1. RULE TYPES IN NGP

Rules that are surface true generalizations can be divided into at least two types on a formal basis and these two types of rules have quite different characteristics. On the one hand there are rules or processes whose statement contains only phonetic information—phonetically based features and the phonetically-motivated boundaries, syllable boundary and pause boundary. These are phonetically-motivated processes and will be referred to as P-rules or processes. On the other hand, there are rules whose statement requires, along with some phonetic reference, the reference to morphological, syntactic and lexical features. These will be referred to as MP-rules.

This formal distinction is very similar (although not identical to) the distinction Stampe (1973b) draws between natural processes and acquired rules. One characteristic of P-rules or processes (following Stampe) is that they are productive and unsuppressible. They apply in loan-word adaptation; they interfere with foreign language acquistion. Further they make a minimal structural change as compared to MP-rules, and finally, they are all “natural,” that is, they are all phonetically-explainable synchronically.

Since P-rules are unsuppressible or automatic, it follows that they will not have exceptions. (This is more strongly maintained in NGP than in Stampe’s theory, since in Stampe’s theory a process can be suppressed or partially suppressed by the subsequent application of another process. Thus surface exceptions are allowed in Stampe’s natural phonology, but not in NGP.) All P-rules are variable to some extent. They are responsible for specifying the shape of the phonetic representation, in which some degree of variation is associated with every feature. Variation should be thought of as the extent to which a feature is altered and not as a matter of whether a rule applies or not. For instance, consider the process of flapping in English and what is sometimes thought of as a separate and highly variable process, flap-deletion. These should not be thought of as two separate processes, one which produces [D] and one that produces Ø. Rather they are one and the same process, a weakening of ambi-syllabic alveolar stops, which produces not just [D] and Ø, but dozens of articulations varying in a continuum from [D] to Ø. The extent to which the articulation is weakened, or the weakening is curtailed, depends on a variety of factors, such as the social situation, the tempo of speech, the degree of stress, and even certain lexical and grammatical factors.

An important way of determining if a P-rule is still a live and productive process is by looking for exceptions to it. Therefore it is important to be able to distinguish true exceptions from the variation inherent in any productive process. But this is a straightforward matter since no lexical item that presents the appropriate phonetic environment for a process can be totally resistant to it. Consider an example: certain words, e.g., veto and certain acronyms such as NATO, seem resistant to flapping. But the resistance is not absolute: [víyDow] and [néyDow] are possible pronunciations. Compare this to an MP-rule that has true exceptions. Consider the alleged S-voicing rule of English, which gives the alternations in con[s]erve and re[z]erve, con[s]ent, re[z]ent, [s]emblance, re[z]emble. Chomsky and Halle (1968) formulate a rule which voices /s/ after a V and a prefix boundary. This rule has dozens of surface exceptions, e.g., descent, recite, assemble, none of which, in even the most unguarded speech will ever undergo voicing to give *de[z]ent, *re[z]ite or *a[z]emble. Nor will the weakening of the boundary in re#sell, re#sew, re#sand ever yield a voiced fricative such as *re[z]ell, etc. The rule is unproductive, it is an MP-rule (if it is a rule at all), and it has true unyielding exceptions.

In the rare cases where it appears that an unproductive rule applies to a new form, such an application can still be distinguished from the application of a P-rule or process. A possible example would be monologuist [g] becoming or varying with monologist [ǰ]. Notice how this differs from the variation or change produced by a P-rule. The variation produced by a P-rule is a phonetic continuum: when flapping affects NATO, a whole range of pronunciations from a full [t^h] to a weak flap are possible. There is no phonetic continuum between monologuist [g] and monologist [ǰ] in present-day English.

In contrast with P-rules or processes, MP-rules make larger structural changes on the whole (e.g., the k → s of electric, electricity, versus the process k → k^y of coo and key).¹ And, as we have seen, they very often have exceptions. If they apply to new forms, it is in the way just explained for monologuist, or it is under morphological conditioning.

