3/17/97

On an Alleged Connection between Indirect Speech

and Theory of Meaning

Herman Cappelen and Ernie Lepore

An overlooked assumption in the semantic literature is about a connection between semantic content (whatever a semantic theory attempts to elucidate) and indirect speech. In simple but clumsy form the assumption is that an adequate semantic theory T for a language L should assign p as the semantic content of an utterance u, by A, of a sentence S in L iff éA said that pù is a true report of u.1 We'll call this assumption MA. So, since Galileo in uttering (1) makes (2) true, by MA, an adequate semantics for Italian must assign to (1) the semantic content that the earth moves.

1. La terra si muove.

2. Galileo said that the earth moves.

MA, and related assumptions, have important implications both for how to conceptualize the semantic program in general and for constraints on the semantics for indirect speech in particular. MA provides semanticists with an intuitive, pre-theoretic foundation for their investigations. The what-was-said by an utterance of a sentence is (alleged to be) something we pre-theoretically grasp. According to MA, a semantic theory attempts to reconstruct what this pre-theoretic notion picks out. In its other direction, MA severely restricts the semantics for indirect speech. A particularly striking implication is that there can be but one true indirect report of any utterance (on the assumption that there is at most one correct ascription of semantic content to an utterance of a sentence).

That a semantic theory should specify what was said by utterances of sentences seems innocent enough, but, so we shall argue, when this assumption is embodied by MA, semanticists both misconstrue the aim of semantics and unreasonably constrain the semantics for indirect speech. Before establishing that (and explaining why) MA fails, we'll make a case for how central MA is in the works of two philosophers with radically different views about the nature of semantic theory, namely, David Kaplan and Donald Davidson.

I. Kaplan and MA

According to Kaplan, a semantic theory must be anchored in speaker intuitions about ``what-was-said''. Here are two representative passages from ``Demonstratives'':

What is said in using a given indexical in different contexts may be different. Thus if I say, today, I was insulted yesterday and you utter the same words tomorrow, what is said is different...Let's call this first kind of meaning--what is said--content [1989, p.500].

If I may wax metaphysical in order to fix and image, let us think of the vehicles of evaluation--the what-was-said in a given context--as propositions [1989, p.494].

In ``Afterthoughts'', Kaplan writes:

The idea of content--the what-is-said on a particular occasion--is central to my account [1989, p.568].2

Scott Soames summarizes Kaplan's lead succinctly:

...the fundamental task of a semantic theory is to tell us what sentences say in various contexts of utterance. On this view, the meaning of a sentence can be thought of as a function from contexts to what is said by the sentences in those contexts [1989, p.394, our emphasis].

This appeal to what is said by an utterance of a sentence is not unimportant. Both writers, and in fact all contemporary intensionalists, develop elaborate technical frameworks within which they attempt to capture the notion of what is said. Were this notion entirely theoretical, it would be unclear exactly which phenomenon the technical apparatus was attempting to explicate or clarify or even describe.

We ourselves don't see how to elicit intuitions about what-is-said by an utterance of a sentence without appealing to intuitions about the accuracy of indirect reports of the form ``he said that....'' or ``what he said is that ...''. So, the intuitive basis for the project, as it applies, for instance, to (1), derives from reports like (2). Underlying this conception of semantics is some version of MA. We suggest that within Kaplan's and Soames' intensionalist framework, MA takes the form:3

The content of (i.e., the proposition expressed by) an utterance u of a sentence S, by A, is p iff éA said that pù is a true report of u.

If, as Soames tells us, ``the fundamental task of a semantic theory is to tell us what sentences say in various contexts of utterance,'' and if our grasp on what is said manifests itself in indirect reports, then MA is essential to this conception of semantics. We turn to another conception.

II. Davidson and MA

Davidson's commitment to MA is found not so much in what he says about the aims of semantics, but in what he says about indirect speech. According to Davidson, truth conditions for (2) are provided by (3) [Davidson, 1968]:

3. ($u)(Ugu & SSu,that): [The earth moves.]

The quantifier ranges over utterances; ``Ugu'' holds just in case Galileo uttered u; ``SS(u,that)'' holds when u and the utterance demonstrated by the occurrence of ``that'' samesay each other, where the demonstratum of ``that'' is an ensuing utterance of the bracketed sentence.

What few remarks Davidson makes about samesaying can be (and usually are) read as endorsing MA. Passages like the following encourage commentators to infer Davidson imposes MA as an adequacy condition on semantic theory.

