Towards a speaker model of FG 359
adequate if it generates precisely the set of well-formed expressions of L.
And a theory of grammar is descriptively adequate if it allows for all and
only the forms and structures that may occur in human language as such.
- This is just an example of a possible tree expansion. We have no intention
here to develop an optimal constituency theory for English or FG for that mat-
ter. That will be a separate and very elaborate exercise, which should be done
on the basis of an in-depth formal morphosyntactic analysis of a typologically
well-founded sample of languages. Much can be gained from the insights of
other functionally based theories which have developed a morphosyntax, such
as Role and Reference Grammar (Van Valin and LaPolla 1999).
- Certain types of speech errors make it clear that phonological information
must be available at a relatively early stage of the production of an utter-
ance, such as the frequently observed reversal of initial consonant clusters
between two words, either accidental or on purpose (‘spoonerisms’).
- It would be a challenge to implement intriguing examples such as (1) and
(2) below (from Nespor and Vogel 1986: 238), where the presence or ab-
sence of flapping in American English depends on the strength of the
semantic or pragmatic link between the two sentences: Turn up the heat. I
am freezing (... hea[s]) vs. Turn up the heat. I am Frances (*... hea[s]).
Technically speaking, ours is by far not the only way the strategy envisaged
here might be implemented. However, in this form we follow models in the
psycholinguistic literature, such as Kempen and Hoenkamp (1987) and
Levelt (1989), thereby adding to the cognitive adequacy of the FG model.
- For a fully worked out implementation of the dynamic model, a feature-
value notation will be assumed including a set of logical operators and a
unification mechanism. Such a formal system is necessary to make the rep-
resentations explicit and unambiguous, and to define the operations on them
in a straightforward way. A proposal for such a system may be found in
Bakker (1989; 1994: 263–296). Since these are mainly technicalities which
may obscure the current discussion we will stick to a more informal version
here. In the representation of nodes we will leave out the fields that have no
value.
- Typically, the models for speaker and hearer are sketched more or less in-
dependently, although they use the same linguistic and non-linguistic
knowledge resources. However, since speakers obviously hear and monitor
their own output a realistic model of the speaker should at least contain a
feedback loop to the hearer model.
- Hesp (1990) is a critical discussion of the MNLU proposal from the per-
spective of the psychology of language. The author shows that the MNLU
model as introduced in Dik (1989) cannot work, and is for a number of rea-
sons not in harmony with the models and experiments found in psycho-
logical literature. See our discussion below and Nuyts (this volume).
