Roger P.G. van Gompel,1 Martin J. Pickering2 & Jamie Pearson1
1 University of Dundee, 2 University of Edinburgh
We report an eye-movement study that provides evidence that the use of subcategorisation information during syntactic ambiguity resolution is delayed. Some studies (Mitchell, 1987; Van Gompel & Pickering, in press; but cf. Adams et al., 1998) have found evidence for reading difficulty at "the vet" in (1a) relative to (1b), suggesting that subcategorisation information is initially ignored.
1a. While the dog struggled the vet took off the muzzle.
1b. While the dog scratched the vet took off the muzzle.
These findings are often taken as evidence against frequency-based accounts of parsing (e.g., MacDonald et al., 1994; Trueswell et al., 1993).
However, some frequency-based models (e.g., Juliano & Tanenhaus, 1996; Tabor & Tanenhaus, 1999) can account for these results, because they claim that the parser employs frequency information associated with specific verbs as well as frequency information associated with verbs in general (cf. Mitchell et al., 1995). Because verbs occur more often transitively than intransitively, the direct object analysis may be activated in (1a).
We tested models that claim that the processor uses both lexically specific and category-general frequency information by investigating sentences like (2):
2a. The nurse heard the man the radio in the opposite room kept annoying in the corridor.
2b. Not surprisingly the man the radio in the opposite room kept annoying complained a lot.
2c. The nurse heard the man who the radio in the opposite room kept annoying in the corridor.
2d. Not surprisingly the man who the radio in the opposite room kept annoying complained a lot.
In (2a), subcategorisation information of the monotransitive verb "heard" rules out the analysis in which "the radio" is the direct object of the preceding verb. This contrasts with ditransitive verbs such as (3):
3. The nurse gave the man the radio in the opposite room.
Importantly, verbs occur monotransitively more often than ditransitively. This rules out the possibility that there is a general verb preference for the direct object analysis. In order to eliminate the possibility that difficulty occurs in (2a) because it is the start of a new clause, we contrasted it with (2b), where "the radio" is the subject and cannot be attached to a preceding verb. Both conditions were compared with baseline conditions (2c) and (2d), which contained "who".
First-pass reading times for the region "in the opposite room" showed a significant interaction between the grammatical role of "the radio" and the presence of "who". Sentence (2a) took longer to read than (2c), whereas no clear difference occurred between (2b) and (2d). This indicates that readers initially analysed "the radio" as the direct object of "heard". We conclude that readers ignore both lexically specific and category-general subcategorisation information. Interestingly, this does not appear to be due to a simple temporal delay in the availability of subcategorisation information. If it were a temporal delay, we would expect that the intervening phrase "the man" provided a sufficient delay for the processor to access the verb information before it encountered "the radio". We argue that inherent properties of the sentence processor preclude the immediate use of subcategorisation information. Our results provide evidence against frequency-based accounts of parsing, and support models that claim that frequency information is initially ignored (Frazier, 1987; Pickering et al., 2000).