Fawcett's 2 Reasons Why Adding The Cardiff Grammar Syntactic Representation To SFL Clause Structures Cannot Reconcile The Two Theories

Fawcett (2010: 273-4):

However, we also saw in Chapter 7 that we cannot reconcile the two versions of the theory by simply adding the syntactic representation of the Cardiff Grammar to the 'multiple structures' of the Sydney Grammar, as a way to integrate them in a single structure. 

The first reason is that in the Cardiff Grammar it is simply not necessary to have any such 'intermediate' instantial representation between (1) the selection expression of features that are the output from the system networks and (2) the single, integrated structure that must be the final structural representation of any text-sentence (e.g., as shown in the upper half of Figure 10 in Chapter 7) — a fact that is demonstrated by the successful operation of the computer version of the Cardiff Grammar. 

The second reason why we cannot simply add the Cardiff representation of syntax to an IFG-style 'multiple structure' representation is that there are major (and probably insuperable) theoretical problems for the generative version of a model of language that is intended first to generate a set of five or more different structures for a clause and then, by the application of some type of 'structure conflation' rule that no SF theorist has yet attempted to formalise, to integrate them all into a single structure. 

It seems from the experience of those who have tried (in the Penman Project as reported in Matthiessen & Bateman 1991, and in the early stages of the COMMUNAL Project as described in Fawcett, Tucker & Lin 1993) that it is just not possible to incorporate 'multiple structures' in a generative SF grammar. The clear conclusion is that such grammars should be based on the concept of 'element conflation' rather than 'structure conflation'.

Blogger Comments:

[1] For the detailed arguments that demonstrate that Fawcett's claims in his Chapter 7 are based on multidimensional misunderstandings of SFL Theory, see the 103 posts here.

[2] To be clear, the unquestioned assumption here is that it would be theoretically advantageous to reconcile the Cardiff Grammar and SFL Theory. As this blog has demonstrated, over and over and over, unknown to Fawcett, the Cardiff Grammar is inconsistent — in terms of both theoretical assumptions and architecture — with both SFL Theory and itself.

[3] To be clear, as demonstrated in the examination of Chapter 7, there is no theoretical requirement that the three metafunctional structures of the clause be integrated into any single structure, let alone that of the Cardiff Grammar, which is itself a confused hybrid of form (Main Verb) interpersonal function (Subject, Complement, Adjunct).

[4] To be clear, Fawcett's first reason for why the Cardiff Grammar syntactic model cannot be added to the SFL model of clause structure is that the architecture of Cardiff Grammar does not require the SFL model of clause structure. This is analogous to arguing that the model of alchemy cannot be added to the model of chemistry because the model of alchemy does not require the model of chemistry.

[5] To be clear, Fawcett's second reason for why the Cardiff Grammar syntactic model cannot be added to the SFL model of clause structure is that the SFL model of clause structure has major theoretical problems. However, as demonstrated in the examination of Chapter 7, this problem arises from Fawcett's misunderstanding of SFL Theory, especially his false claims that (i) the metafunctional clause structures are not syntagmatically integrated, and that (ii) the metafunctional clause structures need to be integrated in a single structure. As previously explained, the metafunctional clause structures are integrated in a syntagm of clause constituents.

[6] To be clear, the reason why 'no SF theorist' has attempted to formalise a structure conflation rule is that structure conflation is not a feature of SFL Theory. Here, also, Fawcett misleads by shifting terminology from structure 'integration' to structure 'conflation' to set up his final misleading point in this extract; see [8] below.

[7] To be clear, on the one hand, as previously demonstrated, this misrepresents Matthiessen & Bateman (1991), and on the other hand, any adaptations of theory to the limitations of computers is not an argument about the validity of the theory itself, since the theory describes what humans can do, not what machines can do.

[8] To be clear, on the one hand, this is a non sequitur, because Fawcett has not presented an argument weighing up the relative theoretical values of structure vs element conflation. On the other hand, it is misleading, because it falsely attributes structure conflation to SFL Theory, and falsely claims that element conflation is not a feature of SFL Theory.

Misrepresenting Halliday (1994) On "Filling" Notation

Fawcett (2010: 251-2):

Interestingly, there is an equivalent gap in the Sydney Grammar's notation for representing the outputs from the grammar. This arises from the surprising fact that there is no diagram in IFG — or in the equivalent diagrams in Matthiessen & Bateman (1991) or Matthiessen (1995) — that shows how such a relationship should be represented in the full analysis of a text-sentence. In all of these works each unit is analysed in its own terms, almost as if the way in which they are to be related to the units above and below them in the structure is self-evident and has no complications. Filling is in fact a complex matter, and it very often happens that the possibilities as to what class of unit may fill an element depends, either in absolute or in probabilisitc [sic] terms, on choices in the generation of the unit above. The most obvious example is the restrictions on what may fill the Complements of particular Main Verbs (for which see Fawcett 1996).

