372 How to describe Inflection* Arnold i1. Zt~ickv Ohio State University 1. An overvieN. The aim of this paper is to sketch a framework for describing systems of inflectional morphologv. In 50 doing I shall be making many implicit claims about the nature of language -not merely about a convenient formalism for statinq generalizations and listing idiosyncratic facts -but the focus ~f mv presentation here will be Simply to show how the framework can be applied to some reasonably complex arrays of inflectional forms. My intellectual debts in this enterprise are considerable. First, to the work of Peter i1atthew5 (j965, 1972), Stephen Anderson (1977, 1982), F:ich Janda (198.3), and others who have resuscitated process morphology within modern approaches to formal grHmar. Second, to Andrew Carstairs (1981. 1983), Wolfgang t~urzel ito appear), Fred Karlsson (to appear), and other writers who have stressed that paradtgms are not mere arravs of forms, but have internal organization of considerable interest. Third. to Paul Kiparsky (1982a, bl and other proponents of 'lexical morphoiogv and phonology, whose work (along wi th Anderson '5) raises urgent questions about the relationships among syntax, Inflectional morphologv. derivational morohology, phonology, and the le:dcon. My focus is on what have been called, by various writers, rules of aiiomOfQhy Dr morpholexical rules, insofar as they concern inflecti~n~l morph~logy. Since th~ terminology in these matters is somewhat confused, with different writers using these two technical terms in distinct ways, I have opted for the term realization rules to refer to principles describing when and how morphosyntacticfeatures are realized as morphological processes. These principles belong In a morphological component [IJ, which follows a syntactic component, in which (among other things) morphosyntactic features are located within syntactic structures, and precedes a phonological component, in which (among other things) some morphophonemic alternations are accounted for by rules altering phonological representations. There are, I claim, two types of realization rules. First there are rules of exponence, describing how certain combinations of morphosyntactic features are realized, in the context of certain other bundles, 25 morphophonological operations. The following principle of English is a typical rule of exponence: In the context of [CAT:verbJ, [VFORM:pastJ is realized by the suffixatIon of /d/. The~ there are rules of referral, stipulating that certain combinations of features have the same realization as certain others. The following principle of English is a typical rule of referral: In the context of [CAT:verbJ, [VFORM:pastprt] has the same realization as [VFORM:pastJ. 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 50 on). [2J The framework distinquishes features, such as CASE and PERSON, from clusters of ~alues that feat~res can take. such as direct versus oblique case, or second versus nonsecond·person, And it permits reference to feature clusters, such as [ASE-GE~DER-NUMBEF:. ... ~ .. Finally, I assume not only a set of realization rules. but also an ordered set of (abstract! slots for inflectional . material. Anv particular rule supplies material for a specified slot or slots, and several distinct rules can supply material for the same slot. The ordering of a rule with respect to others is then governed by the ordering of slots. fhe main features of the framework can now be listed: (a) rules of referral, as well as rules of exponence; (bl extensive use of default settings; Icl a distinction between features. value ~lusters, and feature clusters; and (d) a conceptual separation of rules and slots. In the remainder of this paper I will develop these proposals in some deta11, concentrating on (a) and (b) and illustrating the proposals with a description of a SUbstantial portion of the declensional system of standard German. 2. Fundamental assumptions and conventIons. I presuppose some analysis of the units of synt~x and morphology (constituent types, word classes, base classes) as combinations of feature values, or as bundles, as I shall call them. For the mDment I make the simplifving assumption that a bundle is simply an (unordered) set of (ordered) pairs, each pair associating to some feature one of its values (or, in some cases, a d1sjunction of several of its vaJ ues). Typographical conventions: names of features are in upper case (GENDi; names of values are in IOI'/er Cd.se (fem); a disjunction of values is indicated by a slash between the value names (nom/acc): a pairing of feature and value is indicated by appending the value name to the feature name, with a colon separating them (GEND:fem); names of pairs are combined by means of a separating comma and space (GEND:fem, CASE:nom/acc); and names of bundles have flanking square brackets ([GEND:fem, CASE:nom/acc]l. It lS often convenient to refer to linguistic forms bv means of the values that are realized in them -for instance, to refer to forms In which the values CASE:nom, GEND:fem. and NUH:sg are realized as being in (or ofl the nom fem sg, or simplv as being nom fem sg. The primary tool in describing inflectional systems is the rule of exponence, the function of which is to realize some bundle, in the context of some other bundle. as a morphophonological operation Dr operations. [3] In German, for inst3nc~, the bundle [CASE:nom, GEND:masc, NUM:sgl, in the context of the feature values picking out the "strong' declension of determiners and adjectIves (which I will suppose for the moment is the bundle [CAT:det/adj, CLASS:str]1 I is realized by the suffixation of -er to a base. as in dies-er Hann 'th1S man' and ein alt-er Hann 'an old man'. 3. SYncretism. I begin the main exposit10n with the observation (stressed by CarstairsJ that syncretism is very common in inflectional paradigms. Consider the weak' declension paradigm for German adjectives. given in Table I. Although there are four val Lies for CASE (nom, acc, gen, d t i, three for GENU (rnase, neut, femi, and two for NUrl (sg, pi , making a total of 24 distinct bundles involving t~ese three fea Llres, there are only two distinct forms in the paradigm, -e and -en .374 MASC-5G NEUT-SG FEM-SG PLURAL NOM -e -e -e -en ACe -en -e -e -en GEN -en -en -en -en DAT -en -en -en -en Table