This text might strike contemporary readers as a little unusual. References to lines from old war songs and needle-pointed mottoes hanging on the wall belong to a bygone era and anecdotes relying on them are likely to be somewhat vague. However, what is even more striking about the text is exactly what typical native speakers of English are likely to find unremarkable, namely the numerous, very different interpretations assigned to the single word, over. In this short text, over has four distinct interpretations – overA can be paraphrased by ‘again’, overB by ‘above’, overC by ‘finished’ and overD by ‘in some other place’. For us, the fundamental question that texts such as the one above raise is whether the various meanings regularly associated with a single word are simply accidental (the fact that over has four very different meanings might, after all, be a bizarre accident), or systematically related.

In the previous two chapters we have argued that the semantics of over, under, above and below differentially reference the vertical axis (or what Langacker, 1987, terms ‘oriented space’). As such, these spatial particles serve to partition conceptual space in terms of verticality. In this chapter, we turn to a subset of spatial particles whose semantics crucially involves an additional element, the orientation of either the LM or the TR. We will deal with four forms which relate to an oriented TR: up, down, to and for, and four which relate to an oriented LM: in front of, before, behind and after.

Orientation is a designation of a conceptual front/back, top/bottom, or lateral partitioning of either the TR or the LM. This conceptual partitioning has its basis in a perceived asymmetry of the partitioned entity. Such asymmetry can arise as a result of the way the entity typically stands or sits, its shape (e.g., if an object has a pointed end, it is typically understood as its ‘front’) (Talmy, 2000), its perceived parts (Talmy, 2000), its perceived ‘directedness’ (Talmy, 2000: 199), its movement (Fillmore, 1971), or the way it is used by humans (e.g., the ‘front’ of a building is the side which is accessible for general entry) (Langacker, 1987), as well as its perceived resemblance to human beings or animals (Fillmore, 1971; Herskovits, 1986; Svorou, 1994). Clearly, the attributes which give rise to an object being conceptualized as asymmetric involve how humans both perceive and interact with the object.

In this chapter we will consider three additional spatial particles that involve verticality – above, under and below. The analysis of spatial scenes related to these forms, in conjunction with over, will demonstrate that this subset of spatial particles acts as a particular lexical contrast set. Hence, in semantic terms, the four English lexemes over, above, under and below represent a systematic means of dividing up the vertical axis into four distinct spatial locations. Moreover, the analysis for each of these forms provides further evidence for our representation of proto-scenes for English spatial particles as most appropriately being characterized as involving both a functional component and a conceptual-spatial configuration between a TR and a LM. Our goal, then, in this chapter is to show that working out an appropriate analysis of the distinction between the proto-scenes associated with over versus above and under versus below underscores the efficacy of the model developed in the first four chapters.

Contrast sets

Linguists have long recognized that languages readily add new lexical items to the existing inventory of certain classes of words (open-class words), while being more resistant to adding a new lexical item to the existing set of other classes of words (closed-class words). We argued in chapter 3 that spatial particles represent a closed class of lexemes. They have this status because, in their spatial-physical uses, spatial particles operate within a stable, self-contained conceptual domain (Talmy, 2000).

The focus of this investigation has been to account for the wide array of meanings associated with English spatial particles. In contrast to most previous investigations of spatial particles, we have been particularly concerned with the many non-spatial meanings associated with theseforms. The hypothesis that English spatial particles are polysemous (as opposed to homosemous or monosemous) and therefore that the many-to-one mappings between meanings and form are largely systematic is central to our analysis. In the course of this exploration, we have developed a model which we term principled polysemy.

The principled polysemy model takes a number of key assumptions about language as foundational. Many of these assumptions are more general in nature and must be recognized by any serious approach to language; others are informed by the perspective of cognitive linguistics. The approach rests fundamentally on the assumption that the primary reason humans use language is to communicate with one another and that this motivation constrains semantic extensions in non-trivial ways. Moreover, when lexical items are uttered (or written), they do not occur in isolation. Rather, they are embedded in longer segments of language, that is, naturally occurring language is always contextualized. Language itself radically underdetermines the rich interpretations regularly assigned to naturally occurring utterances (e.g., Green, 1989; Grice, 1975; Gumperz, 1982).

Linguists, psychologists and philosophers have long observed the importance of space and spatial experience for both language and thought. In this book, we examine the nature of human spatio-physical experience and how human conceptualization of spatial relations is reflected in the English language. In particular, we are interested in how spatial concepts are systematically extended to provide a wide array of non-spatial meanings. We do so through a study of English spatial particles, an important subset of which are prepositions.

