Natural Meaning for Natural Language

 

David Cole 5-10-98

 

In Book II of the Essay, at the beginning of his discussion of language in Chapter II ("Of the Signification of Words"), John Locke writes that we humans have a variety of thoughts which might profit others, but that unfortunately these thoughts lie invisible and hidden from others. And so we use language to communicate these thoughts. As a result, "words, in their primary or immediate signification,stand for nothing but the ideas in the mind of him that uses them….". Perhaps this is the most natural view of natural language: language is a communication tool, an arbitrary conventional code, for letting others know what you think. Natural language derives its meaning from the thoughts it represents.

This general view of the status of language has enjoyed a lengthy ascendency. Critics, such as Wittgenstein, have not prevailed. The recent focus by Dretske, Fodor, Pinker, and others, on a semantics for mentalese exists along side followers of Grice, who focus on the role speaker intent plays in determining the meaning of natural language. All hold that the principal bearers and determinants of semantic properties are something other than natural language.

Here I hope to develop an alternative to the view that natural language has meaning only derivatively, as an encoding of thought, the primary bearer of meaning. I’ll not follow Wittgenstein into ways of life, but will try to respect the uses of language. Since I think a primary use of language is to convey information about how things are in the world (and not how things are in the speaker’s head), I will fly in the face of Locke’s dire warning that taking words as standing "for the reality of things" is

a perverting the use of words, and brings unavoidable obscurity and confusion into their signification, whenever we make them stand for anything but those ideas we have in our own minds.

I will try to show how the meaning of natural language can be understood as a special case of natural meaning. Language, of course, is at least complicated by the presence of an intelligent language user. But I will argue that linguistic meaning is as much like natural meaning as it could be, given the presence of an intelligent agent in the causal chain linking world and linguistic representation.

Let us look first at physical systems designed to convey information, non-intelligent instruments that are such that their display states mean one thing or another. Having developed a case for a meaning for linguistic displays that is related to natur al meaning, I’ll set out three ways in which meaning for natural language might be understood as derivative of natural meaning. Then, let us reconsider the arguments at the beginning of Paul Grice’s classic paper, Meaning, to see if there is as gr eat a difference between natural meaning and linguistic meaning as Grice makes out.

NATURAL MEANING

In the 40 years or so since Grice’s "Meaning" appeared, attention to his work has largely focussed on his theory of non-natural meaning. Independently, others have been developing accounts of various types of natural meaning. Much of this w ork has been done by Fred Dretske and Jerry Fodor, who have been interested in understanding the meaning of mental representations in terms of information they carry. While many have rejected Grice’s highly intellectualized theory of meaning, as far as I know they have not discussed Grice’s original distinction of the two types of meaning. That distinction remains very interesting, and it is useful, in light of subsequent developments, to revisit his paper.

Grice begins by making a distinction between two types of meaning, then in the rest of the paper he attends almost exclusively to the second of these, which he calls "non-natural meaning". At the very beginning of the paper, Grice invites re aders to consider three sentences having to do with meaning:

"Those spots mean (meant) measles."

"Those spots didn’t mean anything to me, but to the doctor they meant measles."

"The recent budget means that we shall have a hard year."

These are all cases of natural meaning, or, following Grice, meaning(n). Note that not all bearers of natural meaning are natural events or facts – the last example is a budget. Generally, natural meaning concerns symptoms or signs or some state of a ffairs. Other examples might be "smoke means fire", "that drop in barometric pressure means rain", and "the fact that the car is exhausting blue smoke means that it is burning oil.".

In these cases of natural meaning the connection between the bearer of natural meaning and the state that it indicates does not owe to agency. But even in cases like these one could suppose that although no agent is a component of the causal connection, n evertheless the existence of he causal connection itself is the result of an intelligent agent. For example, one could suppose that God has established the connection between measles and its characteristic red spots, or between fire and smoke, perhaps ju st so that we might know when someone has measles or that there is a fire. If God were to act in these ways, he would be what Dretske has called a "structuring cause" – He would cause it to be the case that measles cause red spots, or that fire causes smoke. The meaning of smoke and spots would still be natural meaning, for God would not on each occasion of the appearance of spots or smoke cause them to appear. Rather He would have arranged things so that in the ordinary course of things, mea sles cause red spots and fire causes smoke, so that from red spots one may infer measles, and from smoke one may infer fire.

