Restrictions on productivity

Rules and restrictions

One of the central questions concerning productivity is whether the productivity of a given rule can be determined solely on the basis of the properties and restrictions of the rule. In other words, is productivity an epiphenomenon that is the predictable result of other, underlying, phenomena? I will argue that structural restrictions are an important factor influencing productivity, but not the only one.

Let us look again at Booij's justification of the view that productivity is a derived notion, which is repeated here for convenience: "The degree of productivity of a WF-rule [word formation rule, I.P.] can be seen as inversely proportional to the amount of competence restrictions on that WF rule" (Booij 1977:5). According to this position, one would only have to define the word formation rule with its proper restrictions and the degree of productivity would naturally fall out. Disregarding the non-trivial question how to measure "the amount" of structural restrictions across different derivational processes, it seems that Booij's position is untenable in this strong form.

However, following Booij (1977) and Plank (1981), I maintain that apparent gaps in derivational patterns can in most cases be explained in terms of the structural properties of the process, so that in many cases reference to language use or 'norm' are premature and ill-justified. Therefore, if we talk about the structural properties of a word formation rule, its productivity is crucially dependent on the restrictions imposed on the bases or the derivatives, and therefore an epiphenomenon. This does not mean, though, that one could entirely predict the application rate of a process on the basis of the structural properties. As we will shortly see, the application rate is indeed often governed by extra-linguistic factors. In terms of the individual speaker's behavior, the implementation of a word formation process depends on whether other speakers also use this process to coin new words. In this perspective, the application rate is a property of morphological rules that cannot be derived from linguistic structure, but only from language usage. Hence, we must distinguish between, on the one hand, the possibility to apply a process to a given word, which is indeed crucially constrained by structural restrictions and therefore a derived property, and, on the other hand, the use of newly coined derivatives in speech that are licensed as possible words by the restrictions. The following examples illustrate this point.

Let us consider two different cases of unproductive categories, nominal -th and verbal -en. In principle, only intensionally defined categories can be productive, but not extensionally defined ones. Hence lexically-governed rules like English nominal -th (as in length) or -al (as in arrival) suffixation are unproductive. But even intensionally-defined categories may be unproductive, as is the case with English verbal -en. Thus, although -en suffixation is a well-defined process in terms of its morphological, semantic, and phonological properties, it is practically dead. This example nicely demonstrates that the non-applicability of a process cannot be inferred on the basis of structural knowledge, and that therefore a statement like Booij's needs further qualification.

If the application rate cannot be inferred from the structural properties of the rule, how do speakers (come to) know that some intensionally-defined categories can be used to coin new lexical items whereas others cannot? To my knowledge, this question has not been addressed systematically by anyone working on productivity. I will therefore venture some hypotheses here.

As was pointed out, productive processes are semantically (and phonologically) transparent and have a high proportion of low frequency types. Psycholinguistic experiments using subjective frequency ratings such as Schreuder and Baayen (1997), Baayen (1997) have shown that, quite surprisingly, speakers have robust knowledge about the frequency of lexical items, which means that speakers are able to glean from their linguistic environment information about the application rate of a given morphological process. If a process does not meet the criteria of transparency and high quantity of low-frequency items, the speaker infers that this process cannot be used to coin new words. This kind of mechanism can also explain the decline of even those morphological categories that are (still) transparent. If, for some reason, the number of transparent low-frequency items in the environment drops, this is an indication for the speaker that the process has become less productive.¹ In this case the speaker may prefer other means of expression, such as rival morphological processes or syntactic constructions.

In any case, what has been labeled 'application rate' is also an inherent and changeable property of morphological processes, and a property which cannot be completely derived from the structural properties of the processes, but which depends also on pragmatic factors.

Perhaps the most obvious of these factors is fashion.² The rise and fall of affixes like mini- or -nik is an example of the result of extra-linguistic developments in society that make certain words or morphological elements desirable to use, and not the consequence of structural mechanisms (although these are operative in constraining the shape of these words). Ruf (1996), for example, finds a remarkable increase in the use of the German augmentative loan prefixes Mega-, Giga- and Supra- in a very short time span, a fact that cannot be explained satisfactorily by structural factors, since these affixes seem to be no more structurally constrained than some of their augmentative competitors. But even in these cases structural fac tors should not be neglected, because on closer inspection the many rival augmentative prefixes are not completely identical in their semantic and phonological properties, which in turn influences their applicability. Thus, the most productive ones seem to be the ones that are also semantically least restricted.

