Forthcoming in MIND


Herman Cappelen and Ernie Lepore

Language is the instrument it is because the same expression, with semantic features (meaning) unchanged, can serve countless purposes - Davidson (1968).

Suppose Alice utters (1). She can be properly quoted by any of (2)-(4):

(1) Life is difficult to understand.

(2) Alice said, `Life is difficult to understand'.

(3) Alice said that life is difficult to understand.

(4) Alice said that life `is difficult to understand'.

(2) quotes Alice by mentioning the words she uttered. This is direct quotation. (3) quotes her, but could be true even if Alice never uttered any word in (3). This is indirect quotation. (4) quotes Alice by reporting what she said, but attributes to her only an utterance of `is difficult to understand'. Call this mixed quotation.1 If Alice utters (5), she uses quotation not to report what another says, but simply to talk about linguistic expressions. Call this pure quotation.

(5) `Life is difficult to understand' is a sentence.

I. Interactions among Varieties of Quotation2

These varieties of quotation interact in interesting and largely overlooked ways. Consider first the following exchange:

A: Alice said that life `is difficult to understand'.

B: She did not; she said that death is difficult to understand.

A and B disagree. One uses mixed quotation, the other indirect quotation. That A and B disagree indicates that mixed and indirect quotation function in overlapping ways. We take this as evidence for:

C1 Mixed and indirect quotation should receive overlapping semantic treatments.

C1 gains additional support from the fact that joint mixed and indirect quotations admit of certain sorts of conjunction reduction. (6) is true only if (7) is:

(6) Alice said that life stinks and she said that life `is difficult to understand'.

(7) Alice said that life both stinks and `is difficult to understand'.

Consider next the dispute between A and C:

A: Alice said that life `is difficult to understand'.

C: No! She said, `Life is not difficult to understand'.

That C disagrees with A shows that direct and mixed quotation must admit overlapping accounts, lending support to:

C2 Direct and mixed quotation should receive overlapping semantic treatments.3

C could disagree with someone who uttered (3) only on the (not uninnocent , at least in this context) assumption that C is directly quoting English,4 and so:

C3 Direct and indirect quotation should receive distinct semantic treatments.

Finally, consider the inference from the direct quote (2) to the pure quote (8); and similarly from the mixed quote (4) to the pure quote (9):

(2) Alice said, `Life is difficult to understand'.

(8) A token of `Life is difficult to understand' was uttered.

(4) Alice said that life `is difficult to understand'.

(9) A token of `is difficult to understand' was uttered.

A natural explanation for these inferences is:

C4 Quotation5 in pure, direct, and mixed quotation should receive overlapping semantic treatments.

Lastly, C1-C4 are evidenced further by considerations about understanding pure, direct, mixed, and indirect quotation. Quotes in (2) function as they do in (5). All one need learn to extend an understanding of pure quotation to direct quotation is an understanding of the verb `say', as it functions in contexts such as (2). Furthermore, understanding (2) and how indirect quotation, as in (3), functions suffices for understanding (4). So, it would seem that all that's needed to understand mixed quotation is a prior understanding of pure, direct, and indirect quotation.

Surprisingly, though much is in print on the semantics of indirect quotation, some on the semantics of pure and direct quotation, and even a little on the semantics of mixed quotation, no one, as far as we know, has ever tried to develop an account that satisfies constraints C1-C4. That is our chief aim in this paper. The assorted data thus far adduced provide strong support for the desirability of an account satisfying C1-C4 and show at least that any semantics not satisfying C1-C4 leaves much unexplained.6

We will offer a semantics for indirect, pure, direct, and mixed quotation, erected upon Davidson's accounts for indirect and pure quotation [Davidson 1968, 1979]. We begin by raising doubts about whether leading semantic accounts for indirect quotation and for pure (and direct) quotation can accommodate mixed cases in any non-ad hoc manner. Since no author attempts to satisfy C1-C4, our discussion will proceed by first showing briskly that the influential views on the semantics of indirect quotation cannot easily be extended or refined to accommodate mixed quotation and then equally briskly showing that the leading accounts of pure (and direct) quotation cannot either.

II. The Semantics of Indirect Quotation

Most accounts of the semantics of propositional attitude reports share features (A) and (B):

(A) Propositional attitude reports assert that a relation obtains between an agent and a proposition (or a proposition-like content); and

(B) A propositional attitude report éA Ved that pù (for some propositional attitude verb V) is true only if the proposition (or proposition-like content) expressed by the complement clause p matches the proposition (or proposition-like content) of the agent's attitude.

