LINGUISTIK UMUM

2
communication and information

2.1. What is communication?

The principal signalling-systems employed by human beings for the transmission of information, though not the only one, are languages. We shall be discussing the similarities and differences that exist between linguistic and non-linguistic signalling-systems in the next chapter. The concepts and terminology introduced here are ⁱntended to be appli‑cable to both. It will be assumed that the sense in which the terms `signal', 'sender', 'receiver' and 'transmission' are being employed in this section is clear enough from the context. They will be introduced and incorporated in a simple model of a signalling-system in the next section; and they will be discussed with particular reference to language in later chapters.

A signal is communicative*, we will say, if it is intended by the sender to make the receiver aware of something of which he was not previously aware. Whether a signal is communicative or not rests, then, upon the possibility of choice, or selectⁱon, on the part of the sender. If the sender cannot but behave in a certain way (i.e. if he cannot choose between alternative kinds of behaviour), then he obviously cannot communicate anything by behaving in that way. This, we say, is obvious; and upon it depends one of the most fundamental principles of semantics – the principle that choice, or the possibility of selection between alternatives, is a necessary, though not a sufficient, condition of meaningfulness. This principle is frequently expressed in terms of the slogan: meaning, or meaningfulness, implies choice.

`Communicative' means "meaningful for the sender". But there is another sense of `meaningful'; and for this we will reserve the term ‘informative’ and the cognate expressions ‘information’ and ‘inform. A signal is informative* if (regardless of the intentions of the sender) it makes the receiver aware of something of which he was not pre­viously aware. 'Informative' therefore means "meaningful to the receiver". If the signal tells him something he knew already, it tells him nothing (to equivocate deliberately with the verb 'tell'): it is uninformative. The generally accepted slogan, that meaningfulness implies choice, can thus be interpreted from either the sender's or the receiver's point of view. It is worth observing, at this point, that sender's meaning involves the notion of intention and receiver's meaning the notion of value, or significance. These two notions were referred to in our preliminary discussion of the meanings of 'meaning' (cf. i.i).

Under a fairly standard idealization of the process of communication, what the sender communicates (the information put into the signal, as it were, by the sender's selection among possible alternatives) and the information derived from the signal by the receiver (which may be thought of as the receiver's selection from the same set of alternatives) are assumed to be^. identical. But there are, in practice, frequent III-stances of misunderstanding; and we must allow for this theoretically.

The communicative component in the use of language, important though it is, should not overemphasized to the neglect of the non-communicative, but nevertheless informative, component which is of such importance in social interaction. All utterances will contain a certain amount of information which, though put there by the speaker, has not been intentionally selected for transmission by him; and the listener will commonly react, in one way or another, to information of this kind. We will come back to this point in the following chapter (3.1).

1A now classic work is Cherry (1956). Smith (1966) gives a broadcoverage of the field, with reprints of some of the most important articles

There are two further points having to do with the notion of com­munication which should be mentioned, though they will not be dis­cussed in detail here. The first has to do with the distinction between the actual and the intended receiver of a signal. It is not uncommon for there to be more than one receiver linked to the sender by a channel of communication and for the sender to be communicating with only one (or some subset) of these receivers. The sender may then include as part of the signal some feature which identifies the intended receiver, or addressee*, and invites him to pay attention to, or respond to, the signal. The most obvious case of this in communication by means of language is when the sender uses a name or some other term of address in what we will later refer to as the vocative* function (7.5)^. But the distinction between receiver and addressee is more widely relevant in communication, since, as we shall see later, the sender will often adjust what he has to say according to his conception of the intended receiver's state of knowledge, social status, and so on (14.2).

The second point is of more general theoretical importance: that successful communication depends, not only upon the receiver's recep­tion of the signal and his appreciation of the fact that it is intended for him rather than for another, but also upon his recognition of the sender's communicative intention and upon his making an appropriate be­havioural or cognitive response to it. This has long been a common­place of non-philosophical treatments of meaning and communication (e.g., Gardin(1932); and it has been forcefully argued more recently, from a philosophical point of view, by such writers as Grice (1957) and Strawson (1964).

As far as statements of fact (or what purport to be statements of fact) are concerned, it is generally the case that the sender will intend that the receiver should believe what he is told: that he should hold it to be true and should store it in memory as a fact. Furthermore, the sender's desire to convince the receiver that such-and-such is true commonly derives from, or is associated with, some other purpose. For example, there are all sorts of reasons why we might wish to draw someone's attention to the fact that _it is raining: we may think that he will be pleased to know that he need not water the garden; we may be con­cerned that he should not forget to take his raincoat or umbrella; we may want him to close the window or bring in the washing. The par­ticular purpose that we have in telling someone that it is raining will vary, but there will usually be some purpose over and above our desire to inform him of a fact of which he was previously ignorant. Indeed, it may be the case (and it commonly is) that what we actually say is of itself uninformative, in that the receiver knows (and we may know that he knows) whatever fact it is that we are drawing to his attention. This does not invalidate in any way the notions of communication and in­formation with which we are operating here. There is nothing para­doxical in the suggestion that a non-informative utterance should be produced with the intention that the receiver should infer from it (and from the fact that, despite its banality, it is uttered) something that is not said and in the context need not be said. It may be assumed, how­ever, that the interpretation of non-informative utterances trades upon our ability to interpret the same utterances in contexts in which they would be informative; so too does our ability to infer the very specific and context-bound purposes that the sender might have had for pro­ducing such-and-such an utterance-token on some particular occasion. The sentence 'It's raining' has a certain constancy of meaning which is independent of the specific purposes that someone might have in uttering it. The question is whether this constant meaning of 'It's raining' and of any arbitrary sentence that might be uttered in order to make a statement of fact can be said to depend intrinsically upon some more general notion of communicative intention.

We will not go into this question here. Certain aspects of the question will be taken up, however, in our discussion of speech-acts* in a later chapter (16.1). Meanwhile, it may be pointed out that, whereas it is clearly not essential to the notion of making a communicative and in­formative statement that the person making the statement should be speaking what he believes to be the truth or should intend the addressee to believe what he is being told, these are arguably the conditions under which the communication of factual information is normally assumed to operate. We will tacitly make this assumption throughout most of the book².