The `that' refers to the second and the first utterance is true iff an utterance of Galileo's was the same in content (`translates') the utterance to which that `that' refers [1976, p.177, our emphasis].4

Platts, taking his cue from these passages, writes ``indirect discourse requires that the reporter reproduce the meaning of the original utterance: the contentsentence employed by the reporter should be a correct translation of the original speaker's utterance...'' [1979, p.126]. Likewise, Burge writes, ``the point of indirect discourse might be fairly taken to be to introduce and produce an utterance that gives the content of the original speaker's utterance...'' [1986, p.196]. More explicitly, Larson and Ludlow maintain ``it is arguable that one of the main charges of any semantic theory is to give...an account of the content of a given utterance: semantic theories should characterize what is said in uttering a given sentence S...In a truth conditional theory...content is ostensibly captured through the truthconditions that are assigned'' [1993, p.334]. And, lastly, McDowell writes:5

The basis of the truth-conditional conception of meaning, as I see it, is the following thought: to specify what would be asserted [i.e., said], in the assertoric utterance of a sentence apt for such use, is to specify a condition under which the sentence (as thus uttered) would be true [1987, p.60].

Other commentators seize upon passing comments Davidson makes about samesaying in relationship to radical interpretation:

The fact that an informal paraphrase of the [samesay] predicate appeals to a relation of sameness of content as between utterances introduces no intentional entities or semantics...It is...worth observing that radical interpretation, if it succeeds, yields an adequate concept of synonymy as between utterances [1968, note 12, p.104].

Rumfitt, inspired by Davidson's brief footnote, reconstrues (3) as (4):

4. ($u)(there is a truth theory T interpretative for Galileo as he produces u, which canonically entails: True(u) iff the earth moves) [1993, p.439].6

In effect, Rumfitt requires radical interpretation to respect MA. He believes that the ``dovetailing of the theory of indirect speech and the theory of understanding ...reflects ... the insight that what is said in the course of an utterance is nothing other than what somebody who understands the utterance understands to be said'' [1993, p.439]. Segal also interprets Davidson to mean that the practice of radical interpretation determines the samesay relationship [1989, pp.8485]. According to Segal, an utterance u of an English sentence s samesays Galileo's utterance of (1) only if there is an interpretive truth theory for Italian in English that assigns truth-conditions to (1) identical to u's (with some variations in `flavour', variations too subtle for the truththeory to detect are allowed) [Segal 1989, p.84; see, also, Seymour, 1994, pp.22, 26].

Our concern is whether any semanticist can sustain inflexible commitment to MA without ignoring our actual practice of indirect reporting. We shall argue MA fails in both its directions. Even superficial attention to our practice of reporting others reveals that, by anyone's estimates, some true reports have complement clauses differing in semantic content from the utterances they report and that some false reports have complement clauses matching in semantic content with utterances they report.

III. Counter-Examples to the Only-If Direction of MA

a. Partial Semantic Overlap

Clear counter-examples to the only if direction of MA are reports where complement clauses only partially agree in semantic content with reported utterances. (5) can be correctly reported by (6) even though its complement clause couldn't have the same semantic content as (5).

5. A: I bought a pair of Bruno Magli shoes and then I ate lunch.

6. A said that he bought a pair of Bruno Magli shoes.

Often reports eliminate adjectival or adverbial modifiers as in exchanges (7)-(8) and (9)-(10), again resulting in complement clauses that differ in semantic content.

7. A: I own a very expensive pair of brown Bruno Magli shoes.

8. A said that he owns a pair of Bruno Magli shoes.

9. A: I slapped him hard in the face.

10. A said he slapped him in the face.

A related class of counter-examples to the only-if direction of MA is so pervasive it's easy to miss. Indirect reports more often than not summarize what speakers say. Suppose Clinton goes on at length to describe his new economic program. A journalist might summarize him by reporting, ``Clinton says he'll cut taxes.'' This can be a true report of what Clinton said even if he never used these words or anything that could correctly be called a translation of them. (11)-(12) is a smaller scale example of this sort of transition.

11. A: At around 11 p.m., I put on a white shirt, a blue suit, dark socks and my brown Bruno Magli shoes. I then got into a waiting limousine and drove off into heavy traffic to the airport, where I just made my midnight flight to Chicago.

12. A said that he dressed around 11 p.m., went to the airport and took the midnight flight to Chicago.

To see how common these kinds of cases are, think about how often you are asked, just after delivering a paper or presenting a view in a conversation, ``So, what you are saying is that.....?'' Sometimes the correct answer is, ``Yes. Exactly''.

Another class of counter-examples involves substituting one expression for a co-referential one (or with a definite description that denotes the same object), as in (13)-(14).

