Disjunctive Lexical Stratification

Abstract

In many languages, loanwords allow patterns that are absent in the native phonology. The phonotactic requirements for loanwords may thus be less restrictive than the ones for the native vocabulary. For instance, native items in Turkish only have voiceless obstruents in codas, while both voiced and voiceless obstruents are found in loanword codas (Inkelas, Orgun, and Zoll 1997). The opposite pattern, in which loanwords only allow a subset of the structures attested in native words, is cross-linguistically rare but attested (Nadasdy 1989, Kawahara, Kohei, and Ono 2003, Kenstowicz 2005). For example, in Latvian, [ae] can only occur in native words, but not in loanwords, where it is replaced by [e], for example, [manxetena] ‘Manhattan’ (Gelbart 2005). Here, I present the third type of language, in which the sets of native and foreign phonotactic patterns are disjunctive. In Slovenian, there is no overlap between the sets of front vowels appearing before tautosyllabic [R] in native words and in loanwords. Thus, loanwords in Slovenian are more exceptional with respect to native words than previously thought possible. The Slovenian data have implications for the theory of exceptionality. Since the advent of Optimality Theory, exceptional loanword patterns have been modelled using either cophonologies (Inkelas, Orgun, and Zoll 1997, Anttila 2002, Inkelas and Zoll 2007) or indexed constraints (Ito and Mester 1995b, 1999, 2001, Pater 2000, 2007, to appear). Cophonologies allow independent constraint rankings for different groups of words (lexical strata). Indexed constraints, on the other hand, are limited in their domain of application to a particular set of morphemes. Pater (2007, to appear) argues that indexed constraints are more restrictive than cophonologies. The crucial difference between the theories is