I will use the term 'Word Syntax' to refer to those morphological theories which rest on the three assumptions which Beard (1995) calls the Lexical Morpheme Hypothesis:
In various forms this set of assumptions underlies Selkirk (1982), Di Sciullo and Williams (1987), Lieber (1992), and the principle work of the advocates of P&P, RG, and LFG. The central problem with Word Syntax is that these assumptions describe only major category items, i.e. lexemes, and fail to account for differences between lexemes and the output of derivational and inflectional morphology. The inclusion of affixation in the lexicon leads to confusion within the model which, as a result, fails to predict major aspects of the behavior of linguistic elements.
If affixes are signs with the same definitional properties as prototypical lexical signs, we would predict the following.
Since all lexemes must be at least partially phonologically specified, neither zero nor empty lexemes can exist. It follows that zero and empty affixes also cannot exist.
Affixes should belong to open classes and hence be subject to the same sort of synchronic lexical derivation which prototypical lexemes undergo.
Affixes should belong to one and only one lexical category as is required of uncontroversial lexical items.
Affixes should be listable items like other lexemes, not operations or processes more akin to syntactic or phonological rules.
If affixes are to be coclassified with prototypical lexemes, we would expect them to share most of the attributes of prototypical lexemes. If they do not—certainly if they share none of these attributes—the argument for a unified lexicon containing a single type of morpheme collapses.
Since it departs from the assumption that affixes are lexical signs, the first challenge to the Lexical Affix Hypothesis of Word Syntax is 'zero' morphology. If affixes are lexemes, since prototypical lexemes are never null or empty by Principle I, zero and empty affixes also should not exist yet both are ubiquitous. One way to avoid the loss of this generalization is to demonstrate the assumption false by arguing that zero lexemes do exist.
The most dramatic claim of zero lexemes is found in Anceaux (1965). Anceaux claims that Nimboran, a language spoken in New Guinea, contains a total of 12 verbs (of the 300 or so he studied) with phonologically null stems. Anceaux's list includes the verb stems for 'bring', 'dream', 'extend', 'go', 'head', 'kiss', 'laugh', 'make cat's cradles', 'say to', and 'sleep'. Anceaux, however, provides no allomorphic rules at all, even though the verbal data he presents is rife with phonological variation, e.g. the three distinct verbs he posits for the single meaning 'to signal': kyéb-, iyé-, yé-.
Anceaux is so struck with the similarities in these three perfectly synonymous stems that he mentions them on p. 144 and again on p. 151; nonetheless, he counts them as three synonyms. On pp. 124-125 and again on p. 151 he admits that one of his zero lexemes is a member of a similar triplet, all meaning 'say to': u-, i- and null. He shows no curiosity at the fact that u- has no Plural and that i- occurs only in the Plural, and simply lists the null variant without comment. In fact he allows that three more of his zero stems have phonologically realized variants: réi- 'sleep' (p. 124, 158), ty- 'hear' (p. 129), and kiú- 'laugh' (p. 129). Anceaux's evidence actually points toward zero allomorphic variants resulting from phonological rules.
Sign-based morphology assumes that lexical and grammatical meaning is directly conveyed by sound; so, a priori, the absence of sound should imply the absence of lexical or grammatical meaning. This assumption not be undermined by zero morphology if it may appeal to another kind of morphological relation altogether to explain it. European morphologists account for zero morphology as a process wholly distinct from derivation, conversion. Conversion implies that the affixed and unaffixed forms of the following Table 1 are unrelated.
|Table 1: Deadjectival Causative Verbs|
If two unrelated productive processes with the same functions existed side by side, one mimicking the other, we would expect them to apply to the same stems with the same lexical results, differing only in the properties of the components to which each belongs. Lexical derivates, for example, like widen and deepen are generated alongside syntactic constructs with the same grammatical meaning and similar form: make/get wide, make/get deep. Duplicate output like this would be convincing evidence for discrete components operating over lexical derivation and conversion, but we do not find it. Table 2 exemplifies the typical case: wherever we find an affixed derivation, we do not find a conversional pair and vice versa.
|Table 2: Conversion vs. Affixation|
|(to) double : *endouble||(to) *noble : ennoble|
|(to) right : *righten||(to) *thick : thicken|
|(to) warm : *warmen||(to) *deep : deepen|
The introduction of conversion into the grammar in fact necessitates an ad hoc convention, complementary blocking, to explain the curious distribution which excludes just those stems undergoing conversion from lexical derivation and vice versa. Such a convention is highly suspect since complementary distribution usually marks an underlying grammatical regularity unifying the data, just the opposite of the contrastive distribution required by the argument.
