THE DISCOURSE SEMANTICS OF ATTITUDINAL RELATIONS: CONTINUING THE STUDY OF LEXIS

This paper explores some aspects of the problem of categorizing attitudinal relations in English, as part of a description of evaluation informed by systemic functional linguistics (SFL) — APPRAISAL. It reviews paradigmatic and syntagmatic orientations to lexis within this tradition, and the development of typological and topological representations of systemic relations. Corpus based argumentation is considered in relation to work on evaluation by Bednarek 2008; and proposals for continuing the study of lexis are suggested, focusing on resources for negotiating sadness and negative reactions to behavior (e.g. embarrassed, ashamed) and the affordances of topological representation. The paper highlights the possibilities and challenges involved in continuing the study of lexis in descriptions using SFL as their informing theory.


THEORY AND DESCRIPTION
In late 2012 I was approached by a very concerned research student who reported that some people were saying 'Appraisal Theory' wasn't a theory at all, but just a description. To which I replied: "Yes, of course. That's right. Systemic Functional Linguistics (hereafter SFL) is the theory. APPRAISAL is a description of resources for evaluation in English".
A comparable confusion around the relation of theory to description arises in the introduction to a recent collection of papers on language education informed by SFL (Whittaker et al. 2009: 2): "While SFL recognised an 'interpersonal' component of meaning, the model as stated did not readily support the analysis of speaker attitudes in text. During the 1990s, Peter White, Jim Martin and others developed an approach to attitudinal analysis, complementary to SFL, called 'Appraisal Theory'". Chapter 1 of Martin & White's The Language of Evaluation on the other hand is quite explicit that its "model of evaluation evolved within the general theoretical model of SFL" (2005: 7) and introduces appraisal resources in relation to relevant dimensions of SFL theory, including metafunction, realization, axis, system, structure, instantiation, genesis, register and genre. It concludes by situating "appraisal as an interpersonal system at the level of discourse semantics" (2005: 33). To my mind, the relation of theory to description is made clear. That said, colleagues working with APPRAISAL, and I include myself among them, have made countless references to 'Appraisal Theory' in presentations and publications, as a short-hand for a 'description of APPRAISAL resources in English within the general theoretical framework of SFL'. We need to be more careful.
The confusion at play here is of course between theory and description, or what Bernstein (2000: 131-141) refers to as L1 (the internal language of description) and L2 (the external language of description). As Maton 2014: 127 explains (elaborating Bernstein's (2000 definitions): "L1 'refers to the syntax whereby a conceptual language is created' or how constituent concepts of a theory are interrelated; and L2 'refers to the syntax whereby the internal language can describe something other than itself' or how a theory's concepts are related to referents". In these terms SFL is the L1 informing APPRAISAL, which is the L2. As Matthiessen & Nesbitt 1996 clarify, there is no such thing as a theory neutral description. Descriptions which purport to be theory neutral are simply assuming a naturalized taken for granted theorisation, much as citizens who claim to be free of ideology enact a naturalized hegemony in which only those trying to redistribute power are viewed as political. I won't pursue the discussion of internal and external language of description here; for elaboration see Muller 2007 on verticality and grammaticality, and Maton 2014 (especially his discussion of specialization, semantic gravity and semantic density). But the distinction between what linguists think of as theory and as description is crucial, and one aspect of the relation between L1 and L2 as far as the categorization of attitudinal relations is concerned is the main focus of this paper.

SFL AS A RELATIONAL THEORY OF MEANING
As is well known, SFL has evolved as a theory of language foregrounding paradigmatic relations as the basic organizing principle of both theory and description. Formalisation of these relations on the basis of the structures through which they are realized gives rise to crucial derived concepts such as rank, metafunction and stratificationwhich function as comments on the bundling of paradigmatic relations in the organisation of language. As such, the theory builds on Saussure's understanding of the sign and valeur, Hjelmslev's interpretation of language as a stratified system of signs and Firth's notion of meaning as function in context. For a basic introduction to these ideas see Halliday 2009 and. The critical concept arising from this intellectual history is the idea that language is a network of relations, an orientation shared with what was originally known as stratificational linguistics (Lamb 1996, Lockwood 1972, which has a comparable theoretical heritage. As far as meaning is concerned, this gives rise to a relational theory of meaning in which meaning is formalized as networks of options. In an L1 of this kind, language is conceived as a resource, and meaning as choice. Asking what a choice means involves explicitly relating that choice to other options in relevant systems (on a higher or lower rank, on a higher or lower stratum, and in one or another metafunction).