The model introduced  here is not restricted in application to communication by means of language. Nor is it intended to cover all aspects, or all kinds, of linguistics communication. The sense in which the term ‘communication’ is being used falls within the scope of the restrictions explained In the previous section; and we are for the present concerned with an even more restricted kind of communication, namely with the communication of what we have been calling factual information.

Our simple model of communication is cast in the terminology of communication-engineering; it is based upon the model described in the now classic work by Shannon and Weaver (1949). Similar block diagrams to the one given here in figure 1 have appeared in a number of

                                                   Signal                      Received

                                              Transmitted                                  Signal

 

X                                      Transmitter                      Chanel                      Receiver                     Y

                                                                            Source of noise

                                                                Figure 1. A model of communication

2lewis (1969) argues, on philosophical grounds, that a convention of  truthfulness is necessary condition for the operation of language

publication in the last twenty years; and the terminology, with certain minor modification, has been widely employed by psychologists, and to a lesser extent by linguists, in general discussions of communication. It is important to realize that terms in question are of much wider applicability than their origin in communication-engineering might suggest; and they should not be thought of as referring solely to some electrical, mechanical or electronic system of signal-transmission. To give just one example from linguistics: Jakobson (1960) has used an analysis of essentially this kind to classify different aspects or functions of language and different kinds of information that is transmitted in language-utterances (cf. 2.4). much general discussion of language was cast in these terms in the late 1950’s.

Let us now interpret the diagram (see figure 1). X is the source* and Y is the destination*. We can describe this by saying, as we have done so far, that X sends some information to Y. but ‘send’ is ambiguous, as far as the present model is concerned, as between the origination of what in communication-engineering is generally referred to as the message* and the actual transmission; and we must distinguish, in principle, between the source of the message and the transmitter. But we can continue to use the word ‘sender’ when, as is commonly the case, the same machine or organism is both source and transmitter, or incorporates both mechanism within it, or when it is not important to distinguish between origination and transmission. The message originated by X is encoded* by the transmitter* into a signal*. The signal is sent over a particular communication channel* to the receiver*. The receiver decodes* the signal into a message and passes the message on to Y. (it would be convenient to have a distinct term subsuming ‘receiver’ and ‘destination’ as ‘sender’ subsumes ‘source’ and ‘transmitter’. We have previously used ‘receiver’ in this wider sense and will continue to do so when there is no conflict between the general sense and the more technical sense just introduced. The distinction between ‘receiver’ and ‘addressee’ was explained in the previous section). It will be observed that the signal transmitted may differ from the signal that is received due to distortions introduced by noise* in the channel. This may, but does not necessarily, lead to a failure in communication. We will come back to this point later: meanwhile we may neglect the source of noise and the differences between the signal transmitted and the receiver signal.

As was said above, we must be careful to interpret all the theoretical terms introduced here in a sufficiently general sense. Telephone signals are transmitted along a wire as a varying electric current; vocal signals, used in speech, are transmitted as sounds waves through the air. Other signaling system make use of oudours, gestures, etc; and different channels may be distinguished according to the nature of the encoding or decoding systems that are used and the operations that are carried out. Some authors have distinguished the channels according to the sense employed by the receiver, and have talked of the visual, aural or tactile channel; but this is perhaps to use the term in a slightly different sense from the sense it bears in communication engineering. A distinction will be drawn later between channel and medium* (3.3).

Not only is it the cast that the source and transmitter, as well as the receiver and destination, may be parts of the same machine or processing mechanism. This is typically so in human and animal communication, where the brain is involved in both the origination and encoding of messages. And there is the further complication that the sender may monitor the signal as he is transmitting it and use feedback from this process, whether consciously or not, to modify  the signal, and even the message, during transmission itself. All these complexities will be ignored for the present.

Some messages, or types of messages, may be channel-dependent, in the sense that they can be signalled only along a particular channel of communication. But most messages of the kind we would describe as factual are, we will assume, channel-free: they could be transmitted along one of various alternative channels, and might be sent simultane­ously along several, or partly along one and partly along another.

Not only one-to-one communication is possible (where both X and Y are single organisms or machines), but also one-to-many and, ill prin­ciple, many-to-one communication. Of these, the former type of communication, where one source, X, sends the same message simul­taneously to several destinations, Y, Y', Y", etc., is common enough: X might be giving a lecture to a group or talking to a number of friends. Many species make use of channels of communication characterized by the properties of broadcast transmission and directional reception. That is to say, the signals are transmitted, like ordinary radio signals, in all directions: they are not beamed to particular points. But the receiver must be turned towards, or otherwise directed to, the transmitter in order to pick up the signals. Broadcast transmission and directional reception are particularly appropriate to one-to-many communication.

More important for semantic theory than properties of the channel, are the following possibilities: (i) that two or more signals may be equi^valent, each encoding the same message; and (ii) that a particular signal may be ambiguous, encoding more than one message. Equiva­lence and ambiguity of utterances are common phenomena in language.

Both the source X, and the destination, Y, will be in a certain state of knowledge ^or belief: each will have stored in his memory or brain a certain number of propositions*. What precisely is meant by the term `proposition' we need not at this point enquire (cf. 6.2). It is sufficient to say that propositions are expressed by sentences (and contained in utterances) and may be either true or false.