Cases where natural meaning is at work as the result of structuring agents are particularly interesting for exploring the connection between semantics and pragmatics, and the relation of natural and non-natural meaning. Accordingly, let us conside r the special case of instruments, devices which are designed to convey information about some condition. These have, as it were, artificial natural meaning.

INDICATING SYSTEMS

The lights and gauges on the instrument panel of a car indicate various facts about the condition of the vehicle. That the leftmost light is on may indicate that the oil pressure is low. Or, as we might also put it, that the light is on means that th e oil pressure is low. Although the light is an artifact, it functions as an indicator or symptom of states of affairs. Statements about what the display states of an instrument mean have the same logic as the natural meaning of red spots on the skin and clouds in the sky. Namely, they support inference to certain states of affairs, and it they do not support inference to anyone meaning anything by turning on the light. It is less clear (to use another one of Grice’s tests for non-natural meaning) whet her they support inference to something of the form:

The fact that the leftmost light is on means "the oil pressure is low".

In part this uncertainty, at least on my part, is because I am not sure that this claim, that a fact means a sentence, is coherent. But perhaps we could say :

The fact that the leftmost light is on means the same as would a talking car that said, "the oil pressure is low".

This would relate the meaning of one state of affairs, the fact that the light is on, to the meaning of another state of affairs, one that involves producing an instance of a sentence. This at least seems coherent, unlike Grice’s own example involving a quoted sentence. In any case, it seems reasonable to suspect that Grice’s example has paraphrases that are coherent in idiolects other than his own.

The oil pressure light can malfunction. Then its being on might not mean that the oil pressure is low. Of course, it is supposed to mean that the oil pressure is low. And it usually means that. But, if it is malfunctioning, that is no t what its being on now means. This may appear to be a difference with some forms of natural meaning, where there is an invariable connection between the thing or state said to have the meaning and the meaning. But it is not unusual even with par adigms of natural meaning for the connection not to be invariable. Clouds that usually mean storms sometimes appear without the storms, (and the storms may arrive unheralded by any clouds). Most disease symptoms can appear in a variety of conditions. S o there can be lack of an invariable connection between bearer of natural meaning and its meaning. We might then say "x usually means p".

Common simple single-purpose dashboard indicators are always each directly connected with the sensors for the conditions they indicate. Such single purpose indicators may be supplanted by more sophisticated indicator systems. And in order to understa nd meaning, it may be useful to consider what would happen to meaning in more sophisticated indicator systems.

Suppose then we replace the multiple gauges and indicator lights that typically are placed on the instrument panel of a car with a single digital display that can display numerals and words – a single channel universal display or "scud". A s mall digital processor, with a simple program, will be part of a scud. Let us turn to some plausible design considerations for a scud.

Traditional multi-display instrument panels display a great deal of information in parallel; a scud is a serial system and can only display one item at a time. It can move through items and display one after another. As a result, scud designers will need to make some decisions as to what will be displayed when, and implement those decisions in the commands used in programming the scud.

The program for the scud must implement rules for what symbols should be displayed under what conditions. For example, the display might be capable of displaying

1. an icon of a car outline, next to icon of a thermometer, next to the numerals 72.

Or, alternatively, the sentence

2. "The outside temperature is 72 degrees F".

When should these icons or this string be displayed?

First, given the point of having an instrument panel on the car, namely that a driver wants information, we need a display that will provide the driver with information. Icons have their place, but they may well be thought excessively ambiguous and d ifficult for a driver to discern and interpret. If the primary users of the vehicles will be English speakers, our designers might prefer a linguistic display such as 2). Suppose then our designers decide to program the scud to that it will display 2) o nly if the temperature is 72 degrees F. They will program the scud computer so that it complies with the rule

Rule: Display (2) only if the outside temperature is 72 degrees F.