13. Francois: Chartreuse is Maria's favorite color.

Someone, knowing his audience unfamiliar with the word ``chartreuse'', employs another means for conveying what Francois said. Demonstrating a chartreuse dress, she reports:

14. Francois said that the color of that dress is Maria's favorite color.

Surely, no adequate semantics interprets Francois' words with the complement clause in (14). Or suppose Orel says, ``I did this,'' pushing the door open. On the assumption that his audience lacks access to what Orel was demonstrating, a reporter may provide a description.

15. Orel said that he pushed the door open.

Suppose Bill says, ``Bob dislikes that guy,'' pointing at Andre. A reporter, realizing an audience more informed than Bill himself, may indirectly quote Bill with any of (16).

16. Bill said that John dislikes like your grandson/Andre/your boyfriend.

b. Non-Semantic Features Encoded in the Complement Clause

With many reports pragmatic non-semantic features of an utterance reported become encoded in the semantics of the complement clause of the report.

17. Pointing at a pink car, A asks, ``Do you like that car?''

18. B replies, ``I hate pink cars.''

19. A, pointing at that same car, reports, ``B said he doesn't like that car.''

No adequate semantics would (or should) interpret B's utterance by the complement of A's report.

When it's clear from context that A's utterance of ``Do you know what time it is?'' is meant as a question about what time it is, it's correct to report him as having asked what time it is. But suppose A utters to his daughter, upon her arrival home two hours after curfew, ``Do you know what time it is?''. It may very well be that he should not be reported as having asked what time it is.

Suppose A, convinced that Stanley is Smith's murderer, looking at Stanley says, ``Smith's murderer didn't comb his hair today''. B wants to report A's utterance to Mathilda, who is convinced Stanley is not Smith's murderer; Mathilda is also unaware of A's contrary belief and is also unfamiliar with the context in which A's utterance was made (staring at Stanley). B might report A to Mathilda with:

20. A said that Stanley didn't comb his hair today.

Even if a philosopher whose deepest conviction is that definite descriptions aren't ambiguous between referential and attributive uses can accept (20) as true.

Larson and Segal [1995] provide the following case. Suppose that X, after a particularly awful philosophy talk, says in a heavily sarcastic tone, ``That was, like, really good.'' X cannot be reported by (21), but can be by (22):

21. X said that talk was really good.

22. X said that he didn't like the talk much.

In these four cases, pragmatic features of a reported utterance (an indirect speech act in the first two cases, speaker meaning in the second two cases) are semantically encoded in the complement of a correct indirect report.

This is by no means the sole way in which indirect reports violate the only-if direction of MA.

c. Mixed Cases

Some counter-examples to the only-if direction of MA are more difficult to categorize. Suppose Professor H, when asked whether Alice passed her exam, answers with (23). (24) appropriately reports him.

23. I didn't fail any students.

24. Professor H said Alice passed her exam.

Suppose Professor H, when asked whether he passed all the students in his class, answers with (25) (when it is mutual knowledge between H and his interlocutor that Sally and Mathilda are his only students). (26) properly reports him:

25. I passed Sally and Mathilda.

26. Professor H said he passed the whole class.

Suppose George says, ``John leaves for Berkeley next week.'' It is not incorrect to report George to someone unfamiliar with Berkeley, seeking a ride to northern California, with (27):

27. George said that John's going to northern California next week.

Our claim is not that this report (or indeed any of the others) would be correct in every single context. All we are claiming, and all we need to claim in order to refute the only-if direction of MA, is that there are contexts in which this and the other reports would be true.

IV. First Generalization

Each of these cases presents an obstacle for MA. We venture further, conjecturing that (G):

G: For any possible utterance u of S in L and for any linguistic item w in S, there is a true indirect report of u not employing (any translation of) w.

The data adduced thus far certainly renders (G) plausible. But how then can MA be maintained?

The picture inherited from Frege is that a reporter is a mere conduit for information; a speaker asserts something with a specific content (e.g., a proposition); it's the reporter's duty to employ words that express exactly what the speaker expressed. What we've been noticing is that this is not at all how our practice of indirect quotation works. What is correct for us, as reporters, to do with a speaker's utterance is determined by all sorts of noninterpretive, non-semantic considerations.

