On The Timing Of The Application Of Realisation Rules

Fawcett (2010: 66-7):

In practice the simplest workable solution is that no realisation rules should be applied until the traversal of the network has been completed, and to make the resulting selection expression of features available to each realisation rule, as it is applied. This is what is done in the Cardiff Grammar and, as we shall see in the next chapter, in the computer implementation of Halliday's grammar too.

Blogger Comments:

To be clear, here Fawcett confuses system networks, the SFL formalism that models human language as potential, with the flowcharts he uses for text generation by computer.  In doing so, he presents his solution to his own technical problem as a solution to a non-existent theoretical problem (as demonstrated in previous posts).

Fawcett's Argument Against Realisation Rules In System Networks

Fawcett (2010: 66-7):

Let us look at a simple example, taken from the little grammar in Appendix B. Consider the realisation rule for the feature [near] in Figure 2 of Appendix B. The rule states that, if either [singular] or [mass] is also chosen, the realisation is that the deictic determiner (dd) will be expounded by the item this, but that if the feature [plural] is co-selected it will be expounded by these. 

In this particular example, the conditional features happen to occur in a sub-network that is 'higher' on the page than the one in which the feature [near] occurs. This might lead you to think that this makes it possible for the realisation rule for the feature [near] to fire as soon as it is chosen, on the grounds that the grammar already 'knows' whether the conditional features have or have not been chosen. However, the features that function as 'conditions' could equally well occur in a part of the network to be traversed later, so that we cannot proceed on this assumption.* 

* Neither approach would be acceptable in the Sydney Grammar, however, because there is a strong insistence on the concept that, in principle, all systems are entered simultaneously. If this is the case, the grammar would not know whether a possible conditional feature had or had not been co-selected at the time when the feature [near] was chosen. (The computer implementation of the Cardiff Grammar currently operates on the assumption that the 'higher' system networks are traversed before the lower ones, but they could be reformulated if it ever became possible to apply the computational concept of 'parallel processing' to system networks.)

Blogger Comments:

[1] Figures 1 and 2 of Appendix A:

To be clear, this network is not consistent with the principles of SFL theory.  For example, the network

• confuses Thing with Deictic + Thing in a nominal group,
• misunderstands delicacy, in that it presents examples of mass nouns as more delicate features of the feature [mass], and examples of count nouns as more delicate features of the feature [count], and
• presents grammatical classes (mass, count) of nouns as semantic features.

That is, Fawcett's argument that realisation rules cannot be located in system networks only applies to misunderstandings of system networks, such as this devised by Fawcett himself.  As such, Fawcett's argument  here is merely another deployment of the Straw Man fallacy.

A system network, featuring realisation statements, that is consistent with the principles of SFL, from Halliday & Matthiessen (2014: 366), is presented below for comparison:



[2] As previously explained, Fawcett misunderstands system networks as flowcharts, and it is this misunderstanding that leads him to be concerned with the temporal order of feature selection.  To be clear, system networks are networks of relations.  It is the instantiation process that unfolds in time, during logogenesis, not traversals of system networks.

Fawcett's Problem With Realisation Rules In System Networks

Fawcett (2010: 66): 

The initially attractive idea that this is intended to represent is that each feature in a system network contributes to the structure that is being built, and that each such rule should 'fire' as soon as its feature is chosen. Representing the realisation rules in this way, then, fits in nicely with the idea that the lexicogrammar is simply all at one level of language — and this is precisely the concept that is required in Halliday's second approach to meaning. 

Ultimately, however, this approach is unworkable. The problem with it is that it depends on the concept that there are no exceptions to the 'typical' effect of choosing a given feature. But if the 'firing' of the realisation rule is dependent upon the co-selection of another feature (as is often the case), it cannot be allowed to fire as soon as the feature is chosen, because a 'conditional feature' may also be selected in another part of the network which might demand that the realisation should be different.

Blogger Comments:

[1] To be clear, in SFL theory, the system of a grammatical rank is realised by the structure of a rank unit.  This is the axial dimension: the relation between the paradigmatic and syntagmatic axes.

On the other hand, the selection of a feature entails the activation of its realisation statements.  This is the dimension of instantiation: the relation between potential and instance.

[2] As previously explained, Halliday has only had one "approach to meaning" in devising SFL theory.  It merely suits Fawcett's argument to believe otherwise.

[3] To be clear, the problems that Fawcett identifies here only arise if there are problems with (the wiring and/or features of) the system network itself.  It will be seen in the next post that Fawcett exemplifies "the problem", not by critiquing one of Halliday's systems, but by reference to one of his own (problematic) systems.

The Impression That Realisation Rules Are Part Of The System Network Itself

Fawcett (2010: 65-6):

However, in some of Halliday's other writings, starting very soon after that time, he began using a representation of the system networks and realisation rules that can be interpreted in a very different manner — though always in introductory grammars or fragments of grammars, as in Halliday (1971/73a:40 and 1977/78:208-22). The same pattern is found in his 1964 networks (which were published in Halliday (1976:135).* And the same pattern is found in his recent work e.g., in the network for the 'verbal group' in Halliday (1996:11). 

Finally, this way of representing the realisation rules is used throughout Matthiessen (1995). In the diagrams in all of these works, any feature for which there is a realisation rule has the rule written in immediately under the feature itself, almost as a footnote on the feature. In other words, the impression is given that the realisation rules are part of the system network itself. 

* These system networks may well date from the time before Halliday realised that they should be regarded as modelling the 'meaning potential' of the language. 

Blogger Comments:

[1] To be clear, in SFL theory, realisation statements are located in system networks.  This is why the publications of Halliday and Matthiessen present them as such.

[2] As previously explained, Fawcett misunderstands Halliday's 'meaning potential', language as system, to mean the stratum of semantics.

Misrepresenting Halliday On The Separation Of System And Realisation Rules

Fawcett (2010: 65):

There was no such problem with Halliday's early systemic functional grammars (e.g., 1969/81 and 1970/76b). Each contains two components: (1) the system network and (2) the realisation rules — very much as in Figure 4 in Chapter 3. In Halliday (1969/81), for example, the system network is shown on page 141, and the 'realisation statements' that convert any selection expression that is chosen in traversing the network into a structure are set out in a table on page 142. Thus the two figures illustrate each of the two components of the model. And the same pattern is found in Halliday (1970/76b) — with some minor changes in the detail of the realisation rules, as is to be expected at this early stage in the development of generative systemic functional grammars. In other words, the components of these grammars and their outputs correspond directly to the two components and two outputs shown in Figure 4 of Section 3.2 of Chapter 3.

Blogger Comments:

This is very misleading indeed.  Here Fawcett refers to the separate display of system and realisation rules in early Halliday publications and misrepresents this visual separation as consistent with the theoretical separation of them his own model (Figure 4), which, unlike Halliday, misconstrues systems and their realisation rules as different levels of symbolic abstraction (meaning and form).