Let us call this rule the semantic condition on displaying. Note that it sets a necessary condition, namely that the temp be 72 degrees, but not a sufficient condition for displaying. It can’t be a sufficient condition because other messages m ust be displayed, messages carrying information that may be more important to the driver – or ones that should be more important to the driver, such as information about abnormally high current coolant temperature, even if the driver is unconcerned .

So the sufficient conditions for displaying a scud message will be much more complicated than the necessary conditions They will be pragmatic. These pragmatic conditions on displaying must take account of when it is appropriate to display different messages.

Another complication arises because automobile instrument displays typically have more than one mode. For example, at startup (key-on, engine not running) traditional automotive instruments may enter a test mode where all the various lights display. T his enables the user to detect burnt out light bulbs and other malfunctions in the instrumentation system. Fancier displays than those typically found on autos may also have a "demo" mode in which they demonstrate their capabilities. Or a suff iciently complex system could have a "simulation" mode in which its displays simulate what would be displayed in interesting but non-actual conditions – a simulation that might be for training or for entertainment purposes. These modes all con trast with the normal functioning of instrument displays in providing information about monitored conditions.

Let us then suppose that the scud system has other modes, including test modes or demo modes. Then the conditions we have discussing apply only to the normal mode of providing information. Let us call that normal mode "assert" mode. Thus t he semantic condition we have discussed applies only in assert mode, as do the pragmatic conditions discussed above. There will be different pragmatic conditions for test, demo and simulation modes. And the semantic conditions will not be operative in t hese modes. The necessary and sufficient conditions discussed above then are conditions for assert mode only.

Are these conditions connected with meaning? Consider a case where the scud displays 2). Then it may well be the case that the fact that the scud displays "The outside temperature is 72 degrees F." means that the outside temperature is 72 d egrees F. In the case of the scud, there is an internal program that contains a functional equivalent of the semantic assert condition. This embodiment of the condition thus plays a causal role in determining what this particular displayed string means. The pragmatic conditions do not play a role in determining the natural meaning of the display. Even if the display malfunctions such that it fails to adhere to the pragmatic conditions and so displays 2) when it should be displaying say road speed, as during hard acceleration, as long as the semantic condition is still operative, the fact that it displays 2) will still mean that the temp is 72. It shouldn’t be displaying 2), but not because it is inaccurate.

A CONJUNCTION PROBLEM

Suppose we also have as a necessary condition on the display that messages be displayed only if the system voltage is between 11 and 14 volts DC – say this is the range in which the system can function reliably. Then can we conclude that the fact that the system displays 1) means that the system voltage is between 11 and 14? Presumably, Yes – it does mean that. However, the voltage condition is of a special type. It concerns a condition for meeting the semantic condition. The semantic condition is concerned with conventions of message – what symbol string shall be indicative of what condition. The voltage condition is concerned with reliability given the conventions. Thus it is more akin to an epistemic condition. Display asserts should adhere to a convention, should be accurate, and should consider users’ need to know. The former is semantic, the latter pragmatic, and the middle is reliability. Reliability conditions are needed for adherence to the semantic convention.

Several counterfactuals are relevant here. Suppose the semantic rule were changed, so, for example, we made the rule

Display "The outside temperature is 22" only if the outside temp is 72 degrees F.

It would still be the case that, given no other changes in the system, we would need the voltage reliability condition. And similarly, if new system components became available that were more tolerant of supply voltage variation, we might relax the re liability condition. Then the scud’s displaying 2) would no longer mean that the voltage is between 11 and 14.

Users are typically interested in the inference from the display of 2) to the ambient temperature. But a technician might be interested in the inference from the same display to a conclusion about the internal electrical conditions in the system. Th ere is an asymmetry between the two inferences: that any message is displayed will mean that the reliability conditions are met, whereas each differing displayed temperature message will mean something different about the temperature.

