The ambiguity advantage effect in reading is due to garden-pathing: New evidence from eye movements

An ambiguously attached relative clause (e.g., The grandmother of the girl who admired herself in the mirror) is on average read faster than one with an unambiguous "high" attachment (The grandfather of the girl who admired himself in the mirror) or an unambiguous "low" attachment (The grandfather of the girl who admired herself in the mirror). A classic account of this "ambiguity advantage effect" (AAE) attributes it to the operation of a serial but stochastic parser: The parser's initial attachment of the relative clause varies from trial to trial, so re-analysis is sometimes necessary upon disambiguation toward either analysis. Because this account claims that the AAE is due to garden-pathing on a subset of unambiguous trials, it predicts that the AAE in eye movements should be due to an increase in the frequency of regressive re-reading in both unambiguous conditions, rather than to a general reading time advantage in the ambiguous condition. In a large (N = 121) eye-tracking experiment, regressive eye movements from both the critical disambiguating reflexive and subsequent sentence regions were more frequent in the unambiguous conditions, leading to sizable go-past reading-time effects. However, first-pass reading times showed at most an extremely small AAE that was restricted to the reflexive itself; in a first-pass analysis that included only non-regressive trials, the effect was not fully credible. This pattern supports an explanation of the AAE as reflecting primarily a re-analysis cost in the unambiguous conditions.