V. Counter Examples to the If-Direction of MA

So far we have alleged that there are correct indirect reports of another's words where the complement clause does not semantically interpret the reported utterance in any obvious sense. The situation is worse; MA fails, surprisingly, in its if-direction as well. Suppose Benson says, ``A fortnight is two weeks.'' On the assumption that ``fortnight'' and ``two weeks'' are semantically equivalent, it would be hard to fault a semantic theory, especially a truth theory, that interprets Benson's words as a fortnight is a fortnight. They express the same proposition; they have the same truth conditions; they are synonymous; they are true in exactly the same worlds, and so on. Though there might be contexts in which (28) is an accurate report of Benson's words, there surely are contexts in which were someone to try to indirectly report Benson with (28), he would fail:

28. Benson said that two weeks are two weeks.

Suppose A is asked what Benson taught him in class today. As a reply, (28) seems to us incorrect. Anyone who knows Benson knows that he would never say anything so useless and uninformative.

Our previous sampling of correct direct/indirect quotation transitions showed it is sometimes unnecessary that the complement clause semantically interpret an original utterance; (28) shows it is sometimes insufficient as well.

For a less familiar counter-example to MA's if-direction, reconsider X's sarcastic remark, ``That was, like, really good'' (made after an especially bad philosophy talk). We argued that (22) incorrectly reports X.

22. X said that the talk was really good.

Suppose A reporter is asked about the audience's reaction to the talk; his task is to provide a quick and accurate account of what each member of the audience said. In this context, (22) would be incorrect. Or, suppose X is pressed about whether he didn't in fact say he really liked the talk. He could in all truthfulness reply that he never said any such thing and that anyone who thought otherwise doesn't understand the first thing about sarcasm.

VI. Two Attempts to Preserve MA

A typical reaction to alleged counter-examples to MA is to insist that any indirect report incompatible with MA is false. This reaction comes in two varieties. On one view, the problematic reports are simply false (call this the ``Strongly Dismissive Reaction''). Alternatively, some grant that as they stand the reports are true, but they become false once ``said'' is modified by ``literally'' or ``strictly speaking'' (call this the ``Modified Dismissive Reaction''). On either view, the aim is to preserve some version of MA.

a. Strongly Dismissive Reaction

One might respond to problematic reports by treating them as merely appropriate and reasonable in the contexts described, but not as true. MA accurately characterizes truth conditions for indirect reports; some pragmatic story can be invoked to account for why we occasionally find false reports still reasonable or appropriate.

We know of no argument for this reaction, unless dogmatic insistence on MA counts as argument. That's dialectically significant since anyone who insists on MA owes us an argument for why a semantic theory should ignore normal speaker intuitions about the conditions under which sentences of their language are true. It is simply a fact about English that indirect speech is used as we reported and that normal speakers of English would judge the indirect reports we advanced to be accurate in the situations we described. A semantic theory that disregards these intuitions faces several challenges.

It must explain why we have these mistaken intuitions. If it is a part of the semantics for English that the complement clauses of true indirect reports must have the same semantic content as the reported utterance, why aren't competent speakers of English aware of this? In explaining this, the theory must also explain why truth is largely irrelevant to the success of our practice of indirect reporting.

Such a theory must also explain why we have a linguistic practice in which people propagate false claims. Is there any functional reason behind this? It would be exceedingly strange if the meaning of our verb ``to say'' were radically different from what we had always thought. Why hasn't its meaning simply changed? Why hasn't our linguistic practice died out, and been replaced by one in which no gap exists between what we think our verb means and what it really means?

However, it is not worth the effort to try to meet these challenges until proponents of the Strongly Dismissive Reaction advance an argument to show that there is some important theoretical advantage associated with denying our shared intuitions. Arguably, a semantic theory must sometimes ignore speaker intuitions. But when it does, it must be for some theoretical advantage.

b. No Arguments for MA

As far as we know, the arguments that seem to underlie commitment to MA are of the form:

If the task of a semantic theory is to...., then MA.

We know of no reasonable substitute for ``...'' from which MA follows; and for reasons we've already cited, it would surprise us were there such a connection. If you doubt us, plug in your own favorite view about the task of semantics into the schema. We'll provide two illustrations.

Suppose the task of a semantic theory for L is to assign truth conditions to utterances of sentences of L. Nothing about indirect speech follows. Nothing in this conception of the task of semantics implies that the complement clause of an indirect report must have the same truth conditions as the utterance reported.