Blogger Comments:

[1] This is misleading, because it is the direct opposite of what is true. Halliday (1994: 109) provides the following diagram illustrating both clause experiential function types and their realisations by classes of forms at the rank of group/phrase:

Halliday & Matthiessen (2004: 169) further elaborates the model for all three metafunctional structures:

[2] This is not misleading, because it is true. In SFL theory, each rank provides the entry condition to the systems of that rank, in which the structures of each rank are specified.

[3] This is misleading because, in SFL Theory, formal constituents are related to each other by the rank scale, and the relation between function structures at a higher rank and formal syntagms at the lower rank is specified as realisation.

[4] To be clear, in contradiction of SFL Theory, Fawcett here gives priority to the view 'from below', classes of form that realise functions, instead of the view 'from above', the functions that are realised by forms.

[5] To be clear, in SFL Theory, Complement is an element of interpersonal structure at clause rank, which may be conflated with most, if not all, types of participant in experiential structures. Any restrictions on the class of unit that realises a Complement thus depend on the type of participant with which it is conflated.

Misrepresenting Halliday (1993) On The Realisation Statement 'Conflate'

Fawcett (2010: 184-5):

Matthiessen & Bateman (1991) and Matthiessen (1995) call Halliday's "Split" operation "Expand", but otherwise there is, in principle, no difference between them and Halliday. In practice, however, they do not fully implement Halliday's concept that the MOOD structure consists primarily of "Mood + Residue" and only secondarily of "Subject + Finite + Complement" (or whatever it happens to be). See the examination of their account of generation in Section 7.4.2 of Chapter 7, which showed that their model is not a 'structure conflation' model, as the IFG model is, but an 'element conflation' model. As I argued there, this fact demonstrates that in the theoretical-generative strand of work in the Sydney Grammar the concept of 'structure conflation' is unnecessary, undesirable, and ultimately unworkable.

Blogger Comments:

[1] Trivially, Halliday (1995 [1993]: 272) specifies an 'Expand' realisation statement, not a "Split" operation:

[2] This is misleading, because it is untrue. It is only Fawcett, not Halliday, who regards these structures as 'primary' and 'secondary'.

[3] This is misleading, because it misrepresents Halliday's model. As previously demonstrated in the examination of Fawcett's Section 7.4.2, and shown above in Halliday's realisation statement (b), in SFL Theory, it is only elements that are conflated, not structures.

Fawcett's Difficulty In Understanding The Absence Of 'Insert Unit' In Halliday's Theory

 Fawcett (2010: 183):

It is hard to understand why the Sydney Grammar lists of realisation operations should not give Operation 1 at least the status of being a separate operation from 'preselection'. For a start, this operation will only generate units for the lower layers of a structure, so their realisation operations appear not to have any means of generating the initial unit of the clause. Moreover, even when we limit ourselves to lower units in the structure (such as how a nominal group comes to fill an element such as Subject, we should note that it is perfectly possible for the grammar to need to select a given feature on the next pass through the network without thereby also inserting a new unit, so that preselection must in any case be treated as a separate operation from the insertion of a new unit. 

It is therefore hard to understand why Halliday, Matthiessen and Bateman have no equivalent of Operation 1. The reason may possibly be connected with the fact that the emphasis in the Sydney Grammar is strongly on the generation of clauses. Perhaps this leads those working in that framework to take it for granted that all the choices being made in the system network are for the clause unless the rules state otherwise, so that the insertion of the initial clause does not get stated explicitly. Indeed, one surprising fact about the literature of the Sydney Grammar is that it contains no examples of generations of texts that involve a further layer of structure beyond that of a single clause — even in the fairly full description of the Penman Project given in Matthiessen & Bateman (1991). So there is no account of how the process of re-entry operates in the Sydney Grammar, whether for a clause or even for a simple nominal group. In practice, however, the computer implementation of the Penman model must have some equivalent operation to our Operation 1, or it would be unable to generate structures with more than a single layer (as it does when used in computer implementations). Indeed, the existing realisation operations refer to the generation of lower units (e.g., "Preselect nominal group" in Halliday (1993:4505) — though there are no worked examples.

Blogger Comments:

[1] As previously noted, Fawcett's Operation 1 is 'Insert unit (to fill an element)'. The reason why SFL Theory has no equivalent is simply because a unit is selected from the rank scale within the system.

Each of these units serves as the entry condition to the systems of that unit. It is the rank scale that anchors the systems of the clause — theme, mood and transitivity — on the stratum of lexicogrammar, rather than semantics. Fawcett's Cardiff Grammar locates these grammatical systems at the level of semantics, and does not include a rank scale.

[2] To be clear, in SFL Theory, a nominal group realises the element Subject.

[3] To be clear, in SFL Theory, a unit such as a clause or group is instantiated in a single 'pass' through the simultaneous systems of that unit, so there is no need for re-entry in the instantiation of a single unit. In cases where units form complexes, the network is re-entered through a system of RECURSION, as illustrated for the highest unit, clause, by Halliday & Matthiessen (2014: 438):

On The Absence Of Fawcett's Operations 1 and 6 In The Realisation Statements Of Halliday (1993)

Fawcett (2010: 182-3):

However, the most striking difference between the two sets of realisation operations is the omission from all three of the Sydney Grammar lists of our Operations 1 and 6. Operations 1 and 2 are in a sense the two fundamental operations, in that each adds one of the two major categories to the structure: Operation 1 adds a unit and Operation 2 an element. For example, the choice of the feature [situation] is realised by inserting the unit 'Cl' (for 'Clause'), and the choice of the feature [thing] is realised by inserting 'ngp' (for 'nominal group'). If there is no unit, there can logically be no elements, and so no structure. Thus the computer implementation of the Sydney Grammar must have such an operation, or it could not work. The odd thing is that it is not listed here. Operation 6 equally is crucial, because if the network is not re-entered only one layer of structure can be generated. 

The clue to the answer comes in Matthiessen and Bateman's description of what is covered by their term "preselection". For them, "preselections are instructions to re-enter the grammar to develop the function constituents of the clause as nominal groups, [...] and so on" (1991:108), so that their "Preselect" operation includes all three of (1) 'preselection' (in the strict sense of the term). (2) 're-entry' and (3) the insertion of a new unit. No doubt we can infer from this that Halliday's "preselect" operation includes the same three steps. 

Blogger Comments:

[1] To be clear, Fawcett's Operation 1 is 'Insert a unit (to fill an element)' (p180). In SFL Theory, a unit (clause, group, word, morpheme) is selected from the rank scale in a system network. SFL Theory models the phenomena of syntax as a rank scale. Fawcett's model of syntax (allegedly) does not include a rank scale.

[2] To be clear, Fawcett's Operation 6 is 'For an element, re-enter the system network' (p180). In SFL Theory, the selection of a given unit entails the selection of the features and the activation of realisation statements that specify all the elements of that unit. This includes structures that Fawcett regards as involving more than one "layer", such as clauses with textual and/or interpersonal Themes, where each is modelled as a simultaneous system within the network. Halliday & Matthiessen (2014: 162):

[3] This is misleading, because it is untrue; see [1] above. Halliday's (1995 [1993]: 273) realisation statement is not concerned with inserting a unit, but with preselecting a feature at a lower rank:

• 'Preselect' some feature at a lower rank (e.g., preselect actor : human collective);

[4] To be clear, Matthiessen & Bateman (1991) involves the adaptation of SFL Theory to the limitations of computers for the purposes of text generation. As such, it is not equivalent to the SFL model of the human use of language.

Misrepresenting Halliday (1993) On The Realisation Statements 'Preselect' And 'Classify'

Fawcett (2010: 182):

Halliday's Operation (f) is roughly equivalent to our Operation 5, with the important difference that ours is expressed in terms of preferences — and preferences can set the probability of a feature being selected on re-entering the system network to any point on a scale from 0% to 100%. In contrast, the more limited concept of preselection implies an absolute probability (i.e., 100%) that the specified feature will be selected. Interestingly, it appears from Halliday's examples that his "Classify" operation is essentially the same as his "Preselect" operation, and both Matthiessen & Bateman (1991) and Matthiessen (1995) simply omit "Classify" from their lists. The two are therefore treated as equivalent in Figure 11.

Blogger Comments:

[1] This is misleading, because it is untrue. Halliday's realisation statement (f) is

• 'Preselect' some feature at a lower rank (e.g., preselect actor : human collective),

whereas Fawcett's Operation 5 is

• Prefer certain features on re-entry to the system network , including preselection (= 100% preference). 

These are not 'roughly equivalent', because Halliday's statement does not involve re-entry, and because Fawcett's Cardiff Grammar does not include a rank scale.

[2] This is misleading, because it is untrue. Halliday's realisation statement (d) is

• 'Classify' an element (e.g., classify process as mental : cognition);

As can be seen, 'classify' affects features of a given rank, whereas 'preselect' affects features of the rank below.

[3] It is therefore misleading to treat the two realisation statements as equivalent. In downplaying these differences between Halliday's realisation statements and Fawcett's realisation operations, Fawcett is downplaying the disadvantages of not including a rank scale in his model.

Fawcett's Summary Of His Argument Against Metafunctional Clause Structures

Fawcett (2010: 143):

In summary, we can say that the main focus of interest among most systemic functional linguists is the description of texts. When one is working on text analysis, what matters most is that one has as one's working tool the most insightful method of analysis that one can find. The approach to representing structure in IFG has one great advantage over others (such as an updated Scale and Category analysis, e.g., as advocated in Morley 2000). This is, of course, that it provides a clear visual image of the multifunctional nature of language — and so a clear mental construct. Moreover, it is currently the only SF framework for describing language that is available in published form (i.e., in IFG and the various derived publications). This fact has led to the widespread acceptance of IFG-style analyses as the only 'correct' SF analyses — and so, coincidentally, to the widespread acceptance of the 'structure conflation' model.

