Phases are Read-Only: Evidence from Hindi-Urdu

Abstract

Under Chomsky (2000, 2001)’s Phase Impenetrability Condition, phases induce Transfer of their complements, rendering the complements inaccessible. As a consequence, cross-phasal dependencies are ruled out. Recent work onphases has suggested that instead of being eliminated, transferred phase complements remain in the syntax (Bo?skovi´c 2003; Obata 2010; Chomsky 2012; Chomsky et al. 2019). In this paper, I expand on the idea of spelled out phase complements being visible but not completely accessible for syntactic processes. I propose a Read-Only view of phases, wherein phase complementsare not deleted from the narrow syntactic derivation for inspection after undergoing Transfer, but the featural content of the phase complement becomes unalterable. The major consequence of this view is a nuanced conception of phase locality, such that some cross-phasal dependencies—namely those that do not require feature valuation of a transferred element—are possible. Cross-phasal dependencies that do value features of transferred elements continue to remain impossible, like in standard phase theory. I show that Hindi-Urdu ?-agreement and case assignment bear out the predictions of Read-Only withregard to cross-phasal dependencies. ?-agreement by a higher probe with a transferred goal, where the goal itself is not altered, is possible in Hindi-Urdu. On the other hand, accusative case assignment into a spelled out phase complement—which involves valuing the case feature of the transferred goal—is impossible. However, the same transferred DP that cannot be accusative is able to condition dative case on a DP in a higher phase. I argue that no notion of phases—other than Read-Only—accounts for the Hindi-Urdu pattern. The phase locality imposed by Read-Only offers a new way of accommodating dependencies between elements belonging to different phases in a principled way.