Robert Thornton and Maryellen C. MacDonald
University of Southern California
A number of recent proposals have implicated a crucial role for phrase length in sentence processing (Fodor, 1998; Gibson, 1998; MacDonald, in press), but few studies have explicitly manipulated this factor to determine its effect on processing. In two experiments, we investigated the role of phrase length in resolving complex verb phrase (VP) modification ambiguites, as in (1-2). Although these constructions have received relatively little experimental investigation, many theories make strong claims as to how they will be resolved. For example, Construal theory (Frazier & Clifton, 1996) predicts that examples such as (1) are handled by late closure such that (1b), in which the modifier attaches to the local site, should initially be significantly easier to process than (1a), in which the modifier attaches to the distant site.
| (1) | a. | VP1 attachment: |
| She [VP1 taught the sixth and seventh grade kids [VP2 to dive] [PP in a single afternoon]] | ||
| b. | VP2 attachment: | |
| She [VP1 taught the sixth and seventh grade kids [VP2 to dive [PP into the deep end]]] |
Moreover, this principle is thought to determine attachment preferences irrespective of length, such that the same pattern of results should occur for (2), namely that (2b) should be significantly easier to process that (2a).
| (2) | a. | VP1 attachment: |
| She [VP1 taught the kids [VP2 to dive] [PP in a single afternoon]] | ||
| b. | VP2 attachment: | |
| She [VP1 taught the kids [VP2 to dive [PP into the deep end]]] |
In contrast, Gibson's (1998) proposal suggests that increased distance between a modifier and its attachment site will result in increased processing difficulty. Thus, while local attachment should be preferred over distant attachment for both (1) and (2), distant attachment should be significantly easier for (2a) than for (1a). Although Fodor's (1998) prosodic account and MacDonald's (in press) distributional account differ substantially from Gibson's proposal and from each other, they make very similar predictions regarding length phenomena.
The results of two self-paced reading experiments support the length-based predictions. Reading times at the disambiguation (the PP) revealed a significant length x disambiguation interaction; there was a strong local attachment preference for sentences like (1), in which the complex VP is significantly longer than the modifier. No such preference was observed for sentences like (2), in which the phrases are about the same length. These data highlight an important role for phrase length in theories of sentence processing, consistent with all three length-sensitive hypotheses.
Having confirmed the role of length in ambiguity resolution in this construction, we next discuss how to distinguish the Gibson, Fodor, and MacDonald hypotheses. We will discuss four domains in which some or all of the hypotheses make different predictions: (1) the role of length in other ambiguous constructions, (2) the importance of absolute versus relative length of constituents, (3) relationships to phrasal ordering phenomena in the production literature, and (4) their motivation as speaker-oriented versus hearer-oriented phenomena.