Morphology

As Aronoff (1976: 4) observes, `Within the generative framework, morphology was for a long time quite successfully ignored. There was a good ideological reason for this: in its zeal, post-Syntactic Structures linguistics saw syntax and phonology everywhere, with the result that morphology was lost somewhere in between.' The inclusion of the traditional substance of morphology within syntax meant that, in the Aspects (Chomsky 1965) model, no distinction was drawn between word-building and sentence-building operations: all distributional regularities were necessarily captured using transformational rules, which derived related surface structures from a common Deep Structure. This methodology, and the large number of surface relations between words and constructions to be accounted for, had two results: the Deep Structures became progressively more remote from these surface representations, and the transformations became more and more complex and unconstrained.

Chomsky's `Remarks on nominalization' (1970) is a first attempt to simplify and reduce the power of the transformational component, at the cost of more complex base rules and an enriched lexicon. The paper focuses on derived nominals, such as criticism, reduction, transmission, recital, although it is clear that these should be regarded as a test-case, and that Chomsky's proposals generalize to all derivational morphology. Chomsky argues that these nominals are unsuited to transformational derivation, since, for example, the processes involved are characteristically unproductive, while the nominals themselves are semantically idiosyncratic. Chomsky concludes that T-rules should be used only to effect fully regular relationships; processes like nominalization, which have lexical exceptions, should instead be handled in the lexicon. In the Aspects model, the lexicon had been seen as simply a repository for idiosyncratic information on lexical items; it was now extended and equipped with lexical rules intended to cope with subregularities. Verbs like criticise, reduce and their derived nominals, criticism and reduction, could then be base-generated, and their lexical entries related using these lexical rules.

Chomsky's (1970) suggestions for the structure of this revised lexicon are extremely sketchy; in retrospect, it is clear that `the significance of ``Remarks'' lies less in what it says itself than in what it caused others to say' (Hoekstra, van der Hulst and Moortgat 1981: 1). The removal of derivational morphology from the scope of the transformations facilitated the reintroduction of morphology as a linguistic subdiscipline separate from phonology and syntax; and the location of morphological processes in the lexicon also gave rise to lexicalist syntaxes (Hoekstra, van der Hulst and Moortgat 1981, Bresnan 1982), and eventually to LP.

However, it is clear that base-generating and storing all word-forms would introduce high levels of redundancy into the grammar. Consequently, most morphological work after `Remarks' (Halle 1973, Siegel 1974, Aronoff 1976, Allen 1978) has proposed that word-formation rules perform morpheme concatenations rather than linking independent lexical entries. The next innovation involves the organization of these word-formation processes in the lexicon. Siegel (1974) observes that derivational affixes in English fall into two classes; Class I affixes include in-,-ity, Adjective-forming-al,-ic and-ate, while Class II includes un-,-ness,-er, Noun-forming-al and-hood. The former set corresponds to the #-boundary affixes of SPE, and the latter to #-boundary affixes. This class division rests on the morphological behavior of the affixes, as well as having phonological consequences.

First, as shown in (2.1), Class I affixes are free to attach to roots, while Class II affixes attach only to words.

(2.1)

inert                               *unert

intrepid                         *untrepid

insipid                            *unsipid

immaculate                   *unmaculate

(from Allen 1978)

Secondly, in multiple affixation, Class I affixes appear nearer the stem, so that a Class II affix can be added `outside' a Class I affix, but not vice versa (2.2).

Siegel proposes that all Class I affixations precede all Class II affixations. This idea is developed and extended by Allen (1978: 6), who reinterprets Siegel's classes as levels, arguing that `the ``level'' designation indicates that the morphology is partitioned into blocks of rules, each block having different morphological characteristics. Furthermore ... the morphology is level-ordered. That is, the levels of rule operation are ordered with respect to each other, although no ordering is imposed on individual rules of word-formation.'

Derivational word-formation rules attaching Class I affixes will there fore be ordered on Level, or Stratum 1 of the lexicon, while Class II affixations will take place on Level 2, as shown in (2.3). Underived stems are acceptable on Level 1, but only words on Level 2. Bound stems must therefore undergo some affixation process on Level 1, or will be ineligible to pass to subsequent levels.