Moral: a displayed message will mean many different things, for different reasons. We typically single out the semantic meaning, the meaning that reflects the semantic condition. The reasons for doing this are non-arbitrary, and have to do with the information we are interested in.

A DISJUNCTION PROBLEM

A second problem may arise from the fact that conditions other than the ambient temp being 72 will cause the scud to display 2): for example, if bright sunlight or other source of radiant heat strikes the temp sensor. This type of problem has been a m ajor concern of advocates of indicator semantics such as Millikan, Dretske and Fodor. The first two have had recourse to some naturalistic way of accounting for what the indicator is supposed to indicate. This is certainly reasonable, and accords with the thought, in the scud example at hand, that the scud is supposed to indicate ambient temp, not level of infrared radiation on its temp sensor. Millikan, interested in biological systems, develops an account of proper function based on evolution , and Dretske, who has interests both in biological systems and in instrumentation, appeals more generally to the "structuring cause", which is what brings it about that the system responds as it does. Both of these accounts are historical. F odor tries an ahistorical account in terms of counterfactuals.

But common to all these accounts, it seems to me, is a neglect of a type distinction: between what a current token indication means, and what indications of a type mean. A display of a particular message on a particular occasion is a token display. I f one asks what it means that the scud is now displaying 2), the answer, given the way the scud is now operating, is that the outside temp is 72. This is compatible with the truth of the counterfactual that if there were an intense source of IR near the temp sensor, the scud would also display 2) even if the outside temp were substantially lower than 72. With natural meaning, things mean what they mean, and not another thing. These red spots mean measles, even though spots of that kind might be produc ed by, say, makeup. Those clouds mean rain, even though clouds of that type might be produced by, say, Industrial Light and Magic. We might say,

Given the way the scud is affected by the world, its display of 2) means the outside temperature is 72.

This is compatible with it meaning something else if the scud were affected by the world in a different way, as by IR irradiation of its sensors.

We can certainly talk about what displays of this type mean (that is, the type of which string 2) is a token). They will mean different things, depending on circumstances. If that seems unhelpful, as well it might, we can usefully report that displays of this type usually mean such and such, or that they are supposed to mean such and such. But talk of the meaning of a particular token display need not take note of such complications. And talk of what the display was intended by its designers to display may not be very helpful in the case of an instrument that is miscalibrated. For example, it may be the case that whenever my scud displays a temperature reading, it is reported 10 degrees higher than the actual temperature. Then someone who drives my car will be better informed by knowing how my scud reports temp rather than the conditions under which it is supposed to be displaying what it is displaying. Indeed, how it is supposed to read is neither here nor there, nor is it relevant what other scuds read. The only thing that matters is what there be lawlike connections between my scuds displays and conditions in the world. Then they mean something.

Moral: we must be careful to distinguish what it means for this string to be displayed now from talk about what this string means simpliciter, abstracting from actual displayings of the string by particular systems.

With a simple scud system, the semantic rules programmed into its computer may link entire sentences with conditions. But with even minimal complexity, syntactic rules and substitutional variables will likely appear. Thus even in the temperature case considered so far, it would be very inefficient program design to write rules for each temperature, each of which completely specified the whole sentence. In standard computing practice, we would specify a rule for the display that had the form:

Display "The outside temperature is" T "degrees."

"T" is in effect a substitutional variable that will determine what numeral gets displayed. If we extend the capabilities of the scud system so that it can report the temperature for not just the outside but also say 22 different components of the vehicle, we may well want new rules that introduce new substitutional variables, replacing "outside". If we add the capacity to report other parameters besides temperature for those same 22 components, we would want a variable instead of "temperature". Thus some syntactic competence would be added to the system as a natural result of enhanced information complexity, and in particular, as the result of being able to represent many values of each of a variety of parameters about a variety of different components and systems.

Summary: So far we have seen that in even a simple information display system, there

Will be meaning with the logic of Grice’s natural meaning.

