Chapter 3
The Theory of Meaning: Framework for the Present Discussion

3.1 Introduction

In my first chapter I discussed the general motivation for and emphasis of a "Gricean" theory of meaning. But in the course of this dissertation I would like to discuss the mechanics of a few theories of meaning and to propose one of my own. All of that requires a more detailed introduction. The present chapter is meant to provide this.

Essentially what I will do in this chapter is present a certain manner of conceiving of - or of talking about, at any rate - the problem of providing a general theory of public-language expression meaning. I will discuss a number of issues concerning public-language meaning - ambiguity, indexicality, non-indicatives, non-literal speech, and sub-sentential expression meaning - and show, either in broad terms or by mere illustration, how the conception I will present here and work with throughout this dissertation is expected to be able to handle those issues. Details about how a specific theory of meaning will deal with the various issues at play will have to await the details of the specific theory, however. I am also going to introduce the notion of a language of thought at the end of this chapter since that notion will be important in the discussions that follow throughout the remainder of the dissertation.

3.2 The Actual-Language Relation

Just about every sentence you can think of will have some indexical element in it. That means that the meaning of just about every utterance will not be determinable independently of features of the context in which the utterance was made. "I think that there is a greatest prime" contains the pronoun "I" and therefore when uttered by different people, means - in a perfectly legitimate ordinary sense of "means" - different things. "Grass is green" evidently means something different depending on when it is said. It is roughly true today, but climactic changes, perhaps, could occur such that in 135 years it is no longer true. So, uttered today it expresses a truth, and uttered in 135 years, it, perhaps, expresses a falsehood. It is, apparently, the present-tense of the verb that is responsible for this shifting semantic value. We might, therefore, identify tense as an indexical element in many sentences in which it occurs. "It is raining" is false if I, sitting in Brooklyn on a clear day, utter it in response to a query about the present weather in Brooklyn, but is true if I, still sitting in Brooklyn on the same clear day, utter it in response to the question "What is the weather like in Prague?" provided, of course, that it is raining in Prague. There is a more subtle indexical element at work here than in the other examples, but, arguably, it is an indexical element all the same: somehow reference to a place is understood in utterances of "it is raining". Perhaps statements of arithmetic such as "2+2=4" don't contain any indexical elements. Perhaps "Grass is green on September 4, 1995 at 7:30 pm" is wholly non-indexical. I am not sure whether it is completely.

But it looks like most ordinary sentences contain some sort of indexicality. This is annoying, but it is true.

Nonetheless, the framework that I would like to work within in this dissertation will have us imagining a class of sentences that have no indexical features. Even if it is hard to come up with definite examples of such sentences from ordinary language, still, it is pretty easy to imagine what they would be like. They would have something like a complete and definite meaning independently of contexts of utterance. "Grass is green on September 4, 1995 at 7:30 pm" perhaps approaches what is imagined.

With or without clear examples from ordinary language, the assumption of non-indexicality is simply a convenience. This will be discussed further below.

Imagine, then, that we are concerned only with languages that have no indexical devices. Imagine also, that we are concerned with languages that only contain indicative sentences. What would a theory of meaning be like for such languages? It would be, essentially, the provision of a satisfactory completion of [M]<1>:

[M](P)(x)(y)(x means among the members of P y ...)

In chapter 1 I discussed a bit what would count as a satisfactory completion of [M] for various styles of Gricean theorists. Here I will talk a bit about another way of talking about providing such a theory of meaning.

A satisfactory completion of [M] will be a three-place predicate that is necessarily<2> coextensive with the three-place predicate [*] and that uses only propositional-attitude and propositional-speech-act notions along with, perhaps, other notions so long as these other notions do not require for their satisfaction the satisfaction of any public-language semantic notions:

[*]x means among the members of P y

For a given population P, the extension of the predicate "x means among the members of P y" will be a set of ordered pairs of non-indexical, indicative sentences and their meanings. Let's temporarily call any set of pairs of non-indexical, indicative sentences a language*. It is harmless to restate the central task of the theory of meaning as the task of finding a satisfactory completion for [U]:

[U](P)(L)(P uses* L just in case ... )

where P ranges over populations, L over languages*, and where uses* is defined as follows<3>:

(P)(L)(P uses* L just in case for every sentence of L and sentence meaning such that is paired with in L, means among the members of P )

Nothing in all of this is very interesting. All that is being done is the introduction of a way of speaking about providing a completion for [M] and thereby about providing a theory of meaning. It should be clear that the notions of use* and of a language* have been set up so that it turns out that providing a completion for [U] is equivalent with no remainder to providing a completion for [M] (and vice versa).