Empty morphology is also a well-established property of affixation which does not characterize lexical stems. Languages brim with examples of affixes with neither grammatical nor semantic content, a striking deviation from the norm for a lexical item, given the otherwise strong semantic character of the lexicon. (1) illustrates obligatory and optional English affixes which are totally empty.
(1) a. dram-at-ic(al) b. syntact-ic(al) c. class-ic(al)
The suffix -at is an imported stem extension which Greek bases sometimes require for reasons germane only to Greek morphology. The suffix -al occasionally offers a means of distinguishing two meanings of an ambiguous adjective, e.g. economic versus economical; however, in (1) it is so empty as to be optional. The other interesting aspect of emptiness is that it may be a variable rather than a constant state of a given affix. Notice that the meaningful suffix in (2), -ic, plays an empty role in the adjectives of (2).
(2) a. Marx-ist(ic) b. social-ist(ic) c. capital-ist(ic)
Again, the evidence is that the properties of morphological objects are radically distinct from those of lexical objects.
Since lexemes are subject to lexical derivation, evidence that affixes are, too, would partially substantiate the claim that affixes are lexemes. Lieber argues that affixes do undergo derivation, referring to the German suffix -ig in A : V pairs like
(3) ängst-ig 'afraid' : ängst-ig-en 'frighten' kräft-ig 'strong' : kräft-ig-en 'strengthen'
Lieber (1981) accounts for these examples by an unmarked operation on the suffix -ig alone, rather than on the stem as a whole. The same rule, A → V, would also account for lexical pairs like grün 'green' : grün-en 'to green', steif 'stiff' : steif-en 'stiffen'.
This argument does not go through, however, because the position which it opposes, that the entire adjective stem, ängst-ig-, undergoes verbalization like the entire stems grün and steif, is equally supported by the data. The point is therefore moot given the fact that such affixal derivation would never be marked by any morphology not equally applicable to the stem as a whole. The only proof of affixal derivation or conversion we could hope for would be an example of an affixed affix like English ?at-ic or ?ic-al, perhaps produced by a compounding lexical rule. But such extended exponence is not derivational since the morphemes involved never express lexical derivational functions. Dram-at-ic does not differ at all derivationally from theatr-ic or semantically from dram-at-ic-al. We must concede that affixes do not derive and rationalize this fact in our theory.
The lexical listing of affixes is supported by one fact in particular: in the structure of a lexically derived word under Word Syntax, the affix may be said to assign the lexical category and subcategorization features to the node dominating the lexical bases to which they attach, e.g. baker is a noun because the head is the noun -er attached to the lexical verb bake. However, if affixes possess lexical category and subcategorization like prototypical lexemes, they would be limited to the possession of only one lexical category since noncontroversial, prototypical lexical items belong only to one lexical category (N, V or A).
Although it is conceivable that an affix might belong to more than one lexical category and assign its lexical categories in dependence on context, such behavior would not be lexemic. Lexemes have only one lexical category and whenever a lexeme ostensibly belongs to more than one lexical category, it is either an accidental homonym (pear : pare : pair) or is involved in a derivational relation like a frame : to frame. Since the preceding section has eliminated the possibility of derived affixes, Word Syntax now predicts that, with the exception of accidental homonyms, affixes must belong to one and only one lexical category.
The evidence against the prediction that affixes are restricted to a unique lexical category can be quite spectacular, as (4-7) demonstrate.
(4) The boy is cutting flowers (Progressive Verbal Aspect)
(5) He ruined his knife cutting flowers (Deverbal Adverb)
(6) a. His cutting the flowers dismayed us (Gerundive)
(6) b. Cutting is for the birds (Imperfective Nominalization)
(6) c. He brought his cuttings in (Resultative Nominalization)
(7) a. The boy cutting the flowers is cute (Present Participle)
(7) b. a very cutting remark (Subjective Adjective) br>
Not only does -ing mark verbs (4), adverbs (5), nouns (6) and adjectives (7), it marks both inflectional (Aspect, participles, deverbal adverbs and gerundives) and lexical derivational categories (Subjective adjectives, Resultative nominals). Other affixes mark narrower ranges of derivational and inflectional relations, e.g. -ed, -s, -en.