The central notion we explore is that of a spatial scene, a conceptualized relation grounded in spatial interaction and experience, involving entities that are related in a particular spatio-configurational way. For instance, in a spatial scene described by: The cup is on the table, the cup is in contact with the upper side of the table. A distinct spatio-configuration is described by the following: The coffee is in the cup. This scene involves the coffee being located inside (as opposed to outside) the cup. However, spatial scenes do not involve only spatio-physical relations or configurations. It turns out that particular spatial relations have non-trivial consequences that are meaningful to humans. The spatial scene involving on also involves a support function between the table and the cup: unless enough of the cup's base is situated on the table, the cup will fall and smash on the floor.

In chapter 1 we saw that a spatial particle such as over has a number of distinct meanings or senses associated with it. We reviewed two positions that have been proposed to account for the relationship between such distinct meanings. The first, homonymy, holds that the meanings associated with a particular form are simply stored in the mental lexicon, as unique entries. While this position may sometimes be justified based on synchronic evidence – for example, the river bank versus the bank of England – in the case of a spatial particle, such as over, it demonstrably is not. For instance, Sandra and Rice (1995), and Rice et al. (1999) have found, based on a series of psycholinguistic experiments, that native speakers of both English and Dutch tend to recognize relationships between distinct meanings associated with the same spatial particle. The homonymy position ignores any shared similarities and interrelationships between the meanings of a form such as over, and would predict that native speakers would fail to find a relationship between the various meanings in a consistent inter-speaker way.

The second position, monosemy, assumes one highly abstract meaning, from which all other meanings are simply contextual variants. However, as we have already observed, some meanings associated with a particular form are context independent. That is, the range of meanings associated with an individual form such as over is so diverse that they cannot be straightforwardly inferred from a single abstract meaning as it occurs in context.

For a model of polysemy, it is not sufficient simply to describe the senses that exist in the network. We must also provide an explanation of how they came to be there and how they are distinct from locally constructed (i.e., on-line) meanings, which are not instantiated in memory. Our account will show how the model outlined in the previous chapter will achieve each of these objectives. For instance, we will employ our methodology for identifying distinct senses. We will illustrate how the notion of pragmatic strengthening serves to conventionalize implicatures that derive from language use and the nature of spatio-physical experience. We will also show how the inferencing strategies we discussed in chapter 3 serve to provide on-line interpretations, created for the purpose of local understanding. However, before proceeding with these issues, we must first identify the proto-scene for over, from which the other senses are derived diachronically in a principled manner. Accordingly, our discussion of over will serve as our primary illustration of how the model adduced in chapter 3 applies. Consequently, this chapter will be somewhat more detailed than most ensuing chapters.

The proto-scene for over

In chapter 3, we provided five linguistic criteria for adducing the primary sense associated with a semantic network. To illustrate the application of each of those criteria, we used over as our primary example.

A number of spatial particles appear to mediate relations in which the dimensions of the LM are essentially irrelevant. As we saw in our discussions of over, the extendedness or dimensionality of the LM appears to have no particular bearing on the relationship mediated by over. In contrast, in this chapter we will explore a set of spatial particles, in, into, out, out of and through, which are sensitive to certain dimensions of the LM, namely the dimensions which collectively give rise to the notion of boundedness.

We define a bounded LM as one that possesses an interior, a boundary and an exterior. Canonically, we think of bounded LMs as three-dimensional objects, such as boxes or rooms. Entities that are typically thought of as having a one-dimensional or two-dimensional structure are not typically conceptualized as a bounded LM; for instance, they are not typically thought of as possessing an interior. However, as we have seen in our previous discussions, humans have the capacity for construing spatial scenes from a variety of perspectives; this ability appears to extend to how the dimensionality of any given entity is construed for the purposes at hand. The linguistic evidence shows that the conceptualization of a particular LM as bounded is determined not in absolute terms by its geometry (although clearly this does play some part), but rather by virtue of the way in which humans experience and interact with the LM in question.

In the previous chapter we argued that our knowledge of the world is indirect because it is constrained by how we experience it. This follows as our experience of the world is always mediated via our uniquely human perceptual system, physiology and neural architecture. A hummingbird's understanding of gravity as a force which can be overcome for extended periods of time, albeit with effort, would be significantly different from a human's. Thus, a hummingbird no doubt experiences and represents the same world to itself in quite different ways from how human beings do; both versions of the world, while presumably very different, are equally ‘real’. As pointed out by the philosopher Hilary Putnam (1981), to claim that we can have direct access to and conscious knowledge of an objective reality (i.e., an objective god's-eye view of the world) is wrongheaded.

