Comprehension of elided structure: Evidence from sluicing

Michael Walsh Dickey, Ann Bunger

Language and Cognitive Processes · 2010 · 103 citations · 32 references

Concepts

Abstract

Abstract In sluiced sentences like "Nolan colored, but he wouldn't tell me what", the elided clause ("what [Nolan colored _]") contains a gap not present in the antecedent clause ("Nolan colored"). Previous work has shown that interpreting such nonparallel sluiced sentences creates measurable processing difficulty, and these processing costs have been interpreted as evidence for an ellipsis-specific "sprouting" operation. The current study compared sluiced sentences and their nonelided counterparts in a self-paced reading task. Results show that nonparallelism was equally costly for both sluiced and nonelided sentences. This finding suggests that these processing costs are as likely to be due to a violation of parallelism expectations for coordinated structures as to be due to ellipsis-specific operations. For nonelided sentences, moreover, the transitivity biases of individual verbs affected the magnitude of these parallelism costs: verb-specific transitivity biases were a reliable predictor of reading times for the nonelided sentences, though not for the sluiced sentences. Implications for the representation and processing of elided structure and the role of verb-specific transitivity biases are discussed. Keywords: Sentence processingEllipsisSluicingLexical information Notes 1Still other accounts claim that sluicing does not involve any unpronounced material (e.g., Culicover & Jackendoff, Citation2005), rather, it involves an unpronounced clausal anaphor or pointer which refers to the antecedent clause. Mismatches between the semantic requirements of the clausal anaphor or pointer and the syntactic form of the antecedent are also presumably costly. These accounts are similar to the sprouting account of Chung et al. (1995) in claiming that the syntactic representation of sluiced clauses is qualitatively different from the representation of otherwise similar nonelided sentences. 2A reviewer raises the question of whether readers may have been assigning implicit prosody to nonelided conditions, which is consistent with a deaccented reading of these sentences. If readers did indeed assign such an implicit prosody to the nonelided condition, that would undercut the strength of the sluicing vs. nonelided comparison, since ellipsis and deaccenting are subject to similar semantic and pragmatic constraints (cf. Merchant, Citation2001; Schwarzchild, Citation1999; Tancredi, Citation1992). There is significant evidence suggesting that comprehenders assign implicit prosodic contours to sentences during silent reading (see Fodor, Citation1998, among many others), and some readers may have implicitly deaccented the relevant material in the nonelided sentences. However, such implicit deaccenting would seem to work against the observed main effect of ellipsis. If readers did implicitly deaccent the material following what, that deaccenting should have made the nonelided condition more similar to the elided condition and thus should have reduced the difference between the elided and nonelided conditions. To the extent that the main effect of ellipsis is a reliable one, it provides suggestive evidence that readers did not engage in much implicit deaccenting, or that the effect of such deaccenting was not sufficient to obscure the (other) differences between the conditions. Nonetheless, it would be preferable to be able to establish that readers did not implicitly deaccent the material in the nonelided condition, for instance in a follow-up study with stimuli presented auditorily and a direct manipulation of prosody. We are grateful to the reviewer for this point.

References

32