The Two Processes Of Generation And Understanding

Fawcett (2010: 104):

The reason why the two processes of generation and understanding cannot be treated as mirror images of each other is that each of the two processes of generation and understanding involves a different kind of 'problem-solving'. This arises because the two processes operate in different directions. More specifically, the evidence that is taken into account when deciding to make one analysis rather than another when parsing the syntax of a text is different from the evidence that is drawn on to determine the choices in generation. In parsing, the available linguistic data are those that are observable in the surrounding text at the level of form; in generation in the decisions are taken at the level of meaning (or in a higher component), so that the data that affect choices are at the level of semantics (or above it).

Blogger Comments:

[1] To be clear, here Fawcett is not concerned with modelling language, but with the problems that a computational linguist needs to solve when trying to use a theory of language for the purposes of text generation and parsing.  Moreover, the theory of language being used is his own 'flowchart' model ('syntax', 'level of form', 'level of semantics'), which, as previously demonstrated, is inconsistent with the architecture of SFL Theory.

[2] To be clear, when applied to language, rather than computation, this is at odds with SFL theory, where structural interpretations are made on the basis of systemic contrasts. Halliday & Matthiessen (2014: 49):

Giving priority to the view ‘from above’ means that the organising principle adopted is that of system: the grammar is seen as a network of interrelated meaningful choices. In other words, the dominant axis is the paradigmatic one: the fundamental components of the grammar are sets of mutually defining contrastive features. Explaining something consists not in stating how it is structured but in showing how it is related to other things: its pattern of systemic relationships, or agnateness (agnation).

[3] To be clear, this confuses linguistic data with linguistic theory. Linguistic data are what is to be modelled, whereas 'level of form' and 'level of meaning' are dimensions of Fawcett's theoretical model (Figure 4).

In terms of SFL Theory, lexicogrammar is a purely abstract level of representation. Halliday & Matthiessen (1999: 613-4):

For these to be possible you need a semiotic of a different kind, one that allows for a purely abstract level of representation "in between" the two faces of the sign, To put this another way (as we did at the beginning of the book), the sign has to be deconstructed so that, instead of content interfacing directly with expression, the relationship is mediated by a systematic organisation of form (a lexicogrammar).