Towards a speaker model of FG 331essarily in their final form. Although it is not clear to us at what stage of
the expression process phonological details of the respective lexical and
grammatical entities are available, we will nevertheless give our terminal
elements a (pseudo)phonological shape indicated by forward slashes, but
will furthermore remain agnostic about the phonology.^6 In underlying
clauses we will follow the practice of assuming that a more abstract repre-
sentation of predicates is present, indicated by the uses of single quotation
marks.
We will organize this process of dynamic tree construction by applying
the following principles. We would like to stress that the model almost
completely ignores the phonological level. Far from assuming that the
phonological module may simply be stuck on as a separate component to
the model presented here, we nevertheless think that the phonology might
be interleaved with the morphosyntax in the same fashion in which we in-
terleave syntax and morphology below.^7
Principle 1: Constituent structures are developed top-down from the
material in the underlying representation. Thus, first the higher syntactic
constituents, such as noun phrases, are developed, then the individual free
morphemes, next the bound morphemes, and finally their phonological
form. In this way, and in contrast to the standard model, the explanatory
hierarchy of FG as discussed in Dik (1986) is respected, which runs from
syntax to morphology to phonology.
Principle 2: Development takes place from left to right. This is the
‘natural’ order in which linguistic forms are uttered in the first place. Left-
to-right ordering may be expected to shape and have shaped language over
time, at least to some extent, and more than any other order.
Principle 3: Development works depth-first. This means that of any two
contiguous elements to be expressed, at whatever syntactic or morphologi-
cal level, the leftmost one will be completely expanded up to its terminal
forms before the rightmost one is considered, and this applies recursively.
An implication of this is that only a fraction of the complete information
which is necessary for the production of the whole utterance will be avail-
able at any one time. This considerably reduces the problem for short-term
memory which is inherent to breadth-first development. In combination
with Principle 2, this may solve a number of problems of the type illus-
trated in example (1) above.
Principle 4: For any node N in the tree, all features found on a direct
path from that node to the top node are available for inheritance by N, at
least in principle. Trivially, this implies that overt primary operators are
available for every node in the tree. However, a distinction will be made