Inflectional paradigms in a given language typically resemble each other in all but a few characteristics. What is needed is a natural way of saying ``this paradigm is just like this paradigm except for this property''. One obvious approach is to use nonmonotonicity and inheritance machinery to capture such matters of family resemblance. Much recent research into the design of representation languages for natural language lexicons has made use of nonmonotonic inheritance networks (or ``semantic nets'') as originally developed for more general representation purposes in Artificial Intelligence (AI). Daelemans et al. (1992) provide a rationale for, and an introduction to, this body of research and we will not rehearse the content of that paper here, nor review the work cited there. Daelemans & Gazdar (1992) and Briscoe et al. (1993) are collections that bring together much recent work on the application of inheritance networks to lexical description.
Recent published work on German inflection has mostly been concerned
with nouns and verbs rather than the word classes considered
here
. Like the descriptive analyses of
German adjective, determiner and pronoun inflection to be found in
traditional grammars, our primary concern is with the obvious
similarities between the inflectional paradigms of these categories.
But unlike these traditional accounts, we provide an axiomatisation
from which the similarities follow.
The theory of German determiner and adjective inflection presented here is an elaboration and development of that proposed by Zwicky (1985). However, we presuppose a lexicon structured as a nonmonotonic inheritance hierarchy, in which information common to most or all of a set of lexical entries is defined at a shared higher point in the hierarchy. This information is then inherited by default by all the entries below it, but with subregular classes and specific lexical exceptions blocking this inheritance lower in the hierarchy. Thus, for example, the assumptions we make about phonological and morphological structure are all encoded towards the top of the hierarchy with specific details (such as vowel choice) filled in at lower points.
We assume that lexemes, not word forms, are the basic unit of lexical
organization. This distinguishes our approach to inflection from
that taken in such works as Karttunen (1984) and Blevins (1995). For
them, the various word forms themselves are what the analysis is
about. Their puzzle is how best to describe the morphosyntactic
properties of each phonologically distinct word form in a paradigm,
either by invoking disjunction and negation (Karttunen) or by
elaborating a type system in which most forms find a single place
(Blevins). For us, pairs consisting of a morphosyntactic feature
specification and a phonologically specified word form are just
properties, generally implicit ones, of lexemes. Such properties are
on a par with all the other the properties of lexemes, syntactic,
semantic, and so on. The various phoneme sequences that correspond
to distinct word forms have no particular ontological status in our
approach. However, given such a sequence as a starting point, our
approach will implicitly define the set of morphosyntactic feature
specifications that map into it. For certain natural language
processing purposes, such as parsing and word form tagging, it may be
desirable for such sets to receive compact descriptions, but this is not our concern here.
Our general approach to inflectional morphology falls within the
tradition that treats paradigms as analytically central rather than epiphenomenal. As regards current work, it is
closely related to Corbett & Fraser's Network
Morphology and, somewhat less closely, to Stump's Paradigm
Function Morphology. The present paper is
entirely about inflectional morphology but, in our approach, abstract
inflectional rules are typically stated in terms of phonological
units, most commonly the syllable and the segment (as in Cahill 1990a,
1990b, 1993). Gibbon and his collaborators in the ILEX (Integrated
Lexicon with EXceptions) project at Bielefeld have pioneered the use of default inheritance
hierarchies for the representation of lexical phonology and
morphophonology. Our work is thus also indebted to theirs.
The relation between morphology and phonology is an intimate one,
both synchronically and diachronically. Allomorphic variation of
affixes is frequently determined by phonological context, and
affixation itself often imposes phonological requirements. There are
also non-affixational morphological relations, such as umlaut, whose
origins are purely phonological. Thus linguistic phenomena which
belong to one domain at one time, may belong to the other at a later
time. The traditional approach to such phenomena is to maintain a
distinction between two levels of description, with morphosyntactic
features triggering morphological processes, such as affixation, and
with phonological processes following. However, recently,
declarative and constraint-based approaches to phonology have begun
to appear, as in Bird and Klein (1994). Despite some differences of
representation, the current account is an instance of this class of
approach. Although we make a distinction between phonological
structure and morphotactics, and between attributes in the
phonological domain and those in the morphological domain, we do not
embrace the traditional notion of level of description, nor its
concomitant notion of rule type (or process) mapping from one level
to another. For us, the linguistic description is just a set of
simultaneously applicable constraints. Such constraints may, for
example, directly connect morphosyntactic attributes to the
individual phonological components of word forms. The explicit
definition of elements of the phonological structure by reference to
morphosyntactic features is fundamental to our account, and is
described in more detail in section 3. Kloeke (1982) provides an
account of German inflection in the Generative Phonology tradition. His
account (part of an analysis of German phonology) also makes a direct
link between morphosyntactic features and phonological realization.
The question of how morphosyntactic features are distributed within the syntax of German is not our concern here. Thus questions of congruence, such as the way in which adjective inflection is determined by the determiner with which it appears, are not discussed, our view being that this is a matter of syntax. What we account for here are the patterns of inflection that apply once morphosyntactic features are determined.
The theory of German inflection presented in the following sections of this paper demonstrates that a default inheritance based approach to the phonological expression of morphosyntactic information can provide a concise and yet formally explicit definition of linguistic facts which are apparently very complex. For example, the definition of the German adjective mixed declension, which involves four distinct suffixes distributed among twenty-four distinct morphosyntactic feature sets, is described with just three simple equations: one says that the mixed declension is, ceteris paribus, identical to the weak declension, and the other two define individual phonological elements, in this case the codas, of the two suffixes which are different from those in the relevant forms in the weak declension.