Some languages* will be functions in the set-theoretic sense. These will be the languages* in which there is no sentence paired with two distinct meanings, that is, in which there are no ambiguous sentences. We can, following David Lewis, call these languages. But, the notion of a language here, just like the notion of a language* is a special technical notion and not some ordinary notion of a language.

Notice that every language* will either be a language or the union of two or more languages. So, if we reconstrue the "L" of [U] as ranging over Lewisian languages rather than languages*, nothing changes essentially with respect to the theory of meaning. We can still say that the central task in providing a theory of meaning is providing a satisfactory completion of [U] where L ranges over languages in the Lewisian sense.

The uses* relation which holds between populations and languages is what Lewis has called the actual-language relation. To complete [U] is to provide a theory of what David Lewis has called the actual-language relation. The main task in providing a theory of meaning, then, can be said to be the task of providing a theory of the actual-language relation (ALR).

I introduce the notion of the ALR here mainly because literature that I will discuss in subsequent chapters is couched within this way of talking about the theory of meaning. But I would like to stress that nothing in such talk adds anything at all of substance to talking about providing a satisfactory completion for [M]. In the introduction to this chapter I spoke of a certain way of conceptualizing the theory of meaning that I would be using throughout this dissertation. What I had in mind was actually this way of talking about the theory of meaning by way of talking about the ALR. Perhaps merely introducing a way of talking shouldn't be called a way of conceptualizing things. But, it should be clear what I am doing here.

3.3 Propositions

The objects of our propositional attitudes and the meanings of sentences are, I will take it, propositions. I need this assumption to make sense of the extension of the meaning relation as I have, that is, as a set of ordered pairs of sentences and things that are their meanings. And there seems to be something of prima facie support for accepting propositions when we consider certain data from natural language. For example, the inference from [a] to [b] here seems a valid one:

[a] Oscar believes that dogs bite.

[b] Therefore, there is something that Oscar believes.<4>

It is hard to account for the apparent validity of this inference without taking the conclusion to involve objectual quantification over things that Oscar can believe.

Of course, there are famous problems associated with propositions. Most notably, nobody seems to have a clue about how we, presumably bits of the natural order, could be related to them, presumably abstract, non-natural objects. Different philosophers have suggested different ways of availing themselves of propositions with a clear metaphysical conscience. I'll remark on a few of these and then suggest my own.

One attitude that can be held here is that there simply are good reasons for accepting propositions - and accepting them as abstract, non-natural objects -, and that the epistemological worries of some philosophers are tied to a naturalistic bias in their epistemology which can be shed by accepting a non-naturalistic epistemology of some sort.<5> But since I find the notion of a non-naturalistic epistemology extremely unclear, I cannot adopt such a view as this.

Fodor has suggested that were there no worries extending from naturalistic epistemological scruples, then we might as well ask, "if physicists have numbers to play with, why shouldn't psychologists have propositions".<6> But I wouldn't be happy with this attitude either since - even apart from its also asking me to put aside worries that extend from a sense that our epistemology should be accounted for naturalistically - it is not clear what the physicists use of numbers amounts too. You can't quell the epistemological concerns by pleading indispensability since one wonders because of the epistemological concerns whether abstract objects indeed are ultimately indispensable.

A moderate approach to propositions has been to identify them with set-theoretic objects of some sort and then to hope, I think the idea is, that somehow set-theoretic notions will escape the sorts of problems that philosophers have had with propositions. For example, David Lewis<7> and others<8> have identified propositions with sets of possible worlds, and Scott Soames<9> and others<10> have suggested understanding propositions as structured entities that might be identifiable in set-theoretic terms as certain sorts of ordered-pairs. Of course, possible-worlds semantics has ontological and epistemological problems of its own. But still, one might feel - or hope - that acceptance of possible worlds is clearer than acceptance of unreduced propositions. So we can imagine, anyway, that someone might try to forestall the epistemological problems with raw propositions by shifting the issue to the epistemology of possible worlds and set theory.<11> The structured-proposition theorist doesn't have the special ontological and epistemological issues to face as the possible-worlds semanticist, and so might have all the more reason to welcome the identification of propositions with set-theoretic stuff, given the independent view that somehow the epistemology of set-theory is not at all as problematic as the epistemology of raw propositions.<12>

The motivations for set-theoretic identifications of propositions largely, if not completely, stem from concerns with the compositionality of natural language semantics. I will discuss such issues in my next chapter. But for now I will simply say that I am reluctant to accept these set-theoretic identifications of propositions even on the assumption that they could help with the epistemological worries about propositions - an assumption that I am not sure should be granted either - since I am not convinced that propositions so understood can help to give an acceptable account of our propositional-attitude ascriptions. If they can't, then any promise they might have offered to help with understanding how we could stand in epistemological relations to such things is beside the point: we might as well have picked trees to be propositions - they surely can be known about and they don't help with belief-ascription either. The issues here are bound up with those of compositionality which I will discuss further in the next chapter.