All alternations that are the residue of unproductive processes are not accounted for by MP-rules. Some alternations amount to no more than lexical correspondences which become more remote as time passes. These are alternations that go along with morphological processes that are no longer productive (Hooper 1976: Chapter 4). Examples are the English vowel shift and velar softening rules. Vennemann (1972a) has proposed that these lexical correspondences be described in viarules, which do not change one form into another, but merely state the relation that holds between the forms both semantically and phonologically.

On the other hand, some alternations occur in paradigms among items that are very closely and productively related semantically. Alternations of this type seem to be governed by rules. These rules must refer to morphological, syntactic or lexical information. These are the MP-rules that we will be discussing in some detail below.

The True Generalization Condition is meant to apply to MP-rules just as it applies to P-rules. However, in order to restrict MP-rules to generalizations that are true about surface forms, it is necessary to conceive of surface representations as containing non-phonetic information, i.e., the morphological and lexical features that are necessary for a well-formed derivation. Such a conception is entirely natural and realistic, since both the speaker and the hearer are fully aware of all of these features. What the True Generalization Condition means for MP-rules is that all surface exceptions must be marked as exceptions, and that generalizations about morphology must be surface-true and not abstract.

One of the most significant results of viewing synchronic phonologies as consisting of rules and processes is the discovery that many languages have identical, or very similar, processes.² The contribution of Natural Phonology is the idea that all such processes are natural, and that certain universal principles can be developed to explain all processes (Stampe 1973b). For instance, studying processes in their productive and non-productive stages, the notion of a “minimal structural change” can be made more precise. The internal organization of syllables is understood to some degree, and principles governing syllable-related processes, e.g., syllable-final weakening, constraints on deletion processes, can be formulated. The processes of palatalization and vowel nasalization have been explored in terms of substantive principles (Chen 1973, 1974a). In short, great progress has been made recently in the investigation of natural processes of phonology, and in the development of substantive universal principles. Such principles will greatly simplify the analyst’s task of identifying and formulating the processes of a language. In addition, we can hope for some explanations for the various processes that we observe. These explanations will be found in phonetics.

It is the proposal of this paper that a corresponding theory of natural morphology can be developed to explain the nature of MP-rules. Such a theory would consist of a set of general principles that govern the speaker’s analysis of the morpho-syntactically motivated alternations of the language. These principles will be distinct from the principles that explain natural phonological processes, for the MP-rules are motivated by meaning, not by phonetics. In this paper, I will argue that the principles previously used in analyzing morpho-syntactically motivated alternations, the principles borrowed from phonological analysis, are largely inappropriate, and must be replaced by principles based on meaning.

2. MORPHOLOGY IN A NATURAL GENERATIVE GRAMMAR

Some general points about the way morpho-syntactically motivated rules are viewed in NGP must be made clear at the outset. Skousen (1975) and Hooper (1976) have presented a large number of examples that support the hypothesis that speakers, when presented with a choice, will prefer to construct a morphologically-motivated analysis over a purely phonological analysis. The reason is that the speaker-hearer’s task is to associate sound and meaning, and we assume that the speaker does this in the most direct way possible, i.e., surface form to meaning.

Because morpho-syntactic alternations are taken by speakers to be a part of the sound-meaning correspondence, rather than motivated by phonetics, it is to be expected that such alternations are phonologically arbitrary in the synchronic grammar. (Of course, they have a phonologically non-arbitrary diachronie source, but the speakers don’t know this.) Therefore, it is often a pointless exercise to seek phonologically motivated generalizations and explanations for MP-rules in a synchronic grammar.

Let me illustrate this point briefly using the well-worked examples of Spanish stem-vowel alternations in verbs. With a handful of exceptions excluded, there are three basic types of stem-vowel alternations, as shown in (1).

(1) (a) First and second conjugation only

(b)Third conjugation only

(c) Third conjugation only

It is lexically arbitrary which verbs have the alternations (since, of course, some verbs don’t), but if a verb has vowel alternations, the alternation will be one of those in (1). Furthermore, which alternation it will be is partially determined by conjugation class.

