Lexicon Optimization Reconsidered

Abstract

This paper takes up the hitherto neglected issue of the lexicon, and shows how the lexicon can be projected from the structure of a language's grammar, seeking to resolve the tension between a desire to minimize lexical specification and a desire to maximize input-output faithfulness in terms of the organization of the lexicon. First, I will show that the earlier version of Lexicon Optimization, proposed by Prince and Smolensky (1993), results in a less economical lexicon and argue that predictable feature values must be unspecified in the lexical representation irrespective of whether they show alternating patterns or not. And then, through a dose examination of various allophonic and allomorphic cases, I will propose that Lexicon Optimization must be properly reformulated from a form-by-form optimization to a global optimization reflecting the segmental inventory and morphemic paradigm. The main point of this paper is that the presence of a lexical economy constraint *SPEC has some consequences for the more restrictive lexicon.