Restricted to indirect speech, (A)-(B) require that (3) is true iff Alice uttered something that matches the propositional content of its complement clause. Theories incorporating (A)-(B) differ on what `matches' means and on what exactly propositions or proposition-like contents are. Mixed cases, we believe, show that any theory incorporating (A)-(B) is inadequate. We shall illustrate the problem by discussing recent influential theories in which (A)-(B) are central components.

According to Soames, (3) is true iff Alice assertively uttered a sentence S in an associated context C such that for some S¢ that can be readily inferred from S, the content of S¢ in C = the content of (1) in the context of the report [1989, p.411]. If Soames' account is extended innocently to mixed quotation, then an utterance of (4) is a true report of Alice's utterance u, say, of (1) iff `Life is `difficult to understand'' expresses the same proposition as (an utterance of) a sentence that can be readily inferred from (1). But which proposition could this be? Assuming quotation marks are functioning normally here (i.e., functioning as a singular noun phrase referring to an expression-type), it is not obvious that `Life is `difficult to understand'' expresses any proposition at all. [Indeed, it's not even clear whether it is well-formed in English.] But even if it expresses a proposition, how could it express anything readily inferable from (1)? After all, aren't (T1) and (T2) obvious truths about (4) and u?

T1 Alice was talking about life; she was not talking about words. In no sense did she express a proposition that was about linguistic entities.

T2 The complement clause of (4) contains quotation marks and is therefore, in some sense, about words; there is some sense in which the proposition expressed by the complement clause of (4) is about linguistic entities.

Hence, if (4) were an example of indirect quotation, then we could indirectly quote another without using a complement clause matching the content of the reported utterance, contrary to Soames' account.

Can the notion of readily inferable rescue (our extension of) Soames' account? He says little about what is required for one sentence to be readily inferable from another. His only example is an inference of a conjunct from a conjunction [1989, p.411]. But is there any sense to the idea of readily inferring from (1) a sentence identical in content with `life is `difficult to understand''? First, as just noted, `life is `difficult to understand'' does not seem to express a proposition. Hence, what could be inferable from it? Also, even if there is a sense in which an utterance of `life is `difficult to understand'' says something, Soames' account can be rescued only if a sentence about words can be readily inferable from a sentence not about words. How this might be is a mystery for us.

For another illustration of the challenge mixed quotation poses for standard accounts of the semantics of indirect speech (i.e., accounts respecting (A)-(B)), consider accounts according to which propositional attitude reports relate an agent to an Interpreted Logical Form (ILF).7 The basic idea is that propositional attitude reports express relations between agents and `annotated constituency graphs or phrase markers whose nodes pair terminal and nonterminal symbols with a semantic value' [Larson and Ludlow, 1993, p.305]. An ILF is a logical form in the sense of Chomsky [1981] augmented by semantic values at each node in the phrase marker. So, an ILF effectively incorporates three types of information: the semantic content of an utterance (the `proposition expressed'), the logical structure of that semantic information, and the lexical means through which that semantic content is passed along. ILF theorists hope to appropriate various successes of diverse traditions in blocking unwanted inferences between propositional attitude reports. Adapting an example from Larson and Segal [1995, pp.438-40], `Peter said that Lori met Cary Grant' is true just in case Peter said


NP:[Lori] VP:[Lori]

N:[Lori] V:[<Lori, Cary Grant>] NP:[Cary Grant]

N:[Cary Grant]

Lori:[Lori] met:[<Lori,Cary Grant>] ary Grant:[Cary Grant]

where semantic values appear in the square brackets to the right of each node.

No ILF theorist discusses mixed, pure, or direct quotation. One thing is clear, however: if an ILF theory is to treat quotation innocently, then, inter alia, quotes must function in the same way and have the same semantic value whatever linguistic context they occur in. (Larson and Segal explicitly endorse semantic innocence [1995, pp.436-7]; so do Larson and Ludlow [1993, p.332].) So, for mixed cases, e.g., (4), expressions quoted must be included as semantic values in the ILF of the complement clause (because the semantic value of a quotation is the quoted expression) and it must contain the quotation itself as a lexical element; e.g., there must be a node in which ``is'' occurs as a lexical item and the semantic value of `is', whatever that may be, occurs as well. On this view, (4) claims Alice stands in the saying-relation to such an ILF. However, to the extent that it is clear what it is to stand in the saying-relation to an ILF, Alice does not bear this relation to any ILF as described above.

ILF theorists disagree among themselves about whether standing in the saying-relation to an ILF is to any extent pragmatically determined. If it were, then we could exploit this feature of their theory to reply to a number of objections (e.g., how it is that Caesar can stand in the saying-relation to an ILF containing English words; how it is that `Cary Grant' and `Archie Leach' can be (sometimes) exchanged without reversing truth-value in indirect speech; how it is that two speakers can samesay each other even though their respective words have distinct references8). Though we endorse appeals to pragmatic considerations in an effort to characterize our practice of indirect quotation (and other propositional attitude attributional practices), such appeals cannot solve the problem posed by mixed quotation. It is not a pragmatic question whether a mixed report requires the reported speaker to stand in the saying-relation to an entity containing lexical elements as semantic values. Typically, a mixed quoted speaker said nothing about words and it is a straightforward semantic fact that he did not. Any theory that does not respect this fact, or tries to dodge it by deporting it to pragmatics, is inadequate.

In broad outline, a number of theories satisfying (A)-(B) differ from Soames' inasmuch as they, so to speak, add elements to the proposition expressed by the complement clause in order to fine-grain a report (or fine-grain it in what they take to be the right way), and then say that the report is true iff the reported speaker stands in the saying-relation to this new entity. In addition to ILF theorists, proponents of such views include Crimmins and Perry [1989] (and Crimmins [1992]), and Richard [1990]. According to all such theories, a propositional attitude report is true just in case an agent stands in a certain relation, e.g., the saying-relation, to the content of the complement clause. Each adds new components¾for example, ILFs, notions, or lexical elements¾to the proposition (or some proposition-like entity) expressed. By letting objects of the attitudes be propositions (or something proposition-like) expressed plus something, it is still a requirement that the speaker, or believer, or... stands in the relevant relation to the entity expressed.

Whatever the merits of these maneuvers, with respect to mixed cases they establish no advance over Soames' less baroque version. The content expressed by the complement clause of a mixed case is about words. The person reported by, say, (4), did not utter anything about words. So, if any theory requiring for the truth of an indirect quotation that the reported speaker stand in the saying-relation to the proposition expressed by the complement clause cannot accommodate mixed cases.

III. A Reply

Are we exaggerating the significance of mixed quotation? A dismissive reaction to mixed cases is the following modification of Soames' view:9 Transfer the contribution made by the quotation marks outside the scope of `says that'. The quotes of mixed quotation behave like an afterword, somewhat like adding `and, by the way, she used these words in saying it'. This observation might lead one to think that even though other theories do not explicitly try to account for cases like (4), it is fairly easy to do so if (4) is construed along the lines of (10):10

(10) Alice said that life is difficult to understand and she said it uttering, in part, the words `is difficult to understand'.

First of all, (10) fails to account for a basic fact about mixed quotation: mixed quotation contains a component that serves two functions concurrently. The quoted part is both employed to report what the speaker said and it is employed to say, at least partially, what the speaker actually uttered. As Davidson notes, `[some tokens] double duty, once as meaningful cogs in the machine of the sentence, once as semantically neutral objects with a useful form' [1979, p.92].

In (4), e.g., `is difficult to understand' serves two functions. It (together with other words) functions to report what Alice said, namely, that life is difficult to understand; but it also functions minimally to report that Alice tokened `is difficult to understand'. So, `is difficult to understand' serves two functions, without incurring ambiguity. In (10), no single component serves these two functions; and so, (10) fails to explain how a token can have this double function; it just states that it has it.11

Here's another objection to (10). Suppose Nicola utters `Alice is a philtosopher'. She can be correctly mixed quoted by (11):

(11) Nicola said that Alice is a `philtosopher'.

Mixed quotes like (11) are not uncommon. Often we hear others using unknown words (i.e., a word not part of our vocabulary). An utterer of (11) might have mixed quoted Nicola's utterance of `Alice is a philtosopher' because he is uncertain about what `philtosopher' means. He might assume that Nicola's vocabulary is larger than his and mixed quotes her to indicate that this is a word unknown to him. Alternatively, he might be convinced that Nicola is linguistically incompetent and wants to make this transparent without himself committing what he thinks is Nicola's mistake. On either scenario, mixed quotation is used to report what someone said when part of what was said is unintelligible to the reporter himself.12

It should be obvious that an account of mixed quotation along the lines of (10) cannot be extended to (11). A natural extension would be (12):

*(12) Nicola said that Alice is a philtosopher and she said it using, in part, `philtosopher'.

But in uttering the first conjunct of (12), a normal English speaker fails to report anything at all. Since a speaker can make a correct report by uttering (11), the account adumbrated by (10) fails as an account of mixed quotation.

We cannot prove that the leading accounts of the semantics of indirect quotation cannot be extended to account for mixed quotation, but if C1-C4 are acceptable constraints, these accounts face a so far unmet challenge.

IV. The Semantics of Pure Quotation

The most influential accounts of pure quotation also fail to extend to the mixed cases. The two prominent accounts of quotation are the proper name account and the description account; both preclude satisfying C1-C4.

According to the proper-name account [e.g., Quine 1961 p.140 and Tarski 1956, p.159], quotations are unstructured proper names of the quoted expressions. There is no systematic correlation between what occurs inside the quotation marks and the semantic value (i.e., the referent) of the entire quotation. Applying the proper name account to (4), unfortunately, results in ungrammaticality. Since quotations are just names, (4), on this account, semantically (and syntactically) parallels (13):13

*(13) Alice said that life Manhattan.

According to the description account [e.g., Quine 1960 p.202, Tarski 1956, p.160, Geach 1957, p.79, and 1970], there is a set of basic units in the language (words, according to Geach; letters, according to Quine). At this basic level, we retain the proper name account, e.g., according to Quine, `a' is a name of one letter, `b' a name of another, etc. Complex quotations, i.e., quotations with more than one basic unit, are understood as descriptions of concatenations of the basic units. So, (4) is construed as (14):

(4) Alice said that life `is difficult to understand'.

*(14) Alice said that life `i'-`s'-` '-`d'-`i'-`f'-`f'-`i'-`c'.....`d',

where `-' is a sign for concatenation. (14) is ungrammatical. Since, according to the description account, when quotes surround basic units, say, `a', `b', etc., the result is a name of the expression, (14) represents the structure of (4) as analogous to `Alice said that life Manhattan-Brooklyn-Bronx-Queens-...', which it clearly is not.

So, prima facie, one general lesson from our brief discussion of these two accounts is that any account of quotation according to which the semantic function of word-tokens inside quotation marks is to refer to word-types (or some other type of linguistic entity) fails to assign correct truth-conditions to (4).

In summary, our criticisms of accounts of indirect, direct, and pure quotation show that in order to account for mixed cases an account must do two things: it must account for how the complement clause of, e.g., (4) can be employed to effect simultaneously a report that Alice uttered the words `is difficult to understand' and one that Alice said life is difficult to understand. We turn to proposals for how to execute these.

V. Davidson's Account of Pure Quotation

The main focus of Davidson's paper `Quotation' is pure quotation. Davidson construes (5) as (15):

(5) `Life is difficult to understand' is a sentence.

(15) Life is difficult to understand. The expression of which this is a token is a sentence.

Since Davidson takes expressions to be shapes or patterns [1979, p.85], (15) is equivalent to (16):

(16) Life is difficult to understand. The shape of which this is a token is a sentence,

where an utterance of the second sentence is accompanied by a demonstration of an utterance of the first. According to Davidson:

...quotation refer to a shape by pointing out something that has it...The singular term is the quotation marks, which may be read `the expression a token of which is here' [1979, p.90].

On his view, quotes are definite descriptions containing demonstratives. The demonstrative picks out the token within the quotation marks and the definite description denotes an expression, i.e., a shape or a pattern, instantiated by the demonstrated token.

Extending Davidson's idea to direct quotation, (2) would be semantically construed as (17):

(2) Alice said, `Life is difficult to understand'.

(17) Alice said (produced) a token of the pattern instantiated by that. Life is difficult to understand,

where an utterance of the first sentence is accompanied by a demonstration of an utterance of the second.14

This unified demonstrative account of pure and direct quotation incorporates four attractive features:

(i) It explains why learning to quote is learning a practice with endless but non-iterable application.

Understanding quotation is understanding `the pattern instantiated by that'. There is no mystery about how we acquire this capacity, nor about how to account for it in a finitely axiomatic semantic theory. Obviously, there is no upper bound on the length of expressions that can be quoted. However, it does not follow, on the demonstrative account, that quotation is a semantically productive device. Every pure quotation ascribing a metalinguistic feature a to a quoted expression b asserts the same thing: that the demonstrated object [i.e., b] is a. However, it does follow, on the demonstrative account, that quotation is not, contrary to a common view, genuinely iterative. Quoted expressions are exhibited so that speakers can talk about the patterns (according to Davidson) they instantiate. The semantic properties of the tokens are not in active use; they are semantically inert (see (ii) below). So, quotation marks within quotation marks are semantically inert. This is why (18) makes sense:

(18) ``oswerk'' is not a quoted expression in Romanian.

It would be a serious error to re-apply Davidson's account of pure quotation to the referenced token in (18), resulting in nonsense like:

*(19) ("x)(ST(x,that) & x is not a quoted expression in Romanian): ("y)ST(y,that): oswerk.

The displayed token in (19), all that succeeds the first colon, is not at all what's quoted in (18).

(ii) It does this while preserving semantic innocence.

A semantic account T for a language L is semantically innocent just in case what an expression of L means according to T does not vary systematically according to context [see, Davidson 1968, p.106; 1975, p.166]. Semantic innocence is preserved at two levels. First, the account does not assume words take on new semantic values when quoted. Secondly, it makes the device of quotation unambiguous; quotes in pure quotation are treated semantically in exactly the same way as quotes in direct quotation, thus, respecting C4.