In the models of lexical representation proposed by Vennemann (1974b) and Hudson (1975), all the allomorphs of a paradigm are listed in the lexicon. Harris (1978) has argued that this type of representation makes it appear arbitrary that all the stems that have vowel alternations have just these alternations. Since the alternants are listed for each verb separately, there could as well, along with the alternations in (1), be alternations of i with u (which these aren’t), alternations of o with e (which these aren’t), or any other logically possible combination of vowels and diphthongs alternating. Since these other conceivable alternations do not occur, a model which records the alternations in (1) in a separate list for each verb stem, is, in Harris’ view, representing the fact that only the alternations in (1) occur as a synchronically arbitrary fact.

This observation is quite correct, and, furthermore, the observed effect of the model is precisely the desired effect. It is only an accident of history that there are not other alternations along with those in (1). It is an arbitrary fact about the sound-meaning correspondence that each verb stem that alternates alternates in just the way it does. It is about as synchronically explainable as the fact that the word in Spanish for table is mesa, and that this noun is feminine.

Let me hasten to add that the fact that stem-vowel alternations are limited to just those in (1) certainly makes Spanish morphology easier to master than if there were dozens of different types of alternations. This advantage is registered in a natural generative grammar of Spanish by the fact that there is only one rule for vowel alternations that must be learned, rather than dozens (see Hooper 1976:Chapter 8).

Along with the stem vowel alternations in (1) there is an alternation of second and third conjugation theme vowels. The alternants are the same as the front vowel alternants of (lb):

The selected forms shown in (2) are enough to show that the appearance of a particular vowel is determined morphologically (cf. comémos and comímos) . However, since the same vowels alternate, albeit under different conditions, Brame and Bordelois (1974) argue that the alternations in (2) should be handled by the same rule(s) as the alternation in (1). Brame and Bordelois feel that an important generalization is being missed if these alternations are not treated as a unified process.

Notice that the only basis for relating the stem and theme vowel alternations is a phonological identity of alternates. From a morphological point of view, that is, from the point of view of what they signal in terms of meaning, there is no relation between the alternations. Since the alternations are, in any analysis, at least partially morphologically conditioned, and since the conditioning factors differ for the two alternations, there is no reason to consider the phonological identity of any significance.

Substantive evidence supports this point of view. Dialectal innovations in the two alternations effect quite different sorts of changes. The leveling of the stem vowel alternation involves the extension of the high vowel to all unstressed syllables, leaving the diphthong in all stressed syllables (Espinosa 1946, Boyd-Bowman 1960). In the theme vowel, the changes show a movement towards eliminating the distinction between second and third conjugation in the first person of the present indicative. In some dialects the e of second conjugation shows up in the third, e.g., vivémos; in others the i of the third conjugation appears in the second, e.g., comímos (Rosenblat 1946, Espinosa 1946, Boyd-Bowman 1960). These developments are quite unrelated: among other differences, we can note that stress is not a factor in the theme vowel change, while it clearly has an effect on the stem vowel change. These two alternations, despite the identity of alternates, are not synchronically related.

There are of course clear cases of phonological factors influencing MP-rules. We will mention some of these in the last section. Phonological factors are not, however, as important as has been claimed (implicitly or explicitly) in generative phonology and its predecessors (see for example Bloomfield 1933). Rather morphological factors should be considered the primary factors governing MP-rules, and phonological factors should be included only where the evidence clearly indicates this necessity.

In the following sections, we will discuss one important hypothesis concerning morphological analysis. Substantive evidence from linguistic change and from language acquisition will be presented to support this hypothesis. Before turning to this discussion, however, it is important to note a recurrent difficulty in dealing with morphology. Morphology is inherently messy. Since in morphology we study the arbitrary sound-meaning correspondence, we find an almost disconcerting tolerance of exceptions, irregularities and competing generalizations. This is true of synchronic and diachronie data alike. However, certain strong tendencies and general principles are evident, especially in the dynamic data. To develop a theory of morphology, we must isolate those principles and learn to apply them to synchronic analysis.