Furthermore, it seems probable that most inquirers into the nature of SFL have ignored works such as Halliday's "Systemic theory" (1993) and Matthiessen & Bateman's Text Generation and Systemic Functional Linguistics. (1991) — and yet it is in such works that we find the statements that define the Sydney version of the theory. On the other hand, even these works fail to bring out the fact (as we have seen it to be) that there is no theoretical-generative SF grammar that is capable of either (1) generating multiple structures of the type illustrated in IFG and (2) integrating them into a single structure.

Blogger Comments:

[1] To be clear, any "correctness" of the model of clause structure in IFG (Halliday 1994) lies in it being consistent with SFL Theory.

[2] This is misleading, because it misrepresents both SFL Theory and the text linguists who apply SFL Theory.  As previously explained, the notion of 'structure conflation' does not feature in SFL Theory, and as such, there is no acceptance of the model whatsoever.

[3] This is misleading. To be clear, "the Sydney version of the theory" is Halliday's own version of the theory he himself painstakingly devised, whereas other "versions" are merely alterations to Halliday's theory based on misunderstandings, such as those exposed on this blog. For "statements that define" Halliday's version of his own theory, see Halliday & Matthiessen (1999) and IFG (Halliday 1985, 1994; Halliday & Matthiessen 2004, 2014).

[4] This is misleading, because it is untrue on two counts, since metafunctional clause systems do "generate" metafunctional clause structures, and since integrating them into a single structure is not a requirement of the theory. To be clear, clause structures are integrated in the formal syntagm (of group/phrase rank units) that realises them.

Misrepresenting Clause Structure

Fawcett (2010: 138):

We can now see that, in terms of the metaphor of a text as a 'rope' of interwoven strands, the 'rope' often consists of just one strand — though it also often has two or three very short multiple 'strands of meaning' which extend over one or two elements. Matthiessen and Bateman's diagram in Figure 9 exemplifies this point in terms of the Sydney Grammar, and Figure 10 of Section 7.8 does so in terms of a Cardiff Grammar representation.

Blogger Comments:

[1] Trivially, the rope metaphor refers to the structure of the clause, not the text.

[2] To be clear, in SFL Theory, a major clause always has three strands of metafunctional meaning. However, just in terms of Fawcett's own argument, if a clause often has more than one strand of meaning, then the model has to accommodate the fact.

[3] This is misleading, because it is untrue. Matthiessen and Bateman's diagram (Figure 9) represents a clause as three strands of meaning, those of THEME, MOOD and TRANSITIVITY:

[4] This is misleading, because it is untrue. The Cardiff Grammar analysis (Figure 10) represents a clause as six strands of meaning (semantics) — in addition to that assigned to syntax.

The Problem Of Conflating Non-Coterminous Elements

Fawcett (2010: 136-7, 137n):

Consider again the nature of the representations in Figure 7, which is a typical IFG-style analysis. The main problem with it is that so few 'functions' in one strand are coterminous with 'functions' in another strand. What, precisely, would the rule be like that 'conflated' all four of the Finite, the Predicator, the Process and the Goal with the Rheme? How would it be formulated in such a way that it could be generalised across many cases? The answer is that we do not know — and I for one do not think that it would be worth spending a lot of time in exploring this route, since we know that there is already in existence a simpler and more insightful alternative.¹⁹

19 It would be quite a simple matter, of course, to add to a computer implementation such as Matthiessen and Bateman's a small program that inserted the label "Rheme" in any box in the line of thematic meaning that was not already labelled, and to do the same for "Given" and "Residue". But this would be simply a cosmetic adjustment to make the output appear even more like that in a typical IFG representation, rather than the positive conflation of elements, such as happens in the 'element conflation' model. Matthiessen and Bateman are to be congratulated for resisting the temptation to make their computer output more like the IFG representations than they really are (e.g., as illustrated in Figure 9).

Blogger Comments:

Reminder:

[1] Fawcett's assumption here, that each element of each function structure must be conflated with an element of every other function structure, derives from his mistaken notion of 'structure conflation'; see previous posts.

[2] To be clear, the model that Fawcett appraises as 'a simpler and more insightful alternative' is his own.

[3] This is misleading. To be clear, the structural representations in Matthiessen & Bateman (1991) and in Halliday (1994) are functionally equivalent. In the former, 'not Theme' (etc.) is represented iconically by the absence of a label, whereas in the latter, 'not Theme' (etc.) is represented by a label: 'Rheme' (etc.). Moreover, the presence or absence of a label is irrelevant to the conflation of elements, since it is the elements that are conflated, not their labels.

[4] Here Fawcett again raises the issue of intellectual integrity. By positively judging Matthiessen & Bateman in this instance, he implicitly raises doubt on the issue with respect to other instances, thereby reinforcing his previous negative judgement (p135):

it seems clear that that Matthiessen and Bateman's only purpose in generating the Mood elements is to generate a second line of interpersonal structure, and so to make available a visual representation that looks a little more like an IFG analysis. In other words, the generation of the 'Mood' element is simply a cosmetic exercise.

Fawcett's Argument Against The 'Structure Conflation' Model

Fawcett (2010: 136):

Note that the structure that is generated in Matthiessen and Batemen's [sic] generator is indeed a single integrated structure. Indeed, none of the grammars referred to in this section generate first a set of different structures (as is implied by representations such as those in Figure 7 of Section 7.2) and then conflate them. As the decriptions [sic] in all of the publications from Halliday (1969/91) to Matthiessen & Bateman (1991) and Fawcett, Tucker & Lin (1993) show, it is a single structure that the grammar builds — and not a multiple one. 

If this is so — and there can be no doubt that it is — any grammarian who wishes to claim that the 'structure conflation' model has a theoretical status (rather than some other value) has a number of problems to solve.

Blogger Comments:

[1] This is misleading because it is untrue. As Fawcett's Figure 9 demonstrates, Matthiessen & Bateman, like Halliday, provide three distinct clause structures, each differentiated from the others by metafunction: theme, mood and transitivity.

As previously explained, the three different function structures of the clause are integrated in the group rank syntagm that realises all three of them. (Fawcett will later argue against the theoretical value of a rank scale.)

[2] To be clear, 'structure conflation' is not a feature of SFL Theory. The notion is Fawcett's only, and arises from his misunderstandings of SFL Theory, as previously explained. That is, Fawcett is merely arguing against his own theoretical misunderstanding (a logically fallacious Straw Man).

[3] As previously noted, Figure 7 is Fawcett's misrepresentation of an SFL analysis in which he misconstrues information as a system of the clause, and the Scope (Mrs Skinner) of the Process as its Goal:

Misrepresenting The Structural Representation Of Matthiessen & Bateman

Fawcett (2010: 136, 136n):

Thus, while the representation of the structure of a clause in Matthiessen & Bateman (1991:109) may look roughly like the representations of clauses in IFG in terms of the number of boxes and in its use of some of the same labels as in an IFG representation, the structures that it generates are in reality more like the representations in the Cardiff Grammar (e.g., the example that we shall meet in Figure 10 in Section 7.8). In other words, the two have in common that (1) the only type of conflation that they have is 'element conflation', and (2) they both leave empty the quite sizeable boxes where an IFG analysis would write in labels such as "Rheme" and "Residue".¹⁸ 

18. The same principle would apply to "Given", if Matthiessen and Bateman's generator generated intonation too (as the Cardiff generator does). This is because, like "Rheme" and "Residue", "Given" is essentially 'that which is not New'. 

Blogger Comments:

[1] This is misleading, because it is untrue. This is easily demonstrated by comparing the Matthiessen & Bateman representation (Figure 9) with the Cardiff Grammar representation (Figure 10).

As can be seen, the two figures represent very different analyses. For example, as well as the different function labels, Figure 10 divides grammatical functions into semantics and syntax, and mistakes information for a system of the clause.

[2] To be clear, what the two representations have in common is a principle of SFL Theory (element conflation) and the non-labelling of negatively-defined elements, namely: 'not Theme' (Rheme) and 'not Mood' (Residue). 

The "Cosmetic Exercise" Of Matthiessen and Bateman

Fawcett (2010: 135):

To summarise: it seems clear that that Matthiessen and Bateman's only purpose in generating the Mood elements is to generate a second line of interpersonal structure, and so to make available a visual representation that looks a little more like an IFG analysis. In other words, the generation of the 'Mood' element is simply a cosmetic exercise. Essentially, then, the Nigel generator that Matthiessen and Bateman describe here is one that generates single structures, just as Halliday's earlier models do (e.g., Halliday (1969/81). In other words, their generator conflates 'functions to generate multiple elements, as shown above, but it does not generate multiple structures. 

Blogger Comments:

[1] This is misleading, because it is untrue. The reason why Matthiessen and Bateman generated Mood elements is because the Mood element is a structural function of the clause, according to the theory they used.

[2] This is misleading, because, in Fawcett's own terms, it is untrue. The text generator of Matthiessen and Bateman produces Subject and Finite as components of the Mood element, which for Fawcett, constitutes "two lines of structure".

[3] To be clear, the conflation of functions does not generate "multiple elements". For example, the conflation of Theme, Subject and Actor does not generate a "multiple element". Each of these is an element of a different structure, a different configuration of relations, and so the notion of a "multiple element" of structure is nonsensical. What is true, in terms of SFL Theory, is that conflated elements are realised by the same unit at the rank below, such as a nominal group, as previously explained.

[4] This is misleading, because it it is untrue. The text generator of Matthiessen and Bateman produces each of the metafunctional structures of the clause: namely, those of theme, mood and transitivity.

Fawcett's Claim That The Finite (Clause) Should Be Conflated With The Event (Verbal Group)

Fawcett (2010: 134-5, 135n):

The reason why it is important to generate the Subject and Finite as direct elements of the clause (rather than the 'Mood' clement') is that it is the Subject and the Finite on which further work must be done. This further work will  generate (1) the nominal group the new system to fill the Theme/Subject/Actor and (2) the item is to expound the Finite/Process.¹⁶ Interestingly, however, Matthiessen and Bateman do not explain why — or even how — these are generated as "subcomponents' of 'Mood'. (However it is done, the same principle presumably applies to the division of the "Moodtag" into its two elements.) 