Will be both semantic and pragmatic rules.

Will likely also be reliability rules akin to epistemic.

May well be both assertoric and other modes with different rules.

Will be multiple things that any particular displayed message means.

Will be multiple things that any particular displayed message could mean, but these will generally be other than what the display of the message does mean.

The meaning typically of greatest interest to users will be a causal result of the implementation of semantic assertion rules which establish the dependence of display of a particular string rather than other strings upon certain conditions in the worl d.

GRICE'S DISCUSSION OF MEANING

As noted above in passing, it seems that even simple traditional indicator systems can meet one of Grice’s conditions on non-natural meaning, namely "the lights’ being on means "the oil pressure is low"". I surmise this is ellipti cal for something like "the fact that the light is on means the same as asserting "the oil pressure is low"".

Now let us turn to reconsider the rest of Grice’s distinction between natural and non-natural meaning. Grice gives two examples of what he subsequently calls non-natural meaning, and then he draws five conclusions about the difference between natural and non-natural meaning. His examples:

"Those three rings on the bell (of the bus) mean that the bus is full."

"That remark, ‘Smith couldn’t get on without his trouble and strife,’ meant that Smith found his wife indispensable."

Regarding these, Grice says that it is characteristic of such cases of non-natural meaning that one can say "x means that p, but in fact not p". Thus his first conclusion is:

1. I can use the first of these and go on to say, "But it isn’t in fact full – the conductor has made a mistake"; and I can use the second and go on, "But in fact Smith deserted her seven years ago."

However it seems to me that this may not be true of the bell ringing example! If the bus is not full, then those three rings on the bell did not mean that the bus is full. Perhaps they are supposed to mean that the bus is full, and perha ps they usually mean that the bus is full, and perhaps they were intended to mean that the bus is full, but they did not in fact mean the bus is full. Similar considerations apply to the second example. The remark – Smith’s remarking – did not mean that Smith finds his wife indispensable, although it is plausible to suppose that it was intended to be taken by Smith’s auditors to mean that.

I’ll return to this below. Also, as mentioned above, there appear to be cases of natural meaning where there is not an invariable connection between a type of sign and a type of indicated state of affairs. Thus, I can say "spots like those usual ly mean measles, but not in this case." And one could say much the same of the three rings of the bell on a bus not yet full.

Grice’s second point about natural and non-natural meaning:

2. I can argue from the first to some statement about "what is (was) meant" by the rings on the bell and from the second to some statement about "what is (was) meant" by the quoted remark.

This point seems correct. However, these sentences seem elliptical for an explicit attribution of agency to the bell ringer and the remarker. It is not that something is meant by the quoted sentence, rather it is meant by the speaker in using the quo ted sentence. Compare a statement about what was accomplished by blows to the midriff – this presumably is a covert attribution of agency, elliptical for what x accomplished by x’s blows to the midriff. Similarly, in speaking about what was meant by rin gs on a bell, I am speaking about what x meant by x’s rings on a bell, or perhaps in the case of meaning, what x intended to mean by x’s rings on a bell. The construction attributing meaning of this sort to bell rings and remarks instead of an agent may be a natural accommodation in cases where we don’t know who the agent is.

Grice continues:

3. I can argue from the first sentence to the conclusion that somebody (viz., the conductor) meant, or at any rate should have meant, by the rings that the bus is full, and I can argue analogously for the second sentence.

Here Grice appears to be aware for the first time that we might say something about what the conductor should have meant, by the rings on the bell. But he does not seem aware that doing this creates difficulties for his theory. For he later go es on to hold that "A meant(nn) something by x" is (roughly) equivalent to

A intended the utterance of x to produce some effect in an audience by means of the recognition of this intention….

But when we substitute this equivalent of "meant" in Grice’s third point, we get

(3a) I can argue from the first sentence to the conclusion that somebody intended the utterance of x to produce some effect in an audience by means of the recognition of this intention, or at least should have intended the utterance of x to produce so me effect in an audience by means of the recognition of this intention.