Suppose, instead, that the task of a semantic theory for L is to determine which proposition is expressed by an utterance of a sentence of L. Again, nothing about indirect speech follows. One could, of course, define ``the proposition expressed'' in such a way that it is inextricably intertwined with indirect speech. If ``the proposition expressed by an utterance u'' is defined as whatever is expressed by the complement clause of a true indirect report of u, then a version of MA is stipulated. Needless to say, it is bizarre to appeal to a stipulative connection as the basis for an argument for overriding speaker intuitions about our actual reporting practices.

c. Modified Dismissive Reaction

Sometimes we encounter the response that the problematic reports, though true, would be false if ``said'' were modified by ``literally'' or ``strictly speaking''. On this view, a special class of indirect reports of the form éA literally/strictly speaking said that pù creates a connection between indirect speech and semantic theory relevantly similar to the connection we've been denying. As a claim about how the expressions ``literally'' and/or ``strictly speaking'' function in our language, this reaction fails.

Recall from above where A says, ``I bought a pair of Bruno Magli shoes and then I ate lunch..'' We claimed he could be accurately reported by (6):

6. A said that he bought a pair of Bruno Magli shoes.

The question, ``Was that literally what A said?'' could be answered negatively only if ``literally'' is understood as transforming our question about A into one about whether (6) is a correct direct quote of his words. If this is right, then a negative answer to the question so understood does not underwrite even our modified MA. (The same could be said about ``Was that strictly speaking what A said?'')

Or take the journalist's report of Clinton's speech. Clinton says he'll cut taxes. Is this literally what Clinton said? If not, then perhaps, again, we're being asked for a direct quote of Clinton's words.

In short, actual usage of the modifier ``literally'' does not support a revised version of MA. If anything, ``literally'' can be used to ask whether an indirect quote is also a direct quote. A proponent of the Modified Dismissive Reaction might insist on using ``literally'' theoretically. If so, no independent support for MA is provided.

VII. Diagnosis

It would be surprising were there a complete overlap in extension of a technical term used by semanticists and the English verb ``to say''. Given how different the aims of semantics and ordinary (indirect) reporting are, we should not expect features of utterances that occupy semantics to be captured exactly by any natural language expression.

Semanticists disagree on what central semantic features are (truth conditions, intensions, extensions, propositions, functions from worlds to truth-values, or whatever), but they tend to agree that semantics is a discipline that aims to characterize systematically certain features of linguistic expressions and to do so in a way that captures general truths about our linguistic practice, not just truths about particular speakers in specific contexts. Indirect speech, on the other hand, is a device reporters use for characterizing acts (utterances) performed by other speakers. In so doing, reporters are interested neither in systematicity nor in generality; they aim is to convey something about a particular act in a particular context C to a particular audience situated in a different context C*. Reporters draw on information about the specific intentions of, knowledge about, and the history of a reported speaker in C and (maybe) similar features of an audience in C*. These are features one does not want to solicit when the aim is systematic and general. For this reason, we should not expect semanticists to be able to exploit the standard usage of the English locution ``says that'' in exacting features of utterances that interest semanticists.

VIII. Second Generalization

Our point is not particularly about ``says that''. Why should any natural language expression select just those features of utterances that engage semanticists? What applies to ``says that'' applies equally well, for example, to ``means that''.

Semanticists are fond of saying their aim is to characterize the meanings of (utterances of) sentences. This might just be an innocent gesture towards the sort of issues that occupy semantic theory, but as an adequacy condition on semantic theory, on the order of M, it is no less misguided than MA.

M: A semantic theory T for a language L assigns p as the semantic content of an utterance U, by A, of a sentence S in L iff éU means that pù is true.

Just as a proponent of MA is mistaken about our practice of indirect reporting, so too a proponent of M would be mistaken about how ``means that'' is actually used. A thorough investigation into how ``means that'' is used exceeds the scope of this paper, but we'll provide a few illustrations (and leave it as an exercise to the reader to generate more).

If, after X's sarcastic utterance of ``That was, like, really good'', someone asks what that utterance means, it is difficult to imagine contexts in which ``It means that that was like really good'' is a correct response, and it's easy to imagine contexts in which ``It means that the talk was awful'' is a correct response. If Clinton describes a complex new property tax, and someone confused by all the technical jargon asks, ``What does that mean?'', a correct answer might be, ``It means he'll raise property taxes'', but it would never be a repetition of Clinton's words. If Ludwig points to Mathilda and utters, ``I like that person'', it is, again, easy to think of contexts in which it is correct that that utterance means that he liked Mathilda, and if Mathilda is Sam's sister, there are contexts in which it is natural to say that it means that Ludwig likes Sam's sister. And so it goes.

