DRAFT - Comments warmly welcomed
ICLC 1999, Stockholm

Ideological ground and relevant interpretation in a cognitive semantics
Peter Grundy University of Durham and Yan Jiang The Hong Kong Polytechnic University

"The perceptual mechanisms - and perceptual salience itself - are relevance oriented" (Sperber & Wilson, 1986/95: 152)


1 Introduction

In this paper, we show how the figure / ground gestalt enables discourse to be interpreted in relation to the background ideological context in which it occurs. We argue that the relation of linguistic figure to contextual ground is indicated by discourse markers which function as viewpoint shifters and space builders enabling contextual ground to be represented in the mental space model of cognitive semantics proposed by Fauconnier. Although it is nowhere explicitly stated in Fauconnier's work, we suggest that his proposals also reflect the figure / ground gestalt in the characterization of the different cognitive functions associated with Focus and Viewpoint spaces.
The data on which we draw are taken from President Clinton's national television address of 18 August 1998 following his testimony to the grand jury in the Monica Lewinsky affair. (See appendix for the full text.) Our analysis provides cognitive plausibility for the well motivated distinction made in relevance theory between conceptual and procedural meaning (Blakemore 1987), and at the same time shows how both linguistic expressions and non-linguistic pragmatic conditions are represented in a single semantics.
Broadly, the intuition is that there is little (if anything) procedural about the utterance

(1) Presidents have private lives.

But when Clinton says

(1') Even presidents have private lives

the procedural use of even constrains the interpretation of Presidents have private lives by restricting the set of contexts which are called up. It is in relation to these contexts that Presidents have private lives is both one amongst a set of possible variables (i.e. many other things might have been said) and a salient figure. In this way, procedural meaning relates a new notion, a variable figure, to an established context, the invariant ground (Talmy, 1978). This ground may well be, and perhaps usually is, in part ideological. Given the obvious ideological context in which Clinton's conduct appears unacceptable, it is hardly surprising that his national television address exhibits a very wide range of metalinguistic and metapragmatic procedural encodings.
The principal focus of this paper is the implications for the nature of a cognitive semantics posed by attempting to model data containing a wide range of procedural forms with space shifting and space building properties. Thus, a complete semantics for the statement / utterance

(2) Indeed, I did have a relationship with Ms Lewinsky that was not appropriate. In fact, it was wrong

would need to model at least how the contexts are constructed which are oriented to by the maxim hedges indeed and in fact, by emphatic did, by the higher level metalinguistic predicates not appropriate and wrong, and by an utterance that glosses the preceding utterance. In this paper, we attempt to model the way in which such metapragmatic phenomena relate conceptual meaning to background ideological context. Readers expecting an ideological critique of a moment in US political history will, therefore, be disappointed.
From a model theoretic point of view, this paper addresses the issue of how non-linguistic elements are represented in cognitive semantics. The work of Fauconnier and others is largely, although not exclusively, concerned with showing how linguistic expressions build mental spaces and how meanings are constructed in them. In this paper, we will attempt to characterize the way in which mental spaces may, and indeed must, include non-linguistic objects which provide a ground in relation to the linguistic figures in focus. It is the fact that the prevailing ideology is relevant in interpreting the Clinton's statement rather than explicitly encoded in it that make the data discussed here especially pertinent.
In the remainder of this paper, we will first delimit the semantic / pragmatic area under consideration and review the ways in which the figure / ground gestalt has informed linguistic analysis. We then will argue that this gestalt can also be productively linked to the relevance theoretic notions of procedural and conceptual encoding and that it is implicit in proposed mental space representation types. After considering the status of cognitive constructions in the light of the arguments presented in the earlier sections, we will show how cognitive semantics allows for the construction of the ideological contexts without which the interpretation of the linguistic figure is at best problematic, and sometimes even impossible.

2 Semantics, pragmatics and cognitive linguistics

The essential facts that a semantics of language production and comprehension has to account for are that a speaker conveys Meaning X in Context Y by means of Form Z and that (all being well) their addressee/s process Form Z, supply Context Y and infer Meaning X. Because the roles of speakers and their addressees are different in that speakers start with a meaning to convey and addressees end by recovering that meaning, pragmatic accounts have typically focused either on production, as in Speech Act theory, or on understanding, as in Grice's theory of conversational implicature and Sperber & Wilson's work on relevance. In the cognitive literature, Langacker at one point suggests that conceptualizing is identified primarily with speakers and secondarily with addressees (1991: 318). As we shall see, there is a considerable advantage in a proposal which recognizes the conceptualizing role of both speakers and hearers.
Cognitive linguists typically reject the semantics / pragmatics distinction on the grounds that logical semantics presupposes a relation of language to an objective world which takes no account of the language user's conceptualization. However, the logical form of an utterance (Form Z) and the logical form of the inferred meaning (Meaning X) are often, perhaps typically, unrelated. Thus a speaker may use the form "I'm tired" to mean I want to go to bed, I want you to come to bed, I don't want to get out of bed, I want you to get out of bed and make me a drink, etc.
Whilst cognitive semantics models pragmatic phenomena such as presupposition and deixis and some cases of implicature, we suggest that the new logical forms associated with (most) implicatures are not so easily captured in this kind of cognitive construction. In any case, it isn't appropriate to do this. If the relevance theoretic notion that implicatures are deductive inferences is right, it seems reasonable to restrict the role of mental space constructions to providing all the information necessary for drawing the deductive inference. This is consistent with the notion that mental space configurations are mental models of discourse (Fauconnier, 1994: xxxix) and not that they represent the overall process of discourse understanding. The absence of any mention of pragmatics in the 'Final remarks' of Concept, Image, and Symbol and the conclusion that "The only elements ascribable to a linguistic system are semantic, phonological, and symbolic structures that occur overtly as (parts) of linguistic expressions, schematizations of such structures, and categorizing relationships" (Langacker, 1991: 343) also supports the suggestion that a cognitive semantics is limited to specifying a meaning construction1 which includes both linguistic forms and pragmatically conditioned contexts, but stops short of modelling the implicature whose propositional form is entirely new.
In this paper, we argue that Form Z is a figure and that context Y is a ground and that both exist as cognitive constructions, but that Meaning X exists, not as a conceptualization, but in rather the way that we would draw a conclusion about what to do from looking through the window in the morning and seeing the sun shine or the clouds gathering:


Cognitive construction Pragmatic inference

What is said Context of utterance What is meant by what is said
Weather observed Knowledge of weather types What to wear, etc.