Stephen Schiffer has argued for what he calls a pleonastic conception of propositions.<13> On this conception, propositions are something like fictional objects: we can truly speak of their existence and of their having certain properties, but this existence and the properties that they have are determined by our linguistic practices alone and not by them themselves. I cannot discuss all the details and merits of Schiffer's discussion of pleonastic propositions here. But, the view is interesting and attractive enough to merit a few comments, even if in the end I will express dissatisfaction with it.

Schiffer tries to make sense of the notion of a proposition by way of making sense of a general group of suspect objects that includes actions, events, and properties<14>. He is impressed by the fact that there is a systematic syntactic relation between the singular-terms that apparently refer to these suspect objects, and certain other linguistic expressions. So, for example, the sentence "snow is white" is not, it would seem, a referential singular-term, and yet, if you stick a "that" in front of it, you get "that snow is white" which arguably is a referential singular-term. And the same is, more-or-less, true for all English indicative sentences (ignoring indexical features): stick "that" in frontof an English indicative and you get something that seems to behave like a referential singular-term. Schiffer thinks it is puzzling that we can somehow get referential singular-terms in this way: "From a true sentence containing no singular term that refers to an entity of the kind in question, we get a singular term that does refer to an entity of the kind in question" (Schiffer (1994b), p. 304). Whether or not this is really puzzling, I am not sure. But let's suppose it is.<15>

Schiffer is interested in examining a position whereby such singular-terms are taken somewhat seriously referentially - they really are referential singular-terms -, but the things that they refer to are treated "in a suitably deflationary, or minimalist, manner" (p. 305):

States, events, properties, and propositions exist all right, but in acknowledging this we are merely playing along with the language games that introduce these notions, and there is nothing more to the natures of these things than these little language games determine. [Ibid.]

No doubt, this is a seductive suggestion. It sounds - even if you ignore the allusion to language games - a bit like what Wittgenstein might have believed to be true about metaphysically problematic entities.

The story that Schiffer suggests to account for the suspect entities that he considers has it, more-or-less, that their existence is a matter of the existence of referential singular-terms that are constructed in a systematic and purely syntactic way from ontologically unsuspicious language.<16> The best way to understand this account is by looking at an example of the construction of a pleonastic object of an entirely new sort. This should be possible if Schiffer's theory is true.<17>

Consider the following syntactic transformation rule, call it the Q-rule: construct the noun-phrase "the 'Q'-quantification of F" from an expression of the form "Q F" where 'Q' is a quantifier and 'F' is a common noun.<18> This rule would allow you to get from "every dog" to the noun-phrase "the 'every'-quantification of dog". And we can apply the rule to noun-phrases that occur in sentences. "The 'some'-quantification of dog is fun" will be true just in case "some dogs are fun". We can also allow quantification into the subject position here. We just need to adopt a rule - call it the QE-rule - that allows this, that is, so that it follows from "the 'some'-quantification of dog is fun" that "there is something that is fun".

So it looks like a few simple syntactic innovations have allowed the appearance of commitment to a new class of entities, quantifications, that do their work, not by somehow being out there in the world constraining thought by their existence, but by having their "existence" wholly determined by linguistic practices that have been adopted parasitic on already existing linguistic practices. But perhaps there is something more complicated is going on than merely paraphrasing familiar sentences with new syntactic arrangements.

Consider the sentence "no dog is fun". Applying the Q-rule to the subject we get, "the 'no'-quantification of dog is fun". Applying the QE-rule here we get "there is something that is fun". This sentence will be true if no dogs are fun, that is, "there is something that is fun" follows from "there is nothing that is fun". The introduction of the Q-rule and the QE-rule seems to have introduced an ambiguity into our talk that wasn't there before. It seems that "fun" in "June is fun" is the ordinary fun that applies to people and things just when they are fun, but not to our new entities, quantifications. The "fun" of "the 'some'-quantification of dog is fun" is a different "fun" that only applies to quantifications and not to people and things. Thus, introducing quantifications also introduced a new sort of verb, one that expresses properties of the new entities. For each verb in English that can have a noun-phrase subject of the form "Q F", there will be another verb that corresponds to it and has in its extension quantifications.

It is interesting to look at another sort of example that brings out this feature of the introduction of pleonastic entities by way of Schiffer's recipe.