Nonetheless, we are not claiming that there is not a world ‘out there’ nor that our experience of it is unimportant. To say that our experience with and perceptions of the world are mediated by our conceptual system, and are fundamentally conceptual in nature, is not to say that the real world and its properties do not largely constitute the nature of our experience. On the contrary, it is the real world which provides the raw substrate for our sensory perceptions and the conceptualizations which arise from them. Accordingly, the spatio-physical properties of the world of humanly perceived experience are fundamental to human cognition.

Research on discourse markers has been dominated by taxonomy and classification. As I have shown in the final chapter of this book, this is matched by a corresponding focus on the classification of coherence or discourse relations. In neither case has this concern with classification yielded an account of discourse markers that is universally accepted by those who are working with them. Indeed, as I remarked in the introduction to this book, it is not even agreed what the set of discourse markers for any given language is. It seems that in general, classifications are made at a descriptive level: the aim is to describe the role that these expressions play either in constructed examples of acceptable uses (see Halliday and Hasan 1976) or in naturally occurring discourse (see Schiffrin 1987, 1994). In these approaches, the evidence for any generalization made is always positive. Thus there is not a single example of an unacceptable use of a conjunctive cohesive device in Halliday and Hasan's book, and, since Schiffrin's examples are all examples of actual uses, there are no unacceptable uses of the expressions she classifies as discourse markers either. However, as is so often the case in linguistics, we often learn more about the meanings of these expressions from the fact that they cannot occur in a particular context than the fact that they can occur in another.

The location of discourse on the theoretical map

As I have said in my introduction, this book is about the two properties which have brought discourse markers into the forefront of pragmatics research. On the one hand, expressions classified as discourse markers are said to be non-truth conditional, which means that they play a role in discussions of the non-unitary nature of linguistic meaning and the relationship between semantics and pragmatics. On the other, they are generally claimed to signal connections in discourse, which means that they play a role in the discussion of how we account for the unity of discourse. So far the emphasis has been on the first property. However, the section on well (4.4) should have reminded us that I have yet to discuss the second.

In fact, it seems that discussions of the non-truth conditionality of discourse markers rarely make reference to their role in discourse, while discussions of role in discourse rarely include an investigation of their non-truth conditionality. Indeed, it seems that some writers whose concern is with their analysis within a discourse perspective classify expressions as discourse markers in a way which cuts across the distinctions that we have been discussing in the previous chapter. For example Knott and Dale (1994) include within their list of expressions they call cue markers both truth conditional expressions (for example, because, and, then) and expressions which are regarded as non-truth conditional (for example, but, furthermore, hence).

Introduction

The expressions which occupy centre stage throughout this book have played a number of different roles. In this part of the book, we shall examine the role they have played in the move towards a non-unitary theory of meaning. As we shall see, this move is itself not always a move towards the same sort of distinction, and the purpose of this and the following two chapters is to tease these distinctions apart, and to argue for a distinction between two kinds of meaning that is grounded in human cognition.

For many writers, this distinction is the distinction between semantics and pragmatics, and the significance of the expressions which I am calling discourse connectives lies in the role they have played in arguments for the existence of pragmatic meaning. Chapter 2 will examine the attempts that have been made to develop the notion of pragmatic meaning within the framework of speech act theory. This chapter focusses on the view of semantics which underlies the argument that expressions such as but and well have pragmatic meaning rather than semantic meaning.

This view is implicit in Gazdar's (1979) definition of pragmatics:

According to this view, discourse connectives such as but must have pragmatic meaning rather than semantic meaning because they do not contribute to the truth conditional content of the utterances that contain them. And indeed, this is usually believed to be the case.

Constraints on relevance: new questions

According to the arguments of the previous chapter, the distinction between conceptual and procedural encoding cross-cuts the speech act theoretic distinction between describing and indicating: not all of the expressions defined within the speech act theoretic framework as indicators can be analysed as encoding procedures, and not all expressions which encode procedures are analysed within the speech act theoretic framework as indicators. In view of the fact that the two distinctions are not co-extensive, the decision to take the relevance theoretic distinction as the fundamental one in a theory of linguistic semantics could be construed as a recommendation to simply forget the speech act theoretic distinction, and in particular, as a recommendation to drop the notion of indicating or signalling or pointing altogether. After all, it seems that we now have something less metaphorical to work with, namely, coded means for constraining the inferential tasks involved in utterance interpretation. However, in this section I shall show that we still have much to learn about what it means for an expression to encode a procedure. Moreover, it seems that it may be illuminating to compare such expressions with natural or non-coded means for pointing to something.