The communication of factual, or propositional, information can now be described, in terms of our simple model, as follows. X has accessible to him a set of propositions (p₁, p₂, p₃, . . ., p_n). He selects one of the set, p_i, and encodes it (or has it encoded) as a signal and transmits it (or has it transmitted) along the channel of communication; the signal is decoded and reaches Y as a message (which \^,.^,c will assume has not been distorted or garbled). Provided that the proposi­tion, p,, is not already included in Y's store of knowledge and is not in conflict with any of the other propositions known to Y (or held to be true by Y), it will be accepted and stored by him; and Y's store of factual knowledge (or belief) will be thereby augmented. As far as normal communication by means of language is concerned, we can say that X makes a statement expressing what he holds to be a true proposition, and that, if the communication is successful, Y believes the proposition and remembers it. If Y is already in possession of p_i, the signal encoding p_i, as we saw in the previous section, is uninformative and Y's state of knowledge is unaffected. But if p_i is in conflict with any other proposi­tion, p_j, in Y's possession, Y can do one of various things: he can reject p_i as untrue; he can store p_i in place of p., which it conflicts with; he can suspend judgement, storing in effect the information that either P_i or P_j (but not both) is true.

The account that has just been given of the communication of pro­positional knowledge is admittedly very schematic and highly idealized. But it is not without value. One of the advantages of putting things in the way that we have done is that, by drawing our attention to the deficiencies of our simple model, it enables us to see more clearly the various factors that are involved in the process of communication and their interaction. Let us now list some of the more obvious deficiencies.

(1) Nothing has been said about the possibility that some propositions may be more accessible than others, and that their accessibility may vary from time to time; and no allowance has been made for the possibility that facts can be temporarily or permanently forgotten. Our assumption has been that all propositions are either held in store and are imme­diately accessible or are not held in store at all. This is clearly un­realistic. If we were to attempt to implement our model of communica­tion oil a general-purpose computer, these questions of accessibility and storage would have to be resolved at the outset.

(ii) No distinction has been drawn between propositions that are known a priori* (i.e. independently of experience) and propositions that are known a posteriori* (i.e. on the basis of experience). The status of the distinction between truths of reason and truths of fact, as Leibniz called them (cf. Russell, 1949: 207), has been a central issue in philosophy from the earliest times; and in one form or another it will be of concern to us throughout this book. Its relevance in the present context is that if two propositions are in conflict, it obviously makes a difference that one should be held to be contingent and the other to be an irrefutable truth of reason. Even if ,we reject the distinction between truths of reason and truths of fact (between the necessary and the contingent, between the analytic* and the synthetic*: cf. _6.5), we must admit that some of the propositions that we hold to be true occupy a more central place than others do in our view of the world; anti holds, not only with respect to propositions that tend to excite the interest of philosophers, but also with respect to the most mundane propositions descriptive of our everyday life. Not all the propositions that are put to us compete for our attention and acceptance on equal terms.

(iii)             No distinction has been drawn within the set of propositions that are held to be true as a matter of empirical fact between those that derive from one's own observation and immediate experience and those that have been learned from others or arrived at by a process of reason­ing. And yet the source of our beliefs will surely make a difference to the degree of conviction with which we adhere to them; and it may make a difference to the form in which they are stored.

(iv) No account has been taken of the fact that the participants in the communicative process will always have some knowledge or beliefs about one another and will be continually adjusting their view of one another (and, in particular, of one another's sincerity and reliability) in the course of their communicative interaction, When X informs Y of p_i, Y will not only accept or reject p_i himself, but (unless he has reason to doubt the sincerity of X) he will also store the fact that X holds pi to be true; and that X holds p_i to be true will itself be a proposition whose truth Y can assume in any further dealings he has with X. Similarly, provided that X has no reason to doubt that lie has been successful in communication, to Y and in getting him to believe it, lie can henceforth act on the assumption, not only that Y now holds p_! to be true, but also that Y knows that X knows that Y holds p_i to be true. Obviously, Y's evaluation of the sincerity and the reliability of his in­formant is going to be an important factor in the determination of whether Y accepts p_i as true or not. It is not only the inherent plausi­bility of p_i or its relationship to other such propositions in Y's store of factual knowledge that is relevant. The same proposition might be communicated to Y, in what are otherwise the same circumstances, by two different informants and be accepted as true from the one, but rejected as false from the other.

These are perhaps the most serious of the deficiencies in the deliberately simplified model of communication that has been presented in this section. Since we are not concerned with the construction of a realistic model of communication as such, but only with the way in which the structure of language is determined by the communicative functions that it is called upon to perform, we shall not go into the questions that have been raised here in any detail. However, the points that have just been made are important and should be borne in mind throughout this chapter. So too should the more general point that not all of our knowledge is correctly described as propositional.

2.3. The quantification of information

So far we have been using the term 'information' in more or less its everyday sense. There is, however, another sense in which the term has come to be used in the study of communication. In order to avoid con­fusion we will draw a terminological distinction between these two senses of 'information'. The first kind we will call signal-information*`; the second we will call semantic information*.

The distinction between these two senses of 'Information' (and they are not always kept clearly apart ⁱⁿ the literature) has to do with the difference, between identifying a signal (as s_i, rather than s_j) and inter­preting it in terms of the message (p_i or p_j) that it encodes. For example, if the sentences Ile had a book and He had a look are produced as spoken utterances (i.e. is signals in the vocal-auditory channel: cf. 3.1) they will differ acoustically in that where one has .what ,%-e may here refer to as a b-sound the- other has an 1-sound. Any physical properties of the sig­nals which enable the receiver to identify a particular sound as b rather than 1 (or any other potentially occurrent sound), and consequently to identify a particular form as book rather than look (or any other poten­tially occurrent form), may be described as signal-information: this is acoustic information in the case of spoken utterances, and visual in­formation in the case of written utterances. 'Semantic information', on the other hand, is closer to, and, if it is defined as a theoretical term, can be said to explicate the non-technical, or everyday, term 'informa­tion'. It will be clear that 'information' was used in the sense of "semantic information" in the first section of this chapter. When we say that a signal is informative we imply that it is conveys, some semantic information to the receiver (that it tells him something). Signal-information and semantic information, though they must be distin­guished, interact, as we shall see, in the process of decoding and interpreting utterances.

It is the notion of signal-information that has been quantified within the mathematical theory of communication: or information-theory*, as it is often called. Information-theory, which was originally developed with reference to the very practical problem of maximizing the effi­ciency of signal transmission in terms of cost and reliability, has not so far justified all the claims that were made for it by some of its early enthusiastic proponents. But it has been the source of much speculation and experiment in the last twenty years or so; and the basic concepts are of the greatest importance _in any discussion of com­munication. Only the very briefest summary need be given here³.