3. THE SEMANTIC TRANSPARENCY HYPOTHESIS

The type of research that is applicable to morphological analysis has been carried on in the past to a limited extent. Kuryɬowicz (1949) and Mańczak (1957) have studied morphology in a diachronie perspective and have attempted to develop general principles that will explain the direction of analogical change. Both hypothesize that the direction of leveling is determined primarily by grammatical category (N.B., not by phonological factors), but they have different notions of how one determines the category to be favored in leveling. Kuryɬowicz (1949, 1968) claims that leveling favors the basic form, which is the semantically neutral form. Mańczak’s principles rely on frequency of use: he predicts that the form to survive leveling is the most frequent form.

Jakobson (1932, 1939) makes extensive use of markedness in analyzing morphological categories and their expression. Jakobson points out that “Morphology is rich in examples of alternate signs which exhibit an equivalent relation between their signantia and signata” (Jakobson 1965:352). A case in point is Greenberg’s (1963) observation that in languages which regularly distinguish singular from plural forms, it is the plural which always carries an overt mark, an additional morpheme, while the singular is under no such requirement.³

These notions are formalized in Vennemann (1972a): “Usually in natural languages, a semantic derivation of secondary conceptual categories from primitive ones, tertiary from secondary ones, etc., is reflected by a parallel syntactic or morpho-phonological derivation” (p. 240). The following diagram illustrates the principle:

Here X is the sign for A, and y is some “mark” added to X which signifies membership in the secondary category A + b. Thus y may be the addition of a morpheme or the application of an MP-rule. Vennemann proposes this principle, named the Semantic Transparency Hypothesis by Baxter (1975), as a principle of synchronic analysis which also explains certain morphological changes, in particular “rule inversions.” This justifies the inclusion of such a principle in a general theory, and its use in synchronic analysis, since linguistic change is viewed as emanating from the same source as the synchronic grammar—the speakers of the language.

The type of change governed by the Semantic Transparency Hypothesis (i.e., rule inversion) can be simply illustrated with an example from Spanish given in Baxter (1975). (See also Kuryɬowicz 1968 and Malkiel 1974.) In early Romance, many verb stems were followed by a back vowel in first singular present indicative and the present subjunctive, and by a front vowel in the remainder of the present indicative, as shown in (4).

(4) Present Indicative

Present Subjunctive

Under phonetic motivation, velars before front vowels became palatalized, and nasal consonants and /l/ coalesced with contiguous palatals to yield the palatal consonants [ñ] and [ʎ]. These changes give the alternations in (5).

(5) 3rd sg. present

Given the unmarked status of the 3rd person singular present indicative, with respect to both the present subjunctive and the first person singular indicative, and given a base form containing a velar consonant, we have the following expression of the semantic relations:

With the phonetically motivated palatalization process applying in the indicative, the phonetic representations do not accurately reflect the semantic relation between the primitive and derived categories. The Semantic Transparency Hypothesis predicts a reanalysis of the paradigm by the speakers. The reanalysis would involve taking the phonetic realization of the primitive category as the basis for the secondary category.

It is important to note the covert nature of the change: given the phonetic forms, either analysis will work. The evidence for the reanalysis of the velar as the marker of subjunctive is massive. A large number of verbs that have no etymological source for a velar in the subjunctive or in the first person singular present indicative now have these forms marked in just this way. The following are some examples.

The ‘Regular’ column shows what the development of the present subjunctive would have been given the application of phonetic processes only. The forms without asterisks are attested, either in Old Spanish or in dialects (Espinosa 1946). The forms without velars also occur in Portuguese, e.g., vinha [vĩña], tenha [tẽña] and valhe [vaʎa].

The forms in (8) suggest, then, that the analysis in (7), where the velar is taken as a marker of the secondary category, is correct. Unfortunately, in analogical change, it is never certain precisely which forms serve as the model for a change. Malkiel (1974) suggests a number of models for this change including the ones given here, as the pattern was clearly present and strong in the language. It is important to note also that some verbs where leveled in favor of forms containing no velar, e.g., uñir ‘to yoke,’tañer ‘to play’ (a musical instrument), and others wavered for a period, e.g., salgo, salo, valgo, valo. Notice that the loss of the velar is a predictable result, since the velar represents suppletion in the representation of the present subjunctive: /+a/ for some second and third conjugation verbs, /+ga/ for others. The change that would contradict the analysis in (7), the one predicted by the phonological analysis in (6), is the restoration of the velar throughout the entire paradigm. It is precisely this possibility that never occurs.