ДИСКУРСИВНАЯ СЕМАНТИКА И ПРАГМАТИКА
It is important to contrast this relational theory of meaning with the common sense referential one that is often taken as the basis for alternative conceptualisations in linguistics, philosophy and psychology. In common sense terms it makes sense to ask what a word means, and answer by pointing to some concrete object it refers to, or, where this is not possible, to offer a definition (the dictionary strategy). In this approach, words for example are conceived as having meaning (as realizing or encoding meaning if we want to say this more formally). From a relational perspective on the other hand, words don't have meaning; rather they do meaning -they mean in relation to the other words that might have been chosen. Similarly groups and phrases mean in relation to other groups and phrases, clauses in relation to other clauses, exchanges in conversation in relation to other exchanges, phases of discourse is relation to other phases, genres in relation to other genres and so on. So the task of description in a relational model of language is to relate choices to one another, as explicitly as possible (as opposed to offering definitions or relating meanings to real world entities or cognitive concepts). As far as APPRAISAL resources are concerned, this means describing how evaluative meanings are related to one another. And if we are focusing on feelings (i.e. ATTITUDE), as we are in this paper, this means building up a picture of the feelings we mean, describing how they are related to one another (looking round), specifying how they are realized (looking down) and outlining what they realize (looking up).
Note in practical terms that a relational theory of meaning implies that a good thesaurus is going to be a much more valuable resource than a dictionary. On the whole, definitions in dictionaries are not very well coordinated with one another, since dictionary makers tend to work a word at a time rather than with sets of related meanings. It also implies that consulting a translator will be far more insightful than introspecting about the meaning of an attitudinal expression. This is because most of us are not very good at bringing the relational meaning of a feeling expression to consciousness, whereas translators spend their whole life worrying relationally about the meaning of a word in one language and its relation to alternative translations in another. Their relational perspective is an invaluable resource in this regard (cf. de Souza 2010 on translating evaluative language from English into Portuguese). The basic message I am trying to get across here is that a relational perspective on meaning means that we need to think relationally. Formally speaking, this means looking closely at how paradigmatic relations are modeling typologically and topologically as far as the discourse semantics of APPRAISAL is concerned.

TYPOLOGY AND TOPOLOGY
SFL's usual strategy for formalizing paradigmatic relations is a system network. A system network is a two-dimensional static display of logical relations among choices for meaning. In Fig. 1 below, the kind of traffic light system used in Hjelmslev 1947 to illustrate Saussure's concepts of the sign and valeur is formalized in SFL terms as a system network with three options (technically features). The basic meaning of each choice is its relation to other choices (its valeur). Since this is a very simple semiotic system, there is not much to say from round about, above or below, since choices do not bundle into metafunctions, ranks or of the sign I have used a yin/yang symbo each sign with respect to its fusion of sig and go/green). The arrow and square bra system you have to choose one of the thr

Fig. 2: Cross classificatio
Turning to the relation of typology to relations is concerned we need to introduce t scale meanings from one pole to another. T kind of system in SFL; in Fig. 3 I have tilte involve a cline rather than a categorical op between features [c] and [d]. Where two s can be re-expressed as a topology -with o and the another clined system as the horizon ings can be graded as more or less [a] or [b speaking a topology is not limited to two di paradigms) a third dimension is challenging a dimensions require considerable ingenuity t perspective is that meanings can be graded i rically opposed.
