Rival morphological processes 1:

The productivity of verb-deriving processes

Introduction

In English there are a number of morphological processes by which verbs can be derived from items of a different syntactic category (mostly adjectives and nouns). Disregarding the affixes that add a negative, privative, or reversative element of meaning (de-, un-), the following category-changing verb-forming affixes are attested: the prefixes en-, em- and be- (sometimes accompanied by the suffixation of -en), the suffixes -en, -ize, -ate, and -ify, and a process of conversion. (In what follows I will refer to the latter process also as a kind of affixation or 'zero-derivation', but only as a matter of terminological convenience and not out of theoretical conviction). It is this group of processes.

The verb-deriving processes in English have been the topic of many studies, but three major problems are still unresolved: the problem of meaning, the problem of rivalry and the problem of productivity. There is only one study to date which has tried to tackle these questions in a systematic fashion across affixes, E. Schneider (1987). In that paper, a number of interesting observations are made, but many questions remain unanswered. Thus, Schneider only investigates the suffixes -en, -ize, and -ify, but not their other potential rivals, eN-, -ate or conversion. Furthermore, Schneider's generalizations (and the numerous exceptions he mentions) remain unexplained. The present investigation is an attempt to provide more satisfactory solutions to the three above-mentioned problems, which will be discussed in the following paragraphs.

The first of these problems is how many rules or processes we are actually confronted with. Are the suffixes allomorphs of one or two underlying morphemes, or are we confronted with a variety of rather independent processes? A crucial point in this discussion is of course the meaning of derived verbs, which are remarkably diverse not only across affixes, but also concerning one single affix, for which often more than half a dozen meanings have been postulated. Derivatives involving the suffix -ize, for example, seem to be quite heterogeneous in their semantics and in the kinds of bases they attach to. Thus, -ize can attach to adjectives and nouns, the resulting derivatives may be transitive or intransitive and they may have one or more of the following meanings (cf. e.g. Marchand 1969:320; X stands for the base word): 'render X, make X, convert into X, put into the form of X, give the character or shape of X, subject to the action, treatment or process of X, subject to a process connected with X, impregnate, treat, combine with X, act in a way characterized by X, imitate the manner of X'. Given a somewhat similar situation with the other affixes, how many processes should we distinguish, in view of the abundance of different affixes, different bases, different meanings and different syntax of the derivatives?

So far, no convincing account of the polysemy or homonymy of the affixes has been proposed, although individual affixes have been studied extensively. Two diverging approaches can be found in the literature. Some studies take the supposedly central meaning as their starting point and investigate the range of affixes that may express this meaning and discuss their formal properties. The problem with such approaches is that the polysemy of the affixes is not taken into account, in spite of the fact that this may influence the behavior of the affixes. Furthermore, the status of such a general rule may be questionable. Gussmann, for example, assumes that there is a so-called rule of derivation, i.e. a "uniform semanto-syntactic formula whereby adjectives become verbalized" (1987:82) with the meaning 'make (more) X'. While the postulated meaning is intuitively plausible, the rule fails to explain why so many of the verbalizing suffixes also take nouns and bound roots as their bases. In such a framework we would have to posit yet other rules, an approach that suspiciously calls for Occam's Razor.

In a more general perspective, Gussmann's analysis represents an approach to morphology that separates the meaning of affixes from their phonological spell-out. The most pronounced theory of this kind is the so-called 'lexeme-morpheme base morphology', developed by Robert Beard over the last two decades (e.g. 1981, 1987a 1987b, 1988, 1990a, 1990b, 1994), slightly modified versions of which have been proposed by Szymanek (1985, 1988), and Don (1993). The central claim of these authors is that the frequent mismatches of form and meaning in morphology are best accounted for if both aspects are strictly separated. This has been come to know as the 'separation hypothesis', which claims that word formation rules split into two different kinds of rules: the categorial and semantic operation is done by so-called lexical (or derivational) rules, whereas the phonological operation is performed by a morphological rule which simply adds the appropriate phonetic material or manipulates the base form in some other phonetically relevant way (e.g. by ablaut). By separating the meaning from the phonetic realization the theory can elegantly account for the synonymy of different affixes and for the polyfunctionality of individual affixes. In the case of derived verbs, there would, for example, be a lexical rule (or 'derivational category' in the terminology of Gussmann or Szymanek) CAUSATIVE which is paired with a number of morphological, or 'affixation', rules, i.e. [aɪz], [ɪfaɪ], [eɪt], [ən] (Gussmann 1987, Don 1993:66). The crucial assumption of the theory is however the absolute synonymy of the processes (e.g. Beard 1990b: 162). As we will see, this assumption is most problematic.

Non-separationist accounts, called 'direct articulation models' by Beard, maintain the idea that there is a direct link between the form and the meaning of affixes but often fail to present an account of whether or how these meanings are related (cf. e.g. Lieber's (1996) study of -ize).

In what follows, the meaning of each affix will be investigated on the basis of a large collection of forms, and an explicit account of their polysemy will be proposed.¹ The comparison of the semantics of the individual affixes will reveal that the affixes are only partially synonymous and that there are compelling arguments for the postulation of a number of different affixation rules for verb-forming affixes.

The second problem is the problem of rivalry. Assuming that at least some of the processes can be shown to be semantically overlapping, the question arises how the affixes select their bases (or the bases select their affixes). The data in (1) illustrate that there must be certain restrictions at work:

Basically, two different kinds of mechanisms are conceivable, syntagmatic and paradigmatic. We will see that it is primarily the properties of the individual processes in combination with the properties of their base words, that are responsible for the distribution of the different affixes. Paradigmatic mechanisms are of less relevance, in fact type-blocking will be demonstrated to play no role at all, which seriously questions its usefulness as a theoretical notion and descriptive tool.

This brings us to the third of the above-mentioned problems, namely productivity, which is of course closely related to the second. Which of the processes are (still) productive, and which factors are responsible for the differences in productivity among those affixes that are still used to coin new words? While it seems clear that some of the processes are more productive than others, a clear assessment of their productivity is still lacking. On the basis of dictionary data and text corpus data we will close this gap in the description of these affixes.

Starting out from these three central problems I will proceed as follows. The productivity of the different processes will be measured, making use of the Cobuild corpus and the OED. We will then scrutinize very closely the structural proper ties of the individual processes, which will finally lead to a solution of the two remaining problems, meaning and rivalry.

As outlined in Productivity: Definitions and measurements, there are a whole range of measures available to the researcher, of which Baayen's corpus-based measures and Neuhaus's dictionary-based counts seem to be the most promising. These two methods will be employed below. In the earlier discussion of the different productivity measures I already briefly hinted at the methodological problems inherent to Baayen's and Neuhaus's measures, and much of the examination to follow will further illustrate these points. Hence, this topic does not only provide the first explicit quantitative account of the productivity of verbal derivation in English, but also contains a detailed discussion of hitherto unmentioned methodological problems involved in the application of productivity measures.

We will start with the dictionary-data, then move on to the text corpus data, and finally compare the results of the two analyses.

¹ An earlier version of my account of the polysemy of -ize derivatives has appeared as Plag (1997).