed against them—or, perhaps better: against people who have already accepted the other three claims that characterize minimalism.)

---

¹Of course, there are all kinds of complications in practice. For example, there might be multiple quantifiers with different domains, so we'll need not just one 'locus' for quantifier domains, but something like a sequence of them. In some cases, we might also want to consider the 'formal context' not as assigning a value directly to an element, but as containing other information sufficient to do so. Thus, 'contexts' are often taken to include the speaker, location, and time of the utterance. These determine what "I", "here", and "now" refer to (if they are used in the utterance).

Maitra first takes up a topic also discussed by Borg: what divides Moderate from Radical Contextualism. Like Borg, she starts with the idea, taken from Cappelen and Lepore (C&L), that the issue concerns the "extent" of context-sensitivity. But Maitra suggests that we should understand this in terms of:

• The Meaning Question: How much does the "standing meaning" of an expression constrain the content it can have on a given occasion of use?
• The Context Question: How rule-governed is the determination of content from context?

The thought, then, is: "The more constrained and rule-governed the semantic contents of an expression are on a given Contextualist view, the less context-sensitive that expression is taken to be" (p. 116). This is an important point: One does not have to choose between the different sorts of views here once and for all, but can hold different views about different sorts of expressions. E.g., almost everyone agrees that "I" is much more constrained than, say, "that" and it seems not unreasonable to think that determining the reference of "I" is more rule-governed than determining the reference of "that".

Since there are two dimensions of variation here, there may be cases where a view is more 'moderate' along one dimension, but more 'radical' along the other. Maitra recognizes this issue but suggests that it will not affect her argument. For what it's worth, I find it hard to see how it could matter for any argument. What's important is that we have correctly identified the dimensions along which views may vary. That matters because Cappelen and Lepore had argued that all 'moderate' views are unstable: They collapse under pressure into 'radical' views. So the key question here (as Borg also points out) is what distinguishes 'moderate' from 'radical' views.

Maitra mentions the term "character", which was introduced in work on demonstratives by David Kaplan. In this sense, 'character' is supposed to model 'standing meaning', that is, the character of an expression is (roughly) the meaning that expression has in its own right (as opposed to on any given occasion). Kaplan treats character as a function from contexts (circumstances in which an utterance is made) to 'contents', so the character of "you", for example, would be a function from contexts to whoever is denoted by "you" in that context (the 'addressee', as it is sometimes put).

Maitra goes on to argue, in §3, that this way of categorizing the different views makes the question how many expressions are context-sensitive not the crucial question (as C&L had taken it to be). What matters is the way in which they are context-sensitive: how much standing meaning constrains semantic content (what is said), and how rule-governed the 'resolution' of context-dependence is.

There are thus some obvious similarities between how Maitra and Borg characterize moderate and radical contextualism. Are there differences as well or do they really amount to the same idea? If there are differences, is there some way to put the two views together to get a stronger one?

Maitra suggests that her way of characterizing moderate and radical contextualism is preferable to C&L's: that it illuminates better what is at issue between, say, Carston and Stanley. How so? Does the question whether there are 'unarticulated constituents' figure into Maitra's characterization? If so, how? If not, has she left out something important?

The main focus of the paper is on what Maitra calls the "Miracle of Communication Argument" (MCA), which C&L use to argue against Radical Contextualism. They also argue, as we have seen, that Moderate Contextualism 'collapses' into Radical Contextualism. Maitra has already argued against this latter thesis. But C&L (or a sympathizer) might argue that this doesn't matter, because the MCA poses just as much of a threat to Moderate Contextualism as to Radical Contextualism. So Maitra will now argue against that claim: Moderate Contextualism, in her sense, is not vulnerable to the MCA.

The idea behind the MCA is that, if pragmatic processes affect semantic content as dramatically as Radical Contextualists propose, then, since what is said can depend upon almost any piece of information one has, it is obscure how speakers and hearers ever manage to converge on a particular interpretation of some bit of language that is context-dependent in the way that Radical Contextualists suppose.

Maitra raises three sorts of objections to this argument.

