Towards a speaker model of FG 339model is the ultimate goal of the enterprise. We will not assume that this
should be a full psychological model of the speaker; this might better be
constructed from the perspective of psychological modelling, and by a psy-
chologist. The FG version would be a linguistic model of the speaker
which is as much as possible in harmony with psycholinguistics.
Just like other generative models of grammar, FDG (Hengeveld this
volume) shows how complete (‘fully specified’) underlying representations
are first constructed and then fed to the expression rules which generate the
corresponding forms. Layered formal representations are introduced for the
respective intermediate structures, which are an extension of the traditional
underlying clause formalism first introduced into FG in Hengeveld (1989).
A novelty vis-à-vis the standard model is the introduction of hierarchical
tree-like structures suggested for the output of the expression rules. Most of
the structures at the respective levels are static and ready made, i.e. nothing
is said about how they come into existence. No explicit claims are made as
to their psychological status, i.e. whether they should be seen as mental
representations or just as formalisms. The latter possibility therefore seems
to be the more likely.^11 Only for the representation at the pragmatic level is
an outside-in history claimed: Move goes before Act and this goes before
Communicated Content. Interestingly, though there are modules for Cogni-
tion and Communicative Context in the model, there is a separate box
labelled ‘Grammar’ around the pragmatic (IL), semantic (RL) and expres-
sion (EL) levels. This suggests that, apart from the lexicon, which is
contained in Cognition, the grammar gets a separate status outside the lat-
ter, with the inclusion of the projection rules and the expression rules
which link IL, RL and EL. Linguistic competence, however, is located
within Cognition. So it may be safe to assume that the three levels are just
levels of representation, while the rules that create them and link them are
stored in Cognition. Probably the label ‘Grammar’ should then be inter-
preted as ‘created by the grammar’.
On the other hand, FDG has several properties that are suggestive of a
model of the language user, more specifically of the speaker. Firstly, in
contrast to the standard grammar model, which takes the lexicon as a point
of departure for the generation of sentences via the respective sets of rules,
FDG introduces Cognition. This is the complete knowledge store of the
speaker, of which the lexicon and the rest of the grammar are just substruc-
tures. Indeed, Cognition controls the generation process at all stages. It
introduces long-term memory to the model, or rather semantic memory.^12
Short-term memory (or rather working memory) is introduced via the
Communicative Context module. This stores both the formal representa-