Separationist morphology and lexical insertion
Recall that in the realizational model outlined by Matthews (1972,1991), an analysis is ‘seeded’ by a pair (C, p}, representing a paradigm cell C and a root form p. A similar pair provides the input to paradigm functions in PFM. Both approaches adopt aversion of the ‘Separation Hypothesis’ (Beard 1995) in which the distinctive feature bundles associated with paradigm cells are defined independently of any particular syntactic context, and are realized by word forms that are entered as wholes into syntactic representations.
A different conception of the Separation Hypothesis is developed within AMM. On this alternative, word-level feature bundles are associated with the preterminal nodes of a syntactic tree. Realization rules apply to preterminal bundles and spell out the forms that are associated with the terminal nodes of a tree. Since preterminals are not associated with a lexical root form, a separate process must obtain the stem that seeds an analysis. For this, AMM invokes a realizational analogue of the ‘lexical insertion’ transformation of Chomsky (1965). Anderson (1992) describes the process in the following terms:
“Lexical insertion” involves finding a lexical item consistent with the position to be interpreted—that is, one whose lexical characteristics are not distinct from those of the position in question, and whose subcategorization requirements are met by the containing Phrase Marker. The corresponding phonological form is then “inflected”, through the
Figure 6.27 Lexical insertion in A-Morphous Morphology
operation of inflectional Word Formation Rules, to reflect the morphosyntactic properties of the position. (Anderson 1992:132)
The process proceeds in the three stages illustrated in Figure 6.27. In the first stage, the grammar defines a partial syntactic representation, which terminates in nodes annotated by the fully specified feature bundles B and C. The spell-out of these bundles is initiated by selecting stem entries, (S, x) and (T, y) in Figure 6.27, from the lexicon. The feature bundles S and T are compared with those of the bundles B and C, as illustrated in the second tree in Figure 6.27. If the properties of an entry are ‘nondistinct’ from those of the corresponding nonterminal, and appropriate to the syntactic context in which the nonterminal occurs, then the entry’s form is inserted below the bundle. This step is illustrated in the third tree in Figure 6.27, where the insertion of x andy provides the input to the realization rules that spell out the features in B and C.
The use of a lexical insertion rule is less of a radical departure than it might appear. Like realization, lexical insertion is a purely interpretive process. It is, in effect, just a special, context-sensitive, case of exponence. The nondistinctness check does not modify the properties of a bundle or entry but merely verifies that there is no conflict and then introduces a stem under a preterminal node. Moreover, the choice between lexical and syntactic rule application ultimately comes down to whether one regards inflectional properties as being present ‘in the lexicon, or ‘in the syntax’. This issue represents a free choice within morphological models, particularly given the difficulty in discerning empirical differences between a system in which realization rules apply ‘late’ in the lexicon and a system in which they apply ‘early’ in the syntactic component.
-  As noted by Ingria (1990), a nondistinctness check has many of the effects of subsumptionor unification in enforcing feature compatibility within a local domain, though the effects of thesemechanisms diverge over larger domains (Blevins 2011).
-  The claim that inflectional properties are syntactic (whereas derivational properties are lexical)is often termed the ‘Split Morphology Hypothesis’ (Perlmutter 1988). See also Booij 1993, 1996 andBlevins 2001 for some discussion.