A man is driving through a part of the country he doesn't know, and he gets lost in what looks to him like the middle of nowhere, completely deserted. Finally, he sees an old man working in a field, and he stops the car and calls out to him, 'Excuse me, how do I get from here to ...?' (the town depends on which country you hear the story in). The old man thinks for a while, and then he says, 'Well now, if I were you I wouldn't start from here'.

I am deliberately starting this book from the wrong place for where I want to get to. This is not being perverse: it is that, in recent decades, there has built up an immensely influential view of what the study of language should involve which insists that there is only one proper place to start - from a view of language as an abstract set of generalised rules detached from any particular context of use. It would be possible to ignore this view and simply start with the approach that I will be setting out in the book - based on a view of how language functions as a system of human communication. However, a comparison of different possible approaches will help us to understand better not only the destinations that each approach allows us to head for but also the reasons why we might choose one of the approaches in preference to another. Therefore, in this chapter I will briefly outline the approach that has been dominant, showing why it has been so attractive but also showing why an increasing number of linguists feel that it does not make it easy for us to talk about many of the central features of language. I will then go on to introduce an alternative approach which focuses on those features, and which offers a more appropriate place to start from if we are interested in language in use.

We can begin by looking informally at a bit of language, selected more or less at random. This comes from an advertisement on nursing (we will look at the complete text in Chapter 4). Before reading on, can you decide what aspects of the sentence you might want to consider in providing a linguistic description of it?

Of course, you're unlikely to be attracted to nursing because of the money.

When I have asked students to do this task, they mostly focus on issues such as who exactly 'you' is (since the writer is not addressing anyone face to face), and why the writer assumes this about 'you' so confidently ('Of course'). Some pick up on 'you're unlikely', which softens the possible arrogance of the writer telling 'you' about 'your' own feelings; others comment on the implication that 'you' are likely to be attracted to nursing for other reasons apart from money; and a few wonder why the writer decided not to say 'nursing is unlikely to attract you'. What all these points have in common is that they are concerned with the function of the sentence, what the writer's purpose is in writing the sentence - in other words, with the meaning. Underlying the points, though not usually made explicit, is also the idea of choice: that there are potentially identifiable reasons why the writer is expressing the message in this particular way rather than in other possible ways.

Other students (often those who have learnt English as a foreign language and therefore have more background in grammatical analysis) take a different tack. They label the parts of the sentence using terms like Subject and Verb, or non-finite verb and prepositional phrase. They comment on the fact that 'to be attracted' is a passive form, and that the understood Subject is 'you', carried over from the Subject of the preceding verb. Some mention that the structure 'be unlikely to be attracted' is not possible in some languages and that, in a way, it is an illogical structure (since it is not 'you' who are 'unlikely', but 'you being attracted to nursing'). What they are essentially focusing on is what the different parts of the sentence are and how they fit together - in other words, the form.

Both of these ways of looking at the sentence tell us something useful about it, and, in the informal descriptions given here at least, there is a good deal of potential overlap. Any full analysis of the sentence will inevitably need to take account of both the meaning and the form (and of the links between them). However, in order to make the analysis fairly rigorous rather than just an unordered list of points about the sentence, we need to decide on a reasonably systematic method; and in practice this involves choosing between form and meaning as our starting point. This may at first seem simply a difference in emphasis, but, if carried through consistently, each approach in fact ends up with a strikingly different kind of description of language.

The most fully-developed and influential version of the approach through form is that proposed by Noam Chomsky and his followers, generally known as the TG (Transformational-Generative) approach. Chomsky insisted that linguistics should go beyond merely describing syntactic structures, and aim to explain why language is structured in the way it is - which includes explaining why other kinds of structures are not found. He argued that, to do this adequately, it was essential to make language description absolutely explicit. Although the aim of TG is not to produce a computer program which can generate language, it is computers which provide the driving metaphor behind the approach. A computer is wonderfully literal: it cannot interpret what you mean, and will do exactly - and only - what you tell it to do. Therefore instructions to the computer have to be explicit and unambiguous: this includes giving them in exactly the right order, so that each step in an operation has the required input from preceding steps, and formulating them so as to avoid triggering any unwanted operations by mistake. TG sets out to provide rules of this kind for the formation of grammatically correct sentences. (Note that the following outline describes TG in its early, less complex, form. The theory has changed fairly radically since the 1960s, but the basic focus has remained essentially the same.)

