Vered Argaman & Neal J. Pearlmutter
Northeastern University
Argaman et al. (1998) examined lexical semantics as a source of argument structure frequency biases by extending lexical semantic theories (e.g., Pinker, 1989) and looking at frequency measures of Sentence Complement (SC)-taking verbs and their corresponding nouns (e.g., propose-proposal). They found overall reliable correlations across verb-noun pairs, and significant differences between semantic categories identified by Levin (1993). The current work extends these results with a larger data set, and with comparison to two additional categorizations:
(1) Similarity judgments for two sets of 35 communication verbs were collected from 139 undergraduates. For each set, a similarity matrix was produced, and submitted to a hierarchical cluster analysis. The results from the two solutions were integrated to extract 9 semantic categories, which resembled, but did not completely overlap with, Levin's (1993) categories.
(2) Verbs and their corresponding nouns are explicitly linked by a derivational morpheme, which can be responsible for a variety of semantic and/or syntactic effects. The choice of nominalizing morpheme may have an effect on argument structure frequency biases. Verbs were categorized to one of 5 morphological categories: -ion (decided-decision); -nce (insured-insurance), -ation (accused-accusation), -ment (judged-judgment), and zero derived (report-report).
Each categorization was used as the grouping variable in one-way ANOVAs for each of three frequency measures. A significant F indicates that the categorization captures a significant amount of the variability in the measure considered.
Fs for one-way ANOVAs for different bias measures and different categorization
(SC=sentence complement, DO=direct object, PP=prepositional phrase)
|
Categorization |
df | %SC | %DO | %PP |
| Levin (1993) | 14,48 | 5.34* | 6.05* | 3.83* |
| Similarity Judgments | 8,38 | 4.78* | 14.37* | 9.41* |
| Morphology | 4,88 | <1 | <1 | 2.62* |
Semantic categorization captured substantial variance in all three frequency measures, while morphological category captured variability only weakly and only in one measure, suggesting that argument structure biases are controlled mainly by semantic category.
To examine semantic category effects in comprehension, Argaman & Pearlmutter (1999) had participants read sentences containing a temporary SC/DO ambiguity, resolved as an SC. The semantic category of the verb was manipulated. While reading times in the disambiguating region were correlated with SC-bias, ambiguity effect size did not interact with semantic category.
We will present additional new results examining the effects of semantic category during processing, using a syntactic priming paradigm. Participants complete sentence-initial fragments (e.g., "Jane decided") after reading a set of 5 prime sentences. The argument structures in the prime set and the semantic relation between the prime set verbs and the target verb (same verb, verbs from same semantic category, verbs from different semantic category) are manipulated. The magnitude of the priming effect for the same category condition relative to the other two will indicate the degree to which semantic categories participate in processing.
References
Argaman, V., Pearlmutter, N. J., Garnsey, S. M., Mendelsohn, A. A., Randall, J., & Myers, E. (1998). Lexical semantics as a basis for argument structure frequency biases. Poster presented at the 11th Annual CUNY Conference on Human Sentence Processing, New Brunswick, NJ.
Argaman, V., & Pearlmutter, N. J. (1999). Verb semantic category and argument structure frequency biases in syntactic ambiguity resolution. Poster presented at the 12th Annual CUNY Conference on Human Sentence Processing, New York, NY.
Levin, B. (1993). English Verb Classes and Alternations. Chicago, IL: University of Chicago Press.
Pinker, S. (1989). Learnability and Cognition: The Acquisition of Argument Structure. Cambridge, MA: MIT Press.