Andrews, C. , Yacovone, A. , Sloggett, S. & Dillon, B.
University of Massachusetts Amherst
Cue-based parsing models (Lewis & Vasishth, 2005; McElree, 2000; van Dyke, 2007) hold that grammatically illicit elements interfere with linguistic dependency formation. Such theories explain the finding that mismatch between a verb and its subject often goes unnoticed if non-subject NPs ("lures") match the verb (i.e. "agreement attraction"; Wagers et al. 2009). However, comparable effects for reflexive pronouns, surface-similar dependencies, have proved elusive. Many studies find that reflexive comprehension, unlike agreement, is strongly grammatically constrained--insensitive to the features of lures (Sturt, 2003; Dillon et al. 2013). However, such studies frequently embedded lures inside relative clauses, a position which weakens agreement attraction (Bock & Miller 1991). We present an eye-tracking study which compares the sensitivity of agreement and reflexives to lures in prepositional phrases (see (1)), a strong agreement attraction position. We find that subject-mismatched verbs are read faster if a feature-matched lure is present (i.e. agreement attraction), while reflexives with mismatched antecedents are read slowly regardless of the lure's features. Thus, reflexives appear insensitive to lures even in strong agreement attraction environments, indicating a qualitative difference in the memory access profile of reflexives and agreement.
(1) *The nurse of the widows {were embarrassed/ embarrassed themselves}...