Distinguishing effects of number interference and number attraction in sentence processing.

Several comprehension studies have found evidence for number attraction effects: Processing a number-inflected verb is harder when a NP with the same number intervenes between the subject and verb (Pearlmutter et al., 1999; Wagers et al., 2009). This predicts that "was" in Example 1 should be harder when the verb is preceded by a plural ("customers" or "directors") than a singular NP. In contrast, similarity-based interference (Dyke & Lewis, 2003; Vasishth & Lewis 2005) predicts retrieval interference when a verb needs to be integrated with its dependents. The retrieval of a dependent should be more difficult when it shares retrieval cues (e.g., number) with another phrase. This predicts longer reading times at "was" with singular interfering NPs (customer or director). 1. The secretary | who argued with | the customer(s) of the director(s) | was on | the train | to the meeting. An eye-tracking experiment showed a number interference effect from the local NP ("director") in first fixation for the train and from the second NP ("customer") in total times for the same region. There was no number attraction effect. One possibility is that we did not find evidence for number attraction because the number attractor was syntactically deeply embedded in a relative clause. To test number attraction from nouns not in a relative clause, the intervening attractor ("farmers") was in a PP in Experiment 2. We also tested whether attraction effects might be part of checking processes after initial structure building. If so, these effects should be strong during reanalysis. Thus, we compared structurally ambiguous (2, without comma) and unambiguous sentences (with comma). The verb "has" not only resolves the ambiguity, it also agrees with "cousin". Thus, when the verb is preceded by a plural number attractor ("farmers"), reanalysis should be harder according to the number attraction account. 2. After Virginia answered(,) the cousin of the farmer(s) | has to | think it | all over again. Ambiguous sentences took longer to read than unambiguous sentences. More importantly, although there was no attraction or interference effect for unambiguous sentences, regression-path times for "has to" in the ambiguous sentences were longer when there was a plural than singular attractor. This suggests that number attraction affects comprehension as part of a checking process. We will discuss the implications of both experiments for number attraction and similarity-based interference.