Thus the purpose of cognitive construction is to resolve indeterminacy and to provide a unique characterization of utterances and their contexts: "Linguistic forms are (partial and underdetermined) instructions for constructing interconnected domains with internal structure" (Fauconnier, 1997: 35).
The distinction between Meaning X on the one hand and a cognitive construction including Form Z (= Figure) and Context Y (= Ground) on the other is reflected in current accounts of cognitive semantics. In his summary of operating principles for natural language semantics, Fauconnier (1997: 111) lists many relevant pragmatic conditions (background knowledge, knowledge of activity types, beliefs, cultural constructions, focusing devices, etc.), and in Mental Spaces (1985/1994) he discusses presupposition and scalar implicatures, but, significantly, none of these pragmatic processes results in the construction of new logical forms of the kind associated with Gricean particularized conversational implicature and with relevance oriented implicatures in the sense defined in Sperber & Wilson. In fact, meaning constructions are not either an underlying form or a linguistic representation; nor are they a representation of real or possible worlds, but rather a conceptualizer's (i.e. non-truth conditional) way of relating language and the world in which it occurs, and thus resolving the indeterminacy associated with Form Z (Fauconnier, 1997: 36).
Resolving linguistic indeterminacy involves inference. For example, Fauconnier (1985/94: 39ff) discusses the sentence The President changes every seven years, and points out that the expression The President will have different values in different places or organizations at different times. Moreover, The President could equally be a referential description referring metonymically to the President's mood, for example, or an attributive description (i.e. whoever is President). Each different determination of the expression The President is also likely to be linked to a different determination of the expression changes.
In order to illustrate the argument that pragmatic phenomena which preserve propositional form are part of mental space meaning constructions and that relevance oriented implicatures are not, we might consider the case of

(1') Even presidents have private lives.

In the Gricean account, even would be treated as a conventional implicature whose literal meaning can be distinguished from its pragmatic function - which is to suggest a scale of individuals who have private lives and advise the hearer that presidents are at one end of this scale. In mental space terms, the scale is part of a larger meaning construction which also represents the ideological context and the different attitudes people are assumed to hold to those at different points on the scale. However, the further inference drawn from what is said and the background knowledge invoked or, strictly, constrained by even will be an implicature with a new propositional form, maybe something along the lines of What I do in my own time is my business not yours, and certainly doesn't interfere with my ability to do a good job as President. Sperber & Wilson see this implicature as a deductive inference which follows from the premises represented in the cognitive construction.2
Although we agree that the distinction between semantics and pragmatics as traditionally drawn is not motivated, we argue that cognitive semantics represents everything that is necessary to deduce Meaning X from Form Z, but not Meaning X itself.