In setting up its rules, TG starts from another deceptively simple insight: that every verb has a Subject, and that understanding a sentence means above all identifying the Subject for each verb. In English, Subjects normally appear in front of the verb, so it might be thought that identifying them would be too easy to be interesting. However, there are many cases where the Subject does not appear in the 'right' position - or does not appear at all (we have already seen that the Subject of 'to be attracted' has to be carried over from a different verb). We are so skilled at understanding who does what in a sentence that we typically do not even notice that in such cases we have to interpret something that is not explicitly said. One famous example used by Chomsky was the pair of sentences:

John is eager to please. John is easy to please.

These appear, on the surface, to have the same structure; but in fact we understand that in the first case it is John who does the pleasing (i.e. is the understood Subject of 'to please'), while in the second it is an unnamed person or thing (and 'John' is understood as the Object of 'to please'). This game of 'hunt the Subject' can become even more complex and exciting - the kind of (invented) sentence that makes TG linguists salivate with delight is the following:

Which burglar did John say Mary thought had shot himself?

Here, we understand that the Subject of 'had shot' is 'which burglar' - even though there are two other clauses ('did John say' and 'Mary thought') between the verb and its Subject. Adding to the excitement is the fact that we also understand that 'himself' refers to the burglar, even though 'John' is closer in the sentence - whereas, if we replaced it with 'him', it could not refer to the burglar.

But how do we understand all this? And how can the linguist show, in an explicit way, what it is that we actually understand? One problem is that, in order to label part of the sentence as 'Subject', we have first had to identify that part as a separate constituent (a noun phrase) - in other words, we have actually jumped over the initial stage. That means that our description is not in fact fully explicit. We need to work with labels that tell us what each constituent is in itself, not what it does in the sentence. At the same time, we also need to show where each constituent fits in the basic structure. Chomsky's famous first rule captures this:

S ® NP VP

This is a non-verbal (and thus less ambiguous) way of saying that every 'sentence' in a language consists of a noun phrase followed by a verb phrase - if it does not show these features it is not a grammatically acceptable 'sentence' (VP here includes everything in the sentence apart from the first NP). Translated into functional terms, it means in effect that every verb must have a Subject. Using this rule, the underlying meanings of our 'burglar' example can be set out as follows (the inverted commas round the words signal that we are dealing with the abstract concepts that the words refer to rather than the words themselves):

S_{1 ®} NP VP

['John'] ['did say' (something)]

S_{2 ®} NP VP

['Mary'] ['thought' (something)]

S_{3 ®} NP VP

['which burglar'] ['had shot himself']

Note that this analysis also begins to elucidate why 'himself' refers to the burglar.

If we accept that this captures what we understand in understanding the sentence, our task now is to show how S_1-3 combine into the sentence as we actually see it. Although this is immensely complex in practice, it is simple in theory: it turns out that we can identify not only a finite set of explicit rules governing the possible combinations (the complexity comes especially from the interaction between the rules), but, more crucially, an even more restricted set of underlying regularities in the type of rules that are possible. Just to give you a taste of how the analysis proceeds, one further rule (in a simplified form) is:

VP ® V S

This means that one permitted kind of verb phrase can itself include not only a verb (V) but another S (this is technically known as recursion). This may be easier to grasp if we revise the analysis of our example to take this new rule into account:

S₁ ® NP VP ® [V S]

['John'] [['did say'] [‘S₂’]]

S₂ ® NP VP ® [V S]

['Mary'] [['thought'] [‘S₃’]]

S₃ ® NP VP

['which burglar'] ['had shot himself']

I have concentrated so far on the Subject, but exactly the same kind of analysis can be done for Objects and other clause constituents that appear in the 'wrong' place or which govern the form and interpretation of other constituents (as 'which burglar' governs the interpretation of 'himself'). What are the S_1-3 underlying this version of the example?

Which burglar did John say Mary told him she had shot?

It is perhaps surprising that, using such apparently marginal examples, the approach should have thrown so much light on how sentences are structured; and yet the insights gained have been extensive and in some ways revolutionary. For our present purposes, however, it is less important to look at these discoveries in any detail than to consider where the approach leads us. The first thing to say is that TG is only interested in what we can call 'propositional meaning' - the ‘content’ of the sentence. The following two sentences have exactly the propositional content and therefore the same analysis in terms of Ss:

John said something. Did John say something?