16. Strictly speaking, in the Sydney Grammar it is the element 'Event' within the 'verbal group' (the unit that fills the clause element 'Predicator') with which the 'Finite' should be conflated, and not the 'Process', as an example with auxiliary verbs such as the new system may be going to be more reliable than the old one demonstrates. See Fawcett (2000 and forthcoming b) for a full critique of Halliday's approach to the 'verbal group' and Appendix C for a summary. 

Blogger Comments:

[1] To be clear, on the one hand, the term 'direct' is a red herring here, since there is no distinction between 'direct' and 'indirect' elements of clause structure in SFL theory. The elements Subject and Finite are no less 'direct' for being components of the Mood element.

On the other hand, Fawcett's use of computerised text generation to argue about theoretical issues is also a red herring, since adjustments of the theory in response to the limitations of computers are irrelevant to the theory as a model of human language.

[2] To be clear, the reason why Matthiessen and Bateman generate Subject and Finite as components of the Mood element is because they are applying SFL theory to computerised text generation, and according to the theory, this is a structural realisation of the systemic feature 'indicative'.

[3] To be clear, the Predicator is realised by all elements of the verbal group except the Finite. In Fawcett's example, the Predicator is realised by the elements Auxiliary₁ ^ Auxiliary₂ ^ Event, as shown below.

the new system	may	be	going to	be	more reliable than the old one
Carrier	Process: attributive				Attribute
Subject	Finite	Predicator			Complement
	verbal group
	Finite	Auxiliary₁	Auxiliary₂	Event

[4] This is misleading, because it misrepresents Halliday's theory.  Fawcett's claim is that, on the SFL model, the Finite (clause rank) should be conflated with the Event (group rank), and that his example demonstrates the fact.  That is, Fawcett's claim is that two syntagmatically separated elements, at different ranks, should be conflated. This demonstrates a serious failure to understand the notion of conflation.

As the analysis above makes clear, in Fawcett's example, the Process (transitivity) conflates with Finite ^ Predicator (mood), both of which are integrated in their realisation as a verbal group.

On There Being "No Conflation Of Non-Coterminous Elements"

Fawcett (2010: 134):

Thus, after the conflation rules have been applied, the structure of the clause the new system is more reliable than the old one (omitting the 'Moodtag isn't it), as generated by Matthiessen and Bateman, can be represented very simply as follows: 

Theme/Subject/Carrier + Finite/Process + Attribute. 

Thus two of the three elements are formed from conflated 'functions' and the third consists of a single 'function'. If you compare this economical representation with the corresponding part of the representation in Figure 9 above, you will see that, from the generative viewpoint, it contains all the necessary information. The key point is that Matthiiessen [sic] and Bateman's generator operates in essentially the same way as the one described in Halliday (1969/81) — and that there is no conflation of non-coterminous elements. 

Indeed, Matthiessen and Bateman do not make clear precisely what role the IFG-like diagram given in Figure 9 plays in the process of generation. It may well be that their main reason for generating it was to provide a satisfactory visual representation of the structure for those who are expecting something that resembles an IFG-style analysis, rather than because it is necessary for the process of generation. 

Blogger Comments:

Reminder:

[1] This is true. In SFL Theory, elements of clause structure can only be conflated if they are integrated by being realised as the same group rank unit, as in the realisation of Theme, Subject and Carrier as the same nominal group. However, Fawcett misleadingly presents this as if it were a problem for the theory, rather than a principle of it.

[2] To be clear, the rôle of such diagrams is to represent a computer-generated clause structure in terms of the model used: SFL Theory.

The Undesirability Of Finite and Subject As Subcomponents Of Mood

Fawcett (2010: 134, 134n):

Unfortunately, Matthiessen and Bateman give no explanation as to why they find it desirable to generate the Finite and the Subject as "subcomponents" of the 'Mood', even though the same effect could have been obtained by simply inserting the Finite and the Subject as direct 'functions' of the clause. The only 'advantage' (if it is one) is that the final structure that is generated resembles slightly more closely the representations shown in IFG than it would otherwise. But there seems to be little point in going to all of this extra work when the 'sister' element to the 'Mood' (i.e., the 'Residue') does not get generated — and when, even if it did, it would serve no useful purpose.ⁱ⁵ 

15. Incidentally, this criticism of the introduction of two layers of structure to model 'interactional' meaning is as valid for text-descriptive work as for theoretical-generative work. 

Blogger Comments:

[1] To be clear, the reason why Matthiessen and Bateman "generate the Finite and the Subject as "subcomponents" of the Mood' is that these are components of the model, SFL Theory, that they are applying to text generation.

