Íslenskt mál og almenn málfræði - 01.01.1980, Blaðsíða 21

A Note on Local Ordering
19
There is therefore a class of derivations possible under the theory of
linear ordering that is incompatible with the theory of local ordering.
Is the converse of this statement true? In principle, it would seem not:
given the possibility of splitting any rule A into two rules A' and A", it
would seem that the linear order theory can replicate any derivation
that the local order theory permits. Thus in principle, it turns out that
the local order theory is the more constrained of the two.
3.
3.1
Despite our findings in section 2, we have not yet established grounds
for deciding between the two theories. While it is true that the linear
order theory, as construed here, is compatible with a wider class of rule
interaction types than is the theory of local order, it is not the case that
we can decide between the two theories on the grounds of their relative
constrainedness. This is because in the present case, „not all else is
equal“. Both theories form part of a larger theory of generative gram-
mar which provides a criterion of evaluation to select among alternative
analyses of a given set of data. All else being equal, the more highly
valued of two alternative analyses is the one that attains the highest
degree of overall simplicity and generality, that is, the one that sub-
sumes the data in question under the most comprehensive, least redund-
ant set of descriptive statements compatible with the principles of
universal grammar. As we recall, it was this principle that militated
against the duplication of rules in the grammar in the first place. Now
it will be noticed that in terms of this general criterion of evaluation,
there are certain imaginable types of rule interaction that would reduce
the linear ordering theory to near-absurdity. For example, suppose we
find a language with the two very natural rules:
(3) A: V -» 0 / [ + sonorant]-----#
B: [ + nasal]->0/—#
and such characteristic alternating forms as ka beside kanamana + t,
nilika beside nilikani + t, etc., where -t is some suffix. The proponent
of linear ordering would be highly discomfited by such forms, as they
would force him to postulate such derivations as the following: