In an eye-tracking experiment that used sentences containing a direct object/complement ambiguity, it was found that readers employed different levels of information at different stages of processing. Thus syntax, semantics, and pragmatic inference were employed differentially.
We compared single meaning sentences such as (1):
in which the disambiguation occurs at the second verb and is complete, with sentences such as (2):
that has two possible interpretations, and which the reader may or may not find ambiguous. Although ``her pony'' is clearly the subject of the main verb ``broke'', there is still a plausible inference that Sally is riding her pony as it breaks into a canter. Sentences, such as (3), which did not contain a garden path, were also read.
Each stimulus sentence was followed by a comprehension question such as (4):
that checked the reader's interpretation of the ambiguous NP.
Reading times were longer for single interpretation sentences than for the dual interpretation sentences. Differences were apparent in rereading and were responsible for the overall difference found in total reading times.
The comprehension questions following sentences with dual interpretations were answered ``True'' about 60% of the time.
Since the initial grouping of words is responsible for the garden-path effect, we considered it important to see how far, if at all, this influenced the final interpretation. In a second experiment the clause order was reversed to establish whether the same inference was made when there is no initial ambiguity but the dual interpretation remains plausible. Responses showed that the effect of inferential information was just as strong as that found for garden path sentences. Participants answered ``True'' slightly more frequently than to the garden path sentences, although the difference was not significant, indicating that the initial garden pathed reading had no noticeable effect on final interpretation. Patterns of rereading were different for the reverse clause sentences, showing that word order influenced the choice of which sentential information was revisited. However, as with the garden path sentences, the reverse clause sentences showed much more rereading in the single meaning condition. Strategies for reanalysis were thus seen to vary depending on sentential information such as word order, and plausible inference.