340 Dik Bakker and Anna Siewierska
tions of the Moves and (certain aspects of?) their expression. Interestingly,
in FDG the Content of the respective Acts is not stored in the Communica-
tive Context, but only influenced by it. Furthermore, there is a link from
this short-term device to the long-term Cognitive module. This feedback
loop accounts for the speaker’s monitoring function. A second point that
gives FDG the flavour of a speaker model is that it takes the Move as its
highest entity of processing, which brings discourse notions into the arena.
The feedback loop via working memory adds to this discourse orientation
by making aspects of previous utterances available to current or following
ones. This caters for problems such as long-distance anaphoric and
metalinguistic reference. Thirdly, in FDG a distinction is made between the
pragmatic and semantic modules. The output of the former partially feeds
into the latter, but both modules have their own, in principle parallel con-
tribution to make to the final output via independent access to the
expression rules. This parallelism introduces the possibility of testing the
model with respect to its performance characteristics.
Assuming, then, that FDG should be perceived as a model of the
speaker rather than of the grammar, we may make some further observa-
tions which in our view should be taken into consideration for future
developments of FDG.
Firstly, there is the role of the Communicative Context. In Hengeveld
(this volume: 3) this is supposed to be “the (short-term) linguistic informa-
tion derivable from the preceding discourse and the non-linguistic,
perceptual information derivable from the speech situation”. The arrows in
Hengeveld's Figure 6 make it clear that the information currently at the
speaker’s IL and EL levels is included in this information. If we are right in
assuming that ‘short-term’ in the quote above should be read as ‘contained
in short-term memory’, then the point is that this information can only be
very restricted, i.e. correspond to a few seconds of the verbal interaction at
most.^13 It is not likely that this will always be the complete intended Move
at hand, including its Central Act and possible subordinate Acts as sug-
gested in Figure 6. The simple reason is that full recursion must be
assumed for Acts since they will get semantic content via referential and
ascriptive acts which in their turn are expressed by recursive linguistic enti-
ties such as NPs, PPs and Ss. Since there seems to be no upper limit to the
length and complexity of a Move, the information contained at the IL and
EL levels, which are thought to feed into the Communicative Context, will
regularly exceed the maximum capacity of short-term memory.
We think that it may in fact be more realistic to assume that what is
contained in the IL, RL and EL levels is actually in working memory, at