3 The figure / ground gestalt in linguistic analysis

In this section, we review the way in which the figure / ground gestalt has been applied in linguistic analysis. The review will show how pervasive the figure / ground relation is in linguistic representation. We will argue additionally that linguistic structure and relevant non-linguistic context are equally accounted for in this way.
Essentially, figures are associated with discreteness, shape and singularity, whereas diffuseness is the principal characteristic associated with ground. This insight is owed the pioneering experimental work of Rubin (1915/1958) in the field of visual perception. What we 'see' is the figure. We assume that the contour marking the boundary of figure and ground belongs to the figure and not to the ground against which the figure is, consequently, profiled. Figures give an impression of solidity, closeness and density of colour relative to ground. The ground appears to continue uninterrupted behind the figure. It is the figure and not the ground which is remembered. However, without ground there can be no figure.
In relation to the 'actual' world, these perceptions are, strictly speaking, illusions. They are, therefore, direct reflections of cognitive processes which impose, even on a flat surface such as the page you are now reading, the impression that some parts of what you see, i.e. the graphemic symbols, are closer to you than the background (we cannot avoid the metaphor) on which they are printed. The salient shapes to which your attention is drawn are those of the symbols and not of the background page. Moreover, you can easily reproduce the symbols, but would have immense difficulty reproducing the shape which shares a common contour with them.
The figure / ground gestalt has been appealed to at a number of linguistic levels. One of the earliest is the Prague School's structuralist characterisation of poetic language, and especially the work of Mukarovsky.3 For Mukarovsky, poetic language consists of the "foregrounding" of phonological, syntactic or even semantic features, whose resulting prominence tends to push meaning into the background (Freeman, 1970: 43ff). Thus the formal means of expression rather than what is expressed is the salient figure: what is drawn to our attention in the case of poetic writing is not what is conveyed, but how it is conveyed. Mukarovsky's insight therefore lays the foundations for Jakobson's functionalist definition of poetic language as "projecting the principle of equivalence from the axis of selection into the axis of combination" (1960: 358).
Many of the later, more overt appeals to the figure / ground gestalt focus on the intra-sentential structural properties of language. For Langacker, figure designates the foregrounded entity in the trajector / landmark profile of a grammatical relation, such as that of subject and predicate (1991: 301). As the term trajector suggests, the figure is dynamic rather than static. In Talmy's account of complex sentences (1978: 628ff), figure is seen as a variable in relation to a ground provided in the subordinate clause. Thus clauses introduced by after and before, for example, and phrases introduced by during provide the ground, or presupposition, in relation to which the rest of the sentence is seen as a figure. This assertion-as-figure / presupposition-as-ground hypothesis is also taken up by Levinson (1983: 180).
At the level of the word or phrase, Hanks (1992) argues that deictics uniquely capture the relation of referential figure to indexical ground in a single linguistic expression. Thus what a demonstrative points to as a figure in an expression like you or this year is related to the indexical ground or deictic anchoring point of the speaker (and sometimes of the hearer) at the time when and in the place where the utterance occurs.
Each of these accounts generalizes a perceptual theory to the understanding of language and shows how salience, a relation of figure to ground, is basic to language. The positions adopted by Talmy and, especially, Hanks strongly suggest that the structures of language must reflect the cognitive structure of the mind in this respect.
Wallace (1982) attempts to draw together and expand on the considerable body of existing work on grammatical categories, which, he argues, demonstrate figure / ground polarity. Thus languages have a range of grammatical forms (perfective / imperfective, eventive / non-eventive, etc.) which are predominantly oriented to expressing figure / ground relations. Wallace therefore invokes the figure / ground gestalt to explain discoursal phenomena: a speaker has the means at each point in a discourse to foreground some element of propositional meaning as figure in relation to some other element as ground.
These and other like accounts may be seen as demonstrations of the linguistic reflexes of a fundamental processing strategy in which salience is perceived as a relation of figure to ground. In these accounts, ground is typically represented as a co-textual phenomenon (Langacker, Talmy) or as the linguistic option which is not chosen (Mukarovsky, Wallace). By contrast, Hanks treats context as the ground which enables the identification of the demonstratum by virtue of their relatedness. Langacker also acknowledges this type of contextual ground, stating that it includes the "speech event, its participants, and its immediate circumstances (such as the time and place of speaking)" (1991: 318).
The table below summarizes these accounts of how figure / ground relations have been invoked in linguistic analysis:


Level Author Figure which is Ground against which the
remembered Figure stands out

Word/phrase Hanks Deictic reference Deictic origo

Sentence Langacker Trajector Landmark
Sentence Talmy Content of asserted Content of presupposed
sentence sentence

Discourse Mukarovsky Form of poetic text Meaning of poetic text

Discourse Wallace (E.g.) a perfective in an (E.g.) the imperfective
imperfective discourse discourse surrounding a
perfective item

Discourse Langacker Speech act4 The speech event, its
participants and its
immediate circumstances


In this paper we extend the application of the figure / ground gestalt to show how the broader contextual, and particularly the ideological, ground is relevant in processing foregrounded linguistic phenomena.

4 Procedural and conceptual encoding and the figure / ground gestalt

So far as we know, encodings of propositional attitude have not previously been discussed in relation to the figure / ground gestalt, despite their roles in limiting the ground in relation to which the figure appears salient.
Expressions of propositional attitude are a long recognized category and include, for example, hedges and intensifiers on Gricean maxims. Thus, expressions such as I'm told, all I know is that, by the way and I mean advise the hearer of the extent to which the speaker is committed to the well-foundedness, informativeness, relevance and perspicuity, respectively, of the propositions to which they are attached. They show speaker viewpoint and advise the hearer how to take what is in focus. In Fauconnier's model of cognitive semantics, as discourse unfolds, mental spaces are constructed, each of which relates to one or more items in the propositional content of utterances. Whilst new spaces are introduced to reflect changes of propositional notions such as time and location, propositional attitude is indicated by the space taken as viewpoint.
Meta-talk is by no means limited to hedges and intensifiers on Gricean maxims. Schiffrin (1987) draws attention to a range of phenomena, including higher level predicates such as right, wrong, for example, and like, which modify propositions in the text and thus show the speaker's evaluation of the stated proposition. And in his more general work on reflexivity, Lucy (1993) brings together deictics, reported speech, gloss, mention and a range of other phenomena where it is possible to distinguish propositional description from talk about talk. All these expressions of propositional attitude invoke ground in relation to which the propositions to which they are attached or on which they comment may be seen as figures.
As well as these meta-talk phenomena, there is another important category, usually known as discourse particles, whose function has been especially productively studied within the relevance theoretic framework, and particularly in the ground-breaking work of Blakemore (1987). This has led to a recognition of the distinction between conceptual and procedural meaning. Utterances typically contain both conceptual and procedural encodings, or, as Wilson and Sperber put it, "information about the representations to be manipulated, and information about how to manipulate them" (1993: 2). This second, computational, type of encoding (i.e. information about how to manipulate representations) is held to constrain the interpretation of conceptual meaning by limiting the available ground in relation to which it is to be interpreted. Whereas to say