But the last clause doesn’t make sense – the conductor presumably did intend the rings of the bell to produce an effect in an audience, so what is the point of saying that he should have done this? After all, if he should have intended, but did not intend, to produce an effect in the audience, then presumably his rings on the bell don’t mean(NN) anything at all. He was just ringing for the joy of it, or perhaps by accident. On the other hand, if we interpret the "means" in 3 as natural meaning, then it does make sense to say that his rings should have meant(n) that the bus is full – they should have indicated that the bus is full, they should have been equivalent in meaning(n) to the bus making a certain groaning no ise that it makes only when full. Presumably the only reason for having bus conductors ring bells to indicate busses are full is because busses do not reliably make certain sounds all by themselves only when full. Thus it seems Grice is conflicte d in point three, wishing to steer the discussion to speaker meaning, but also aware that it seems to make sense to talk about what someone’s signalling should have meant. But he seems not to notice that this talk of what should have been meant seems mor e naturally understood as indicative of something akin to natural meaning and not his own analysis of speaker intentions.

Grice’s fourth point:

(4) The first sentence can be restated in a form in which the verb "mean" is followed by a phrase in inverted commas, that is, "Those three rings of the bell mean ‘the bus is full.’" So also can the second sentence.

As I have mentioned earlier, I find this an odd construction, saying that rings of a bell mean a sentence. The rings of a bell might have the same meaning as that had by a sentence, but I don’t see how the meaning could just be a sentence. If it were so, if meanings could be sentences, then if would seem it should make sense to say that the three rings of the bell mean something with four words, or mean something with no x’s, or in this example mean something that is English only. If the thre e rings of the bell mean, they don’t mean something proprietary to English, such as a particular English sentence.

So I don’t know what to make of this fourth point. Grice goes on to give an analysis of it later in the paper (at least, that is what I suppose he is doing):

"x means(nn) (timeless) that so-and-so" might as a first shot be equated with some statement or disjunction of statements about what "people" (vague) intend … to effect by x."

But this won’t do as stated, because it is about meaning that such and such, and Grice’s fourth point was about "x means ‘s’". I’m left without knowing what he could have meant in saying that the rings on the bell mean "the bus is full", apart from the paraphrases earlier, to wit: Those three rings on the bell mean the same as saying "the bus is full". But that is compatible with the meaning involved being natural meaning. The bus could have an automatic bell t hat rings when it is suitably weighted down -- or an electronic voice

uttering "the bus is full". Or a human conductor making the utterance. In all cases, the point is to provide information to the passengers, and it seems only economics and the current state of technology might make the choice of means go on e way rather than another.

And now Grice’s final claim:

(5) Such a statement as "The fact that the bell has been rung three times means that the bus is full" is not a restatement of the meaning of the first sentence. Both may be true, but they do not have, even approximately, the same meaning.

This of course is a straightforward rejection of one way of attempting to assimilate meaning(nn), such as the meaning of the rings on the bell, to natural meaning. However, Grice does not offer us any explanation of why (5) is true. And he does not p ursue the point to see if we can’t capture the way in which the rings on the bell mean as a manifestation of a univocal sense of meaning. Of course, by the end of the paper we are clear that this is because Grice is quite sure that there are (at least) t wo distinct senses of "mean", and that his interest appears to be entirely in what he identifies as the second sense, nonnatural meaning.

Grice also says that he thinks the distinction between natural and nonnatural meaning is what people are getting at when "they display an interest in a distinction between "natural" and "conventional" signs." But as we ha ve see in this very last point, this is not clear. Grice’s example in point 5, "The fact that the bell has been rung three times means that the bus is full", appears to centrally involve conventional signs, namely bell rings by a conductor, but Grice says it is not equivalent in meaning to the attribution of meaning that he calls nonnatural. He does not say what this sentence does mean, but it seems pretty clear that it is closer to the "logic" of natural meaning. (In saying t hat the fact that the bell has been rung three times means that the bus is full, I presumably support the inference to a conclusion that the bus is full, which is a feature of natural meaning and is not a feature of meaning(nn); see Grice’s point (1) abov e)). If that is true, then even on his own account, the distinction between natural meaning and nonnatural meaning does not parallel that between natural and conventional signs, contrary to what Grice says here.