[2] To be clear, the 'advantage' of identifying the Mood element of a clause is provided by its explanatory power. Halliday & Matthiessen (2014: 150) identify both the semantic function of the Mood element:

Hence the Mood element has a clearly defined semantic function: it carries the burden of the clause as an interactive event. So it remains constant, as the nub of the proposition, unless some positive step is taken to change it…

and (op. cit.: 170) its rôle in realising systemic distinctions:

In the grammar of MOOD in English, the general principle is that less delicate distinctions in mood are realised through the Mood element – its presence and the nature and relative sequence of its element, Subject and Finite, plus the presence of the WH- element…

[3] To be clear, the underlying assumption here is that expediency in the generation of texts by computer is the criterion of theoretical validity for a model of human language.

[4] This is misleading. As Figure 9 demonstrates, the Residue is generated, but since it is simply the negative of Mood, for expediency it isn't labelled.



[5] This is true.

Misrepresenting The Theoretical Status Of The Finite Element

Fawcett (2010: 133-4, 133n):

The procedure described in Matthiessen & Bateman (1991) is first to generate the 'function' of 'Mood' in the 'primary' structure of the interpersonal line of structure, and then, apparently straight away (judging by the 'blackboard' representation in their Figure 7.15 on page 108), they insert the 'function' of 'Finite' by the use of what they term the "Expand" operator. The effect of this is to build the Finite into the secondary structure as what they term a "subconstituent" of the higher 'function' of 'Mood'.¹⁴ A later realisation statement then adds the 'Subject' as a second 'subconstituent' of the 'Mood' 'function' in the same way, and then another finally orders them with respect to each other.

14. The Finite is therefore in effect a 'function', while not being, oddly, a direct 'function' in a 'unit', in the usual manner. However, it seems that in some unspecified way it must still count as a direct 'function' of the clause, because it later gets conflated with the direct clause 'function' of Process. (This is necessary because the word is in the text is both the Finite and the Process, as Figure 9 shows.)

Blogger Comments:

[1] This is misleading. To be clear, this describes an application of SFL Theory to the generation of texts by computer. It is not a description of the theory as a  model of human language.

[2] This is misleading. The word 'direct' is a red herring here. The Finite is an element of the interpersonal function structure of the unit 'clause', and in this instance, is realised by a verbal group. The fact that the Finite is a component of the element 'Mood', does not make any less direct an element of clause structure, and Fawcett offers no argument in support of his bare assertion.

Fawcett's Argument Against The Mood Element

Fawcett (2010: 133):

However, there is a third point of interest in Matthiessen and Bateman's diagram. It is the fact that they have felt obliged to make some show of generating what for Halliday is the 'primary structure' in the analysis of MOOD. This is the 'function' 'Mood', which in an IFG-style description must be represented in addition to the 'Subject' and the 'Finite', which are shown as its two 'subcomponents". But the fact is that it is only the 'Subject' and the 'Finite' that are needed for (1) conflation with other 'functions' from other lines of the analysis, and (2) the next stage of the process of generation. In other words, the 'function' of 'Mood' plays no part in the generation of the clause. Clearly, it would be far simpler to generate the Finite and the Subject as direct 'functions' of the clause, so that they can be directly conflated with the appropriate other 'functions'. 

Blogger Comments:

Reminder:

[1] This wording seems to insinuate an accusation of insincerity on the part of Matthiessen and Bateman, especially when considered in the light of the earlier judgement of the diagram as 'remarkably honest' (p131).

[2] To be clear, here Fawcett argues against the theoretical value of the Mood element of the clause, in modelling language, on the basis of degrees of expediency in the generation of texts by computers. The theoretical value of the Mood element is in its explanatory power in understanding language, not in whether or not it conflates with the element of another structure in the process of computerised text generation.

Halliday & Matthiessen (2014: 150) identify both the semantic function of the Mood element:

Hence the Mood element has a clearly defined semantic function: it carries the burden of the clause as an interactive event. So it remains constant, as the nub of the proposition, unless some positive step is taken to change it…

and (op. cit.: 170) its rôle in realising systemic distinctions:

In the grammar of MOOD in English, the general principle is that less delicate distinctions in mood are realised through the Mood element – its presence and the nature and relative sequence of its element, Subject and Finite, plus the presence of the WH- element…

[3] To be clear, Subject and Finite are "direct" functions of the clause, as components of the Mood element, and the Subject, more usually, is "directly" conflated with other functions, with this typically depending on systemic selections such as those of THEME, VOICE and MOOD.

The Integration Of Different Function Structures (And The Analysis Of Moodtags)

Fawcett (2010: 132-3, 133n):

The importance of establishing these facts is as follows. Despite the  similarity in appeararance [sic] of the representation in Figure 9 to those in IFG, it is clear from both the verbal description in Matthiessen & Bateman (1991) and from the diagram showing "successive states of the blackboard" (their Figure 9.15 on their page 108) that in their account of the generation of a clause (or any other unit) there is no stage at which there are co-existing sets of different functional structures which must later be integrated. And this is the case with all SF generators. 13  

13. For our present purposes, we shall ignore the "Moodtag" in the Matthiessen and Bateman analysis, because it raises additional theoretical and descriptive problems. These arise because their analysis suggests that the elements labelled "Moodfinite" and "Moodsubject" function directly as elements of the clause, rather than being elements of a clause that is embedded within the 'Tag', as it would be in the Cardiff Grammar analysis. See Fawcett (1999) and (in press) for the Cardiff Grammar's analysis of such 'tags', and for the evidence that they should be modelled as embedded clause of a 'truncated' type. 