3A more popular exposition than Shannon & Weaver (1949) is Weaver (1949). For further details and possible applications, cf. Miller (1951), Cherry1957), Smith (1966). Hockett (1953) did much to familiarize linguists with the general principles.

Let us begin by supposing that there is a fixed and finite set of poten­tial messages any one of which X may wish to send to Y. Let us further suppose that each message can be encoded by means of one and only one signal. Y knows that X is about to send him a signal, but lie does not know which one of the inventory of signals it is to be. We will now define signal-information content, as a function of Y's expectancy, interpreting 'expectancy' in terms of probability of occurrence. We will first assume that Y's expectancy is the same for all signals; that is to say, lie no more expects to receive any one signal than he does any of the others. Identifying expectancy with probability, we can say that Y regards all the signals as being equiprobable. Y's knowledge or calculation of the probabilities might depend, in principle, on various factors. However, it is customary to define 'probability' in this context as being equivalent to 'statistical probability': i.e. in terms of frequency of occurrence in a statistically stable system. This means, in effect, that Y's expectancy is assumed to be-determined solely by the relative frequency with which he has received such signals in the past. Given that the signals are equiprobable we can say that they all contain the same amount of signal-information.

How signal-information content is quantified (in terms of bits*) is of no consequence for our present purpose. The important point is that signal-information content is inversely proportionate to probabilit^y of occurrence: the greater a signal's probability of occurrence the less signal-information it contains; an,^; if the occurrence of a particular signal is totally predictable (i.e. if it has a probability of i) it carries no signal-information. Given that we have some measure of signal-information based on these principles, we can calculate both the capacity of the channel and the signal- information content of any signal. If a sig­nal takes up more of the channel-capacity than is necessary, it will be to that degree redundant*. Redundancy in a signalling-system is measured as the difference between its maximum potential capacity and its actual capacity. By reducing the redundancy in a system we reduce the cost of transmission; but, as we shall see, we also lessen its reliability. To sum­marize: there are two principles of general importance deriving from information-theory. The first is the principle that information content is inversely proportionate to probability of occurrence; the second – to which we will return – is that a certain amount of redundancy is not only inevitable, but desirable.

So far we have taken messages as unanalysable wholes and have thought of them as being encoded holistically as signals. Linguistic signals, however, are not holistic; or, at least, not characteristically so. Utterances have an internal structure. In fact, they have two levels* of internal structure: the level of forms and the level of sounds (^cf. 3.4 Statistical considerations are relevant at both levels.

Let us first take the lower, or secondary, level and, for simplicity, illustrate its statistical structure with reference to written English. It is well known that different letters of the alphabet vary in their overall frequency of occurrence in any sufficiently large and representative sample of English: e occurs more often that t, t more often than a, a more often than i, and so on. We can calculate these relative frequencies of occurrence and assign to each letter an overall probability of occurrence: distinguished from the overall probability of occurrence, however, are various kinds of conditional* probabilities, which we may group under two main heads: positional* and contextual*. By the positional fre­quency of occurrence we mean the relative frequency with which a given letter occurs in a particular position in the structure of a word-form: in initial position, in final position, and so on. For example, in English b has a high probability of occurrence in word-initial position, but a low probability in word-final position. By the contextual frequency of occurrence is meant the relative frequency with which a given letter occurs in the context of one or more other letters. It is important to appreciate that 'context' does not here imply contiguity. Where the context is contiguous with the letter whose contextual probability we are calculating we may use the more specific term transitional* probability. For example, the transitional probability of occurrence of r in the context of an immediately preceding t is relatively high; but in the context of an immediately preceding s (in the same word-form) it is very low.

Applying the general principle that signal-information content is inversely proportionate to statistical probability and using the same formula as before, we can calculate the signal-information content of any given letter for particular structural positions and particular con­texts. In the limiting case the conditional probability of a given letter sill be 1 (i.e. its occurrence will be completely determined by either structural position or context). For example, this holds in English (apart from a small number of borrowed words and transliterated proper names) with respect to the letter u after an immediately preceding q. In this context, therefore, u has no signal-information content: it is completely redundant.

Exactly the same principles apply at the level of word-forms. The overall probabilities of particular word-forms in English have been calculated for different kinds of texts and published in various so called word counts. It is more tedious to calculate positional and contextual probabilities; and very little useful information about conditional probabilities at this level is available. However, it is intuitively clear that our ability to guess what word has been omitted from a text is increased by our being able to draw upon our knowledge of its structural position and context; and this ability has been experimentally verified on many occasions. Just as the letter u is redundant after q, so the word-form to is redundant in a context like want... come' (regardless of what precedes want or follows come) provided that it is known that only a single word-form (or alternatively a form of such-and-such a length in terms of transmission-time or the number of lower-level units it contains) has occurred: in contexts like this to has no signal-information content. It could be omitted without affecting the message that is being sent.

We come now to the second main point in our discussion of informa­tion-theory: this is that a certain degree of redundancy* is essential, not only in language, but in any communication system, in order to counteract the distorting effects of noise*. The term 'noise', which is here being employed in the technical sense that it bears in communica­tions-engineering, refers to any disturbances or defects in the system which interfere with the faithful transmission of signals. Every channel of communication, whatever its physical properties, is subject to a greater or less amount of random noise; and the noise will obliterate some of the signal-information transmitted along the channel. Conse­quently, the information content of the received signal will differ, to a greater or less degree, from that of the transmitted signal (see figure 1). If the transmitted signal were free of redundancy, the information lost in the course of transmission could not be recovered by the receiver and the signal would be incorrectly decoded. Misprints in the written language fall within the scope of the concept of noise: they are often not noticed by the reader because the redundancy of even quite small sections of written text is high enough to outbalance the loss of informa­tion. As far as the spoken language is concerned, noise can be taken to comprehend any kind of distortion of the utterance, whether this is due to the imperfect performance of the speaker and hearer or to the acoustic conditions of the physical environment in which the utterance is produced.