S₁ ® NP VP

['John'] [‘said something’]

The difference in surface form ('John said' vs 'Did John say') results from rules which allow the auxiliary 'did' to appear in front of the NP as the Ss transform into the sentence. On the other hand, the fact that a statement and a question serve entirely different functions in communication is regarded as irrelevant in the grammatical analysis - it is taken into account in a different part of the linguistic description (though there has so far been little interest in developing that part within TG). Chomsky made a principled decision to exclude how we use sentences in communication (e.g. as statements or questions): the model is not designed to show, for example, that one sentence functions as the answer to a preceding question. The aim is to discover the rules which govern how constituents can be put together to form grammatically correct sentences, and to formulate these rules in as general a way as possible (ideally, so that they apply to all human language rather than just individual languages); therefore each sentence is analysed in complete isolation, both from other sentences and from the situations in which it might be used. This limitation is self-imposed because TG linguists feel that it is only worth describing those aspects of language which can be described 'scientifically' (i.e. with absolute explicitness). The ways in which language is used are, unfortunately, too messy and are therefore ignored, at least until someone can find a way of describing them according to scientific general laws.

But if the path towards an examination of use is blocked off, where else can we go? The answer is inwards, into the mind. The fact that we as language users can make the complex transitions from Ss to sentences and vice versa - i.e. can identify the separate constituents in the sentence and assign them to their place in the structure of the appropriate S - tells us a great deal about how our minds must work. At the same time, the fact that we do not need to be explicitly taught how to do this means that we must in some way be born with the required mental capacities. Thus a rigorously formal approach to the description of language leads us towards psychology and genetics. Clearly, these are fascinating and worthwhile areas, but they do involve giving up any idea of looking at language in use.

It may well be possible, and intellectually productive, to view language, as the TG approach does, as a system of abstract rules which are applied in order to end up with a grammatically correct sentence; but there are doubts about whether this view captures to any useful extent the psychological processes involved when users actually produce or understand language. More importantly, there is little doubt that it does not reflect how the users themselves view language. They respond above all to the meanings that are expressed and the ways in which those meanings are expressed. For the user, despite the clear similarities in terms of propositional content, the following sentences have very different meanings because they are designed to elicit different responses from the addressee (acknowledging, agreeing or informing):

Similarly, there are important differences between the following sentences because of the speaker's choice of a formal or intimate wording:

The syntactic underpinning in the examples above is of course important in expressing the different meanings, but only as a tool which enables the primary function of language - communicating meanings in particular contexts - to be carried out. As always, the exact nature of the tool used depends on the task in hand. In linguistic terms, we can express this as the assumption that, since language has evolved for the function of communication, this must have a direct and controlling effect on its design features - in other words, the form of language can be substantially explained by examining its functions. Of course, we need to take into account the constraints of the 'raw materials': the pre-determined (genetic) characteristics of the human mind which allow or encourage certain kinds of language forms, and disallow or discourage other kinds. TG provides a possible way of investigating those characteristics. But they clearly represent only half the story: we still need to examine the formative influences of the uses to which language is put. (We can see the contrast between the two approaches as a reflection of the old dichotomy of nature vs nurture - and, as always, the answer is most likely to lie in a combination of both.)

What happens, then, if we take the other tack and (like language users) start from meaning? The meanings that we may want to express, or the uses to which we may want to put language, are clearly 'messy': they appear so varied and so dependent on the infinite range of different contexts that it is difficult at first to see how we might impose some order on them. However, if we look at the grammatical options open to us, we can in fact relate those options fairly systematically to different kinds of meanings. Let us take just two examples of areas that we will examine in more detail later. We can relate the presence of modal verbs to the expression of the speaker's personal view of events and states in the external world. In the following example, the speaker evaluates 'this seeming strange at first' as only potentially valid ('may'):

And we can relate the ordering of parts of the clause to the speaker's desire to signal how this message fits in with the preceding message(s). Compare what comes first in the second sentences in each of these pairs (and think about why the order is different, and whether the second sentences could be swapped):

It may seem odd to say that ordering in the clause has 'meaning'; but it is only odd if we restrict meaning to 'propositional meaning' - which, as I have suggested, is a narrower definition than we want. If we take meaning as being the sum of what the speaker wants the hearer to understand - in other words, if we equate the meaning of a sentence with its function - then understanding how the present message fits in its context is clearly part of the meaning, just as the difference between a statement and a question is part of the meaning.

