|Title:||Deletion, Insertion, and Symmetrical Identity|
|Comment:||Appeared in Harvard Working Papers in Linguistics 7, 1999.|
|Abstract:||In the Correspondence Theory of faithfulness as originally proposed by McCarthy & Prince (1995), faithfulness to a segment's feature specifications is regulated by symmetrical IDENT[f] constraints: a change within a segment from [+f] to [-f] or from [-f] to [+f] is assumed to violate one and the same IDENT[f] constraint. This view of featural faithfulness, which deprives features of much of their autosegmental independence, has been argued against by a number of authors. For instance, Pater (to appear) gives an argument for asymmetrical IDENT[f] constraints, still binding featural faithfulness to segmental correspondence but partially recognizing the independence of opposite feature values (see also McCarthy & Prince 1995:5.1, to appear:5.4). More radically, Lombardi (1995 et seq.) and others have argued for full-blown featural correspondence, freeing features completely from their segmental anchors. In this paper I bring another set of facts to bear on the question of featural faithfulness, arguing in favor of the original, symmetrical IDENT[f]. This set of facts concerns the distribution of [r] in Eastern Massachusetts English (Whorf 1943, Venneman 1972, Kahn 1976, McCarthy 1991, 1993, Blevins 1997, Halle & Idsardi 1997).
The well-known facts concerning the distribution of [r] in Eastern Massachusetts and other dialects of English have resisted explanatorily adequate analysis primarily because they involve the insertion of a generally unexpected epenthetic consonant, [r]. The fact that underlying [r] is also deleted in a complementary set of environments (and retained otherwise) is clearly relevant, as originally noted by Vennemann (1972), but no synchronic analysis of these facts to date has connected them to each other in a satisfactory way, claims to the contrary notwithstanding (see e.g. Halle & Idsardi 1997). I propose a novel interpretation of these facts within OT and lay out the details of an analysis of them that meets satisfactory standards of both descriptive and explanatory adequacy.
Following up on a proposal originally made by Kahn (1976:69-70; see also Broadbent 1991 and Gnanadesikan 1997:159-162), I analyze [r]-insertion as the diphthongization of a vowel, where diphthongization is here technically understood as a relation between one segment in the input and two segments in the output. This imperfect correspondence violates some IDENT[f] constraint(s), where 'f' is a feature or features not shared between [r] and the vowel it forms a diphthong with. Similarly, [r]-deletion is analyzed as the coalescence of [r] with a preceding vowel, where coalescence is a relation between two segments in the input and one segment in the output. This equally imperfect correspondence violates the same IDENT[f] constraint(s) that [r]-insertion violates, and it is this connection between the two processes that explains why [r] and not some other consonant is inserted in Eastern Massachusetts English speech.