The ideal system is one which encodes just enough redundant in­formation in signals to enable the receiver to recover any information lost as a result of noise. If the conditions of transmission are relatively constant, and if the degree of reliability expected is also relatively constant and does not vary according to the kind of message that is being transmitted, it is possible to design a system which approximates to this ideal. It should be clear that the conditions under which linguistic sig­nals are transmitted vary enormously from one occasion to another, and that hearing exactly what is said is more important in some situations than in others. One might not, therefore, expect languages to approxi­mate very closely to the ideal of signalling efficiency. That some general principle of signalling efficiency is operative in the historical develop­ment of languages has often been suggested, though it has never been convincingly demonstrated that signalling efficiency, in this sense, is a major determinant of language-change. One manifestation of the principle of efficiency is the tendency to shorten the most frequently used forms; and the operation of this factor ⁱⁿ the historical develop­ment of languages, on a small scale at least, is well attested.

Signal-information content, as measured by the mathematical theory of communication, has frequently been referred to as surprise-value; and it is this aspect of the theory, if any, which links the two senses of `information', which we are distinguishing by means of the terms `signal-information' and 'semantic information'. For, in a general way, it does seem reasonable to say that the greater the surprise-value of a certain item of news, the more significant it is (in some sense of `significant'). "Man bites dog", as they say, is a more significant item of news than "Dog bites man". As we have seen, when a signal (or some portion of a signal) has a probability of I (i.e. when its occurrence is totally predictable), it is defined to carry no signal-information. This seem^s to be intuitively acceptable also as far as its semantic information is concerned. If the receiver is aware that a certain message will neces­sarily be transmitted, reception of the signal which encodes that message, when it is transmitted, will not augment his store of knowledge. The information-theory principle that what is completely by its context carries no information is in accord, therefore, with the principle introduced in the previous section that meaningfulness implies choice.

The illustration given earlier of information-theory principles was artificially simplified. It was assumed that there was a fixed and finite number of potential messages and that each of these could be encoded holistically into one and only one signal. There are certain situations in which a restricted part of a language is used in this way. But they can hardly be regarded as typical of the use of language. There are alter­native ways of encoding the same message (i.e. different utterances can have the same meaning) and one signal can encode more than one message (i.e. utterances can be ambiguous); and it is certainly not the case that speaker and hearer have stored in their brain a table listing all possible messages together with the appropriate signals for encoding them. Furthermore, the reception of utterances cannot be split sharply into two distinct processes: first the identification of the signal and then its interpretation. In the decoding of the acoustic signal, the listener draws upon his knowledge of the positional and contextual probabilities of words even for the identification of sounds; and his calculation of these conditional probabilities of words is not determined solely by his knowledge of the statistical structure of the language, if indeed lie can be truly said to know the statistical structure of his language. He is influenced also by his more general expectations of what the speaker wants to say in the particular situation; that is to say, he decodes the signal, partly at least, in the light of what he thinks the message will be. That is to say, signal-information and semantic information interact, in a very complex manner, in the processing of language-utterances; and any theoretical model of the production and reception of speech must take account of this. Because of this complex interaction between signal-information and semantic information, and for other reasons too, there are immense, and perhaps insuperable, problems attaching to any pre­cise application of information-theory to the processing of language-utterances. It does not follow, however, that the general principles are not applicable; in particular, the principle that signal-information con­tent is inversely proportionate to the expectancy of the receiver. The main difficulty is that the receiver's expectancy is not solely a function of the statistical probabilities of sounds (or letters) and words. Prob­abilities of a different, and perhaps more subjective, kind are also, relevant.

We may now raise the question whether semantic information content is quantifiable, as signal- information content is. Certain proposals have been made relating to the quantification of semantic information. Although they have not so far been developed to the point that they can be profitably applied to the everyday use of language (and it is not clear that they are in principle capable of being developed to this point), they are worth discussing briefly for their explication of at least certain aspects of what it means to inform someone of a fact of which he was not previously aware. We will restrict ourselves to a very informal account of the theory of semantic information put forward some years ago by Bar-Hillel and Carnap (1952).

Let us consider a situation in which X is describing some state-of-­affairs to Y; and let us assume that this state-of-affairs is totally describ­able by means of the assertion of just four propositions: p, P21 P3 and P4- We might, for example, be concerned with the question whether each o' four person, (a, b, c and d) is married or not. Y's initial state of knowledge with respect to this question, we will assume, is minimal. There are sixteen possible states-of-affairs, any one of which might be the actual state-of-affairs. Consider now the effect of communicating to Y any one of the four propositions: e.g. the proposition p, that a is married. If he accepts this s as true he will put it into his store of factual knowledge; in effect, he will eliminate from the set of sixteen possible states-of-affairs each of the eight which, is incompatible with the fact that a is married. His prior doubt about the actual state-of-­affairs is halved. Let us now suppose that subsequently another and different proposition, P2, that b is mar: led, is communicated to Y and accepted as true. The set of states-of-affairs that he is prepared to enter­tain as possible is again reduced by a factor of two (from eight to four). If p, and p, had been combined and communicated together, the effect, it will be noted, would have been the same.