NATURAL MEANING FOR NATURAL LANGUAGE

Grice’s point in "Meaning" was to drive a wedge between natural meaning (as in "Those spots mean measles") and the meaning of language, which he called "non-natural meaning" . But the discussion above suggests that the pr oject may not be successful. And there have emerged three ways in which we might understand linguistic meaning in terms of natural meaning. Let "means(nn)" represent nonnatural meaning, "means(n)" represent natural meaning. There is yet another sense of "mean" that is common, and may be seen to be at work in attributions of meaning that involves speakers and other humans agents. Agents can mean to do such and such – they can intent to do such and such. So let "means(i )" represent meaning in the sense of intending, as in "x means to reply to her critics":

Then we can provide three possible analyses of non-natural meaning in terms of natural meaning:

Assertion s means(nn) that p means

1) assertion of s usually means(n) that p, or

2) speaker S means(i) his assertion/utterance of s to mean(n) that p, or

3) assertion of s is supposed to mean(n) that p.

Note first that all of these are compatible with p not following from a particular utterance of s. In the first analysis, the connection between the utterance and the indicated condition is made probabilistic. In the second two, the connection is imb edded in an intensional context (?), intentional and deontic respectively.

There is something to be said for each. Almost all complex indicating systems sometimes fail, so the connection between the indicator _type_ and the type of the condition indicated will be probabilistic. That the phone is ringing usually means that s omeone is calling the number – but not always. That Jim asserts that the phone is ringing usually means that the phone is ringing, but not always. Both may issue false positives.

Of course, if there is no agent producing s, one can’t infer that someone means something by s. And when there is an agent involved in meaning, an agent with a choice of indicators, we infer that the agent means something by assertion of s. But that does not mean that there are multiple kinds of semantic meaning at work.

Let us look more closely at agent or speaker meaning, as in "x means something" and "x means something by asserting s". We need not take this construction to be complete and explicit. It may be elliptical. It appears to have to do with intentions. Perhaps then it is reasonably interpreted as claiming "x means to indicate something (e.g. that p) by asserting s". And if indicating can be understood as a form of natural meaning, then speaker meaning of this sort can be unde rstood as equivalent to:

x means(i) to mean(n) that p by asserting s.

It is certainly reasonable to suppose that if it _does_ mean this, ordinary use would very quickly collapse it to a shorter form, reducing "means to mean" to "means". Or, more in the spirit of Grice, "x means something by s&qu ot; might even mean the same as

"x means(i) to be understood as meaning(i) to mean(n) that p by asserting s."

This makes it a second order intention to mean(n). Either way, on these interpretations natural meaning is at the root of speaker meaning.

The third listed way of understanding "x means that p" looks to the fact that language is normative: phonology, orthography, syntax – all are governed by norms. It would certainly be surprising if semantics was not also governed by norms. N aturalistic philosophers have not been fond of norms, but there is no doubt that norms govern behavior in many areas, including almost all aspects of language. In asking about meaning abstracted from speakers, as in "what does ‘Es Regnet’ mean (in G erman)", we aren’t interested in what any particular speaker might have meant or intended to mean on some particular occasion of uttering the expression. We might mean to ask about what such utterances usually mean, or are intended to mean or are int ended to be taken as meaning. But it may well be that we are asking about proper German usage, German regarded as a rule governed system, and are inquiring about what the German expression is supposed to mean. And again it seems plausible in all these i nquiries to interpret the meaning as natural meaning. Above in the discussion of the SCUD indicator system, we saw that it was reasonable to talk about what a particular display type was supposed to mean, and also what displays of that type usually mean.