Interpersonal meanings in particular 3 ( MODALITY in English in Chapter 10 of Hall to description of this kind. For further discu paradigmatic relations as typology and topol As indicated in Fig. 3, the regions construed terpreted as having their own centre/margin types (e.g. prototypical [a/c] below) and the or almost [b]  ances of a two-dimensional page or screen. ross-classify any number of systems using cifies the [a/b] system as simultaneous with n in relation to a paradigm o topology as far as modeling paradigmatic the notion of clined systems. Clined systems There is no standard representation for this ed the systems clock-wise to signal that they pposition between features [a] and [b] and systems are involved, this kind of network one clined system taken as the vertical axis ntal one. This creates a space in which meanb] and more or less [c] or [d]. Theoretically imensions; but practically speaking (as with as far as viewing is concerned, and additional to display. The advantage of the topological in relation to one another instead of catego- The complementarity of typology a their introductory courses in phonetics an presented typologically in terms of ideal cally in relation to tongue position. For have the same vowel system phonemical in Fig. 4 might be used to show how New further back than Australian ones (so that as /seks/, New Zealand /seks/ as /siks/ an delight in such misunderstandings). and topology will be familiar to linguists from nd phonology, where vowel systems tend to be types, and their phonetic realisations topologir Australian and New Zealand English, which lly speaking, a topological diagram such as that w Zealand short vowels are spoken higher and t an Australian might hear New Zealand /saeks/ nd New Zealand /siks/ as /suks/ and so on and display for vowel articulations he complementarity of typology and topology 007. Below I will review the two modeling stran particular AFFECT, as presented in Martin & ypology and topology across a fuller range of

THE STUDY OF LEXIS
first principle of phonological and grammatical ure and system" (1957/1968: 186). As far as the phasized the importance of studying mutual ex- [a] ДИСКУРСИВНАЯ СЕМАНТИКА И ПРАГМАТИКА pectancies between words as texts unfold, which expectancies he referred to as collocation. This corpus perspective on lexis was developed by Sinclair and his colleagues at Birmingham, beginning with his seminal 1966 paper 'Beginning the study of lexis'. Hunston 2011 presents an overview of the contributions of this work to our understanding of evaluative language, including critical contributions by Bednarek (2006. The complementary lexis as system perspective was developed by Halliday (1961Halliday ( , 1966, in relation to his proposal that the "grammarian's dream is (and must be, such is the nature of grammar) of constant territorial expansion. He would like to turn the whole of linguistic form into grammar, hoping to show that lexis can be defined as "most delicate grammar"" (1961: 267). This proposal was insightfully explored by Hasan 1987 4 in relation to a small set of material processes (gather, collect, accumulate; scatter, divide, distribute; strew, spill, share). Her typology takes material processes of disposal and their interaction with benefaction as a starting point and pushes the description in delicacy until the realization of choices can be specified in terms of specific disposal lexis. I'll use an interpersonal example here to illustrate this conception of lexis as delicate grammar -drawing on Halliday & Matthiessen's classification of Comment Adjuncts (2014: 190-193). Their first distinction is between what they call propositional and speech-functional comment (I use the features [feeling] vs [dialogism] for this in order to orient the discussion towards work on the appraisal system ENGAGEMENT). The more attitudinal comments only appear in statements (group 1 below), whereas the dialogic ones position a speaker's voice in statements and invite the addressee to position hers in questions (group 2 below).
Fortunately, we won the match. *Fortunately, did we win the match? *Fortunately, win the match.
Honestly, they won the match. Honestly, did they win the match? *Honestly, win the match.
Halliday & Matthiessen then break the dialogic comments down into a what they call a qualified and an unqualified comment; the criteria they use for this distinction is the ability of the qualified type to be followed by the word speaking: e.g. generally speaking, frankly speaking, strictly speaking (cf. *admittedly speaking, *actually speaking).
Honestly speaking, I doubt they'll win. Strictly speaking, it's invoking not inscribing feeling.
The qualified type is subsequently s ally, broadly, roughly etc.) and personal duced to this point are formalized as a sy fied point of origin the feature [indicat clauses cannot be commented on).   ДИСКУРСИВНАЯ СЕМАНТИКА И ПРАГМАТИКА that as we move from relatively closed system items to word classes with more members, the kind of motivations for features we are used to deploying for more general 'grammar' systems get harder to find.

CONTINUING THE STUDY OF LEXIS: ATTITUDE
In section 4 above we exemplified the way in which an analysis of feeling can be approached from the perspective of lexis as delicate grammar. Significantly, this meant taking one dimension of grammar (indicative mood to be precise) as a starting point and asking how Comment Adjuncts can be deployed to negotiate an attitude towards a declarative or interrogative clause -for example Sadly, they lost. But unhappiness can be realized through a number of grammatical resources, not just Comment Adjuncts. Across languages, nominal groups appear to provide the richest lexical resources for expressing feeling, through attitudinal Epithets (e.g. a sad fan). But feelings can also be realized as Circumstances of manner (e.g. they walked home sadly), and as mental processes (e.g. the loss distressed them) or behavioural ones (they frowned).