In describing the various kinds of meanings in this fairly general way (e.g. 'signalling how this message fits in with the preceding message(s)'), we are already beginning to set up categories of functions that we perform through language; and we can then go back to texts to see if there are other grammatical features which seem to be performing the same kind of function. But we are still in danger of ending up with a fairly random-seeming list of functions. Is there any way of arriving at an even more generalised grouping of meaning types, so that we can start to explain why we find the particular kinds of functions that we do? For this, we need to step back and, rather than looking at language structures, think about what we do with language. In the broadest terms, we use language to talk about things and events ('It's raining') and to get things done ('Sit down'). As we shall see, these are not mutually exclusive (the command 'Sit down' involves reference to the particular event of sitting rather than any other; and telling someone that it’s raining has the effect of changing their knowledge): indeed, the basic principle is that every time we use language we are doing both simultaneously. We will also see that we need to add a third major function, a kind of language-internal 'service function'; but, having simply established here that it is possible to identify a very small number of broad functions, we can leave further specification until, in Chapter 3, we start exploring how these major functions can be used to illuminate and explain the choices that are available in language.

I have at several points used the term 'choice' in discussing meanings. If we want to examine what a piece of language is intended to do (i.e. its function), we cannot avoid thinking in terms of choice. Clearly, speakers do not go round producing de-contextualised grammatically correct sentences: they have reasons for saying something, and for saying it in the way they do. To take a simple example, if you want to find out some information you are most likely to ask a question rather than make a statement; and, at a more detailed level, you are more likely to use an informal wording if you are talking to a friend rather than a formal one:

But note that, in describing the example in this way, we have in fact set up two sets of context-dependent choices: question vs statement, informal vs formal. If you have reasons for doing (saying) one thing, the implication is that you could have done (said) something else if the reasons (the context) had been different.

Functional grammar sets out to investigate what the range of relevant choices are, both in the kinds of meanings that we might want to express (or functions that we might want to perform) and in the kinds of wordings that we can use to express these meanings; and to match these two sets of choices. In order to identify meaning choices, we have to look outwards at the context: what, in the kind of society we live in, do we typically need or want to say? What are the contextual factors which make one set of meanings more appropriate or likely to be expressed than another? But at the same time we need to identify the linguistic options (i.e. the lexical and structural possibilities that the language system offers for use), and to explore the meanings that each option expresses. These are complementary perspectives on the same phenomenon, one, as it were, from the bottom up - from wording to context - and the other from the top down - from context to wording. Looking from the bottom up, the use of the 'the hell' in the question above means - i.e. has the function of expressing - informality (amongst other things): in other words, one thing that our grammatical description must account for is the lexical and structural means by which different degrees of formality are expressed. Looking from the top down, the fact that the speaker is talking to a friend makes appropriate the use of informal wordings: in other words, we need a description of the social context which includes degrees of familiarity between people interacting with each other as a relevant factor influencing language choices.

Note that the use of the term ‘choice’ does not necessarily imply a conscious process of selection by the speaker: what we aim to uncover through a functional analysis are the reasons why the speaker produces a particular wording rather than any other in a particular context (in some ways, it would almost be as true to talk of the wording choosing the speaker). In writing this book, there are certain choices that I am very aware of making - e.g. the decision to sometimes address 'you' directly rather than always avoiding this by using passives, etc. (both decisions are possible in a textbook, and I have consciously set out to sound 'interactive' in parts). But there are many 'choices' that I am constrained to make by the kind of context in which I am using language: for example, it is very unlikely that I will use the structures associated with swearing - except perhaps in quotes. It is only in by consciously trying to imagine the ‘wrong’ choices that such choices even present themselves as possible: but the choice not to swear has nevertheless been made (or, rather, made for me). These are deliberately crude examples; but the principle applies in every detail of the wordings that I 'choose'.

One important implication of the functional view of language is that context and language are interdependent. This might seem too strong a way of putting it: it looks as though language could be seen as dependent on context - because the speaker is in an informal context, he is likely to use informal language - but not necessarily the other way round. But note that the use of informal language also contributes to creating the informal context: if the example question above were uttered by, say, a TV journalist interviewing a government minister, it would sound odd precisely because it would conjure up an inappropriate context. We can use the term 'construe' to talk about this kind of reflexivity. The wording 'the hell' construes informality: that is, it simultaneously reflects and constructs an informal context. Similarly, 'the glass broke' construes a slightly different view of events from 'I broke the glass' (hinging on the question of agency - see Chapter 5).

At a broader level, our experiences in the world clearly influence what we normally talk about and the way we talk about it: for example, we are not impassive observers of the world, and we typically evaluate events that we talk about (and the lexico-grammatical resources of the language therefore offer ways of performing this evaluation). At the same time, the way we normally talk about these experiences (and the way we hear other people talk about them) influences the way we see them: for example, we generally accept without conscious query the fact that advertisers talk about their products as solutions to our problems (as opposed to talking about our willingness to pay for the products as the solution to the advertisers’ problems, which is at least equally valid a view).

