The core of Optimality Theory (Prince and Smolensky 1993) resides in the hypothesis that constraints are universal, potentially conflicting, and ranked. Constraints can be violated in well-formed sentences. Where two constraints conflict on a given input it is the ranking of the constraints that determines which of the available options is grammatical. Grammars are nothing more than a ranking of the set of universal constraints. Within this general theoretical framework, this talk offers an analysis of several cases of apparently odd lexical choices. Why does English "choose" to have an auxiliary "do" and an expletive "it"? Why do the Romance clitic combinations show unexpected forms?

First, I will show that the distribution of the English auxiliary "do" can be understood quite precisely in terms of constraint conflict. The occurrence of "do" violates a constraint of Full-Interpretation, because "do" has no semantic analysis. The verb thus occurs only when a higher ranked constraint, such as Obligatory-Heads, is satisfied by its presence and violated in its absence. "do" is thus possible only when necessary. From this perspective it is not a lexical accident that English "has" a semantically empty auxiliary. Rather it is a consequence of the grammar of the language, i.e. the ranking of the constraints of UG, which forces the regular verb "do" to appear, but without its meaning. The hypothesis is that "do" minimally violates Full-Interpretation: any other verb would violate it more, having a more highly specified semantics which is unparsed, or unanalyzed, when the verb is meaningless. Every language with the (relevant) constraint rankings must have "do", no language with crucially different rankings can have it. The appearance of empty "do" is far from being a language particular lexical fact.

A similar point holds for other expletive elements. Based on joint work with Vieri Samek-Lodovici, I will argue that the appearance of the expletive "it" in English has the same analysis, fundamentally. English ranks Full-Interpretation below the constraint(s) requiring a filled subject position. Hence the grammar of English prefers using a nominal without its meaning to leaving the subject position unfilled. Italian, with a different constraint ranking, makes the opposite choice. But this is not a fact about the lexicons of English and Italian, it is a fact about their grammars. Here again, the hypothesis is that "it" rather than some other nominal element occurs because it represents minimal violation: stripping the meaningful pronoun "it" of its semantics is a lesser violation than stripping any other nominal of the language.

In order to explore this idea further, and achieve a more principled view of the way optimization affects lexical choice, I will extend the basic idea to the "opaque" clitics in Romance (Bonet, NLLT 1995). These are cases where the clitic pronoun occurring in a sequence of clitics is not the one to be expected on the basis of the clitic pronouns as they occur in isolation. I will argue that an optimality theoretic account can explain some of the fundamental properties of this system. The basic idea is that there is a constraint against adjacent occurrences of identical forms (*XX). Whether the identity at issue is phonological, morphological, or both, is an issue to be discussed. This constraint conflicts with faithfulness constraints, which require the clitic which best analyzes (is most faithful to) the input to be selected. When *XX dominates, an opaque clitic must occur.

An example can be found in Italian, where the impersonal subject clitic pronoun is "si", but this is also the 3rd person reflexive. When both occur in a single sentence, instead of "si si" we find "ci si", where the impersonal subject is not realized as it would be in isolation. The hypothesis is, then, that the perfect clitic, namely "si", is not available in this situation, because of the effects of *XX. The clitic which does occur represents the best the language can do in the situation. The chosen clitic involves minimal violation of the morphological faithfulness constraints. Constraint re-ranking explains the existence of considerable cross-dialectal and cross-linguistic variation in this system.

In all of these cases, if the argument is correct, the actual choice of a lexical item is determined by the grammar of the language. In all of these cases, the lexical item that occurs is the one that minimally violates the regulating constraints: Full-Interpretation for "do" and "it" and the faithfulness constraints in the case of the opaque clitics. This proposal, in which lexical choice is systematically optimized, cannot be instantiated without a well-defined theory of optimality.