(1) Presidents have private lives

is to encode a conceptual meaning, to say

(1') Even Presidents have private lives

is to encode not only a conceptual meaning but also to constrain the background context in relation to which the conceptual meaning is to be interpreted by implying that the explicated proposition that Presidents of the United States such as the speaker are entitled to extra-marital relationships (inferred from "Presidents have private lives") would be low on a scale of expectability given the prevailing ideology which constitutes the contextual ground. It is in relation to this prevailing ideology that Presidents have private lives is both one amongst a set of possible variables and a salient figure. In this way, procedural meaning relates a new notion, a variable figure, to an established context, the invariant ground. This ground may well be, and perhaps usually is, in part ideological. In Sperber & Wilson (1986/95), the relevance of the variable figure is guaranteed, but is only proved by the recovery of the appropriate ground, which in this case is made possible by even.
Put simply, it is much more difficult to know in what way the would-be figure Presidents have private lives could be relevant (i.e. what might be meant by uttering it) when recovery of the relevant ground is not triggered by the procedural even. Thus procedural encodings provide the hearer with an indication of how to limit the potentially infinite set of contexts in relation which Meaning X is to be inferred. They therefore enable a hearer to recover the ground in relation to which an utterance can be a figure in a more economical way.
For these reasons, we suggest that Blakemore's relevance theoretic distinction between conceptual and procedural meaning should also be considered in relation to the figure / ground gestalt. And given the distinctly ideological context in which Clinton's conduct appears unacceptable, it is hardly surprising that his national television address exhibits a very wide range of metalinguistic and metapragmatic procedural phenomena which act as space shifters and space builders for pragmatically conditioned material, thus establishing viewpoints not set up in the meaning construction built from previous discourse.
Finally, it turns out that the figure / ground representational phenomena described in this section parallel Hanks's three-part Denotatum type - Relational type - Indexical type categorization of the figure / ground properties of deictics discussed in the previous section. The table below includes three representative examples from Hanks's original table showing the relational structures of deictic reference with respect to figure (Denotatum type) and ground (Indexical type) (1992: 52):


Form Denotatum type Relational type Indexical type

this = "the one Proximal to me"
here = "the region Immediate to you"
now = "the time Immediate to this utterance"


Both higher-level predicates such as not appropriate and the relational function of even establish a relation between figure and ground comparable to that encoded in demonstratives:


Figure Relation Ground
(Focus space) (Connector) (Viewpoint space)

the one Proximal to me (this)
Clinton's conduct Not appropriate to the cultural context
P has private life <relator = even> A 1 , A 2, A 3 ,... have private lives


5 Mental space representations of figure / ground phenomena: the case of deixis

The question then is how a cognitive semantics represents the relative salience which is a fundamental feature of perceptual cognition in general (Rubin, 1915/1958), and which, according to Wallace, determines the very linguistic categories available to us. We suggest that Fauconnier's definitions of Focus space as the space "where meaning is currently being constructed" (1997: 72) and Viewpoint as "the space from which others are accessed and structured or set up" (1997: 49) precisely allows for the representation of a figure / ground relation. This is hardly surprising since cognitive linguists seek to show how linguistic expressions evoke conceptual structures as natural reflections of such cognitive abilities as grounding, i.e. relating language (and other) events to the perspective of "the conceptualizer [who] chooses to construe the situation and portray it for expressive purposes" (Langacker, 1991: 315).
A detailed proposal for representing ground in cognitive constructions is made by Grundy & Jiang for deictics. They suggest a re-analysis of an example discussed by Rubba5 in order to account for the anaphoric as well as the deictic reading of this in the utterance

(3) ...or the same with when I go to, like, a Spanish part of
town, you know, see everything in Spanish, and I say, well,
you know, this is not where I belong

They propose that when this arises in Focus space, the index associated with the demonstrative must either find an antecedent in the same Focus space as itself, or, where this default mechanism fails, float up to Viewpoint space (by analogy with presupposition float) where it acts as a space builder for pragmatically conditioned material. How then might we get either a deictic or an anaphoric reading of this in this example?
In Rubba's account there are four spaces:
- a Base space containing the utterance situation, i.e. speaker, hearer, etc.
- a Time space opened by WHEN I go
- a Location space opened by TO like a Spanish part of town see everything in Spanish..
- and a Quotation space opened by SAY for well you know this is not where I belong.
When the Location space is Viewpoint and the Quotation space is Focus, this cannot find an antecedent in Focus and is therefore treated as deictic. When this is the case, according to Grundy & Jiang's proposal, the deictic index floats up to Viewpoint. When the index reaches Viewpoint, it builds a new mental space to be filled with pragmatic material which the space-building index causes to be recovered from encyclopedic knowledge, processing of local context, etc. and which constitutes the interpretation6. This proposal acknowledges that a deictic figure necessarily first occurs in the Focus space, but that its interpretation crucially depends on the indexical ground which will be in Viewpoint space. Index-instantiation thus becomes a matter of choosing the relevant ground among the available spaces that serve as potential alternate grounds. Thus the indexical element of a demonstrative is space building just in case it cannot co-index with a linguistic item in a Focus space, and a deictic interpretation is then assigned. In addition, treating indexes as space builders provides us with a principled way of showing how non-sentential pragmatic material is incorporated in mental spaces.
Although Rubba does not consider the possibility of an anaphoric interpretation, Grundy & Jiang suggest that the uses of the procedurals "you know" and "well" in Rubba's example shift the Viewpoint back to the Base. In this case, the index attaches to a co-referential item "the Spanish part of town" in the enlarged Focus which includes Rubba's time, location and quotation spaces. We believe that this is the preferred reading and that the deictic reading is hard (but not impossible) to recover.
The ease with which these two interpretations are modelled seems to show the advantage of mental space theory, which captures in a maximally economical manner the relatedness of deictic and anaphoric reference by showing how an index is either instantiated into a contextually inferred interpretation (deictic reference) or attached to an antecedent item in the linguistic co-text (anaphoric reference), depending on the space taken as ground or Viewpoint.7
This is a radical proposal for two reasons: Firstly, the new space created is built by the index, i.e. the deictic element of a demonstratum. Secondly, a new space is built from Viewpoint, contrary to the standard position that new spaces can only be built, for apparently obvious reasons, from Base or Focus spaces (Cutrer 1994, in Fauconnier 1997: 83). However, Grundy & Jiang's proposal seems intuitively suited to a mental space theory which allows for the incorporation of not only linguistic but also pragmatically conditioned contextual material. Indeed Grundy & Jiang argue precisely that the default for a demonstrative, such as this, is that it should find an linguistic antecedent. But where the default interpretation is impossible, the demonstrative is a space builder for non-linguistic material which is necessarily in the viewpoint of the speaker, a viewpoint which constitutes the deictic origo in relation to which the figure is in focus.
The intuitively persuasive notion that linguistically filled spaces are built from Focus and Base spaces and that pragmatically conditioned spaces are built from Viewpoint will turn out to be an important principle in accounting for data contained in Clinton's address.
At the beginning of this section, we cited Fauconnier's definition of Viewpoint as "the space from which others are accessed and structured or set up" (1997: 49). We are now in a position to expand this definition so that the spaces accessed and structured from Viewpoint include a construction of the conceptualizer's understanding of the relevant context and the instantiation of pragmatically conditioned structure. The whole mental space lattice then invites a further conceptualizing inference which results in the new propositional form of a relevance oriented implicature.

