Abstract
This paper elaborates on Chomsky's (2013, 2015) framework and revisits sentential subject constructions in English. The first part of the paper refines the Labeling Algorithm, demonstrating that feature sharing requires agreement in which each of the merged non-heads bears an unvalued feature that is consequently deleted. The second part of the paper analyzes sentential subject constructions in terms of the refined the Labeling Algorithm. A novel analysis is proposed in which C and sentential subjects in SPEC-T are required to bear topic-related features for labeling success. In conjunction with the refined Labeling Algorithm, the proposed analysis successfully captures a number of peculiar properties of sentential subjects.
