Tessa Warren1 & Ted Gibson2
1 University of Massachusetts, Amherst, 2 Massachusetts Institute of Technology
Recent work has demonstrated that the length of structural dependencies affects sentence complexity and reading times (e.g., Gibson, 1998; Grodner, Watson & Gibson, 2000). Past investigations of distance have focused on properties of words intervening between the endpoints of dependencies, such as the amount of referential processing they require (Warren, 2001). In this poster, we investigate an alternative distance metric that past work has not ruled out.
In his (1998) paper, Gibson calculated the length of syntactic dependencies according to the distance between the heads of the dependent clauses. But alternatively, the length of a syntactic dependency could reflect the distance between complete constituents rather than their heads. In this case, distance would be calculated between the right edge of the first constituent and the left edge of the second constituent. One location where these distance metrics make divergent predictions is at the matrix verb of a sentence whose NP subject is modified by a constituent intervening between its head and the upcoming verb. If the distance between complete constituents affects the difficulty of integration then including a modifier will not affect reading times at the matrix verb, because the modifier is part of the subject NP constituent. Alternatively, if the distance between heads affects the difficulty of integration, then including a modifier will cause increased reading times on the matrix verb. Previous studies have manipulated the amount of discourse processing required in a subject-modifying constituent and shown that reading times on the matrix verb are longer when more processing is required during the NP, as predicted by the head-based distance metric. But in most of these studies, the embedded verb and matrix verb are in consecutive positions, so the effects observed at the matrix verb could be spill-over from the embedded verb.
The two experiments in this poster tested sentences like (a), where an object-extracted RC modifies the head of the subject NP. The subject of the RC was either new to the discourse or had previously appeared in context. A phrase intervened between the two verbs to eliminate the possibility of reading time spill-over.
(a) The suspect, who the/a detective had sighted the previous Wednesday, struggled but was eventually subdued.
Both experiments showed reliable effects at the embedded verb, such that conditions requiring more discourse processing in the subject modifier had longer reading times. But neither experiment showed any differences in reading times at the matrix verb. This finding supports a constituent-based distance metric over a head-based distance metric.
References
Gibson, E. (1998). Syntactic complexity: Locality of syntactic dependencies. Cognition, 68(1), 1-76.
Grodner, D., Watson, D., & Gibson, E. (2000). Locality effects in processing unambiguous sentences. Talk at 13th Annual CUNY Conference on Human Sentence Processing, San Diego, CA.
Warren, T. (2001). Understanding the Role of Referential Processing in Sentence Complexity. Ph.D. thesis, MIT.