Consider the A-rule: Construct "the Adj Adv" from sentence of the form "Q Adj F V Adv" where 'Q' is a quantifier, 'Adj' is an adjective, 'F' is a common noun, 'Adv' is an adverb, and 'V' is a verb. Thus from "Every boring lecturer eats quickly" you get "the boring quickly". We can introduce a rule - call it the AS-rule - that allows us to use such noun-phrases in sentences: For any noun-phrase x constructed from a sentence S by the A-rule, there is a sentence "x is the Q F V" which is true just in case S is true. Thus, "the boring quickly is every lecturer eats" is a sentence in an English augmented with the AS-rule and it is true just in case every boring lecturer eats quickly. Nothing Schiffer says rules out this sort of formation. And I am not sure that it should. Such rules can be adopted and used with a little work, and we could probably communicate with each other very well with their results.

In the case of the A-rule and the SA-rule, the new "predicates" we get are clearly not like the predicates of English prior to the adoption of these rules. The ambiguity we saw in the "fun" example above is not present here. But, in a sense, we see now that it is not the ambiguity that was really interesting in the "fun" example, but the fact that entirely new predicates have to be admitted into the language to accept reference to pleonastic entities, whether these predicates are syntactically identical to old predicates, thereby introducing ambiguity, or not.

It seems to me that this all should be allowable given what Schiffer states in his gloss of pleonastic propositions. But I am not sure that this is unproblematic, though I cannot investigate the matter any further here.

But there is a somewhat more picky problem I have with Schiffer's conception of pleonastic propositions. Schiffer tries to trace our apparent acceptance of certain abstract objects into our ontology to our acceptance of syntactic procedures that allow for apparent talk of such objects. But, if this is the case, there will be limits set on our ability to legitimately accept abstract objects into our ontology due to whatever limits there are on our abilities to deal with syntactic objects. If our syntactic competence is limited in the construction of natural-language sentences so that we could only, even under ideal conditions that ignore performance constraints, ever construct countably many sentences, then we should only be able to legitimately conceive of countably many propositions. It would seem that we will only have as many propositions as there are sentences in the language for which the syntactic-transformation rules that Schiffer's view requires are stated. If human languages - natural or artificial - only contain countably many sentences, it will follow from Schiffer's gloss of propositions that there are less propositions than there are mathematical truths, at least on a certain vague but intuitive conception of the notion of mathematical truth. But propositions, classically understood, are supposed to be the things that are mathematical truths. Thus, for each real number r there is a truth "r is not a member of the empty-set". Each one of these truths will not be expressible in a natural language if natural languages can have only countably many sentences.<19> And classically, propositions have been the sort of thing that can be identified with these truths even when we run out of sentences. So if we accept Schiffer's strategy for understanding propositions, then either we have to accept that mathematical truths are not propositions, or that they are propositions, but there aren't as many as we thought, or that they are propositions, and there are more than countably many of them, but that the human languages that we know have the resources to allow for more than countably many sentences<20>. It feels odd to me that an account of propositions should force me to make such choices.

Though I don't know that these observations about Schiffer's conception of propositions might lead to objections that are fatal to it, they strike me as making the conception problematic enough I would rather not avail myself of it. So I have the following suggestion about how to feel okay about my talking about propositions throughout the remainder of this dissertation.

To speak of the belief that snow is white, is to speak of a property of a certain sort, namely, a specific belief-property. Properties are problematic entities, just like propositions, but if belief-reporting sentences one day enjoy explication, I believe, talk of belief-properties will be explainable as well. But it may or may not be that this future explication of belief-reporting sentences will employ a notion of a proposition usefully - for all we know right now, belief-reporting sentences will one day enjoy explications that do not require any talk of propositions. That would be a nice thing. But whether or not there ever will be such economical explications of belief-reporting sentences, it seems that the possibility of them makes talk of belief properties neutral about the existence of propositions. And since the problematic aspects of speaking of belief-properties will dissolve with an explication of belief-reporting sentences, it seems that it is completely legitimate ontologically for an intentional realist to speak as freely as possible about belief-properties.

But it turns out that every sentence that reports a propositional-attitude in English can be pleonastically restated with a sentence that talks about belief-properties. For example, [e], [f], and [g] can be restated as [e'], [f'], and [g']:

[e] June desires that she drinks some water.
[f] June expects that she will drink some water.
[g] June intends that she will drink some water.
[e'] June desires that the belief that she drinks some water be true.
[f'] June expects that the belief that she drinks some water will be true.
[g'] June intends that the believe that she drinks some water will be true.

Even belief-reporting sentences themselves can be harmlessly restated in terms of belief-properties, even if these have a sort of redundancy in them:

[h] Bingo believes that June is thirsty.
[h'] Bingo believes that the belief that June is thirsty is true.

The inference from [i] to [j] can be restated as an inference from [i'] to [j']:

[i] Bingo believes that June is dancing.
[j] Therefore, there is something that Bingo believes.

[i'] Bingo believes that the belief that June is dancing is true.
[j'] Therefore, there is something that Bingo believes.

