Towards a speaker model of FG 345then projection from IL to RL would in fact be trivial: the R and T vari-
ables would simply be linking devices between UC-like mental
representations and newly made underlying clauses, via discourse variables
of the right type.
However, such an extreme relativistic position seems to be untenable.
There are a number of arguments, both logical and experimental in nature,
which point towards a non-linguistic, abstract representation of knowledge.
We will mention but a few. If language were to be the code in which (con-
ceptual) knowledge was stored, it would not be clear how knowledge about
language itself (or rather, a specific language) could be acquired by the
pre-linguistic brain. There must be some built-in mode of operation, neces-
sarily prelinguistic, with which to bootstrap the brain to thinking and
knowledge acquisition. There must therefore be some kind of thinking be-
fore language, and indeed without language altogether, since the
congenitally deaf do think, and not in a dramatically different way from the
non-deaf (cf. Furth 1966). Furthermore, there is no evidence for differences
in the way conceptual knowledge on the one hand and knowledge acquired
through perception on the other are stored and processed in the mind. The
same goes for declarative knowledge versus procedural knowledge. For
most types of knowledge that we possess, we do not remember how it was
acquired, nor can we determine by ourselves how it is internally repre-
sented. Furthermore, we seem to reason and to process it all in much the
same way. This all points in the direction of a unified form of representa-
tion, both for the knowledge that is linguistically expressible, like most
declarative knowledge, and knowledge that is not, or only partially so, like
procedural and perceptual knowledge. For further arguments we refer to
Clark and Clark (1977) and Fodor (1983) and references therein.
In what follows, our assumption will be that the planning of utterances
is largely pre- and indeed non-linguistic, i.e. it proceeds in terms of ab-
stract, language-independent concepts. This means that the mapping from
IL to RL involves translation from such abstract concepts to predicates of
the language concerned. The crucial link for this translation process lies in
the meaning definitions of the predicates in the lexicon. We will assume
that these are constructed on the basis of such abstract concepts, possibly
mixed with language-specific elements. Arguably, the ongoing discourse –
the complete Communicative Context – as represented in semantic mem-
ory is coded in the same, more or less language-independent fashion.^16
This concludes our critical discussion of the more general aspects of the
FDG model. In the next section we will see how FDG, with the provisions
made above, may be linked to our dynamic model for the expression rules.