All three of the ways of understanding natural language meaning base that meaning upon natural meaning. And all three invoke additional elements that a) seem required for understanding aspects of natural language and b) account for the differences Gri ce noted between natural meaning and meaning as it pertains to natural languages. Utterances will not always mean what one most naturally (!) take them to mean – the connection with the world will be probabilistic. Secondly, they will be uttered with sp ecific conversational intent. And finally, there are norms governing how one should use language in indicating. Humpty Dumpty, in ignoring the norms, is useless, or worse, as a conversant.

CONCLUSION

We have seen three ways of understanding non-natural meaning, or the meaning of natural language use, that base that meaning upon natural meaning. We can understand the attributions of meaning to utterances as probabilistic natural meaning, as reflec ting speaker intention, at one or two removes, to naturally mean, or as normative, as reporting what the utterance is supposed to mean. Which of these analyses is best?

With regard to speaker meaning, an appeal to intention seems appropriate. So "x means something by s" seems to be captured by appeal to the speaker’s intentions, often are often intentions to indicate or at least be taken to be indicating som ething about the world. In Grice’s example, the bus conductor means something by ringing the bell – namely he intends to indicate the bus is full. To move to the second order, and hold that he means to be understood as intending to indicate that the bus is full suggests a certain deviousness. And he may not care what auditors understand, he may just be doing his job, which in part is to indicate that the bus is full when it is full. On the other hand, perhaps his job is not merely to indicate that the bus is full, but to ensure that the bus is not filled past capacity, in which case his intention is that passengers not board when he rings the bell. In that case, he may have the second order intention, namely that he be understood by passengers as int ending to indicate that the bus is full. Or he may merely intend that passengers know that the bus is full from his indicating, by ringing the bell, that it is full. In the latter case, he need have no intention that they recognize his intention – passen gers may not know that a human causes the bell to ring. They can still know that three rings mean that the bus is full.

In the case of linguistic meaning for a type, where we are interested in what "Es regnet" means, or what three rings on an English bus bell mean, abstracted from any particular situation, we may be interested in what they usually mean, or wha t they are usually used to indicate. But a pernicious disjunction problem sets it – an utterance of "es Regnet" usually means that it is raining, and it usually means that the speaker speaks German, is alive, thinks it is raining, etc.

Hence we can either think of attributions of meaning to sentence production as picking out the operation of a semantic condition on the production of the sentence. And where speakers are involved, we can interpret such attributions of linguistic mean ing normatively. In asking what "Es regnet" means, we are interested in what it is supposed to indicate, in the semantic component of its assertion conditions. "Es regnet" is supposed to be used to indicate that it is raining. It’s use will indicate that the speaker is alive, but that is not its semantic purpose. (It could be it’s pragmatic purpose in a particular situation.) This brings us full circle to the meaning of displays of a flexible instrument such as the SCUD.

Clever creatures that we are, we can (wink, wink, nudge, nudge) do many ironic, witty and indirect things with language. But language will be of value to even the most naïve, upon whom all this cleverness is lost. Language embodies information a bout the world. As a result, it is suitable as a medium of communication -- and thought.

BIBLBIOGRAPHY

Cole, David 1997 "Hearing Yourself Think: natural language, inner speech, and thought" http://www.d.umn.edu/~dcole

Dretske, Fred 1988 Explaining Behavior. MIT Press.

Dretske, Fred 1995 Naturalizing the Mind. MIT Press.

Fodor, Jerry 1987 Psychosemantics. MIT Press.

Fodor, Jerry 1990 A Theory of Content and Other Essays. MIT Press.

Fodor Jerry 1994 The Elm and the Expert: Mentalese and Its Semantics. MIT Press

Gauker, Chris 1994 Thinking Out Loud. Princeton University Press.

Grice, H. Paul 1957 "Meaning" Philosophical Review 66: 377-88.

Locke, John (1690) An Essay Concerning Human Understanding. Dover.

Millikan, Ruth 1984 Language, Thought and Other Biological Categories.