338 Dik Bakker and Anna Siewierska
speaker should give a precise representation of the process of language
production, running from the initial, prelinguistic intention to the eventual
utterance, while specifying all relevant intermediate steps in the right order.
This information flow is mainly top-down, as in the generative grammar
model. However, psycholinguistic research strongly suggests that the pro-
duction of an utterance is not a purely linear process but an incremental
one, managed by several more or less independent modules which work in
parallel, at least partially, rather than consecutively. It is above all speech
error data which are suggestive of this parallelism, as well as the occur-
rence of hesitations, pauses, repetitions, etc. A model of the speaker should
cater for the extremely complex data flow during such a process as well as
for the ‘performance’ phenomena related to it. Finally, this kind of model
should include a representation of the speaker’s knowledge about the con-
versational partner(s). Hearers mainly use the same knowledge resources as
speakers do. However, the procedural aspects of a model of the hearer are
rather different from those of speaking. The hearing mode interleaves bot-
tom-up processes with top-down strategies, such as the construction of
cohorts of phonologically similar strings, or the priming of semantically
related entities during recognition. Another example of this is the choice of
the most promising syntactic structure at an early stage of speech analysis
via probabilistic devices. In the (rare) case of erroneous intermediate
choices, backtracking mechanisms are employed. One of the differences
between speaker and hearer that the respective models should make clear is
the often observed gap between a language user’s passive knowledge and
active use of a language. Eventually, both models should make clear how
first and second language acquisition take place.^9
Hengeveld contrasts the FDG model sharply with the grammar model in
Dik (1997). The latter is seen – mistakenly, we think – as a speech produc-
tion model which lacks psychological adequacy, and therefore runs counter
to the adequacy requirements of the theory (cf. Dik 1997: 13f.). Dik (1989)
makes it clear, however that an FG model of the speaker necessarily has
other properties than the standard model of the theory and then sketches
such a Model of the Natural Language User.^10 Hengeveld (this volume) ob-
serves that FDG is an improvement on the FG grammar model precisely
with respect to psychological adequacy. This, and the fact that the model
could be described in terms of the decisions a speaker makes during lan-
guage production, is highly indicative of the fact that FDG should be seen
as a model of the speaker rather than of the grammar. However, when we
take a closer look, it seems to have aspects of both models. We will briefly
mention these, and then assume for the rest of the discussion that a speaker