1. It is also true that what is implicated when a particular utterance is made can depend upon almost any piece of information one possesses, and it is not at all clear that there are 'rules' we follow to determine implicatures (although there are guidelines of the sort Grice elaborated). Yet we do regularly manage to communicate by implicature. If the MCA were correct, then, it would imply that it is a 'miracle' that we ever communicate this way—which it is not. So something is wrong with the MCA.
2. The second point is ad hominem and less important for our purposes. But, on C&L's own view, 'what is asserted' (which is what Borg called "speech act content") depends upon the same sort of information on which Radical Contextualists think what is said depends. But C&L think we nonetheless regularly manage to converge on speech act content. (That's what, on their view, communication typically requires.) So it is hard to see how even they can endorse the MCA in full generality.
3. To determine the reference of many uncontroversially context-sensitive expressions, one can need to know a great deal. Indeed, there seem to be very few expressions in the "basic set" whose content on a given occasion of utterance is completely determined by rule: "I", "today", "tomorrow", and "yesterday" seem plausible candidates. (These are the so called 'automatic indexicals'.) But neither "that" nor "we" has its reference completely determined by rule.

Regarding (1), Maitra's argument so far has shown, at best, that there must be some explanation to be given of how we manage to communicate by implicature. And, the thought seems to be, whatever that explanation is, it will also work for context-dependent expressions. But Maitra has not told us what the explanation is. What might it be? How do we manage successfully to communicate by implicature when the implicated content is so 'unconstrained'?

Regarding (2), since C&L are minimalists, they are committed to think of, e.g., the semantic content of "Every bottle is empty" as always being that every bottle in the universe is empty. Since this is not what is usually communicated by such an utterance, they need some other notion, besides semantic content, to play this latter role. Whatever it is, it appears that it will depend upon the same sorts of things that non-minimalists think what is said depends upon.

Regarding (3), Maitra does not really develop the example she gives to show that determining the reference of "we" can depend upon a wide variety of information. Can you fill in some of the details? Can you develop similar examples for "here" and "now"? (In what respects are "here" and "now" like "we"?)

The optional paper develops a version of this third argument at length, focusing specifically on demonstratives.

Still, Maitra thinks (or at least concedes) that the MCA does pose some sort of challenge to Contextualism. In particular, we need to "explain why hearers are generally more confident about what is communicated via semantic contents, than about what is communicated in other ways" (p. 125), such as implicature.

Maitra argues in §5 that an appropriately Moderate form of Contextualism has an explanation to offer. Focusing on comparative adjectives, like "tall", she suggests that:

1. Their standing meaning highly constrains their content, since the only locus of variation is in the comparison class. I.e., "tall" always means: tall for an F, and the only issue is what F is.
2. It might be possible to say something fairly definite about how different contexts make "natural" comparison classes available.

Maitra argues that a view with these features can answer the MCA: The work one has to do to understand a sentence containing a comparative adjective is limited to figuring out what the comparison class is, and one can 'default' to more natural comparison classes. By contrast, nothing of this sort is true for implicatures and other non-literal forms of communication.

Regarding (i): One might have thought that the standing meaning of "that" hardly constrains its content at all, since "that" can be used to refer to any object at all. But, in a different way, one might think that the standing meaning constrains its content quite a bit, since the only locus of variation is which object is referred to. What should Maitra say about this? (Hint: Remember exactly what she is trying to argue here.)

Regarding (ii): What does Maitra mean by a "natural" comparison class? What does she mean by "natural" readings of sentences? (How might this compare to an example mentioned earlier, according to which there are 1000 possible concepts humans can have, and they are all innate?)

Regarding (ii) again: Is anything like this true of ordinary uses of demonstratives? Are there 'natural referents' for uses of "that"? If not, what might that imply about Maitra's response to the MCA?

On p. 128, Maitra considers an objection that a Radical Contextualist might reasonably bring: Even once we know what the comparison class for "fast", say, is, "there are many ways of being fast for a snail". So the worry is that, if we just specify the comparison class as being snails, we have still yet to specify a truth-evaluable content. Maitra offers two replies, the first of which is directed at C&L and is ad hominem. What is the second reply and how effective is it?

This question is for those familiar with degree-theoretic treatments of comparative adjectives: Are such accounts vulnerable to this kind of objection? Why or why not?

Finally, Maitra considers the question whether a Contextualist might concede that, since so much information is potentially relevant for determining the comparison class, say, there will be failures of perfect communication. Nonetheless, the thought is, communication does not need to be perfect to be successful. She does not really develop an example to illustrate this possibility. Can you develop one using, say, "fast" or "tall"?

Maitra remarks that Moderate Contextualism "makes it very easy for hearers to have good but approximate knowledge of semantic contents in pretty much every actual context" (p. 130, emphasis added). Why is that enough to answer the MCA?

The central question of Hom's paper is how we should think about the derogatory content of racial epithets. In fact, the discussion seems to apply more broadly, to other slurs, such as "faggot". The main question Hom wants to discuss is: Is the derogatory content of such terms part of what is said when someone uses them? Or is it just part of what is meant?