Cases with quantification into "that"-clauses are also capable of being restated with expressions reporting belief-properties. Thus, [k] and [k'] are equivalent:

[k] There is something that June believes is charging.
[k'] There is something such that June believes that the belief that it is charging is true.

It is my view that every sentence with a propositional-attitude verb in it can be restated in terms of belief-properties in some way similar to the above examples. And that seems to indicate that all talk of propositions can be replaced by talk of belief-properties. Thus, all talk of propositions is just as safe as talk of belief-properties. And for the intentional realist that will mean, all talk of propositions is completely safe ontologically.

Perhaps the view I suggest here is not wholly unproblematic. There will be issues about the semantics of the restatements, I believe, that I cannot investigate now. But I think the view ultimately will prove a happy one. So I accept it here. All this acceptance will mean for the current dissertation is that I will talk freely in terms of propositions.

3.4 Ambiguity

In section 3.2 above I discussed how the goals of a theory of meaning could be expressed in terms of Lewis's notion of the actual-language relation. But I had to abstract away the fact that groups of speakers often have sentences with more than onemeaning for them. So, it might seem at first glance that we should have to augment the specification of the actual-language relation in some way to account for ambiguity. But this is not going to be true. For an ambiguous sentence can be treated as two sentences of two separate languages and nothing stands in the way of a population using two separate languages. So, the specification of the actual-language relation already treats of ambiguous sentences without augmentation.

Let me illustrate this point. Suppose the only sentence that was ambiguous for English speakers was the sentence "Sinatra really swings" which means for them both that Sinatra really swings and that Paris is a cool city.<21> We can think of English speakers as speaking two languages here. One of these languages contains all the non-ambiguous sentences paired with their meanings plus the sentence "Sinatra really swings" paired with the proposition that Sinatra really swings. The other of these languages contains just the one sentence "Sinatra really swings" paired with the proposition that Paris is a cool city. English speakers will stand in the actual-language relation to both of these languages, and thus the sentence "Sinatra really swings" will be ambiguous for them.<22>

Thus, a general definition of ambiguity can be given in terms of the actual-language relation:

A sentence is ambiguous for a population P just in case there are two languages L and L' such that (1) both L and L' contain but are such that L() L'(), and (2) P uses both L and L'.<23>

David Lewis gives a slightly different account of ambiguity in Lewis (1975).<24> There he complicates the notion of a language so that it is not just a function from sentences to meanings, but a function from sentences to sets of meanings. That account is fine, it seems to me, but it is unnecessary. Everything needed to explain ambiguity is already available with the simpler notion of a language and the actual-language relation.

3.5 Indexicality

When indexical features of sentences are addressed, it seems best not to take sentence meanings as propositions, but rather as functions from contexts of utterance to propositions. It is common to call such a function a character after Kaplan's use of that term.<25> So, indexicals, presumably can be treated by redefining a language as a function from sentences to characters and then redefining the actual-language relation in the following way:

A sentence means among a population P a character just in case P uses L and = L().

The question is how to get from an account of the actual-language relation in the original sense in which indexicality is ignored to an account of the actual-language relation in this new sense. I don't know any general way of stating how to do this, but I will give the following example.

Suppose that someone comes up with the following theory of the actual-language relation in the original sense of that notion which excluded consideration of indexicals:

P uses L just in case for each L-sentence , members of P utter just in case they are trying to say L().

We could get a theory of the actual-language relation in the new sense glossed just above in the following way:

P uses L just in case for each L-sentence , and for each context of utterance c, members of P utter in c just in case they are trying to say L()(c).

I can think of no reason why we should expect that some particular theory of the actual-language relation - in the original sense - will be such that this sort of extension will be unavailable. So, supposing that the notion of a character can be appropriately understood, it seems that if we have provided a theory of the actual-language relation - again, in the original sense -, then we will have provided enough to state a theory of meaning that also dealt with the indexical features of languages.

3.6 Non-Indicatives

Non-indicative sentences are meaningful too and a theory of meaning should explain wherein lies their meaning. For the present discussion I will restrict my comments to questions and commands, but what I say, hopefully, will apply to other non-indicative moods as well with a little adjustment.

It is arguable that [l],

[l] "'snow is white' means among English speakers that snow is white"

is an ordinary-language sentence that reports the meaning of the sentence "snow is white". But there doesn't seem to be equally unproblematic sentences of ordinary-language that report the meanings of questions and commands, not to mention other moods. Both [m] and [n] are really strange:

[m] "'Is snow white?' means among English speakers that is snow white.
[n] "'Close the door' means among English speakers that close the door.

I think [o] and [p] are moderately good colloquial English:

[o] "'Is snow white?' means among English speakers the question whether snow is white.
[p] "'Close the door' means among English speakers the command to close the door.

