We adopt Zwicky's (1985) use of the term realization rule to refer to rules which map morphosyntactic specifications into morphological forms. Like him, we distinguish two kinds of realization rule: (i) rules of exponence which map directly to concrete morphological forms, and (ii) rules of referral (or syncretism) which equate the form(s) associated with one morphosyntactic class with the form(s) associated with another such class. And, for us also, ``all realization rules are treated as expressing defaults, which are automatically overridden by more specific rules (and these in turn by still more specific rules, and so on).'' (1985, 372). All these points can be conveniently illustrated by our treatment of the German weak declension: what we want to say is that the default form is -en, that the nominative singular coda is null, that accusative forms default to their nominative counterparts, but that the accusative singular masculine form employs the general default form (i.e., -en). We can say this as follows:
Weak:
<> == Suffix_en
<phn coda nom sing> == Null
<phn syll acc> == "<phn syll nom>"
<phn syll acc sing masc> == Suffix_en.
The definition for this declension node consists of just four realization
rules
. The first, and most general,
is a (default) rule of exponence that maps any
morphosyntactic
specification into an -en suffix, inherited from the node for
that suffix. This is then partly overridden by the second, more
specific, realization rule which is a rule of exponence that has the
ultimate effect of mapping nominative singulars to the -e form.
It does this by overriding the definition of the coda which will
otherwise be inherited from Suffix_en (as is the peak). The
first rule is also overridden by the third rule which is a rule of
referral that says that accusatives are to be realized in the same
way as the corresponding nominatives. As Zwicky points out, ``we ..
want to say that there is a nom/acc identity, period'' (1985, 377).
It is exactly this major generalization that our third rule captures,
with the other attributes indicating that we are talking about the
phonological realisation of the form (phn) and, more
specifically, about the syllable structure (syll). This rule
is, in turn, partly overridden by the highly specific fourth and
final realization rule.
There are twenty four distinct relevant morphosyntactic
specifications in the domain of the Weak node and these four
rules map all twenty four to -e or -en, as appropriate
(see Section 6, below).
The three German adjective declensions are commonly called weak, mixed and strong. The weak declension endings apply to adjectives occurring after one of a set of articles including the definite article; the mixed declension endings apply to adjectives occurring after one of another set of articles including the indefinite article; and the strong endings apply to certain articles as well as to adjectives which appear on their own, without an article. The twenty four mixed endings are largely identical to the weak endings with a few different forms. What we want to say is that the mixed declension is just like the weak except for a couple of nominative singular forms (and an accusative). We express this as follows, with just three DATR equations:
Mixed:
<> == Weak
<phn coda nom sing masc> == r
<phn coda nom sing neut> == s.
The first equation is a very general rule of referral that simply
says that, by default, the rules that govern the mixed declension are
the same as those that govern the weak declension. And the second
and third are rules of exponence for two of the three possible
nominative singular forms. These two very specific rules take
precedence over the first rule, of course. Taken together with the
rules already provided in the Weak node, these three rules
suffice to map the twenty four distinct Mixed morphosyntactic
specifications to -e, -en, -er, or -es, as
appropriate. Note that nothing explicit has to be said here about
the realization of mixed accusative forms - they follow from the
interaction of the rules already given. Specifically, the third equation in
the Weak node which defines the whole syllable of accusatives
as being the same by default as the nominative, results in the new coda
definitions for the nominative providing the same values to the accusative
forms.
The (first) strong declension is not wholly dissimilar to the mixed declension but a fifth suffix appears (-em) and there is a significant amount of additional syncretism, as expressed in the rules given for the Strong_1 node, below:
Strong_1:
<> == Mixed
<phn coda dat sing masc> == m
<phn syll gen sing femn> == <phn syll nom sing masc>
<phn syll gen sing neut> == "<phn syll gen sing masc>"
<phn syll dat sing femn> == "<phn syll gen sing femn>"
<phn syll dat sing neut> == "<phn syll dat sing masc>"
<phn syll nom plur> == "<phn syll nom sing femn>"
<phn syll gen plur> == "<phn syll gen sing femn>".
Notice in particular how the nominative and genitive plural forms are determined by the corresponding feminine singulars.
The strong declension shown above is directly responsible for the strong inflection of adjectives and a small class of determiners. But there are two further strong declensions (which Zwicky (1986, 960) calls S-ES and S-ES-Z) which determine the inflection of major classes of determiners. The first of these is just like declension Strong_1, above, except that a couple of genitive singular forms appear as -es. We can define the declension (which we will call Strong_2) as follows:
Strong_2:
<> == Strong_1
<phn coda gen sing masc> == s.
Although this definition does not say anything explicit about the
genitive singular neuter form, it is, in fact, implicitly defined to
be the same as the genitive singular masculine form (thus -es)
in virtue of one of the syncretic equations given in our treatment of
the Strong_1 declension.
The final strong declension is just like declension Strong_2, above, except that a couple of nominative singular forms (and an accusative) are null. We can define the declension (which we will call Strong_3) as follows:
Strong_3:
<> == Strong_2
<phn syll nom sing masc> == Null
<phn syll nom sing neut> == Null.
As previously, nothing explicit needs to be said about the
corresponding accusative singulars since their forms are correctly
determined by the rules of syncretism given
in our treatment of the Weak
declension.