In §I, Hom outlines the view he wants to defend. This is the 'semantic' view that, to first approximation, the word "faggot" means: gay man (that part is the 'non-pejorative correlate') and despicable because of it. Such an account faces two main challenges, Hom suggests: (i) explaining why only some uses of epithets derogate (in particular, how there can be 'appropriated' uses, if the derogatory content is part of the literal meaning); and (ii) how different epithets can have different force (e.g., why "faggot" is worse than "fairy").

Hom also discusses and dismisses what he calls 'silentist' accounts. The way he introduces this view does not seem to match what he ultimately takes it to be. What he initially says is that silentists hold that the meaning of a racial slur "is so strong that it derogates its intended targets on every occasion of use" (p. 417). But many views, as we shall see, agree with that claim. Silentism seems, rather, to be the view that there is something sui generis about slurs that cannot be captured by ascribing any familiar sort of 'content' to them. Since Hom does not really develop this view, we won't worry too much about it.

In §II, Hom considers (and criticizes) various 'pragmatic' accounts of epithets, though these are not all 'pragmatic' in the same sense. The first of these is a 'radical contextualist' account that holds that the meaning of a epithet varies according to the context. Hom's complaint about it is that it doesn't specify a 'rule' that would allow us to predict the meaning of a given utterance of an epithet.

Generally speaking, contextualists have not been worried by such complaints—in part, as we have seen, because it is not at all clear that there are 'rules' for determining the reference of "we" and "now". Hom does give some reasons that we should demand a rule in this case. How good are those reasons? (It will be worth asking, later, whether Hom gives us a rule that distinguishes derogatory from non-derogatory uses of slurs.)

Note, by the way, that this 'contextualist' view could be understood either as a view about what is said or as one about what is meant. That is one of the reasons I said, above, that not all of the views considered in this section are 'pragmatic' in the same sense.

The second pragmatic strategy borrows from Frege, who suggested that synonymous words sometimes differ in 'tone' or 'coloring'. My own view is that what Frege meant here is probably best elaborated in terms of conventional implicature, which is the third view Hom discusses. Here, Hom is trying to see if there is a different way of understanding the view. So, to a large extent, this discussion can be skimmed.

As for the third view just mentioned, Hom has several criticisms to make of it. The first is that, if the derogatory content of a word is a conventional implicature, then it is not 'cancelable' and so would have always to be present. But Hom claims (and will argue later) that there are non-derogatory uses of words like "faggot" (and not just appropriated uses).

It will be worth asking, later, when we see what sorts of non-derogatory uses Hom has in mind, whether standard cases of conventional implicatures pattern similarly. That is: Is it obvious that conventional implicatures must always be present, whenever a word that allegedly has such an implicature is used?

Another objection is that, on this view, "Male homosexuals are faggots" is not only true but analytic, since it is synonymous with "Male homosexuals are gay men". Hom, however, seems to think the sentence is not analytic (and, on his view, it is not even true). A third objection concerns pairs like:

1. Alex said that Fred is a gay man.
2. Alex said that Fred is a faggot.
3. Alex believes that gay men are gay men.
4. Alex believes that gay men are faggots.

One might think that (1) and (2), and (3) and (4), can have different truth-values although, again, on the view we are considering, one might expect them to be synonymous, since we have just replaced one word with a synonym.

Note that these two objections will arise for anyone who thinks that the difference between "faggot" and "gay man" is pragmatic rather than semantic. In that sense, it is a very strong objection. But defenders of a pragmatic approach can, it seems to me, simply note the similarity between this example and so-called Mates cases (for which see the optional paper by him). Consider, e.g., "A fortnight is two weeks". That is the definition of "fortnight", but it also seems as if the statement could be informative. Can you develop this analogy?

The last objection, concerning indirect speech reports and belief-attributions, may be difficult for us to evaluate, since how one responds to it may depend upon views about those constructions. On some views, for example, the specific words used in an indirect speech report can be important; such views might not have as hard a time with (1) and (2) as others.

In §III, Hom lays out a number of facts he thinks a good theory of epithets will need to explain. Many of these are fairly obvious, but a couple merit special attention. What Hom calls "derogatory autonomy" is the fact that the power of an epithet does not seem to depend upon a speaker's own attitudes but seems to be some kind of social fact (possibly even a convention, in Lewis's sense). What he calls "evolution" is related: the fact that the derogatory force of an epithet can change over time. Hom also thinks it is important, as was said above, that we be able to account for 'appropriated' uses of epithets.

