Analogy as optimization: “exceptions” to Sievers’ Law in Gothic Paul Kiparsky Stanford University 1. Lexical representations as a site of optimization Suppose the phonological rules/constraints of the language are such that under- lying /A/ and /B/ lead to to the same output [A]. [1] Underlying: /A/ /B/ Output: [A] Lexical Phonology and Morphology (LPM) dictates that non-alternating [A] is then analyzed a. as underlying /A/, other things being equal, but b. as underlying /B/, if /B/ conforms better to the constraints on underlying representations. Case (a) has been familiar for a long time, and is supported by a fair amount of historical evidence (Kiparsky 1968, 1973). It was adopted by NGG (Vennemann 1972, Hooper 1976) and by Natural Phonology (Stampe 1972/1980). Prince & Smolensky 1993 dub it lexicon optimization, and show that it is a consequence of basic assumptions of OT. It is case (b) that is controversial. Although it follows from LPM, where constraints on the phonological inventory or morpheme structure of a language are defined by its lexical phonology and morphology, it does not follow from theories such as those assumed in much current OT phonology, which define optimality only on output representations, and claim that the structure of the lexical input is derivative just from those constraints. Therefore evidence for (b) also calls into question the adequacy of such output-oriented theories. This paper will contribute such evidence, in the form of analogical changes at the level of lexical (underlying) representations, driven by constraints dominated 1