Recursive linearization

Abstract

Linguistic structures are represented as hierarchical tree structures, although utterances are linear. To deal with this discrepancy, the grammar model needs to specify in which way linear order is derived from hierarchical structure. This article examines the idea that linearization is a PF procedure, implemented as a recursive procedure searching the hierarchical tree top-down for terminal elements. It is shown that we can describe linearization by positing one linearization principle and two parameters, one a modified head parameter, and one parameter to linearize adjuncts. The resulting system is applied to the domain of the Arabic and English noun phrases. Although these domains have vastly different linear orders, with the proposed model their syntactic structures turn out to be only minimally different.

As developed here, the model makes a fundamental prediction about language: order variation only exists in positions created by External Merge. That is, an element that is moved will invariably shift toward the left in the linear string. It is tentatively suggested that this property results from efficiency requirements of the parser, which would show that the language faculty is perhaps primarily but not solely adapted to the requirements imposed by the C-I interface.