1
Introduction
1.1 Aims and Assumptions
The aim of this book is to propose and defend certain generalisations about morphological behaviour. These generalisations are intended to be valid for all languages which exhibit morphological behaviour of the relevant kinds and are therefore, in that sense, claims about linguistic universals. They concern inflexional morphology, and more particularly the relationship between morphological âexpressionâ and âcontentâ (or âsignifiantâ and âsignifiĂ©â).
All languages relate sounds to meanings, and do so partly through attributing significance to the order of meaningful units smaller than the total utterance. The first of these remarks is quite banal. The second is somewhat less so; it is not logically necessary that a communication system for use by human beings should be âarticulatedâ (in Martinetâs sense) at two levels, the phonological and the syntactic; but the fact that language is so articulated is one of the few elements of common ground among all serious students of language. It follows that the description of any language will involve a distinction between its phonology on the one hand and what we can loosely call its syntactic-semantic apparatus on the other. By no means all languages, however, display the sort of behaviour which in a traditional grammatical description of Greek or Latin is treated under the heading âinflexional morphologyâ. It has, moreover, been notoriously difficult to arrive at a satisfactory general definition of the term âwordâ, designating the linguistic unit whose internal structure is the subject-matter of morphology. These are two of the reasons why some linguists have not merely neglected morphology as uninteresting but actually denied its existence as a distinct component of grammar altogether. The first assumption that I will make is that this is incorrect, and that in many languages one can identify grammatical units â âwordsâ â with an internal structure which differs more or less from that of sentences and which therefore cannot be described adequately by reference only to the rules of sentence structure or syntax. This assumption is scarcely controversial among generative grammarians today, since the âlexicalist hypothesisâ about word-formation has displaced the âtransformationalist hypothesisâ (cf. Chomsky 1970; Scalise 1984); still, for arguments to back up the assumption, nongenerativists and persistent sceptics can turn to Peter Matthewsâ work on morphology (e.g. Matthews 1974: 2-8).
I assume also Matthewsâ notions morphosyntactic category and morphosyntactic property (1972b: 161-2; 1974: 66, 136). Morphosyntactic properties are what inflexions express or realise, such as Masculine Gender, Past Tense or Accusative Case; I regard them as constituting the inflexional âcontentâ (as opposed to âexpressionâ) referred to earlier. Morphosyntactic categories are classes of contrasting and mutually exclusive morphosyntactic properties, such as, in Latin, Gender, Tense and Case. Each category, together with the properties it contains, is applicable to one or more parts of speech or wordclasses. I adopt here, as I do throughout, Matthewsâ practice of giving a capital initial letter to the names of morphosyntactic categories and properties. For brevity, I will often omit the word âmorphosyntacticâ, but all references to categories and properties should be understood as references to morphosyntactic ones unless I make it plain that I am using these terms in some other way. In particular, I will not use âcategoryâ in the sense of âword-classâ or âpart of speechâ.
The set of categories and properties relevant to one language is not necessarily the same as that relevant to the next. This could hardly be otherwise, given that there are âisolatingâ languages which have no inflexion at all and consequently no morphosyntactic properties, according to my definition; that is, in an isolating language like Vietnamese, for example, verbal tenses (if they exist) must be purely syntactic or semantic and cannot be called morphosyntactic. The non-universality of categories and properties, in this sense, is so obvious as to be hardly worth mentioning. But it leads directly to a problem which is far from banal, namely: what are the criteria for identifying the morphosyntactic properties and categories relevant to a given language? My answer to this question resembles the answer that I will give to various other fundamental questions of definition. To arrive at a watertight set of criteria would involve discussion of, and decisions about, a number of problems quite far removed from the aim of this book, such as the handling of syntactic âcooccurrence restrictionsâ in the widest sense (including concord and âsequence of tensesâ). But there are enough clear examples of inflexion, involving morphosyntactic properties that are fairly straightforwardly identifiable, to provide us with a core of material to begin our investigation. Refining the criteria to cope with the more controversial penumbra can wait until we know whether our study of the core material looks like yielding profitable results in the shape of interesting (i.e. readily falsifiable but nevertheless unfalsified) generalisations; and at that stage we can legitimately allow our provisional results to influence our decisions.
There is, however, one characteristic of morphosyntactic properties which we can regard as definitive straight away. If morphosyntactic properties are what inflexions realise, then a distinction between two properties which is never manifested in any distinction between inflected word-forms is impossible. One may, of course, want to recognise, even in an inflected language, syntactically relevant âpropertiesâ or âfeaturesâ which are never expressed morphologically. âPropertiesâ of this kind will include, for example, many of Fillmoreâs (1968) âcasesâ, which are explicitly more abstract entities than the traditional morphological Cases of a language such as Latin. Under this heading, too, comes (for example) Dixonâs (1972) distinction between instrumental and ergative âcasesâ in Dyirbal. Dixon claims that there are good syntactic grounds for distinguishing these two âcasesâ; but, since there is never any overt morphological distinction between them, we are not entitled to recognise here more than one morphosyntactic Case, any more than the syntactic distinction between the object of transitive verbs and the subject of embedded infinitival sentences in Latin justifies us in recognising more than one morphosyntactic property Accusative, which happens to be manifested by nouns in these two distinct syntactic contexts. A necessary condition, then, for a clear example of a morphosyntactic property is that it should have an overt inflexional manifestation in at least some members of the appropriate word-class.