And I will suppose that these are paradigmatic meaning-reporting sentences for questions and commands. I will take it, that is, that these are the sorts of sentences that a theory of meaning should aim to clarify to account for the meaning of questions and commands.<26>

Since "the question whether snow is white" and "the command to close the door" are descriptions, it looks like [o] and [p] are committed to new sorts of things, questions and commands, that can be the meanings of sentences. So it looks like we have three sorts of things that can be the meanings of sentences: propositions, questions, and commands. But these three sorts of things can be dealt with pretty economically in the following way.<27> Let each of these types of meanings be an ordered-pair of some arbitrary code and a proposition (in the original sense). Thus, the meaning of "snow is white" can be identified with <1, that snow is white>; the meaning of "is snow white?" can be identified with <2, that snow is white>; and the meaning of "close the door" can be identified with <3, that the door is closed>. Thus, the code "1" is associated with indicatives, "2" is associated with questions, and "3" is associated with commands. If there were further moods, hopefully, we could analyze these by way of pairs of mood-codes and a propositional contents as well. For ease, we can introduce the functors "m₁" and "m₂" to denote the functions from sentences to their mood-code and from sentences to their propositional content respectively. So, m₁("is snow white?") = 2 and m₂("is snow white?") = the proposition that snow is white.

With such identifications in hand a language can be redefined to include the non-indicatives as well: a language is now a function from sentences to their meanings in the new sense.<28> The actual-language relation will have to be adjusted to accommodate the new sorts of meanings. So, suppose again that someone came up with the following theory of the actual-language relation for a language without indexicals and non-indicatives:

P uses L just in case for each L-sentence , members of P utter just in case they are trying to say L().

To handle a language in the new sense just suggested, this account can be extended in the following way:

P uses L just in case for each L-sentence , members of P utter just in case either

(1) m₁() = 1 and they are trying to say m₂(),
(2) m₁() = 2 and they are trying to ask whether m₂() is true, or
(3) m₁() = 3 and they are trying to command m₂() be made true.

This is perhaps a bit clunky sounding, but I take it that this sort of approach is more-or-less right and will be able to be used to handle all non-indicatives.<29>

Thus, it should be somewhat clear that providing the actual-language relation for the artificial case of a language with no indexicals or non-indicatives makes impressive headway into the problem of supplying a general theory of sentence-meaning.<30>

3.7 Sub-Sentential Expression Meaning

So far nothing has been said about word-meaning or the meanings of other sub-sentential expressions. But an account of these is expected from a general theory of meaning as well. And if this dissertation will focus for the most part on issues concerning sentence-meaning it is because I take it that sub-sentential expressions have as their meanings whatever it is that they contribute to sentence meanings. So, with respect to sub-sentential expression meaning, little can be said in advance of a theory of sentence-meaning.

3.8 Non-Literal Meaning

There are two main issues with respect to non-literal meaning. One concerns how meaning can be possible without the use of any conventional devices at all, that is, how what I have called language-less speech might be possible. The other concerns how conventional devices can be used in non-conventional ways.

The first of these issues will be addressed by a theory of propositional-speech acts, presuming that such a theory will not need to employ notions of conventional meaning, that is, notions of literal meaning. Grice worked on such a theory in trying to analyze the concept of speaker-meaning. But the present dissertation is concerned with understanding the notion of literal meaning itself and so, the project of trying to come up with a theory of propositional-speech acts is not necessary here.

The second of the issues mentioned above is addressed by a theory that says how we create and interpret non-literal uses of conventional expressions. Grice concerned himself with this issue in part in his famous work on conversational-implicature. Others have dealt with how the use of metaphor, metonymy, and other non-literal devices works. A theory of literal meaning would be pointless if there was no way to have it interface with a theory of non-literal use of conventional expressions. The literal meaning of an expression is, presumably, what we start off with when we interpret an utterance of the expression: if it turned out that we had a theory of literal meaning that didn't help in our understanding of how non-literal usage worked, then, in fact, we had an incorrect theory of literal meaning. In the final chapter of this dissertation I will make a proposal that is directly informed by concerns with the accommodation of non-literal usage.

Still, I will not be able to discuss in this dissertation the details of any theory of non-literal use of conventional devices. I mention this issue only to acknowledge its importance to a complete theory of meaning.

3.9 The Language of Thought Hypothesis

The language of thought (LOT) hypothesis figures into the discussion of the following chapters in a number of different ways. In this section I would like to describe this hypothesis and then show how it follows from another hypothesis which is extremely plausible and widely accepted.