6 Instruction in Fauconnier's cognitive semantics

In his proposals for a cognitive semantics, Fauconnier posits a restricted set of frames and space types sufficient to represent all possible meaning potentials: "What human grammar reflects is a small number of general frames and space builders which can apply to organize the very large numbers of situations that we encounter or imagine" (1997: 190). The generative position in which an autonomous syntax is semantically interpreted to give a context-free truth-conditional meaning, which is itself subject to pragmatic processes resulting in a context-bound meaning is rejected (1997: 34, 111). Rather, knowledge of language involves knowing "how to apply partial grammatical instructions in context to provide appropriate cognitive configurations" (1997: 189). As in Relevance Theory, this definition treats grammar as less than fully determining of structure, recognizes the computational nature of grammatical instructions, and acknowledges the role of context in determining meaning. However, because our knowledge of the way in which context contributes to the elaboration of constructions is tacit and because, unlike grammatical instructions, context lacks any kind of formal instructions for its recovery or specification beyond the constraining effect of discourse particles, precise proposals for how context is included in a semantics are not easy to decide.
Cognitive semantics, therefore, treats sentences as sets of "(underspecified) instructions for cognitive construction at many different levels" (1997: 40). It seems to us that there will need to be two kinds of instructions: those which enable us to specify the cognitive configurations that relate linguistic material, and those which specify how frames and other kinds of schemata or ICMs necessary for the successful recovery of meaning in context are related to the cognitive configurations containing linguistic material.
Cognitive construction takes place at a level which is neither a representation of language nor a representation of models of the world:

"constructions at level C ... are a function of the language expressions that come in, the state of the cognitive construction when the language expression arises, and the context of the discourse; this includes social framing, pragmatic conditions such as relevance, and real-world events perceived by the participants" (Fauconnier, 1997: 36).
Even though space building is driven by linguistic information, the spaces themselves are not linguistically filled, because they are by nature part of a mental representation or "language thought" in the sense of Pinker (1994). Thus the mental space built by the linguistic content of the utterance is only the initial cognitive context, and can be enriched by factors in the non-linguistic context which are cognitively salient. New elements are added to spaces "by linguistic expressions (e.g., indefinites) or by non-linguistic pragmatic conditions (e.g., objects which are salient in the interaction that produces the discourse)" (Fauconnier, 1997: 39).
There is always a Base space or starting point for the construction, a Focus space in which meaning is currently being constructed, and a Viewpoint space "from which others are accessed and structured or set up" (Fauconnier, 1997: 49). These spaces may be, and typically will be, reassigned as the discourse unfolds. There are a number of processes which ensure that structuring in one space is accessible in another. The purpose of these representations is to constrain potentially available interpretations to that intended, or assumed to be intended, by the speaker.

Since new spaces may not be built from Viewpoint (except when co-incidentally it is also Base or Focus space), its function may appear somewhat underspecified. However, there is clear intuitive support for the notion of Viewpoint, which is motivated as a construct by the way in which natural language sentences express attitudes to and instructions for manipulating the propositions which they contain. It is for this reason that we propose that spaces built from Viewpoint should be available for non-linguistic pragmatic conditions, often in the form of ICMs, as motivated in our earlier discussion of deictics. Despite stating that space building is "determined by linguistic and non-linguistic features of the ongoing discourse and discourse setting" (1997: 131) and hinting that matching allows for various pragmatic parameters for a single cognitive construction (1997: 143), Fauconnier provides only a limited number of illustrations of the operation of space building in relation to the non-linguistic pragmatic conditions without which the linguistic expressions are not properly interpretable.
The remainder of this paper, therefore, focuses on the cognitive semantic treatment of discourse particles, which in Relevance Theory are taken as instructions for accessing a limited (i.e. processable) set of contexts which prove the relevance (i.e. interpretability) of the conceptual content of the utterance.
In Mappings in Thought and Language, Fauconnier mentions a number of discourse particles and discusses their function: "words like even, but, already ... typically signal implicit scales for reasoning and argumentation" (1997: 40), and "words like therefore signal deductive relationships that may not have been explicitly stated" (1997: 70).8
The essential fact to be taken into account in the case of even is that the speaker wishes to convey to the hearer that the proposition to which it is attached is at the end of a scale9 of expectability, even a , even b , even c ..even n . The cognitive construction must also include ICMs which provide background knowledge about the set of paradigmatic cases of which even P is an end-of-scale member. This is the viewpoint in relation to which the proposition in focus is understood. Taken together, proposition and context constitute the premises which yield Meaning X as a deductive inference. In the case of