The diagram in (2.3) incorporates a number of more or less controversial assumptions on the organization of the lexicon, especially concerning the storage and attachment of affixes. There are two opposing views here, represented by Lieber (1981) and Mohanan (1982, 1986) on one hand, and Kiparsky (1982; partly after Aronoff 1976) on the other.

Lieber argues that both stems and affixes are lexically stored, with appropriately specified features and labels: thus, the suffix-ness would be labelled ]A -]N, showing that it is added to an adjective to create a noun, while the verbal suffix-ed would carry the label ]V -]V and the feature [+ past]. Unlabelled binary branching trees, generated by a single context-free rewrite rule, represent the internal structure of words. Formatives are inserted from the lexicon under the terminal nodes of these trees, and features are transferred to higher nodes by Feature Percolation Conventions. In Lieber's model, affixes are heads, and the final affix determines the category and features of the word (2.4).

Mohanan (see 1986: 16) appears to accept a version of Lieber's proposal. He assumes that stems and affixes are stored in a single morpheme list, and are undifferentiated in terms of bracketing. This lack of differentiation extends also to compounding and affixation, as shown in (2.5); note that the single ] bracket of LP replaces the + and # boundaries of SPE.

(2.5) [happy]                                 stem

         [un], [ness]                           affixes

         [[happy][ness]]                    affixation

         [[green][house]]                  compounding

Mohanan (1986: 16±17) further suggests that the information given for each affix in the morpheme list includes a specification of the domain of application for the rule attaching that affix: this domain may be a single stratum, or a set of continuous strata.

On the other hand, Kiparsky (1982) proposes that stems alone should be stored, and that affixes are introduced by word-formation rules, which again will be marked for their domain of application: `affixes will not then be lexical entries, and they will have no lexical features either inherently or by percolation' (1982: 6). Restrictions on the environment in which the affix may be attached, corresponding to Lieber's subcategorization frame and categorial specification, are instead construed as contextual restrictions on the affix-insertion rules, as shown in (2.6).

(2.6)     General affixation rule: Insert A in env. [Y-Z]X

             Plural:

             Insert-en in env. [ox-]N, +Pl.

             Insert-s in env. [X-]N, +Pl.

Kiparsky also distinguishes stems (which are stored) from affixes (which are not) by bracketing, and in his model, the outputs of affixation and compounding will also be distinct in terms of bracket configurations, as (2.7) shows. For the moment, I follow Kiparsky's assumptions on bracketing and affix insertion: I shall justify this decision more fully, and make some revisions.

(2.7) [happy]                           stem

         [un[happy]]                    prefixed

         [[happy]ness]                 suffixed

         [[green][house]]            compound

Finally, Mohanan and Kiparsky agree that, although word-internal structure is relevant within the stratum on which it is created, it should not be accessible to rules on subsequent strata. A Bracket Erasure Convention therefore removes all word-internal brackets at the end of each level: this `opacity principle' (Mohanan 1982: 7) will be further justified below in terms of the interaction of phonological and morphological processes.

The extension of the lexicalist hypothesis since Chomsky (1970) has led to the inclusion of morphological processes other than derivation in the expanded lexicon. Allen (1978) proposes that compounding, as well as derivational affixation, should be regarded as lexical, and introduces a third morphological stratum for compounding processes. Halle had already argued that a generative model of morphology should not be limited to derivation, but that `facts that traditionally have been treated under the separate heading of inflectional morphology must be handled in completely parallel fashion' (1973: 6); Lieber (1981) follows this lead and adds inflectional affixation to the inventory of lexical processes, on the grounds that inflectional stem allomorphs may form the input to derivation and compounding, so that all these word-formation processes should take place in the same component. The assumption that all morphology is lexical is one shared by most proponents of LP, including Kiparsky (1982, 1985), Mohanan (1982, 1986) and Halle and Mohanan (1985). There have been attempts to argue that inflection should be regarded as syntactic (and therefore postlexical); Anderson (1982), for instance, presents an analysis of Breton verb agreement which relies on the interaction of inflectional morphology and syntax. However, Ander son's proposals are countered by Jensen and Stong-Jensen (1984), and further persuasive arguments for parallel treatment of inflectional and derivational morphology can be found in Halle (1973) and Miller (1985). I shall therefore adopt the view that processes of inflection, derivation and compounding all take place within the lexicon. To indicate the composition of such a morphological model, I give in (2.8) the lexical organization proposed in Kiparsky's early (1982) work on English; note that this is included simply for illustration, and will be amended later.

