Copy invisibility and (non-)categorial labeling

Abstract

In contrast to dominant views that the labeling algorithm (LA) detects (i) only the structurally highest copy of a moved object, or (ii) detects all copies, we propose and defend a third option: (iii) all copies are invisible to LA. The most immediate consequence of this is that objects formed by Internal Merge cannot serve as labels. We relate this proposal to a particular reinterpretation of LA theory such that LA constructs only categorial labels, barring the construction of <Q, Q> and <phi, phi> configurations. We then propose an interface condition, Equal Embedding (EE), under which agreeing features must be equally as embedded in order for interpretation to be licensed. We argue that EE appears to fall out of minimal search requirements. We then propose a principled distinction between Agree and LA, based on their sensitivity to copies and interface relations: Both Agree and LA involve minimal search (Probe-Goal for Agree; categorial feature-detection for LA); however, copies are invisible to LA but not to Agree, and LA involves a CI relation (category-specific interpretation) whereas Agree involves an SM relation (the morpho-phonological process of feature-valuation). (University of Texas Health Science Center . Gyeongsang National University)