It is on the basis of considerations such as these that Carnap and Bar­Hillel have developed their theory of semantic information. They first define a state-description* as a complete set of propositions describing some possible state-of-affairs, and then define the semantic content of a proposition to be the set of state-descriptions that it eliminates. What this means should be clear from the simple example just given. It should also be clear that, although we do not usually talk in this way about what it means to be informed of some fact, this notion of semantic content agrees well enough with our pre-theoretical notion of what constitutes the semantically informative aspect of a proposition. A tautology, such as the proposition that a is either married or not, tells the recipient of a signal communicating to him such a proposition nothing of which lie was not previously aware; and it would fail to exclude any of the sixteen states of the universe which Y is prepared to entertain, initially, as possible. A contradiction, such as the statement that a is both married and not married, is uninformative, however, in the dif­ferent, and somewhat paradoxical, sense that it has too much content: "it excludes too much, and is incompatible with any state of the universe" (Bar-Hillel, 1964: 301)^. To put it in psychological terms, it baffles him by purporting to tell film more than he can accommodate in his conceptual scheme. Finally, if the class of state-descriptions excluded by proposition p includes, and is not included in, the class of state-descriptions excluded by another proposition q, then p is seman­tically more informative than q. Thus p-and-q is semantically more informative than either-p-or-q. Carnap and Bar-Hillel deliberately refrain from making their notion of semantic content relative to a re­cipient's state of knowledge. It seems clear, however, that, within certain limits at least, it can be interpreted in this way. For example, it can be maintained that, after Y has been informed that p is true the subsequent transmission to him of both p and q is no more informative than would be the communication of just q.

At this point, a cautionary remark should be made in qualification of the principle that tautologies are semantically uninformative. Whilst this is intuitively acceptable as far as many tautologies are concerned, there are certain propositions, which are tautologous, or necessarily` true, in that they follow logically from other propositions taken to be axiomatic and which nonetheless would normally be held to be in­formative. The most obvious examples are mathematical propositions. For example, it is demonstrable that "(x²—y²) — (x+y) = (x— y)" is necessarily true for all values of x and y; and yet the communication of this proposition to someone who had not previously known or realized that it was true would generally be taken to augment his store of knowledge. The status of such tautologous, but apparently informative, propositions has been of central importance in modern philosophical semantics (cf. 6.5 and 7.3).

Carnap and Bar-Hillel go on to define the amount of semantic in­formation conveyed by a proposition in terms of Carnap's (195o) notion of logical probability. As we have seen, the notion of probability upon which the Shannon–Weaver measure of signal-information content is founded is defined in terms of relative frequency: we have referred to this as statistical probability. Many modern treatments of the theory of probability suggest or imply that there can be no objective definition of probability other than in terms of relative frequency of occurrence. Carnap disagrees, and claims that two fundamentally dis­tinct senses of the pre-theoretical expression 'probable' are being confused and that each has its own range of applicability. When we say The probability of throwing a six with this dice is 1/6 we are pre­sumably making an appeal to the notion of statistical probability. But when we say The probability of rain (on the evidence of certain meteoro­logical observations) is 1/6 we are ascribing a certain logical relationship (of a_ll inductive kind) to two propositions: the first proposition is the hypothesis "It will rain"; the second is the evidence reporting the relevant meteorological observations (this will normally be a complex proposition composed of many simple propositions). This kind of prob­ability, which Carnap calls logical and which, in his view, is the basis for inductive inference, is often described as subjective and explained in terms of degree of belief or certainty. But Carnap defines it, in terms of the degree of confirmation of a hypothesis with respect to a given body of evidence. It would be a separate question whether anybody's degree of belief in the hypothesis, however measured, was equal to the degree of confirmation. The point is that logical probability can be defined, and measured, as a property of some system of propositions in abstraction from the beliefs of the users of the system; and Carnap's theory is intended to measure logical, or inductive, probability     in this sense. We will make reference to this notion of logical probability in our discussion of modality in a later chapter (cf. 17.1).

It is within this general framework that Carnap and Bar-Hillel define their notion of semantic information. The basic idea is that semantic information, like signal-information, is equivalent to the elimination of uncertainty. The difference between the two kinds of information can be expressed by saying that the one eliminates uncertainty as to what the signal is and the other uncertainty as to what the message is. In both cases, however, there will be the same kind of inverse relationship between probability and information content; the greater the statistical probability of a certain signal, the smaller will be its signal-information content; the greater the logical probability of a proposition (whether transmitted as a message or not), the smaller will be its semantic in­formation content. It turns out, however, that there are alternative ways of interpreting, and then measuring, the semantic information contained in a proposition, according to whether we have in mind the absolute number of state-descriptions that it enables us to eliminate or the relati^ve number of state-descriptions with respect to the number of state-descriptions not eliminated by previously given propositions. Without going into the numerical relationship which holds between these two measures of semantic information in terms of the theory pro­posed by Carnap and Bar-Hillel, we can perhaps see that it would be reasonable to distinguish two senses of 'informative' along these general lines, and thus to make precise (in perhaps various ways) that particular sense of 'informative' (or 'significant') in which we say that some facts are more informative (or significant) than others. It is in this sense that semantic information content is analogous to signal-information content. Whether it should be measured in terms of logi­cal probability, or in relation to some other notion of expectancy is, however, another matter; and we will not pursue the question further. It is the notion of a state-description as a set of propositions describing some actual or possible state-of-affairs that will be taken up later (cf. 6.5).

2.4. Descriptive, social and expressive information

So far in this chapter we have deliberately restricted our attention to a consideration of what is involved _in the transmission of factual, or pro­positional, information. It is difficult, and at this stage of the discussion it would be impossible, to make more precise than we have done already what is meant by factual information. For the present, we will rest con­tent with the statement that a piece of information is factual if it pur­ports to describe some state-of-affairs.

Many semanticists have talked as if language was used solely, or primarily, for the communication of factual information. Others have maintained that making statements descriptive of states-of-affairs is but one of the functions of language; that it also serves, as do our other customs and patterns of behaviour, for the establishment and main­tenance of social relationships and for the expression of our attitudes and personality. We will not go into this question in detail at this point. Let us simply assume that these are three more or less distinguishable functions: the descriptive*, the social* and the expressive*. Correlated with these three different functions we can recognize three different kinds of semantic information encodable in language-utterances. De­scriptive information (or descriptive meaning) is factual in the sense explained above: it can be explicitly asserted or denied and, in the most favourable instances at least, it can be objectively verified. An example of an utterance with descriptive meaning is the statement It is raining here in Edinburgh at the moment. Whether this utterance necessarily or normally contains any additional non-descriptive information is a ques‑tion we may leave on one side for the present. It is descriptive meaning, as we shall see later, that has been of central concern in philosophical semantics. Other terms that have been used in the literature for this aspect of meaning ng include 'referential', 'cognitive', 'propositional', `ideational' and 'designative'.