(1) Presidents have private lives

no constraint on relevance-making contexts is available and the hearer will have a hard time drawing the appropriate inference as to what was meant by what was said, unless the discourse context provides very considerable help. This is because the hearer lacks a sufficiently rich perspective or viewpoint. However, in the case of Clinton's utterance

(1') Even Presidents have private lives

the existence of a pragmatic scale is suggested to the hearer containing sets of representative individual referents who "have private lives". It is this ground which enables the hearer to infer what Clinton meant by what he said. The question then is how to represent this scale with its particularly interesting ideological content in the cognitive construction.
This leads naturally to the related question of how frames or schemata are represented in meaning constructions. Fauconnier discusses the sentence

(4) In France, Watergate wouldn't have done Nixon any harm

at some length. Much of his work on this example addresses issues of counterfactual representation, which are not directly relevant to the issue under consideration here. However, in his discussion, Fauconnier states that the background knowledge required to make sense of the sentence "is not in any way conveyed by it" and suggests that when the sentence is processed, "the space builder in France ... is going to bring in two new spaces. First it beings in a space G (as in Gallic) corresponding to relevant partial background information about the French political system" (1997: 107).
It seems to us that such data are not a special case at all. The relevance theoretic construct of explicature enables us to enrich "In France" to a full propositional form something like In French political culture and "Watergate" to a full propositional form something like The Watergate break-in in which Republican Party officials were instructed to burgle the office of their Democrat opponents and steal information from them. Metonymies such as these occur in most utterances. Because the metonymic items are conceptual encodings and do not invoke pragmatic contexts beyond those required to elaborate them, they do not require the same kind of cognitive construction as those triggered by procedural encodings (such as even) that invoke, or, more accurately, constrain, contexts. In fact, metonymies are more comparable to deictics - at Viewpoint the metonymic 'index' is instantiated in an interpretation of the kind suggested above for "In France" and "Watergate".
How would this work, then?
Imagine an 'index' associated with expressions as a space builder for ICMs which the speaker as conceptualizer supposes that the hearer as conceptualizer can recover.10 Thus explicatures of linguistic representations are constructed at Viewpoint where encyclopedic knowledge is recovered in the form of ICMs to resolve the indeterminacy of the meanings being structured in Focus space.
A plausible model might see a mental space configuration for Even Presidents have private lives along these lines:

Base space (also, co-incidentally, the Viewpoint space):
- the discourse context (following Rubba), including Clinton and his television audience

Focus space = Figure:
- Presidents have private lives (conceptual content)
- Even (procedural content); an instruction to build new structure from Viewpoint which elaborates the scale associated with even

Viewpoint space = Ground (also, co-incidentally, the Base space):
- through access to relevant ICMs, the conceptual content in Focus is explicated or enriched to give the full propositional form Presidents of the USA such as the speaker are entitled to extra-marital relationships)
- the existence of a paradigmatic scale even a , even b , even c ..even n
- ICMs which provide background knowledge about the set of paradigmatic cases
- a consequential inference as to what it means to be at the end of scale of expectability (perhaps in the form of an ICM).

This mental model then provides the premises for a deductive inference which is guaranteed to produce the most relevant way of understanding what is meant by saying Even Presidents have private lives. There is nothing especially remarkable about linguistic ostention in this respect. Deciding meaning, like deciding to overtake when driving a car, is a decision taken in relation to a mental model constructed as a representation of all the relevant data.

7 Cognitive pragmatics and President Clinton's television address

President Clinton's address of 18 August 1998 was printed in The Times of 19 August under the headline "Reading between the lines of TV address". The address itself was prefaced by the sub-heading "What Clinton said". To the right of the address, the sub-heading which prefaced the interpretation offered by The Times was "What Clinton meant". Although The Times' interpretation was concerned only with such matters of conceptual content as, for example, the significance of the distinction between "legally accurate" answers on the one hand and not having "volunteer[ed] information" on the other, it is our opinion that another important aspect of "What Clinton meant" can only be captured by understanding his awareness of the significance of the ideological ground constructed by his use of procedural space builders and meta-talk in general. It is this awareness which we now explore.
Most attempts to model the cognitive construction of utterances work with examples like

(4) In France, Watergate wouldn't have done Nixon any harm

which are straightforward encodings of conceptual meaning requiring only the kind of explication suggested earlier. This contrasts with Clinton's utterance,

(2) Indeed, I did have a relationship with Ms Lewinsky that was not appropriate. In fact, it was wrong

which contains several overt comments on the proposition that he had a relationship with Ms Lewinsky: indeed and in fact and the emphatic use of the epenthetic auxiliary did advise us of the extent to which the propositional information is to be regarded as reliable; the higher level predicates not appropriate and wrong are speaker comments on the proposition; in addition, the repair "In fact, it was wrong" is evidence of the speaker's belief that he could not get away with styling the relationship "not appropriate".
The question then is how mental space construction is determined by meta-talk and how it represents the pragmatic background evoked by the speaker's realisation that the predicate "not appropriate" is inadequate to the ideological context in which it occurs.
Let's begin with considering how a an utterance like

(5) I had a relationship with Ms Lewinsky that was not appropriate

might be represented as a mental space configuration. Following Rubba, we begin with a Base space representing the discourse context, including President Clinton and his audience. A new space, which we will call R, is then opened and contains I had a relationship with Ms Lewinsky. A further comment space, which we will call C, is then opened. C inherits the material in R by optimization, or downward spreading, and adds the higher-level predicate which was not appropriate.
The issue now is how this construction differs from the construction for

(5') Indeed, I did have a relationship with Ms Lewinsky that was not appropriate.