a sad fan (nominal group Epithet) they walked home sadly (Manner circumstance) the loss distressed them (mental process) they frowned (behavioural process) And grammatical metaphor can of course be deployed to reconstrue any of these realisations of unhappiness as a Thing in a nominal group: It is with great sadness that I have to inform you that they lost. The fans' sadness... They walked home in sadness. Their distress at the loss... Their frown... This means that as far as feeling is concerned the grammarian's dream has to be pursued in several regions of a grammar, each it must be acknowledged with a distinctive set of relational resources for negotiating feeling. That said, positioning lexis as delicate grammar means we cannot in lexicogrammar generalise the kinds of attitude that may be realised across different lexicogrammatical systems. To capture these generalisations we have to move up a level in abstraction to discourse semantics and make room for APPRAISAL. We move in other words from the grammarian's dream to a discourse analyst's nightmare! Not knowing quite where else to turn, our basic strategy for proposing attitudinal relations was to lean on grammar, implicitly based on the feeling that if the grammar can be bothered generalising parameters related to evaluation, they might prove useful. Note in passing the assumption here of a 'natural' relation between lexicogrammar and discourse semantics at play, in relation to SFL's conception of a stratified content plane (as lexicogrammar and discourse semantics in a model such as that proposed in Martin 1992 and assumed in Rose 2003/2007 andhere). Although we will focus on just AFFECT at this point in the discussion, space precludes a detailed presentation of the relevant grammatical parameters. In short, as summarized in Table 1 below, our [irrealis/realis] opposition derives from the distinction between desiderative and emotive mental process (I wanted them to win/I like them winning); our [desire/fear] opposition from the distinction between positive and negative expanding purpose clauses (They played aggressively so that they'd win/they played conservatively lest they lose); our [surge/disposition] opposition from the distinction between behavioural and mental processes (I cried when they lost/It upset me that they lost); our [mood/directed at] opposition from the distinction between relational and mental processes (I felt sad (but wasn't sure what made me feel that way)/The loss upset me); our [high/median/low] opposition on MODALITY (They're certainly/probably/possibly upset); and our [positive/ negative] opposition on POLARITY (I was/wasn't sad). This left us with the problem of sorting out kinds of emotion, for which the grammar didn't seem to be offering generalizable support. I was parenting a small child at the time and suggested categories based on my reading of his emotional repertoire in relation to his parents coping (or not) with his moments of distress -basically asking whether he was unhappy because he wanted his mother or father (contented sociability), or because he wanted the comfort of his security blanket (which he called 'baggy'), or because he wanted the satisfaction of his bottle ('bopple'). This gave us the [unhappiness/happiness], [insecurity/security] and [dissatisfaction/satisfaction] oppositions outlined in Table 2.   [un/happiness, triggered, positive]. It is important to note that these terms are not lexical items exemplifying the realization of discourse semantic features; they are in fact a short-hand for specific ATTITUDE oppositions. For the remainder of this paper I enclose these feature consolidating discourse semantic terms in single quotes to help avoid confusing them with the lexical items which realize them.
In order to emphasise that the lexical items included in the paradigms were simply graded examples of relevant realisations, Martin & White 2005: 51 drew on Roget's Thesaurus to illustrate the range of alternatives at play, focusing on moody unhappiness. This cell is blown up in Table 4 below (although by no means exhaustively), and gives us some indication of the scope of the task of developing the description of AFFECT to the point where it differentiates all the core and non-core lexical items realizing unhappiness from one another. It is also important to clarify at this point that the lexical items in the paradigms exemplify how discourse semantic systems are realised in lexicogrammar; the paradigms thus relate one stratum of meaning to another.

CORPUS BASED ARGUMENTATION
In section 4 above I raised the issue of motivating features in delicate lexicogrammatical systems; the same kind of problem arises for discourse semantic ones. One possible recourse is to bring corpus evidence to bear on classification schemes, as exemplified in  We'll deal with her discussion of 'fear' and 'surprise' here.