Semantic innocence so construed, however, is compatible with there being contexts in which what an expression means is not in active use. So, even though `the United States' denotes the United States, it is semantically inert in (20):

(20) `the United States' is a linguistic expression.

(iii) It explains why quotational contexts are opaque.

Sentences containing demonstratives need not preserve their truth value when different objects are demonstrated. If you substitute a word-token of one type for another of a different type as the demonstrated object, different objects are demonstrated and thus the truth-value of the original (utterance of that) sentence may change (see conclusion below).

(iv) It explains why quantifying into quotes in natural language produces absurd results.

`$y(`boxey' is a word)' cannot be inferred from ``boxer' is a word' nor can `$x(`x' is a word)'. The account explains why these inferences fail; since it makes no sense to quantify into a demonstrated object, it makes no sense to quantify into quotes on this account.15

Attractive as (i)-(iv) are, the account requires an important modification. Consider Alice's spoken utterance of (1). (2) directly reports Alice. According to our modest extension of Davidson's account, (2) is construed as (17). But Alice did no such thing. Whatever sounds Alice said (produced) do not instantiate the pattern demonstrated by an utterance of (17) in any obvious sense.

This objection is by no means fatal. After all, Davidson only says `we may take [an expression] to be an abstract shape' [1979, p.85]. His theory is compatible with expressions being something else. We need only find something that can be instantiated by things radically differently shaped. In other words, one response is to identify an appropriate object to serve as the expression instantiated by the demonstrated token. This object must be such that written tokens, spoken tokens, Braille tokens, Semaphore tokens, finger language tokens, and any other way in which words can be produced, can be instantiated by it. Moreover, since we can, and constantly do, develop new ways of producing words (we develop new sign systems for blind people, for computer languages, etc.), this entity must be instantiable by tokens not yet conceived.

Though we doubt such entities exist, they might and if they do, they might ultimately play a role in the metaphysics of language. But even if they do, it is by no means clear that this issue should be settled by the semantics of quotation. A semantics should avoid countenancing (quantification over) dubious metaphysical entities in the metalanguage unless absolutely necessary. Happily, Davidson's account can be modified so as not to quantify over expressions. It could equally well treat quotes as quantifying over tokens that stand in a certain relation, call it the same-tokening relation, to the demonstrated token. This suggests construing (5) as (21):

(21) "x(ST(x,that) -> Sx). Life is difficult to understand,

where an utterance of the first sentence demonstrates the exhibited token of (1), `ST' means same-tokens, and `S' means is a sentence (token). Rather than quotes demonstrating a token and denoting some abstract object it instantiates, they are expressions that quantify over tokens that stand in a same-tokening relation to the demonstrated token. Whether two entities stand in a same-tokening relation to each other is not settled by the semantics. It might involve appeal to an abstract object, but, then again, it might not.

The demonstrative account of pure quotation can be extended naturally to direct quotation. (2) would be construed as (22) (alternatively, as (23)):

(2) Alice said, `Life is difficult to understand'.

(22) $u(Says(a,u) & "y(ST(y,these) -> ST(u,y))). Life is difficult to understand.

(23) $u(Says(a,u) & ST(u,these)). Life is difficult to understand.16

`Says' means says. So, Alice said a token that same-tokens the demonstrated object.

It goes beyond the scope of this paper to elaborate on and fully defend the view that quotes are quantified expressions.17 One obvious advantage of this view is that it construes sentences containing quotes as being about concrete particulars and not about objects existing outside space and time (i.e., sounds, ink marks, etc.) or they can be psychological states, tokens in a language of thought, or what have you. The semantics leaves this open. Glancing forward, just as Davidson's semantics isn't committed to the existence of propositions as what determines the samesaying relationship, a semantics need not be committed to abstract entities, expression-types, as what determines the same-tokening relationship.

VI. Davidson's Account of Indirect Quotation

In `On Saying That', Davidson paraphrases (3) as (24):

(24) Life is difficult to understand. Alice said that,

where `that' is accompanied by a demonstration of the first utterance and that second utterance is true just in case Alice said something that samesays the demonstrated utterance. We prefer to ignore Davidson's distinction between analysis and logical form and represent (3) as (25):18

(25) $u(Says(u,a) & SS(u,that)). Life is difficult to understand,

where an utterance of the first sentence demonstrates an utterance of the latter sentence, `Says' still means says, and `SS' means samesays.