A number of examples of this sort have been discussed recently. There is the famous Maori example discussed by Hale (1971), the case of French liaison, shown to be an inverted rule by Baxter (1975) and by Klausenberger (1976b), Latin rhotacism, analyzed as an inverted rule by Vennemann (1974b) and Klausenberger (1976a), not to mention the many examples presented by Skousen (1975), and the examples discussed in the original work by Vennemann. Although these cases are generally accepted as valid, I do not think their significance has been properly appreciated. These are not rare and scattered cases where morphological (instead of phonological) principles are seen in action. Rather in every case where substantive evidence is available we see the MP-rules are governed by the Semantic Transparancy Hypothesis. This principle is not, then, just a way of explaining a few cases of analogical change, but rather a general principle that should guide synchronic analyses of morphologically motivated alternations even where additional evidence is not available.

Because MP-rules are conditioned by semantic categories, it follows that the principles which guide their formulation should be based on semantics, not on phonology. The Semantic Transparency formula should replace principles of analysis that are based on phonology. Thus, for example, the most widely used method of determining the base form of a paradigm is to choose as the base form the allomorph from which all other allomorphs are phonologically predictable, i.e., the one in which there are no neutralized contrasts (or a “patched” form, lacking neutralization, if no one form contains all contrastive features). When this principle is applied to the formulation of P-rules, the correct results emerge, because P-rules can neutralize a contrast, but never undo one. The correct direction and even the correct conditioning environment are very often discovered by formulating a P-rule so that it applies to an underlying contrast and neutralizes it. (This, of course, does not imply that speakers use this principle, only that it is useful for linguists.)

The same principle, however, gives the wrong results for MP-rules, unless it happens that the form from which the others are predictable is the semantically unmarked form. All of the cases of rule inversion that have been uncovered show precisely this: that semantic principles override the criterion of predictability. However, the predictability principle seems difficult to give up: Kurylowicz (1968) invokes it, Jakobson (1948:120-121) cites it as a basic principle for morphological analysis (following Bloomfield 1933, who proposes predictability in the name of simplicity), and Vennemann (1974b: 139) lists it as a principle determining rule inversion.

Vennemann cites the example of the so-called loss of final devoicing in Yiddish as a case in which analogical leveling (and rule inversion) are determined by phonological predictability. In this case, however, there is no evidence that the alternation which is alleged to have been leveled ever actually existed. Sapir (1915) was one of the first to notice that certain dialects of Yiddish do not exhibit the voiced/voiceless alternation found in MHG and NHG. Thus the following North East Yiddish forms correspond to NHG forms:

Sapir proposed that the alternation had once existed in NE Yiddish, but had been leveled analogically. The direction of leveling in this case is towards the marked category. It can be explained as the restoration of an underlying form, since these paradigms must have a voiced obstruent underlying to keep them distinct from paradigms with voiceless obstruents throughout.

Sapir’s evidence that the alternation once existed comes from a few forms in which devoicing has clearly applied: mit ‘with’, af ‘upon’, op ‘down’ ˂ MHG abe, and avek ‘away’ (=NHG weg) . Sadock (1973), however, points out that these forms can be matched with other non-alternating forms with voiced finals: oyb ‘if’, iz ‘is’, az ‘as’, and biz ‘up to, until.’ Both sets of forms can be explained if it is assumed that final devoicing applied, but was never a fully generalized process. There is then no reason to posit a stage in which the forms in (9) alternated, and no need to suppose an analogical leveling. This interpretation is further supported by noun forms that did indeed undergo devoicing, nouns ending originally in nd and ld. In some of these forms a voiceless stem-final obstruent is found in both singular and plural, although its original phonetic environment occurred only in the singular: hant, hent ‘hand’, vant, vent ‘wall’, funt, funtn ‘pound’, bunt, buntn (dim. bintl) ‘bundle’, vint, vintn ‘wind’, fant, fantn ‘pledge’. These forms show leveling in favor of the unmarked category, not the phonologically predictable underlying form. Other stems in -nd retained the voiced obstruent throughout, just as the forms in (9) do.