4Benedetto Croce is very rarely mentioned by English-speaking linguists and philosophers of language. On the other hand, as Lepschy (1966: 98) points out, he is one of the very few scholars whose influence Sapir (1921) explicitly acknowledges. Croce's work is set in the more general context of the develop­ment of semantics by De Mauro (1965). Although I have contrasted Croce and Vossler with Firth and Malinowski, no-one can read the works of any of these four scholars (or Sapir) without realizing that each of them acknow- ledges the role both of social constraints and of individual creativity in the use and development of language: it is at most a question of balance. The Prague School linguists, too, have always been appreciative of the comple­mentary roles of the social and the expressive functions of language (cf. Garvin, 1955; Vachek, 1964).

The distinction between expressive and social meaning is far from clear-cut, and many authors have subsumed both tinder a single term ('emotive', 'attitudinal', 'interpersonal', 'expressive', etc.). If we define expressive meaning (in a narrower sense than it often bears) to be that aspect of meaning which "covaries with characteristics of the speaker" (Brown, x958: 307) and social meaning to be that aspect which serves to establish and maintain social relations, it is clear that the two are inter­connected. For it is only by virtue of our membership of social groups that we arc able to interact with others and, in doing so, to establish our individual identity and personality (cf. Argyle, 1969). The most appro­priate term for what is common to the social and expressive functions of language (and of other human signalling-systems) is interpersonal* (cf. Halliday, 1970: 143). It is convenient, however, to allow for the terminological distinction of the two functions, since one of the points of controversy in linguistic theory is the degree to which the individual is constrained by social conventions in the use of language. Such writers as Croce (1902) and Vossler (1932) have perhaps exaggerated the role of individual creative expression in language, whereas others, like Malinowski (1935) and to a lesser extent Firth (1950), may have given undue emphasis to the force of social constraints.⁴

A somewhat different tripartite classification of the functions of lan­guage from the one that we have adopted here was put forward some years ago by BÜhler (1934) and figures prominently in many influential treatments of language. Two of I361 lei's functions, for which fie em­ployed the German terms 'Darstellung' ("representation") and 'Aus­druck' ("expression"), correspond closely with what we are calling the descriptive and the expressive functions, respectively. The third, for which BÜhler used the term 'Appell', is what we will call the vocative* function. BÜhler's classification is based upon his analysis of the typical speech-act (Sprechakt) in terms of three essential components: the speaker, the addressee and the external situation to which reference may be made in the utterance.⁵ According to whether reference is made primarily to one rather than the other two of these three components, so the utterance will be primarily expressive, vocative or descriptive in function. There is an obvious connexion between BÜhler 's analysis and the traditional analysis of the typical situation of utterance as a drama in which three roles are given grammatical recognition by means of the category of person* (cf. 15.1); and BÜhler and his followers have explicitly mentioned this connexion. They have emphasized, however, that it is not only utterances with a first-person subject that are expres­sive and not only utterances whose subject is a second-person pronoun that have a vocative function. They have also stressed the fact, as we are doing here, that few, if any, utterances have one function to the exclusion of the others. As we shall see later, BÜhler 's tripartite classifica­tion is also relevant to his distinction of symptoms, symbols and signals: every utterance is, in general and regardless of its more specific function, an expressive symptom of what is in the speaker's mind, a symbol descriptive of what is signified and a vocative signal that is addressed to the receiver (cf. 4.1).

BÜhler 's scheme has been modified and extended by Jakobson (196o). The principal modification consists in the substitution of 'conative' for 'vocative' (i.e. 'Appell'). This is not a purely terminological sub­stitution (as is perhaps Jakobson's substitution of 'referential' for 'representational' and 'emotive' for 'expressive'). By using the term 'conative' and explicitly associating it with BÜhler 's notion of orienta­tion towards the addressee, Jakobson is presumably implying (as others have done) that it is primarily as an instrument in the satisfaction of the speaker's wishes and desires that the addressee is invoked. The conative* function of language is thus closely linked with what is commonly called its instrumental* function: i.e. its being used in order to achieve some practical effect. Furthermore, as we shall see in our discussion of mood and modality later, it is not always possible to draw a sharp distinction between utterances expressive of the speaker's wishes and utterances which serve as directives imposing upon the addressee some obligation (cf. ^I S. i). But we may leave this particular point for the present, noting only that what Jakobson and others have referred to as the conative function of language merges with the expressive function, on the one hand, and the instrumental function, on the other.

BÜhler 's scheme is extended by Jakobson by bringing into considera­tion three further components of the communicative process and recog­nizing that each of these may be the focal point, as it were, of the utterance. The first of these additional components is the language that is used (or, in Jakobson's terms, the code*). Any utterance whose primary function it is to verify that the interlocutors are using the same language or dialect, or using expressions of the language in the same way, is said to be metalinguistic*. Enough has been said about this func­tion of language in the previous chapter (1.2; 1-3).

6Laver (1975). argues that the notion of 'phatic communion' should be given a somewhat broader interpretation.

5This is a different sense of 'speech-act' from the sense in which it is used in work based on Austin (1962): cf. 16.1.