The decision to concentrate on uncontroversial or âcoreâ instances of inflexion enables us largely to skirt around the issue of distinguishing between inflexion and derivation. This longstanding problem has been tackled by numerous recent writers on morphology (e.g. Matthews 1974; M. Allen 1979; Plank 1981; Anderson 1982; Scalise 1984), and, at first sight, it may seem imperative that I should circumscribe the subject-matter of this book by offering clear criteria for the distinction. But the fact that I do not do so is not a serious deficiency, precisely because none of the claims that I will put forward hinges on where one draws the line between inflexion and derivation, or even on acknowledging that there is a line to be drawn. In other words, none of my generalisations, as presented, depends crucially on excluding âderivationalâ phenomena from its scope, and I leave open the possibility that these generalisations may be applicable to morphological behaviour which would traditionally be labelled âderivationalâ.
That said, one can nevertheless identify a kind of spectrum of morphological behaviour with âderivationalâ and âinflexionalâ extremes. Most linguists will probably agree in calling a morphological process (of affixation, for example) âinflexionalâ if it has all the following characteristics:
(a) it expresses a meaning (or realises a property) which all members of the relevant word-class can manifest (that is, the expression of that meaning is totally âproductiveâ);
(b) it is in complementary distribution with some other process or processes which realise the same property (that is, allomorphy is involved);
(c) the property which it realises is one of a finite set (or âcategoryâ) of mutually exclusive properties, one of which must be manifested in every word-form belonging to the relevant word-class;
(d) it does not alter the word-class membership of the forms to which it applies;
(e) it is syntactically relevant in the sense that the property it realises is involved in quite precisely specifiable âcooccurrence restrictionsâ with properties realised elsewhere in the sentence (for example, restrictions due to concord, government or âsequence of tensesâ).
In contrast, most linguists will probably agree in calling a process âderivationalâ if it has all the following characteristics:
(f) it is not fully productive (that is, there are some members of the relevant word-class to which it idiosyncratically fails to apply);
(g) no single property or âmeaningâ can be associated with it;
(h) it alters the word-class membership of the forms to which it applies;
(i) it is not syntactically relevant in the sense of (e) (except insofar as characteristic (h) implies syntactic relevance).1
The traditional difficulty of demarcation arises from the fact that few morphological processes display all and only the characteristics (a)-(e) or (f)-(i) respectively, and many display some characteristics taken from both sets. For example, the process of affixing the âagentiveâ suffix -er to verbs in English, which would traditionally be called âderivationalâ, does indeed have characteristics (h) and (i) but lacks characteristic (g) and would seem to possess characteristic (a). It may also lack characteristic (f), if we are prepared to accept in some contexts agent nouns in -er formed even from those verbs for which the usual corresponding agent noun has some other form (e.g. cycle, type). In contrast, the suffixation of -e to form the Plural of Afrikaans nouns, which would traditionally be called an inflexional process, does indeed have properties (a), (b) and (d), since all Afrikaans âcount nounsâ (as one might expect) can form a Plural which is syntactically still a noun, but only some of them do so by adding -e; on the other hand, this process lacks characteristic (e), since, perhaps alone among Indo-European languages, Afrikaans has no âNumber concordâ of any kind. âCoreâ examples of inflexional morphology, I suggest, are ones which share most of characteristics (a)-(e) and lack most of characteristics (f)-(i). The great majority of the morphological examples which I will be discussing will be unequivocally inflexional in this sense; but, again, nothing in the claims and suggestions that I will be putting forward makes it vital that I should avoid straying occasionally towards the derivational end of the spectrum.
I have designated morphosyntactic properties as the basic units of morphological content. What about the basic units of morphological expression (the morphological âsignifiantsâ)? In discussing the characteristics typical of the two ends of the morphological spectrum (inflexional and derivational), I referred to âmorphological processesâ such as affixation which might ârealiseâ morphosyntactic properties. I will in fact generally refer to morphological âsignificantsâ as inflexional realisations or inflexional exponents2 of morphosyntactic properties, or sometimes simply as inflexions. These apparently rather cumbersome terms are chosen in preference to, for example, âmorphemeâ or âmorphâ because they seem appropriate cover terms not only for affixation but also for such processes as infixation, ablaut, consonantal alternation, tonal alternation and reduplication, all of which may play a part in inflexion. For example, in the English words dogs, I would say that the morphosyntactic property Plural is realised by (or has as its inflexional exponent) the suffix -s (or [z]), while in the word men it is realised by ablaut or, more specifically, the substitution of -e- for the Singular formâs -a-. My definitions thus do not commit me to trying to identify a Plural âmorphemeâ or âmorphâ on the level of expression in a word-form such as man, where inflexion does not involve affixation. Another reason for avoiding the term âmorphemeâ is purely practical; it has been used in so many different senses that its use here would carry too much risk of confusion and misunderstanding, even if I defined carefully at the outset the sense in which I intended to use it myself. To a lesser extent, this is also true of the term âformativeâ, which I likewise avoid.
The term âallomorphyâ, which appears in the title of this book, is to be understood by reference to the more precise questions which I will be posing presently about the relationship between morphosyntactic properties and their exponents. To anticipate somewhat, I will be looking for evidence of constraints on certain deviations from the simplest conceivable pattern of exponence; and the deviations which I will have most to say about all involve the sort of behaviour that would traditionally be called âallomorphicâ. âConstraints on allomorphyâ is therefore a useful and relatively comprehensible shorthand for what, in my terminology, should more strictly be called âconstraints on deviation from the simplest conceivable pattern of relationship between morphosyntactic properties and their inflexional exponentsâ.
1.2 Method
My aim, as stated at the beginning of the introduction, is to propose and defend certain empirical generalisations about the relationship between morphosyntactic properties and their exponents. Any generalis...