It is tempting to interpret some questions as about speaker meaning or as requests for explanation; and neither of these is what proponents of M seek. They seek pure intuitions about what (utterances of) sentences mean. But the actual use of locutions like ``means that'' (and ``said that'' and ``believes that'') reveals no ``pure'' semantic intuitions about these uses. Intuitions about when such reports are true, it seems, is always cut with non-semantic material: often they are about what a speaker said, sometimes about what he meant, sometimes attempts to explain what was said, sometimes they provide lexical or syntactic or pragmatic information, almost invariably they are a mix of several of these.

Philosophers like to think they can overcome these problems by ``refining'' intuition; if intuitions are refined in just the right way they become pure, i.e., they isolate the genuine ``means that'' (or ``says that'') reports. The problem with such refinement is that there is no difference between it and developing a theory. To ``refine'' intuitions is to become more discriminating about which features our pre-refined selves should respect in making ``means that'' reports. If actual ``means that'' reports are sensitive to features F1, F2,...,Fn, a refined ``means that'' reporter might be sensitive, say, only to F1 and F2. Is that refined reporter making more genuine ``means that'' reports than non-refined reporters? Of course not. What he has done is developed a theory in which it's important to ignore F3,...,Fn. There is no pre-theoretic phenomenon, the meaning, he segregates better than his non-refined counterparts.7

As semanticists we begin with a pre-theoretic notion of what interests us. This can no doubt be approximated by some cluster of natural language terms. As our project becomes refined, there is no reason to wed the discipline either to one particular (cluster of) natural language terms or to intuitions about correct usage of these terms. There is no reason why the features that interest semanticists should be all and only those of the objects in the extension of some one particular (or even a cluster of) natural language expression(s).

IX. Implications of Renouncing MA

Relinquishing MA has implications both for the semantics of indirect speech and for methodological questions about the aims of semantics.

a. Implications for the Semantics of Indirect Speech

Reports like those we have discussed make evident that indirect reports are sensitive to innumerable non-semantic features of reported utterances and even of the context of the report itself. As a result, typically there will be indefinitely many correct indirect reports of any particular utterance. Commitment to MA forces us either to conclude that every sentence has indefinitely many semantic contents or just to ignore this basic fact about indirect speech. But, then, once we give up MA, what becomes of the relation between semantic theory and indirect speech? Minimally, of course, a semantic theory must assign semantic contents to indirect quotes. No positive theory of indirect quotation follows from forfeiting MA. We believe, though we won't argue for it in detail here, that without MA it's easier to defend a revised version of Davidson's paratactic account of the semantics for indirect speech.8 As we acknowledged above, such theories invoke a samesay relation. But unlike what others conclude, we believe our data establishes that the perfectly viable practice of indirect speech requires the samesay relation to be broader than MA permits; it's no role for semantic theory to place a priori constraints on what can samesay what. Whether two utterances samesay each other often depends on nonsemantic considerations. Competent speakers of English, those who competently use the ``says that'' locution, are able to render such judgments. Having such competence consists, in part, in being able to judge whether a given report is correct or incorrect, accurate or inaccurate, misleading or exactly what the speaker said. Therefore, the task of characterizing our competence with ``says that'' locutions will be no simpler than, say, accounting for how we can make judgments of similarity or how we can classify objects as being of the same color.

Also, without MA, we have a reply to what many find the most persuasive objection against Davidson's own account of indirect quotation. According to Loar [reported in Schiffer 1987, pp.131133; see also Burge 1986, pp.193194, Blackburn 1975, p.184, and Haack 1971, pp.35657], Davidson's account results in a violation of principle (P):

P: If the occurrence of t in éA said that...t...ù is primary and refers to x, then that sentence is true only if A referred to x,

where an occurrence of a singular term in a sentence is primary iff it is not properly contained within the occurrence of another singular term. Suppose Galileo's utterance of (1) is reported with (2). According to (P), (2) is true only if Galileo referred to the earth in making his utterance, which he did. However, an utterance of (29) may be true even though Laplace never referred to an utterance of English.

29. Laplace said that Galileo said that the earth moves.

But, so Loar's argument goes, the Davidson paraphrase of (29) is (30):

30. ($u)(U(l,u) & SS(u,that)): [Galileo said that the earth moves]

According to (P), an utterance of (30) is true only if Laplace referred to whatever the second occurrence of ``that'' demonstrates, since it has primary reference. But its referent is an utterance of ``the earth moves'' and Laplace never made any such reference. Since (an utterance of) (29) has no such implication, Davidson's account must be false.

Why should anyone endorse (P)? It is a corollary to MA; so, commitment to MA is commitment to (P). But since MA is false, appeal to it cannot underwrite (P). Is there independent evidence for it? Surely, some semantic properties of the complement clause of an indirect quotation must match those of the reported utterance. It's just not obvious to us that the referential properties of the complement clause of an indirect quotation must match those of the reported utterances, especially since the very sorts of arguments we ran against MA suffice to undermine (P).

