328 Dik Bakker and Anna Siewierska
(2) a. Jij speel-t goed vandaag!
you play-2SG well today
‘You are playing well today!’
b. Vandaag speel je goed!
today play-ø you well
‘Today you are playing well!’
c. Ik zie dat jij vandaag goed speel-t!
I see that you today well play-2SG
‘I see that you are playing well today!’
In the (a) examples, with SV order, we find a 2nd person singular subject
agreement marker on the verb. In the (b) examples, with VS order, there is
no such marker; the verb appears as a bare stem. In the complement clause
of (2c), with its more or less fixed SOV order, the agreement marker is al-
ways expressed. If the expression rules first generate the grammatical
forms and only then establish linear order, as in the standard model, there
is no way to determine whether the suffix should be generated or not. The
only solution would be to look at the factors that determine the eventual
constituent order, in this case pragmatic aspects of the subject constituent,
the availability of another potential P1 filler, and the level of embedding.
But this would imply ‘prerunning’ the ordering rules in some way or
other.^3
A second problem that we encountered with the expression rules in their
current form is that they overgenerate, i.e. they will produce all kinds of
forms that actually do not occur in any existing language. At first sight,
overgeneration seems to be a less serious problem than undergeneration:
we may simply assume that the theory will prevent such ‘impossible’ ex-
pressions from arising at all because they express ‘impossible’ underlying
representations. However, under the assumption that there is a niche in the
grammar which allows for a certain amount of autonomy within morpho-
syntax there might indeed exist certain (universal or typological)
constraints on what can be a well-formed expression in languages which
are not directly reducible to functional criteria (cf. Croft 1995). A linguistic
theory should make these constraints explicit, much as it should be descrip-
tively adequate in the more obvious sense of generating the forms that do
occur.^4 But overgeneration is not only a theoretical issue. An overgenerat-
ing expression component puts into question the learnability of a FG
grammar, and as such endangers the cognitive adequacy of the theory.
In Bakker (2001) a model is proposed for the expression component
that tries to repair these shortcomings. In correspondence with the standard