The most important of Hom's desiderata is the eighth: Hom claims that there are non-derogatory, non-appropriated uses of epithets (what he calls NDNA uses). Hom gives a number of examples to support this claim. How compelling do you find them to be? It's worth reformulating these for yourself in terms of other epithets. It's also worth considering what might seem to be minor variants of Hom's examples, such as "Is Naomi Osaka a chink?" or "Naomi Osaka is not a chink" or "Are Japanese people chinks?"

In thinking about some of these examples, it's worth be aware of what is known as 'meta-linguistic negation'. This is the sort of use of negation that one finds in a dialog like:
A: Sally is knocked up.
B: Sally isn't knocked up. She's pregnant.
(Statements like B's often feature special intonation, such as focal stress on "knocked up".) It's often said that, since "knocked-up" just does mean pregnant, the initial negation can't be an ordinary negation. Rather, it means something like: You shouldn't say it that way. It's the specific words that are negated, so to speak. Some of Hom's examples seem to be 'meta-linguistic' in this way. Which of them aren't? (Of course, it is open to Hom to regard "knocked up" as a derogatory term for pregnant and so to disagree with this particular example. But there are other examples, and meta-linguistic negation is a pretty well-established phenomenon.)

Hom's last desideratum is controversial. One might reasonably think that "cunt" and "faggot" are derogatory in the same way racial slurs are. But it's much less obvious that expletives like "damn" or "fucking" work the same way, let alone that 'approbatives' do. In any event, we will not be concerned with this issue.

In §IV, Hom develops his positive view, which he calls "combinatorial externalism". Hom's view is that the derogatory force of a racial epithet, at a given time, and in a given society, derives, in large part, from racist ideologies and racist practices prevalent in that society at that time: the "social institutions of racism" (p. 430). So his view is that what the epithet means is: a person who is a member of a certain group (that is the non-pejorative part) and who, as such, ought to be treated in certain ways (that is the practice part) because members of that group have certain properties (that is the ideology part). It's crucial to Hom's view that, although this content is always present when the epithet is used, uses of it are not always derogatory.

Hom's view implies that a racial slur derogates only in so far as there are 'social practices of racism' associated with it, and similarly for other sorts of slurs. Does this seem correct?

In the last section, Hom argues that his view does better with the desiderata listed in §III than the other views he considered. I'll leave it to you to comment upon these and to raise any objections to Hom's view that might seem to you to emerge from that. (I find it hard myself to understand what Hom wants to say about appropriation. If you do understand it, please feel free to explain.)

At one point, Hom writes: "Racists are not only wrong in the normative sense, but also wrong about the world in falsely attributing racist properties to people" (p. 437). Explain why he thinks this claim supports his account. Does it?

One welcome aspect of A&L's paper is how explicit they are about what they want to explain: the offensiveness of slurs. (We'll return to this point.) Their goal is (i) to argue against all views according to which the offensiveness of slurs is to be explained in terms of some offensive content that the slur conventionally expresses (in one way or another) and thereby (ii) to motivate their alternative view according to which the offensiveness of slurs is to be explained simply in terms of the fact that they are taboo, i.e., that their use is prohibited by social norms.

Section 2 considers whether (e.g.) "faggot" and "gay man" 'express the same concept' or 'have the same meaning'. (Their view is that they do—or better, that, however they differ, it is not in meaning.) A&L's first complaint is that everything that anyone has ever been said about how these words differ in meaning would make "faggot" and "fairy" have the same meaning, even though the former is much more offensive than the latter. (They give a different example.)

We have, so far, only considered one view, Hom's. Recall that he does aim to distinguish the strength of slurs for different groups, e.g., to explain why "faggot" is more derogatory than "limey". Can he also explain why "faggot" is more derogatory than "fairy"? (It will be worth keeping this question in mind as we proceed to discuss other views.)

A second worry is that, on Hom's view, it seems as if

(a) Elton John is not a faggot.

should (assuming the negation is not 'meta-linguistic') mean something like: Elton John is not a gay man who should be discriminated against for that reason. That seems straightforwardly true and, indeed, laudatory, whereas A&L claim that (a) is in fact every bit as offensive as "Elton John is a faggot".

There is, then, also the case of meta-linguistic negation to consider. I suggest that it will not help A&L's opponent to focus on the use of (a) in which the negation is meta-linguistic. Why not?

