Encoding word order by making reference to a separate feature has the advantage that it is possible to formulate word order constraints, while at the same time lexical entries can specify an exceptional word order. As will be explained in section 6, such exceptions do exist.
The default values for DIR (for verbal heads) are right for verbal complements and left for all other complements. That is, verbal complements normally occur right of the head, whereas NP's and other non-verbal complements occur left of the head.
The value of SUBCAT is an ordered list, and thus, the order of complements on SUBCAT is significant. We assume that complements are ordered in terms of ``obliqueness'', with the least oblique elements occurring first on the list, and with the most oblique elements occurring last. This notion of obliqueness is important for our account of word order, since we assume that word order in Dutch is determined by the obliqueness ordering. In particular, we require that more oblique complements appear closer to the head.
In order to implement this observation we assume the following rule, which is an instantiation of the HEAD-COMPLEMENT schema (12).
![\enumsentence{
\parbox[t]{10cm}{
{\sc Rule 1}
Rule~1 is an instance of the head-...
...s $r_2$\ iff $r_1$occurs after $r_2$\ on the {\sc subcat} list of the head.
}}](img15.png)
Note that the word-order in (14) is the only
possible word-order that can be derived by this rule.
Thus, the obligatory cross-serial dependency word-order is obtained by assigning a flat structure (by the requirement that the verbal argument of a verb-raiser is lexical); furthermore all arguments of a head are ordered in terms of obliqueness: the more oblique elements occur closer to the head.
In section 6 we discuss a number of exceptions to the word-order generalizations given above, and we show how the analysis can be extended to treat these exceptions. In [2] our analysis of word-order in verb-clustering constructions is extended and applied to German as well.