Recall George's utterance of ``John leaves for Berkeley next week.'' As noted above, it need not be incorrect to report George to someone seeking a ride to northern California with (27):

27. George said that John's going to northern California next week.

Recall also, when Professor H was asked whether Alice failed her exam, he replied ``I didn't fail any students'' and was correctly reported by (24):

24. Professor H said Alice passed her exam.

Or suppose Billy says, ``I want to visit mom and dad during the holidays.'' In response to cynics who think no one still visits family during the holidays, we can correctly report Billy with (31):

31. Billy says that he will visit his family these holidays.

Or if in reply to a concerned friend's query, ``How is your arm?'', Albert replies, ``I'll have surgery tomorrow,'' we can report Albert correctly with (32).

32. Albert said that he'll have surgery on his arm tomorrow.

Similar stories can be devised for transition pairs (33)(34 and (35)(36):

33. Clinton said, ``I'm will send troops to Bosnia.''

34. Clinton said that the United States will send troops to Bosnia.

35. Michael said, ``Names are not rigid designators.''

36. Michael said that Kripke is wrong about names.

None of this means the semantic features of the original utterance are never relevant in determining what samesays what. Rather, again, the lesson to be learned from closely attending to our actual practice of indirect reporting is that those features which relevant are and those which are not is not determinable a priori.

b. Implications for Semantics

That MA (and relatives like M) fails establishes only that certain conceptions of semantics are wrong; nothing follows about what semanticists should be doing. Throughout this paper, we assumed that semantics aims to characterize systematically certain features of linguistic items and to do so in a way that captures general truths about linguistic practice, not just truths about a particular speaker in a specific context.

Even if the output of linguistic activity (i.e., what was said, meant, asserted, claimed) is, as we have argued, context sensitive to the extreme, it does not follow that there is nothing general, systematic, or non-context sensitive to resolve about how we determine what-was-said (or meant, asserted or claimed). Quite the contrary, in every conversational contexts we described, English speakers shared a core of linguistic knowledge and abilities, i.e., they shared linguistic competence. That shared linguistic competence (partially) underwrites agreement about what a speaker said in a particular context. Without shared linguistic competence, common interpretations would be miraculous.

Even if non-semantic features are more important determinants of what was said than semanticists under the spell of MA have tended to assume, it doesn't follow that words and sentences lack stable cross-contextual features, features knowledge of which together with non-linguistic knowledge enable speakers to use those words and sentences to say things. It's these cross-contextual stable features that semantics is about.

Which features are these? No easy answer follows simply from denying MA, but it's worth noting that a Davidson-like position gains considerable support from its denial. According to Davidson, a theory of meaning for a language should take the form of a Tarski-like truth theory [1967].We suggest this be understood as:

The features words and sentences share across contexts, features which together with non-linguistic information enable speakers to interpret another's utterances, are the business of a truth theory.

When Davidson's position is detached from MA, it becomes considerably easier to defend. The claim is no longer that the truth theory alone suffices to capture the extension of the intuitive notion of what-was-said, or what-was-meant. The point of our paper is that it can't. Rather, our claim is that a truth theory characterizes only part of the linguistic competence which enables speakers to determine what-was-said, what-was-meant. Seeing this is so enables us once and for all to put to rest Foster's problem -- the bugaboo of the entire Davidsonian semantic program.

Foster [1976], Loar [1976], and Fodor and Lepore [1992], among many others, object that the trouble with an extensional semantic theory, for example, a truth theory, is that its theorems are too weak. A truth theory for Italian should issue in (37):

37. ``La terra si muove'' is true in Italian iff the earth moves.

But (37) says nothing about what (1) means; it discloses no more than that it's left and right hand sides are materially (or extensionally) equivalent. But, how then can a truth theory for Italian be used to interpret utterances of (1)?

Suppose Galileo utters (1). Anyone who understands Italian is in a position to indirectly report Galileo with (2). How on earth can he do that if all she has available is information as impotent as (37)? (37) says no more about the meaning of (1) than does (38):

38. ``La terra si muove'' is true in Italian iff the moon moves.

Both (37) and (38) are true since the left and right hand sides of each are true.

