|Title:||Compensatory Lengthening in Harmonic Serialism|
|Abstract:||Informally, compensatory lengthening (CL) is a phenomenon that consists of two parts: the deletion of a weight-bearing segment, and the concomitant lengthening of another, usually adjacent segment. Unsurprisingly, analyses of CL are usually derivational in that lengthening only occurs after it is triggered by deletion of a neighboring segment. Less discussed is the fact that, because CL is triggered only by the loss of moraic segments, both of these steps must refer to a syllabified and moraically specified form. Only coda consonants can be associated with moras, and a particular segment’s status as a coda can only be determined after the form has been syllabified. It is crucial, therefore, that syllabification and assignment of moras happen before segmental deletion.
In constraint-based phonology, this requirement for prosodic structure means that faithfulness vio- lations must be evaluated with respect to a form that contains prosodic structure at least up to the level of the syllable. Doing so is not possible in classic Optimality Theory (OT, Prince and Smolensky 1993/2004), a framework that disallows intermediate levels of representation. Moreover, it is not immediately obvious that Harmonic Serialism (HS; McCarthy 2000, et seq.), an iterative version of OT, allows for this particular intermediate stage. Even in HS, there is no guarantee that the correct syllabic and moraic structure will be built before segmental changes occur. I will show that a ranking paradox prevents the building of prosodic structure from intrinsically preceding deletion of segments. Similarly, it is not possible to use gradual deletion of consonants to force the consonant to remain in the derivation long enough to obtain the proper moraic specification. I argue, though, that an analysis of CL is possible in HS if we make the auxiliary assumption that the derivation begins with a fully faithful candidate (FFC, McCarthy 2007). I will then show that HS can derive the most canonical type of CL using a mora-sharing approach. Some of the more complex interactions between deletion and lengthening can also be analyzed straightforwardly in this framework. Cases that the analysis cannot account for, namely nonlocal interactions, are extremely rare, and it is a strength of the analysis that it can account for most variations on canonical CL without difficulty.