We cannot emphasize enough that we do not intend here to engage in an evaluation of Davidson's (or our version of his) account of indirect quotation.19 Our aim, instead, is to show how the accounts of direct and indirect quotation can be exploited and developed so as (at least) to satisfy C1-C4. Notice straight away that C3 is satisfied, i.e., indirect quotation is treated differently from direct quotation. The former invokes same-tokening; whereas the latter invokes samesaying.20

VII. Mixed Quotation

Earlier we argued that the available semantic accounts of pure (and direct) and indirect quotation do not integrate mixed quotation and therefore fail to satisfy C1-C4. The semantic theories of indirect speech canvassed earlier treat the complement clause as a semantic unit referring to (or in some other way determining) a proposition (or something proposition-like). The semantic theories of quotation canvassed earlier treat pure quotes as singular terms referring to abstract objects, expression types. For anyone receptive to either of these ideas mixed cases remain enigmatic. From these points of view mixed quoting seems to involve two entirely different activities taking place at the same time in the same place. How could the complement clause of (4) both determine a proposition not about words, and, concomitantly, refer to words Alice used?

Merging the two demonstrative accounts supplies an ingenious reply. Since the complement clause is in effect semantically excised from (4) and merely demonstrated, we can ascribe different properties to it. With one utterance we can say both that the demonstrated token samesays one of Alice's utterances and say that it (or parts of it) same-tokens that utterance. Our suggestion, then, is to construe (4) as (26):21

(26) $u(Says(a,u) & SS(u,that) & ST(u,these)). Life is difficult to understand,

where an utterance of the first demonstrative demonstrates an entire utterance of (1) and an utterance of the second demonstrative demonstrates (only) the (sub)utterance of `difficult to understand'.22 According to the unified account, mixed cases like (4) can be utilized both to attribute the same-tokening relationship between one of Alice's utterances and the demonstrated (sub)utterance and to attribute a samesaying relationship between Alice's utterance and the demonstrated utterance.23 , 24

There are cases where quotes in the complement clause do not indicate mixed quoting. If Alice asserts (5), we can report her with (27). On our account this is unproblematic. (27) is construed as (28):

(27) Alice said that `life is difficult to understand' is a sentence.

(28) $u(Says(a,u) & SS(u,that)). `Life is difficult to understand' is a sentence.

In some cases the syntax of an indirect report does not reveal whether the report is mixed or indirect. (29) could report what Alice said about the name `Butcher Bob' (she might have uttered `'Butcher Bob' is ugly') and therefore be construed as indirect quotation, i.e., as (30); or it could be a mixed report of what Alice said about Butcher Bob, but where the reporter may want to distance himself from the use of the name `Butcher Bob', and therefore, he employs a mixed case, i.e., as in (31):

(29) Alice said that `Butcher Bob' is ugly.

(30) $u(Says(a,u) & SS(u,that)). `Butcher Bob' is ugly.

(31) $u(Says(a,u) & SS(u,that) & ST(u,these)). Butcher Bob is ugly.

In (31), the first demonstrative demonstrates an entire token of `Butcher Bob is ugly' and the second demonstrates only the token of `Butcher Bob'.

Based on surface syntax alone, whether an utterance of (29) will be construed as (30) or (31) depends upon what intentions an interpreter ascribes to the utterer. This is no different from ordinary cases of ambiguity. We cannot determine simply on the basis of a written or spoken utterance of an ambiguous sentence what the speaker intended. What is special about mixed cases is that they can be used to attribute both the same-tokening and the samesaying relation between the same two utterances. Though special, there's nothing problematic here. In fact, given that it is a very efficient way of performing a certain kinds of speech act, it is exactly what we expect.25

In criticizing other accounts, we appealed to reports such as (11):

(11) Nicola said that Alice is a `philtosopher'.

Our account treats (11) as (32):

(32) $u(Says(n,u) & SS(u,that) & ST(u,these)). Alice is a philtosopher,

where `that' is accompanied by a demonstration of the token of `Alice is a philtosopher' and `these' by a demonstration of the token of `philtosopher'.

Is our account better positioned to account for the truth of (11) than the account (i.e., (12)) we criticized earlier? After all, an utterance t of (11), construed as (32), is true only if Nicola's utterance samesays t's sub-utterance of `Alice is a philtosopher'. But an utterance of `Alice is a philtosopher' by a normal English speaker can't express anything since `philtosopher' is not English; but, then, how can an utterance that fails to express anything samesay anything?

Anyone who has this concern has not understood how the extension of the samesay relation is determined. It's the actual practice of making indirect reports of others that fixes that extension. There are no a priori constraints on what can samesay what. If a certain sort of report is an important part of that practice, then it's philosophically prejudicial to bar it as illegitimate. Mixed quotes similar to (11) form an important subpart of our practice of indirect reporting. We often encounter speakers whose utterances partly make sense to us and partly don't; we need a reporting device to indicate which part made sense and which was odd. Mixed quotation serves exactly that function. Consider common-place cases like (33)-(37):

(33) Max said that Alice is an `oenophile', but I'm not sure what that means.