Section 3 addresses issues about embedding and speech reports. The point here is that, if "faggot" had a different meaning from "gay man", then something like:

(b) Eric said that Elton John is a faggot.

should, one might have thought, simply attribute such an utterance to Eric. But A&L argue that it is the person who utters (b) who slurs and, correlatively, that (b) need not attribute a slur to Eric (though it might suggest it). The data here seem very muddy indeed: As A&L note, there seems to be some disagreement about both these points. In that sense, (b) cannot easily be used to argue for the claim that slurring is semantic; but it also, it seems to me, cannot be used to argue that it is not. A&L do, however, seem to rely upon claims about (b) later in their paper.

Section 4 discusses presuppostional views. There are many cases in which use of a term seems to 'presuppose' certain facts. Thus, even the question "Has Bill stopped smoking?" 'presupposes' that Bill has smoked in the past; the same goes for "Bill has not stopped smoking". But presuppositions do not always 'project out', and one case where they do not are speech reports, as A&L note. So if the use of "faggot" presupposes that gay men are vile, then one would expect (b) above not to presupppose that, and so not to be offensive. But, A&L claim, it is. (This argument, then, inherits the problems of the argument in section 3.)

As A&L note, presuppositions can be satisfied by explicit suppositions, even ones introduced in the antecedents of conditionals. As A&L note, the data as regards indicative conditionals is once again murky. But we might also consider counterfactuals. If someone says, "If Bill had ever smoked crack, he would have stopped by now", they do not presuppose that Bill ever smoked crack. (In fact, they presuppose the opposite.) So consider: "If gay men were vile and inferior on account of being gay, then Elton John would be a faggot". What should we say about that sentence?

Those who know something about the 'projection problem' for presuppositions might consider the behavior of slurs in other contexts in which presuppositions do not typically project.

Section 5 considers, and quickly dismisses, the view that the offensive content of a slur traces to what Frege called its 'tone'. You can skip it, if you wish.

Sections 6-7 consider the view that slurs conventionally implicate offensive content. A&L note several advantages of this view: Most importantly, it explains why the offensive content of slurs tends to 'project out' from so many sorts of contexts (indeed, on their view, always, even from quotations). What they think is the most plausible form of the view is that slurs work something like appositives, so (a) would mean something like: Elton John is not among the gay men, who are vile and inferior in virtue of being gay.

A&L's criticism of this view has two components. The first is that the view over-generates, since there are, they think, non-offensive uses of slurs, including appropriated uses (which they discuss in detail in section 9) and uses in what they call 'didactic contexts'. Unfortunately, they do not explain this latter term, though they may have in mind the occurrences of slurs in their own paper. Second, if what makes a slur offensive is something about its meaning, then it is puzzling why the mere mention (quotation) of a slur might be offensive, since its meaning is irrelevant when it is just mentioned.

An interesting issue emerges here. In a footnote that Whiting seems to have added late to the paper listed as optional, he writes:

Though I do not doubt that the mere mention of a slur can cause offence, I doubt that merely to mention a slur is really to derogate. (p. 369, fn. 11)

Whiting's point is that the offensiveness of slurs is not what he is trying to explain. His interest, rather, is in the way slurs can be used to derogate, i.e., belittle or demean. That is quite different. (As we'll see, a similar issue arises with respect to Elisabeth Camp's discussion: Some of her examples of 'non-derogatory' uses are quite similar to A&L's (28)-(30).)

So what should we be trying to explain? One possibility would be to let many flowers bloom: There are just different projects here. But one might also think that slurs' capacity to give offense is not entirely independent of the way they can be used to derogate. That different theorists have different explanatory ambitions may also affect their evaluations of indirect speech reports and other data, such as (b) above.

Section 8 present A&L's positive view, that "slurs are prohibited words" (p. 38). A&L suggest that there are lots of different reasons that a word can become prohibited. They also suggest that the group that prohibits the use, in such cases, is usually the 'target' group (though it need not always be). So the view is that "faggot" is a slur, and its uses are offensive, because use of that word is prohibited, most likely by gay men themselves. The thought seems to be something like: Just as each of us gets to decide the name by which we will be known, the members of a certain group get to decide by what names that group will, and will not, be known. Not to respect such a prohibition is to thus evince disrespect for members of that group.

Exactly what is A&L's explanation for the difference in 'intensity' of "faggot" and "fairy"?

It seems a delicate question what sort of authority one has to have to 'prohibit' the use of a word. Could we rephrase A&L's view in other terms? Whether gay men can 'prohibit' the use of "faggot", one might think it clear that they do not wish to be referred to in that way. What alternative views might there be here?