The LOT hypothesis says that we think in a LOT. For simplicity, in this dissertation, I will assume that each person thinks using the same LOT, and I will call this LOT, following custom, Mentalese, or just M. Mentalese is a language whose sentence tokens are neural events or states of some sort. Roughly, the idea of the LOT hypothesis is that propositional-attitude states like believing, desiring, expecting, etc. are realized in us by the tokening of Mentalese sentences in certain ways. There are three main theses of the LOT hypothesis that I will quickly present. I will call these the language-thesis, attitude-thesis and the general-content-thesis. Let me give a brief description of these in turn.

The language-thesis has it that Mentalese consists of things that can be called sentences. M-sentences, however, are not phonological, orthographic, or gestural types as are the sentences of most public languages, but rather they are types the tokens of which are states of or events within our nervous system, or within the physical stuff that makes up some other creature capable of having propositional attitudes.<31> All the sentences of Mentalese, according to the language-thesis, are constructed, in some sense of this term, from a finite stock of items which we may call the simple lexical items or words of Mentalese. The M-words are arranged in M-sentences in accordance with a finite set of rules for combining M-words to make M-sentences. These rules are the syntactic rules of Mentalese. When all the sentences of a language are constructed from a finite set of simple lexical items in accordance with a finite set of syntactic rules it is common to say that that language has a finitely statable syntax. Thus, the language-thesis of the LOT hypothesis also has it that Mentalese is conceived of as having a finitely statable syntax.

The attitude-thesis has it that for each propositional-attitude type there is a unique physicalistic property that an M-sentence token may have the having of which is metaphysically sufficient for the token to realize a propositional-attitude of that type. Thus, according to the LOT hypothesis, there is a specific physicalistic property, call it _B, such that for any person A, a metaphysically sufficient condition for A's having a belief is that both a token s of an M-sentence occurs in A's head and s has the property _B. It is common among philosophers today to simplify talk by referring to the property _B as the property of being in the belief-box.<32> Thus we can say that a metaphysically sufficient condition for A's having a belief is that a token of an M-sentence occurs in A's belief-box. What I have just said about belief will apply to the other propositional-attitudes as well, and thus we end up talking about the desire-box, the expectation-box, etc. For ease of presentation I will throughout only speak of believing and the belief-box, but what I say about these should be understood as applying, mutatis mutandis, to the other propositional-attitudes and boxes as well.

Of course, when people have beliefs, they believe some particular thing, and the LOT-hypothesis seeks to explain what determines the particular thing believed. That's what the general-content-thesis is about. The general-content-thesis has it that for any proposition that a person can believe, there is an M-sentence and a physicalistic property such that has and having and being tokened in the belief-box is metaphysically sufficient for believing .<33> Thus each M-sentence has its own unique property, let's call this the M-sentence's content-determining property, which determines a unique proposition which is, roughly, the proposition that would be believed if the sentence were to be tokened in the belief-box.<34> Often we speak of the proposition thus correlated with a specific M-sentence as the meaning of the M-sentence. But it is best to keep in mind that saying that an M-sentence means some proposition is just a short-hand way of saying that there is some physicalistic property that has which is such that having that property and being tokened in the belief-box is metaphysically sufficient for believing . Notice that I have said nothing about how, given one of these content-determining properties of an M-sentence we might compute what proposition would be believed were the sentence tokened in the belief-box. The general-content-thesis doesn't try to say anything about this - that's why I picked the term general-content-thesis to name this part of the LOT hypothesis. It tells you that each M-sentence has a property that determines the content of any propositional-attitude realized by a tokening of the M-sentence, but it doesn't tell you how that property in fact determines that content. What you need to explain how these content-determining properties in fact determine the contents of propositional-attitudes is what is often called a theory of content. It may be that the attitude-thesis and the general-content-thesis are correct, but that no theory of content can be given. The truth and falsity of the LOT hypothesis does not rest on the existence or non-existence of a theory of content.

I believe that the LOT hypothesis as just stated follows from a widely held and extremely plausible principle, namely, [S]:

This principle, which I will sometimes refer to as the supervenience principle, has it that proposition-attitude properties supervene on physical properties. Let me now say why I think that the LOT hypothesis follows from this supervenience principle.

What I will do is show that if [S] is true, then each of the theses of the LOT hypothesis is true.