Why can't we simply posit homonymous variants in cases like these? There are three types of evidence of homonymy: diachronic, derivational, and allomorphic. Diachronic evidence, such as the spelling variation in pair, pear, pare, demonstrates a clear historical distinction among synchronic homonyms, reflected in previous phonological representations which were distinct. Such diachronic evidence does suggest that two homonymous suffixes are historically responsible for English -er. In Old English the Comparative suffix was written -re or -ra and the Subjective suffix, -ere. The historical status of these suffixes offers no proof of their synchronic status.
There are allomorphic variations in some US dialects. These dialects drop the velar before the comparative -er, e.g. long-er, but not before the subjective, e.g. singg-er. So affix homonymy can be demonstrated on diachronic and synchronic grounds but such evidence is available only for -er, and that is weak and not general. Moreover, this faint evidence of homonymous suffixes expressed by -er distinguishes only the Comparative adjective suffix from the deverbal nominal; the Subjective, Objective, and Instrumental functions of the nominal -er have never exhibited any evidence of homonymy, nor do most other ostensibly homonymous affixes.
There is a difference between lexical homonymy and any definition of homonymy based on data like that of Table 3.
|Table 3: Three English Derivations|
|Affix||Qual. Adj.||Subj. Noun||Obj. Noun|
|-ee||. . .||escap-ee||draft-ee|
|-ist||Marx-ist||Marx-ist||. . .|
The function of the nominal -er, for example, does not drift randomly, in Sapir's (1921) sense, or expand according to the semantic extension schemas of Booij (1986). The meanings of affixes skip cleanly from one derivational function to another: Subjective, Objective, Instrumental. Even the highly idiomatic functions of -er generally come from this limited set of derivational functions, e.g. Locative: diner, Action nominal: merger, Objective nominal: loaner. The referential range of the set of English affixes is thus determined by the range of meaning in the closed set of grammatical categories. Lexical homonyms exhibit no such semantic consistency, no categorial parallels whatever: bear (verb), bear (noun), bare; pair, pear, pare.
>In conclusion, then, the problem of polyfunctional affixes like English -ing cannot be solved by homonymy in an Word Syntax framework; introducing it would be a purely theoretical maneuver. No consistent allomorphic or other empirical evidence supports affix homonymy and the attributes of affix homonymy diverge widely from those of prototypical lexical homonymy.
The only unassailable shared trait of lexemes and affixes would seem to be their necessary phonological realization. However, even this apparent common property dissipates under scrutiny. We have already examined phonologically null and empty affixes and noted the absence of corresponding null and empty major category items. Another threat to the claim that affixes are lexical items is what seem to be morphological operations on stems. Affixes resemble lexemes to the extent that they may be copied onto lexeme stems by the same sort of rule that copies stems to their structural position. However, operations like reduplication resist description as listable items of any sort, despite attempts to describe them as lexemes.
McCarthy (1981), Marantz (1982), Yip (1982), and Archangeli (1983) are supporters of Word Syntax who are recasting the phonological issues of morphology to account for processing as the selection of phonologically underspecified lexical items. Their Autosegmental Phonology allows phonologically partially specified items comprising only CV templates to be listed in the lexicon along with fully specified items. By extending the domain of phonology to allow copying from the stem to these templates or from vowel clusters to stem internal positions, morphological processes become a combination of lexical selection and phonological operations.
The ultimate point is that prototypical major category items, whose nature is not in dispute, cannot be represented phonologically by revowelling or reduplication. All prototypical major category items, which affixes must match in all essential respects to achieve lexical status, must be phonologically prespecified in the lexicon and matched with a cognitive concept. Like fully specified affixes, revowelling and reduplication presuppose prototypical lexemes in that they are defined wholly in terms of some class of lexemes. However we notationally characterize reduplication, it remains a process which mimics some part of a lexical stem and attaches the result of that mimicry to some part of a lexical stem. If one set of objects presupposes another, the two sets must be distinguished and one must logically precede the other. The effect of the Autosegmental approach to reduplication and revowelling then is to conceal a distinction critical to the effectiveness of the model.
A morphology which represents all morphemes as processes rather than items will be much more internally consistent than morphology even extended by an autosegmental treatment of revowelling and reduplication.
If we do not begin with the assumption of lexically listed affixes, the operations proposed by McCarthy and Marantz become quite natural types of stem modification and lose their markedness vis-à-vis uncontroversial lexemes. Reduplicating a prefix or suffix is an unsurprising variation of spelling one out; revowelling is a mundane variation of infixation. However, these operations are radically different from the simple selection and copying of N, V, and A stems which are always fully prespecified phonologically. The autosegmental approach cannot capture this fundamental difference between lexical and morphological relations and operations as well as can LMBM with its sharp distinction between lexemes and morphemes.