Another general pragmatic constraint is that there must be a need for a new form. This need is a reflection of two main functions new words can have, their capacity of labeling a new concept or referent, and the condensation of information, for example for stylistic purposes or reasons of text cohesion. Kastovsky (1986) calls this latter function 'recategorization' because very often the syntactic recategorization of a word, e.g. the nominalization of an adjective, serves exactly this purpose. The example in (1) is a case in point:

With respect to the relation of these functions to productivity, Kastovsky claims that the recategorization function is the most frequent in speech, so that processes serving this function tend to have a high application rate. Another function (not discussed by Kastovsky) is especially prominent with evaluative affixes, the attitudinal function. By attaching, say, a diminutive suffix to a name, the speakers primarily express their attitude, and do not aim at labeling a new concept or making the discourse more coherent (see below for more discussion of diminutives). Interestingly, diminutive affixes are not category-changing, which could be interpreted as the direct consequence of the lack of the other two functions (concept formation and information condensation).

Another pragmatic restriction on productivity is discussed in Lipka (1977), who introduces the concept of hypostatization, which basically says that the existence of a word suggests the existence of an entity to which it refers or which it denotes. Hence, one can only label something that does exist (in a wide sense, allowing also for fictional existence), and any new derivative must have some kind of referent or denotatum. The crucial problem with this constraint is, however, that it does not really constrain anything, since, as Lipka points out, speakers may 'create' a referent by forming a new derivative. This phenomenon is also known as reification and can be observed on a daily basis in political discourse.

Another non-structural requirement new lexemes must meet is that they denote something nameable. Although the nameability requirement is rather ill-defined, it captures a significant insight: the concepts encoded by derivational categories are rather simple and general and may not be highly specific or complex, as illustrated in the famous example of an unlikely denominal verb forming category given by Rose (1973:516): "grasp NOUN in the left hand and shake vigorously while standing on the right foot in a 2.5 gallon galvanized pail of corn-meal-mush". An operationalized definition of the nameability requirement remains to be devised.

Language purists may also show evidence of a pragmatic restriction referred to as neophobia, a notion whose content and theoretical status is utterly obscure. An equally vague notion is euphony, which should be replaced by explicit accounts of the phonological restrictions at work. In other words, euphony is the result of structural, and not of pragmatic restrictions.

Given the rather unspecific and ill-defined nature of most of the pragmatic restrictions proposed in the literature, it seems that a lot remains to be done to pin down these factors in a more explicit manner. I will, however, not engage in this enterprise, but turn instead to structural restrictions. The following general considerations are responsible for choosing this focus. No matter which determinants govern the application rate of a rule, it seems that we cannot arrive at significant insights into the role of aspects of language use in morphology without determining which structural constraints operate on that rule. This position is most radically formulated (again) in Booij (1977:6), who states that "we cannot say anything specific about the role of performance factors before we have investigated which competence factors restrict the productivity of a rule". In other words, we should first aim at describing the class of possible derivatives of a given category as precisely as possible in structural terms, and then ask ourselves which pragmatic factors influence its degree of exhaustion.

Let me illustrate this point with an instructive example of the interaction of the two types of restrictions, as provided by K. P. Schneider (1997), who investigates both the structural and pragmatic aspects of diminutives in English.³ After having outlined the phonological, morphological, semantic and syntactic properties of the different suffixes, he investigates their use in speech, and demonstrates that the application rate of English diminutives is actually constrained by the type of speech act and the kinds of participants involved in the verbal interaction.⁴ Crucially, however, the structural restrictions on the affixes he discusses are obeyed no matter how the diminutives are used. K. P. Schneider's work demonstrates that when starting out from a clear picture of the structural aspects, researchers do not easily fall victim to the fallacy of explaining certain phenomena as conditioned by language use or by some rather ill-defined notion of norm. As will become clear, even in a well-described language like English a lot of work remains to be done to describe more accurately the structural properties of many derivational categories. We attempt to provide such a description for some of these categories, thereby shedding some new light on two major problems concerning the structural properties of word-formation rules in English, namely the problem of stacking restrictions and the problem of rival derivational processes. Stacking restrictions regulate the combinability of affixes amongst each other, and it is far from clear which mechanisms are responsible for the observable patterns. An equally notorious problem of morphological theory is the nature and distribution of rival morphological processes. Before we turn to the empirical investigation of these problems, I will outline some of the theoretical notions and issues that are relevant for an understanding of the kinds of restrictions and mechanisms that we will deal with.

¹ One reason could be that, for example, the number of possible bases has decreased. In the case of -en, for example, the number of monosyllabic adjectives is not infinite, so that, after the majority has undergone the process, new formations are increasingly rare.

² Cf. the title of Algeo (1971): "The voguish uses of non-"

³ K. P. Schneider's investigation shows, among other things, that English has a whole range of productive diminutive suffixes, contrary to the commonly held assumption that this language largely lacks expressive or evaluative word-formation.

⁴ See Dressier and Barbaresi (1994) for another pragmatically-oriented approach.