Consider first the attitude-thesis. This thesis, again, has it that there is a physical property such that it is metaphysically sufficient for having a belief that an M-sentence has that property and gets tokened in the belief-box. But this will be trivially true if [S] is true. Notice first that there may be a number of different physical properties that are each metaphysically sufficient for having a belief. Their disjunction will itself be a physical property. Thus, that there is some physical property the having of which is metaphysically sufficient for having a belief is guaranteed by the supervenience principle.<35> Call this property _B. All we need to do is to show that this property amounts to the property of having a sentence in the belief-box. All that needs to be done here is to say so. _B is the property of having a sentence in the belief-box.<36>

This might seem strange because when you first hear the LOT hypothesis you expect that its truth will depend on all sorts of complicated detailed facts about neurons and functional states and what-have-you. But none of that is really the case. This will become more and more apparent, I hope, as I go on. I will discuss this sense of the LOT hypothesis more below. For now I will just ask that you keep in mind that the LOT hypothesis is pitched at such a high level of abstraction that it is open to all sorts of interpretations. I am just providing one on which it will be true if the supervenience principle is true.

Let's consider the general-content-thesis next. This says, roughly, that for any proposition that can be believed there will be a physical property that some M-sentence has such that having that property and being in the belief-box is metaphysically sufficient for producing a belief that that proposition holds. But this also will be trivially true if [S] is. For, for any proposition p, [S] guarantees the existence of a physical property that is metaphysically sufficient for believing p, albeit, most likely, a highly disjunctive one. An M-sentence can be understood as any configuration whatsoever of the nervous system or the physical stuff that makes up a believer - nothing in the LOT hypothesis constrains this. Suppose that ₆₉ is the physical property that is metaphysically sufficient for believing that dancing is fun. Suppose that for June to have this property is for her nervous system to be in either state S₁, or S₂, or.... There will have to be some such disjunction if June has a nervous system at all. I will call this disjunction S_D and I will identify it with the M-sentence that means that dancing is fun. For each one of the states in this disjunction will have to be physically related to the whole rest of the world in such a way that ₆₉ holds. And that is just to say that each one of those states can be said to have a property such that ₆₉ holds. But since the general-content-thesis required that for each believable proposition there should be some such property, and since what I have said for the proposition that dancing is fun will just as well hold for any proposition, I have shown that the general-content-thesis is true if the supervenience principle is true.

All that I need do now is show that the language-thesis also follows from [S] and I will have thereby shown that the LOT hypothesis follows from [S]. But since I have taken M-sentences to be disjunctions of states of nervous systems or other physical stuff that some non-nervous believer - as we might call such - might be made up from, and since any such system will be constructed of only finitely many elements, it will follow that all M-sentences are made up from finitely many parts arranged in finitely many relations. But this being true is sufficient for Mentalese having a finitely-statable syntax. So it looks like the language-thesis is true if [S] is as well.

It might be objected here that I have misconstrued the point of the LOT hypothesis wanting a finitely-statable syntax for M. For the point is supposed to be that the finitely many parts contribute to the meaning of the whole sentence somehow and I have not shown that this is the case in just pointing out that any nervous system, or other sort of system of a finite non-nervous believer, will be constructed of finitely many physical parts arranged in finitely many physical relationships. This objection is not difficult to address.

Suppose that we identify the simple lexical items and syntactic structures of M with disjunctions of states of nervous systems in the following way. There will be some disjunction of states such that one of the disjuncts holds just in case a person is having a belief about cats. Let this disjunction be identified with the M-word for cat. Likewise, there will be some disjunction of states such that one of the disjuncts holds just in case a person is having a simple two-place-relational belief where the relation is the is on-relation. Etc. For each sort of thought that a person can have, the disjunction of states that are metaphysically sufficient for having a thought of that sort can be identified with the M-expression or bit of syntax. Clearly, any belief-making state of the central nervous system will be made up of words and syntactic structures so construed. These words and syntactic structures trivially have properties that make them contribute to the meanings of the states which they form parts of. So it seems that Mentalese does have a finitely-statable syntax in the sense required by the objection after all.

This all may seem outlandish at first sight. But I think it is a harmless bit of interpretation that demonstrates the vacuity of the LOT hypothesis baldly stated as I did above. It is an altogether different question whether the LOT hypothesis shown here to be trivially true if the supervenience principle is true can serve the purposes which folks who have been interested in the LOT hypothesis have wanted it to serve.<37> Most notably, it might be wondered whether the LOT hypothesis as construed here can serve for the purpose of making sense of psychological explanation. I think the answer is no. A much hardier LOT hypothesis will be necessary for such purposes. But, it turns out, the weak LOT hypothesis argued for here actually can do some work for us as I hope will be seen in what follows.

3.10 Summary

In this chapter I have laid out what I will take a comprehensive theory of meaning to be like. I have also discussed the LOT hypothesis and I hope that I have successfully made the case that a very weak version of this hypothesis follows from the supervenience principle [S] which holds that propositional-attitude properties supervene on physical properties. I have not mentioned in setting down the framework that I will work within the notion of a compositional-semantic theory at all and this might be puzzling to some who have believed that compositional-semantic theories must play an essential role in a theory of meaning. So in the next chapter I will explain my omitting talk of compositional semantics here.