The relevant paradigm for her discussion of 'fear' is presented as Table 5 below. For Martin & White the [positive/negative] opposition at play here is between emotional reactions to things we want to happen and things we don't -between 'desire' and 'fear'. As far as surges of 'desire' are concerned, they suggest verbal process realisations graded according to the strength of the feeling they invoke. On the basis of corpus evidence Bednarek argues that realisations of 'fear' combine freely with triggers that are already present (e.g. the noise frightened her) and that such emotions are therefore not irrealis. As a first step in exploring this concern let's deal with the labeling issue. As noted above, the grammatical opposition inspiring the [realis/ irrealis] affect opposition does indeed involve what linguists regularly term irrealis meaning. In an enhancing clause complex context the opposition is clearly between what we want to happen and what we don't -positive and negative 'purpose' if you will. I studied so that I'd pass : lest I fail :: I studied because I wanted to pass : out of fear of failing 8 This grammatical opposition is then recontextualised by Martin & White to oppose feelings about what we do and don't want to happen to others. Since the terms realis and irrealis hadn't in fact been set up as features in Halliday's functional grammar (e.g. Halliday & Matthiessen 2014) they adopted the terms. It is in this discourse semantics context that they suggest that 'fear' (i.e. [irrealis/negative/triggered] emotion) concerns what might happen or not, not whether a trigger is materially present or not. In other words, when someone frightens us, are we afraid of them, or are we afraid because of what they might do? It follows that the most likely reading of It frightened me that he'd come, to my mind, is 'it frightened me that he would come', not 'it frightened me that he had come'. For It startled me that he'd come on the other hand, which realizes 'surprise' (i.e. triggered realis insecurity), the most likely reading, to my mind, is 'it startled me he had come', not 'it startled me he would come.' The alternative readings are possible; but It frightened me that he had come implies, for me, fear about what he might do. Similarly, It startled me that he would come makes sense, for me, in a context where it implies that it was hearing the news that he would come that startled me.
We also need to keep in mind at this point in the discussion that in a relational theory of meaning removing 'fear' from negative irrealis affect 9 means putting it somewhere else -re-grouping it perhaps as a parameter of insecurity. At stake here is our reading of a text like the following, from a children's picture book (Wolfer & Harrison-Lever 2005). Martin & White would read shock as realizing insecurity, in relation to the soldiers having been wounded, and terror as realizing 'fear', in relation to what might come (pain, death, capture, defeat etc.). Reworking 'fear' as a dimension of insecurity raises the question of how exactly shock differs from terror, in terms of force perhaps (high, median, low) or some other yet to be established parameter.
Jack fired his gun. He saw shock and terror in the Japanese soldier's eyes as they fell. Jack wanted to drop his rifle and cover his ears, but it was impossible to block the cries of the injured and dying men. [Wolfer & Harrison-Lever 2005] My basic point here is that labeling is not defining. Terms for classifying AFFECT have to come from somewhere, and we don't in linguistics have much terminological heritage to draw on in this regard. A term like irrealis needs to be carefully interpreted with regard to the realis feelings it opposes (not just in terms of the meaning of imperfective and perfective clauses in the grammar of ASPECT), just as a term like positive has to be interpreted in relation to the negative feelings it opposes (not simply in terms of positive or negative POLARITY and the grammar of MOOD). So what we really need to know from corpus evidence is not whether a trigger is materially present or not as far as realisations of 'fear' are concerned, but whether 'fear' can be shown (or not) to be about what might happen -as opposed to 'surprise', which is arguably about what has already occurred. How exactly such a study might be formulated as a piece of corpus research I am not sure.
Another of Bednarek's concerns has to do with 'surprise', specifically with whether it in fact realizes negative in/security. The relevant dimensions of insecurity are outlined in Table 6 below.  hoping to surprise, wanted to surprise, urge someone to surprise, it would be nice to surprise iv. as an Epithet, the lexical item surprise can modify both positive and negative lexis surprise party/surprise attack Taking the lack of 'mirroring', and corpus evidence from i-iv into account, Bednarek argues 'surprise' should be removed from [in/security] and set up as a separate category of AFFECT. Her proposed revision is outlined in Fig. 8 below.