Thus this case is not at all clear, because we do not know if the forms alleged to have been leveled ever in fact alternated. Given the a priori implausibility of phonological predictability as a criterion in morphological change, a very clear and well-documented case is needed.

If the Semantic Transparency Hypothesis is correct, then a number of phonological criteria traditionally used in morphological analysis are inappropriate. Besides the predictability criterion, other suspicious criteria are phonological naturalness and simplicity, symmetry of the underlying segment inventory, and the criterion of the independently necessary rule. Unfortunately, there is not space here to discuss these other principles. However, if my main point is clear, that morphological criteria are of primary importance for morphological analysis, then it will be clear that these other phonological criteria may well be irrelevant also.

4. LANGUAGE ACQUISITION AND THE SEMANTIC TRANSPARENCY HYPOTHESIS

The original rationale behind the Semantic Transparency Hypothesis as explained in Vennemann (1972a) is that it reflects a language acquisition strategy. The presumed strategy is that the basic category is learned first, and, when necessary, certain modifications of the form of this basic category are made to produce the secondary ones.

Of course, more work needs to be done on the acquisition of inflectional systems and MP-rules. Most studies in the acquisition of morphology deal with inflections rather than explicitly with allomorphy or MP alternations. One study, however, illustrates clearly the hypothesized strategy. Stoel-Gammon (1976) reports on the acquisition of certain paradigms in Brazilian Portuguese, in particular on verbal and adjectival paradigms that show (in the adult language) an alternation of the mid vowels [e] and [o] with the low vowels [ԑ] and [ɔ]. The following present tense verbal forms illustrate a very general pattern of alternation:

In the first singular forms the stem vowel varies according to conjugation class, low for first conjugation, mid for second, and high for third. The third singular form always has the low vowel. The first plural form has a mid vowel, which could be predicted from the low vowel, since there is a phonetically-conditioned, productive neutralization of [e] and [o] with [ԑ] and [ɔ] in unstressed syllables.

In the analyses Stoel-Gammon discusses, the mid vowel is taken as underlying, and a lowering rule is postulated to derive third singular forms. In both analyses cited (Harris 1974 and Imanishi 1975), morphological information is included in the lowering rule. The directionality of the rule (i.e., lowering from underlying mid vowel, rather than raising from underlying low vowel) is decided on by Harris (1974) on the basis of the existence of certain related nouns and adjectives with mid vowels, as show in (11)

In Hooper (1976) I have argued that derivational relations such as those between nouns, adjectives and verbs are often not productive relations, and may not be relevant to decisions about the base form of a paradigm. Since the rules in question are undoubtedly MP-rules, we expect only paradigm-internal factors to be at work. Furthermore, we expect semantic factors to be the major determining factors.

The language acquisition data Stoel-Gammon presents and her analysis of these data support these expectations. In longitudinal studies of several children, she finds that the third singular present indicative form is the first to be used and is substituted for all other forms. Then gradually, the child begins to mark person with a pronoun, an inflection or both. The first forms inflected for first person singular have the distinctive inflection -o, but the stem appears in the third person form, as the following examples show:

Stoel-Gammon analyzes the process as follows: “we can say that the child’s rule would be something like: to form first person verbs, take the vowel off the end of the neutral form (the 3rd sg. -JBH) and add -o” (p. 6). Stoel-Gammon says that the child later produces the adult form by a rule which raises the vowel of the first singular form. In other words, she is describing a process by which the primitive category is taken as basic and the secondary categories derived from it.⁴ Furthermore, the vowels that occur in related nouns and adjectives seem to have no relevance for the child’s analysis.

