The Origins of Obligatoriness: Evidence from Agreement Failures

Friday, 8 October 2010, Syntax Colloquia

Omer Preminger, MIT

Abstract

Within the generative tradition, the obligatory nature of agreement in phi-features (number, person, gender, etc.) has often been captured in terms of what I will call “derivational time-bombs”: elements of the initial representation that cannot be part of a well-formed, end-of-the-derivation representation (e.g., ‘uninterpretable’ features; Chomsky 2000, 2001). These “time-bombs”, in turn, can only be “defused” by successfully establishing a phi-agreement relation.

In this talk, I present evidence from Hebrew showing that this way of relating phi-agreement to grammaticality is lacking: there are certain empirical patterns that cannot be captured without explicit recourse to an obligatory *operation*, which is responsible for phi-agreement, and whose obligatory status cannot be subsumed by appealing to the initial state of the phi-features on the putative bearer of agreement-morphology. Furthermore, I argue that the “derivational time-bomb” approach cannot be salvaged – at least as far as Hebrew is concerned – by appealing to mechanisms that recast apparent agreement failures as agreement with alternate targets (e.g., agreement-attraction, expletive/default agreement, or agreement with intervening functional structure).

This result has consequences for the way we conceive of obligatoriness in syntax – and in fact, the way we conceive of the syntactic computation in general. Recent work in generative linguistics has seen a tendency to eliminate “operations”, as such. Instead, there has been a shift towards a model of grammar that relies on more articulated *representations*, while adopting a radically reduced inventory of *operations* (perhaps as few as two: Agree and Merge; Chomsky 2008). These fundamental operations are in turn seen as neither “obligatory” nor “optional”, unto themselves; they are subservient to creating a well-formed representation, and are freely deployed by the computational system to that end.

If the arguments here are correct, this shift is ill-conceived, at least in the domain of phi-agreement.