Earlier we suggested that discourse particles such as you know and well shift the Viewpoint back to the Base. As the Viewpoint is already the Base space in the example under consideration, this doesn't enable us to distinguish Indeed-R from R. Moreover, there is a distinction between the discourse context as represented in the Base and the Clinton's perception of the discourse context. We therefore suggest a modification to allow such particles to open up a new Viewpoint space which inherits the conceptualizer's view of the initial discourse context based on ICMs from Base and represents the conceptualizer's view of the discourse context at the moment of conceptualization. Thus we can distinguish the R version of this utterance in which Base is also Viewpoint from the Indeed-R version in which Base and Viewpoint are distinct spaces. But how can we justify such structure building, apart from by appeal to its intuitive rightness?
Grundy & Jiang (1998) discuss the representation of anomalous sentences which are typically found as public address messages in Hong Kong, such as

(6) Last bus had departed

and

(7) This passage was closed

They argue that what makes them anomalous in the contexts in which they are encountered is that it is impossible to recover a reference time in relation to which the events are located. They then try to show how mental space configurations are able to represent this anomaly. Their suggestion is that the index associated with the deictic tense form acts as a space builder. In non-anomalous utterances such as

(6') When we arrived, the last bus had (already) departed

this space contains linguistic material, i.e. when we arrived, and is the Viewpoint or ground in relation to which the figure, the last bus had departed, is in Focus. The space builder is therefore an index, the deictic element of the tense, which, in non-anomalous utterances, will be instantiated in a linguistic form such as when we arrived to provide a ground or reference time in relation to which (6') is interpretable. Their characterization of an anomalous utterance such as (6) is, therefore, that there is a Viewpoint space opened by a pragmatic index which remains empty of linguistic material.
We now wish to appeal to this kind of cognitive construction as a way of accounting for the difference between R and Indeed-R type utterances, and at the same time to refine Grundy and Jiang's characterization of pragmatic anomaly: an anomaly occurs when a Viewpoint space is built and remains empty of either linguistic or pragmatically conditioned material. This is illustrated in the case of (6), which is anomalous when displayed as a public address message at a bus station (as observed originally), but which poses no such problems as the opening sentence of a novel, precisely because the reader is able to supply sufficient pragmatically conditioned material to provide a reference time. Moreover, the sentence

(6") The last bus had already departed

isn't anomalous, again because already acts as a space builder for the Viewpoint space which the hearer or reader is able to fill with pragmatically conditioned material.
We therefore suggest that expressions like indeed in (5') are space builders for a Viewpoint space which contains the speaker's apparent representation of the discourse context inherited from the Base space at the moment of conceptualization. This space enables the hearer to reconstruct the speaker's viewpoint of, or perspective on, the space R currently being constructed - in this case that the speaker wants to assure the hearer that what is asserted in R is reliable information. Emphatic did works in the same kind of way and opens a further Viewpoint space, again filled with pragmatically recovered material. Although it is sometimes possible for conflicting viewpoints to be constructed, in this case the Viewpoint spaces opened by indeed and did offer consistent perspectives on the proposition in the space being constructed.
The material in the new Viewpoint space 'I' (for indeed) will constrain the interpretation of the linguistic material in R and as a result enable the hearer to draw a conclusion as to what the speaker means by uttering R. We assume that indeed is therefore also a space builder for an ideological context recovered as an ICM in relation to which the speaker wishes to mediate his utterance of R. The distinction between the B context and the I context is that the I context contains the knowledge the speaker has of Base which is essentially ideological. Thus it is more difficult to interpret I had a relationship with Ms Lewinsky than it is to interpret I did have a relationship with Ms Lewinsky, and more difficult to interpret I did have a relationship with Ms Lewinsky than it is to interpret Indeed, I did have a relationship with Ms Lewinsky.
Knowing that he cannot get away with describing this relationship as "not appropriate", Clinton then constructs a further Viewpoint space opened by "In fact" and a further Comment space to include "was wrong", as well as a Focus that inherits the antecedent structure in respect of which "It" is an anaphor. The space opened by "In fact" is filled with a slightly different speaker conceptualization of the ideological ground from that already opened by "Indeed", and it is with respect to his perception of this new ground that the speaker assures his audience of the status of his representation, "it was wrong", as a well-founded comment.
The same kinds of analysis are appropriate to the other examples of meta-talk in Clinton's address. Consider the case of

(8) Still, I must take full responsibility for all my actions..

Fauconnier (1985/94: 114ff) suggests that the function of the arguably more propositional use of still in the apodosis of (counterfactual) conditionals is to cancel an implicature that would otherwise have arisen on the basis of expectability. Thus

If A, B

gives rise to the expectation that If A were to occur, then so would B. Thus the unexpected case

If A, B

is often metalinguistically marked, to give

If A, still B.

(8) follows the self-congratulatory statement, "I answered their questions truthfully, including questions about my private life, questions no American citizen would ever want to answer." The expectation of further statements in this vein, presumably constructed from Viewpoint as a kind of template for the new meaning to be constructed in Focus space, presupposes the same Base and Viewpoint. We, therefore, suggest that the procedural encoding, still, cancels this template and reassigns the separate Base, Viewpoint and Focus spaces of the preceding construction so that they are now understood as Base (and, consequently, Viewpoint) in relation to which a contrasting new meaning will be structured in Focus space.
As you know in

(9) As you know, in a deposition in January, I was asked questions..

returns the Viewpoint to Base and the ideological context in which the speaker is the President of the United States and not the person who in the previous Focus space has just stated "I did not volunteer information".
Although but has the same truth function as and, it implies a contrast between the conjoined stretches of discourse. Fauconnier treats but as an explicit warning against a likely implicature (1985/94: 110, 113), and in some cases against the expectation of optimization. Thus in