Bednarerk's use of corpus evidence in relation to classifying AFFECT is an important step as far as developing argumentation in relation to categorizing meanings is concerned and lays the foundation for important developments along various lines. The problem here is that we don't examples of this range of realisations. S which occurred frequently enough for B TUDE is concerned, a large corpus of sp genesis of core attitudinal lexis would b study (2003). Data of this kind is unfortu and one of the more unlikely kinds of da It would also be useful to know wha tic prosody' (e.g. Stewart 2010) has on of the lexical item surprise in recurrent ev prising success, mean that a positive ap rubs off on this particular lexical item an positive/negative associations for surpris

A TOPOLOG
Since the promise of corpus-based in the short term, due to lack of appropri explore the data we have!), in this section ely arises however is that of core and non-core in relation to the discourse semantic category m surprise, with one exception (shock, which onger case for re-categorisation could be made nsidered, showing for example that all of the ssociated with both positive and negative emould find quite surprising): nned, astounded, shaken, rattled, shattered, owled over, caught unawares, caught napping, ed.
yet have corpora big enough to give enough Surprise was apparently the only lexical item Bednarek to establish patterns. As far as ATTIoken pre-school discourse revealing the ontobe ideal -by way of extending Painter's case unately the most costly kind of data to compile ata to be funded by commercial interests. at effect the corpus linguistic concept of 'semanparticular lexical items. Does the occurrence valuations such as what a nice surprise, or a surppreciation of a thing or event in some sense nd is over time in part responsible for the mixed se in Bednarek's corpus?

GICAL PERSPECTIVE
d argumentation seems unlikely to be fulfilled ate data (and enough researchers to thoroughly n I'll explore a topological perspective on ATTI- ДИСК TUDE a little further -as one possible direc of lexis. We'll begin with the feelings outline ings there are reconfigured as a topology in gative] and [surge/disposition] oppositions to exemplify the realisation of the relevant vant region of the topology.

Fig. 9: ATTITUDE topolo
If we then home in on the [negative/di right quadrant), the force of the feelings an privileged as axes (Fig. 10 below). This allo feelings on a cline from high to low (e.g. mis makes room for additional realisations (not sp ferent points along the horizontal axis. This gets us in position to begin to tackl introduced in Table 4  ction in which we might continue the study ed as a paradigm in Table 3 above. The feeln Fig. 9 below, privileging the [positive/neas axes. The lexical items used in Table 3 t feelings have been positioned in the releogy for [un/happiness] isposition] region of the topology (its upper nd the [moody/triggered] opposition can be ows us to arrange lexical items exemplifying serable, sad, down; abhor, hate, dislike), and pecified in Fig. 10 presented in Table 3. The horizontal axis bodied feelings (e.g. heart-broken, gutted where the embodied feelings are lexical physiology. The vertical axis proposes a about something we had (e.g. heart-broke thing we didn't achieve (e.g. gutted, deje able) which might be associated with ei at a half-way point on the vertical axis. I won't pursue development of this t has to do with proposing appropriate axe lexical items. As Bednarek has shown, th we find corpus evidence to show that w had and lost, but despondent about som have loved and lost than never to have l evidence, can we set up convincing fram He was grief-stricken about her dea ?He was grief-stricken about having He was despondent about having fa ?He was despondent about her deat A key problem here is that example 'right' or 'wrong', so that the grammaria grammatical structures to define the mea to deploy. Perhaps the best we can do at of attitude that serve discourse analysts' register, in relation to one or another pra In this regard let's consider an unusu propose as arguably realising both AFFE follows: guilty, embarrassed, proud, jea f embodied heart broken, broken hearted, sick at heart gutted of Linguistics, 2017, 21 (1), 22-47 CS 39 s proposes a relational parameter opposing emd) to general ones (e.g. mournful, despondent), lized in relation to some dimension of human a relational parameter opposing feelings of loss en, mournful) to feelings of failure about someected). General feelings (e.g. wretched, miserither loss or frustration can then be positioned logy for [high/moody] feelings topology any further here. The critical problem es on the basis of arguable oppositions among he nature of triggers is potentially criterial; can we feel grief-stricken about something we have mething we wanted but never got (a 'better to loved at all' opposition)? And pending corpus mes to test the relevance of an axis? ath. g failed.