(34) Max said that Alice is an `oenophile', but I think that what he means by `oenophile' is not what we mean by it.

(35) Max said that Alice is an `oenophile', but I don't think he knows what that means.

(36) Max said that Alice is an `oenophile', but I don't think that's a word.

(37) Max said that Alice is an `oenophile', but that isn't a word.

In (33), the reporter is uncertain about the meaning of an expression, but leaves open the possibility that this due to his own linguistic ignorance; whereas the difference between (34) and (37) concerns the speaker's degree of certainty about `oenophile'-s place or non-place in the lexicon.

(33)-(37) communicate efficiently that a certain part of Max's utterance made sense, but that one specific part, his use of `oenophile', was odd. These five cases begin to indicate the wide variety of reasons we have for finding, and pointing out that, the use of certain expressions are odd. (36)-(37) are, perhaps, closer to (11). Notice how we can extend (11) to (11¢):

(11¢) Nicola said that Alice is a `philtosopher', but I don't think that's a word, but that isn't a word, etc.

What all these examples show is that since mixed quotes like (11) are an important part of our indirect reporting practice and since the extension of the samesay relation is determined by our actual practice of indirect reporting, there can be no further question whether the demonstrated sub-utterance of an utterance of (32) can samesay an utterance of Nicola's. Therefore, any account of indirect speech that ignores these sorts of cases is incomplete.

What about the question how it is possible for a demonstrated utterance of `Alice is a philtosopher' to samesay Nicola's utterance? If this question is asking how samesaying can have this sort of extension, we don't know how to answer it. It has the extension it has and we can understand why that is useful and important in our linguistic practice. Still, one might wonder how, assuming that (11) and its like are linguistically acceptable mixed quotations, do we understand them? In order to understand (11), don't we need to understand its complement clause? But its complement clause contains an expression we might not understand, namely, `philtosopher'. And isn't this particularly damaging for us since our chief complaint against the alternative account (12) is that its reporter ends up asserting an ill-formed meaningless English sentence?

Since we do claim we understand (11), there is a challenge here, but that challenge is an objection to us only if our theory makes that challenge particularly difficult to meet. We don't think it does. According to the unified demonstrative account, the complement clause is no part, at least not `from a semantical point of view', of (11). To understand an utterance of (11) is to understand the utterance of its main clause (the sentence containing the demonstrative). Having understood an utterance t of that main clause, one can go on to determine whether the demonstrated sub-utterances samesays Nicola's utterance in order to determine the truth of t. How we actually do that, i.e., how we determine whether two utterances samesay each other, is a question that takes us well beyond the scope of this paper.26 For our purposes, all we need to establish is that this is something we do for utterances of sentences like (11). But anyone who agrees that (an utterance of) (11) is linguistically acceptable and potentially true must concede that this is something we do do.27

VIII. Conclusion

We have shown at least this much: there are interesting interactions among the varieties of quotation. Semantic accounts of pure, direct, mixed, and indirect quotation must acknowledge these interactions, i.e., must satisfy C1-C4. A myopically developed account of either indirect or pure quotation is unlikely to be correct. Our joint account, as far as we know, is the only one available satisfying all four constraints.

Also worth noticing is that this unified account satisfies C1-C4 with a simplicity and elegance not likely to be shared by any competing account. According to C4, quotes in pure, direct, and mixed quotation should be treated uniformly. It is hard to evaluate alternative accounts before they are supplemented with an account of direct and pure quotation. The joint modified Davidsonian account of pure and direct quotation is, for reasons discussed above enormously attractive. If you share this view, it is certainly natural, though not necessary, to combine it with Davidson's account of indirect quotation.

What makes it almost irresistible to join a demonstrative account of pure and direct quotation with a demonstrative account of indirect quotation is that this results in a unified account of opacity. Pure and indirect quotation are paradigms of opaque contexts. From a methodological point of view, it is both plausible and desirable that there be a common explanation of their opacity. This probably accounts for so many efforts in the history of this subject to assimilate indirect quotation to direct quotation (Carnap [1937, p.248; 1947], Scheffler [1954], Quine [1956, 1960, secs. 30-32], Sellars [1955], even Church [1954] given his metalinguistic solution to Mates' problem). Though the unified demonstrative account provides a uniform account of quotation, it does not do so by assimilating either form of quotation to the other and therefore, does not fall prey, as did its predecessors, to the standard Church arguments [Church 1950, p.97].

