Towards a speaker model of FG 341least partially, with the inclusion of the rules that operate on them. The
complete set of materials necessary for the construction of these entities is
stored more permanently in semantic memory. Alongside the grammar and
the lexicon, this is both the ‘full story’ that is being told and the scripts,
scenarios or mental models that are relevant for it (cf. Schank and Abelson
1977; Johnson-Laird 1983). A central planning mechanism sends coherent
bits of this material to working memory for linguistic expression, where it
is operated upon in three stages, leading to the representations at the IL, RL
and EL levels. In its EL form, it is finally send to the articulator, which
runs it as a set of instructions for the speech apparatus. The feedback loop
from Communicative Context to Cognition would then mean that whatever
information is held in working memory at any one time is fed back to se-
mantic memory for more permanent storage, or at least certain aspects of it.
In this respect it is interesting that there is no feeding arrow from the
semantics of the RL level to the communicative context. Rather, RL gets
part of its input from the Communicative Context, probably anaphoric in-
formation above all. Since Communicative Context is related to short-term
memory only, this must be very local information. This would imply that
no (direct) information about the precise semantics of an utterance finds its
way back to semantic (sic!) memory, but only information about the prag-
matics and the form. This is not a very probable state of affairs. The
extensive literature on recall experiments shows that hearers and speakers
retain the content of what they hear, read or say rather than the form. See
Brandsford and Franks (1971), Schweller et al. (1976) and Johnson-Laird
et al. (1974) for classical experiments that refute the recollection of syntac-
tic structure, precise words and parts of speech, respectively. But even
though it is the meaning of an utterance rather than anything else that is
remembered, it is not necessarily the concrete linguistic meaning in terms
of the meaning representation as given by an Underlying Clause, for in-
stance. Rather, it is a broad interpretation of the meaning, the gist or
intention of what was said. Therefore, speakers will not normally remem-
ber whether they have said (8a) or (8b).
(8) a. The window was open.
b. The window was not closed.
Taking a different perspective on things, one might argue that the con-
tents of what is being said are stored in semantic memory even before
speaking starts, and that there is no need for them to be fed back. This may
be true to some extent for a coherent and structured story or event in the