es seem more or less likely, rather than clearly an's strategy of exploring grammatical and unaning potential of a language is very awkward t this stage is to propose topologies for regions needs, as part of their focus on one or another actical concern. ual set of lexis that Martin & White (2005: 60) ECT and JUDGEMENT, which they exemplify as alous, envious, ashamed, resentful, contemptu-loss frustration general mournful, grief stricken, woe begone despondent, dejected wretched, miserable ous.... This region of meaning deals in particular with emotional reactions to social behaviour, most of which are oriented to disaffiliation (i.e. social bonds at risk; for discussions of bonding in relation to identity in SFL see Stenglin 2009, Knight 2013, Martin 2010b. Within this set proud seems to be the odd term out, since it enacts satisfaction with one's achievements in positive 10 terms; the other terms negotiate negative reactions to behavior. As far as the negative reactions are concerned, a number of possible axes of opposition suggest themselves here. . What is interesting here is that the 'positive' reactions are positioned as excessive -as inappropriate desire (e.g. jealous), inappropriate affection (e.g. dote on), inappropriate trust (e.g. credulous) or inappropriate pleasure (e.g. smug; awestruck). The negative irrealis reactions also focus on excess (e.g. paranoid as 'too fearful'); the realis reactions are triggered by misbehavior and/or flawed character. The misbehaviour and/or flawed character triggering these reactions may relate either to behavior and character for which the emoter is responsible, or to behavior and character beyond their purview. Reactions are reclassified in these terms in Table 8 below. The analysis for irrealis reactions suggests opposing fears about what one has to do (daunted, intimidated) to fears about what someone else might do (paranoid, phobic), even though in both cases someone or something else triggers the fear. As a final step, as far as the analysis here is being pursued, Table 9 considers the kind of judgements 11 triggering the reactions canvassed in Tables 7 and 8. The Table  suggests that reactions and judgements do not freely combine, but in the absence of corpus evidence it is hard to know whether we are talking about tendencies or categorical distinctions. For example, the Table proposes that we can feel aggrieved about someone else's dishonesty or impropriety (social sanction), but not about their cowardice, stupidity or misfortune (social esteem). What, one day, might a corpus large enough to reveal patterns about non-core items such as aggrieved tell us about how they are in fact used? As Table 9 in effect acknowledges, the lexical items negotiating feeling in this region of meaning can arguably be double-coded as inscribing both AFFECT and JUDGEMENT. The possibility of blending thus acknowledged, it is important to note that these items generally fit snugly into our most effective colligational frame for AFFECT and are out of place in our most effective one for JUDGEMENT; it is this patterning that underlies the inscribed AFFECT, invoked JUDGEMENT analysis suggested in Martin & White 2005: 68): I felt angry that I did that. AFFECT I felt guilty that I did that.
It was brave of them to do that. JUDGEMENT *It was guilty of them to do that.
Reviewing this exercise, a number of points arise from the proposals encoded in Tables 7, 8 and 9 -many of which call Malinowski's comments on the 'gaps, gluts ДИСКУРСИВНАЯ СЕМАНТИКА И ПРАГМАТИКА and vagaries' of Trobriand Island gardening terminology to mind (1935: 65). As far as 'gaps and gluts' are concerned, the feelings at play here are overwhelmingly negative; pride is arguably the only 'feel good' reaction we negotiate about our achievements or others. Beyond this, as far as negative reactions are concerned, the cells in Tables 7, 8 and 9 are populated very differently -some with few realisations and others with several (the more populous cells of course call out for further exploration, probably along the lines of that modeled in Figures 9, 10 and 11 above). The teleological orientation of the affect category [dis/satisfaction] perhaps explains some of the skewing, since it deals with emotions arising from participation in one or another activity sequence. But a more general account of 'gaps and gluts' is well beyond, and perhaps forever beyond, our understanding of the contextual history of the lexical items involved. Perhaps a corpus revealing the ontogenesis of this region of meaning could give us a glimmer of understanding; but as noted above, corpora monitoring language development are currently prohibitively costly to assemble.
