Workspace as a source of hierarchy in extended projections

Extended projections (EPs) in natural languages have several properties which have not yet been explained. (i) EPs conform to a Hierarchy of Projections (HoP), a crosslinguistically similar hierarchical arrangement of semantically grounded categories. (ii) Each token of an EP is linear, in the sense that it has a single dimension (with specifiers and adjuncts stripped out, it is a string). (iii) HoPs are rooted in a lexical category with conceptual content at the bottom, and a succession of functional elements above. I argue that these properties motivate a particular architecture of the workspaces in which sentences are constructed. I model the workspace as a Finite State Automaton (FSA) with a monotonicity property which underpins hierarchy. The FSA starts from a lexical category (cf. (iii)), `projecting' it into an EP. Transitions correspond to applications of Merge, and states are stages in the derivation. The sequence of states in a path from start to accepting state is a string, the EP (cf. (ii)). The HoP is then the entire FSA, arranged with the start at the bottom and the final state, the complete clause or noun phrase, at the top (cf. (i)).