Is it really true that use of the N-word is offensive because African-Americans have prohibited its use? (Whiting raises this question in the optional reading.) One might think that Prohibitionism gets the explanatory order backwards: Slurs don't offend because they are taboo; they are taboo because they offend (better: because they derogate and therefore offend). How can this question be decided?

As is made clear in the book of Randall Kennedy's listed as optional, it wasn't very long ago that lots of white people used the N-word routinely. They thought using it was perfectly fine. Many of them were racists and actually did think black people were inferior, etc. But, for many of them, it was just the way they spoke. If there was a "prohibition" on use of the N-word, that would have been news to them. Some of them were genuinely puzzled about why the use of the word caused offense. (As Geoffrey Nunberg points out in the paper listed as optional, slurs are most often used in casual conversation between people who endorse their use, not as weapons.)

The word "retard" has only recently come to be regarded as a slur. Was that because its use only recently became prohibited? By whom? (This might be an example of a slur that might not be prohibited by the targeted group but by someone else: see note 37.) Was its use offensive only after it became prohibited?

Generally speaking, A&L's view is that the offensiveness of slurs is really a sociological fact, not a linguistic one—not a matter of what they 'mean', in any sense. That kind of view could be developed in a variety of ways, not just in terms of prohibition. What might some different approaches be like? (A&L's view is similar in this respect that the view developed by Nunberg, in the paper listed as optional.)

Camp offers in this paper an account of slurs in which they encode what she calls 'perspectives'. She here attempts to remain neutral on how such perspectives are 'encoded': whether as part of what's said or in some other way. Her claims here are meant only to concern, as she puts it, "what the 'other' component of slurs is" (p. 331), besides what is present in the 'non-pejorative correlate'.

Many of the views Camp discusses treat slurs as what are sometimes called 'thick' terms: they have a descriptive component, and also an evaluative one. (See the SEP article on thick terms for more on this notion.) As she notes, though, many of the terms that have been claimed to be 'thick', such as "brave", do not seem to have non-thick correlates, so the analogy is not perfect.

One important observation here (which is not original to Camp) is that the derogatory aspect of slurs seems to 'project out' from negation, conditionals, and many other constructions. That is, these three sentences:

(1) John is a faggot.

(2) John is not a faggot.

(3) If John is a faggot, then ....

might all seem equally offensive. If that's right, it's a problem for views like Hom's: If the derogatory aspect is just part of what's said, then it seems as if I should be able straightforwardly to negate it. I.e., (2) should mean something like: It is not the case that John is a gay man and, as such, is an appropriate target of discrimination; if gay men are never appropriate targets of discrimination as such, then all atomic predications like (2) are false (and that, indeed, is Hom's view). But, one might think, what (2) denies is simply that John is a gay man, and the derogatory element is still present.

Camp claims that we cannot use (2) to 'reject' (1) "unless we want to buy into [the bigoted] perspective ourselves" (p. 330). Is that true? Is there a 'non-derogatory, non-appropriated' use of (2), as Hom claims? If so, what sort of use might that be?

In section 2, Camp attempts to map the terrain of possible views. On one end, we find expressivist views, according to which the use of a slur expresses (in the sense of moral expressivism) a negative attitude about the target group. So (1) would mean something like: John is a gay man, yuck! Camp lodges two objections against this sort of view. The more interesting of these is that not all uses of slurs seem to express negative attitudes. Someone who utters

(4) Faggots are great interior decorators.

does not obviously express a negative attitude towards gay men.

I find this kind of example somewhat difficult to analyze, but it's important to keep in mind here that, at least as Camp understands the view she is criticizing, uses of slurs are supposed to express negative attitudes. This needs to be distinguished from whether use of a slur reveals such attitudes. One might think that someone who utters (4) does reveal such an attitude, but it's a different question whether they thereby express it. (One might also want to raise questions here about what Hom called 'derogatory autonomy'.)

On the other end are views like Hom's. Camp does a nice job explaining why there is a sense that someone who uses a slur (in the typical way) is making a mistake. But she notes that it is difficult, at best, to say exactly what, besides membership in a given group, the use of a slur asserts.