By formulating our approach to linguistic description in the kind of terms used above - choices amongst relevant options in context - we are deliberately opening up the path towards grammatically-based text analysis (where 'text' means any instance of language in use): at each stage, we can ask why the writer or speaker is expressing this particular meaning in this particular way at this particular point. I mentioned earlier that TG takes linguistics towards biology; functional grammar takes it towards sociology: the systematic study of relevant features in the culture and society which form the context in which language is used. Both approaches, through form and meaning, ask essentially the same question about language: how can we explain why language has the main features that it does? But whereas the form-based approach finds the answer in the way our brains are structured, the meaning-based approach finds it in the way our social context is structured. (Of course, the different answers depend very largely on the fact that each approach takes a different view of the 'main features' that need to be explained.) Although our focus in the rest of the book will be on choices within the grammatical systems, we shall be regularly looking outwards towards the wider contextual factors that are construed by these choices.

If it is true that language and context are inextricably linked, any stretch of language should, to a greater or lesser extent, come trailing clouds of context with it: we should be able to deduce a great deal about the context in which the language was produced, the purpose for which it was produced, and the reasons why it was expressed in the way it was. We can check this in a preliminary way by looking at a simple example. I have deliberately chosen one which conjures up a very clear context; but can you go from that to explain as much as possible about the language choices in terms of who the interactants are and what the speaker's purposes are? My commentary follows, but you will find it useful to try your own analysis before reading it.

The context is obviously a fairy story, probably told by an adult to a young child. This is most clearly signalled by 'once upon a time', which is used almost only in fairy stories (so much so that, if used in another context, it imports the very specific fairy tale context, however fleetingly). The individual story teller hardly needs to 'choose' this opening: he knows that this is how fairy stories start. However, it is worth considering why this type of narrative should have such an immediately recognizable opening. One important factor is the addressee: a relatively unsophisticated language user, for whom very clear signals of purpose are necessary. The conventional opening signals something like: 'I'm not going to tell you to do anything; I'm not going to scold you; all you need to do is to sit back and enjoy the story that is coming up.' In addition, although the expression belongs grammatically to the group of adverbials which specify time ('Once', 'Yesterday', 'Three years ago', etc.), it clearly does not in fact specify a real time. It thus signals that the narrative is a fictional one rather than, say, an account of what the teller did last year.

The clause structure ('there was ...') is an existential one (see 5.2.4). It introduces one of the main characters without saying that the bear was involved in any particular action - the action will presumably start in the next clause. Thus it stages the information, building up the story in increments that are manageable to the inexperienced language processor to whom the story is addressed. What we are told about the bear apart from its existence is that it is big and bad. The alliteration is obviously striking here: it appeals to children's pleasure in incidental patternings of sound, rather like wordplay at a more sophisticated level (in adult texts we are more likely to rewrite something to remove alliteration if it happens to occur). At the same time, it serves to reinforce the non-real, poetic nature of the story, perhaps reducing the potential frighteningness of the animal (cf. the effect of 'an enormous, savage bear'). It is also worth commenting on the fact that the speaker evaluates the character as he introduces it. In sophisticated narratives such as novels, we expect to be skilfully guided towards an evaluation of characters without having the author's evaluation thrust upon us; but here the child is told in advance that the bear is bad. The adult takes on the responsibility of setting out the required set of values for the child, partly no doubt as a reflection of his assessment of the child's restricted ability to do the necessary inferencing for himself (we will see a similar lack of confidence in young language users' expertise in processing language in 6.12). In addition, the evaluation opens up generic expectations of how the story will unfold: the bear will somehow cause problems for the good characters who will be along in a moment, but will in the end be defeated. Children learn very rapidly to recognize conventional story lines, as long as the signals are clear enough.

These are only some of the main points that can be made about how this piece of language works in its context - I have not, for example, touched on the broader issues of the role of story-telling in the socialisation of children. I have deliberately outlined the points as informally as I can; but what I hope the discussion shows is the kind of thing that we want to be able to discuss in a more formalised way. The grammatical system that we set up should provide categories that relate to the communicative purposes and choices that we have identified. In the rest of the book, I shall be setting out a functional approach based closely on Michael Halliday's work, which allows us to do this in a systematic and satisfying way.

• Refer to Exercise 1.1