Davidson, in his reply to Foster, accepts the challenge and modifies his theory to accommodate the weakness of extensional semantic theories. Davidson adds that in order to know what is said by (utterances of) sentences of L it's not enough to know a true truth theory for L, one must also know that that truth theory meets certain empirical constraints [Davidson 1976]. Armed with this additional constraint, one recognizes that theorems like (37) are not mere material equivalences, but are rather semantic laws.

Both the initial criticism and Davidson's reply (and every other we are aware of) disclose a misconception about the goals of semantic theory. Someone not already in the grips of MA could never be seduced by Foster's alleged problem. Recall, according to Davidson, the logical form of (2) is (3):

3. ($u)(U(g,u) & SS(u,that)): [The earth moves]

An utterance u of (2) is true, according to his account, just in case an Galileo's utterance of (1) samesays u's subutterance of ``the earth moves.'' The mistake everyone party to this dispute makes is to assume that it's the semantics that's supposed to specify the complement clauses of socalled ``intensional'' idioms like ``says that'' (or ``can be interpreted as''). Foster must be worrying how a purely extensional truth theory can underwrite substitutions into such ``intensional'' contexts? Since it cannot, he concludes such theories fail to be interpretive. But this just is MA. Give it up, and the proper response to Foster's problem is neither to strengthen the semantic theory (a la Foster 1976) nor the requisite epistemic attitude toward the theory (a la Davidson 1976), but rather to see the problem as a pseudoproblem.9

Bibliography

Arnauld, R, ``Sentence, Utterance, and Samesayer'', Nous 10, 1976, pp.283-304.

Baldwin, T. ``Prior and Davidson on Indirect Speech'', Philosophical Studies 42, 1982, pp.255-282.

Blackburn, S., Spreading the Word, Clarendon Press, Oxford, 1984.

Burge, T., ``On Davidson's 'Saying that'``, in Truth and Interpretation, Basil Blackwell, Oxford, ed., E. Lepore, 1986, pp.190-208.

Cappelen, H. and E. Lepore, ``Varieties of Quotation'', Mind, forthcoming, 1997.

Davidson, D., ``Truth and Meaning'', reprinted in IT, 1967, pp.17-36.

Davidson, D., ``On Saying That'', reprinted in IT, 1968, pp.93-108.

Davidson, D., ``Thought and Talk'', reprinted in IT, 1975, pp.155-170.

Davidson, D., ``Reply to Foster'', reprinted in IT, 1976, pp.171-180.

Davidson, D., (IT) Inquiries into Truth and Interpretation, OUP, Oxford, 1984.

Evans, G. and J. McDowell, eds., Truth and Meaning, OUP, Oxford, 1976.

Fodor, J. and E. Lepore, Holism: A Shopper's Guide, Basil Blackwell, Oxford, 1992.

Foster, J., ``Meaning and Truth Theory'', in Evans and McDowell, 1976, pp.1-32.

Haack, R.J., ``On Davidson's Paratactic Theory of Oblique Contexts'', Nous 5, 1971, pp.351-61.

Hand, M., ``On Saying That Again'', Linguistics and Philosophy 14, 1991, pp.349-365.

Kaplan, D., ``Demonstratives'', in J. Almog, et. al., (eds), Themes from Kaplan, 1989.

Larson, R. and P. Ludlow, ``Interpreted Logical Forms'', Synthese 95, 1993, pp.305-356.

Larson, R. and G. Segal, Knowledge and Meaning, MIT Press, Cambridge, MA., 1995.

Lepore, E. and B. Loewer, ``You can say that again'', Midwest Studies in Philosophy, eds., H. Wettstein, et. al., 1989, pp.338-356.

Loar, B., ``Two Theories of Meaning'', in Evans and McDowell, 1976, pp.138-161.

McDowell, J., ``Truth Conditions, Bivalence, and Verificationism'', in Evans and McDowell, 1976.

McDowell, J., ``In Defence of Modesty'', in Michael Dummett, eds. B.M.Taylor, Martinus Nijhoff Publishers, Dordrecht, 1987, pp.59-80.

Platts, M., Ways of Meaning, RKP, London, 1979.

Rumfitt, I., ``Content and Context: The Paratactic Theory Revisited and Revised'', Mind 102, 1993, pp.429-453.

Schiffer, S., Remnants of Meaning, MIT Press, Cambridge, MA., 1987.

Segal, G., ``A Preference for Sense and Reference'', Journal of Philosophy 86, 1989, pp.73-89.

Seymour, M., ``Indirect Discourse and Quotation'', Philosophical Studies 74, 1994, pp.1-38.

Soames, S., ``Direct Reference, Propositional Attitudes, and Semantic Content'', Philosophical Topics 15, 1989, pp.47-87.