Kiparsky (1982) has thoroughly investigated the morphological con sequences of the level-ordering hypothesis. We have already mentioned the phenomenon of stacking (the fact that affixes from a later stratum may be attached only `outside' those attached earlier in the lexicon, not nearer the stem; this has become known as the Affix Ordering Generalization (Selkirk 1982a)), and also the ability of Level 1 affixes alone to attach to bound stems. I shall consider one further example of the morphological predictions of the lexicalist model, namely blocking.

The blocking effect, which Aronoff (1976) calls `pre-emption by synonymy', has two subcases:

(1) Forms may not usually receive two alternative affixes with the same semantic content. So, we have feet and oxen but not additionally *foots, *oxes, and zero-derived guide, spy, but not *guider,*spier.

(2) Lexical items with some inherently marked morphological feature do not additionally acquire an affix which marks this feature. Thus, people, which is already inherently [+ plural], does not receive plural-s. Linked to this is the failure of semantically equivalent affixes to accumulate on a single stem; so, oxen does not undergo regular plural suffixation to give *oxens. This generalization does sometimes break down: English children, Dutch kinderen, lammeren and Afrikaans kinders, eiers would all be exceptions at least historically.

Kiparsky (1982) argues that these blocking phenomena can be readily explained within the lexicalist model, by two slightly different strategies.

(a) Doublets are prohibited by making morphological rules obligatory in the unmarked case: so, ox, if it carries the feature [+ plural], is marked to undergo a special Level 1 rule attaching-en. The form is not then eligible to undergo the Level 3 regular plural rule. In cases where doublets do obtain, as with indices-indexes, Kiparsky assumes that the special rule is exceptionally speaker-specific. The system for blocking derivational doublets is identical (although less rigid): the deverbal agent noun spy is zero-derived on Level 1, and may not also acquire the functionally identical Level 2 agentive marker-er. Blocking is therefore seen as `pre-emption by prior application' (Kiparsky 1982: 8). Kiparsky uses these facts to support a number of hypotheses on the organization of the lexicon: notably, he argues that when a set of processes is involved in a blocking relationship, special rules with restricted applicability must precede general, regular processes. Hence rules on later levels are more productive, and more semantically uniform, than those higher in the lexicon.

(b) The exclusion of functionally equivalent stacked affixes and double marking of features is rather more complex, and requires the introduction of one of the principal constraints of LP, the Elsewhere Condition (henceforth EC). The EC governs disjunctive application of rules, and is given in (2.9).

(2.9) Rules A, B in the same component apply disjunctively to a form  if and only if

(i) The structural description of A (the special rule) properly includes the structural description of B (the general rule).

(ii) The result of applying A to  is distinct from the result of applying B to . In that case, A is applied first, and if it takes effect, then B is not applied. (Kiparsky 1982: 8)

Kiparsky makes the further assumption that every lexical entry, and the output of every layer of derivation, is an identity rule L, where the structural description and structural change of L are both L. The lexical entry for people is then inherently marked [+ plural], so that L = [people]+N, +Pl.. L in this case is disjunctive with the regular plural rule by (2.9): the rule [people]+N, +Pl. properly includes the structural description [X - ]+N, +Pl., and the outputs, people and peoples, are distinct. The identity rule, as the special rule, then takes precedence. Similarly, *oxens is impossible, since the Level 1 derived lexical entry [oxen]+N, +Pl. is again disjunctive with the regular plural rule. The EC has had profound consequences for the development of lexicalist theory.