The acquisition of the alternation in adjectives shows the same strategy. In the adjective paradigms alternations of mid and low vowels are as follows:

The choice of an underlying form for this alternation is phonologically arbitrary, since non-alternating adjectives with [o] and with [ɔ] also occur. Morphological criteria are unambiguous, however, and once again, these seem to be the criteria used by children.

Stoel-Gammon reports that in the first stage a single constant form is used, usually the masculine singular form. Next gender, but not number, is differentiated. In this stage “adjectives with vowel alternations in the masculine and feminine seem to be produced correctly, e.g., vestido n[o]vo, camisa n[ɔ]va” (Stoel-Gammon 1976:7). When the plural is learned, it appears as the suffix /+s/, and the vowel alternations between singular and plural do not appear right away. In fact, Stoel-Gammon indicated that the adult plural forms are sometimes not acquired until after the child has been in school several years. The forms for this stage are

The forms marked with * have a low vowel in the adult forms. These forms show that the child takes the primitive category (singular) as basic and derives the secondary category from it.

Another interesting point is suggested by these data. The various alternations involving [o] and [ɔ] may not be considered related by the child. The singular forms of the adjective have the adult vowels as soon as gender is differentiated, but the plural forms lag far behind. This suggests that the masculine/feminine relation is of a different sort than the singular/plural relation, an idea which is not too surprising from a semantic point of view. It further supports a point made earlier in this paper, that phonological identity of alternating segments does not necessarily indicate a single source and a single MP-rule for all instances of the alternation. Stoel-Gammon suggests (very cautiously) that the masculine and feminine may have separate underlying forms, while plural forms are produced by rule. This may, in fact, be the way such alternations need to be differentiated.

The acquisition data shows, then, that children do implement a strategy much like the Semantic Transparency Hypothesis. This does not necessarily mean that adult grammars are formulated on the same basis. However, until evidence appears which shows that the acquisition process includes a total revision of the child underlying forms and rules before adulthood, so that child and adult grammars differ radically, we can hypothesize that adult grammars, like child grammars, are based on the Semantic Transparency Hypothesis.

5. FURTHER PROBLEMS

Given the notorious irregularity of morphological form, it would be surprising if any principles of morphological analysis could be upheld as anything but general tendencies. For instance, it is not always the case that the neutral or unmarked category is the category expressed with the fewest overt markers (Jakobson 1939). If the Semantic Transparency Hypothesis is a strategy of analysis used by speakers, what is the response to a situation in which the primitive category is expressed by an overt marker and the secondary category by zero? A logical possibility would be the analysis of the secondary category by deletion, so that the "mark" of the secondary category is the application of an MP deletion rule. However, it seems highly implausible that a morphologically-conditioned alternation of a segment with zero would be accounte for by a deletion rule. The reason is that if a segment appears in one category but not in another, then that segment will be taken as the marker of the category it appears in. This follows directly from the premise that speakers account for the sound-meaning correspondence in the most direct way possible.

For instance, the English present tense forms would not be analyzed with the third singular form talks taken as basic, and the other forms derived by deletion of s. Kuryɬowicz’s solution seems better in such a case: “in derivations like nominative singular: other case forms, present: other tenses, indicative: other moods, third person singular: other persons, the formal surplus of the basic form may be ignored” (1968:75). In other words, he advocates an analysis which isolates a base form of the stem.

It seems appropriate here to defer judgment until more evidence is examined, for even the case of talk/talks is not clear. In varieties of English that have leveled the present tense forms, the base form with no third singular marker is often found, e.g., in Black English, but a generalized form with the third singular suffix for all persons also occurs, e.g., in Appalachian (Wolfram and Fasold 1974). In many varieties of English the present tense of the verb say is leveled to says (sez) (“and then I sez”) when this verb is used for reporting conversation.⁵

Problems such as this one, then, need to be studied further. Even a brief look at morphological data reveals numerous problems of just this type that need to be approached through the study of historical and psycho-linguistic data. I will mention only two more very important problems.