(10) But I told the grand jury today and I say to you now that..

the previous Focus space has become a Viewpoint space for the materials being constructed in (10). The expectation derived from Clinton's admission in the previous sentence that what he had done "constituted a critical lapse in judgement and a personal failure on my part for which I am solely and completely responsible" sets up a template for a further admission. This template is cancelled by but, and the previously constructed spaces reassigned as a new Base, and therefore Viewpoint. The expectation templates cancelled by still and but are constructions based on ideological perspectives which Clinton wishes to challenge.
The factive know in

(11) I know that my public comments..

opens a space for a presupposition which reflects the ideological perspective (the President holds) of the President's audience.
In

(12) I misled people, including even my wife

even is a space builder for a scale constructed from Viewpoint where such an act of betrayal is extremely unexpectable and of which Clinton will say, "I deeply regret that". This creates an expectation template which awaits an explanation. The explanation is provided by

(13) I can only tell you I was motivated by many factors

Only constructs a scale from viewpoint. Given such a scale, onlyp is not an especially convincing explanation - but then given the prevailing ideology in relation to which it is to be interpreted, there cannot be much of a justifying explanation.
The penultimate sentence of the address begins

(14) And so tonight, I ask you to turn away from the spectacle of the past seven months..

In the Conversation Analytic literature, so is a conventional way of signalling an upcoming formulation. Typically, formulations follow accounts and attempt to summarize the relevance or procedural consequentiality of these accounts. Thus so reassigns all the spaces containing the preceding account to a new Base (and consequently Viewpoint) in relation to which a formulation is offered. Moreover, the close connection between the account-bearing cognitive construction and the new Focus space is indicated by and.11
As we see, these examples contain space builders and space shifters which do not contribute to the conceptual or propositional content of Clinton's address, but instead provide instructions for the cognitive construction of ideological ground.

8 Conclusion

Most existing accounts of how the figure / ground gestalt is linguistically realised have focused on conceptual meaning. To be satisfied with this is tantamount to treating communicating meaning as a simple matter of encoding and decoding linguistic form rather than causing an addressee to draw inferences from explicated linguistic and inferred non-linguistic premises. We have been able to show that mental space constructions neatly allow for the construction of linguistic figure in Focus space and contextual ground in Viewpoint space. In doing this, we demonstrate how mental space representations are uniquely able to represent in a single account phenomena treated counter-intuitively (and certainly non-cognitively) as either semantic or pragmatic in other theories.

References

Blakemore, Diane 1987 Semantic Constraints on Relevance Oxford: Blackwell

Cutrer M 1994 Time and Tense in Narratives and Everyday Language PhD diss. University of California, San Diego

Fauconnier, Gilles 1975 Pragmatic scales and logical structure Linguistic Inquiry 6 353-75

Fauconnier, Gilles 1985 Mental Spaces Cambridge, Mass.: MIT Press (Revised edition CUP 1994)

Fauconnier, Gilles 1997 Mappings in Thought and Language Cambridge: Cambridge University Press

Grundy, Peter & Yan Jiang 1998 Cognitive semantics and deictic reference Working Papers in Chinese and Bilingual Studies 1, 85-103 Hong Kong: The Hong Kong Polytechnic University

Hanks, William H 1992 The indexical ground of deictic reference In Duranti Alessandro & Charles Goodwin (Eds.) Rethinking Context Cambridge: Cambridge University Press 43-76

Jakobson, Roman 1960 Closing statement: Linguistics and poetics In Thomas A Sebeok (Ed.) Style in Language Cambridge, Mass.: MIT Press, 350-77

Kaplan, David 'Demonstratives' (1977) and 'Afterthoughts' (1989) In Themes from Kaplan Almog Joseph, John Perry & Howard Wettstein (Eds) New York / Oxford: Oxford University Press 1989

Langacker, Ronald W 1991 Concept, Image, and Symbol: The Cognitive Basis of Grammar Berlin: Mouton de Gruyter

Levinson, Stephen C 1983 Pragmatics Cambridge: Cambridge University Press

Lucy, John A (Ed.) 1993 Reflexive Language: Reported Speech and Metapragmatics Cambridge: Cambridge University Press

Mukarovsky, Jan 1964 (English version) Standard language and poetic language In Donald C Freeman (Ed.) Linguistics and Literary Style New York: Holt, Rinehart and Winston, 40-56

Nunberg, Geoffrey 1993 Indexicality and deixis Linguistics and Philosophy 68, 1-43

Pinker, Steven 1994 The Language Instinct New York: W Morrow and Co.

Rubba, Jo 1996 Alternate grounds in the interpretation of deictic expressions In Fauconnier Gilles & Eve Sweetser (Eds) Spaces, Worlds and Grammar Chicago: University of Chicago Press, 227-61

Rubin, Edgar 1915 Figure and ground In Beardslee, DC & Michael Wertheimer (Eds) Readings in Perception New York: Van Nostrand 1958, 194-203

Schiffrin, Deborah 1987 Meta-talk: Organization and evaluative brackets in discourse Sociological Inquiry 20, 199-236

Sperber, Dan & Deirdre Wilson 1986/1995 Relevance: Communication and Cognition Oxford: Blackwell

Talmy, Leonard 1978 Figure and ground in complex sentences In Greenberg JH (Ed.) Universals of Human Language Stanford: University of California Press, 625-49

The Times 19 August 1998 London: Times Newspapers Ltd

Wallace, Stephen 1982 Figure and ground In Hopper PJ (Ed.) Tense-aspect: Between semantics and pragmatics Amsterdam: John Benjamins, 201-23

Wilson, Deirdre & Dan Sperber 1993 Linguistic form and relevance Lingua 90, 1-15