In section 3, Camp begins developing her view, "that slurs are so rhetorically powerful because they signal allegiance to a perspective: an integrated, intuitive way of cognizing members of the targeted group" (p. 335). Section 3.1 is devoted to outlining what a perspective is: a conception, usually implicit, of what features of members of the target group (in this case) are most important, in the sense that they explain or ground other features. (You might think about this in connection with implicit bias: The influence of perspectives on our thinking is meant to be "only partly under conscious control" (p. 336).) Perspectives, Camp thinks, structure our thoughts about a particular type of thing (including types of people): They highlight which properties of those things are important, and thus draw our attention to those properties; those same properties are taken to be central in the sense that they "cause, motivate, or otherwise explain many others" (p. 336). It will be important later that perspectives do not necessarily involve commit to any specific claims. They 'organize' our thoughts. They are not themselves thoughts or beliefs.

Camp also insists that "getting a perspective", that is, understanding what it is, "even temporarily, requires actually structuring one's thoughts in the relevant" way (p. 336); that is, it involves (at least temporarily) adopting that perspective oneself. This will turn out to be one of those cases where a philosopher says something, almost in passing, that one might not even have noticed but which turns out to be really important.

Does Camp give any argument for this claim? If so, what is it? Is it an empirical claim? or a conceptual one?

In section 3.2, Camp applies this notion to slurs, arguing that use of a slur "signals a commitment to an overarching perspective on the targeted group as a whole" (p. 337) or, at least, to the claim that being a member of that group is an important feature of a member of it, one that explains or grounds other features: It's because one is a member of the group (or because one has properties distinctive of the group) that one has other properties. In the case of bigoted perspectives, these other properties will be ones with negative valence. Bigoted perspectives thus ascribe negative properties to members of the group because those people are members of that group. Someone who holds or endorse such a perspective thus evinces disrespect for members of the target group that use of the slur derogates.

Camp takes disrespect to be the central attitude that slurs express. In particular, contempt and derogation are meant to be less central. Why? (See in particular pp. 338-9.)

Is it clear that racial bias (and other forms of bias) must have the sort of explanatory structure that Camp assumes? Must membership in the group, or possession of the properties that make one a member of the group, explain or ground the negative properties attributed to members of the group? How important to Camp's view is the claim that it must? (It's often important to disentangle different aspects of a view: What can one reject without rejecting the overall spirit of the view?)

In section 3.3, Camp argues that the connection between slurs and perspectives is, in some sense, semantic, i.e., that the associated perspective is part of the meaning of the slur itself. (Note that this does not commit Camp to Hom's view, since there can be other aspects of meaning besides truth-conditions.) The main argument here is simply that use of that word, as opposed to the neutral counterpart, seems to 'insert a way of thinking about the group into the conversation'. Precisely how the perspective is part of meaning, Camp does not say, but she does identify it as 'not at issue content', one standard example of which is conventional implicature (so one might, for the moment, take that to be Camp's view).

In what sense might it be conventional that the use of a slur 'signals' allegiance to a particular perspective? How specific does the perspective need to be?

Camp also suggests that we can explain, in such terms, "why the [slurs] produce a feeling of complicity in their hearers..." (p. 343). Camp identifies two forms of complicity. First, someone who hears the slur is, in effect, forced to adopt, even if only temporarily, the perspective associated with it; this is 'cognitive' complicity. Second, the utterance of the slur is a reminder of the social structures that give it weight. And silence equals agreement: "[I]f the hearer allows the slur's use to stand uncontested, she implicitly conspires in perpetuating its associated perspective's public currency" (p. 343). This is 'social' complicity.

Do both of these points seem right? In the latter case, one might add that, if one does not object to a slur, then one might in effect signal that one shares the perspective it evokes, thus inviting additional racist content. But the claim about cognitive complicity seems less secure. Camp cites some empirical literature in favor of the claim that at least some slurs have this kind of effect. But it is here that her earlier claim, that 'getting' a perspective requires at least temporarily adopting it, becomes important. How plausible is that claim?

---

If you read a lot of the footnotes, you can see that the view that Camp develops in this paper is a successor to a view she held earlier. It seems to me that, in some ways, it is actually easier to understand her current view if you see it in that light. So let me fill in this part of the story.

Camp's original view seems to have been as follows. Each slur is associated with a stereotype of the members of the group to which the slur purports to apply. I am guessing, given what else I know about Camp's views, that the notion of "stereotype" she had in mind was not (just) the ordinary one, but that it at least had elements of the notion going by that name that psychologists sometimes discuss. Stereotypes, in this sense, are supposed to play a role in how we sort objects into categories, how we organize our thought about them (what's typical, what goes with what), and so forth. We might then think of the "pejorative aspect" of a slur as, e.g., signaling a commitment to the appropriateness of a certain stereotype. And one can now imagine relatively easily what kind of story one might tell about why the use of slurs typically derogates the targeted group.

