Binding reflexives and pronouns in picture noun phrases

Jeffrey T. Runner, Rachel S. Sussman and Michael K. Tanenhaus
University of Rochester

We investigated the interpretation of reflexives in picture noun phrases with possessors (PPNPs), e.g., "Harry's picture of himself", by having participants follow spoken instructions to manipulate an array of dolls and pictures belonging to those dolls while we monitored their eye movements. The questions of interest were (1) would reflexives in PPNPs be interpreted as referring to the possessor (e.g., Harry's picture of Harry), as predicted by current Binding Theories (BTs) and (2) would pronouns and reflexives exhibit the complementarity predicted by BT. Sample materials are:

1. Look at Joe. Pick up Ken. Have Ken touch Harry's picture of him/himself.

We manipulated the accessibility of the subject (e.g., Ken) as a potential referent for a subsequent pronoun by varying the order of the "Look at" and "Pick up" instructions.

The complementarity predicted by BT was not confirmed. On pronoun trials, 88.9 % of the choices were BT-compatible (i.e., the picture of the subject, Ken, or lead-in, Joe), compared to only 68.9% for the reflexives (the possessor's picture of the possessor), resulting in an interaction of Type of anaphor and BT-compatibility for both choices and looks (F(1,17)=9.08, MSE=.72, p<.01; (F(1,17)=39.63, MSE=.68, p<.01). Planned comparisons demonstrated that the lead-in manipulation affected the proportion of subject choices and looks for the pronouns, but not for the reflexives. The timing and pattern of fixations were closely time-locked to the referential expressions and ruled out two explanations that might reconcile the data with BT:

(1) There was not an initial stage of processing in which participants considered only BT-compatible referents for reflexives.
(2) Binding-incompatible looks did not reflect a lexical garden-path in which participants responded to "himself" as if it were a pronoun because of the embedded "him".

These results pose two problems for an approach based on standard BT. First, the interpretation of reflexives in PPNPs frequently differs from that predicted by BT. Second, BT accounts propose a complementary relationship between pronouns and reflexives, predicting that reflexives may take a non-BT appropriate referent, so long as this referent is
unavailable as a referent for a pronoun. Our accessibility manipulation strongly affected the likelihood that the subject was selected as the referent of the pronoun, while having no effect on whether the subject was chosen as the referent for the reflexive. On these grounds, we argue that reflexives in PPNPs are better viewed as "logophors", and develop the representational and processing implications of this view.