Unlike vowel-final dactyls, vowel-final trochees never lose their final vowel under -ize attachment (cf. dándyìze). According to the constraints introduced so far, this contrast is unexpected since the deletion of the final vowel (as in *dándìze) would equally lead to a better satisfaction of the higher ranked R-ALIGN-HEAD. In view of these facts there must be yet another constraint that supersedes R-ALIGN-HEAD, thereby mitigating against truncation. I propose that truncated candidates like *dàndìze are ruled out because they violate a constraint against the adjacency of a stressed syllable and the prosodic head of the word, *CLASH-HEAD.

(1) 

This constraint is identical with Pater's STRESS-WELL (see also Halle and Vergnaud 1987:238), but I have chosen what I consider a more mnemonic label for it. In essence, * CLASH-HEAD is a stronger version of the well-known prohibition of adjacent stresses (see Prince 1983, Hammond 1984, or Raffelsiefen's *CLASH constraint). *CLASH-HEAD must be ranked above R-ALIGN-HEAD, to the effect that the better satisfaction of R-ALIGN-HEAD does not save the candidate violating *CLASH-HEAD:

Dandy-ize¹

There is reason to believe, however, that an additional constraint is at work which only becomes visible with disyllabic schwa-final bases (and with bases ending in a stressed vowel, as we will see below). This constraint has the effect that it prohibits the adjacency of schwa and a following vowel. Although no examples of schwa-final disyllables can be found among the neologisms, evidence for the operation of such constraints can be gleaned from older attested forms and from nonce words. The OED lists the following formations on the basis of schwa-final bases (none of them 20th century innovations): hebraize, judaize, Mithraize, Utopia-ize. The interesting thing now is that the first three forms are given with a pronunciation in which the schwa-syllable of the base is pronounced [eɪ] in the derived word. The OED does not give a pronunciation of Utopia-ize, but its spelling with a hyphen strongly suggests a phonological oddity, namely the insertion of a glottal stop. If forced to pronounce nonce words such as !moraize, native speakers insert a glottal stop between base and suffix. Thus, it seems that concerning new formations, consonant epenthesis (i.e. insertion of a glottal stop) is optimal with schwa-final trochees of non-Greek origin. In contrast to that, schwa deletion is optimal with dactylic bases.

The constraint prohibiting the adjacency of schwa and -ize is a special kind of hiatus restriction, which is most likely the effect of a number of different constraints on featural and syllabic constraints (see, for example, Casali's 1997 OT account of hiatus resolution in a number of different languages). Since the details of the English schwa-vowel hiatus restriction are beyond the scope of this topic, I will use *SCHWA-V as an abbreviation of what is probably the combined effect of a number of interacting constraints:

(3) 

*SCHWA-V as a constraint has not been proposed before in the OT literature, but can be independently justified by the allmorphy of articles in English. As is well-known, both definite and indefinite articles have two allomorphs, whose use is governed by the following segment. A vowel triggers [ði] or [ən], respectively, whereas a following consonant demands the allomorphs [ðə] and [ə], respectively. What unites the different allomorphies of definite and indefinite articles is the fact that schwa may never precede a vowel. I argue that this generalization can be captured by the proposed constraint *SCHWA-V, which is not only operative with articles but also with (certain) suffixes. Note that the constraint becomes active only below the level of the prosodic word and the clitic group (Nespor and Vogel 1986), as is shown by the unproblematic occurrence of schwa-vowel between phonological words, as in Indian[ə ɪ]s wonderful,² Suffixes like -ize only expand prosodic words, but are not prosodic words themselves.

With regard to the ranking of *SCHWA-V, the surface form of the potential word Imóraìze shows that *SCHWA-V must be ranked above R-ALIGN-HEAD:

*SCHWA-V cannot be replaced by a more general constraint that prohibits hiatus (such as Raffelsiefen 's *VV), because, as we have seen, vowel-final trochees that end in a different vowel behave differently. The optimal candidate in (4) obviously violates the constraint against epenthesis, DEP, which must be ranked below *SCHWA-V and *CLASH-HEAD. DEP does not interact with R-ALIGN-HEAD:

That glottal stop insertion is triggered by *SCHWA-V and *CLASH-HEAD, and not by a high-ranking ONSET is evidenced by verbs such as dándyìze, which do not necessitate glide insertion.³

The proposed constraint rankings solve the problem of the non-unifomity of vowel-deletion with V-final trochaic bases and V-final dactylic bases. In essence, disyllabic bases are never truncated because this would lead to a clash of secondary and primary stress in the derivative.

¹ Dandyize is chosen as the canonical example of a vowel-final trochaic base although it is not a 20th century neologism. The reason is that, in spite of their being fully systematic coinages, the corpus did not contain the relevant forms. As pointed out above, the corpus of attested participials features the parallel form dolbyize, and similar nonce forms are equally possible.

² If such sequences occur across phonological words, so-called 'intrusive r' can be inserted in non-formal speech (e.g. idea[:]of). Note, however, that the class of possible vowels allowing intrusive r is not restricted to schwa but includes all non-high vowels (e.g. Giegerich 1992:283).

³ There is a pronunciation possible which involves glide insertion, as in dandy[j]ize. In this case ONSET would rank above DEP but still below *CLASH.