As far as 'vagaries' are concerned, the doubts I raised above about the placement of lexical realisations in Table 9 indicate the usefulness of a topological perspective alongside a typological one -since realisations can then be positioned along clined axes (e.g. as reacting to a greater or lesser extent to [normality], [capacity], [tenacity] and so on). That said I have not attempted a topological display for the meanings at stake in this region, in part because my account is a partial one, and in part because, in spite of this, there are several simultaneous axes already in play (i.e. types of AFFECT, positive or negative, in relation to one's own behavior or that of others, in relation to kinds of JUDGEMENT) -and I have no principled basis for privileging one or another of these axes in the kind of displays presented in Figures 9-11 above (where the privileging was equally arbitrary). As noted above, this is not a theoretical issue; a topology is in principle an 'x'-dimensional space. Rather the problem is representational. What is needed perhaps is a form of electronic representation which allows different axes to be foregrounded, in effect affording multiple windows of perspective on the complex agnation involved. This would rework the arbitrary privileging of axes in Figures 9 through 11 as a question of perspective, in relation to a discourse analyst's concerns. For recent developments in representation moving beyond the affordances of a 2-dimensional diagram on page or screen see Almutairi 2013, Zappvigna 2011 The multidimensionality involved here recalls van Leeuwen's work on what he calls parametric systems (van Leeuwen 2009, Martin 2011) -semiotic resources involving a number of simultaneous systems, consisting of two terms, which are graded in relation to one another. In his work on voice quality, colour and typography the systems tend to freely combine, and so a typological representation such as that introduced in Fig. 3 above is appropriate. The 'gaps and gluts' of lexical realisations means however that a representation of this kind overgeneralizes the meanings involved, proposing too many feature combinations that don't get realized and not providing enough delicacy for combinations that do. In this regard it is instructive to reflect on the complexity of the wiring in Hasan's 1987 lexis as delicate grammar initiative (e.g. her Fig. 4.2) where the possibilities afforded by simultaneous systems are all constrained with complex left-facing wiring so that only lexicalized meanings are realized.

A GRAMMARIAN'S VISION (AND BEYOND)
In this paper we have explored some of the issues arising from what Halliday 1961 has characterized as the grammarian's dream of formalizing lexis as delicate grammar. As far as attitudinal lexis is concerned we have in fact shifted our focus from lexicogrammar to discourse semantics, in order to generalize across the range of systems enacting attitude -from the grammarian's dream to a discourse analyst's nightmare.
Why nightmare? My hunch is that the bad dreams derive in part from grammarians' vision of the nature of SFL as an L1. SFL's conceptual architecture is basically derived from work on grammar -on axial relations (the particular complementarity of system and structure engineered by Halliday and his colleagues in the 1960s) and the conception of rank, metafunction and stratification arising directly from SFL's distinctive privileging of paradigmatic relations (for foundational papers see Halliday & Martin 1981). Representation was a key part of this enterprise, with system networks evolving as a formalization of systemic relations -canonically for English clauses and verbal groups. Critically a tradition of cryptogrammatical reasoning (Davidse 1998) evolved which gave rise to networks cross-classifying a small number of more general systems (e.g. PROCESS TYPE and AGENCY, MOOD and POLARITY, or THEME and INFOR-MATION) and then extending these systems and their interactions in delicacy until relevant structural distinctions had been accounted for. Lexical insertion rules did arise as part of this process, for closed system items such as English do; but for the most part the formalization of lexical relations was positioned as a second step, dependent (in delicacy) on the general grammatical relations just reviewed.
One result of this is that a robust tradition of reasoning about lexical relations has not developed in SFL; there is nothing comparable to the decades of cryptogrammatical reasoning about grammatical relations in English and other languages. And uncertainty about how to motivate distinctions undermines our work on lexical relations whether we attempt to formalize these as delicate grammatical or discourse semantic oppositions. Work in corpus linguistics has shown us one possible path forward, as illustrated from  above; but corpora aren't anywhere near big enough at present to support the kind of fine-grained analysis we need. We know that we have to think relationally, and that the meaning of a word is its relationship with other words. But in the absence of corpus evidence, we don't know how to argue for one kind of relation or another, and for one kind of relation among relations or another. Clearly we need to move beyond a grammarian's vision of SFL; but how can we best prod our L1 to evolve?
As implicated in this paper, and the work inspired by Martin & White 2005, the development of L2s addressing lexical relations will be a critical part of this processespecially where the L2s are designed for text analysis (and especially where the text analysis is oriented to social problems arising in fields such as educational, clinical or forensic linguistics). For attitudinal relations, topology appears to be a more promising form of representation than typology -since there are so many relevant axes to consider and so many of them are clines. This reflects perhaps the sense in which lexical relations are a qualitatively different kind of phenomenon than grammatical ones. Lexis after all fine-tunes the meaning potential of a culture; there are many more lexical distinc-tions than grammatical ones. And lexis is also at a culture's cutting edge; words come and go as social practices ebb and flow. So the gaps, gluts and vagaries that currently frustrate our SFL L1 in fact afford our culture. We need to embrace this challenge, not hide from it -continuing to develop L2s that confound our L1. Otherwise most of the fine-gained meaning potential of a culture will remain untheorised. As functional linguists and semioticians, we need our L1 to do better than that.