A Restricted Model of UR Discovery: Evidence from Lakhota∗

Under the standard approach to UR discovery (Chomsky and Halle 1968; Kenstowicz and Kisseberth 1977), it is assumed that whenever possible, surface contrasts should be derived in a lawful way from underlying distinctions. In other words, if a surface contrast ([A] vs. [B]) can be distinguished using an underlying difference (/A/ vs. /B/), this is preferable to using some other technique, such as marking all B’s with a diacritic (/A/ vs. /A/+[A→B]) or using one UR (/A/) and listing all B’s as lexical exceptions. In its most extreme version, this bias has lead some analysts to posit abstract, underspecified archiphonemes to avoid listing exceptions. In this paper, I argue against this use of underspecification. I propose a more restrictive model of UR discovery, which cannot use underspecified archiphonemes, but relies instead on exceptions. I discuss two separate but related changes in the history of Lakhota which are unexpected under an underspecification account, but which are predicted by an exception-based analysis.