The first concerns the roles played by phonological factors in morphological systems. Of course, a good deal of allomorphy is due to phonetically motivated processes, and the role of phonology here is indisputable. However, some alternations that are restricted to certain morphemes are also clearly based on phonological principles. Alternations of this sort preserve the general syllable structure conditions and phonotactic constraints of the language without obliterating the morphological markers. An example is the alternation in the English plural morpheme of /z/ with /ɨz/ and the past /d/ with /ɨd/. (The variants /s/ and /t/ are phonetically conditioned.) The barred i in these morphemes appears in order to separate the consonant of the suffix from a similar stem-final consonant (e.g. kisses [klsiz], wanted [wãtɨd]). Phonological information concerning the end of the stem is necessary, and a phonological generalization is made. Yet this alternation is restricted to these few morphemes. When in other instances two similar obstruents came together, their assimilation and degemination is allowed, as in horseshoe or maddog.

The extension of the velar subjunctives in the second and third conjugation in Spanish seems to follow some phonological pattern as well. The stems that acquired the velar ended originally in /n/, /ñ/, /l/, /ʎ/, /s/, a vowel, or a palatal glide. The extension in these environments is quite general if we include dialectal data, where, e.g., vaya (subjunctive) “go” is vaiga, crea (subjunctive) “believe” is creiga, and so on. This is a reasonably natural class of segments, but the question arises, why just this phonological class? The answer has to do with syllable structure again. Vowels, glides, nasals, liquids and /s/ happen to be just those segments that can end syllables in Spanish. Note that with the inserted velar, the stem-final segment becomes syllable-final. If the velar were inserted after an obstruent, or a cluster, an unacceptable syllable would result. (I should note, as does Baxter (1975: 192-193) that there is no reason why /r/ should not belong to this class. However, Baxter notes that a few examples have been found, e.g., Old Spanish ferir, firgades, fiergan, and the velar insert appears after /r/ in Old French dialects.) In this case, then, it seems to be the general syllable structure conditions of the language that determine the morphological class membership.

A second use of phonology and phonological shape in MP-rules is in the designation of morphological class membership. An example is the English strong verb class exemplified by sing, sang and sung. This is one of the few strong verb classes that attracts any new members. Among its newer members are fling, flang, flung and the alternate past form of bring, brung. It is clear why bring and fling are assigned to this class of alternating verbs: ending in a nasal, particularly a velar nasal, makes them seem to belong to this class.

Finally, there is an important question remaining concerning further sub-classification of rules with morphosyntactic conditioning. There seem to be two major types: those which govern alternations that are viewed primarily as suppletive, and are subject to leveling, and those which govern alternations that correlate with morphological categories, and are subject to analogical extension. This distinction is made by Klausenberger (1976a) and also attributed to Kruszewski (Klausenberger 1977). Rules of both types usually arise in the same manner, by morphologization of an originally phonetically-conditioned process, but we do not know how to predict which type of rule will arise from a given alternation. While it is clear that there must be some reasonably strong correlation with morphological categories, this correlation need not be perfect. The problem can be approached now through historical data. When we understand which alternations speakers consider to be suppletive and which they consider to be a signal for a meaningful category, we will understand a great deal more about how speakers analyze morphology.

NOTES

I am grateful to Andy Baxter and Carol Stoel-Gammon for comments on an earlier version of this paper, and to the participants in the Conference, in particular David Stampe and Linda Waugh, for comments and suggestions concerning the points made here.

1. In some cases the SC made by a P-rule and an MP-rule are the same. For example, intervocalic voicing of obstruents is attested as a P-rule in some languages, as an MP-rule in others.

2. Of course, Trubetzkoy was interested in neutralization processes, and American structuralists studied natural process under the heading of “phonetic similarity” (Austin 1957), but in both of these paradigms the emphasis was on the phonemic inventory, not the processes themselves.

3. Of course, as Jakobson (1939) points out, and as we will see below, the zero mark does not always coincide with the neutral category.

4. See Simões and Stoel-Gammon (1977) for a discussion of some of the possible reasons that children acquire the primitive category first.

5. I am grateful to David Stampe for pointing out the relevance of sez.