It would seem, however, that Camp became convinced that that view is too narrow: Some slurs, the claim seems to have been, are not associated with stereotypes. So Camp reformulates the view so that the role previously played by stereotypes is now played by what she calls 'perspectives'. These play a similar psychological role, and one might think of stereotypes as a kind of perspective. So much of the structure of the original view is preserved. But one might wonder if something else has been lost: Is it a good thing or a bad thing if these 'perspectives' can be so thin that there is little more to them than that a person's being a member of the 'targeted group' is supposed to be regarded as an important fact about them (e.g., important to who that person is)?

In this paper, Jorgensen offers an account of the offensiveness of slurs that is entirely grounded in pragmatics. It is important to understand that her goal is limited in precisely that respect: It is only the offensiveness of slurs for which Jorgensen is aiming to account. Slurs may have many other features, and she even mentions some of them on p. 2, and then again in the conclusion. Perhaps a semantic account is needed to explain some of those features. But Jorgensen is claiming that no semantic story is required to account for offensiveness. In that respect, her view agrees with Anderson and Lepore's, though the details differ.

Section 1 makes a number of useful distinctions concerning the notion of offense itself. In §1.1, Jorgensen distinguishes actual offense from (morally) warranted and (epistemically) rational offense. In §1.2, she notes that the hearer's offense may have its source either in the speaker's intention to offend, or in the inappropriateness of a remark (e.g., if I were use (not mention) the word "fuck" during class), or in various associations the word has. The latter are what matter most to Jorgensen. As she notes, almost all views regard there as being some such association between slurs and derogation. Her idea, basically, is to take that association as a basic sociological fact, and use it to explain offensiveness.

The central idea behind Jorgensen's account is that the "offense-generation profile" of slurs is strikingly similar to that of "rude" expressions, which include such terms as "damn", "bastard", and "motherfucker". Hence, she claims, we ought to seek an account of offensiveness that covers both rude terms and slurs.

In the case of rude terms, Jorgensen claims (drawing on work on impoliteness) that there is, simply as a matter of empirical fact, a significant correlation between the use of such terms and the holding of certain sorts of attitudes (e.g., anger, frustration, etc). If this correlation is widely enough known, then the use of a particular expletive may reliably 'signal' that the user holds those attitudes.

By contrast, we might also ask whether anything so strong as a convention is required here. Would a regularity alone be sufficient? Why or why not? If not, what more than a regularity seems to be required? That is: Are some but not all of the conditions on a convention required here?

Jorgensen's account of slurs, which she develops in §4, is similar. To borrow some language from Geoffrey Nunberg (in a paper listed as optional for Anderson and Lepore), the idea is that bigots do not use slurs because they are offensive, but rather: slurs are offensive because they are the words bigots use. For example, someone's using the term "faggot" to refer to gay men is reliably correlated with their holding hateful attitudes towards gay men, and that fact is widely known (as are the exceptions). To use that term, rather than "gay man" or some other non-pejorative equivalent, thus 'signals' that one holds such attitudes, precisely because that correlation is widely known.

In section 5, Jorgensen argues that her account provides a good explanation of the facts about offense she had articulated earlier. Maybe the most obvious place her account has an advantage is in explaining why 'forced choice' mentions of slurs are not offensive: If one did not have the option of using (or mentioning) the non-pejorative correlate, then no significance can attach to one's using (or mentioning) the slur (instead).

Jorgensen claims that her account explains why some slurs are more offensive than others: They are more strongly associated with more negative attitudes. (So there are two dimensions to the account.) Does this seem like the right explanation?

Perhaps the most striking Jorgensen's explanations is in section 5.4, where she discusses speech reports and quotation, and perhaps most interestingly the distinction between offensive and inoffensive mentions of a slur. In this case, the 'contrastive' aspect of Jorgensen's account is doing a lot of work. How so?

As noted, Jorgensen is out to explain the offensiveness of slurs. But a good account should also explain why slurs are such an appropriate vehicle for derogation, which is a focus of many other accounts. One possibility is that this is simply an independent matter. But one might also think that it can't simply be independent. Is there a way to extend Jorgensen's account to derogation?

There's a very interesting discussion of the uses of slurs in fiction, theater, and the like in section 6. I'll leave it to you, though, to comment on it should you wish to do so.

In §6.3.1, Jorgensen offers an account of reclamation. The idea is that use of the slur by people who do not hold the negative attitudes usually associated with its use "undermin[es] the signal strength" of that association (p. 17). Is that an adequate account?