The second additional factor is the channel of communication (cf. 2.2). Many utterances of everyday conversation have as their primary communicative function that of opening up or keeping open the channel. For example, there are all sorts of conventional greetings (Good morning!, etc.) or ritualized gambits (Wonderful weather we are having!, etc.) with which we can initiate a conversation. There are others with which we can bring a conversation to a mutually acceptable conclusion (It's been lovely to see you again!, Give my regards to your wife, etc.); and others that serve to prolong the conversation or to indicate to the speaker that the addressee is still ⁱⁿ contact and following what is being said. Much of this interaction-management information, as it has been called (i.e. "the information that the participants exchange in order to collaborate with each other in ordering the temporal progress of the interaction": Laver & Hutcheson, 1972: 12), is transmitted by means of paralinguistic* signals (eye-movements, gestures, posture, etc.: cf. 3.2). But some of it is encoded in the verbal component of language-utterances. Malinowski (1930) had coined the term 'phatic communion' for that kind of speech "in which ties of union are created by a mere exchange of words" – a kind of speech which, he says, "serves to establish bonds of personal union between people brought together by the mere need of companionship and does not serve any purpose of communicating ideas"; and Jakobson borrows the term phatic* in order to refer, more particularly, to that function of language which is channel-oriented in that it contributes to the establishment and maintenance of communicative contact. Thus interpreted, the phatic function is very close to, or at least is a very important part of, what we have been calling the social function of language.⁶

Finally, there is what Jakobson calls the poetic* function – in a very broad sense of 'poetic', which relates it, not just to poetry, but to the artistic or creative use of language in general. The poetic function is defined in terms of its orientation towards what Jakobson calls the message, but which i^s perhaps better referred to as the message-­encoded-as-a-signal. For it is characteristic of the poetic use of language that it tends to blur, if it does not obliterate, the simple distinction of form and meaning in terms of which the structure of language is so often analysed. Many of the devices of poetry – rhythm, rhyme, assonance, alliteration, metre, chiasmus, etc. – exploit the properties of the medium* (cf. 3.3); and it is a commonplace of literary criticism that a line like Tennyson's And murmuring of innumerable bees would lose much of its meaning (in some relevant sense of 'meaning': cf. i.1) if murmuring or innumerable were replaced with word-forms that did riot manifest the same pattern of sounds, even if the forms that were substituted for them were forms of words that had the same meaning generally as `murmur' and `innumerable'. All that needs to be added to this commonplace observation is that, if Moliere's Monsieur Jourdain had spoken prose all his life without knowing it, we all go through life speaking poetry at times, whether we do so deliberately or not and whether we are conscious of doing so or not: we all exploit, to some degree, those resources of our native language which depend upon the properties of the medium in which language is manifest. In the poetic use of language signal-information and semantic information tend to be fused in a way that our simple model of communication does not allow for (cf. 2.3).

BÜhler's analysis of the functions of language and Jakobson's modifi­cation and extension of this have been briefly summarized here pri­marily because they have been very influential and a knowledge of the terms in which their analyses are formulated is often taken for granted by later writers. Another reason is that our discussion of these two approaches, brief though it has been, has shown us that there are several ways in which utterances, and the various kinds of information encoded in utterances, can be classified in terms of an analysis of the necessary, or at least typical, components of an act of communication. There is perhaps no single classificatory scheme that can be described as the only one that is correct. Furthermore, it requires little reflexion to see that the metalinguistic and the poetic function are closely connected: it is riot always possible in the everyday use of language to draw a sharp distinction between object-language and metalanguage, and it may be an important part of the poetic employment of a particular form that attention should be directed to the form itself rather than to what would normally be thought of as the meaning of the expression of which it is a form. Similarly, it is difficult to draw a sharp distinction between the metalinguistic and the phatic function, or between the phatic and the conative. If X uses the word `sesquipedalian' and Y asks him what it means, Y is clearly making use of the metalinguistic, or reflexive, function of language (cf. 1.2). But he may also be trying to prevent the breakdown of communication; and he is presumably making an appeal to the addressee.

The fact that there is perhaps no unique and obviously correct classi­ficatory scheme does not render the various schemes that have been put forward valueless. Many of the terms that have been introduced here for different kinds of utterances, and for different kinds of information encoded in the same utterance, will be utilized in later chapters. Mean­while, however, we will operate with the global tripartite distinction of expressive, social and descriptive information; and we will use the term 'interpersonal' (as far as language is concerned) to subsume both `expressive' and 'social'.

Granted that language has both a descriptive and an interpersonal function, various questions arise. Do any semiotic systems other than languages (or systems derived from or parasitic upon languages) have the one function to the exclusion of the other? This is a question that will be taken up in the following chapter. A further, and more complex, question has to do with the way in which the descriptive and inter­personal functions of language are interrelated. Utterances can be classi­fied, partly in terms of the grammatical structure of the sentences uttered in making them, as statements, questions, commands, requests, wishes, exclamations, and so on. Of these, it is only statements that can ever be said to describe states-of-affairs. It does not make sense to enquire whether a question, command, wish or exclamation, is true or false; it does not therefore make sense to ask what proposition is ex­plicitly asserted by utterances other than statements. However, as we shall see later, we can enquire what propositions are implied or pre­supposed by certain utterances other than statements (and also, what propositions are implied or presupposed, in addition to those that are explicitly asserted, by certain statements). We can also ask whether there is some notion analogous to the logical notion of truth in terms of which questions, commands, etc., are validated or judged successful. Intuitively, it seems that the refusal to obey a command is analogous to the denial of an assertion. These topics will be dealt with in a later chapter (cf. 16.2). It will be obvious even now that questioning and answering, as well as the issuing of commands, necessarily involves the social function of language; and that any discussion of such utterances as wishes and exclamations must appeal to the expressive function.

Here it may also be mentioned, in anticipation of what will be dis­cussed in greater detail later, that some of the most interesting of recent work in semantics draws its inspiration from Austin's (1962) thesis that to make a statement descriptive of some state-of-affairs is to engage in a particular kind of social activity regulated by conventions similar to, and in part identical with, those which regulate such other acts as making promises, asking questions or issuing commands. If we accept this point of view, however narrowly we circumscribe the descriptive function of language, we must grant that it depends upon, and to that extent is less basic than, the social and expressive function; and this is the view taken by perhaps the majority of linguists, anthropologists and social psychologists who have been concerned with semantics. It can be argued (and this point will in fact be made in the following chapter) that the descriptive function of language is more distinctive of natural languages than is the interpersonal function, which is shared by other human and non-human signalling systems. But this does not mean that the descriptive function is more basic than the other functions. This point must be borne in mind when we come to deal with logical semantics (in chapter 6).