Blogger Comments:

[1] To be clear, Fawcett is here arguing against a theory of language on the basis of how the theory is adjusted to fit the limitations of computers in generating texts. More importantly, he misunderstands the model, since clause structures are integrated in the syntagm of group rank units that realise them. (Strategically, Fawcett again switches terminology from the 'conflation' of structures, the misunderstanding that invalidates his argument, to the 'integration' of structures, a genuine principle of the theory.)

[2] To be clear, this is how Moodtags are analysed in SFL Theory: as functional elements of the clause whose Subject and Finite they reprise, and whose presence and polarity depend on the mood features of that clause.

[3] To be clear, the question here is whether the 'Tag' itself is an element of the clause whose Mood element it reprises. If it is, then the 'Tag' functions "directly" as an element of the clause, whether or not it is said to be realised as an embedded clause. If it is not, then Fawcett's nonsensical suggestion is that a 'Tag' is a clause realised by an embedded clause.

The Problem Of Limitations On 'Potential For Conflation'

Fawcett (2010: 132):

The second important difference in Figure 9 from the standard IFG-style analysis follows from the first. It is that the potential for 'conflation' among the 'functions' in Matthiessen and Bateman's "output" diagram (and in fact in all other published accounts of SF generation) is very limited indeed. In the generation of their worked example (shown in Figure 9) the only conflations that occur are (1) the conflation of the elements "Theme", "Subject" and "Carrier", and (2) the conflation of the elements "Finite" and "Process". More specifically, notice that the only 'functions' that Figure 9 shows to be conflated are ones that are coterminous with each other.

Blogger Comments:

Reminder:

[1] To be clear, here Fawcett is comparing representations of structures, the one in the modelling of human language, the other in the generation of texts by computer — rather than examining the theoretical criteria on which structures are based — as a means of questioning the theoretical validity of structures. Again, this makes Fawcett's argument an instance of the Red Herring fallacy, since how structures are represented (even within the same field) is irrelevant to the the theoretical validity of the structures themselves.

[2] This is misleading. To be clear, here Fawcett falsely presents an aspect of SFL Theory, the conflation of coterminous elements, as if it were in contradiction with the theory itself, and as if it supported his own argument; see the preceding and following posts.

Misconstruing An Absence Of Label For An Absence Of Function

Fawcett (2010: 132, 132n):

From our present perspective, the first important difference about Figure 9 is the blank boxes in the THEME and MOOD lines of the representation. Most of the THEME line of analysis is a long blank box, covering several elements, in which the label "Rheme" would be written if this was an IFG analysis. And there is similar blank box in the MOOD line of Figure 9, where "Residue" would be written in an IFG analysis (between "Mood" and "Moodtag"). This is not the trivial difference that it might at first appear, because when the blank boxes are labelled this suggests that there are 'functions' in these lines of analysis that correspond to them — and it is this fact that in turn allows one to read into such diagrams the idea that they represent a 'structure'. In other words, I suggest that there is not in fact a single 'functional element' called a "Rheme" that corresponds to several other clause elements; that there is no 'functional element' of 'Residue' that similarly covers one or more elements; and, in cases where there is an INFORMATION line of analysis, that there is no 'functional element' of 'Given' that covers one or more elements.¹²

¹² As you may have noticed, the right hand box in the TRANSITIVITY line of analysis is also blank, but in this case an IFG-style analysis would not label it. This is because this element only has 'interpersonal' meaning. Since the tag isn't it is automatically derived from the main clause it has no 'experiential' meaning, so that there is nothing to say about it in terms of the TRANSITIVITY line of analysis. 

Blogger Comments:

Reminder:

[1] This is misleading, because the difference is indeed trivial. The difference here is between labelling a negative — Rheme or Residue — and not labelling it. 'Rheme' labels the portion of a clause as message that is not Theme: the point of departure of the message. 'Residue' labels the portion of a clause as exchange that is not Mood: the carrier of 'the burden of the clause as an interactive event' (Halliday & Matthiessen 2014: 150).  The unlabelled box after 'Theme' signifies 'not Theme', which is equivalent to 'Rheme'; the absence of a label after 'Mood' signifies 'not Mood', which is equivalent to 'Residue'.

[2] To be clear, a structure is the relation between elements, and the relations obtain if both elements are labelled or if one of them is iconically represented as a negative.

[3] To be clear, this is a non-sequitur. Here Fawcett has invalidly concluded that the absence of a label necessarily entails the absence of a function. Moreover, this conclusion leads to an absurdity, since, without at least the possibility of an element that is not Theme (Rheme), not Mood (Residue), not New (Given), there can be no contrasting element that is Theme, Mood, New, respectively.

[4] To be clear, the reason why a Moodtag 'has no experiential meaning' is because it is not a representation of experience.