@inproceedings{LCNL:136,
	author={Michael S. Amato and Jon A. Willits and Maryellen C. MacDonald},
	year={2009},
	title={Verb aspect and argument activation: World vs. Word Knowledge},
	journal={CUNY Conference on Human Sentence Processing 2009},
	abstract={Much sentence processing work has investigated the role of structural patterns vs. item-specific information (e.g., verb transitivity). Relatively little work has examined information that is neither item-specific nor structural.  We address this gap in four studies examining the role of aspect in verb argument processing.
Ferretti, Kutas, and McRae (2007) found shorter naming RTs for target locations (e.g., kitchen) following related imperfect aspect verbs (IA: was cooking) compared to perfect aspect (PA: had cooked). They argued that IA verbs evoke a situation model that selectively activates likely location arguments from an event schema, while PA does not. This suggests an abstract effect of aspect that is independent of the specific verb being used.
 Study 1 replicated Ferretti et al.’s materials and procedure and found the same effect of aspect.  Study 2 was a corpus analysis investigating aspect-verb-noun co-occurrence probabilities, namely whether kitchen is more likely given was cooking than had cooked. We counted target locations within a 9-word window of the IA/PA in two corpra: a 532 million word corpus of Wikipedia articles and the 360 million word Corpus of Contemporary American English. Location targets were more probable in the context of the IA than the PA verbs, suggesting the IA advantage in naming could arise from participant sensitivity to conditional probabilities between specific pairs of verb-forms and nouns. 
Study 3 used corpus data to manipulate aspect and co-occurrence probability independently. 30 target locations were selected, each with 6 verb primes crossing aspect (IA/PA) and verb-location relationship (related high-probability, related low-probability, unrelated). Chosen verbs had large probability differences in IA and PA to the same target: 
primes for Ocean and P(ocean|prime)	high-prob	low-prob	unrelated
	was drifting (1.11%)	was sinking (0.37%)	was assaulting (0.00%)
	had sunk (1.73%)	had drifted (0.20%)	had assaulted (0.00%)

In norming, high and low probability primes were judged equally related to the target, and more so than were unrelated primes. The same naming task was used with these new items (n=78). ANOVA revealed a main effect of relatedness; RTs were shorter following related primes, F1(2,72)=6.57, p<.01, F2(2,24)=8.18, p<.01; and a marginal interaction: low-probability verb primes sped target naming only in IA, F1(2,72)=2.89, p=.06, F2(2,24)=4.83, p=.02. There was no main effect of aspect, Fs<1. These results suggest primes activated arguments that were statistically likely for the specific verb regardless of aspect, but that low probability locations got a boost from IA primes. 
Study 4 used sentence completion to test the effects of aspect and co-occurrence probability in sentence contexts. 95 participants completed sentence fragments containing a subject NP and IA/PA phrase (Mary was drifting...). The original Ferretti et al. verbs elicited more location completions in IA than PA, t(20)=.2.75, p<.05. However the Study 3 verbs, deliberately chosen to eliminate the IA probability advantage for one particular location, elicited an equal number of locations across aspects t<1.
These results confirm that aspect influences argument activation during sentence processing, but also demonstrate that the effect is contingent on verb-specific information. Implications for generalizations from item-specific constraints are discussed.
},
	language={English},
	keywords={word vs. world, aspect, item specific},
	URL={http://lcnl.wisc.edu/publications/archive/136.pdf},
}