Towards well-formedness conditions on -ize derivatives

Hiatus and stress

The first major problem we will discuss is the *VV constraint, postulated by Raffelsiefen to bar forms with adjacent vowels, such as *summaryize or truncated forms like *emphasiize. As we will shortly see, this constraint involves a number of empirical and conceptual problems.

Raffelsiefen herself mentions a few counterexamples (dandyize, Toryize, ghettoize, zeroize, echoize, statuize, virtuize, Zuluize). In view of these formations, Raffelsiefen tries to save her constraint by proposing a "correlative pattern Xize -> Xism" (p.201, note 21), i.e. for each word in -ism there is a word in -ize, such as Toryize - Toryism. It is, however, unclear what the status of correlative pairs should be in Raffelsiefen's model. Can correlative pairs override any constraint? And why should they be allowed to? It seems that Raffelsiefen evokes correlative pairs as an escape hatch whenever the model encounters systematic constraint violations that would otherwise falsify the proposed constraints (see, for example, the discussion of t-epenthesis below). The existence of correlative pairs is of course interesting, but it is a fact that is in need of an explanation itself. Thus, one could venture the hypothesis that the phonological and semantic restrictions -ize and -ism impose on their bases or derivatives are very similar or identical, which then inevitably leads to the existence of correlative pairs. In other words, correlative pairs present evidence for the operation of (similar or identical) constraints and not for their violation.

These problems aside, there is more empirical evidence against *VV that can neither be explained away with the notion of correlative pairs, nor can it be accounted for by the kinds of constraints proposed by Raffelsiefen. The following 20th century neologisms can be added to the *VV violators mentioned above: Maoize, radioize, jumboize, scenarioize, virtuize (Raffelsiefen's ghettoize and zeroize are also in the OED list of neologisms). Omitting words labeled as rare or obsolete, there are in addition numerous non-20th-century forms listed in the OED which further illustrate the proliferance of adjacent vowels in -ize words: altruize, centoize, cockneyize, flunkeyize, Harveyize, heroize, hinduize, libraryize, shintoize, toddyize, trollyize, vestryize. Among the -ize derivatives that are only attested as participials, we find dolbyize. From such data one can only conclude that *VV is either not operative, or that it must be ranked lower than IDENT, in order to allow the existence of adjacent vowels. But the truncation of vowels in words like summarize, memorize had required the opposite constraint ranking, so that Raffelsiefen's account faces an empirical problem here. Alternatively, the counterexamples would have to be labeled as idiosyncratic, which, as we will shortly see, they are not.¹

The counterexamples display a regularity which calls for an explanation. Thus, all of the attested *VV violators (except libraryize, a nonce formation in the OED's eyes, and scenarioize, both to be discussed below) share the striking characteristic that the base words are all disyllabic with initial stress, i.e. they have trochaic rhythm. This stands in remarkable contrast to the syllabic make-up of all the forms cited by Raffelsiefen to support *VV, which are all polysyllabic with antepenultimate primary stress, followed by two unstressed syllables (mémory, jéopardy, epítomy, su̒mmary, fa̒ntasy, etc.). In other words, it is only with the latter group of bases that the final vowel is truncated, but not with the former. Under Raffelsiefen's approach, these facts are coincidental, and the disyllabic non-truncating forms must be assigned the status of unsystematic exceptions.

It seems, though, that it is neither the constraint *VV nor the sheer number of base syllables that is responsible for the toleration of *VV violations, but rather the stress pattern. Thus, we can make the generalization that -ize attaches to vowels only if the base has penultimate main stress, which is trivially the case with trochaic bases. Although not attested in the OED data, it seems that trisyllabic vowel-final bases with penultimate stress may indeed take -ize, again contra the *VV predictions. This can be illustrated by the invented examples !muláttoize, !conféttiìze,² which appear to be at least phonologically, if not also semantically, well-formed. But even if some speakers reject these words, they would probably still prefer them to truncated forms like *mulàttìze or conféttìze, which would exhibit a stress clash. Notably, under Raffelsiefen's analysis the null-parsed candidate would emerge as optimal because M-PARSE is ranked lower than both * CLASH and *VV. Hence Raffelsiefen would wrongly predict that there is a gap.

Thus, it is clear that the truncation of final vowels cannot be the effect of a putative *VV constraint alone but must be the result of some interaction with other constraints. For the moment, let us capture the facts by the following generalization:

(1) 

For illustration of (1), consider a form such as fe̒deralı̒ze, which shows a stress lapse, and a form like *súmmaryı̒ze, which is ruled out by *VV in Raffelsiefen's model. The base féderal ends in a consonant, the stress lapse is therefore permissible. Items like *súmmaryı̒ze violate (1), because the last two syllables are unstressed. Derivatives with disyllabic trochaic bases such as dandyize do not show a stress lapse and are therefore in accordance with the generalization in (1), and therefore possible words (if they are otherwise well-formed).

We may now turn to another problem that occurs with vowel-final bases. It can be observed that there is one vowel that never occurs immediately before -ize, namely schwa. This generalization is formulated in (2):

(2) 

Violations of this constraints are avoided either by the occurrence of [t] between the putative base word and the suffix, as in rhematize, stigmatize, etc, or by truncation of schwa as in patinize. Raffelsiefen claims that "there is an undominated constraint ruling out any type of epenthesis in -ize formations" (p.200, note 19) and argues against a phonological rule of tepenthesis, "since it is unlikely that the nasal in -ma conditions tepenthesis". Again she attributes the data to the existence of correlative pairs (Xmatize -> Xma), a notion whose questionable value we have already discussed above.

In my view, two other observations are important. First, intermediate [t] does not only occur with base words ending in -ma, as claimed by Raffelsiefen. Thus, the OED also lists Asiatize. Second, i-epenthesis is restricted to words of Greek origin, which have stem allomorphs ending in [t]. This kind of stem allomorph does not only occur with -ize, but also with other suffixes, such as -ic (cf. asiatic, asthmatic, aromatic, etc. with concurrent stress shift) or -ous (as in eczematous, hematomatous). In all such cases we are dealing with Greek bases that have a bound stem allomorph which is exclusively used in word-formation.³

The preliminary generalization in (2) will be further discussed below. With regard to the insertion of consonantal material, Kettemann proposes a rule of n-epenthesis for items such as platonize or solemnize. Although there are at least two more forms with this kind of alternation (the OED lists Neronize, Plutonize in addition to the two mentioned by Kettemann), both the scarcity and the age of the derivatives (first attestation in the 14th to 17th century) indicate a lexicalized alternation.⁴ Kettemann's rule of [ɪn] insertion in derivatives like attitudinize is also lexically governed and does not occur in the neologism corpus.

¹ Note also that, contra Raffelsiefen, most of the putative counterexamples do not have a correlating form in -ism.

² I use '!' to mark unattested, but possible forms. Asterisks indicate impossible words.

³ It can be assumed that occasionally non-Greek base words like opera can behave like Greek words on the basis of analogy, if their phonological structure is similar. Opera - operatize is a case in point (the only one listed in the OED).

⁴ Note that the much younger form autumnize (first attested in 1829) does not show [n] insertion.