Anyone who finds the account of quotation served in this paper unpalatable (perhaps because of the numerous objections to Davidson's account of indirect quotation) needs an alternative. But the situation here is unlike indirect quotation. There are no large number of more or less acceptable competing accounts to choose from. In other words, if C1-C4 are acceptable constraints on a general account of reported speech, then any account of indirect quotation is incomplete until supplemented with an account of pure and direct quotation.

Still, we'd like to end with a challenge for those philosophers unalterably convinced that Davidson's account of indirect quotation is wrong. 1. Take your favorite theory of indirect quotation and show that it can (be extended to) account for C1-C4. 2. Either develop a theory of pure quotation that combines with your favorite theory of indirect quotation to yield a unified account of opacity or explain why opacity doesn't admit of a unified account. When you've completed tasks (1) and (2), compare your results with the unified account with respect to simplicity and elegance.28

Herman Cappelen

Ernie Lepore

Philosophy Department

Vassar College

Poughkeepsie, NY


Center for Cognitive Science

Rutgers University

New Brunswick, NJ 08903


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

Cappelen, H. 1997: Signs, Doctoral Dissertation, University of California-Berkeley.

Cappelen, H. and E. Lepore: `On an Alleged Connection between Semantic Theory and Indirect Quotation,' forthcoming.

Carnap, R. 1937: Logical Syntax of Language, London, Routledge and Kegan Paul.

Carnap, R. 1947: Meaning and Necessity, Chicago, University of Chicago Press.

Chomsky, N. 1981: Lectures on Government and Binding, Foris, Dordrecht.

Church, A. 1950: `On Carnap's Analysis of Statements of Assertion and Belief', Analysis, X, pp.97-99.

Church, A. 1954: `Intentional Isomorphism and Identity of Belief', Philosophical Studies, V, 5, pp.65-73.

Crimmins, M. 1992: Talk About Belief, Cambridge, Mass., MIT Press.

Crimmins. M. and J. Perry 1989: `The Prince and the Phone Booth', Journal of Philosophy, 86, pp.685-711.

Davidson, D. 1968: `On Saying That', in Inquiries Into Truth and Interpretation, Oxford, Oxford University Press, pp.93-108.

Davidson, D. 1975: `Thought and Talk', in Inquiries Into Truth and Interpretation, Oxford, Oxford University Press, pp.155-170.

Davidson, D. 1979: `Quotation', in Inquiries Into Truth and Interpretation, Oxford, Oxford University Press, pp.79-92.

Geach, P. 1957: Mental Acts, London, Routledge Kegan Paul.

Geach, P. 1970: `Quotation and Quantification', in Logic Matters, Oxford, Basil Blackwell.

Higginbotham, J. 1986: `Linguistic Theory and Davidson's Semantic Program', in Truth and Interpretation, ed., E. Lepore, Oxford, Basil Blackwell, pp.29-48.

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

Larson, R. and G. Segal 1995: Knowledge of Meaning, Cambridge, Mass., MIT Press.

Lepore, E., and B. Loewer 1989: `You Can Say That Again', Midwest Studies in Philosophy, eds., H. Wettstein, et. al., Notre Dame University Press, 1989, pp.338-356.

Lepore, E., and B. Loewer 1990: `A Study in Comparative Semantics', in Propositional Attitudes, eds., C. Anderson and J. Owens, Stanford, CSLI.

Partee, B. 1973: `The Syntax and Semantics of Quotation' A Festschrift for Morris Halle, eds. S.R. Anderson and P. Kiparsky, New York, Holt, Reinehart and Winston, pp.410-418.

Platts, M. 1979: Ways of Meaning, London, Routledge Kegan Paul.

Quine, W.V. 1956: `Quantifiers and Propositional Attitudes', Journal of Philosophy.

Quine, W.V. 1960: Word And Object, Cambridge, Mass., MIT Press.

Quine, W.V. 1961: From A Logical Point Of View, Cambridge, Mass., Harvard University Press.

Quine, W.V. 1961: `Reference and Modality', in From a Logical Point of View, pp.139-159.

Richard, M. 1990: Propositional Attitudes: An Essay On Thoughts and How We Ascribe Them, Cambridge, Cambridge University Press.

Scheffler, I. 1954: `An Inscriptional Approach to Indirect Quotation', Analysis X, pp.83-90.

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

Sellars, W. 1955: `Putnam on Synonymity and Belief', Analysis, xv, 5, pp.117-120.

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

Soames, S. 1989: `Direct Reference and Propositional Attitudes', in Themes From Kaplan, eds., J. Almog, J. Perry, H. Wettstein, New York, Oxford University Press, pp.393-419.

Tarski, A. 1956: `Concept of Truth in Formalized Languages', in Logic, Semantics and Metamathematics, Indianapolis, Hackett.

Washington, C. 1992: `The Identity Theory of Quotation', Journal of Philosophy, 89, pp.582-605.